Original Research

Sleep apnea is a heterogeneous group of conditions that may be attributable to a wide array of underlying conditions, with varying contributions of obstructive or central sleep-disordered breathing. The spectrum from obstructive sleep apnea (OSA) to central sleep apnea (CSA) includes mixed sleep apnea, treatment-emergent CSA (TECSA), and Cheyne-Stokes respiration (CSR).¹ The pathophysiologic causes of CSA can be attributed to delayed cardiopulmonary circulation in heart failure, decreased brainstem ventilatory response due to stroke, blunting of central chemoreceptors in chronic opioid use, and/or stimulation of the Hering-Breuer reflex from activation of pulmonary stretch receptors after initiating positive airway pressure (PAP) for treatment of OSA.^2,3 Medications are commonly implicated in many forms of sleep-disordered breathing; importantly, opioids and benzodiazepines may blunt the respiratory drive, leading to CSA, and/or impair upper airway patency, resulting in or worsening OSA.

Continuous positive airway pressure (CPAP) therapy is largely ineffective in correcting CSA or improving outcomes and is often poorly tolerated in these patients.⁴ Adaptive servo-ventilation (ASV) is a form of bilevel PAP (BPAP) therapy that delivers variable adjusting pressure support, primarily to treat CSA. PAP also may relieve upper airway obstructions, thereby effectively treating any comorbid obstructive component. ASV has been well documented to improve sleep-related disorders and improve apnea-hypopnea index (AHI) in patients with CSA. However, longitudinal data have demonstrated increased mortality in patients with heart failure with reduced ejection fraction (HFrEF) who were treated with ASV.⁵ Since the SERVE-HF trial results came to light in 2015, there has been no consensus regarding the optimal use, if any, of ASV therapy.^6-8 This is partly related to the inability to fully explain the study’s major findings, which were unexpected at the time, and partly due to the absence of similar relevant mortality data in patients with CSA but without HFrEF.

TECSA may present in some patients with OSA who are new to PAP therapy. These events are frequently seen during PAP titration sleep studies, though patients can also experience significant TECSA shortly after initiating home PAP therapy. TECSA is felt to result from a combination of stimulating pulmonary stretch receptors and lowering arterial carbon dioxide below the apneic threshold. Chemoreceptors located in the medulla respond by attenuating the respiratory drive.⁹ Previous studies have shown most cases of mild TECSA resolve over time with CPAP treatment. However, in patients with persistent or worsening TECSA, ASV may be considered as an alternative to CPAP.

The prevalence of OSA in the veteran population is estimated to be as high as 60%, considerably higher than the general population estimation.¹⁰ Patients with more significant comorbidities may also experience a higher frequency of central events. Patients with CSA have also been shown to have a higher risk for cardiac-related hospital admissions, providing plausible justification for correcting CSA.¹⁰

In the current study, we aim to characterize the group of patients using ASV therapy in the modern era. We will assess the objective efficacy and adherence of ASV therapy in patients with primarily CSA compared with those having primarily OSA (ie, TECSA). Secondarily, we aim to identify baseline clinical and polysomnographic features that may be predictive of ASV adherence, as a surrogate for subjective benefit.¹¹ In the wake of the SERVE-HF study, the sleep medicine community has paused prescribing ASV therapy for CSA. We hope to provide more perspective on the treatment of veterans with CSA and identify the patient groups that would benefit most from ASV therapy.

 

Methods

This retrospective chart review examined patients prescribed ASV therapy at the Hampton Veterans Affairs Medical Center (HVAMC) in Virginia who had therapy data between January 1, 2015, and April 30, 2020. The start date was chosen to approximate the phase-in of wireless PAP devices at HVAMC and to correspond with the release of preliminary results from the SERVE-HF trial.

Patients were initially identified through a query into commercial wireless PAP management databases and cross-referenced with HVAMC patients. Adherence and efficacy data were obtained from the most recent clinical PAP data, which allowed for the evaluation of patients who discontinued therapy for reasons other than intolerance. Clinical, demographic, and polysomnography (PSG) data were obtained from the electronic health record. One patient, identified through the database query but not found in the electronic health record, was excluded. In cases of missing PSG data, especially AHI or similar values, all attempts were made to calculate the data with other provided values. This study was determined to be exempt by the HVAMC Institutional Review Board (protocol #20-01).

Statistics

Statistical analyses were designed to compare clinical characteristics and adherence to therapy of those with primarily CSA on PSG and those with primarily OSA. Because it was not currently known how many patients would fit into each of these categories, we also planned secondary comparisons of the clinical and PSG characteristics of those patients who were adherent with therapy and those who were not. Adherence with ASV therapy was defined as device use for ≥ 4 hours for ≥ 70% of nights.

Comparisons between the means of 2 normally distributed groups were performed with an unpaired t test. Comparisons between 2 nonnormally distributed groups and groups of dates were done with the Mann-Whitney U test. The normality of a group distribution was determined using D’Agostino-Pearson omnibus normality test. Two groups of dichotomous variables were compared with the Fisher exact test. P value < .05 was considered statistically significant.

Results

Thirty-one patients were prescribed ASV therapy and had follow-up at HVAMC since 2015. All patients were male. The mean (SD) age was 67.2 (11.4) years, mean body mass index (BMI) was 34.0 (5.9), and the mean (SD) Epworth Sleepiness Scale (ESS) score was 10.9 (5.8). Patient comorbidities included 30 (97%) with hypertension, 17 (55%) with diabetes mellitus, 16 (52%) with coronary artery disease, and 11 (35%) with congestive heart failure. Three patients had no echocardiogram or other documentation of left ventricular ejection fraction (LVEF). One of these patients had voluntarily stopped using PAP therapy, another had been erroneously started on ASV (ordered for fixed BPAP), and the third had since been retitrated to CPAP. In the 28 patients with documented LVEF, the mean (SD) LVEF was 61.8% (6.9). Ten patients (32%) had opioids documented on their medication lists and 6 (19%) had benzodiazepines.

The median date of diagnostic sleep testing was January 9, 2015, and testing was completed after the release of the initial field safety notice regarding the SERVE-HF trial preliminary findings May 13, 2015, for 14 patients (45%).¹² On diagnostic sleep testing, the mean (SD) AHI was 47.3 (25.6) events/h and the median (IQR) oxygen saturation (SpO₂) nadir was 82% (78-84). Three patients (10%) were initially diagnosed with CSA, 19 (61%) with OSA, and 9 (29%) with both. Sixteen patients (52%) had ASV with fixed expiratory PAP (EPAP), and 15 (48%) had variable adjusting EPAP. Mean (SD) usage of ASV was 6.5 (2.6) hours and 66.0% (34.2) of nights for ≥ 4 hours. Mean (SD) titrated EPAP (set or 90th/95th percentile autotitrated) was 10.1 (3.4) cm H₂O and inspiratory PAP (IPAP) (90th/95th percentile) was 17.1 (3.3) cm H₂O. The median (IQR) residual AHI on ASV was 2.7 events/h (1.1-5.1), apnea index (AI) was 0.4 (0.1-1.0), and hypopnea index (HI) was 1.4 (1.0-3.2); the residual central and obstructive events were not available in most cases.

 

Adherence

Seventeen patients (55%) met the minimum adherence criteria of ≥ 4 hours of usage for ≥ 70% of the nights. There were no significant differences in age, BMI, sex, race, comorbidities, medications, or ESS when comparing patients who were adherent to ASC and those who were not (Table 1). The date of diagnostic sleep testing and sleep architecture, including sleep latency, total sleep time, sleep efficiency, wake after sleep onset, arousal index, and percentage of rapid eye movement stage sleep were similar between the adherent patients and nonadherent patients (Table 2).

The overall AHI mean (SD) on diagnostic PSG were also similar between the adherent group (52.3 [24.8] events/h) and nonadherent group (45.1 [27.0] events/h) (P = .47). The mean (SD) for obstructive AHI (obstructive apneas, mixed apneas, obstructive hypopneas, and undifferentiated hypopneas per hour of sleep) were higher in the adherent group as a percentage of the total AHI: 81.5% (27.9) in the adherent group vs 46.7% (38.4%) in the nonadherent group; P = .02) (Figure 1).

This difference was primarily driven by a difference in mean (SD) HI: 29.7 (16.5) in the adherent group vs. 15.3 (12.1) in the nonadherent group (P = .04).

There were no significant differences between the proportions of patients on ASV with set EPAP or the titrated EPAP and IPAP. The median (IQR) residual AHI was lower in the adherent group compared with the nonadherent group, both in absolute values (1.7 [0.9-3.2] events/h vs 4.7 [2.4-10.3] events/h, respectively [P = .004]), and as a percentage of the pretreatment AHI (3.1% [2.5-6.0] vs 10.2% [5.3-34.4], respectively; P = .002) (Figure 2).

Primarily Obstructive Sleep Apnea

Sleep apnea was a mixed picture of obstructive and central events in many patients. Only 3 patients had “pure” CSA. Thus, we were unable to define discrete comparison groups based on the sleep-disordered breathing phenotype. We identified 19 patients with primarily OSA (ie, initially diagnosed with OSA, OSA with TECSA, or complex sleep apnea). The mean (SD) age was 66.1 (12.8) years, BMI was 36.2 (4.7), and ESS was 11.4 (5.6). The mean (SD) baseline AHI was 46.9 (29.5), obstructive AHI was 40.5 (30.4), and central AHI was 0.4 (1.2); the median (IQR) SpO₂ nadir was 81% (78%-84%). The mean (SD) titrated EPAP was 10.2 (3.5) cm H₂O, and the 90th/95th percentile IPAP was 17.9 (3.5) cm H₂O. The mean (SD) usage of ASV was 7.9 (5.3) hours with 11 patients (58%) meeting the minimum standard for adherence to ASV therapy.

No significant differences were seen between the adherent and nonadherent groups in clinical or demographic characteristics or date of diagnostic sleep testing (eAppendix, available online at doi:10.12788/fp.0374). In baseline sleep studies the mean (SD) HI was 32.3 (15.8) in the adherent group compared with 14.7 (8.8) in the nonadherent group (P = .049). In contrast, obstructive AHI was not significantly lower in the adherent group: 51.9 (30.9) in the adherent group compared with 22.2 (20.6) in the nonadherent group (P = .09). The median (IQR) residual AHI on ASV as a percentage of the pretreatment AHI was 3.0% (2.4%-6.5%) in the adherent group compared with 11.3% (5.4%-89.1%) in the nonadherent group, a statistically significant difference (P = .01). No other significant differences were seen between the groups.

 

Discussion

This study describes a real-world cohort of patients using ASV therapy and the characteristics associated with benefit from therapy. The patients that were prescribed and started ASV therapy most often had a significant degree of obstructive component to sleep-disordered breathing, whether primary OSA with TECSA or comorbid OSA and CSA. Moreover, we found that a higher obstructive AHI on the baseline PSG was associated with adherence to ASV therapy. Another important finding was that a lower residual AHI on ASV as a proportion of the baseline was associated with PAP adherence. Adherent patients had similar clinical characteristics as the nonadherent patients, including comorbidities, severity of sleep-disordered breathing, and obesity.

Though the results of the SERVE-HF trial have dampened the enthusiasm somewhat, ASV therapy has long been considered an effective and well-tolerated treatment for many types of CSA.¹³ In fact, treatments that can eliminate the central AHI are fairly limited.^4,14 Our data suggest that ASV is also effective and tolerated in OSA with TECSA and/or comorbid CSA. Recent studies suggest that CSA resolves spontaneously in a majority of TECSA patients within 2 to 3 months of regular CPAP use.¹⁵ Other estimates suggest that persistent TESCA may be present in 2% of patients with OSA on treatment.¹⁶

Given the high and rising prevalence of OSA, many people are at risk for clinically significant TESCA. Another retrospective case series found that 72% of patients that failed treatment with CPAP or BPAP during PSG, met diagnostic criteria (at the time) for CSA; ASV was objectively beneficial in these patients.¹⁷ ASV can be an especially useful modality to treat OSA in patients with CSA that either prevents tolerance of standard therapies or causes clinical consequences, presuming the patient does not also have HFrEF.¹⁸ The long-term outcomes of treatment with ASV therapy remain a matter of debate.

The SERVE-HF trial remains among the only studies that have assessed the mortality effects of CSA treatments, with unfavorable findings. Treatment of OSA has been associated with favorable chronic health benefits, though recent studies have questioned the degree of benefit attributable to OSA treatment.^19-24 Similar studies have not been done for comorbidities represented by our study cohort (ie, OSA with TECSA and/or comorbid CSA).

The lack of CSA diagnosis alone in our cohort may be partially attributable to changing practice patterns following the SERVE-HF trial, though it is not clear from these data why a higher baseline obstructive AHI was associated with adherence to ASV therapy. Our data in this regard are somewhat at odds with the preliminary results of the ADVENT-HF trial. In that study, adherence to ASV therapy in patients with predominantly OSA declined significantly more than in patients with predominantly CSA.²⁵ Most of our patients were diagnosed with predominantly OSA, so a direct comparison with the CSA group is problematic; additionally, the primary brand and the pressure adjustments algorithm used in our study differed from the ADVENT-HF trial.

OSA and CSA may present with similar clinical symptoms, including sleep fragmentation, insomnia, and excessive daytime sleepiness; however, the degree of symptomatology, especially daytime sleepiness, and the response to treatment, may be less in CSA.^2,26 Both the subjective report of symptoms (ESS) and PSG measures of sleep fragmentation were similar in our patients, again likely explained by the predominance of obstructive events.

The pathophysiology of CSA is more varied than OSA, which is probably relevant in this case. ASV was originally designed for the management of CSA with CSR, accomplishing this goal by stabilizing the periods of central apnea and hyperpnea characteristic of CSR.²⁷ Although other forms of CSA demonstrate breathing patterns distinct from CSR, ASV has become an accepted treatment for most of these. It is plausible that the long-term subjective benefit and tolerance of ASV in CSA without CSR is less than for CSA with CSR or OSA. None of the patients in our study had CSA with CSR.

Ultimately, it may be the objective treatment effect that lends to adherence, as has been shown previously in OSA patients; our group of adherent patients showed a greater improvement in AHI, relative to baseline, than the nonadherent patients did.²⁸ The technology behind ASV therapy can greatly reduce the frequencies of central apneas, yet this same treatment effectively splints the upper airway and even more effectively eliminates obstructive apneas and hypopneas. Variable adjusting EPAP devices would plausibly provide even more benefit in these patients, as has been shown in prior studies.²⁹ To the contrary, our small sample of patients with TESCA showed a nonsignificant trend toward adherence with fixed EPAP ASV.

Opioid use was substantial in our population, without significant differences between the groups. CPAP therapy is ineffective in improving opioid-associated CSA. In a recent study, 20 patients on opioid therapy with CSA were treated with CPAP therapy; after several weeks, the average therapeutic use was 4 to 5 hours per night and CPAP was abandoned in favor of ASV therapy due to persistent central apnea. ASV treatment was associated with a considerable reduction in central apnea index, AHI, arousal index, and oxygen desaturations in a remarkable improvement over CPAP.³⁰

 

Limitations and Future Directions

This retrospective, single-center study may have limited applicability to other populations. Adherence was used as a surrogate for subjective benefit from treatment, though benefit was not confirmed by the patients directly. Only patients seen in follow-up for documentation of the ASV download were identified for inclusion and data analysis. As a single center, we risk homogeneity in the treatment algorithms, though sleep medicine treatments are often decided at the time of the sleep studies. Studies and treatment recommendations were made at a variety of sites, including our sleep center, other US Department of Veterans Affairs hospitals, in the community network, and at US Department of Defense centers. Our population was homogenous in some ways; notably, 100% of our group was male, which is substantially higher than both the veteran population and the general population. Risk factors for OSA and CSA are more common in male patients, which may partially explain this anomaly. Lastly, with our small sample size, there is increased risk that the results seen occurred by chance.

There are several areas for further study. A larger multicenter study may permit these results to be generalized to the population and should include subjective measures of benefit. Patients with primarily CSA were largely absent in our group and may be the focus of future studies; data on predictors of treatment adherence in CSA are lacking. With the availability of consistent older adherence data, comparisons may be made between the efficacies of clinical practice habits, including treatment efficacy, before and after the results of the SERVE-HF trial became known.

Conclusions

In selected patients with preserved LVEF, ASV therapy appears especially effective in patients with OSA combined with CSA. Adherence to ASV treatment was associated with higher obstructive AHI during the baseline PSG and with a greater reduction in the AHI. This understanding may help guide sleep specialists in personalizing treatments for sleep-disordered breathing. Because objective efficacy appears to be important for therapy adherence, clinicians should be able to consistently determine the obstructive and central components of the residual AHI, thus taking all information into account when optimizing the treatment. Additionally, both OSA and CSA pressure requirements should be considered when developing ASV devices.

Acknowledgments

We thank Martha Harper, RRT, of Hampton Veterans Affairs Medical Center (HVAMC) for helping to identify our patients and assisting with data collection. This material is the result of work supported with resources and the use of HVAMC facilities.

Sleep apnea is a heterogeneous group of conditions that may be attributable to a wide array of underlying conditions, with varying contributions of obstructive or central sleep-disordered breathing. The spectrum from obstructive sleep apnea (OSA) to central sleep apnea (CSA) includes mixed sleep apnea, treatment-emergent CSA (TECSA), and Cheyne-Stokes respiration (CSR).¹ The pathophysiologic causes of CSA can be attributed to delayed cardiopulmonary circulation in heart failure, decreased brainstem ventilatory response due to stroke, blunting of central chemoreceptors in chronic opioid use, and/or stimulation of the Hering-Breuer reflex from activation of pulmonary stretch receptors after initiating positive airway pressure (PAP) for treatment of OSA.^2,3 Medications are commonly implicated in many forms of sleep-disordered breathing; importantly, opioids and benzodiazepines may blunt the respiratory drive, leading to CSA, and/or impair upper airway patency, resulting in or worsening OSA.

Continuous positive airway pressure (CPAP) therapy is largely ineffective in correcting CSA or improving outcomes and is often poorly tolerated in these patients.⁴ Adaptive servo-ventilation (ASV) is a form of bilevel PAP (BPAP) therapy that delivers variable adjusting pressure support, primarily to treat CSA. PAP also may relieve upper airway obstructions, thereby effectively treating any comorbid obstructive component. ASV has been well documented to improve sleep-related disorders and improve apnea-hypopnea index (AHI) in patients with CSA. However, longitudinal data have demonstrated increased mortality in patients with heart failure with reduced ejection fraction (HFrEF) who were treated with ASV.⁵ Since the SERVE-HF trial results came to light in 2015, there has been no consensus regarding the optimal use, if any, of ASV therapy.^6-8 This is partly related to the inability to fully explain the study’s major findings, which were unexpected at the time, and partly due to the absence of similar relevant mortality data in patients with CSA but without HFrEF.

TECSA may present in some patients with OSA who are new to PAP therapy. These events are frequently seen during PAP titration sleep studies, though patients can also experience significant TECSA shortly after initiating home PAP therapy. TECSA is felt to result from a combination of stimulating pulmonary stretch receptors and lowering arterial carbon dioxide below the apneic threshold. Chemoreceptors located in the medulla respond by attenuating the respiratory drive.⁹ Previous studies have shown most cases of mild TECSA resolve over time with CPAP treatment. However, in patients with persistent or worsening TECSA, ASV may be considered as an alternative to CPAP.

The prevalence of OSA in the veteran population is estimated to be as high as 60%, considerably higher than the general population estimation.¹⁰ Patients with more significant comorbidities may also experience a higher frequency of central events. Patients with CSA have also been shown to have a higher risk for cardiac-related hospital admissions, providing plausible justification for correcting CSA.¹⁰

In the current study, we aim to characterize the group of patients using ASV therapy in the modern era. We will assess the objective efficacy and adherence of ASV therapy in patients with primarily CSA compared with those having primarily OSA (ie, TECSA). Secondarily, we aim to identify baseline clinical and polysomnographic features that may be predictive of ASV adherence, as a surrogate for subjective benefit.¹¹ In the wake of the SERVE-HF study, the sleep medicine community has paused prescribing ASV therapy for CSA. We hope to provide more perspective on the treatment of veterans with CSA and identify the patient groups that would benefit most from ASV therapy.

 

Methods

This retrospective chart review examined patients prescribed ASV therapy at the Hampton Veterans Affairs Medical Center (HVAMC) in Virginia who had therapy data between January 1, 2015, and April 30, 2020. The start date was chosen to approximate the phase-in of wireless PAP devices at HVAMC and to correspond with the release of preliminary results from the SERVE-HF trial.

Patients were initially identified through a query into commercial wireless PAP management databases and cross-referenced with HVAMC patients. Adherence and efficacy data were obtained from the most recent clinical PAP data, which allowed for the evaluation of patients who discontinued therapy for reasons other than intolerance. Clinical, demographic, and polysomnography (PSG) data were obtained from the electronic health record. One patient, identified through the database query but not found in the electronic health record, was excluded. In cases of missing PSG data, especially AHI or similar values, all attempts were made to calculate the data with other provided values. This study was determined to be exempt by the HVAMC Institutional Review Board (protocol #20-01).

Statistics

Statistical analyses were designed to compare clinical characteristics and adherence to therapy of those with primarily CSA on PSG and those with primarily OSA. Because it was not currently known how many patients would fit into each of these categories, we also planned secondary comparisons of the clinical and PSG characteristics of those patients who were adherent with therapy and those who were not. Adherence with ASV therapy was defined as device use for ≥ 4 hours for ≥ 70% of nights.

Comparisons between the means of 2 normally distributed groups were performed with an unpaired t test. Comparisons between 2 nonnormally distributed groups and groups of dates were done with the Mann-Whitney U test. The normality of a group distribution was determined using D’Agostino-Pearson omnibus normality test. Two groups of dichotomous variables were compared with the Fisher exact test. P value < .05 was considered statistically significant.

Results

Thirty-one patients were prescribed ASV therapy and had follow-up at HVAMC since 2015. All patients were male. The mean (SD) age was 67.2 (11.4) years, mean body mass index (BMI) was 34.0 (5.9), and the mean (SD) Epworth Sleepiness Scale (ESS) score was 10.9 (5.8). Patient comorbidities included 30 (97%) with hypertension, 17 (55%) with diabetes mellitus, 16 (52%) with coronary artery disease, and 11 (35%) with congestive heart failure. Three patients had no echocardiogram or other documentation of left ventricular ejection fraction (LVEF). One of these patients had voluntarily stopped using PAP therapy, another had been erroneously started on ASV (ordered for fixed BPAP), and the third had since been retitrated to CPAP. In the 28 patients with documented LVEF, the mean (SD) LVEF was 61.8% (6.9). Ten patients (32%) had opioids documented on their medication lists and 6 (19%) had benzodiazepines.

The median date of diagnostic sleep testing was January 9, 2015, and testing was completed after the release of the initial field safety notice regarding the SERVE-HF trial preliminary findings May 13, 2015, for 14 patients (45%).¹² On diagnostic sleep testing, the mean (SD) AHI was 47.3 (25.6) events/h and the median (IQR) oxygen saturation (SpO₂) nadir was 82% (78-84). Three patients (10%) were initially diagnosed with CSA, 19 (61%) with OSA, and 9 (29%) with both. Sixteen patients (52%) had ASV with fixed expiratory PAP (EPAP), and 15 (48%) had variable adjusting EPAP. Mean (SD) usage of ASV was 6.5 (2.6) hours and 66.0% (34.2) of nights for ≥ 4 hours. Mean (SD) titrated EPAP (set or 90th/95th percentile autotitrated) was 10.1 (3.4) cm H₂O and inspiratory PAP (IPAP) (90th/95th percentile) was 17.1 (3.3) cm H₂O. The median (IQR) residual AHI on ASV was 2.7 events/h (1.1-5.1), apnea index (AI) was 0.4 (0.1-1.0), and hypopnea index (HI) was 1.4 (1.0-3.2); the residual central and obstructive events were not available in most cases.

 

Adherence

Seventeen patients (55%) met the minimum adherence criteria of ≥ 4 hours of usage for ≥ 70% of the nights. There were no significant differences in age, BMI, sex, race, comorbidities, medications, or ESS when comparing patients who were adherent to ASC and those who were not (Table 1). The date of diagnostic sleep testing and sleep architecture, including sleep latency, total sleep time, sleep efficiency, wake after sleep onset, arousal index, and percentage of rapid eye movement stage sleep were similar between the adherent patients and nonadherent patients (Table 2).

The overall AHI mean (SD) on diagnostic PSG were also similar between the adherent group (52.3 [24.8] events/h) and nonadherent group (45.1 [27.0] events/h) (P = .47). The mean (SD) for obstructive AHI (obstructive apneas, mixed apneas, obstructive hypopneas, and undifferentiated hypopneas per hour of sleep) were higher in the adherent group as a percentage of the total AHI: 81.5% (27.9) in the adherent group vs 46.7% (38.4%) in the nonadherent group; P = .02) (Figure 1).

This difference was primarily driven by a difference in mean (SD) HI: 29.7 (16.5) in the adherent group vs. 15.3 (12.1) in the nonadherent group (P = .04).

There were no significant differences between the proportions of patients on ASV with set EPAP or the titrated EPAP and IPAP. The median (IQR) residual AHI was lower in the adherent group compared with the nonadherent group, both in absolute values (1.7 [0.9-3.2] events/h vs 4.7 [2.4-10.3] events/h, respectively [P = .004]), and as a percentage of the pretreatment AHI (3.1% [2.5-6.0] vs 10.2% [5.3-34.4], respectively; P = .002) (Figure 2).

Primarily Obstructive Sleep Apnea

Sleep apnea was a mixed picture of obstructive and central events in many patients. Only 3 patients had “pure” CSA. Thus, we were unable to define discrete comparison groups based on the sleep-disordered breathing phenotype. We identified 19 patients with primarily OSA (ie, initially diagnosed with OSA, OSA with TECSA, or complex sleep apnea). The mean (SD) age was 66.1 (12.8) years, BMI was 36.2 (4.7), and ESS was 11.4 (5.6). The mean (SD) baseline AHI was 46.9 (29.5), obstructive AHI was 40.5 (30.4), and central AHI was 0.4 (1.2); the median (IQR) SpO₂ nadir was 81% (78%-84%). The mean (SD) titrated EPAP was 10.2 (3.5) cm H₂O, and the 90th/95th percentile IPAP was 17.9 (3.5) cm H₂O. The mean (SD) usage of ASV was 7.9 (5.3) hours with 11 patients (58%) meeting the minimum standard for adherence to ASV therapy.

No significant differences were seen between the adherent and nonadherent groups in clinical or demographic characteristics or date of diagnostic sleep testing (eAppendix, available online at doi:10.12788/fp.0374). In baseline sleep studies the mean (SD) HI was 32.3 (15.8) in the adherent group compared with 14.7 (8.8) in the nonadherent group (P = .049). In contrast, obstructive AHI was not significantly lower in the adherent group: 51.9 (30.9) in the adherent group compared with 22.2 (20.6) in the nonadherent group (P = .09). The median (IQR) residual AHI on ASV as a percentage of the pretreatment AHI was 3.0% (2.4%-6.5%) in the adherent group compared with 11.3% (5.4%-89.1%) in the nonadherent group, a statistically significant difference (P = .01). No other significant differences were seen between the groups.

 

Discussion

This study describes a real-world cohort of patients using ASV therapy and the characteristics associated with benefit from therapy. The patients that were prescribed and started ASV therapy most often had a significant degree of obstructive component to sleep-disordered breathing, whether primary OSA with TECSA or comorbid OSA and CSA. Moreover, we found that a higher obstructive AHI on the baseline PSG was associated with adherence to ASV therapy. Another important finding was that a lower residual AHI on ASV as a proportion of the baseline was associated with PAP adherence. Adherent patients had similar clinical characteristics as the nonadherent patients, including comorbidities, severity of sleep-disordered breathing, and obesity.

Though the results of the SERVE-HF trial have dampened the enthusiasm somewhat, ASV therapy has long been considered an effective and well-tolerated treatment for many types of CSA.¹³ In fact, treatments that can eliminate the central AHI are fairly limited.^4,14 Our data suggest that ASV is also effective and tolerated in OSA with TECSA and/or comorbid CSA. Recent studies suggest that CSA resolves spontaneously in a majority of TECSA patients within 2 to 3 months of regular CPAP use.¹⁵ Other estimates suggest that persistent TESCA may be present in 2% of patients with OSA on treatment.¹⁶

Given the high and rising prevalence of OSA, many people are at risk for clinically significant TESCA. Another retrospective case series found that 72% of patients that failed treatment with CPAP or BPAP during PSG, met diagnostic criteria (at the time) for CSA; ASV was objectively beneficial in these patients.¹⁷ ASV can be an especially useful modality to treat OSA in patients with CSA that either prevents tolerance of standard therapies or causes clinical consequences, presuming the patient does not also have HFrEF.¹⁸ The long-term outcomes of treatment with ASV therapy remain a matter of debate.

The SERVE-HF trial remains among the only studies that have assessed the mortality effects of CSA treatments, with unfavorable findings. Treatment of OSA has been associated with favorable chronic health benefits, though recent studies have questioned the degree of benefit attributable to OSA treatment.^19-24 Similar studies have not been done for comorbidities represented by our study cohort (ie, OSA with TECSA and/or comorbid CSA).

The lack of CSA diagnosis alone in our cohort may be partially attributable to changing practice patterns following the SERVE-HF trial, though it is not clear from these data why a higher baseline obstructive AHI was associated with adherence to ASV therapy. Our data in this regard are somewhat at odds with the preliminary results of the ADVENT-HF trial. In that study, adherence to ASV therapy in patients with predominantly OSA declined significantly more than in patients with predominantly CSA.²⁵ Most of our patients were diagnosed with predominantly OSA, so a direct comparison with the CSA group is problematic; additionally, the primary brand and the pressure adjustments algorithm used in our study differed from the ADVENT-HF trial.

OSA and CSA may present with similar clinical symptoms, including sleep fragmentation, insomnia, and excessive daytime sleepiness; however, the degree of symptomatology, especially daytime sleepiness, and the response to treatment, may be less in CSA.^2,26 Both the subjective report of symptoms (ESS) and PSG measures of sleep fragmentation were similar in our patients, again likely explained by the predominance of obstructive events.

The pathophysiology of CSA is more varied than OSA, which is probably relevant in this case. ASV was originally designed for the management of CSA with CSR, accomplishing this goal by stabilizing the periods of central apnea and hyperpnea characteristic of CSR.²⁷ Although other forms of CSA demonstrate breathing patterns distinct from CSR, ASV has become an accepted treatment for most of these. It is plausible that the long-term subjective benefit and tolerance of ASV in CSA without CSR is less than for CSA with CSR or OSA. None of the patients in our study had CSA with CSR.

Ultimately, it may be the objective treatment effect that lends to adherence, as has been shown previously in OSA patients; our group of adherent patients showed a greater improvement in AHI, relative to baseline, than the nonadherent patients did.²⁸ The technology behind ASV therapy can greatly reduce the frequencies of central apneas, yet this same treatment effectively splints the upper airway and even more effectively eliminates obstructive apneas and hypopneas. Variable adjusting EPAP devices would plausibly provide even more benefit in these patients, as has been shown in prior studies.²⁹ To the contrary, our small sample of patients with TESCA showed a nonsignificant trend toward adherence with fixed EPAP ASV.

Opioid use was substantial in our population, without significant differences between the groups. CPAP therapy is ineffective in improving opioid-associated CSA. In a recent study, 20 patients on opioid therapy with CSA were treated with CPAP therapy; after several weeks, the average therapeutic use was 4 to 5 hours per night and CPAP was abandoned in favor of ASV therapy due to persistent central apnea. ASV treatment was associated with a considerable reduction in central apnea index, AHI, arousal index, and oxygen desaturations in a remarkable improvement over CPAP.³⁰

 

Limitations and Future Directions

This retrospective, single-center study may have limited applicability to other populations. Adherence was used as a surrogate for subjective benefit from treatment, though benefit was not confirmed by the patients directly. Only patients seen in follow-up for documentation of the ASV download were identified for inclusion and data analysis. As a single center, we risk homogeneity in the treatment algorithms, though sleep medicine treatments are often decided at the time of the sleep studies. Studies and treatment recommendations were made at a variety of sites, including our sleep center, other US Department of Veterans Affairs hospitals, in the community network, and at US Department of Defense centers. Our population was homogenous in some ways; notably, 100% of our group was male, which is substantially higher than both the veteran population and the general population. Risk factors for OSA and CSA are more common in male patients, which may partially explain this anomaly. Lastly, with our small sample size, there is increased risk that the results seen occurred by chance.

There are several areas for further study. A larger multicenter study may permit these results to be generalized to the population and should include subjective measures of benefit. Patients with primarily CSA were largely absent in our group and may be the focus of future studies; data on predictors of treatment adherence in CSA are lacking. With the availability of consistent older adherence data, comparisons may be made between the efficacies of clinical practice habits, including treatment efficacy, before and after the results of the SERVE-HF trial became known.

Conclusions

In selected patients with preserved LVEF, ASV therapy appears especially effective in patients with OSA combined with CSA. Adherence to ASV treatment was associated with higher obstructive AHI during the baseline PSG and with a greater reduction in the AHI. This understanding may help guide sleep specialists in personalizing treatments for sleep-disordered breathing. Because objective efficacy appears to be important for therapy adherence, clinicians should be able to consistently determine the obstructive and central components of the residual AHI, thus taking all information into account when optimizing the treatment. Additionally, both OSA and CSA pressure requirements should be considered when developing ASV devices.

Acknowledgments

We thank Martha Harper, RRT, of Hampton Veterans Affairs Medical Center (HVAMC) for helping to identify our patients and assisting with data collection. This material is the result of work supported with resources and the use of HVAMC facilities.

Sleep apnea is a heterogeneous group of conditions that may be attributable to a wide array of underlying conditions, with varying contributions of obstructive or central sleep-disordered breathing. The spectrum from obstructive sleep apnea (OSA) to central sleep apnea (CSA) includes mixed sleep apnea, treatment-emergent CSA (TECSA), and Cheyne-Stokes respiration (CSR).¹ The pathophysiologic causes of CSA can be attributed to delayed cardiopulmonary circulation in heart failure, decreased brainstem ventilatory response due to stroke, blunting of central chemoreceptors in chronic opioid use, and/or stimulation of the Hering-Breuer reflex from activation of pulmonary stretch receptors after initiating positive airway pressure (PAP) for treatment of OSA.^2,3 Medications are commonly implicated in many forms of sleep-disordered breathing; importantly, opioids and benzodiazepines may blunt the respiratory drive, leading to CSA, and/or impair upper airway patency, resulting in or worsening OSA.

Continuous positive airway pressure (CPAP) therapy is largely ineffective in correcting CSA or improving outcomes and is often poorly tolerated in these patients.⁴ Adaptive servo-ventilation (ASV) is a form of bilevel PAP (BPAP) therapy that delivers variable adjusting pressure support, primarily to treat CSA. PAP also may relieve upper airway obstructions, thereby effectively treating any comorbid obstructive component. ASV has been well documented to improve sleep-related disorders and improve apnea-hypopnea index (AHI) in patients with CSA. However, longitudinal data have demonstrated increased mortality in patients with heart failure with reduced ejection fraction (HFrEF) who were treated with ASV.⁵ Since the SERVE-HF trial results came to light in 2015, there has been no consensus regarding the optimal use, if any, of ASV therapy.^6-8 This is partly related to the inability to fully explain the study’s major findings, which were unexpected at the time, and partly due to the absence of similar relevant mortality data in patients with CSA but without HFrEF.

TECSA may present in some patients with OSA who are new to PAP therapy. These events are frequently seen during PAP titration sleep studies, though patients can also experience significant TECSA shortly after initiating home PAP therapy. TECSA is felt to result from a combination of stimulating pulmonary stretch receptors and lowering arterial carbon dioxide below the apneic threshold. Chemoreceptors located in the medulla respond by attenuating the respiratory drive.⁹ Previous studies have shown most cases of mild TECSA resolve over time with CPAP treatment. However, in patients with persistent or worsening TECSA, ASV may be considered as an alternative to CPAP.

The prevalence of OSA in the veteran population is estimated to be as high as 60%, considerably higher than the general population estimation.¹⁰ Patients with more significant comorbidities may also experience a higher frequency of central events. Patients with CSA have also been shown to have a higher risk for cardiac-related hospital admissions, providing plausible justification for correcting CSA.¹⁰

In the current study, we aim to characterize the group of patients using ASV therapy in the modern era. We will assess the objective efficacy and adherence of ASV therapy in patients with primarily CSA compared with those having primarily OSA (ie, TECSA). Secondarily, we aim to identify baseline clinical and polysomnographic features that may be predictive of ASV adherence, as a surrogate for subjective benefit.¹¹ In the wake of the SERVE-HF study, the sleep medicine community has paused prescribing ASV therapy for CSA. We hope to provide more perspective on the treatment of veterans with CSA and identify the patient groups that would benefit most from ASV therapy.

 

Methods

This retrospective chart review examined patients prescribed ASV therapy at the Hampton Veterans Affairs Medical Center (HVAMC) in Virginia who had therapy data between January 1, 2015, and April 30, 2020. The start date was chosen to approximate the phase-in of wireless PAP devices at HVAMC and to correspond with the release of preliminary results from the SERVE-HF trial.

Patients were initially identified through a query into commercial wireless PAP management databases and cross-referenced with HVAMC patients. Adherence and efficacy data were obtained from the most recent clinical PAP data, which allowed for the evaluation of patients who discontinued therapy for reasons other than intolerance. Clinical, demographic, and polysomnography (PSG) data were obtained from the electronic health record. One patient, identified through the database query but not found in the electronic health record, was excluded. In cases of missing PSG data, especially AHI or similar values, all attempts were made to calculate the data with other provided values. This study was determined to be exempt by the HVAMC Institutional Review Board (protocol #20-01).

Statistics

Statistical analyses were designed to compare clinical characteristics and adherence to therapy of those with primarily CSA on PSG and those with primarily OSA. Because it was not currently known how many patients would fit into each of these categories, we also planned secondary comparisons of the clinical and PSG characteristics of those patients who were adherent with therapy and those who were not. Adherence with ASV therapy was defined as device use for ≥ 4 hours for ≥ 70% of nights.

Comparisons between the means of 2 normally distributed groups were performed with an unpaired t test. Comparisons between 2 nonnormally distributed groups and groups of dates were done with the Mann-Whitney U test. The normality of a group distribution was determined using D’Agostino-Pearson omnibus normality test. Two groups of dichotomous variables were compared with the Fisher exact test. P value < .05 was considered statistically significant.

Results

Thirty-one patients were prescribed ASV therapy and had follow-up at HVAMC since 2015. All patients were male. The mean (SD) age was 67.2 (11.4) years, mean body mass index (BMI) was 34.0 (5.9), and the mean (SD) Epworth Sleepiness Scale (ESS) score was 10.9 (5.8). Patient comorbidities included 30 (97%) with hypertension, 17 (55%) with diabetes mellitus, 16 (52%) with coronary artery disease, and 11 (35%) with congestive heart failure. Three patients had no echocardiogram or other documentation of left ventricular ejection fraction (LVEF). One of these patients had voluntarily stopped using PAP therapy, another had been erroneously started on ASV (ordered for fixed BPAP), and the third had since been retitrated to CPAP. In the 28 patients with documented LVEF, the mean (SD) LVEF was 61.8% (6.9). Ten patients (32%) had opioids documented on their medication lists and 6 (19%) had benzodiazepines.

The median date of diagnostic sleep testing was January 9, 2015, and testing was completed after the release of the initial field safety notice regarding the SERVE-HF trial preliminary findings May 13, 2015, for 14 patients (45%).¹² On diagnostic sleep testing, the mean (SD) AHI was 47.3 (25.6) events/h and the median (IQR) oxygen saturation (SpO₂) nadir was 82% (78-84). Three patients (10%) were initially diagnosed with CSA, 19 (61%) with OSA, and 9 (29%) with both. Sixteen patients (52%) had ASV with fixed expiratory PAP (EPAP), and 15 (48%) had variable adjusting EPAP. Mean (SD) usage of ASV was 6.5 (2.6) hours and 66.0% (34.2) of nights for ≥ 4 hours. Mean (SD) titrated EPAP (set or 90th/95th percentile autotitrated) was 10.1 (3.4) cm H₂O and inspiratory PAP (IPAP) (90th/95th percentile) was 17.1 (3.3) cm H₂O. The median (IQR) residual AHI on ASV was 2.7 events/h (1.1-5.1), apnea index (AI) was 0.4 (0.1-1.0), and hypopnea index (HI) was 1.4 (1.0-3.2); the residual central and obstructive events were not available in most cases.

 

Adherence

Seventeen patients (55%) met the minimum adherence criteria of ≥ 4 hours of usage for ≥ 70% of the nights. There were no significant differences in age, BMI, sex, race, comorbidities, medications, or ESS when comparing patients who were adherent to ASC and those who were not (Table 1). The date of diagnostic sleep testing and sleep architecture, including sleep latency, total sleep time, sleep efficiency, wake after sleep onset, arousal index, and percentage of rapid eye movement stage sleep were similar between the adherent patients and nonadherent patients (Table 2).

The overall AHI mean (SD) on diagnostic PSG were also similar between the adherent group (52.3 [24.8] events/h) and nonadherent group (45.1 [27.0] events/h) (P = .47). The mean (SD) for obstructive AHI (obstructive apneas, mixed apneas, obstructive hypopneas, and undifferentiated hypopneas per hour of sleep) were higher in the adherent group as a percentage of the total AHI: 81.5% (27.9) in the adherent group vs 46.7% (38.4%) in the nonadherent group; P = .02) (Figure 1).

This difference was primarily driven by a difference in mean (SD) HI: 29.7 (16.5) in the adherent group vs. 15.3 (12.1) in the nonadherent group (P = .04).

There were no significant differences between the proportions of patients on ASV with set EPAP or the titrated EPAP and IPAP. The median (IQR) residual AHI was lower in the adherent group compared with the nonadherent group, both in absolute values (1.7 [0.9-3.2] events/h vs 4.7 [2.4-10.3] events/h, respectively [P = .004]), and as a percentage of the pretreatment AHI (3.1% [2.5-6.0] vs 10.2% [5.3-34.4], respectively; P = .002) (Figure 2).

Primarily Obstructive Sleep Apnea

Sleep apnea was a mixed picture of obstructive and central events in many patients. Only 3 patients had “pure” CSA. Thus, we were unable to define discrete comparison groups based on the sleep-disordered breathing phenotype. We identified 19 patients with primarily OSA (ie, initially diagnosed with OSA, OSA with TECSA, or complex sleep apnea). The mean (SD) age was 66.1 (12.8) years, BMI was 36.2 (4.7), and ESS was 11.4 (5.6). The mean (SD) baseline AHI was 46.9 (29.5), obstructive AHI was 40.5 (30.4), and central AHI was 0.4 (1.2); the median (IQR) SpO₂ nadir was 81% (78%-84%). The mean (SD) titrated EPAP was 10.2 (3.5) cm H₂O, and the 90th/95th percentile IPAP was 17.9 (3.5) cm H₂O. The mean (SD) usage of ASV was 7.9 (5.3) hours with 11 patients (58%) meeting the minimum standard for adherence to ASV therapy.

No significant differences were seen between the adherent and nonadherent groups in clinical or demographic characteristics or date of diagnostic sleep testing (eAppendix, available online at doi:10.12788/fp.0374). In baseline sleep studies the mean (SD) HI was 32.3 (15.8) in the adherent group compared with 14.7 (8.8) in the nonadherent group (P = .049). In contrast, obstructive AHI was not significantly lower in the adherent group: 51.9 (30.9) in the adherent group compared with 22.2 (20.6) in the nonadherent group (P = .09). The median (IQR) residual AHI on ASV as a percentage of the pretreatment AHI was 3.0% (2.4%-6.5%) in the adherent group compared with 11.3% (5.4%-89.1%) in the nonadherent group, a statistically significant difference (P = .01). No other significant differences were seen between the groups.

 

Discussion

This study describes a real-world cohort of patients using ASV therapy and the characteristics associated with benefit from therapy. The patients that were prescribed and started ASV therapy most often had a significant degree of obstructive component to sleep-disordered breathing, whether primary OSA with TECSA or comorbid OSA and CSA. Moreover, we found that a higher obstructive AHI on the baseline PSG was associated with adherence to ASV therapy. Another important finding was that a lower residual AHI on ASV as a proportion of the baseline was associated with PAP adherence. Adherent patients had similar clinical characteristics as the nonadherent patients, including comorbidities, severity of sleep-disordered breathing, and obesity.

Though the results of the SERVE-HF trial have dampened the enthusiasm somewhat, ASV therapy has long been considered an effective and well-tolerated treatment for many types of CSA.¹³ In fact, treatments that can eliminate the central AHI are fairly limited.^4,14 Our data suggest that ASV is also effective and tolerated in OSA with TECSA and/or comorbid CSA. Recent studies suggest that CSA resolves spontaneously in a majority of TECSA patients within 2 to 3 months of regular CPAP use.¹⁵ Other estimates suggest that persistent TESCA may be present in 2% of patients with OSA on treatment.¹⁶

Given the high and rising prevalence of OSA, many people are at risk for clinically significant TESCA. Another retrospective case series found that 72% of patients that failed treatment with CPAP or BPAP during PSG, met diagnostic criteria (at the time) for CSA; ASV was objectively beneficial in these patients.¹⁷ ASV can be an especially useful modality to treat OSA in patients with CSA that either prevents tolerance of standard therapies or causes clinical consequences, presuming the patient does not also have HFrEF.¹⁸ The long-term outcomes of treatment with ASV therapy remain a matter of debate.

The SERVE-HF trial remains among the only studies that have assessed the mortality effects of CSA treatments, with unfavorable findings. Treatment of OSA has been associated with favorable chronic health benefits, though recent studies have questioned the degree of benefit attributable to OSA treatment.^19-24 Similar studies have not been done for comorbidities represented by our study cohort (ie, OSA with TECSA and/or comorbid CSA).

The lack of CSA diagnosis alone in our cohort may be partially attributable to changing practice patterns following the SERVE-HF trial, though it is not clear from these data why a higher baseline obstructive AHI was associated with adherence to ASV therapy. Our data in this regard are somewhat at odds with the preliminary results of the ADVENT-HF trial. In that study, adherence to ASV therapy in patients with predominantly OSA declined significantly more than in patients with predominantly CSA.²⁵ Most of our patients were diagnosed with predominantly OSA, so a direct comparison with the CSA group is problematic; additionally, the primary brand and the pressure adjustments algorithm used in our study differed from the ADVENT-HF trial.

OSA and CSA may present with similar clinical symptoms, including sleep fragmentation, insomnia, and excessive daytime sleepiness; however, the degree of symptomatology, especially daytime sleepiness, and the response to treatment, may be less in CSA.^2,26 Both the subjective report of symptoms (ESS) and PSG measures of sleep fragmentation were similar in our patients, again likely explained by the predominance of obstructive events.

The pathophysiology of CSA is more varied than OSA, which is probably relevant in this case. ASV was originally designed for the management of CSA with CSR, accomplishing this goal by stabilizing the periods of central apnea and hyperpnea characteristic of CSR.²⁷ Although other forms of CSA demonstrate breathing patterns distinct from CSR, ASV has become an accepted treatment for most of these. It is plausible that the long-term subjective benefit and tolerance of ASV in CSA without CSR is less than for CSA with CSR or OSA. None of the patients in our study had CSA with CSR.

Ultimately, it may be the objective treatment effect that lends to adherence, as has been shown previously in OSA patients; our group of adherent patients showed a greater improvement in AHI, relative to baseline, than the nonadherent patients did.²⁸ The technology behind ASV therapy can greatly reduce the frequencies of central apneas, yet this same treatment effectively splints the upper airway and even more effectively eliminates obstructive apneas and hypopneas. Variable adjusting EPAP devices would plausibly provide even more benefit in these patients, as has been shown in prior studies.²⁹ To the contrary, our small sample of patients with TESCA showed a nonsignificant trend toward adherence with fixed EPAP ASV.

Opioid use was substantial in our population, without significant differences between the groups. CPAP therapy is ineffective in improving opioid-associated CSA. In a recent study, 20 patients on opioid therapy with CSA were treated with CPAP therapy; after several weeks, the average therapeutic use was 4 to 5 hours per night and CPAP was abandoned in favor of ASV therapy due to persistent central apnea. ASV treatment was associated with a considerable reduction in central apnea index, AHI, arousal index, and oxygen desaturations in a remarkable improvement over CPAP.³⁰

 

Limitations and Future Directions

This retrospective, single-center study may have limited applicability to other populations. Adherence was used as a surrogate for subjective benefit from treatment, though benefit was not confirmed by the patients directly. Only patients seen in follow-up for documentation of the ASV download were identified for inclusion and data analysis. As a single center, we risk homogeneity in the treatment algorithms, though sleep medicine treatments are often decided at the time of the sleep studies. Studies and treatment recommendations were made at a variety of sites, including our sleep center, other US Department of Veterans Affairs hospitals, in the community network, and at US Department of Defense centers. Our population was homogenous in some ways; notably, 100% of our group was male, which is substantially higher than both the veteran population and the general population. Risk factors for OSA and CSA are more common in male patients, which may partially explain this anomaly. Lastly, with our small sample size, there is increased risk that the results seen occurred by chance.

There are several areas for further study. A larger multicenter study may permit these results to be generalized to the population and should include subjective measures of benefit. Patients with primarily CSA were largely absent in our group and may be the focus of future studies; data on predictors of treatment adherence in CSA are lacking. With the availability of consistent older adherence data, comparisons may be made between the efficacies of clinical practice habits, including treatment efficacy, before and after the results of the SERVE-HF trial became known.

Conclusions

In selected patients with preserved LVEF, ASV therapy appears especially effective in patients with OSA combined with CSA. Adherence to ASV treatment was associated with higher obstructive AHI during the baseline PSG and with a greater reduction in the AHI. This understanding may help guide sleep specialists in personalizing treatments for sleep-disordered breathing. Because objective efficacy appears to be important for therapy adherence, clinicians should be able to consistently determine the obstructive and central components of the residual AHI, thus taking all information into account when optimizing the treatment. Additionally, both OSA and CSA pressure requirements should be considered when developing ASV devices.

Acknowledgments

We thank Martha Harper, RRT, of Hampton Veterans Affairs Medical Center (HVAMC) for helping to identify our patients and assisting with data collection. This material is the result of work supported with resources and the use of HVAMC facilities.

The inappropriate use of antibiotics is associated with an increased risk of antibiotic resistance, health care costs, and risk of adverse drug reactions.¹ According to the Centers for Disease Control and Prevention (CDC), a 10% decrease in overall antibiotic use across different wards was associated with a 34% decrease in Clostridioides difficile (C difficile) infections.² In addition, antimicrobial resistance accounts for > 2.8 million infections and > 35,000 deaths each year.³ The estimated total economic costs of antibiotic resistance to the US economy have ranged as high as $20 billion in excess direct health care costs.⁴ A main goal of an antimicrobial stewardship program (ASP) is to optimize antibiotic use to prevent the adverse consequences of inappropriate antibiotic prescribing.

During the COVID-19 pandemic, increased use of empiric antibiotic therapy has been observed. According to the CDC, almost 80% of patients hospitalized with COVID-19 received an antibiotic from March 2020 to October 2020.⁵ Studies were conducted to investigate the prevalence of bacterial coinfection in patients with COVID-19 and whether antibiotics were indicated in this patient population. A United Kingdom multicenter, prospective cohort study showed a high proportion of patients hospitalized with COVID-19 received antimicrobials despite microbiologically confirmed bacterial infections being rare and more likely to be secondary infections.⁶

Many other studies have reported similar findings. Langord and colleagues found the prevalence of bacterial coinfection in patients with COVID-19 was 3.5% but that 71.9% received antibiotics.⁷ Coenen and colleagues identified 12.4% of the patients with possible and 1.1% of patients with probable bacterial coinfection, while 81% of the study population and 78% of patients were classified as unlikely bacterial coinfection received antibiotics.⁸

At Veterans Affairs Southern Nevada Healthcare System (VASNHS), an ASP team consisting of an infectious disease (ID) physician and 2 pharmacists provide daily prospective audits with intervention and feedback along with other interventions, such as providing restricted order menus, institutional treatment guidelines, and staff education to help improve antibiotic prescribing. The ASP pharmacists have a scope of practice to make changes to anti-infective therapies. The purpose of this study was to describe antibiotic prescribing in patients hospitalized with COVID-19 from November 1, 2020, to January 31, 2021, in an ASP setting led by pharmacists.

 

Methods

This retrospective descriptive study included patients who were hospitalized for the treatment of laboratory-confirmed COVID-19 infection. The Theradoc clinical surveillance system was used to retrieve a list of patients who were admitted to VASNHS from November 1, 2020, to January 31, 2021, and tested positive for COVID-19. Patients with incidental positive COVID-19 test results or those who received antibiotics for extrapulmonary indications on hospital admission were excluded.

Each patient chart was reviewed and data, including clinical presentations, procalcitonin (PCT), the requirement of supplemental oxygen, vital signs, imaging findings, antibiotic orders on admission, ASP interventions such as discontinuation or changes to antibiotic therapy during the first 72 hours of hospital admission, clinical outcomes, culture results, and readmission rate, defined as any hospital admission related to COVID-19 or respiratory tract infection within 30 days from the previous discharge, were collected.

The primary objective of the study was to describe antibiotic prescribing in patients hospitalized with COVID-19. The secondary outcomes included the prevalence of bacterial coinfection and nosocomial bacterial infection in patients hospitalized with COVID-19.

Results

A total of 199 patients were admitted to the hospital for laboratory-confirmed COVID-19 infection from November 1, 2020, to January 31, 2021. Sixty-one patients (31%) received at least 1 antibiotic on hospital admission. Among those patients who received empiric antibiotic treatment, 29 patients (48%) met the Systemic Inflammatory Response Syndrome (SIRS) criteria. Fifty-six patients (92%) had ≥ 1 PCT level obtained, and 26 of those (46%) presented with elevated PCT levels (PCT > 0.25). Fifty patients (82%) required oxygen supplement and 49 (80%) presented with remarkable imaging findings. Of 138 patients who did not receive empiric antibiotic therapy within 72 hours of hospital admission, 56 (41%) met the SIRS criteria, 31 (29%) had elevated PCT levels, 100 (72%) required oxygen supplement, and 79 (59%) presented with remarkable imaging findings.

Antibiotic Prescribing

Forty-six of 61 patients (75%) received antibiotic treatment for community-acquired pneumonia (CAP) that included ceftriaxone and azithromycin. Three patients (5%) received ≥ 1 broad-spectrum antibiotic (4th generation cephalosporin [cefepime] or piperacillin-tazobactam), 2 (3%) received vancomycin, and 1 (2%) received a fluoroquinolone (levofloxacin) on admission.

Among 61 patients who received empiric antibiotics, the readmission rate was 6%. The mortality rate was 20%, and the mean (SD) duration of hospital stay was 13.1 (12.5) days. There were 47 patients (77%) who had antibiotics discontinued by the ASP within 72 hours of admission. Among these patients, the readmission rate was 8% and the mortality rate was 15%. The mean (SD) duration of hospital stay was 12.7 (13.3) days. In the group of patients without empiric antibiotic therapy, the readmission rate was 6%, the mortality rate was 10%, and the mean (SD) duration of hospital stay was 10.1 (9.5) days.

Six of 199 patients (3%) had microbiologically confirmed bacterial coinfection on hospital admission: 3 were Pseudomonas aeruginosa (P aeruginosa) and 2 were Klebsiella oxytoca (Table 1).

A total of 16 patients (8%) developed a nosocomial infection, and 10 had suspected hospital-acquired or ventilator-acquired pneumonia (Table 2).

Eight patients had positive respiratory cultures, and P aeruginosa was the most common. Five patients had bacteremia, including 1 vancomycin-resistant Enterococcus, 1 methicillin-resistant Staphylococcus aureus, Enterococcus spp, 1 culture grew both Escherichia coli and methicillin-susceptible Staphylococcus aureus (MSSA), and 1 culture grew MSSA and Streptococcus salivarius. Five patients had candidemia during their hospital stay: 3 Candida albicans, 1 Candida tropicalis, and 1 Candida glabrata isolate (eAppendix, available online at doi:10.12788/fp.0380).

 

Discussion

Prospective audit and feedback and preauthorization are recommended in guidelines as “core components of any stewardship program.”⁹ At VASNHS, the ASP performs daily prospective audits with intervention and feedback. Efforts have been made to maintain daily ASP activities during the pandemic. This study aimed to describe antibiotic prescribing for patients hospitalized with COVID-19 in a pharmacist-led ASP setting. It was found that up to 31% of the patients received ≥ 1 antibiotic on admission for empiric treatment of bacterial coinfection. About half of these patients met the SIRS criteria. Most of these patients received ceftriaxone and azithromycin for concern of CAP. ASP discontinued antibiotics within 72 hours in most of the patients. Chart review and discussion with ID physicians and/or hospitalists determined the probability of bacterial coinfection as well as any potential complication or patient-specific risk factor. It is important to note that most patients who received antibiotics on admission had ≥ 1 PCT level and up to 46% of them had a PCT level > 0.25. However, according to Relph and colleagues, PCT may not be a reliable indicator of bacterial infection in severe viral diseases with raised interleukin-6 levels.¹⁰ An elevated PCT level should not be the sole indicator for empiric antibiotic treatment.

Study findings confirmed the low prevalence of bacterial coinfection in patients hospitalized with COVID-19. The overuse of empiric antibiotics in a patient population unlikely to present with bacterial coinfection is concerning. It is essential to continue promoting antimicrobial stewardship during the COVID-19 pandemic to ensure appropriate and responsible antimicrobial prescribing. A thorough clinical assessment consisting of comorbidities, clinical symptoms, radiologic and microbiologic findings, as well as other relevant workup or biomarker results is crucial to determine whether the antibiotic is strongly indicated in patients hospitalized with COVID-19. Empiric antibiotic therapy should be considered only in patients with clinical findings suggestive of bacterial coinfection.

Limitations

Limitations of our study included the study design (single-center, retrospective review, lack of comparative group) and small sample size with a 3-month study period. In addition, respiratory cultures are not commonly obtained in patients who present with mild-to-moderate CAP. Using culture results solely to confirm bacterial coinfection in patients with COVID-19 could have underestimated the prevalence of bacterial infection. Developing diagnostic criteria that include clinical signs and symptoms, imaging findings, and laboratory results as well as culture results would help to better assess the presence of bacterial coinfection in this patient population.

Conclusions

The study findings showed that up to 30% of patients hospitalized for COVID-19 infection received empiric antibiotic treatment for concern of bacterial coinfection. A pharmacist-led ASP provided interventions, including early discontinuation of antibiotics in 77% of these patients.

A low prevalence of bacterial coinfection (3%) in patients hospitalized with COVID-19 also was reported. A thorough clinical workup to determine the risk of bacterial coinfection in patients with COVID-19 is important before starting empiric antibiotic therapy. Continuing to promote the ASP during the COVID-19 pandemic to ensure responsible antibiotic use and prevent antimicrobial resistance is essential.

The inappropriate use of antibiotics is associated with an increased risk of antibiotic resistance, health care costs, and risk of adverse drug reactions.¹ According to the Centers for Disease Control and Prevention (CDC), a 10% decrease in overall antibiotic use across different wards was associated with a 34% decrease in Clostridioides difficile (C difficile) infections.² In addition, antimicrobial resistance accounts for > 2.8 million infections and > 35,000 deaths each year.³ The estimated total economic costs of antibiotic resistance to the US economy have ranged as high as $20 billion in excess direct health care costs.⁴ A main goal of an antimicrobial stewardship program (ASP) is to optimize antibiotic use to prevent the adverse consequences of inappropriate antibiotic prescribing.

During the COVID-19 pandemic, increased use of empiric antibiotic therapy has been observed. According to the CDC, almost 80% of patients hospitalized with COVID-19 received an antibiotic from March 2020 to October 2020.⁵ Studies were conducted to investigate the prevalence of bacterial coinfection in patients with COVID-19 and whether antibiotics were indicated in this patient population. A United Kingdom multicenter, prospective cohort study showed a high proportion of patients hospitalized with COVID-19 received antimicrobials despite microbiologically confirmed bacterial infections being rare and more likely to be secondary infections.⁶

Many other studies have reported similar findings. Langord and colleagues found the prevalence of bacterial coinfection in patients with COVID-19 was 3.5% but that 71.9% received antibiotics.⁷ Coenen and colleagues identified 12.4% of the patients with possible and 1.1% of patients with probable bacterial coinfection, while 81% of the study population and 78% of patients were classified as unlikely bacterial coinfection received antibiotics.⁸

At Veterans Affairs Southern Nevada Healthcare System (VASNHS), an ASP team consisting of an infectious disease (ID) physician and 2 pharmacists provide daily prospective audits with intervention and feedback along with other interventions, such as providing restricted order menus, institutional treatment guidelines, and staff education to help improve antibiotic prescribing. The ASP pharmacists have a scope of practice to make changes to anti-infective therapies. The purpose of this study was to describe antibiotic prescribing in patients hospitalized with COVID-19 from November 1, 2020, to January 31, 2021, in an ASP setting led by pharmacists.

 

Methods

This retrospective descriptive study included patients who were hospitalized for the treatment of laboratory-confirmed COVID-19 infection. The Theradoc clinical surveillance system was used to retrieve a list of patients who were admitted to VASNHS from November 1, 2020, to January 31, 2021, and tested positive for COVID-19. Patients with incidental positive COVID-19 test results or those who received antibiotics for extrapulmonary indications on hospital admission were excluded.

Each patient chart was reviewed and data, including clinical presentations, procalcitonin (PCT), the requirement of supplemental oxygen, vital signs, imaging findings, antibiotic orders on admission, ASP interventions such as discontinuation or changes to antibiotic therapy during the first 72 hours of hospital admission, clinical outcomes, culture results, and readmission rate, defined as any hospital admission related to COVID-19 or respiratory tract infection within 30 days from the previous discharge, were collected.

The primary objective of the study was to describe antibiotic prescribing in patients hospitalized with COVID-19. The secondary outcomes included the prevalence of bacterial coinfection and nosocomial bacterial infection in patients hospitalized with COVID-19.

Results

A total of 199 patients were admitted to the hospital for laboratory-confirmed COVID-19 infection from November 1, 2020, to January 31, 2021. Sixty-one patients (31%) received at least 1 antibiotic on hospital admission. Among those patients who received empiric antibiotic treatment, 29 patients (48%) met the Systemic Inflammatory Response Syndrome (SIRS) criteria. Fifty-six patients (92%) had ≥ 1 PCT level obtained, and 26 of those (46%) presented with elevated PCT levels (PCT > 0.25). Fifty patients (82%) required oxygen supplement and 49 (80%) presented with remarkable imaging findings. Of 138 patients who did not receive empiric antibiotic therapy within 72 hours of hospital admission, 56 (41%) met the SIRS criteria, 31 (29%) had elevated PCT levels, 100 (72%) required oxygen supplement, and 79 (59%) presented with remarkable imaging findings.

Antibiotic Prescribing

Forty-six of 61 patients (75%) received antibiotic treatment for community-acquired pneumonia (CAP) that included ceftriaxone and azithromycin. Three patients (5%) received ≥ 1 broad-spectrum antibiotic (4th generation cephalosporin [cefepime] or piperacillin-tazobactam), 2 (3%) received vancomycin, and 1 (2%) received a fluoroquinolone (levofloxacin) on admission.

Among 61 patients who received empiric antibiotics, the readmission rate was 6%. The mortality rate was 20%, and the mean (SD) duration of hospital stay was 13.1 (12.5) days. There were 47 patients (77%) who had antibiotics discontinued by the ASP within 72 hours of admission. Among these patients, the readmission rate was 8% and the mortality rate was 15%. The mean (SD) duration of hospital stay was 12.7 (13.3) days. In the group of patients without empiric antibiotic therapy, the readmission rate was 6%, the mortality rate was 10%, and the mean (SD) duration of hospital stay was 10.1 (9.5) days.

Six of 199 patients (3%) had microbiologically confirmed bacterial coinfection on hospital admission: 3 were Pseudomonas aeruginosa (P aeruginosa) and 2 were Klebsiella oxytoca (Table 1).

A total of 16 patients (8%) developed a nosocomial infection, and 10 had suspected hospital-acquired or ventilator-acquired pneumonia (Table 2).

Eight patients had positive respiratory cultures, and P aeruginosa was the most common. Five patients had bacteremia, including 1 vancomycin-resistant Enterococcus, 1 methicillin-resistant Staphylococcus aureus, Enterococcus spp, 1 culture grew both Escherichia coli and methicillin-susceptible Staphylococcus aureus (MSSA), and 1 culture grew MSSA and Streptococcus salivarius. Five patients had candidemia during their hospital stay: 3 Candida albicans, 1 Candida tropicalis, and 1 Candida glabrata isolate (eAppendix, available online at doi:10.12788/fp.0380).

 

Discussion

Prospective audit and feedback and preauthorization are recommended in guidelines as “core components of any stewardship program.”⁹ At VASNHS, the ASP performs daily prospective audits with intervention and feedback. Efforts have been made to maintain daily ASP activities during the pandemic. This study aimed to describe antibiotic prescribing for patients hospitalized with COVID-19 in a pharmacist-led ASP setting. It was found that up to 31% of the patients received ≥ 1 antibiotic on admission for empiric treatment of bacterial coinfection. About half of these patients met the SIRS criteria. Most of these patients received ceftriaxone and azithromycin for concern of CAP. ASP discontinued antibiotics within 72 hours in most of the patients. Chart review and discussion with ID physicians and/or hospitalists determined the probability of bacterial coinfection as well as any potential complication or patient-specific risk factor. It is important to note that most patients who received antibiotics on admission had ≥ 1 PCT level and up to 46% of them had a PCT level > 0.25. However, according to Relph and colleagues, PCT may not be a reliable indicator of bacterial infection in severe viral diseases with raised interleukin-6 levels.¹⁰ An elevated PCT level should not be the sole indicator for empiric antibiotic treatment.

Study findings confirmed the low prevalence of bacterial coinfection in patients hospitalized with COVID-19. The overuse of empiric antibiotics in a patient population unlikely to present with bacterial coinfection is concerning. It is essential to continue promoting antimicrobial stewardship during the COVID-19 pandemic to ensure appropriate and responsible antimicrobial prescribing. A thorough clinical assessment consisting of comorbidities, clinical symptoms, radiologic and microbiologic findings, as well as other relevant workup or biomarker results is crucial to determine whether the antibiotic is strongly indicated in patients hospitalized with COVID-19. Empiric antibiotic therapy should be considered only in patients with clinical findings suggestive of bacterial coinfection.

Limitations

Limitations of our study included the study design (single-center, retrospective review, lack of comparative group) and small sample size with a 3-month study period. In addition, respiratory cultures are not commonly obtained in patients who present with mild-to-moderate CAP. Using culture results solely to confirm bacterial coinfection in patients with COVID-19 could have underestimated the prevalence of bacterial infection. Developing diagnostic criteria that include clinical signs and symptoms, imaging findings, and laboratory results as well as culture results would help to better assess the presence of bacterial coinfection in this patient population.

Conclusions

The study findings showed that up to 30% of patients hospitalized for COVID-19 infection received empiric antibiotic treatment for concern of bacterial coinfection. A pharmacist-led ASP provided interventions, including early discontinuation of antibiotics in 77% of these patients.

A low prevalence of bacterial coinfection (3%) in patients hospitalized with COVID-19 also was reported. A thorough clinical workup to determine the risk of bacterial coinfection in patients with COVID-19 is important before starting empiric antibiotic therapy. Continuing to promote the ASP during the COVID-19 pandemic to ensure responsible antibiotic use and prevent antimicrobial resistance is essential.

The inappropriate use of antibiotics is associated with an increased risk of antibiotic resistance, health care costs, and risk of adverse drug reactions.¹ According to the Centers for Disease Control and Prevention (CDC), a 10% decrease in overall antibiotic use across different wards was associated with a 34% decrease in Clostridioides difficile (C difficile) infections.² In addition, antimicrobial resistance accounts for > 2.8 million infections and > 35,000 deaths each year.³ The estimated total economic costs of antibiotic resistance to the US economy have ranged as high as $20 billion in excess direct health care costs.⁴ A main goal of an antimicrobial stewardship program (ASP) is to optimize antibiotic use to prevent the adverse consequences of inappropriate antibiotic prescribing.

During the COVID-19 pandemic, increased use of empiric antibiotic therapy has been observed. According to the CDC, almost 80% of patients hospitalized with COVID-19 received an antibiotic from March 2020 to October 2020.⁵ Studies were conducted to investigate the prevalence of bacterial coinfection in patients with COVID-19 and whether antibiotics were indicated in this patient population. A United Kingdom multicenter, prospective cohort study showed a high proportion of patients hospitalized with COVID-19 received antimicrobials despite microbiologically confirmed bacterial infections being rare and more likely to be secondary infections.⁶

Many other studies have reported similar findings. Langord and colleagues found the prevalence of bacterial coinfection in patients with COVID-19 was 3.5% but that 71.9% received antibiotics.⁷ Coenen and colleagues identified 12.4% of the patients with possible and 1.1% of patients with probable bacterial coinfection, while 81% of the study population and 78% of patients were classified as unlikely bacterial coinfection received antibiotics.⁸

At Veterans Affairs Southern Nevada Healthcare System (VASNHS), an ASP team consisting of an infectious disease (ID) physician and 2 pharmacists provide daily prospective audits with intervention and feedback along with other interventions, such as providing restricted order menus, institutional treatment guidelines, and staff education to help improve antibiotic prescribing. The ASP pharmacists have a scope of practice to make changes to anti-infective therapies. The purpose of this study was to describe antibiotic prescribing in patients hospitalized with COVID-19 from November 1, 2020, to January 31, 2021, in an ASP setting led by pharmacists.

 

Methods

This retrospective descriptive study included patients who were hospitalized for the treatment of laboratory-confirmed COVID-19 infection. The Theradoc clinical surveillance system was used to retrieve a list of patients who were admitted to VASNHS from November 1, 2020, to January 31, 2021, and tested positive for COVID-19. Patients with incidental positive COVID-19 test results or those who received antibiotics for extrapulmonary indications on hospital admission were excluded.

Each patient chart was reviewed and data, including clinical presentations, procalcitonin (PCT), the requirement of supplemental oxygen, vital signs, imaging findings, antibiotic orders on admission, ASP interventions such as discontinuation or changes to antibiotic therapy during the first 72 hours of hospital admission, clinical outcomes, culture results, and readmission rate, defined as any hospital admission related to COVID-19 or respiratory tract infection within 30 days from the previous discharge, were collected.

The primary objective of the study was to describe antibiotic prescribing in patients hospitalized with COVID-19. The secondary outcomes included the prevalence of bacterial coinfection and nosocomial bacterial infection in patients hospitalized with COVID-19.

Results

A total of 199 patients were admitted to the hospital for laboratory-confirmed COVID-19 infection from November 1, 2020, to January 31, 2021. Sixty-one patients (31%) received at least 1 antibiotic on hospital admission. Among those patients who received empiric antibiotic treatment, 29 patients (48%) met the Systemic Inflammatory Response Syndrome (SIRS) criteria. Fifty-six patients (92%) had ≥ 1 PCT level obtained, and 26 of those (46%) presented with elevated PCT levels (PCT > 0.25). Fifty patients (82%) required oxygen supplement and 49 (80%) presented with remarkable imaging findings. Of 138 patients who did not receive empiric antibiotic therapy within 72 hours of hospital admission, 56 (41%) met the SIRS criteria, 31 (29%) had elevated PCT levels, 100 (72%) required oxygen supplement, and 79 (59%) presented with remarkable imaging findings.

Antibiotic Prescribing

Forty-six of 61 patients (75%) received antibiotic treatment for community-acquired pneumonia (CAP) that included ceftriaxone and azithromycin. Three patients (5%) received ≥ 1 broad-spectrum antibiotic (4th generation cephalosporin [cefepime] or piperacillin-tazobactam), 2 (3%) received vancomycin, and 1 (2%) received a fluoroquinolone (levofloxacin) on admission.

Among 61 patients who received empiric antibiotics, the readmission rate was 6%. The mortality rate was 20%, and the mean (SD) duration of hospital stay was 13.1 (12.5) days. There were 47 patients (77%) who had antibiotics discontinued by the ASP within 72 hours of admission. Among these patients, the readmission rate was 8% and the mortality rate was 15%. The mean (SD) duration of hospital stay was 12.7 (13.3) days. In the group of patients without empiric antibiotic therapy, the readmission rate was 6%, the mortality rate was 10%, and the mean (SD) duration of hospital stay was 10.1 (9.5) days.

Six of 199 patients (3%) had microbiologically confirmed bacterial coinfection on hospital admission: 3 were Pseudomonas aeruginosa (P aeruginosa) and 2 were Klebsiella oxytoca (Table 1).

A total of 16 patients (8%) developed a nosocomial infection, and 10 had suspected hospital-acquired or ventilator-acquired pneumonia (Table 2).

Eight patients had positive respiratory cultures, and P aeruginosa was the most common. Five patients had bacteremia, including 1 vancomycin-resistant Enterococcus, 1 methicillin-resistant Staphylococcus aureus, Enterococcus spp, 1 culture grew both Escherichia coli and methicillin-susceptible Staphylococcus aureus (MSSA), and 1 culture grew MSSA and Streptococcus salivarius. Five patients had candidemia during their hospital stay: 3 Candida albicans, 1 Candida tropicalis, and 1 Candida glabrata isolate (eAppendix, available online at doi:10.12788/fp.0380).

 

Discussion

Prospective audit and feedback and preauthorization are recommended in guidelines as “core components of any stewardship program.”⁹ At VASNHS, the ASP performs daily prospective audits with intervention and feedback. Efforts have been made to maintain daily ASP activities during the pandemic. This study aimed to describe antibiotic prescribing for patients hospitalized with COVID-19 in a pharmacist-led ASP setting. It was found that up to 31% of the patients received ≥ 1 antibiotic on admission for empiric treatment of bacterial coinfection. About half of these patients met the SIRS criteria. Most of these patients received ceftriaxone and azithromycin for concern of CAP. ASP discontinued antibiotics within 72 hours in most of the patients. Chart review and discussion with ID physicians and/or hospitalists determined the probability of bacterial coinfection as well as any potential complication or patient-specific risk factor. It is important to note that most patients who received antibiotics on admission had ≥ 1 PCT level and up to 46% of them had a PCT level > 0.25. However, according to Relph and colleagues, PCT may not be a reliable indicator of bacterial infection in severe viral diseases with raised interleukin-6 levels.¹⁰ An elevated PCT level should not be the sole indicator for empiric antibiotic treatment.

Study findings confirmed the low prevalence of bacterial coinfection in patients hospitalized with COVID-19. The overuse of empiric antibiotics in a patient population unlikely to present with bacterial coinfection is concerning. It is essential to continue promoting antimicrobial stewardship during the COVID-19 pandemic to ensure appropriate and responsible antimicrobial prescribing. A thorough clinical assessment consisting of comorbidities, clinical symptoms, radiologic and microbiologic findings, as well as other relevant workup or biomarker results is crucial to determine whether the antibiotic is strongly indicated in patients hospitalized with COVID-19. Empiric antibiotic therapy should be considered only in patients with clinical findings suggestive of bacterial coinfection.

Limitations

Limitations of our study included the study design (single-center, retrospective review, lack of comparative group) and small sample size with a 3-month study period. In addition, respiratory cultures are not commonly obtained in patients who present with mild-to-moderate CAP. Using culture results solely to confirm bacterial coinfection in patients with COVID-19 could have underestimated the prevalence of bacterial infection. Developing diagnostic criteria that include clinical signs and symptoms, imaging findings, and laboratory results as well as culture results would help to better assess the presence of bacterial coinfection in this patient population.

Conclusions

The study findings showed that up to 30% of patients hospitalized for COVID-19 infection received empiric antibiotic treatment for concern of bacterial coinfection. A pharmacist-led ASP provided interventions, including early discontinuation of antibiotics in 77% of these patients.

A low prevalence of bacterial coinfection (3%) in patients hospitalized with COVID-19 also was reported. A thorough clinical workup to determine the risk of bacterial coinfection in patients with COVID-19 is important before starting empiric antibiotic therapy. Continuing to promote the ASP during the COVID-19 pandemic to ensure responsible antibiotic use and prevent antimicrobial resistance is essential.

Infectious diseases are the most common reason for rehospitalization among patients with spinal cord injuries (SCI), regardless of the number of years postinjury.¹ The appropriate use and selection of antibiotics for properly diagnosed infectious diseases is especially important for this population. This principle helps to avoid the development of drug-resistant organisms and reduces the risk of recurrent infections, aligning with antibiotic stewardship.

Antibiotics are the most common class of drug allergies in the general population, and penicillin is the most frequently reported allergen (up to 10%).² Prescription drug–induced anaphylaxis is severe and life threatening with a reported frequency of 1.1%. Penicillin and sulfonamide (46 and 15 per 10,000 patients, respectively) are the most common allergens.³ Although there is a significant difference between an adverse drug reaction (ADR) and true hypersensitivity, once documented in the electronic health record (EHR) as an allergy, this information deters use of the listed drugs.

Genitourinary, skin, and respiratory diseases are the leading causes for rehospitalization in patients with SCI.¹ A large proportion of these are infectious in etiology and require antibiotic treatment. In fact, persons with SCI are at high risk for antibiotic overuse and hospital-acquired infection due to chronic bacteriuria, frequent health care exposure, implanted medical devices, and other factors.⁴ Concurrently, there is a crisis of antibiotic-resistant bacteria proliferation, described as a threat to patient safety and public health.^5,6 Its severity is illustrated by the report that 38% of the cultures from patients with spinal cord injury are multidrug resistant gram-negative organisms.⁷

The SCI center at James A. Haley Veterans’ Hospital (JAHVH) in Tampa, Florida, serves a high concentration of active-duty military members and veterans with SCI. A study that reviews the exact frequency of antibiotic drug allergies listed on the EHR would be a key first step to identify the magnitude of this issue. The results could guide investigation into differentiating true allergies from ADRs, thereby widening the options for potentially life-saving antibiotic treatment.

 

Methods

We performed a retrospective chart review of patients included in the local SCI registry between October 1, 2015, and September 30, 2017. We collected data on patient demographics (age, sex, race and ethnicity) and a description of patients’ injuries (International Standards for Neurological Classification of Spinal Cord Injury [ISNCSCI] and etiology of injury [traumatic vs atraumatic]). The outcomes included antibiotic allergy and ADRs.

In the EHR, allergies can be listed toward an antibiotic class or a specific antibiotic. An allergy to each specific antibiotic would be recorded separately; however, overlap among antibiotic classes was not duplicated. For example, if a subject has a listed antibiotic allergy to ceftriaxone and cefepime with listed reactions, we would record allergies to each of these antibiotics but would only report a single allergy to the cephalosporin subclass.

Since we did not differentiate hypersensitivity reactions (HSRs) from other ADRs, the reported reactions were grouped by signs and symptoms. There is a variety of terms used to report similar reactions, and best efforts were made to record the data as accurately as possible. Patient-reported history for risk stratification is a tool we used to group these historical reactions into high- vs low-risk for severe reactions. High-risk signs are those listed as anaphylaxis; anaphylactic reactions; angioedema presenting as swelling of mouth, eyes, lips, or tongue; blisters or ulcers involving the lips, mouth, eyes, urethra, vagina, or peeling skin; respiratory changes; shortness of breath; dyspnea; hypotension; or organ involvement (kidneys, lungs, liver).⁶

Inclusion criteria were all veterans who were diagnosed with tetraplegia or paraplegia and received annual evaluation between October 1, 2015, and September 30, 2017. We chose this period because it was the beginning of a financial year at the JAHVH SCI department using the SCI registry. The SCI annual evaluation is a routine practitioner encounter with the veteran, along with appropriate laboratory testing and imaging to follow up potential chronic health issues specific to patients with SCI. Annual evaluations provide an opportunity to maintain routine health screening and preventive care. Patients who had significant portions of data missing or missing elements of primary outcomes were excluded from analysis. The study was reviewed and approved by the University of South Florida Institutional Review Board (VA IRBNet #1573370-4 on September 9, 2019).

Results

Of 1866 patients reviewed, 207 (11.1%) were excluded due to missing data, resulting in 1659 records that were analyzed. Mean age was 64 years, and male to female ratio was about 10 to 1. Most of the SCI or diseases were classified as incomplete (n = 1249) per ISNCSCI (absence of sensory and motor function in the lowest sacral segments) compared with 373 classified as complete.

Thirty-seven subjects were without a recorded classification (Table 1).

Of the 1659 patients, 494 (29.8%) had a recorded allergy to antibiotics. The most frequently recorded were 217 penicillin (13.1%), 159 sulfa drugs (9.6%), 75 fluoroquinolone (4.5%), 66 cephalosporin (4.0%), and 44 vancomycin (2.7%) allergies.

Of the 217 recorded penicillin allergies, 26 (11.9%) were considered high-risk reactions (Table 2). Anaphylaxis was the recorded leading reaction.

 

Discussion

In this study, we evaluated the frequency and characteristics of antibiotic allergies at a single SCI center to better identify potential areas for quality improvement when recording drug allergies. A study in the general population used self-reported methods to collect such information found about a 15% prevalence of antibiotic allergy, which was lower than the 29.8% prevalence noted in our study.⁸

Regarding the most common antibiotic allergies, one study reported allergy to penicillin in the EHR in 12.8% of patients at a major US regional health care system, while 13.1% of patients with SCI had documented allergy to penicillin in our study.⁹ Regarding the other antibiotic classes, the percentage of allergies were higher than those reported in the general population: sulfonamide (9.6% vs 7.4%), fluoroquinolones (4.5% vs 1.3%), and cephalosporins (4.0% vs 1.7%).¹⁰ The EHR appears to capture a much higher rate of antibiotic allergies than that in self-reported studies, such as a study of self-reported allergy in the general adult population in Portugal, where only 4.5% of patients reported allergy to any β-lactam medications.¹⁰

The prevalence of an antibiotic allergy could be affected by the health care setting and sex distribution. For example, the Zhou and colleagues’ study conducted in the Greater Boston area showed higher reported antibiotic rates than those in a study from a Southern California medical group. The higher proportion of tertiary referral patients in that specific network was suggested to be the cause of the difference.^8,9 Our results in the SCI population are more comparable to that in a tertiary setting. This is consistent with the fact that persons with SCI generally have more exposure to antibiotics and consequently a higher reported rate of allergic reactions to antibiotics.

Similarly, the same study in Southern California noted that female patients use more antibiotics than do male patients, thus potentially contributing to higher rates of reported allergy toward all classes of antibiotics.⁸ Our study did not investigate antibiotic allergy by sex; however, the significantly higher proportion of male sex among the veteran population would have impacted these results.

Limitations

Our study was limited as a single-center retrospective study. However, our center is one of the major SCI specialty hubs, and the results should be somewhat reflective of those in the veterans with SCI population. Veterans under the US Department of Veterans Affairs (VA) medical care have the option to seek care or procedures in non-VA facilities. If allergies to antibiotics occurred outside of the VA system, there is no mechanism to automatically merge with the VA EHR allergy list, unless they are later recorded and added to the VA EHR. Thus, there is potential for underreporting.

Drug anaphylaxis incidence was noted to change over time.^4,8,9 For example, a downtrend of reported antibiotic allergy was reported between 1990 and 2013.¹⁰ Our study only reflects an overall prevalence of a single cohort, without demonstration of relationship to time.

Lastly, this study did not aim to differentiate HSRs from other ADRs. This is exactly the point of the study, which investigated the frequency of EHR-recorded antibiotic allergies in our SCI population and reflects the issue with indiscriminate recording of ADRs and HSRs under the umbrella of allergy in the EHR. Further diagnosing true allergies should be considered in the SCI population after weighing the risks and benefits of assessment, aligning with the wishes of the veteran, obtaining informed consent, and addressing the cost-effectiveness of specific tests. We suggest that primary care practitioners work closely with allergy specialists to formulate a mechanism to diagnose various antibiotic allergic reactions, including serum tryptase, epicutaneous skin testing, intradermal skin testing, patch testing, delayed intradermal testing, and drug challenge as appropriate. It is also possible that in cases where very mild reactions/adverse effects of antibiotics were recorded in the EHR, the clinicians and veterans may discuss reintroducing the same antibiotics or proceeding with further testing if necessary. In contrast, the 12% of those with a high risk of severe allergic reactions to penicillin in our study would benefit from allergist evaluation and access to epinephrine auto-injectors at all times. Differentiating true allergy is the only clear way to deter unnecessary avoidance of first-line therapies for antibiotic treatment and avoid promotion of antibiotic resistance.

Future studies can analyze antibiotic allergy based on demographics, including sex and age difference, as well as exploring outpatient vs inpatient settings. Aside from prevalence, we hope to demonstrate antibiotic allergy over time, especially after integration of diagnostic allergy testing, to evaluate the impact to EHR-recorded allergies.

 

Conclusions

Almost 30% of patients with SCI had a recorded allergy to at least 1 antibiotic. The most common allergy was to penicillin, which is similar to what has previously been reported for the general adult US population. However, only 12% of those with a penicillin allergy were considered high risk of true allergic reactions. Consequently, there are opportunities to examine whether approaches to confirm true reactions (such as skin testing) would help to mitigate unnecessary avoidance of certain antibiotic classes due to mild ADRs, rather than a true allergy, in persons with SCI. This would be an important effort to combat both individual safety concerns and the public health crisis of antibiotic resistance. Given the available evidence, it is reasonable for SCI health care practitioners to discuss the potential risks and benefits of allergy testing with patients with SCI; this maintains a patient-centered approach that can ensure judicious use of antibiotics when necessary.

Acknowledgments

This material is based on work supported (or supported in part) with resources and the use of facilities at the James A. Haley Veterans’ Hospital

Infectious diseases are the most common reason for rehospitalization among patients with spinal cord injuries (SCI), regardless of the number of years postinjury.¹ The appropriate use and selection of antibiotics for properly diagnosed infectious diseases is especially important for this population. This principle helps to avoid the development of drug-resistant organisms and reduces the risk of recurrent infections, aligning with antibiotic stewardship.

Antibiotics are the most common class of drug allergies in the general population, and penicillin is the most frequently reported allergen (up to 10%).² Prescription drug–induced anaphylaxis is severe and life threatening with a reported frequency of 1.1%. Penicillin and sulfonamide (46 and 15 per 10,000 patients, respectively) are the most common allergens.³ Although there is a significant difference between an adverse drug reaction (ADR) and true hypersensitivity, once documented in the electronic health record (EHR) as an allergy, this information deters use of the listed drugs.

Genitourinary, skin, and respiratory diseases are the leading causes for rehospitalization in patients with SCI.¹ A large proportion of these are infectious in etiology and require antibiotic treatment. In fact, persons with SCI are at high risk for antibiotic overuse and hospital-acquired infection due to chronic bacteriuria, frequent health care exposure, implanted medical devices, and other factors.⁴ Concurrently, there is a crisis of antibiotic-resistant bacteria proliferation, described as a threat to patient safety and public health.^5,6 Its severity is illustrated by the report that 38% of the cultures from patients with spinal cord injury are multidrug resistant gram-negative organisms.⁷

The SCI center at James A. Haley Veterans’ Hospital (JAHVH) in Tampa, Florida, serves a high concentration of active-duty military members and veterans with SCI. A study that reviews the exact frequency of antibiotic drug allergies listed on the EHR would be a key first step to identify the magnitude of this issue. The results could guide investigation into differentiating true allergies from ADRs, thereby widening the options for potentially life-saving antibiotic treatment.

 

Methods

We performed a retrospective chart review of patients included in the local SCI registry between October 1, 2015, and September 30, 2017. We collected data on patient demographics (age, sex, race and ethnicity) and a description of patients’ injuries (International Standards for Neurological Classification of Spinal Cord Injury [ISNCSCI] and etiology of injury [traumatic vs atraumatic]). The outcomes included antibiotic allergy and ADRs.

In the EHR, allergies can be listed toward an antibiotic class or a specific antibiotic. An allergy to each specific antibiotic would be recorded separately; however, overlap among antibiotic classes was not duplicated. For example, if a subject has a listed antibiotic allergy to ceftriaxone and cefepime with listed reactions, we would record allergies to each of these antibiotics but would only report a single allergy to the cephalosporin subclass.

Since we did not differentiate hypersensitivity reactions (HSRs) from other ADRs, the reported reactions were grouped by signs and symptoms. There is a variety of terms used to report similar reactions, and best efforts were made to record the data as accurately as possible. Patient-reported history for risk stratification is a tool we used to group these historical reactions into high- vs low-risk for severe reactions. High-risk signs are those listed as anaphylaxis; anaphylactic reactions; angioedema presenting as swelling of mouth, eyes, lips, or tongue; blisters or ulcers involving the lips, mouth, eyes, urethra, vagina, or peeling skin; respiratory changes; shortness of breath; dyspnea; hypotension; or organ involvement (kidneys, lungs, liver).⁶

Inclusion criteria were all veterans who were diagnosed with tetraplegia or paraplegia and received annual evaluation between October 1, 2015, and September 30, 2017. We chose this period because it was the beginning of a financial year at the JAHVH SCI department using the SCI registry. The SCI annual evaluation is a routine practitioner encounter with the veteran, along with appropriate laboratory testing and imaging to follow up potential chronic health issues specific to patients with SCI. Annual evaluations provide an opportunity to maintain routine health screening and preventive care. Patients who had significant portions of data missing or missing elements of primary outcomes were excluded from analysis. The study was reviewed and approved by the University of South Florida Institutional Review Board (VA IRBNet #1573370-4 on September 9, 2019).

Results

Of 1866 patients reviewed, 207 (11.1%) were excluded due to missing data, resulting in 1659 records that were analyzed. Mean age was 64 years, and male to female ratio was about 10 to 1. Most of the SCI or diseases were classified as incomplete (n = 1249) per ISNCSCI (absence of sensory and motor function in the lowest sacral segments) compared with 373 classified as complete.

Thirty-seven subjects were without a recorded classification (Table 1).

Of the 1659 patients, 494 (29.8%) had a recorded allergy to antibiotics. The most frequently recorded were 217 penicillin (13.1%), 159 sulfa drugs (9.6%), 75 fluoroquinolone (4.5%), 66 cephalosporin (4.0%), and 44 vancomycin (2.7%) allergies.

Of the 217 recorded penicillin allergies, 26 (11.9%) were considered high-risk reactions (Table 2). Anaphylaxis was the recorded leading reaction.

 

Discussion

In this study, we evaluated the frequency and characteristics of antibiotic allergies at a single SCI center to better identify potential areas for quality improvement when recording drug allergies. A study in the general population used self-reported methods to collect such information found about a 15% prevalence of antibiotic allergy, which was lower than the 29.8% prevalence noted in our study.⁸

Regarding the most common antibiotic allergies, one study reported allergy to penicillin in the EHR in 12.8% of patients at a major US regional health care system, while 13.1% of patients with SCI had documented allergy to penicillin in our study.⁹ Regarding the other antibiotic classes, the percentage of allergies were higher than those reported in the general population: sulfonamide (9.6% vs 7.4%), fluoroquinolones (4.5% vs 1.3%), and cephalosporins (4.0% vs 1.7%).¹⁰ The EHR appears to capture a much higher rate of antibiotic allergies than that in self-reported studies, such as a study of self-reported allergy in the general adult population in Portugal, where only 4.5% of patients reported allergy to any β-lactam medications.¹⁰

The prevalence of an antibiotic allergy could be affected by the health care setting and sex distribution. For example, the Zhou and colleagues’ study conducted in the Greater Boston area showed higher reported antibiotic rates than those in a study from a Southern California medical group. The higher proportion of tertiary referral patients in that specific network was suggested to be the cause of the difference.^8,9 Our results in the SCI population are more comparable to that in a tertiary setting. This is consistent with the fact that persons with SCI generally have more exposure to antibiotics and consequently a higher reported rate of allergic reactions to antibiotics.

Similarly, the same study in Southern California noted that female patients use more antibiotics than do male patients, thus potentially contributing to higher rates of reported allergy toward all classes of antibiotics.⁸ Our study did not investigate antibiotic allergy by sex; however, the significantly higher proportion of male sex among the veteran population would have impacted these results.

Limitations

Our study was limited as a single-center retrospective study. However, our center is one of the major SCI specialty hubs, and the results should be somewhat reflective of those in the veterans with SCI population. Veterans under the US Department of Veterans Affairs (VA) medical care have the option to seek care or procedures in non-VA facilities. If allergies to antibiotics occurred outside of the VA system, there is no mechanism to automatically merge with the VA EHR allergy list, unless they are later recorded and added to the VA EHR. Thus, there is potential for underreporting.

Drug anaphylaxis incidence was noted to change over time.^4,8,9 For example, a downtrend of reported antibiotic allergy was reported between 1990 and 2013.¹⁰ Our study only reflects an overall prevalence of a single cohort, without demonstration of relationship to time.

Lastly, this study did not aim to differentiate HSRs from other ADRs. This is exactly the point of the study, which investigated the frequency of EHR-recorded antibiotic allergies in our SCI population and reflects the issue with indiscriminate recording of ADRs and HSRs under the umbrella of allergy in the EHR. Further diagnosing true allergies should be considered in the SCI population after weighing the risks and benefits of assessment, aligning with the wishes of the veteran, obtaining informed consent, and addressing the cost-effectiveness of specific tests. We suggest that primary care practitioners work closely with allergy specialists to formulate a mechanism to diagnose various antibiotic allergic reactions, including serum tryptase, epicutaneous skin testing, intradermal skin testing, patch testing, delayed intradermal testing, and drug challenge as appropriate. It is also possible that in cases where very mild reactions/adverse effects of antibiotics were recorded in the EHR, the clinicians and veterans may discuss reintroducing the same antibiotics or proceeding with further testing if necessary. In contrast, the 12% of those with a high risk of severe allergic reactions to penicillin in our study would benefit from allergist evaluation and access to epinephrine auto-injectors at all times. Differentiating true allergy is the only clear way to deter unnecessary avoidance of first-line therapies for antibiotic treatment and avoid promotion of antibiotic resistance.

Future studies can analyze antibiotic allergy based on demographics, including sex and age difference, as well as exploring outpatient vs inpatient settings. Aside from prevalence, we hope to demonstrate antibiotic allergy over time, especially after integration of diagnostic allergy testing, to evaluate the impact to EHR-recorded allergies.

 

Conclusions

Almost 30% of patients with SCI had a recorded allergy to at least 1 antibiotic. The most common allergy was to penicillin, which is similar to what has previously been reported for the general adult US population. However, only 12% of those with a penicillin allergy were considered high risk of true allergic reactions. Consequently, there are opportunities to examine whether approaches to confirm true reactions (such as skin testing) would help to mitigate unnecessary avoidance of certain antibiotic classes due to mild ADRs, rather than a true allergy, in persons with SCI. This would be an important effort to combat both individual safety concerns and the public health crisis of antibiotic resistance. Given the available evidence, it is reasonable for SCI health care practitioners to discuss the potential risks and benefits of allergy testing with patients with SCI; this maintains a patient-centered approach that can ensure judicious use of antibiotics when necessary.

Acknowledgments

This material is based on work supported (or supported in part) with resources and the use of facilities at the James A. Haley Veterans’ Hospital

Infectious diseases are the most common reason for rehospitalization among patients with spinal cord injuries (SCI), regardless of the number of years postinjury.¹ The appropriate use and selection of antibiotics for properly diagnosed infectious diseases is especially important for this population. This principle helps to avoid the development of drug-resistant organisms and reduces the risk of recurrent infections, aligning with antibiotic stewardship.

Antibiotics are the most common class of drug allergies in the general population, and penicillin is the most frequently reported allergen (up to 10%).² Prescription drug–induced anaphylaxis is severe and life threatening with a reported frequency of 1.1%. Penicillin and sulfonamide (46 and 15 per 10,000 patients, respectively) are the most common allergens.³ Although there is a significant difference between an adverse drug reaction (ADR) and true hypersensitivity, once documented in the electronic health record (EHR) as an allergy, this information deters use of the listed drugs.

Genitourinary, skin, and respiratory diseases are the leading causes for rehospitalization in patients with SCI.¹ A large proportion of these are infectious in etiology and require antibiotic treatment. In fact, persons with SCI are at high risk for antibiotic overuse and hospital-acquired infection due to chronic bacteriuria, frequent health care exposure, implanted medical devices, and other factors.⁴ Concurrently, there is a crisis of antibiotic-resistant bacteria proliferation, described as a threat to patient safety and public health.^5,6 Its severity is illustrated by the report that 38% of the cultures from patients with spinal cord injury are multidrug resistant gram-negative organisms.⁷

The SCI center at James A. Haley Veterans’ Hospital (JAHVH) in Tampa, Florida, serves a high concentration of active-duty military members and veterans with SCI. A study that reviews the exact frequency of antibiotic drug allergies listed on the EHR would be a key first step to identify the magnitude of this issue. The results could guide investigation into differentiating true allergies from ADRs, thereby widening the options for potentially life-saving antibiotic treatment.

 

Methods

We performed a retrospective chart review of patients included in the local SCI registry between October 1, 2015, and September 30, 2017. We collected data on patient demographics (age, sex, race and ethnicity) and a description of patients’ injuries (International Standards for Neurological Classification of Spinal Cord Injury [ISNCSCI] and etiology of injury [traumatic vs atraumatic]). The outcomes included antibiotic allergy and ADRs.

In the EHR, allergies can be listed toward an antibiotic class or a specific antibiotic. An allergy to each specific antibiotic would be recorded separately; however, overlap among antibiotic classes was not duplicated. For example, if a subject has a listed antibiotic allergy to ceftriaxone and cefepime with listed reactions, we would record allergies to each of these antibiotics but would only report a single allergy to the cephalosporin subclass.

Since we did not differentiate hypersensitivity reactions (HSRs) from other ADRs, the reported reactions were grouped by signs and symptoms. There is a variety of terms used to report similar reactions, and best efforts were made to record the data as accurately as possible. Patient-reported history for risk stratification is a tool we used to group these historical reactions into high- vs low-risk for severe reactions. High-risk signs are those listed as anaphylaxis; anaphylactic reactions; angioedema presenting as swelling of mouth, eyes, lips, or tongue; blisters or ulcers involving the lips, mouth, eyes, urethra, vagina, or peeling skin; respiratory changes; shortness of breath; dyspnea; hypotension; or organ involvement (kidneys, lungs, liver).⁶

Inclusion criteria were all veterans who were diagnosed with tetraplegia or paraplegia and received annual evaluation between October 1, 2015, and September 30, 2017. We chose this period because it was the beginning of a financial year at the JAHVH SCI department using the SCI registry. The SCI annual evaluation is a routine practitioner encounter with the veteran, along with appropriate laboratory testing and imaging to follow up potential chronic health issues specific to patients with SCI. Annual evaluations provide an opportunity to maintain routine health screening and preventive care. Patients who had significant portions of data missing or missing elements of primary outcomes were excluded from analysis. The study was reviewed and approved by the University of South Florida Institutional Review Board (VA IRBNet #1573370-4 on September 9, 2019).

Results

Of 1866 patients reviewed, 207 (11.1%) were excluded due to missing data, resulting in 1659 records that were analyzed. Mean age was 64 years, and male to female ratio was about 10 to 1. Most of the SCI or diseases were classified as incomplete (n = 1249) per ISNCSCI (absence of sensory and motor function in the lowest sacral segments) compared with 373 classified as complete.

Thirty-seven subjects were without a recorded classification (Table 1).

Of the 1659 patients, 494 (29.8%) had a recorded allergy to antibiotics. The most frequently recorded were 217 penicillin (13.1%), 159 sulfa drugs (9.6%), 75 fluoroquinolone (4.5%), 66 cephalosporin (4.0%), and 44 vancomycin (2.7%) allergies.

Of the 217 recorded penicillin allergies, 26 (11.9%) were considered high-risk reactions (Table 2). Anaphylaxis was the recorded leading reaction.

 

Discussion

In this study, we evaluated the frequency and characteristics of antibiotic allergies at a single SCI center to better identify potential areas for quality improvement when recording drug allergies. A study in the general population used self-reported methods to collect such information found about a 15% prevalence of antibiotic allergy, which was lower than the 29.8% prevalence noted in our study.⁸

Regarding the most common antibiotic allergies, one study reported allergy to penicillin in the EHR in 12.8% of patients at a major US regional health care system, while 13.1% of patients with SCI had documented allergy to penicillin in our study.⁹ Regarding the other antibiotic classes, the percentage of allergies were higher than those reported in the general population: sulfonamide (9.6% vs 7.4%), fluoroquinolones (4.5% vs 1.3%), and cephalosporins (4.0% vs 1.7%).¹⁰ The EHR appears to capture a much higher rate of antibiotic allergies than that in self-reported studies, such as a study of self-reported allergy in the general adult population in Portugal, where only 4.5% of patients reported allergy to any β-lactam medications.¹⁰

The prevalence of an antibiotic allergy could be affected by the health care setting and sex distribution. For example, the Zhou and colleagues’ study conducted in the Greater Boston area showed higher reported antibiotic rates than those in a study from a Southern California medical group. The higher proportion of tertiary referral patients in that specific network was suggested to be the cause of the difference.^8,9 Our results in the SCI population are more comparable to that in a tertiary setting. This is consistent with the fact that persons with SCI generally have more exposure to antibiotics and consequently a higher reported rate of allergic reactions to antibiotics.

Similarly, the same study in Southern California noted that female patients use more antibiotics than do male patients, thus potentially contributing to higher rates of reported allergy toward all classes of antibiotics.⁸ Our study did not investigate antibiotic allergy by sex; however, the significantly higher proportion of male sex among the veteran population would have impacted these results.

Limitations

Our study was limited as a single-center retrospective study. However, our center is one of the major SCI specialty hubs, and the results should be somewhat reflective of those in the veterans with SCI population. Veterans under the US Department of Veterans Affairs (VA) medical care have the option to seek care or procedures in non-VA facilities. If allergies to antibiotics occurred outside of the VA system, there is no mechanism to automatically merge with the VA EHR allergy list, unless they are later recorded and added to the VA EHR. Thus, there is potential for underreporting.

Drug anaphylaxis incidence was noted to change over time.^4,8,9 For example, a downtrend of reported antibiotic allergy was reported between 1990 and 2013.¹⁰ Our study only reflects an overall prevalence of a single cohort, without demonstration of relationship to time.

Lastly, this study did not aim to differentiate HSRs from other ADRs. This is exactly the point of the study, which investigated the frequency of EHR-recorded antibiotic allergies in our SCI population and reflects the issue with indiscriminate recording of ADRs and HSRs under the umbrella of allergy in the EHR. Further diagnosing true allergies should be considered in the SCI population after weighing the risks and benefits of assessment, aligning with the wishes of the veteran, obtaining informed consent, and addressing the cost-effectiveness of specific tests. We suggest that primary care practitioners work closely with allergy specialists to formulate a mechanism to diagnose various antibiotic allergic reactions, including serum tryptase, epicutaneous skin testing, intradermal skin testing, patch testing, delayed intradermal testing, and drug challenge as appropriate. It is also possible that in cases where very mild reactions/adverse effects of antibiotics were recorded in the EHR, the clinicians and veterans may discuss reintroducing the same antibiotics or proceeding with further testing if necessary. In contrast, the 12% of those with a high risk of severe allergic reactions to penicillin in our study would benefit from allergist evaluation and access to epinephrine auto-injectors at all times. Differentiating true allergy is the only clear way to deter unnecessary avoidance of first-line therapies for antibiotic treatment and avoid promotion of antibiotic resistance.

Future studies can analyze antibiotic allergy based on demographics, including sex and age difference, as well as exploring outpatient vs inpatient settings. Aside from prevalence, we hope to demonstrate antibiotic allergy over time, especially after integration of diagnostic allergy testing, to evaluate the impact to EHR-recorded allergies.

 

Conclusions

Almost 30% of patients with SCI had a recorded allergy to at least 1 antibiotic. The most common allergy was to penicillin, which is similar to what has previously been reported for the general adult US population. However, only 12% of those with a penicillin allergy were considered high risk of true allergic reactions. Consequently, there are opportunities to examine whether approaches to confirm true reactions (such as skin testing) would help to mitigate unnecessary avoidance of certain antibiotic classes due to mild ADRs, rather than a true allergy, in persons with SCI. This would be an important effort to combat both individual safety concerns and the public health crisis of antibiotic resistance. Given the available evidence, it is reasonable for SCI health care practitioners to discuss the potential risks and benefits of allergy testing with patients with SCI; this maintains a patient-centered approach that can ensure judicious use of antibiotics when necessary.

Acknowledgments

This material is based on work supported (or supported in part) with resources and the use of facilities at the James A. Haley Veterans’ Hospital

Since 1983, the correlation between head and neck squamous cell carcinoma (SCC) and human papillomavirus (HPV) has been of great interest to head and neck oncologists.¹ In 1998, Smith and colleagues provided evidence of HPV as an independent risk factor for the development of head and neck SCC.² HPV-associated head and neck SCC accounts for between 30% and 64% of oropharyngeal SCC, depending on the published study; tonsil primaries account for the majority of these cancers.^3,4

The presence of HPV E6 and E7 oncoproteins leads to the inactivation of p53 and pRb tumor suppressors. Furthermore, Ragin and colleagues discussed a distinct molecular pathway specific to HPV-associated head and neck SCC, which was different from non–HPV-associated head and neck SCC, involving genetic mutations in CDKN2A/p16.⁵

Current methods in correlating the presence of HPV infection in head and neck SCC have centered on p16^INK4a (p16) immunohistochemistry (IHC) staining and DNA in situ hybridization (ISH) for specific HPV DNA types. IHC staining for p16 involves a monoclonal antibody specific to p16. The usefulness of this test relies on p16 overexpression due to the inactivation of pRb by the HPV E7 oncoprotein. This test is readily performed on archived tissue and has a documented sensitivity and specificity of 100% and 79%, respectively, as reported by Singhi and Westra in 2010.⁶ HPV DNA fluorescence in situ hybridization is the gold standard for determining the presence of specific types of HPV DNA; however, p16 IHC can serve as a rapid, less costly means of studying archived tissue, lending its utility to retrospective population-based studies.

METHODS

A retrospective study was designed to determine the proportion of HPV-associated oropharyngeal SCC in a US Department of Veterans Affairs (VA) population, using p16 antigen IHC on paraffin-embedded tissue as the surrogate marker for the presence of HPV infection. Patients consisted of veterans who were treated for oropharyngeal SCC at Veterans Affairs Memphis Healthcare System (VAMHS) in Tennessee between January 1, 2000, and December 31, 2008. This data range allowed for at least 5 years of follow-up. Patients were excluded who lacked enough tissue specimens for analysis. Measurement outcomes included p16 expression, with subset analysis by race and ethnicity, degree of tobacco and alcohol use, tumor location, stage, age at diagnosis, and survival outcome. Microsoft Excel was used to calculate Fisher exact test, Student t test, and χ² statistics. Significance was set at P < .05. This study received institutional review board approval from the University of Tennessee Health Science Center and the VAMHS.

RESULTS

We identified 66 total cases of oropharyngeal SCC; 19 cases (29%) were positive for p16. The mean age at diagnosis for the p16-positive cohort was 59 years vs 61 years for the p16-negative cohort (P = .22; Table 1).

Of the patients with p16-positive status, most were White veterans compared with those with p16-negative status, which consisted of more African Americans veterans (P = .04). Smoking data were available for 18 of 19 patients with p16-positive status and 46 of 47 patients whose status was p16 negative. Four patients (22%) with p16-positive status were tobacco naïve compared with none of the patients with p16-negative status (P = .005). Alcohol use data were available for 17 of 19 patients in the p16-positive cohort and 46 of 47 patients in the p16-negative cohort. Three patients (18%) with p16-positive status were alcohol naïve compared with 2 patients (4%) with p16-negative status (P = .12). Of the patients in the study, 65 of 66 died during the study period, 5 (28%) of the p16-positive cohort and 17 (36%) of the p16-negative cohort were directly attributed to oropharyngeal SCC (P = .52).

Although the tonsil was the most common site of tumor origin in both the p16-positive and negative cohorts (63% vs 51%, respectively), our analysis showed no statistically significant difference in sites of origin (P = .69) (Table 2).

There was no difference in the stage at presentation between the 2 cohorts, with the most presenting with stage III or IV disease (P = .75).

 

DISCUSSION

The VAMHS population in our study had a lower proportion of HPV-associated oropharyngeal SCC compared with studies on nonveteran populations (29% vs 40%-80%, respectively).^5,6 This disparity may indicate a true difference in these populations or may be related to a decreased prevalence of HPV infection in the population served by the VAMHS. This single-institution population did not completely correlate with previous population studies. Specifically, age at presentation (equivalent to patients with p16-negative status rather than earlier age at onset), disease stage at presentation (lower stage for patients with p16-positive status), and disease-specific survival (not improved compared with patients with p16-negative status in other studies) were dissimilar to previous investigations.^2,3

The increased age and staging at presentation could be related in these patients with p16-positive status, which may further account for the lack of improved survival. Furthermore, both groups tended to use alcohol at a high proportion; whereas other populations have had a lesser degree of alcohol intake with p16 positivity.^1-4 These differences may be due to variations in the habits and behavior of VA patients compared with non-VA patients.^3,4

HPV-associated oropharyngeal SCC in published data has been associated with high-risk sexual behavior, lower age, and less tobacco and alcohol use.^5,6 No difference was noted in tumor site predilection; however, the small size of our study could explain the lack of finding site preference shown in previous studies.^2,3Other veteran-specific factors are absent in the at-large population, such as Agent Orange exposure. More than 8 million veterans (22%) from the Vietnam era self-reported Agent Orange exposure.⁷ Agent Orange exposure significantly predicted developing upper aerodigestive tract cancer. Oropharyngeal, nasopharyngeal, laryngeal, and thyroid cancers were significantly associated with Agent Orange exposure. Interestingly, these patients experienced an improved 10-year survival rate compared with patients not exposed to Agent Orange. This finding contrasts with our patients, who did not experience improved outcomes vs nonveteran patients with head and neck cancer.⁷

Suicide in veterans with head and neck cancer has been evaluated and was found at an incidence of 0.7%. Survivors of head and neck cancer are almost twice as likely to die by suicide compared with other cancer survivors. These patients have a higher rate of mental health disorders, substance misuse, and use of palliative care services.⁸ Sixty-five of 66 of our patients died during the 5-year observation period, although none died by suicide.

In a 2022 cohort study by Sun and colleagues, upfront surgical treatment was associated with a 23% reduced risk of stroke compared with definitive chemoradiotherapy in US veterans with oropharyngeal carcinoma.⁹ In our study, 58 of 66 patients (88%) received concurrent chemoradiation, possibly reflecting the more advanced stage of diagnosis in our study population. This was due to comorbidities and other health and economic factors. In our study, 43 patients (65%) died of factors not related to the disease, reflecting the overall comorbidity burden of this population. Seven patients (11%) in our 5-year study died of a documented stroke. In the study of veterans by Sun and colleagues, the 10-year cumulative incidence of stroke was 12.5% and death was 57.3%.⁹ Our veteran population experienced a similar incidence of strokes. These findings may need to be included when discussing the risk-benefit aspects of different treatment options with our veteran patients with oropharyngeal cancer.

To understand the influence of HPV infection on the course of oropharyngeal SCC in the VA patient population and to apply this understanding to future individualized treatment paradigms, this study can be expanded to a greater number of VA patients. p16 immunoexpression appears to be a useful surrogate for high-risk HPV infection in oropharyngeal SCC, and its ease of use supports its feasibility in further VA population analysis.¹⁰ While realizing that the veteran HPV-associated oropharyngeal SCC population differs from the civilian HPV-associated oropharyngeal SCC population, we also have realized that other unique considerations in the veteran population, such as chemical warfare exposure, mental illness, and vascular disease, complicate treatment decisions in these patients.

CONCLUSIONS

Disparities in racial distribution and tobacco use between patients with p16-positive and p16-negative status are similar to those reported in non-VA populations. In contrast, the frequently reported younger age at presentation and better disease outcomes seen in non-VA patients were not observed, perhaps due to the lower percentage of p16 expression in VA patients with oropharyngeal SCC. Whereas de-intensification of therapy may be considered for many patients with oropharygeal cancer that is HPV-associated because of improved prognosis, this approach should be undertaken with great care in this group of patients. Personalization of therapy for these HPV-associated oropharyngeal SCC in the veteran population must be adapted to mitigate this critical disparity.

Since 1983, the correlation between head and neck squamous cell carcinoma (SCC) and human papillomavirus (HPV) has been of great interest to head and neck oncologists.¹ In 1998, Smith and colleagues provided evidence of HPV as an independent risk factor for the development of head and neck SCC.² HPV-associated head and neck SCC accounts for between 30% and 64% of oropharyngeal SCC, depending on the published study; tonsil primaries account for the majority of these cancers.^3,4

The presence of HPV E6 and E7 oncoproteins leads to the inactivation of p53 and pRb tumor suppressors. Furthermore, Ragin and colleagues discussed a distinct molecular pathway specific to HPV-associated head and neck SCC, which was different from non–HPV-associated head and neck SCC, involving genetic mutations in CDKN2A/p16.⁵

Current methods in correlating the presence of HPV infection in head and neck SCC have centered on p16^INK4a (p16) immunohistochemistry (IHC) staining and DNA in situ hybridization (ISH) for specific HPV DNA types. IHC staining for p16 involves a monoclonal antibody specific to p16. The usefulness of this test relies on p16 overexpression due to the inactivation of pRb by the HPV E7 oncoprotein. This test is readily performed on archived tissue and has a documented sensitivity and specificity of 100% and 79%, respectively, as reported by Singhi and Westra in 2010.⁶ HPV DNA fluorescence in situ hybridization is the gold standard for determining the presence of specific types of HPV DNA; however, p16 IHC can serve as a rapid, less costly means of studying archived tissue, lending its utility to retrospective population-based studies.

METHODS

A retrospective study was designed to determine the proportion of HPV-associated oropharyngeal SCC in a US Department of Veterans Affairs (VA) population, using p16 antigen IHC on paraffin-embedded tissue as the surrogate marker for the presence of HPV infection. Patients consisted of veterans who were treated for oropharyngeal SCC at Veterans Affairs Memphis Healthcare System (VAMHS) in Tennessee between January 1, 2000, and December 31, 2008. This data range allowed for at least 5 years of follow-up. Patients were excluded who lacked enough tissue specimens for analysis. Measurement outcomes included p16 expression, with subset analysis by race and ethnicity, degree of tobacco and alcohol use, tumor location, stage, age at diagnosis, and survival outcome. Microsoft Excel was used to calculate Fisher exact test, Student t test, and χ² statistics. Significance was set at P < .05. This study received institutional review board approval from the University of Tennessee Health Science Center and the VAMHS.

RESULTS

We identified 66 total cases of oropharyngeal SCC; 19 cases (29%) were positive for p16. The mean age at diagnosis for the p16-positive cohort was 59 years vs 61 years for the p16-negative cohort (P = .22; Table 1).

Of the patients with p16-positive status, most were White veterans compared with those with p16-negative status, which consisted of more African Americans veterans (P = .04). Smoking data were available for 18 of 19 patients with p16-positive status and 46 of 47 patients whose status was p16 negative. Four patients (22%) with p16-positive status were tobacco naïve compared with none of the patients with p16-negative status (P = .005). Alcohol use data were available for 17 of 19 patients in the p16-positive cohort and 46 of 47 patients in the p16-negative cohort. Three patients (18%) with p16-positive status were alcohol naïve compared with 2 patients (4%) with p16-negative status (P = .12). Of the patients in the study, 65 of 66 died during the study period, 5 (28%) of the p16-positive cohort and 17 (36%) of the p16-negative cohort were directly attributed to oropharyngeal SCC (P = .52).

Although the tonsil was the most common site of tumor origin in both the p16-positive and negative cohorts (63% vs 51%, respectively), our analysis showed no statistically significant difference in sites of origin (P = .69) (Table 2).

There was no difference in the stage at presentation between the 2 cohorts, with the most presenting with stage III or IV disease (P = .75).

 

DISCUSSION

The VAMHS population in our study had a lower proportion of HPV-associated oropharyngeal SCC compared with studies on nonveteran populations (29% vs 40%-80%, respectively).^5,6 This disparity may indicate a true difference in these populations or may be related to a decreased prevalence of HPV infection in the population served by the VAMHS. This single-institution population did not completely correlate with previous population studies. Specifically, age at presentation (equivalent to patients with p16-negative status rather than earlier age at onset), disease stage at presentation (lower stage for patients with p16-positive status), and disease-specific survival (not improved compared with patients with p16-negative status in other studies) were dissimilar to previous investigations.^2,3

The increased age and staging at presentation could be related in these patients with p16-positive status, which may further account for the lack of improved survival. Furthermore, both groups tended to use alcohol at a high proportion; whereas other populations have had a lesser degree of alcohol intake with p16 positivity.^1-4 These differences may be due to variations in the habits and behavior of VA patients compared with non-VA patients.^3,4

HPV-associated oropharyngeal SCC in published data has been associated with high-risk sexual behavior, lower age, and less tobacco and alcohol use.^5,6 No difference was noted in tumor site predilection; however, the small size of our study could explain the lack of finding site preference shown in previous studies.^2,3Other veteran-specific factors are absent in the at-large population, such as Agent Orange exposure. More than 8 million veterans (22%) from the Vietnam era self-reported Agent Orange exposure.⁷ Agent Orange exposure significantly predicted developing upper aerodigestive tract cancer. Oropharyngeal, nasopharyngeal, laryngeal, and thyroid cancers were significantly associated with Agent Orange exposure. Interestingly, these patients experienced an improved 10-year survival rate compared with patients not exposed to Agent Orange. This finding contrasts with our patients, who did not experience improved outcomes vs nonveteran patients with head and neck cancer.⁷

Suicide in veterans with head and neck cancer has been evaluated and was found at an incidence of 0.7%. Survivors of head and neck cancer are almost twice as likely to die by suicide compared with other cancer survivors. These patients have a higher rate of mental health disorders, substance misuse, and use of palliative care services.⁸ Sixty-five of 66 of our patients died during the 5-year observation period, although none died by suicide.

In a 2022 cohort study by Sun and colleagues, upfront surgical treatment was associated with a 23% reduced risk of stroke compared with definitive chemoradiotherapy in US veterans with oropharyngeal carcinoma.⁹ In our study, 58 of 66 patients (88%) received concurrent chemoradiation, possibly reflecting the more advanced stage of diagnosis in our study population. This was due to comorbidities and other health and economic factors. In our study, 43 patients (65%) died of factors not related to the disease, reflecting the overall comorbidity burden of this population. Seven patients (11%) in our 5-year study died of a documented stroke. In the study of veterans by Sun and colleagues, the 10-year cumulative incidence of stroke was 12.5% and death was 57.3%.⁹ Our veteran population experienced a similar incidence of strokes. These findings may need to be included when discussing the risk-benefit aspects of different treatment options with our veteran patients with oropharyngeal cancer.

To understand the influence of HPV infection on the course of oropharyngeal SCC in the VA patient population and to apply this understanding to future individualized treatment paradigms, this study can be expanded to a greater number of VA patients. p16 immunoexpression appears to be a useful surrogate for high-risk HPV infection in oropharyngeal SCC, and its ease of use supports its feasibility in further VA population analysis.¹⁰ While realizing that the veteran HPV-associated oropharyngeal SCC population differs from the civilian HPV-associated oropharyngeal SCC population, we also have realized that other unique considerations in the veteran population, such as chemical warfare exposure, mental illness, and vascular disease, complicate treatment decisions in these patients.

CONCLUSIONS

Disparities in racial distribution and tobacco use between patients with p16-positive and p16-negative status are similar to those reported in non-VA populations. In contrast, the frequently reported younger age at presentation and better disease outcomes seen in non-VA patients were not observed, perhaps due to the lower percentage of p16 expression in VA patients with oropharyngeal SCC. Whereas de-intensification of therapy may be considered for many patients with oropharygeal cancer that is HPV-associated because of improved prognosis, this approach should be undertaken with great care in this group of patients. Personalization of therapy for these HPV-associated oropharyngeal SCC in the veteran population must be adapted to mitigate this critical disparity.

Since 1983, the correlation between head and neck squamous cell carcinoma (SCC) and human papillomavirus (HPV) has been of great interest to head and neck oncologists.¹ In 1998, Smith and colleagues provided evidence of HPV as an independent risk factor for the development of head and neck SCC.² HPV-associated head and neck SCC accounts for between 30% and 64% of oropharyngeal SCC, depending on the published study; tonsil primaries account for the majority of these cancers.^3,4

The presence of HPV E6 and E7 oncoproteins leads to the inactivation of p53 and pRb tumor suppressors. Furthermore, Ragin and colleagues discussed a distinct molecular pathway specific to HPV-associated head and neck SCC, which was different from non–HPV-associated head and neck SCC, involving genetic mutations in CDKN2A/p16.⁵

Current methods in correlating the presence of HPV infection in head and neck SCC have centered on p16^INK4a (p16) immunohistochemistry (IHC) staining and DNA in situ hybridization (ISH) for specific HPV DNA types. IHC staining for p16 involves a monoclonal antibody specific to p16. The usefulness of this test relies on p16 overexpression due to the inactivation of pRb by the HPV E7 oncoprotein. This test is readily performed on archived tissue and has a documented sensitivity and specificity of 100% and 79%, respectively, as reported by Singhi and Westra in 2010.⁶ HPV DNA fluorescence in situ hybridization is the gold standard for determining the presence of specific types of HPV DNA; however, p16 IHC can serve as a rapid, less costly means of studying archived tissue, lending its utility to retrospective population-based studies.

METHODS

A retrospective study was designed to determine the proportion of HPV-associated oropharyngeal SCC in a US Department of Veterans Affairs (VA) population, using p16 antigen IHC on paraffin-embedded tissue as the surrogate marker for the presence of HPV infection. Patients consisted of veterans who were treated for oropharyngeal SCC at Veterans Affairs Memphis Healthcare System (VAMHS) in Tennessee between January 1, 2000, and December 31, 2008. This data range allowed for at least 5 years of follow-up. Patients were excluded who lacked enough tissue specimens for analysis. Measurement outcomes included p16 expression, with subset analysis by race and ethnicity, degree of tobacco and alcohol use, tumor location, stage, age at diagnosis, and survival outcome. Microsoft Excel was used to calculate Fisher exact test, Student t test, and χ² statistics. Significance was set at P < .05. This study received institutional review board approval from the University of Tennessee Health Science Center and the VAMHS.

RESULTS

We identified 66 total cases of oropharyngeal SCC; 19 cases (29%) were positive for p16. The mean age at diagnosis for the p16-positive cohort was 59 years vs 61 years for the p16-negative cohort (P = .22; Table 1).

Of the patients with p16-positive status, most were White veterans compared with those with p16-negative status, which consisted of more African Americans veterans (P = .04). Smoking data were available for 18 of 19 patients with p16-positive status and 46 of 47 patients whose status was p16 negative. Four patients (22%) with p16-positive status were tobacco naïve compared with none of the patients with p16-negative status (P = .005). Alcohol use data were available for 17 of 19 patients in the p16-positive cohort and 46 of 47 patients in the p16-negative cohort. Three patients (18%) with p16-positive status were alcohol naïve compared with 2 patients (4%) with p16-negative status (P = .12). Of the patients in the study, 65 of 66 died during the study period, 5 (28%) of the p16-positive cohort and 17 (36%) of the p16-negative cohort were directly attributed to oropharyngeal SCC (P = .52).

Although the tonsil was the most common site of tumor origin in both the p16-positive and negative cohorts (63% vs 51%, respectively), our analysis showed no statistically significant difference in sites of origin (P = .69) (Table 2).

There was no difference in the stage at presentation between the 2 cohorts, with the most presenting with stage III or IV disease (P = .75).

 

DISCUSSION

The VAMHS population in our study had a lower proportion of HPV-associated oropharyngeal SCC compared with studies on nonveteran populations (29% vs 40%-80%, respectively).^5,6 This disparity may indicate a true difference in these populations or may be related to a decreased prevalence of HPV infection in the population served by the VAMHS. This single-institution population did not completely correlate with previous population studies. Specifically, age at presentation (equivalent to patients with p16-negative status rather than earlier age at onset), disease stage at presentation (lower stage for patients with p16-positive status), and disease-specific survival (not improved compared with patients with p16-negative status in other studies) were dissimilar to previous investigations.^2,3

The increased age and staging at presentation could be related in these patients with p16-positive status, which may further account for the lack of improved survival. Furthermore, both groups tended to use alcohol at a high proportion; whereas other populations have had a lesser degree of alcohol intake with p16 positivity.^1-4 These differences may be due to variations in the habits and behavior of VA patients compared with non-VA patients.^3,4

HPV-associated oropharyngeal SCC in published data has been associated with high-risk sexual behavior, lower age, and less tobacco and alcohol use.^5,6 No difference was noted in tumor site predilection; however, the small size of our study could explain the lack of finding site preference shown in previous studies.^2,3Other veteran-specific factors are absent in the at-large population, such as Agent Orange exposure. More than 8 million veterans (22%) from the Vietnam era self-reported Agent Orange exposure.⁷ Agent Orange exposure significantly predicted developing upper aerodigestive tract cancer. Oropharyngeal, nasopharyngeal, laryngeal, and thyroid cancers were significantly associated with Agent Orange exposure. Interestingly, these patients experienced an improved 10-year survival rate compared with patients not exposed to Agent Orange. This finding contrasts with our patients, who did not experience improved outcomes vs nonveteran patients with head and neck cancer.⁷

Suicide in veterans with head and neck cancer has been evaluated and was found at an incidence of 0.7%. Survivors of head and neck cancer are almost twice as likely to die by suicide compared with other cancer survivors. These patients have a higher rate of mental health disorders, substance misuse, and use of palliative care services.⁸ Sixty-five of 66 of our patients died during the 5-year observation period, although none died by suicide.

In a 2022 cohort study by Sun and colleagues, upfront surgical treatment was associated with a 23% reduced risk of stroke compared with definitive chemoradiotherapy in US veterans with oropharyngeal carcinoma.⁹ In our study, 58 of 66 patients (88%) received concurrent chemoradiation, possibly reflecting the more advanced stage of diagnosis in our study population. This was due to comorbidities and other health and economic factors. In our study, 43 patients (65%) died of factors not related to the disease, reflecting the overall comorbidity burden of this population. Seven patients (11%) in our 5-year study died of a documented stroke. In the study of veterans by Sun and colleagues, the 10-year cumulative incidence of stroke was 12.5% and death was 57.3%.⁹ Our veteran population experienced a similar incidence of strokes. These findings may need to be included when discussing the risk-benefit aspects of different treatment options with our veteran patients with oropharyngeal cancer.

To understand the influence of HPV infection on the course of oropharyngeal SCC in the VA patient population and to apply this understanding to future individualized treatment paradigms, this study can be expanded to a greater number of VA patients. p16 immunoexpression appears to be a useful surrogate for high-risk HPV infection in oropharyngeal SCC, and its ease of use supports its feasibility in further VA population analysis.¹⁰ While realizing that the veteran HPV-associated oropharyngeal SCC population differs from the civilian HPV-associated oropharyngeal SCC population, we also have realized that other unique considerations in the veteran population, such as chemical warfare exposure, mental illness, and vascular disease, complicate treatment decisions in these patients.

CONCLUSIONS

Disparities in racial distribution and tobacco use between patients with p16-positive and p16-negative status are similar to those reported in non-VA populations. In contrast, the frequently reported younger age at presentation and better disease outcomes seen in non-VA patients were not observed, perhaps due to the lower percentage of p16 expression in VA patients with oropharyngeal SCC. Whereas de-intensification of therapy may be considered for many patients with oropharygeal cancer that is HPV-associated because of improved prognosis, this approach should be undertaken with great care in this group of patients. Personalization of therapy for these HPV-associated oropharyngeal SCC in the veteran population must be adapted to mitigate this critical disparity.

Febrile neutropenia (FN) frequently occurs in patients receiving chemotherapy, with the greatest risk of complications occurring in those who experience profound and prolonged neutropenia. Although granulocyte colony-stimulating factor (G-CSF) prophylaxis has been shown to reduce the risk and duration of chemotherapy-induced neutropenia and FN, there is no well-established optimal regimen.¹ The 2022 National Comprehensive Cancer Network guidelines for hematopoietic growth factors recommend prophylaxis with G-CSF in at-risk patients receiving chemotherapy, specifically in chemotherapy regimens considered high risk for FN (incidence > 20%) or intermediate risk for FN (incidence 10%-20%) with additional patient risk factors.² The incidence of developing FN with at least 1 chemotherapy cycle is estimated at 10% to 50% of patients with solid tumors and > 80% of patients with hematologic malignancies.³ The rate of major complications (eg, hypotension, acute renal, respiratory, or heart failure) in the context of FN is 25% to 30%, and mortality is reported up to 11% in this population.⁴

Because of the significant consequences of neutropenia and FN, prevention is imperative due to the increase in morbidity and mortality, including chemotherapy delays, increased hospitalizations, chemotherapy dose reductions, and discontinuations that cause delays in care.⁵ In patients with curative malignancies, these consequences can negatively impact treatment efficacy and overall survival. Additionally, infections occur in 20% to 30% of patients with febrile episodes. Although fever is often the only clinical sign or symptom of infection, patients who are profoundly neutropenic may present with suspected infection and be afebrile or hypothermic.³

For filgrastim, the National Comprehensive Cancer Network guidelines do not specify the total days of required injections but state that a daily dose should be given until the postnadir absolute neutrophil count (ANC) recovers to normal or near normal levels by laboratory standards.² It is uncommon in clinical practice to track postnadir ANCs due to frequent laboratory monitoring. Clinical trial data suggest an average duration of 11 days of daily filgrastim injections for ANC recovery; however, real-world data exist supporting a range from 4 to 10 days with a median of 7 injections per cycle for prevention of neutropenia or FN.^6,7

At the South Texas Veterans Health Care System in San Antonio, daily filgrastim injections are preferred due to cost; patients typically receive a 7-day course for primary prophylaxis for FN. In our study, we aimed to determine the outcomes in patients receiving daily filgrastim injections with a curative cancer diagnosis and a chemotherapy regimen with either high risk for FN, or a chemotherapy regimen with an intermediate risk for FN and additional patient risk factors. Before the initiation of data collection, this study was reviewed and determined to be exempt by the University of Texas Health Science Center at San Antonio Institutional Review Board.

METHODS

Electronic health record reviews at the South Texas Veterans Health Care System were performed to identify patients who received filgrastim primary prophylaxis (defined as filgrastim, tbo-filgrastim, or filgrastim-sndz) for a curative cancer diagnosis. Primary prophylaxis refers to the administration of G-CSF in the first cycle of chemotherapy before the onset of neutropenia. Patients received filgrastim prophylaxis if they were undergoing treatment with a chemotherapy regimen with either high risk for FN or a chemotherapy regimen with an intermediate risk for FN and additional patient risk factors. Risk factors for patients included prior chemotherapy or radiation therapy; persistent neutropenia; bone marrow involvement by tumor; recent surgery and/or open wounds; liver dysfunction (defined as total bilirubin > 2 mg/dL); renal dysfunction (defined as creatinine clearance < 50 mL/min); and those aged > 65 years receiving full chemotherapy dose intensity. Neutropenia is defined as a decrease in ANC < 1000 neutrophils/μL, whereas FN is defined as a single temperature of > 38.3 °C or > 38.0 °C for longer than 1 hour with < 500 neutrophils/μL or < 1000 neutrophils/μL predicted to decline to < 500 neutrophils/μL over the next 48 hours. All patients had their filgrastim dispensed for home administration during their chemotherapy appointment.

 

Patients were included if they received filgrastim for primary prophylaxis of FN with a curative cancer diagnosis. Patients receiving salvage chemotherapy for hematologic malignancies with intent to proceed to curative transplant were also included. Bone marrow involvement of the tumor is a FN risk factor. Only patients with hematologic malignancies and bone marrow involvement were included. Patients wereexcluded if they received filgrastim for secondary prophylaxis of neutropenia or FN, a noncurative cancer diagnosis, stem cell transplant mobilization and engraftment, or nononcologic neutropenia.

The primary outcome for this study was the incidence of neutropenia or FN leading to treatment delays despite the use of primary prophylaxis with filgrastim. A dose delay was defined as a delay of planned chemotherapy by ≥ 3 days. Secondary outcomes included chemotherapy dose decreases or discontinuations, hospitalizations, days of hospitalization, infections, extended duration of filgrastim, and transitions to pegfilgrastim due to neutropenia or FN. Documented infections were defined in patients with a positive culture, laboratory testing, or imaging consistent with infection. Extended durations of filgrastim or transitions to pegfilgrastim were patient specific and upon clinician discretion.

Descriptive statistics were used to summarize the study population and their health outcomes. Fisher exact test was used to compare FN incidence for high- and intermediate-risk FN groups.

RESULTS

Between September 1, 2015, and September 24, 2020, 381 patients received filgrastim. Of these patients, 59 met the inclusion criteria. Patients receiving filgrastim were excluded due to stem cell transplant mobilization/engraftment (n = 145), a noncurative cancer diagnosis (n = 134), use as a secondary prophylaxis (n = 33), and nononcologic neutropenia (n = 8). Additionally, 2 patients initially received pegfilgrastim and were not included in this data set.

The median (IQR) age was 64 (55-70) years and 42 patients (71%) were male (Table 1).

Thirty patients (51%) had non-Hodgkin lymphoma and 19 (32%) had breast cancer. There were 33 patients (56%) with high-risk chemotherapy regimens and 26 (44%) with an intermediate-risk regimen. Overall, 48 patients (81%) received 7 or 10 days. and 11 patients (19%) received 5 days of filgrastim therapy.

Ten patients (17%) experienced dose delays despite filgrastim use (Table 2).

This included 7 patients (21%) in the high risk for FN group and 3 patients (12%) in the intermediate risk for FN group (P = .49) (Figure).

Additionally, 15 patients (25%) were hospitalized with either neutropenia or FN despite filgrastim use. This included 11 patients (33%) in the high risk for FN group and 4 patients (15%) in the intermediate risk for FN group (P = .14). The median (IQR) duration of hospitalization was 5 (4-7) days. Two patients with acute lymphocytic leukemia and acute myeloid leukemia on regimens deemed high risk for FN had multiple hospitalizations despite filgrastim use and were hospitalized for a total of 16 and 17 days, respectively. Both transitioned to pegfilgrastim after their subsequent hospitalizations with successful continuation of treatment.

Nine patients (15%) had the number of filgrastim injections per chemotherapy cycle extended due to various reasons. Five patients required extended days after hospitalization for FN, 3 patients for dose delays due to neutropenia with the previous cycle, and 1 patient with an undocumented reason outside of the prespecified outcomes. Two of these patients experienced continued neutropenia and dose delays after extending filgrastim from 5 to 7 days or 7 to 10 days. One patient who experienced continued neutropenia after extending filgrastim to 10 days was subsequently transitioned to pegfilgrastim without further episodes of neutropenia. The other patient who still experienced neutropenia after extending filgrastim to 7 days was receiving the last chemotherapy cycle and did not require subsequent doses of filgrastim.

Two additional patients were not included in the hospitalizations. The first was a patient on a chemotherapy regimen with a high risk for FN who presented to the emergency department with documented FN but was never admitted since the patient elected to not be hospitalized. This patient developed oral, anal, and vaginal candidiasis, and it was noted by the oncologist at the next clinic visit that this was likely secondary to grade 4 neutropenia (ANC < 500 neutrophils/μL). The second was a patient on a chemotherapy regimen with an intermediate risk for FN who was already hospitalized but had developed FN and sepsis despite filgrastim use.

Finally, out of the hospitalized patients, 9 (15%) had infections. This included 6 patients (18%) in the high risk for FN group and 3 patients (12%) in the intermediate risk for FN group (P = .72). Six patients transitioned to pegfilgrastim for hospitalization, 2 for neutropenia, and 1 for an unspecified reason. Nine patients (15%) who received filgrastim ended up transitioning to pegfilgrastim; 6 (67%) of these patients were transitioned due to hospitalization for FN. Of all the patients who transitioned to pegfilgrastim, 1 patient on a high risk for FN regimen developed sepsis due to herpes zoster in the setting of neutropenia after the previous cycle of chemotherapy.

 

DISCUSSION

Real-world data are limited regarding G-CSF practice patterns; however, available data demonstrate patients may receive suboptimal treatment courses of filgrastim leading to increased complications associated with neutropenia and FN, such as dose delays and hospitalizations.^8,9 At the South Texas Veterans Health Care System, 48 patients (81%) received a filgrastim course of ≥ 7 days as an initial course for primary prophylaxis. Multivariate analyses performed by Weycker and colleagues described a decreased risk of hospitalization for neutropenia or FN with each additional day of filgrastim prophylaxis; however, such analysis could not be performed in our data set due to the small sample size.⁸ In this review, 10 patients (17%) experienced treatment delays due to neutropenia or FN, mirroring previously published data. The hospitalization rate of 25% is higher than the published incidence of 5.2% of cancer-related hospitalizations among adults.^7,10 This difference may be explained by a difference in health care access for the veteran population.

As an alternative to daily filgrastim injections, the National Comprehensive Cancer Network also recommends a single dose of pegfilgrastim for primary prevention of FN. Efficacy benefits of pegfilgrastim use include increased patient adherence due to a single injection, a reduction in FN incidence and FN-related hospitalizations, and improved time to ANC recovery compared with filgrastim.¹¹ There are reports suggesting pegfilgrastim significantly reduces neutropenia and FN incidence to a greater extent compared with daily filgrastim injections.⁶ In patients with breast cancer receiving dose-dense adjuvant chemotherapy, there are data demonstrating that patients who received filgrastim were more likely to experience severe neutropenia, dose reductions, and treatment delays leading to lower dose density compared with pegfilgrastim.¹² Of the 19 patients with breast cancer included in our population, 26% experienced one of the previously described outcomes leading to either extensions of daily filgrastim injections or transitions to pegfilgrastim to successfully maintain dose density. In patients with acute myeloid leukemia receiving consolidation chemotherapy, filgrastim was found to be associated with a statistically significant increased risk of hospitalizations compared with pegfilgrastim.¹³ The one patient with acute myeloid leukemia included in our study did not require additional hospitalizations for neutropenia or FN after transitioning to pegfilgrastim.

Given the cost advantage, the South Texas Veterans Health Care System continues to prefer daily filgrastim injections. A recent survey demonstrated that 73% of patients at 23 sites in the Veterans Health Administration used filgrastim rather than pegfilgrastim for cost savings, although it is recognized that daily filgrastim injections are less convenient for patients.¹⁴ This analysis did not review costs associated with hospitalization for FN or the appropriateness of G-CSF use. Cancer-related neutropenia accounts for 8.3% of all cancer-related hospitalization costs among adults; the average hospitalization costs nearly $25,000 per stay and about $2.3 billion among adult patients with cancer annually.^10,15

Limitations

This study has limitations that affected the applicability and interpretation of the results. This included the study design since it was a retrospective, single-center, descriptive cohort study. Patient adherence to daily filgrastim injections could not be assessed due to the retrospective nature of the study. The small sample size of 59 patients was prohibitive for utilization of additional analytical tools. Additionally, the predominately male veteran population may make applicability to non-VA populations restrictive.

 

CONCLUSIONS

Based on the incidence of primary and secondary outcomes associated with using daily filgrastim injections as primary prophylaxis in this study, additional measures such as tracking postnadir ANCs should be performed to ensure patients receive an appropriate number of filgrastim doses to prevent complications associated with neutropenia.

Acknowledgments

We thank Eric Dougherty, PharmD, for assistance in producing granulocyte colony-stimulating factor data.

Febrile neutropenia (FN) frequently occurs in patients receiving chemotherapy, with the greatest risk of complications occurring in those who experience profound and prolonged neutropenia. Although granulocyte colony-stimulating factor (G-CSF) prophylaxis has been shown to reduce the risk and duration of chemotherapy-induced neutropenia and FN, there is no well-established optimal regimen.¹ The 2022 National Comprehensive Cancer Network guidelines for hematopoietic growth factors recommend prophylaxis with G-CSF in at-risk patients receiving chemotherapy, specifically in chemotherapy regimens considered high risk for FN (incidence > 20%) or intermediate risk for FN (incidence 10%-20%) with additional patient risk factors.² The incidence of developing FN with at least 1 chemotherapy cycle is estimated at 10% to 50% of patients with solid tumors and > 80% of patients with hematologic malignancies.³ The rate of major complications (eg, hypotension, acute renal, respiratory, or heart failure) in the context of FN is 25% to 30%, and mortality is reported up to 11% in this population.⁴

Because of the significant consequences of neutropenia and FN, prevention is imperative due to the increase in morbidity and mortality, including chemotherapy delays, increased hospitalizations, chemotherapy dose reductions, and discontinuations that cause delays in care.⁵ In patients with curative malignancies, these consequences can negatively impact treatment efficacy and overall survival. Additionally, infections occur in 20% to 30% of patients with febrile episodes. Although fever is often the only clinical sign or symptom of infection, patients who are profoundly neutropenic may present with suspected infection and be afebrile or hypothermic.³

For filgrastim, the National Comprehensive Cancer Network guidelines do not specify the total days of required injections but state that a daily dose should be given until the postnadir absolute neutrophil count (ANC) recovers to normal or near normal levels by laboratory standards.² It is uncommon in clinical practice to track postnadir ANCs due to frequent laboratory monitoring. Clinical trial data suggest an average duration of 11 days of daily filgrastim injections for ANC recovery; however, real-world data exist supporting a range from 4 to 10 days with a median of 7 injections per cycle for prevention of neutropenia or FN.^6,7

At the South Texas Veterans Health Care System in San Antonio, daily filgrastim injections are preferred due to cost; patients typically receive a 7-day course for primary prophylaxis for FN. In our study, we aimed to determine the outcomes in patients receiving daily filgrastim injections with a curative cancer diagnosis and a chemotherapy regimen with either high risk for FN, or a chemotherapy regimen with an intermediate risk for FN and additional patient risk factors. Before the initiation of data collection, this study was reviewed and determined to be exempt by the University of Texas Health Science Center at San Antonio Institutional Review Board.

METHODS

Electronic health record reviews at the South Texas Veterans Health Care System were performed to identify patients who received filgrastim primary prophylaxis (defined as filgrastim, tbo-filgrastim, or filgrastim-sndz) for a curative cancer diagnosis. Primary prophylaxis refers to the administration of G-CSF in the first cycle of chemotherapy before the onset of neutropenia. Patients received filgrastim prophylaxis if they were undergoing treatment with a chemotherapy regimen with either high risk for FN or a chemotherapy regimen with an intermediate risk for FN and additional patient risk factors. Risk factors for patients included prior chemotherapy or radiation therapy; persistent neutropenia; bone marrow involvement by tumor; recent surgery and/or open wounds; liver dysfunction (defined as total bilirubin > 2 mg/dL); renal dysfunction (defined as creatinine clearance < 50 mL/min); and those aged > 65 years receiving full chemotherapy dose intensity. Neutropenia is defined as a decrease in ANC < 1000 neutrophils/μL, whereas FN is defined as a single temperature of > 38.3 °C or > 38.0 °C for longer than 1 hour with < 500 neutrophils/μL or < 1000 neutrophils/μL predicted to decline to < 500 neutrophils/μL over the next 48 hours. All patients had their filgrastim dispensed for home administration during their chemotherapy appointment.

 

Patients were included if they received filgrastim for primary prophylaxis of FN with a curative cancer diagnosis. Patients receiving salvage chemotherapy for hematologic malignancies with intent to proceed to curative transplant were also included. Bone marrow involvement of the tumor is a FN risk factor. Only patients with hematologic malignancies and bone marrow involvement were included. Patients wereexcluded if they received filgrastim for secondary prophylaxis of neutropenia or FN, a noncurative cancer diagnosis, stem cell transplant mobilization and engraftment, or nononcologic neutropenia.

The primary outcome for this study was the incidence of neutropenia or FN leading to treatment delays despite the use of primary prophylaxis with filgrastim. A dose delay was defined as a delay of planned chemotherapy by ≥ 3 days. Secondary outcomes included chemotherapy dose decreases or discontinuations, hospitalizations, days of hospitalization, infections, extended duration of filgrastim, and transitions to pegfilgrastim due to neutropenia or FN. Documented infections were defined in patients with a positive culture, laboratory testing, or imaging consistent with infection. Extended durations of filgrastim or transitions to pegfilgrastim were patient specific and upon clinician discretion.

Descriptive statistics were used to summarize the study population and their health outcomes. Fisher exact test was used to compare FN incidence for high- and intermediate-risk FN groups.

RESULTS

Between September 1, 2015, and September 24, 2020, 381 patients received filgrastim. Of these patients, 59 met the inclusion criteria. Patients receiving filgrastim were excluded due to stem cell transplant mobilization/engraftment (n = 145), a noncurative cancer diagnosis (n = 134), use as a secondary prophylaxis (n = 33), and nononcologic neutropenia (n = 8). Additionally, 2 patients initially received pegfilgrastim and were not included in this data set.

The median (IQR) age was 64 (55-70) years and 42 patients (71%) were male (Table 1).

Thirty patients (51%) had non-Hodgkin lymphoma and 19 (32%) had breast cancer. There were 33 patients (56%) with high-risk chemotherapy regimens and 26 (44%) with an intermediate-risk regimen. Overall, 48 patients (81%) received 7 or 10 days. and 11 patients (19%) received 5 days of filgrastim therapy.

Ten patients (17%) experienced dose delays despite filgrastim use (Table 2).

This included 7 patients (21%) in the high risk for FN group and 3 patients (12%) in the intermediate risk for FN group (P = .49) (Figure).

Additionally, 15 patients (25%) were hospitalized with either neutropenia or FN despite filgrastim use. This included 11 patients (33%) in the high risk for FN group and 4 patients (15%) in the intermediate risk for FN group (P = .14). The median (IQR) duration of hospitalization was 5 (4-7) days. Two patients with acute lymphocytic leukemia and acute myeloid leukemia on regimens deemed high risk for FN had multiple hospitalizations despite filgrastim use and were hospitalized for a total of 16 and 17 days, respectively. Both transitioned to pegfilgrastim after their subsequent hospitalizations with successful continuation of treatment.

Nine patients (15%) had the number of filgrastim injections per chemotherapy cycle extended due to various reasons. Five patients required extended days after hospitalization for FN, 3 patients for dose delays due to neutropenia with the previous cycle, and 1 patient with an undocumented reason outside of the prespecified outcomes. Two of these patients experienced continued neutropenia and dose delays after extending filgrastim from 5 to 7 days or 7 to 10 days. One patient who experienced continued neutropenia after extending filgrastim to 10 days was subsequently transitioned to pegfilgrastim without further episodes of neutropenia. The other patient who still experienced neutropenia after extending filgrastim to 7 days was receiving the last chemotherapy cycle and did not require subsequent doses of filgrastim.

Two additional patients were not included in the hospitalizations. The first was a patient on a chemotherapy regimen with a high risk for FN who presented to the emergency department with documented FN but was never admitted since the patient elected to not be hospitalized. This patient developed oral, anal, and vaginal candidiasis, and it was noted by the oncologist at the next clinic visit that this was likely secondary to grade 4 neutropenia (ANC < 500 neutrophils/μL). The second was a patient on a chemotherapy regimen with an intermediate risk for FN who was already hospitalized but had developed FN and sepsis despite filgrastim use.

Finally, out of the hospitalized patients, 9 (15%) had infections. This included 6 patients (18%) in the high risk for FN group and 3 patients (12%) in the intermediate risk for FN group (P = .72). Six patients transitioned to pegfilgrastim for hospitalization, 2 for neutropenia, and 1 for an unspecified reason. Nine patients (15%) who received filgrastim ended up transitioning to pegfilgrastim; 6 (67%) of these patients were transitioned due to hospitalization for FN. Of all the patients who transitioned to pegfilgrastim, 1 patient on a high risk for FN regimen developed sepsis due to herpes zoster in the setting of neutropenia after the previous cycle of chemotherapy.

 

DISCUSSION

Real-world data are limited regarding G-CSF practice patterns; however, available data demonstrate patients may receive suboptimal treatment courses of filgrastim leading to increased complications associated with neutropenia and FN, such as dose delays and hospitalizations.^8,9 At the South Texas Veterans Health Care System, 48 patients (81%) received a filgrastim course of ≥ 7 days as an initial course for primary prophylaxis. Multivariate analyses performed by Weycker and colleagues described a decreased risk of hospitalization for neutropenia or FN with each additional day of filgrastim prophylaxis; however, such analysis could not be performed in our data set due to the small sample size.⁸ In this review, 10 patients (17%) experienced treatment delays due to neutropenia or FN, mirroring previously published data. The hospitalization rate of 25% is higher than the published incidence of 5.2% of cancer-related hospitalizations among adults.^7,10 This difference may be explained by a difference in health care access for the veteran population.

As an alternative to daily filgrastim injections, the National Comprehensive Cancer Network also recommends a single dose of pegfilgrastim for primary prevention of FN. Efficacy benefits of pegfilgrastim use include increased patient adherence due to a single injection, a reduction in FN incidence and FN-related hospitalizations, and improved time to ANC recovery compared with filgrastim.¹¹ There are reports suggesting pegfilgrastim significantly reduces neutropenia and FN incidence to a greater extent compared with daily filgrastim injections.⁶ In patients with breast cancer receiving dose-dense adjuvant chemotherapy, there are data demonstrating that patients who received filgrastim were more likely to experience severe neutropenia, dose reductions, and treatment delays leading to lower dose density compared with pegfilgrastim.¹² Of the 19 patients with breast cancer included in our population, 26% experienced one of the previously described outcomes leading to either extensions of daily filgrastim injections or transitions to pegfilgrastim to successfully maintain dose density. In patients with acute myeloid leukemia receiving consolidation chemotherapy, filgrastim was found to be associated with a statistically significant increased risk of hospitalizations compared with pegfilgrastim.¹³ The one patient with acute myeloid leukemia included in our study did not require additional hospitalizations for neutropenia or FN after transitioning to pegfilgrastim.

Given the cost advantage, the South Texas Veterans Health Care System continues to prefer daily filgrastim injections. A recent survey demonstrated that 73% of patients at 23 sites in the Veterans Health Administration used filgrastim rather than pegfilgrastim for cost savings, although it is recognized that daily filgrastim injections are less convenient for patients.¹⁴ This analysis did not review costs associated with hospitalization for FN or the appropriateness of G-CSF use. Cancer-related neutropenia accounts for 8.3% of all cancer-related hospitalization costs among adults; the average hospitalization costs nearly $25,000 per stay and about $2.3 billion among adult patients with cancer annually.^10,15

Limitations

This study has limitations that affected the applicability and interpretation of the results. This included the study design since it was a retrospective, single-center, descriptive cohort study. Patient adherence to daily filgrastim injections could not be assessed due to the retrospective nature of the study. The small sample size of 59 patients was prohibitive for utilization of additional analytical tools. Additionally, the predominately male veteran population may make applicability to non-VA populations restrictive.

 

CONCLUSIONS

Based on the incidence of primary and secondary outcomes associated with using daily filgrastim injections as primary prophylaxis in this study, additional measures such as tracking postnadir ANCs should be performed to ensure patients receive an appropriate number of filgrastim doses to prevent complications associated with neutropenia.

Acknowledgments

We thank Eric Dougherty, PharmD, for assistance in producing granulocyte colony-stimulating factor data.

Febrile neutropenia (FN) frequently occurs in patients receiving chemotherapy, with the greatest risk of complications occurring in those who experience profound and prolonged neutropenia. Although granulocyte colony-stimulating factor (G-CSF) prophylaxis has been shown to reduce the risk and duration of chemotherapy-induced neutropenia and FN, there is no well-established optimal regimen.¹ The 2022 National Comprehensive Cancer Network guidelines for hematopoietic growth factors recommend prophylaxis with G-CSF in at-risk patients receiving chemotherapy, specifically in chemotherapy regimens considered high risk for FN (incidence > 20%) or intermediate risk for FN (incidence 10%-20%) with additional patient risk factors.² The incidence of developing FN with at least 1 chemotherapy cycle is estimated at 10% to 50% of patients with solid tumors and > 80% of patients with hematologic malignancies.³ The rate of major complications (eg, hypotension, acute renal, respiratory, or heart failure) in the context of FN is 25% to 30%, and mortality is reported up to 11% in this population.⁴

Because of the significant consequences of neutropenia and FN, prevention is imperative due to the increase in morbidity and mortality, including chemotherapy delays, increased hospitalizations, chemotherapy dose reductions, and discontinuations that cause delays in care.⁵ In patients with curative malignancies, these consequences can negatively impact treatment efficacy and overall survival. Additionally, infections occur in 20% to 30% of patients with febrile episodes. Although fever is often the only clinical sign or symptom of infection, patients who are profoundly neutropenic may present with suspected infection and be afebrile or hypothermic.³

For filgrastim, the National Comprehensive Cancer Network guidelines do not specify the total days of required injections but state that a daily dose should be given until the postnadir absolute neutrophil count (ANC) recovers to normal or near normal levels by laboratory standards.² It is uncommon in clinical practice to track postnadir ANCs due to frequent laboratory monitoring. Clinical trial data suggest an average duration of 11 days of daily filgrastim injections for ANC recovery; however, real-world data exist supporting a range from 4 to 10 days with a median of 7 injections per cycle for prevention of neutropenia or FN.^6,7

At the South Texas Veterans Health Care System in San Antonio, daily filgrastim injections are preferred due to cost; patients typically receive a 7-day course for primary prophylaxis for FN. In our study, we aimed to determine the outcomes in patients receiving daily filgrastim injections with a curative cancer diagnosis and a chemotherapy regimen with either high risk for FN, or a chemotherapy regimen with an intermediate risk for FN and additional patient risk factors. Before the initiation of data collection, this study was reviewed and determined to be exempt by the University of Texas Health Science Center at San Antonio Institutional Review Board.

METHODS

Electronic health record reviews at the South Texas Veterans Health Care System were performed to identify patients who received filgrastim primary prophylaxis (defined as filgrastim, tbo-filgrastim, or filgrastim-sndz) for a curative cancer diagnosis. Primary prophylaxis refers to the administration of G-CSF in the first cycle of chemotherapy before the onset of neutropenia. Patients received filgrastim prophylaxis if they were undergoing treatment with a chemotherapy regimen with either high risk for FN or a chemotherapy regimen with an intermediate risk for FN and additional patient risk factors. Risk factors for patients included prior chemotherapy or radiation therapy; persistent neutropenia; bone marrow involvement by tumor; recent surgery and/or open wounds; liver dysfunction (defined as total bilirubin > 2 mg/dL); renal dysfunction (defined as creatinine clearance < 50 mL/min); and those aged > 65 years receiving full chemotherapy dose intensity. Neutropenia is defined as a decrease in ANC < 1000 neutrophils/μL, whereas FN is defined as a single temperature of > 38.3 °C or > 38.0 °C for longer than 1 hour with < 500 neutrophils/μL or < 1000 neutrophils/μL predicted to decline to < 500 neutrophils/μL over the next 48 hours. All patients had their filgrastim dispensed for home administration during their chemotherapy appointment.

 

Patients were included if they received filgrastim for primary prophylaxis of FN with a curative cancer diagnosis. Patients receiving salvage chemotherapy for hematologic malignancies with intent to proceed to curative transplant were also included. Bone marrow involvement of the tumor is a FN risk factor. Only patients with hematologic malignancies and bone marrow involvement were included. Patients wereexcluded if they received filgrastim for secondary prophylaxis of neutropenia or FN, a noncurative cancer diagnosis, stem cell transplant mobilization and engraftment, or nononcologic neutropenia.

The primary outcome for this study was the incidence of neutropenia or FN leading to treatment delays despite the use of primary prophylaxis with filgrastim. A dose delay was defined as a delay of planned chemotherapy by ≥ 3 days. Secondary outcomes included chemotherapy dose decreases or discontinuations, hospitalizations, days of hospitalization, infections, extended duration of filgrastim, and transitions to pegfilgrastim due to neutropenia or FN. Documented infections were defined in patients with a positive culture, laboratory testing, or imaging consistent with infection. Extended durations of filgrastim or transitions to pegfilgrastim were patient specific and upon clinician discretion.

Descriptive statistics were used to summarize the study population and their health outcomes. Fisher exact test was used to compare FN incidence for high- and intermediate-risk FN groups.

RESULTS

Between September 1, 2015, and September 24, 2020, 381 patients received filgrastim. Of these patients, 59 met the inclusion criteria. Patients receiving filgrastim were excluded due to stem cell transplant mobilization/engraftment (n = 145), a noncurative cancer diagnosis (n = 134), use as a secondary prophylaxis (n = 33), and nononcologic neutropenia (n = 8). Additionally, 2 patients initially received pegfilgrastim and were not included in this data set.

The median (IQR) age was 64 (55-70) years and 42 patients (71%) were male (Table 1).

Thirty patients (51%) had non-Hodgkin lymphoma and 19 (32%) had breast cancer. There were 33 patients (56%) with high-risk chemotherapy regimens and 26 (44%) with an intermediate-risk regimen. Overall, 48 patients (81%) received 7 or 10 days. and 11 patients (19%) received 5 days of filgrastim therapy.

Ten patients (17%) experienced dose delays despite filgrastim use (Table 2).

This included 7 patients (21%) in the high risk for FN group and 3 patients (12%) in the intermediate risk for FN group (P = .49) (Figure).

Additionally, 15 patients (25%) were hospitalized with either neutropenia or FN despite filgrastim use. This included 11 patients (33%) in the high risk for FN group and 4 patients (15%) in the intermediate risk for FN group (P = .14). The median (IQR) duration of hospitalization was 5 (4-7) days. Two patients with acute lymphocytic leukemia and acute myeloid leukemia on regimens deemed high risk for FN had multiple hospitalizations despite filgrastim use and were hospitalized for a total of 16 and 17 days, respectively. Both transitioned to pegfilgrastim after their subsequent hospitalizations with successful continuation of treatment.

Nine patients (15%) had the number of filgrastim injections per chemotherapy cycle extended due to various reasons. Five patients required extended days after hospitalization for FN, 3 patients for dose delays due to neutropenia with the previous cycle, and 1 patient with an undocumented reason outside of the prespecified outcomes. Two of these patients experienced continued neutropenia and dose delays after extending filgrastim from 5 to 7 days or 7 to 10 days. One patient who experienced continued neutropenia after extending filgrastim to 10 days was subsequently transitioned to pegfilgrastim without further episodes of neutropenia. The other patient who still experienced neutropenia after extending filgrastim to 7 days was receiving the last chemotherapy cycle and did not require subsequent doses of filgrastim.

Two additional patients were not included in the hospitalizations. The first was a patient on a chemotherapy regimen with a high risk for FN who presented to the emergency department with documented FN but was never admitted since the patient elected to not be hospitalized. This patient developed oral, anal, and vaginal candidiasis, and it was noted by the oncologist at the next clinic visit that this was likely secondary to grade 4 neutropenia (ANC < 500 neutrophils/μL). The second was a patient on a chemotherapy regimen with an intermediate risk for FN who was already hospitalized but had developed FN and sepsis despite filgrastim use.

Finally, out of the hospitalized patients, 9 (15%) had infections. This included 6 patients (18%) in the high risk for FN group and 3 patients (12%) in the intermediate risk for FN group (P = .72). Six patients transitioned to pegfilgrastim for hospitalization, 2 for neutropenia, and 1 for an unspecified reason. Nine patients (15%) who received filgrastim ended up transitioning to pegfilgrastim; 6 (67%) of these patients were transitioned due to hospitalization for FN. Of all the patients who transitioned to pegfilgrastim, 1 patient on a high risk for FN regimen developed sepsis due to herpes zoster in the setting of neutropenia after the previous cycle of chemotherapy.

 

DISCUSSION

Real-world data are limited regarding G-CSF practice patterns; however, available data demonstrate patients may receive suboptimal treatment courses of filgrastim leading to increased complications associated with neutropenia and FN, such as dose delays and hospitalizations.^8,9 At the South Texas Veterans Health Care System, 48 patients (81%) received a filgrastim course of ≥ 7 days as an initial course for primary prophylaxis. Multivariate analyses performed by Weycker and colleagues described a decreased risk of hospitalization for neutropenia or FN with each additional day of filgrastim prophylaxis; however, such analysis could not be performed in our data set due to the small sample size.⁸ In this review, 10 patients (17%) experienced treatment delays due to neutropenia or FN, mirroring previously published data. The hospitalization rate of 25% is higher than the published incidence of 5.2% of cancer-related hospitalizations among adults.^7,10 This difference may be explained by a difference in health care access for the veteran population.

As an alternative to daily filgrastim injections, the National Comprehensive Cancer Network also recommends a single dose of pegfilgrastim for primary prevention of FN. Efficacy benefits of pegfilgrastim use include increased patient adherence due to a single injection, a reduction in FN incidence and FN-related hospitalizations, and improved time to ANC recovery compared with filgrastim.¹¹ There are reports suggesting pegfilgrastim significantly reduces neutropenia and FN incidence to a greater extent compared with daily filgrastim injections.⁶ In patients with breast cancer receiving dose-dense adjuvant chemotherapy, there are data demonstrating that patients who received filgrastim were more likely to experience severe neutropenia, dose reductions, and treatment delays leading to lower dose density compared with pegfilgrastim.¹² Of the 19 patients with breast cancer included in our population, 26% experienced one of the previously described outcomes leading to either extensions of daily filgrastim injections or transitions to pegfilgrastim to successfully maintain dose density. In patients with acute myeloid leukemia receiving consolidation chemotherapy, filgrastim was found to be associated with a statistically significant increased risk of hospitalizations compared with pegfilgrastim.¹³ The one patient with acute myeloid leukemia included in our study did not require additional hospitalizations for neutropenia or FN after transitioning to pegfilgrastim.

Given the cost advantage, the South Texas Veterans Health Care System continues to prefer daily filgrastim injections. A recent survey demonstrated that 73% of patients at 23 sites in the Veterans Health Administration used filgrastim rather than pegfilgrastim for cost savings, although it is recognized that daily filgrastim injections are less convenient for patients.¹⁴ This analysis did not review costs associated with hospitalization for FN or the appropriateness of G-CSF use. Cancer-related neutropenia accounts for 8.3% of all cancer-related hospitalization costs among adults; the average hospitalization costs nearly $25,000 per stay and about $2.3 billion among adult patients with cancer annually.^10,15

Limitations

This study has limitations that affected the applicability and interpretation of the results. This included the study design since it was a retrospective, single-center, descriptive cohort study. Patient adherence to daily filgrastim injections could not be assessed due to the retrospective nature of the study. The small sample size of 59 patients was prohibitive for utilization of additional analytical tools. Additionally, the predominately male veteran population may make applicability to non-VA populations restrictive.

 

CONCLUSIONS

Based on the incidence of primary and secondary outcomes associated with using daily filgrastim injections as primary prophylaxis in this study, additional measures such as tracking postnadir ANCs should be performed to ensure patients receive an appropriate number of filgrastim doses to prevent complications associated with neutropenia.

Acknowledgments

We thank Eric Dougherty, PharmD, for assistance in producing granulocyte colony-stimulating factor data.

Radiation therapy can cause long-term dysphagia that seriously affects quality of life for survivors of head and neck cancer. ^1-3 Numerous studies have linked pharyngeal constrictor dose to long-term dysphagia, but conclusions about the dose distribution that can be safely tolerated have been inconsistent. For example, a group from the Netherlands found that the mean dose to the superior pharyngeal constrictor muscle and the supraglottic larynx were each predictive of dysphagia. ⁴ A subsequent Vanderbilt study refuted these findings, reporting that these structures were not predictive but that dose to the inferior pharyngeal constrictor muscle was. ⁵ Other studies have connected late dysphagia with dose to the middle pharyngeal constrictor muscle, total larynx, oral cavity, contralateral submandibular gland, contralateral parotid gland, or a combination of these structures. ^6-14 NRG Oncology trials commonly evaluate dose to the “uninvolved pharynx,” which is the total pharyngeal constrictor muscle volume minus the planning target volume (PTV) for the lowest dose target volume. NRG head and neck trials 3, 4, 5, 6, 8, and 9 all use uninvolved pharynx mean dose ≤ 45 Gy as a constraint to judge radiation plan quality.

Differences in methodology or patient population may explain the inconsistency of prior studies on dosimetric predictors of dysphagia, but it is possible that these studies did not evaluate the optimal metric for dysphagia. This study evaluates a novel organ at risk, the contralateral pharyngeal constrictor muscle, to determine whether dose to this structure is predictive of late swallowing function. The study also compares a constraint based on this structure to the NRG uninvolved pharynx constraint mentioned earlier.

Methods

This study is a retrospective review of patients treated at the Richard L. Roudebush Veterans Affairs (VA) Medical Center in Indianapolis, Indiana. Patients were identified by searching the VA Cancer Registry for patients treated for head and neck squamous cell carcinoma between September 1, 2016, and August 30, 2019. Eligible sites included cancers of the nasopharynx, oropharynx, hypopharynx, larynx and oral cavity, as well as head and neck cancer of an unknown primary site. Only patients treated with primary radiation with concurrent systemic therapy were included. Patients were excluded if they had prior surgery or radiation to the head and neck.

The pharyngeal constrictor muscles were contoured per the techniques described by Bhide and colleagues.¹¹ The contralateral constrictor was defined as the half of the constrictor volume contralateral to the primary site. For midline tumors, the side of the neck with a lower volume of lymph node metastases was judged to be the contralateral side.

Air cavity editing was assessed by making an auto-expansion of the gross tumor volume (GTV) to match the boost volume clinical target value (CTV), then determining whether the size of this CTV was decreased in an air cavity on any axial slice. In patients with air cavity editing, the CTV was not completely cropped out of air, just reduced relative to the expansion used in soft tissue (Figure 1).

One-year dysphagia was defined as having a gastronomy tube (G-tube) in place or an abnormal modified barium swallow (MBS) ≥ 12 months after the completion of radiation. At the study institution, MBS is not routinely done after therapy but is ordered if a patient or clinician has concerns about swallowing function. MBS was considered abnormal if there was laryngeal penetration that reached the level of the glottis or was not ejected from the larynx.

 

Results

The VA Cancer Registry identified 113 patients treated for head and neck cancer during the study period. Of these, 55 patients met the inclusion criteria. No patients were lost to follow-up. The median follow-up was 29 months. The median age was 67 years (range, 41-83) (Table 1).

Oropharyngeal cancer was the most common primary site, accounting for 36 patients (65%).

All patients were treated with intensity-modulated radiotherapy. Patients treated with a sequential boost had an initial dose of 54 Gy and/or 50 Gy, followed by a boost to a total of 70 Gy at 2 Gy per fraction. Patients treated with a simultaneous integrated boost (SIB) technique received 70.0 Gy in 33 fractions, with elective volumes treated to 54.5 Gy in 33 fractions. Both patients with nasopharyngeal cancer were treated with SIB plans and had an intermediate dose volume of 59.4 Gy.

Systemic therapy was weekly cisplatin in 41 patients (75%) and cetuximab in 14 (25%). Twenty percent of patients receiving cisplatin switched to an alternative agent during treatment, most commonly carboplatin.

Forty-nine patients (89%) had a G-tube placed before starting radiation. G-tubes were in place for an interval of 0 to 47 months (mean, 8.6); 12 (22%) had a G-tube > 12 months. After completion of radiation, 18 patients (33%) had an abnormal MBS. These were done 1 to 50 months (mean, 14.8) after completion of radiation. Abnormal MBS occurred ≥ 12 months after radiation in 9 patients, 5 of whom had their G-tube in place for less than a year.

Forty-six patients (84%) survived more than 1 year and could be evaluated for late swallowing function. One-year dysphagia was seen in 17 (37%) of these patients. Recurrence was seen in 20 patients (36%), with locoregional recurrence in 12 (60%) of these cases. Recurrence occurred at a range of 0 to 15 months (mean, 5.6). Neither recurrence (P = .69) nor locoregional recurrence (P = .11) was associated with increased dysphagia at 1 year.

In patients who could be evaluated for long-term swallowing function, contralateral constrictor V60 ranged from 0% to 100% (median, 51%). V60 was < 40% in 18 patients (39%). With V60 < 40%, there was a 6% rate of 1-year dysphagia compared with 57% for V60 ≥ 40% (P < .001).

Patients with contralateral constrictor V60 < 40 and V60 ≥ 40 both had a mean age of 65 years. χ² analysis did not show a difference in T stage or systemic treatment but did show that patients with V60 < 40% were more likely to have N1 disease (P = .01), and less likely to have N2 disease (P = .01) compared with patients with V60 ≥ 40%. The difference in 1-year dysphagia between N0 to N1 patients (27%) and N2 to N3 patients (46%) was not statistically significant (P = .19).

In patients who could be evaluated for long-term swallowing function, the uninvolved pharynx volume median of the total constrictor volume was 32% (range, < 1%-62%). The uninvolved pharynx mean dose ranged from 28 to 68 Gy (median, 45). When the uninvolved pharynx mean dose was < 45 Gy, 1-year dysphagia was 22% compared with 52% with a dose ≥ 45 Gy (P = .03).

Table 2 compares constraints based on uninvolved pharynx with a constraint based on the contralateral constrictor.

Air cavity editing was performed in 27 patients (49%). One-year survival was 93% with air cavity editing, and 75% without, which was not statistically significant. Locoregional recurrence occurred in 3 patients (11%) with air cavity editing, and 9 (32%) without, which was not statistically significant. In patients surviving at least 1 year, contralateral constrictor V60 averaged 33% with editing and 62% without editing (P < .001). One-year dysphagia was 12% with air cavity editing and 67% without editing (P < .001).

An SIB technique was done in 26 patients (47%). One-year survival was 85% (n = 22) with SIB and 83% (n = 24) with sequential boost, which was not statistically significant. Locoregional recurrence occurred in 19% with SIB, and 32% with sequential boost, which was not statistically significant. For SIB patients alive at 1 year, the median contralateral V60 was 28%, compared with 66% for patients treated with sequential technique. Seventeen patients (77%) with SIB had V60 < 40%. Nineteen (86%) of SIB plans also had air cavity editing. One patient (5%) with SIB had dysphagia at 1 year compared with 16 (67%) sequential patients (P < .001).

 

Discussion

This is the first study to link contralateral constrictor dose to long-term dysphagia in patients treated with radiation for head and neck cancer. Editing the boost volume off air cavities was associated with lower contralateral constrictor V60 and with less long-term dysphagia. This may indicate that optimizing plans to meet a contralateral constrictor constraint can reduce rates of long-term dysphagia.

The most useful clinical predictors are those that identify a patient at low risk for toxicity. These constraints are useful because they reassure physicians that treatments will have a favorable risk/benefit ratio while identifying plans that may need modification before starting treatment.

The contralateral constrictor outperformed the uninvolved pharynx in identifying patients at low risk for long-term dysphagia. This difference could not be overcome by decreasing the threshold of the pharynx constraint, as 17% of patients with dysphagia had a mean dose of < 40 Gy to the uninvolved pharynx, which was not statistically significant. An advantage of contralateral constrictor is that it is independent of PTV size. The uninvolved pharynx structure depends on the PTV contour, so it may obscure a connection between PTV size and dysphagia.

In the context of a clinical trial, only measuring dose to the uninvolved pharynx may allow more plans to meet constraints, but even in NRG trials, physicians have some control over target volumes. For example, NRG HN009, a national trial for patients with head and neck cancer, recommends editing the CTV_7000 (clinical target volume treated to 70 Gy) off air cavities but does not define how much the volume should be cropped or specify protocol violations if the volume is not cropped.¹⁵ Furthermore, constraints used in clinical trials are often adopted for use outside the trial, where physicians have extensive control over target volumes.

The broad range of uninvolved pharynx volume relative to total constrictor volume confounds predictions using this variable. For example, according to the NRG constraint, a patient with an uninvolved pharynx mean dose of 44 Gy will have a low risk of dysphagia even if this structure is only 1% of the total constrictor. The contralateral constrictor is always about 50% of the total constrictor volume, which means that predictions using this structure will not be confounded by the same variation in volume size.

Figure 2 shows a representative patient who met the NRG uninvolved pharynx constraint but developed long-term dysphagia.

This patient had an uninvolved pharynx mean dose of only 33 Gy, but this volume was only 31% of his total constrictor volume. This plan shows that on axial slices containing the GTV, nearly the entire constrictor was within the PTV and received at least 60 Gy. These areas of overlap and the dose they receive are not included in the uninvolved pharynx volume. The contralateral constrictor V60 for this patient was 52%, so the patient would have been in the high-risk group for dysphagia based on this structure’s constraint.

Pharyngoesophageal stricture is a common cause of dysphagia after intensity-modulated radiotherapy for head and neck cancer.¹⁶ Radiation has been shown to decrease pharyngeal function in patients with head and neck cancer.¹⁷ Sparing one side of the pharynx may allow for better pharyngeal compliance throughout the length of the pharynx, possibly decreasing the rate of pharyngoesophageal stricture. Additionally, constraining the contralateral constrictor may preserve strength on this side, allowing it to compensate for weakness on the side of the primary cancer. An exercise sometimes used for dysphagia involves head rotation toward the affected side during swallowing. This technique has been shown to cause food to move to the unaffected side.¹⁸ Sparing the contralateral constrictor may help such techniques work better in patients with head and neck cancer.

Few studies have commented specifically on dose to swallowing structures contralateral to the primary tumor. Two studies have proposed contralateral submandibular gland constraints for dysphagia (not xerostomia), but neither measured the dose to the contralateral constrictor muscle.^9,10 Although the contralateral submandibular dose may correlate with dose to the constrictor on that side, the submandibular gland may have a less direct impact on swallowing than the constrictor muscle, and its limited dimensions may make constraints based on the gland less robust for cancers outside the oropharynx.

Another study reported improved quality of life in patients who were not treated with elective contralateral retropharyngeal radiation.¹⁹ Although it is likely that doses to the contralateral constrictor were lower in patients who did not receive elective radiation to this area, this study did not measure or constrain doses to the contralateral constrictors.

 

Limitations

This study is limited by its single institution, retrospective design, small sample size, and by all patients being male. The high correlation between air cavity editing and the use of SIB makes it impossible to assess the impact of each technique individually. Patients with contralateral constrictor V60 < 40% were less likely to have N2 disease, but N2 to N3 disease did not predict higher 1-year dysphagia, so the difference in N-category cannot fully explain the difference in 1-year dysphagia. It is possible that unreported factors, such as CTV, may contribute significantly to swallowing function. Nevertheless, within the study population, contralateral constrictor dose was able to identify a group with a low rate of long-term dysphagia.

Conclusions

Contralateral constrictor dose is a promising predictor of late dysphagia for patients with head and neck cancer treated with radiation with concurrent systemic therapy. Contralateral constrictor V60 < 40% was able to identify a group of patients with a low rate of 1-year dysphagia in this single-center retrospective study. The correlation between air cavity editing and contralateral constrictor V60 suggests that contralateral constrictor dose may depend partly on technique. Further studies are needed to see if the contralateral constrictor dose can be used to predict long-term dysphagia prospectively and in other patient populations.

Radiation therapy can cause long-term dysphagia that seriously affects quality of life for survivors of head and neck cancer. ^1-3 Numerous studies have linked pharyngeal constrictor dose to long-term dysphagia, but conclusions about the dose distribution that can be safely tolerated have been inconsistent. For example, a group from the Netherlands found that the mean dose to the superior pharyngeal constrictor muscle and the supraglottic larynx were each predictive of dysphagia. ⁴ A subsequent Vanderbilt study refuted these findings, reporting that these structures were not predictive but that dose to the inferior pharyngeal constrictor muscle was. ⁵ Other studies have connected late dysphagia with dose to the middle pharyngeal constrictor muscle, total larynx, oral cavity, contralateral submandibular gland, contralateral parotid gland, or a combination of these structures. ^6-14 NRG Oncology trials commonly evaluate dose to the “uninvolved pharynx,” which is the total pharyngeal constrictor muscle volume minus the planning target volume (PTV) for the lowest dose target volume. NRG head and neck trials 3, 4, 5, 6, 8, and 9 all use uninvolved pharynx mean dose ≤ 45 Gy as a constraint to judge radiation plan quality.

Differences in methodology or patient population may explain the inconsistency of prior studies on dosimetric predictors of dysphagia, but it is possible that these studies did not evaluate the optimal metric for dysphagia. This study evaluates a novel organ at risk, the contralateral pharyngeal constrictor muscle, to determine whether dose to this structure is predictive of late swallowing function. The study also compares a constraint based on this structure to the NRG uninvolved pharynx constraint mentioned earlier.

Methods

This study is a retrospective review of patients treated at the Richard L. Roudebush Veterans Affairs (VA) Medical Center in Indianapolis, Indiana. Patients were identified by searching the VA Cancer Registry for patients treated for head and neck squamous cell carcinoma between September 1, 2016, and August 30, 2019. Eligible sites included cancers of the nasopharynx, oropharynx, hypopharynx, larynx and oral cavity, as well as head and neck cancer of an unknown primary site. Only patients treated with primary radiation with concurrent systemic therapy were included. Patients were excluded if they had prior surgery or radiation to the head and neck.

The pharyngeal constrictor muscles were contoured per the techniques described by Bhide and colleagues.¹¹ The contralateral constrictor was defined as the half of the constrictor volume contralateral to the primary site. For midline tumors, the side of the neck with a lower volume of lymph node metastases was judged to be the contralateral side.

Air cavity editing was assessed by making an auto-expansion of the gross tumor volume (GTV) to match the boost volume clinical target value (CTV), then determining whether the size of this CTV was decreased in an air cavity on any axial slice. In patients with air cavity editing, the CTV was not completely cropped out of air, just reduced relative to the expansion used in soft tissue (Figure 1).

One-year dysphagia was defined as having a gastronomy tube (G-tube) in place or an abnormal modified barium swallow (MBS) ≥ 12 months after the completion of radiation. At the study institution, MBS is not routinely done after therapy but is ordered if a patient or clinician has concerns about swallowing function. MBS was considered abnormal if there was laryngeal penetration that reached the level of the glottis or was not ejected from the larynx.

 

Results

The VA Cancer Registry identified 113 patients treated for head and neck cancer during the study period. Of these, 55 patients met the inclusion criteria. No patients were lost to follow-up. The median follow-up was 29 months. The median age was 67 years (range, 41-83) (Table 1).

Oropharyngeal cancer was the most common primary site, accounting for 36 patients (65%).

All patients were treated with intensity-modulated radiotherapy. Patients treated with a sequential boost had an initial dose of 54 Gy and/or 50 Gy, followed by a boost to a total of 70 Gy at 2 Gy per fraction. Patients treated with a simultaneous integrated boost (SIB) technique received 70.0 Gy in 33 fractions, with elective volumes treated to 54.5 Gy in 33 fractions. Both patients with nasopharyngeal cancer were treated with SIB plans and had an intermediate dose volume of 59.4 Gy.

Systemic therapy was weekly cisplatin in 41 patients (75%) and cetuximab in 14 (25%). Twenty percent of patients receiving cisplatin switched to an alternative agent during treatment, most commonly carboplatin.

Forty-nine patients (89%) had a G-tube placed before starting radiation. G-tubes were in place for an interval of 0 to 47 months (mean, 8.6); 12 (22%) had a G-tube > 12 months. After completion of radiation, 18 patients (33%) had an abnormal MBS. These were done 1 to 50 months (mean, 14.8) after completion of radiation. Abnormal MBS occurred ≥ 12 months after radiation in 9 patients, 5 of whom had their G-tube in place for less than a year.

Forty-six patients (84%) survived more than 1 year and could be evaluated for late swallowing function. One-year dysphagia was seen in 17 (37%) of these patients. Recurrence was seen in 20 patients (36%), with locoregional recurrence in 12 (60%) of these cases. Recurrence occurred at a range of 0 to 15 months (mean, 5.6). Neither recurrence (P = .69) nor locoregional recurrence (P = .11) was associated with increased dysphagia at 1 year.

In patients who could be evaluated for long-term swallowing function, contralateral constrictor V60 ranged from 0% to 100% (median, 51%). V60 was < 40% in 18 patients (39%). With V60 < 40%, there was a 6% rate of 1-year dysphagia compared with 57% for V60 ≥ 40% (P < .001).

Patients with contralateral constrictor V60 < 40 and V60 ≥ 40 both had a mean age of 65 years. χ² analysis did not show a difference in T stage or systemic treatment but did show that patients with V60 < 40% were more likely to have N1 disease (P = .01), and less likely to have N2 disease (P = .01) compared with patients with V60 ≥ 40%. The difference in 1-year dysphagia between N0 to N1 patients (27%) and N2 to N3 patients (46%) was not statistically significant (P = .19).

In patients who could be evaluated for long-term swallowing function, the uninvolved pharynx volume median of the total constrictor volume was 32% (range, < 1%-62%). The uninvolved pharynx mean dose ranged from 28 to 68 Gy (median, 45). When the uninvolved pharynx mean dose was < 45 Gy, 1-year dysphagia was 22% compared with 52% with a dose ≥ 45 Gy (P = .03).

Table 2 compares constraints based on uninvolved pharynx with a constraint based on the contralateral constrictor.

Air cavity editing was performed in 27 patients (49%). One-year survival was 93% with air cavity editing, and 75% without, which was not statistically significant. Locoregional recurrence occurred in 3 patients (11%) with air cavity editing, and 9 (32%) without, which was not statistically significant. In patients surviving at least 1 year, contralateral constrictor V60 averaged 33% with editing and 62% without editing (P < .001). One-year dysphagia was 12% with air cavity editing and 67% without editing (P < .001).

An SIB technique was done in 26 patients (47%). One-year survival was 85% (n = 22) with SIB and 83% (n = 24) with sequential boost, which was not statistically significant. Locoregional recurrence occurred in 19% with SIB, and 32% with sequential boost, which was not statistically significant. For SIB patients alive at 1 year, the median contralateral V60 was 28%, compared with 66% for patients treated with sequential technique. Seventeen patients (77%) with SIB had V60 < 40%. Nineteen (86%) of SIB plans also had air cavity editing. One patient (5%) with SIB had dysphagia at 1 year compared with 16 (67%) sequential patients (P < .001).

 

Discussion

This is the first study to link contralateral constrictor dose to long-term dysphagia in patients treated with radiation for head and neck cancer. Editing the boost volume off air cavities was associated with lower contralateral constrictor V60 and with less long-term dysphagia. This may indicate that optimizing plans to meet a contralateral constrictor constraint can reduce rates of long-term dysphagia.

The most useful clinical predictors are those that identify a patient at low risk for toxicity. These constraints are useful because they reassure physicians that treatments will have a favorable risk/benefit ratio while identifying plans that may need modification before starting treatment.

The contralateral constrictor outperformed the uninvolved pharynx in identifying patients at low risk for long-term dysphagia. This difference could not be overcome by decreasing the threshold of the pharynx constraint, as 17% of patients with dysphagia had a mean dose of < 40 Gy to the uninvolved pharynx, which was not statistically significant. An advantage of contralateral constrictor is that it is independent of PTV size. The uninvolved pharynx structure depends on the PTV contour, so it may obscure a connection between PTV size and dysphagia.

In the context of a clinical trial, only measuring dose to the uninvolved pharynx may allow more plans to meet constraints, but even in NRG trials, physicians have some control over target volumes. For example, NRG HN009, a national trial for patients with head and neck cancer, recommends editing the CTV_7000 (clinical target volume treated to 70 Gy) off air cavities but does not define how much the volume should be cropped or specify protocol violations if the volume is not cropped.¹⁵ Furthermore, constraints used in clinical trials are often adopted for use outside the trial, where physicians have extensive control over target volumes.

The broad range of uninvolved pharynx volume relative to total constrictor volume confounds predictions using this variable. For example, according to the NRG constraint, a patient with an uninvolved pharynx mean dose of 44 Gy will have a low risk of dysphagia even if this structure is only 1% of the total constrictor. The contralateral constrictor is always about 50% of the total constrictor volume, which means that predictions using this structure will not be confounded by the same variation in volume size.

Figure 2 shows a representative patient who met the NRG uninvolved pharynx constraint but developed long-term dysphagia.

This patient had an uninvolved pharynx mean dose of only 33 Gy, but this volume was only 31% of his total constrictor volume. This plan shows that on axial slices containing the GTV, nearly the entire constrictor was within the PTV and received at least 60 Gy. These areas of overlap and the dose they receive are not included in the uninvolved pharynx volume. The contralateral constrictor V60 for this patient was 52%, so the patient would have been in the high-risk group for dysphagia based on this structure’s constraint.

Pharyngoesophageal stricture is a common cause of dysphagia after intensity-modulated radiotherapy for head and neck cancer.¹⁶ Radiation has been shown to decrease pharyngeal function in patients with head and neck cancer.¹⁷ Sparing one side of the pharynx may allow for better pharyngeal compliance throughout the length of the pharynx, possibly decreasing the rate of pharyngoesophageal stricture. Additionally, constraining the contralateral constrictor may preserve strength on this side, allowing it to compensate for weakness on the side of the primary cancer. An exercise sometimes used for dysphagia involves head rotation toward the affected side during swallowing. This technique has been shown to cause food to move to the unaffected side.¹⁸ Sparing the contralateral constrictor may help such techniques work better in patients with head and neck cancer.

Few studies have commented specifically on dose to swallowing structures contralateral to the primary tumor. Two studies have proposed contralateral submandibular gland constraints for dysphagia (not xerostomia), but neither measured the dose to the contralateral constrictor muscle.^9,10 Although the contralateral submandibular dose may correlate with dose to the constrictor on that side, the submandibular gland may have a less direct impact on swallowing than the constrictor muscle, and its limited dimensions may make constraints based on the gland less robust for cancers outside the oropharynx.

Another study reported improved quality of life in patients who were not treated with elective contralateral retropharyngeal radiation.¹⁹ Although it is likely that doses to the contralateral constrictor were lower in patients who did not receive elective radiation to this area, this study did not measure or constrain doses to the contralateral constrictors.

 

Limitations

This study is limited by its single institution, retrospective design, small sample size, and by all patients being male. The high correlation between air cavity editing and the use of SIB makes it impossible to assess the impact of each technique individually. Patients with contralateral constrictor V60 < 40% were less likely to have N2 disease, but N2 to N3 disease did not predict higher 1-year dysphagia, so the difference in N-category cannot fully explain the difference in 1-year dysphagia. It is possible that unreported factors, such as CTV, may contribute significantly to swallowing function. Nevertheless, within the study population, contralateral constrictor dose was able to identify a group with a low rate of long-term dysphagia.

Conclusions

Contralateral constrictor dose is a promising predictor of late dysphagia for patients with head and neck cancer treated with radiation with concurrent systemic therapy. Contralateral constrictor V60 < 40% was able to identify a group of patients with a low rate of 1-year dysphagia in this single-center retrospective study. The correlation between air cavity editing and contralateral constrictor V60 suggests that contralateral constrictor dose may depend partly on technique. Further studies are needed to see if the contralateral constrictor dose can be used to predict long-term dysphagia prospectively and in other patient populations.

Radiation therapy can cause long-term dysphagia that seriously affects quality of life for survivors of head and neck cancer. ^1-3 Numerous studies have linked pharyngeal constrictor dose to long-term dysphagia, but conclusions about the dose distribution that can be safely tolerated have been inconsistent. For example, a group from the Netherlands found that the mean dose to the superior pharyngeal constrictor muscle and the supraglottic larynx were each predictive of dysphagia. ⁴ A subsequent Vanderbilt study refuted these findings, reporting that these structures were not predictive but that dose to the inferior pharyngeal constrictor muscle was. ⁵ Other studies have connected late dysphagia with dose to the middle pharyngeal constrictor muscle, total larynx, oral cavity, contralateral submandibular gland, contralateral parotid gland, or a combination of these structures. ^6-14 NRG Oncology trials commonly evaluate dose to the “uninvolved pharynx,” which is the total pharyngeal constrictor muscle volume minus the planning target volume (PTV) for the lowest dose target volume. NRG head and neck trials 3, 4, 5, 6, 8, and 9 all use uninvolved pharynx mean dose ≤ 45 Gy as a constraint to judge radiation plan quality.

Differences in methodology or patient population may explain the inconsistency of prior studies on dosimetric predictors of dysphagia, but it is possible that these studies did not evaluate the optimal metric for dysphagia. This study evaluates a novel organ at risk, the contralateral pharyngeal constrictor muscle, to determine whether dose to this structure is predictive of late swallowing function. The study also compares a constraint based on this structure to the NRG uninvolved pharynx constraint mentioned earlier.

Methods

This study is a retrospective review of patients treated at the Richard L. Roudebush Veterans Affairs (VA) Medical Center in Indianapolis, Indiana. Patients were identified by searching the VA Cancer Registry for patients treated for head and neck squamous cell carcinoma between September 1, 2016, and August 30, 2019. Eligible sites included cancers of the nasopharynx, oropharynx, hypopharynx, larynx and oral cavity, as well as head and neck cancer of an unknown primary site. Only patients treated with primary radiation with concurrent systemic therapy were included. Patients were excluded if they had prior surgery or radiation to the head and neck.

The pharyngeal constrictor muscles were contoured per the techniques described by Bhide and colleagues.¹¹ The contralateral constrictor was defined as the half of the constrictor volume contralateral to the primary site. For midline tumors, the side of the neck with a lower volume of lymph node metastases was judged to be the contralateral side.

Air cavity editing was assessed by making an auto-expansion of the gross tumor volume (GTV) to match the boost volume clinical target value (CTV), then determining whether the size of this CTV was decreased in an air cavity on any axial slice. In patients with air cavity editing, the CTV was not completely cropped out of air, just reduced relative to the expansion used in soft tissue (Figure 1).

One-year dysphagia was defined as having a gastronomy tube (G-tube) in place or an abnormal modified barium swallow (MBS) ≥ 12 months after the completion of radiation. At the study institution, MBS is not routinely done after therapy but is ordered if a patient or clinician has concerns about swallowing function. MBS was considered abnormal if there was laryngeal penetration that reached the level of the glottis or was not ejected from the larynx.

 

Results

The VA Cancer Registry identified 113 patients treated for head and neck cancer during the study period. Of these, 55 patients met the inclusion criteria. No patients were lost to follow-up. The median follow-up was 29 months. The median age was 67 years (range, 41-83) (Table 1).

Oropharyngeal cancer was the most common primary site, accounting for 36 patients (65%).

All patients were treated with intensity-modulated radiotherapy. Patients treated with a sequential boost had an initial dose of 54 Gy and/or 50 Gy, followed by a boost to a total of 70 Gy at 2 Gy per fraction. Patients treated with a simultaneous integrated boost (SIB) technique received 70.0 Gy in 33 fractions, with elective volumes treated to 54.5 Gy in 33 fractions. Both patients with nasopharyngeal cancer were treated with SIB plans and had an intermediate dose volume of 59.4 Gy.

Systemic therapy was weekly cisplatin in 41 patients (75%) and cetuximab in 14 (25%). Twenty percent of patients receiving cisplatin switched to an alternative agent during treatment, most commonly carboplatin.

Forty-nine patients (89%) had a G-tube placed before starting radiation. G-tubes were in place for an interval of 0 to 47 months (mean, 8.6); 12 (22%) had a G-tube > 12 months. After completion of radiation, 18 patients (33%) had an abnormal MBS. These were done 1 to 50 months (mean, 14.8) after completion of radiation. Abnormal MBS occurred ≥ 12 months after radiation in 9 patients, 5 of whom had their G-tube in place for less than a year.

Forty-six patients (84%) survived more than 1 year and could be evaluated for late swallowing function. One-year dysphagia was seen in 17 (37%) of these patients. Recurrence was seen in 20 patients (36%), with locoregional recurrence in 12 (60%) of these cases. Recurrence occurred at a range of 0 to 15 months (mean, 5.6). Neither recurrence (P = .69) nor locoregional recurrence (P = .11) was associated with increased dysphagia at 1 year.

In patients who could be evaluated for long-term swallowing function, contralateral constrictor V60 ranged from 0% to 100% (median, 51%). V60 was < 40% in 18 patients (39%). With V60 < 40%, there was a 6% rate of 1-year dysphagia compared with 57% for V60 ≥ 40% (P < .001).

Patients with contralateral constrictor V60 < 40 and V60 ≥ 40 both had a mean age of 65 years. χ² analysis did not show a difference in T stage or systemic treatment but did show that patients with V60 < 40% were more likely to have N1 disease (P = .01), and less likely to have N2 disease (P = .01) compared with patients with V60 ≥ 40%. The difference in 1-year dysphagia between N0 to N1 patients (27%) and N2 to N3 patients (46%) was not statistically significant (P = .19).

In patients who could be evaluated for long-term swallowing function, the uninvolved pharynx volume median of the total constrictor volume was 32% (range, < 1%-62%). The uninvolved pharynx mean dose ranged from 28 to 68 Gy (median, 45). When the uninvolved pharynx mean dose was < 45 Gy, 1-year dysphagia was 22% compared with 52% with a dose ≥ 45 Gy (P = .03).

Table 2 compares constraints based on uninvolved pharynx with a constraint based on the contralateral constrictor.

Air cavity editing was performed in 27 patients (49%). One-year survival was 93% with air cavity editing, and 75% without, which was not statistically significant. Locoregional recurrence occurred in 3 patients (11%) with air cavity editing, and 9 (32%) without, which was not statistically significant. In patients surviving at least 1 year, contralateral constrictor V60 averaged 33% with editing and 62% without editing (P < .001). One-year dysphagia was 12% with air cavity editing and 67% without editing (P < .001).

An SIB technique was done in 26 patients (47%). One-year survival was 85% (n = 22) with SIB and 83% (n = 24) with sequential boost, which was not statistically significant. Locoregional recurrence occurred in 19% with SIB, and 32% with sequential boost, which was not statistically significant. For SIB patients alive at 1 year, the median contralateral V60 was 28%, compared with 66% for patients treated with sequential technique. Seventeen patients (77%) with SIB had V60 < 40%. Nineteen (86%) of SIB plans also had air cavity editing. One patient (5%) with SIB had dysphagia at 1 year compared with 16 (67%) sequential patients (P < .001).

 

Discussion

This is the first study to link contralateral constrictor dose to long-term dysphagia in patients treated with radiation for head and neck cancer. Editing the boost volume off air cavities was associated with lower contralateral constrictor V60 and with less long-term dysphagia. This may indicate that optimizing plans to meet a contralateral constrictor constraint can reduce rates of long-term dysphagia.

The most useful clinical predictors are those that identify a patient at low risk for toxicity. These constraints are useful because they reassure physicians that treatments will have a favorable risk/benefit ratio while identifying plans that may need modification before starting treatment.

The contralateral constrictor outperformed the uninvolved pharynx in identifying patients at low risk for long-term dysphagia. This difference could not be overcome by decreasing the threshold of the pharynx constraint, as 17% of patients with dysphagia had a mean dose of < 40 Gy to the uninvolved pharynx, which was not statistically significant. An advantage of contralateral constrictor is that it is independent of PTV size. The uninvolved pharynx structure depends on the PTV contour, so it may obscure a connection between PTV size and dysphagia.

In the context of a clinical trial, only measuring dose to the uninvolved pharynx may allow more plans to meet constraints, but even in NRG trials, physicians have some control over target volumes. For example, NRG HN009, a national trial for patients with head and neck cancer, recommends editing the CTV_7000 (clinical target volume treated to 70 Gy) off air cavities but does not define how much the volume should be cropped or specify protocol violations if the volume is not cropped.¹⁵ Furthermore, constraints used in clinical trials are often adopted for use outside the trial, where physicians have extensive control over target volumes.

The broad range of uninvolved pharynx volume relative to total constrictor volume confounds predictions using this variable. For example, according to the NRG constraint, a patient with an uninvolved pharynx mean dose of 44 Gy will have a low risk of dysphagia even if this structure is only 1% of the total constrictor. The contralateral constrictor is always about 50% of the total constrictor volume, which means that predictions using this structure will not be confounded by the same variation in volume size.

Figure 2 shows a representative patient who met the NRG uninvolved pharynx constraint but developed long-term dysphagia.

This patient had an uninvolved pharynx mean dose of only 33 Gy, but this volume was only 31% of his total constrictor volume. This plan shows that on axial slices containing the GTV, nearly the entire constrictor was within the PTV and received at least 60 Gy. These areas of overlap and the dose they receive are not included in the uninvolved pharynx volume. The contralateral constrictor V60 for this patient was 52%, so the patient would have been in the high-risk group for dysphagia based on this structure’s constraint.

Pharyngoesophageal stricture is a common cause of dysphagia after intensity-modulated radiotherapy for head and neck cancer.¹⁶ Radiation has been shown to decrease pharyngeal function in patients with head and neck cancer.¹⁷ Sparing one side of the pharynx may allow for better pharyngeal compliance throughout the length of the pharynx, possibly decreasing the rate of pharyngoesophageal stricture. Additionally, constraining the contralateral constrictor may preserve strength on this side, allowing it to compensate for weakness on the side of the primary cancer. An exercise sometimes used for dysphagia involves head rotation toward the affected side during swallowing. This technique has been shown to cause food to move to the unaffected side.¹⁸ Sparing the contralateral constrictor may help such techniques work better in patients with head and neck cancer.

Few studies have commented specifically on dose to swallowing structures contralateral to the primary tumor. Two studies have proposed contralateral submandibular gland constraints for dysphagia (not xerostomia), but neither measured the dose to the contralateral constrictor muscle.^9,10 Although the contralateral submandibular dose may correlate with dose to the constrictor on that side, the submandibular gland may have a less direct impact on swallowing than the constrictor muscle, and its limited dimensions may make constraints based on the gland less robust for cancers outside the oropharynx.

Another study reported improved quality of life in patients who were not treated with elective contralateral retropharyngeal radiation.¹⁹ Although it is likely that doses to the contralateral constrictor were lower in patients who did not receive elective radiation to this area, this study did not measure or constrain doses to the contralateral constrictors.

 

Limitations

This study is limited by its single institution, retrospective design, small sample size, and by all patients being male. The high correlation between air cavity editing and the use of SIB makes it impossible to assess the impact of each technique individually. Patients with contralateral constrictor V60 < 40% were less likely to have N2 disease, but N2 to N3 disease did not predict higher 1-year dysphagia, so the difference in N-category cannot fully explain the difference in 1-year dysphagia. It is possible that unreported factors, such as CTV, may contribute significantly to swallowing function. Nevertheless, within the study population, contralateral constrictor dose was able to identify a group with a low rate of long-term dysphagia.

Conclusions

Contralateral constrictor dose is a promising predictor of late dysphagia for patients with head and neck cancer treated with radiation with concurrent systemic therapy. Contralateral constrictor V60 < 40% was able to identify a group of patients with a low rate of 1-year dysphagia in this single-center retrospective study. The correlation between air cavity editing and contralateral constrictor V60 suggests that contralateral constrictor dose may depend partly on technique. Further studies are needed to see if the contralateral constrictor dose can be used to predict long-term dysphagia prospectively and in other patient populations.

In patients with gastrointestinal (GI) malignancies, malnutrition is common. In addition, it has various negative implications, including high risk for surgical complications, prolonged hospitalization, decreased quality of life (QOL), increased mortality, and poor tolerance for treatments such as chemotherapy and radiotherapy.¹

A 2014 French study of 1903 patients hospitalized for cancer reported a 39% overall prevalence of malnutrition; 39% in patients with cancers of the colon/rectum, 60% for pancreatic cancer, and 67% for cancers of the esophagus/stomach.² Malnutrition was defined as body mass index (BMI) < 18.5 for individuals aged < 75 years or BMI < 21 for individuals aged ≥ 75 years, and/or weight loss > 10% since disease onset. Malnutrition also was strongly associated with worsened performance status.

The etiology of malnutrition in GI cancers is often multifactorial. It includes systemic tumor effects, such as inflammatory mediators contributing to hypermetabolism and cachexia, local tumor-associated mechanical obstruction, GI toxicities caused by antineoplastic therapy or other medications, and psychological factors that contribute to anorexia.³ Patient-related risk factors such as older age, other chronic diseases, and history of other GI surgeries also play a role.¹

Other studies have demonstrated that malnutrition in patients with GI malignancies undergoing surgical resection is associated with high rates of severe postoperative complications, increased length of stay (LOS) and time on a ventilator for patients treated in the intensive care unit, and poor QOL in the postoperative survival period.^4-6 Several randomized controlled trials conducted in patients with GI cancers have shown that enteral and parenteral nutrition supplementations in the perioperative period improve various outcomes, such as reduction of postoperative complication rates, fewer readmissions, improved chemotherapy tolerance, and improved QOL.^7-10 Thus, in the management of patients with GI malignancies, it is highly important to implement early nutritional screening and establish a diagnosis of malnutrition to intervene and reduce postoperative morbidity and mortality.¹

However, tools and predictors of malnutrition are often imperfect. The Academy of Nutrition and Dietetics and the American Society for Parenteral and Enteral Nutrition (AND/ASPEN) weight-based criteria define malnutrition and nutritionally-at-risk as BMI < 18.5, involuntary loss of at least 10% of body weight within 6 months or 5% within 1 month, or loss of 10 lb within 6 months.¹¹ While the ASPEN criteria are often used to define malnourishment, they may not fully capture the population at risk, and there does not exist a gold-standard tool for nutritional screening. A 2002 study that performed a critical appraisal of 44 nutritional screening tools found that no single tool was fully sufficient for application, development, evaluation, and consistent screening.¹² As such, consistently screening for malnutrition to target interventions in the perioperative period for GI surgical oncology has been challenging.¹³ More recent tools such as the perioperative nutrition screen (PONS) have been validated as rapid, effective screening tools to predict postoperative outcomes.¹⁴ Additionally, implementation of perioperative nutritional protocols, such as enhanced recovery after surgery (ERAS) in colon cancer (CC) surgery, also has shown improved perioperative care and outcomes.¹⁵

Preoperative nutritional interventions have been implemented in practice and have focused mostly on the immediate perioperative period. This has been shown to improve surgical outcomes. The Veterans Health Administration (VHA) provides comprehensive care to patients in a single-payer system, allowing for capture of perioperative data and the opportunity for focused preoperative interventions to improve outcomes.

 

Methods

This was a retrospective record review of colorectal malignancies treated with curative intent at the Veterans Affairs Ann Arbor Healthcare System (VAAAHS) in Michigan between January 1, 2015, and December 31, 2019. We examined nutritional status, degree of longitudinal weight loss, and subsequent clinical outcomes, including delayed postoperative recovery and delays in chemotherapy in 115 patients with CC and 33 patients with rectal cancer (RC) undergoing curative surgical resection at VAAAHS. To avoid additional confounding effects of advanced cancer, only early-stage, curable disease was included. This study was approved by the VAAAHS Institutional Review Board.

Patient cases were identified in the VAAAHS cancer registry, which included 115 patients with colon primary tumors and 33 patients with rectal tumors. According to the VAAAHS standard of care, patients with CC did not require neoadjuvant therapy while patients with RC cohort did (Figure). The CC cohort was defined as patients who had an adenocarcinoma, mucinous adenocarcinoma, or carcinoid tumor of the colon or rectosigmoid junction. These patients did not receive neoadjuvant therapy and underwent curative-intent surgical resection of their tumor. The RC cohort was defined as patients who had adenocarcinoma, mucinous adenocarcinoma, or signet ring cell carcinoma of the rectum. These patients received neoadjuvant chemoradiation followed by curative-intent surgical resection of their tumor.

Patients with postoperative follow-up outside of VAAAHS were excluded. Patients were excluded if their surgery had noncurative intent or if they had distant metastatic disease. Data on patient weights, laboratory results, nutrition consultations, postoperative complications, delayed recovery, readmissions, and chemotherapy tolerance were abstracted by patient chart review in the VHA Computerized Patient Record System and Joint Legacy Viewer by 2 researchers.

Delayed recovery was defined as any abnormal clinical development described in inpatient progress notes, outpatient follow-up notes within 60 days, or in hospital discharge summaries. Excluded were psychiatric events without additional medical complications, postoperative bleeding not requiring an invasive intervention, urinary retention, postoperative glycemic control difficulties, cardiac events that happened before postoperative hospital discharge and not requiring readmission, and postoperative alcohol withdrawal. Complications were defined similarly to delayed recovery but excluded isolated prolonged postoperative ileus. LOS was defined in days as time from admission to discharge.

Adjuvant management course was derived from reviewing documentation from medical oncology consultations and progress notes. In patients for whom adjuvant chemotherapy was indicated and prescribed, chemotherapy was considered complete if chemotherapy was started and completed as indicated. Adjuvant chemotherapy was considered incomplete if the patient declined chemotherapy, if chemotherapy was not started when indicated, or if chemotherapy was not completed as indicated. Neoadjuvant therapy data were abstracted from medical and radiation oncology notes.

Recorded data were collected on both weight and BMI. Weights were extracted as follows: Weight 1 year before time of diagnosis, ± 4 months; weight 6 months before diagnosis ± 3 months; weight at time of diagnosis ± 2 weeks; weight at time of surgery ± 2 weeks; weight 30 days postsurgery ± 2 weeks; weight 60 days postsurgery ± 2 weeks; weight 1 year postsurgery ± 4 months. Mean percent change in weight was calculated from recorded weights between each allocated time point. A weight loss of ≥ 3% was found to be clinically relevant and was chosen as the minimal cutoff value when analyzing outcomes associated with weight trends.

Nutrition consultations were abstracted as follows: Preoperative nutrition consultations were defined as occurring between time of cancer diagnosis and surgery in either the inpatient or outpatient setting; inpatient postoperative nutrition consultations occurred during admission for surgery; readmission nutrition consultations occurred on readmission in inpatient setting, if applicable; outpatient postoperative nutrition consultations were defined as occurring up to 2 months postdischarge in the outpatient setting.

Albumin values were extracted as follows: Preoperative albumin levels were defined as up to 4 months prior to diagnosis, and postoperative albumin levels were defined as 2 to 6 months after surgery.

 

Analysis

The data were described using mean (SD) for continuous variables and number and percentages for categorical variables. Where appropriate, Fisher exact test, Pearson χ² test, Spearman ρ, and Mann-Whitney U test were used for tests of significance. SAS (SAS Institute) was utilized for multivariable analysis. The significance level was P = .05 for all tests.

Results

There were 115 patients in the CC cohort and 33 in the RC cohort. The mean (SD) age at diagnosis was 70 (9.1) for CC group and 59 (1.4) for RC group (Table 1). 

For the CC and RC cohorts, the mean (SD) pathologic tumor stage was 2.3 (0.8) and 1.6 (1.2), respectively; the mean (SD) nodal stages was 0.4 (0.7) for both groups. Most patients (84 [73%] in the CC cohort and 29 [88%] in the RC cohort) had at least 1 recorded comorbidity (ie, diabetes mellitus, peripheral arterial disease, coronary artery disease, or history of cerebral vascular accident/transient ischemic attack). Malnutrition as determined by BMI criteria was not highly prevalent in the patient cohort. At time of diagnosis, none of the patients were underweight (BMI < 18.5); 24 patients with CC (27%) and 5 patients with RC (15%) were overweight (BMI 25.0-29.9); and 43 patients with CC (48%) and 15 patients with RC (45%) were obese (BMI ≥ 30). CC and RC cohorts had a mean preoperative albumin of 3.71 and 3.60, respectively. Low preoperative albumin (< 3.5), was present in 25 patients with CC (22%) and in 11 patients with RC (33%).

Weight Trends

At time of diagnosis, the mean (SD) BMI was 29.9 (7.1) for the CC group and 30.9 (7.4) for the RC group. The mean (SD) time in days from diagnosis to the date of surgery for the CC group was 43.9 (26.8) and 172.1 (39.1) for the RC group. Mean changes in BMI ranged from -7.0% to +4.9% (Table 2).

From 1 year to 6 months before diagnosis, 40 of 80 patients lost weight in the CC cohort (mean change, +1.9%) and 6 of 22 patients lost weight in the RC cohort (mean change, + 0.5%). From 6 months before diagnosis to time of diagnosis, 47 of 74 patients lost weight in the CC cohort (mean change, -1.5%) and 14 of 21 patients lost weight in the RC cohort (mean change, -2.5%). From time of diagnosis to time of surgery, 36 of 104 patients with CC and 14 of 32 patients with RC lost weight with a mean weight change of and +0.1% and -0.3%, respectively. In the 6 months before surgery, any amount of weight loss was observed in 58 patients (66%) in the CC group and in 12 patients (57%) in the RC group. In this time frame, in the CC cohort, 32 patients (36%) were observed to have at least 3% weight loss, and 23 (26%) were observed to have at least 5% weight loss (Table 3). 

In the 6 months before surgery, 20 of 88 patients (23%) in the CC cohort lost ≥ 10 lb.

In patients who completed adjuvant chemotherapy in the CC group, mean (SD) BMI at the beginning and end of chemotherapy was 32.6 (8.6) and 33.1 (8.7), respectively, and a -0.3% mean change in weight was observed. In the RC group, mean (SD) BMI was 28.2 (5.0) at the initiation of adjuvant chemotherapy and 28.4 (5.0) at its completion, with a +2.6% mean change in weight.

In the immediate postoperative period, most patients were losing weight in both the CC and RC groups (mean, -3.5% and -7.0% at 1 month postoperative, respectively). At 1-year after surgery, patients had modest mean increases in weight: +1.3% for patients with CC and +4.9% for patients with RC.

A relatively large proportion of patients had missing data on weights at various data points (Table 4). 

Preoperative weight trends were hampered by this limitation. Of the 115 patients, only 62 (54%) had data available to evaluate weight dynamics from 1 year prior to time of diagnosis, 74 (64%) from 6 months prior to diagnosis, and 73 (63%) for 1 year prior to time of surgery. The trend from 6 months pre-diagnosis until the time of surgery allowed for the most complete analysis: Data were available for 88 of 115 patients (77%). Data were missing in 33 patients with RC as well; thus, data analysis is focused on the CC cohort.

 

Nutrition Consultations

In the CC group, preoperative nutrition consultations (either inpatient or outpatient) occurred in 17 patients (15%). Inpatient postoperative nutrition evaluations occurred in 110 patients (96%) (Table 5). 

  Of these inpatient postoperative consultations, either oral or enteral nutritional supplements were prescribed 26 times (24%). Patients had a postoperative outpatient nutrition follow-up within 2 months postdischarge in only 14 cases (12%). Of 15 patients who were readmitted to the hospital, 11 (69%) had a nutrition reconsultation on readmission.

In the RC group, preoperative inpatient or outpatient nutrition consultations occurred in 12 patients (36%). Eight of those occurred before initiation of neoadjuvant chemoradiotherapy. All 33 patients received an inpatient postoperative nutrition evaluation during admission. Oral or enteral nutrition supplements were prescribed 19 times (58%). Postoperative outpatient nutrition consultations occurred for 24 patients (73%). Of the 19 patients who were readmitted to the hospital, 3 (16%) had a nutrition reconsultation on readmission.

Outcomes

The primary outcomes observed were delayed recovery, hospital readmission and LOS, and completion of adjuvant chemotherapy as indicated. Delayed recovery was observed in 35 patients with CC (40%) and 21 patients with RC (64%). Multivariable analysis in the CC cohort demonstrated that weight change was significantly associated with delayed recovery. Among those with ≥ 3% weight loss in the 6-month preoperative period (the weight measurement 6 months prior to diagnosis to date of surgery), 20 patients (63%) had delayed recovery compared with 15 patients (27%) without ≥ 3% weight loss who experienced delayed recovery (χ² = 10.84; P < .001).

Weight loss of ≥ 3% in the 6-month preoperative period also was significantly associated with complications. Of patients with at least 3% preoperative weight loss, 16 (50%) experienced complications, while 8 (14%) with < 3% preoperative weight loss experienced complications (χ² = 11.20; P < .001). Notably, ≥ 3% weight loss in the 1-year preoperative period before surgery was not significantly associated with delayed recovery. Any degree of 30-day postoperative weight loss was not correlated with delayed recovery. Finally, low preoperative albumin also was not correlated with delayed recovery (Fisher exact; P = .13). Table 3 displays differences based on presence of delayed recovery in the 88 patients with CC 6 months before surgery. Of note, ≥ 10-lb weight loss in the 6 months preceding surgery also correlated with delayed recovery (P = .01).In our cohort, 3% weight loss over 6 months had a sensitivity of 57%, specificity of 77%, positive predictive value 63%, and negative predictive value 73% for delayed recovery. By comparison, a 10-lb weight loss in 6 months per ASPEN criteria had a sensitivity of 40%, specificity of 85%, positive predictive value 64%, and negative predictive value 68% for delayed recovery.

Hospital Readmissions and LOS

Hospital readmissions occurred within the first 30 days in 11 patients (10%) in the CC cohort and 12 patients (36%) in the RC cohort. Readmissions occurred between 31 and 60 days in 4 (3%) and 7 (21%) of CC and RC cohorts, respectively. The presence of ≥ 3% weight loss in the 6-month preoperative period was significantly associated with 60-day readmissions. Among those with ≥ 3% preoperative weight loss, 8 patients (25%) had readmissions within 60 days vs 3 patients (5%) without 3% preoperative weight loss who had readmissions within 60 days (Fisher exact; P = .02).

 

Mean (SD) LOS was 6.4 (4.7) days (range, 1-28) for patients with CC and 8.8 (5.1) days (range, 3-23) for patients with RC. Mean (SD) LOS increased to 10.2 (4.3) days and 9.7 (6.0) days in patients with delayed recovery in the CC and RC cohorts, respectively. The mean (SD) LOS was 5.2 (2.8) days and 6.3 (2.2) days in patients without delayed recovery in the CC and RC cohorts, respectively. There was no significant difference when examining association between percent weight change and LOS for either initial admission (r_s = -0.1409; 2-tailed P = .19) or for initial and readmission combined (r_s = -0.13532; 2-tailed P = .21) within the CC cohort.

Chemotherapy

Within the CC cohort, 31 patients (27%) had an indication for adjuvant chemotherapy. Of these, 25 of 31 (81%) started chemotherapy within 12 weeks of surgical resection, and of these, 17 of 25 patients (68%) completed chemotherapy as indicated. Within the RC cohort all 33 patients had an indication for adjuvant chemotherapy, of these 18 of 33 patients (55%) began within 12 weeks of surgical resection, and 10 of 18 (56%) completed chemotherapy as indicated.

Among the CC cohort who began but did not complete adjuvant chemotherapy, there was no significant association between completion of chemotherapy and preoperative weight loss. Of patients with ≥ 3% preoperative weight loss, 3 (43%) did not complete chemotherapy vs 3 patients (27%) without preoperative weight loss who did not complete chemotherapy as indicated (Fisher exact; P = .63). Finally, low preoperative albumin did not significantly correlate with lack of completion of adjuvant chemotherapy (Fisher exact; P = .99).

Discussion

This study highlights several important findings. There were no patients in our cohort that met ASPEN malnourishment criteria with a BMI < 18.5. Twenty percent of patients lost at least 10 lb in 6 months before the operation. Notably, patients had significant associations with adverse outcomes with less pronounced weight loss than previously noted. As has been established previously, malnourishment can be difficult to screen for, and BMI also is often an imprecise tool.¹² In the CC cohort, weight loss ≥ 3% from 6 months before surgery was significantly associated with delayed recovery, complications, and hospital readmissions. We did not identify a statistically significant effect on chemotherapy completion. However, the numerical difference was suggestive of a difference, and a type 2 error is possible due to our limited sample size.

Our findings imply that the effects of even mild malnutrition are even more profound than previously thought. Significantly, this applies to overweight and obese patients as well, as these constituted a significant fraction of our cohort. A finding of ≥ 3% weight loss at the time of CC diagnosis may provide an opportunity for a focused nutrition intervention up to the time of surgery. Second, although nutrition consultation was frequent in the inpatient setting during the hospital admission (96%-100%), rates of nutrition evaluation were as low as 15% before surgery and 12% after surgery, representing a key area for improvement and focused intervention. An optimal time for intervention and nutrition prehabilitation would be at time of diagnosis before surgery with plans for continued aggressive monitoring and subsequent follow-up. Our finding seems to provide a more sensitive tool to identify patients at risk for delayed recovery compared with the ASPEN-driven assessment. Given the simplicity and the clinical significance, our test consisting of 3% weight loss over 6 months, with its sensitivity of 57%, may be superior to the ASPEN 10-lb weight loss, with its sensitivity of 40% in our cohort.

 

Previous Studies

Our findings are consistent with previous studies that have demonstrated that perioperative weight loss and malnutrition are correlated with delayed recovery and complications, such as wound healing, in patients with GI cancer.^2,4,5,8 In a retrospective study of more than 7000 patients with CC, those who were overweight or obese were found to have an improved overall survival compared with other BMI categories, and those who were underweight had an increased 30-day mortality and postoperative complications.¹⁶

In another retrospective study of 3799 patients with CC, those who were overweight and obese had an improved 5-year survival rate compared with patients whose weight was normal or underweight. Outcomes were found to be stage dependent.¹⁷ In this study cohort, all patients were either overweight or obese and remained in that category even with weight loss. This may have contributed to overall improved outcomes.

Implications and Next Steps

Our study has several implications. One is that BMI criteria < 18.5 may not be a good measure for malnutrition given that about 75% of the patients in our cohort were overweight or obese and none were underweight. We also show a concrete, easily identifiable finding of percent weight change that could be addressed as an automated electronic notification and potentially identify a patient at risk and serve as a trigger for both timely and early nutrition intervention. It seems to be more sensitive than the ASPEN criterion of 10-lb weight loss in 6 months before surgery. Sensitivity is especially appealing given the ease and potential of embedding this tool in an electronic health record and the clinical importance of the consequent intervention. Preoperative as opposed to perioperative nutrition optimization at time of CC diagnosis is essential, as it may help improve postsurgical outcomes as well as oncologic outcomes, including completion of adjuvant chemotherapy. Finally, although our study found that rates of inpatient postoperative nutrition consultation were high, rates of outpatient nutrition consultation in the preoperative period were low. This represents a missed opportunity for intervention before surgery. Similarly, rates of postoperative nutrition follow-up period were low, which points to an area for improvement in longitudinal and holistic care.

We suggest modifications to nutrition intervention protocols, such as ERAS, which should start at the time of GI malignancy diagnosis.¹⁸ Other suggestions include standard involvement of nutritionists in inpatient and outpatient settings with longitudinal follow-up in the preoperative and postoperative periods and patient enrollment in a nutrition program with monitoring at time of diagnosis at the VHA. Our findings as well as previous literature suggest that the preoperative period is the most important time to intervene with regard to nutrition optimization and represents an opportunity for intensive prehabilitation. Future areas of research include incorporating other important measures of malnourishment independent of BMI into future study designs, such as sarcopenia and adipose tissue density, to better assess body composition and predict prognostic risk in CC.^18,19

Strengths and Limitations

This study is limited by its single-center, retrospective design and small sample sizes, and we acknowledge the limitations of our data set. However, the strength of this VHA-based study is that the single-payer system allows for complete capture of perioperative data as well as the opportunity for focused preoperative interventions to improve outcomes. To our knowledge, there is no currently existing literature on improving nutrition protocols at the VHA for patients with a GI malignancy. These retrospective data will help inform current gaps in quality improvement and supportive oncology as it relates to optimizing malnourishment in veterans undergoing surgical resection for their cancer.

 

Conclusions

In the CC cohort, weight loss of ≥ 3% from 6 months prior to time of surgery was significantly associated with delayed recovery, complications, and hospital readmissions. Our findings suggest that patients with CC undergoing surgery may benefit from an intensive, early nutrition prehabilitation. Preoperative nutrition optimization may help improve postsurgical outcomes as well as oncologic outcomes, including completion of adjuvant chemotherapy. Further research would be able to clarify these hypotheses.

In patients with gastrointestinal (GI) malignancies, malnutrition is common. In addition, it has various negative implications, including high risk for surgical complications, prolonged hospitalization, decreased quality of life (QOL), increased mortality, and poor tolerance for treatments such as chemotherapy and radiotherapy.¹

A 2014 French study of 1903 patients hospitalized for cancer reported a 39% overall prevalence of malnutrition; 39% in patients with cancers of the colon/rectum, 60% for pancreatic cancer, and 67% for cancers of the esophagus/stomach.² Malnutrition was defined as body mass index (BMI) < 18.5 for individuals aged < 75 years or BMI < 21 for individuals aged ≥ 75 years, and/or weight loss > 10% since disease onset. Malnutrition also was strongly associated with worsened performance status.

The etiology of malnutrition in GI cancers is often multifactorial. It includes systemic tumor effects, such as inflammatory mediators contributing to hypermetabolism and cachexia, local tumor-associated mechanical obstruction, GI toxicities caused by antineoplastic therapy or other medications, and psychological factors that contribute to anorexia.³ Patient-related risk factors such as older age, other chronic diseases, and history of other GI surgeries also play a role.¹

Other studies have demonstrated that malnutrition in patients with GI malignancies undergoing surgical resection is associated with high rates of severe postoperative complications, increased length of stay (LOS) and time on a ventilator for patients treated in the intensive care unit, and poor QOL in the postoperative survival period.^4-6 Several randomized controlled trials conducted in patients with GI cancers have shown that enteral and parenteral nutrition supplementations in the perioperative period improve various outcomes, such as reduction of postoperative complication rates, fewer readmissions, improved chemotherapy tolerance, and improved QOL.^7-10 Thus, in the management of patients with GI malignancies, it is highly important to implement early nutritional screening and establish a diagnosis of malnutrition to intervene and reduce postoperative morbidity and mortality.¹

However, tools and predictors of malnutrition are often imperfect. The Academy of Nutrition and Dietetics and the American Society for Parenteral and Enteral Nutrition (AND/ASPEN) weight-based criteria define malnutrition and nutritionally-at-risk as BMI < 18.5, involuntary loss of at least 10% of body weight within 6 months or 5% within 1 month, or loss of 10 lb within 6 months.¹¹ While the ASPEN criteria are often used to define malnourishment, they may not fully capture the population at risk, and there does not exist a gold-standard tool for nutritional screening. A 2002 study that performed a critical appraisal of 44 nutritional screening tools found that no single tool was fully sufficient for application, development, evaluation, and consistent screening.¹² As such, consistently screening for malnutrition to target interventions in the perioperative period for GI surgical oncology has been challenging.¹³ More recent tools such as the perioperative nutrition screen (PONS) have been validated as rapid, effective screening tools to predict postoperative outcomes.¹⁴ Additionally, implementation of perioperative nutritional protocols, such as enhanced recovery after surgery (ERAS) in colon cancer (CC) surgery, also has shown improved perioperative care and outcomes.¹⁵

Preoperative nutritional interventions have been implemented in practice and have focused mostly on the immediate perioperative period. This has been shown to improve surgical outcomes. The Veterans Health Administration (VHA) provides comprehensive care to patients in a single-payer system, allowing for capture of perioperative data and the opportunity for focused preoperative interventions to improve outcomes.

 

Methods

This was a retrospective record review of colorectal malignancies treated with curative intent at the Veterans Affairs Ann Arbor Healthcare System (VAAAHS) in Michigan between January 1, 2015, and December 31, 2019. We examined nutritional status, degree of longitudinal weight loss, and subsequent clinical outcomes, including delayed postoperative recovery and delays in chemotherapy in 115 patients with CC and 33 patients with rectal cancer (RC) undergoing curative surgical resection at VAAAHS. To avoid additional confounding effects of advanced cancer, only early-stage, curable disease was included. This study was approved by the VAAAHS Institutional Review Board.

Patient cases were identified in the VAAAHS cancer registry, which included 115 patients with colon primary tumors and 33 patients with rectal tumors. According to the VAAAHS standard of care, patients with CC did not require neoadjuvant therapy while patients with RC cohort did (Figure). The CC cohort was defined as patients who had an adenocarcinoma, mucinous adenocarcinoma, or carcinoid tumor of the colon or rectosigmoid junction. These patients did not receive neoadjuvant therapy and underwent curative-intent surgical resection of their tumor. The RC cohort was defined as patients who had adenocarcinoma, mucinous adenocarcinoma, or signet ring cell carcinoma of the rectum. These patients received neoadjuvant chemoradiation followed by curative-intent surgical resection of their tumor.

Patients with postoperative follow-up outside of VAAAHS were excluded. Patients were excluded if their surgery had noncurative intent or if they had distant metastatic disease. Data on patient weights, laboratory results, nutrition consultations, postoperative complications, delayed recovery, readmissions, and chemotherapy tolerance were abstracted by patient chart review in the VHA Computerized Patient Record System and Joint Legacy Viewer by 2 researchers.

Delayed recovery was defined as any abnormal clinical development described in inpatient progress notes, outpatient follow-up notes within 60 days, or in hospital discharge summaries. Excluded were psychiatric events without additional medical complications, postoperative bleeding not requiring an invasive intervention, urinary retention, postoperative glycemic control difficulties, cardiac events that happened before postoperative hospital discharge and not requiring readmission, and postoperative alcohol withdrawal. Complications were defined similarly to delayed recovery but excluded isolated prolonged postoperative ileus. LOS was defined in days as time from admission to discharge.

Adjuvant management course was derived from reviewing documentation from medical oncology consultations and progress notes. In patients for whom adjuvant chemotherapy was indicated and prescribed, chemotherapy was considered complete if chemotherapy was started and completed as indicated. Adjuvant chemotherapy was considered incomplete if the patient declined chemotherapy, if chemotherapy was not started when indicated, or if chemotherapy was not completed as indicated. Neoadjuvant therapy data were abstracted from medical and radiation oncology notes.

Recorded data were collected on both weight and BMI. Weights were extracted as follows: Weight 1 year before time of diagnosis, ± 4 months; weight 6 months before diagnosis ± 3 months; weight at time of diagnosis ± 2 weeks; weight at time of surgery ± 2 weeks; weight 30 days postsurgery ± 2 weeks; weight 60 days postsurgery ± 2 weeks; weight 1 year postsurgery ± 4 months. Mean percent change in weight was calculated from recorded weights between each allocated time point. A weight loss of ≥ 3% was found to be clinically relevant and was chosen as the minimal cutoff value when analyzing outcomes associated with weight trends.

Nutrition consultations were abstracted as follows: Preoperative nutrition consultations were defined as occurring between time of cancer diagnosis and surgery in either the inpatient or outpatient setting; inpatient postoperative nutrition consultations occurred during admission for surgery; readmission nutrition consultations occurred on readmission in inpatient setting, if applicable; outpatient postoperative nutrition consultations were defined as occurring up to 2 months postdischarge in the outpatient setting.

Albumin values were extracted as follows: Preoperative albumin levels were defined as up to 4 months prior to diagnosis, and postoperative albumin levels were defined as 2 to 6 months after surgery.

 

Analysis

The data were described using mean (SD) for continuous variables and number and percentages for categorical variables. Where appropriate, Fisher exact test, Pearson χ² test, Spearman ρ, and Mann-Whitney U test were used for tests of significance. SAS (SAS Institute) was utilized for multivariable analysis. The significance level was P = .05 for all tests.

Results

There were 115 patients in the CC cohort and 33 in the RC cohort. The mean (SD) age at diagnosis was 70 (9.1) for CC group and 59 (1.4) for RC group (Table 1). 

For the CC and RC cohorts, the mean (SD) pathologic tumor stage was 2.3 (0.8) and 1.6 (1.2), respectively; the mean (SD) nodal stages was 0.4 (0.7) for both groups. Most patients (84 [73%] in the CC cohort and 29 [88%] in the RC cohort) had at least 1 recorded comorbidity (ie, diabetes mellitus, peripheral arterial disease, coronary artery disease, or history of cerebral vascular accident/transient ischemic attack). Malnutrition as determined by BMI criteria was not highly prevalent in the patient cohort. At time of diagnosis, none of the patients were underweight (BMI < 18.5); 24 patients with CC (27%) and 5 patients with RC (15%) were overweight (BMI 25.0-29.9); and 43 patients with CC (48%) and 15 patients with RC (45%) were obese (BMI ≥ 30). CC and RC cohorts had a mean preoperative albumin of 3.71 and 3.60, respectively. Low preoperative albumin (< 3.5), was present in 25 patients with CC (22%) and in 11 patients with RC (33%).

Weight Trends

At time of diagnosis, the mean (SD) BMI was 29.9 (7.1) for the CC group and 30.9 (7.4) for the RC group. The mean (SD) time in days from diagnosis to the date of surgery for the CC group was 43.9 (26.8) and 172.1 (39.1) for the RC group. Mean changes in BMI ranged from -7.0% to +4.9% (Table 2).

From 1 year to 6 months before diagnosis, 40 of 80 patients lost weight in the CC cohort (mean change, +1.9%) and 6 of 22 patients lost weight in the RC cohort (mean change, + 0.5%). From 6 months before diagnosis to time of diagnosis, 47 of 74 patients lost weight in the CC cohort (mean change, -1.5%) and 14 of 21 patients lost weight in the RC cohort (mean change, -2.5%). From time of diagnosis to time of surgery, 36 of 104 patients with CC and 14 of 32 patients with RC lost weight with a mean weight change of and +0.1% and -0.3%, respectively. In the 6 months before surgery, any amount of weight loss was observed in 58 patients (66%) in the CC group and in 12 patients (57%) in the RC group. In this time frame, in the CC cohort, 32 patients (36%) were observed to have at least 3% weight loss, and 23 (26%) were observed to have at least 5% weight loss (Table 3). 

In the 6 months before surgery, 20 of 88 patients (23%) in the CC cohort lost ≥ 10 lb.

In patients who completed adjuvant chemotherapy in the CC group, mean (SD) BMI at the beginning and end of chemotherapy was 32.6 (8.6) and 33.1 (8.7), respectively, and a -0.3% mean change in weight was observed. In the RC group, mean (SD) BMI was 28.2 (5.0) at the initiation of adjuvant chemotherapy and 28.4 (5.0) at its completion, with a +2.6% mean change in weight.

In the immediate postoperative period, most patients were losing weight in both the CC and RC groups (mean, -3.5% and -7.0% at 1 month postoperative, respectively). At 1-year after surgery, patients had modest mean increases in weight: +1.3% for patients with CC and +4.9% for patients with RC.

A relatively large proportion of patients had missing data on weights at various data points (Table 4). 

Preoperative weight trends were hampered by this limitation. Of the 115 patients, only 62 (54%) had data available to evaluate weight dynamics from 1 year prior to time of diagnosis, 74 (64%) from 6 months prior to diagnosis, and 73 (63%) for 1 year prior to time of surgery. The trend from 6 months pre-diagnosis until the time of surgery allowed for the most complete analysis: Data were available for 88 of 115 patients (77%). Data were missing in 33 patients with RC as well; thus, data analysis is focused on the CC cohort.

 

Nutrition Consultations

In the CC group, preoperative nutrition consultations (either inpatient or outpatient) occurred in 17 patients (15%). Inpatient postoperative nutrition evaluations occurred in 110 patients (96%) (Table 5). 

  Of these inpatient postoperative consultations, either oral or enteral nutritional supplements were prescribed 26 times (24%). Patients had a postoperative outpatient nutrition follow-up within 2 months postdischarge in only 14 cases (12%). Of 15 patients who were readmitted to the hospital, 11 (69%) had a nutrition reconsultation on readmission.

In the RC group, preoperative inpatient or outpatient nutrition consultations occurred in 12 patients (36%). Eight of those occurred before initiation of neoadjuvant chemoradiotherapy. All 33 patients received an inpatient postoperative nutrition evaluation during admission. Oral or enteral nutrition supplements were prescribed 19 times (58%). Postoperative outpatient nutrition consultations occurred for 24 patients (73%). Of the 19 patients who were readmitted to the hospital, 3 (16%) had a nutrition reconsultation on readmission.

Outcomes

The primary outcomes observed were delayed recovery, hospital readmission and LOS, and completion of adjuvant chemotherapy as indicated. Delayed recovery was observed in 35 patients with CC (40%) and 21 patients with RC (64%). Multivariable analysis in the CC cohort demonstrated that weight change was significantly associated with delayed recovery. Among those with ≥ 3% weight loss in the 6-month preoperative period (the weight measurement 6 months prior to diagnosis to date of surgery), 20 patients (63%) had delayed recovery compared with 15 patients (27%) without ≥ 3% weight loss who experienced delayed recovery (χ² = 10.84; P < .001).

Weight loss of ≥ 3% in the 6-month preoperative period also was significantly associated with complications. Of patients with at least 3% preoperative weight loss, 16 (50%) experienced complications, while 8 (14%) with < 3% preoperative weight loss experienced complications (χ² = 11.20; P < .001). Notably, ≥ 3% weight loss in the 1-year preoperative period before surgery was not significantly associated with delayed recovery. Any degree of 30-day postoperative weight loss was not correlated with delayed recovery. Finally, low preoperative albumin also was not correlated with delayed recovery (Fisher exact; P = .13). Table 3 displays differences based on presence of delayed recovery in the 88 patients with CC 6 months before surgery. Of note, ≥ 10-lb weight loss in the 6 months preceding surgery also correlated with delayed recovery (P = .01).In our cohort, 3% weight loss over 6 months had a sensitivity of 57%, specificity of 77%, positive predictive value 63%, and negative predictive value 73% for delayed recovery. By comparison, a 10-lb weight loss in 6 months per ASPEN criteria had a sensitivity of 40%, specificity of 85%, positive predictive value 64%, and negative predictive value 68% for delayed recovery.

Hospital Readmissions and LOS

Hospital readmissions occurred within the first 30 days in 11 patients (10%) in the CC cohort and 12 patients (36%) in the RC cohort. Readmissions occurred between 31 and 60 days in 4 (3%) and 7 (21%) of CC and RC cohorts, respectively. The presence of ≥ 3% weight loss in the 6-month preoperative period was significantly associated with 60-day readmissions. Among those with ≥ 3% preoperative weight loss, 8 patients (25%) had readmissions within 60 days vs 3 patients (5%) without 3% preoperative weight loss who had readmissions within 60 days (Fisher exact; P = .02).

 

Mean (SD) LOS was 6.4 (4.7) days (range, 1-28) for patients with CC and 8.8 (5.1) days (range, 3-23) for patients with RC. Mean (SD) LOS increased to 10.2 (4.3) days and 9.7 (6.0) days in patients with delayed recovery in the CC and RC cohorts, respectively. The mean (SD) LOS was 5.2 (2.8) days and 6.3 (2.2) days in patients without delayed recovery in the CC and RC cohorts, respectively. There was no significant difference when examining association between percent weight change and LOS for either initial admission (r_s = -0.1409; 2-tailed P = .19) or for initial and readmission combined (r_s = -0.13532; 2-tailed P = .21) within the CC cohort.

Chemotherapy

Within the CC cohort, 31 patients (27%) had an indication for adjuvant chemotherapy. Of these, 25 of 31 (81%) started chemotherapy within 12 weeks of surgical resection, and of these, 17 of 25 patients (68%) completed chemotherapy as indicated. Within the RC cohort all 33 patients had an indication for adjuvant chemotherapy, of these 18 of 33 patients (55%) began within 12 weeks of surgical resection, and 10 of 18 (56%) completed chemotherapy as indicated.

Among the CC cohort who began but did not complete adjuvant chemotherapy, there was no significant association between completion of chemotherapy and preoperative weight loss. Of patients with ≥ 3% preoperative weight loss, 3 (43%) did not complete chemotherapy vs 3 patients (27%) without preoperative weight loss who did not complete chemotherapy as indicated (Fisher exact; P = .63). Finally, low preoperative albumin did not significantly correlate with lack of completion of adjuvant chemotherapy (Fisher exact; P = .99).

Discussion

This study highlights several important findings. There were no patients in our cohort that met ASPEN malnourishment criteria with a BMI < 18.5. Twenty percent of patients lost at least 10 lb in 6 months before the operation. Notably, patients had significant associations with adverse outcomes with less pronounced weight loss than previously noted. As has been established previously, malnourishment can be difficult to screen for, and BMI also is often an imprecise tool.¹² In the CC cohort, weight loss ≥ 3% from 6 months before surgery was significantly associated with delayed recovery, complications, and hospital readmissions. We did not identify a statistically significant effect on chemotherapy completion. However, the numerical difference was suggestive of a difference, and a type 2 error is possible due to our limited sample size.

Our findings imply that the effects of even mild malnutrition are even more profound than previously thought. Significantly, this applies to overweight and obese patients as well, as these constituted a significant fraction of our cohort. A finding of ≥ 3% weight loss at the time of CC diagnosis may provide an opportunity for a focused nutrition intervention up to the time of surgery. Second, although nutrition consultation was frequent in the inpatient setting during the hospital admission (96%-100%), rates of nutrition evaluation were as low as 15% before surgery and 12% after surgery, representing a key area for improvement and focused intervention. An optimal time for intervention and nutrition prehabilitation would be at time of diagnosis before surgery with plans for continued aggressive monitoring and subsequent follow-up. Our finding seems to provide a more sensitive tool to identify patients at risk for delayed recovery compared with the ASPEN-driven assessment. Given the simplicity and the clinical significance, our test consisting of 3% weight loss over 6 months, with its sensitivity of 57%, may be superior to the ASPEN 10-lb weight loss, with its sensitivity of 40% in our cohort.

 

Previous Studies

Our findings are consistent with previous studies that have demonstrated that perioperative weight loss and malnutrition are correlated with delayed recovery and complications, such as wound healing, in patients with GI cancer.^2,4,5,8 In a retrospective study of more than 7000 patients with CC, those who were overweight or obese were found to have an improved overall survival compared with other BMI categories, and those who were underweight had an increased 30-day mortality and postoperative complications.¹⁶

In another retrospective study of 3799 patients with CC, those who were overweight and obese had an improved 5-year survival rate compared with patients whose weight was normal or underweight. Outcomes were found to be stage dependent.¹⁷ In this study cohort, all patients were either overweight or obese and remained in that category even with weight loss. This may have contributed to overall improved outcomes.

Implications and Next Steps

Our study has several implications. One is that BMI criteria < 18.5 may not be a good measure for malnutrition given that about 75% of the patients in our cohort were overweight or obese and none were underweight. We also show a concrete, easily identifiable finding of percent weight change that could be addressed as an automated electronic notification and potentially identify a patient at risk and serve as a trigger for both timely and early nutrition intervention. It seems to be more sensitive than the ASPEN criterion of 10-lb weight loss in 6 months before surgery. Sensitivity is especially appealing given the ease and potential of embedding this tool in an electronic health record and the clinical importance of the consequent intervention. Preoperative as opposed to perioperative nutrition optimization at time of CC diagnosis is essential, as it may help improve postsurgical outcomes as well as oncologic outcomes, including completion of adjuvant chemotherapy. Finally, although our study found that rates of inpatient postoperative nutrition consultation were high, rates of outpatient nutrition consultation in the preoperative period were low. This represents a missed opportunity for intervention before surgery. Similarly, rates of postoperative nutrition follow-up period were low, which points to an area for improvement in longitudinal and holistic care.

We suggest modifications to nutrition intervention protocols, such as ERAS, which should start at the time of GI malignancy diagnosis.¹⁸ Other suggestions include standard involvement of nutritionists in inpatient and outpatient settings with longitudinal follow-up in the preoperative and postoperative periods and patient enrollment in a nutrition program with monitoring at time of diagnosis at the VHA. Our findings as well as previous literature suggest that the preoperative period is the most important time to intervene with regard to nutrition optimization and represents an opportunity for intensive prehabilitation. Future areas of research include incorporating other important measures of malnourishment independent of BMI into future study designs, such as sarcopenia and adipose tissue density, to better assess body composition and predict prognostic risk in CC.^18,19

Strengths and Limitations

This study is limited by its single-center, retrospective design and small sample sizes, and we acknowledge the limitations of our data set. However, the strength of this VHA-based study is that the single-payer system allows for complete capture of perioperative data as well as the opportunity for focused preoperative interventions to improve outcomes. To our knowledge, there is no currently existing literature on improving nutrition protocols at the VHA for patients with a GI malignancy. These retrospective data will help inform current gaps in quality improvement and supportive oncology as it relates to optimizing malnourishment in veterans undergoing surgical resection for their cancer.

 

Conclusions

In the CC cohort, weight loss of ≥ 3% from 6 months prior to time of surgery was significantly associated with delayed recovery, complications, and hospital readmissions. Our findings suggest that patients with CC undergoing surgery may benefit from an intensive, early nutrition prehabilitation. Preoperative nutrition optimization may help improve postsurgical outcomes as well as oncologic outcomes, including completion of adjuvant chemotherapy. Further research would be able to clarify these hypotheses.

In patients with gastrointestinal (GI) malignancies, malnutrition is common. In addition, it has various negative implications, including high risk for surgical complications, prolonged hospitalization, decreased quality of life (QOL), increased mortality, and poor tolerance for treatments such as chemotherapy and radiotherapy.¹

A 2014 French study of 1903 patients hospitalized for cancer reported a 39% overall prevalence of malnutrition; 39% in patients with cancers of the colon/rectum, 60% for pancreatic cancer, and 67% for cancers of the esophagus/stomach.² Malnutrition was defined as body mass index (BMI) < 18.5 for individuals aged < 75 years or BMI < 21 for individuals aged ≥ 75 years, and/or weight loss > 10% since disease onset. Malnutrition also was strongly associated with worsened performance status.

The etiology of malnutrition in GI cancers is often multifactorial. It includes systemic tumor effects, such as inflammatory mediators contributing to hypermetabolism and cachexia, local tumor-associated mechanical obstruction, GI toxicities caused by antineoplastic therapy or other medications, and psychological factors that contribute to anorexia.³ Patient-related risk factors such as older age, other chronic diseases, and history of other GI surgeries also play a role.¹

Other studies have demonstrated that malnutrition in patients with GI malignancies undergoing surgical resection is associated with high rates of severe postoperative complications, increased length of stay (LOS) and time on a ventilator for patients treated in the intensive care unit, and poor QOL in the postoperative survival period.^4-6 Several randomized controlled trials conducted in patients with GI cancers have shown that enteral and parenteral nutrition supplementations in the perioperative period improve various outcomes, such as reduction of postoperative complication rates, fewer readmissions, improved chemotherapy tolerance, and improved QOL.^7-10 Thus, in the management of patients with GI malignancies, it is highly important to implement early nutritional screening and establish a diagnosis of malnutrition to intervene and reduce postoperative morbidity and mortality.¹

However, tools and predictors of malnutrition are often imperfect. The Academy of Nutrition and Dietetics and the American Society for Parenteral and Enteral Nutrition (AND/ASPEN) weight-based criteria define malnutrition and nutritionally-at-risk as BMI < 18.5, involuntary loss of at least 10% of body weight within 6 months or 5% within 1 month, or loss of 10 lb within 6 months.¹¹ While the ASPEN criteria are often used to define malnourishment, they may not fully capture the population at risk, and there does not exist a gold-standard tool for nutritional screening. A 2002 study that performed a critical appraisal of 44 nutritional screening tools found that no single tool was fully sufficient for application, development, evaluation, and consistent screening.¹² As such, consistently screening for malnutrition to target interventions in the perioperative period for GI surgical oncology has been challenging.¹³ More recent tools such as the perioperative nutrition screen (PONS) have been validated as rapid, effective screening tools to predict postoperative outcomes.¹⁴ Additionally, implementation of perioperative nutritional protocols, such as enhanced recovery after surgery (ERAS) in colon cancer (CC) surgery, also has shown improved perioperative care and outcomes.¹⁵

Preoperative nutritional interventions have been implemented in practice and have focused mostly on the immediate perioperative period. This has been shown to improve surgical outcomes. The Veterans Health Administration (VHA) provides comprehensive care to patients in a single-payer system, allowing for capture of perioperative data and the opportunity for focused preoperative interventions to improve outcomes.

 

Methods

This was a retrospective record review of colorectal malignancies treated with curative intent at the Veterans Affairs Ann Arbor Healthcare System (VAAAHS) in Michigan between January 1, 2015, and December 31, 2019. We examined nutritional status, degree of longitudinal weight loss, and subsequent clinical outcomes, including delayed postoperative recovery and delays in chemotherapy in 115 patients with CC and 33 patients with rectal cancer (RC) undergoing curative surgical resection at VAAAHS. To avoid additional confounding effects of advanced cancer, only early-stage, curable disease was included. This study was approved by the VAAAHS Institutional Review Board.

Patient cases were identified in the VAAAHS cancer registry, which included 115 patients with colon primary tumors and 33 patients with rectal tumors. According to the VAAAHS standard of care, patients with CC did not require neoadjuvant therapy while patients with RC cohort did (Figure). The CC cohort was defined as patients who had an adenocarcinoma, mucinous adenocarcinoma, or carcinoid tumor of the colon or rectosigmoid junction. These patients did not receive neoadjuvant therapy and underwent curative-intent surgical resection of their tumor. The RC cohort was defined as patients who had adenocarcinoma, mucinous adenocarcinoma, or signet ring cell carcinoma of the rectum. These patients received neoadjuvant chemoradiation followed by curative-intent surgical resection of their tumor.

Patients with postoperative follow-up outside of VAAAHS were excluded. Patients were excluded if their surgery had noncurative intent or if they had distant metastatic disease. Data on patient weights, laboratory results, nutrition consultations, postoperative complications, delayed recovery, readmissions, and chemotherapy tolerance were abstracted by patient chart review in the VHA Computerized Patient Record System and Joint Legacy Viewer by 2 researchers.

Delayed recovery was defined as any abnormal clinical development described in inpatient progress notes, outpatient follow-up notes within 60 days, or in hospital discharge summaries. Excluded were psychiatric events without additional medical complications, postoperative bleeding not requiring an invasive intervention, urinary retention, postoperative glycemic control difficulties, cardiac events that happened before postoperative hospital discharge and not requiring readmission, and postoperative alcohol withdrawal. Complications were defined similarly to delayed recovery but excluded isolated prolonged postoperative ileus. LOS was defined in days as time from admission to discharge.

Adjuvant management course was derived from reviewing documentation from medical oncology consultations and progress notes. In patients for whom adjuvant chemotherapy was indicated and prescribed, chemotherapy was considered complete if chemotherapy was started and completed as indicated. Adjuvant chemotherapy was considered incomplete if the patient declined chemotherapy, if chemotherapy was not started when indicated, or if chemotherapy was not completed as indicated. Neoadjuvant therapy data were abstracted from medical and radiation oncology notes.

Recorded data were collected on both weight and BMI. Weights were extracted as follows: Weight 1 year before time of diagnosis, ± 4 months; weight 6 months before diagnosis ± 3 months; weight at time of diagnosis ± 2 weeks; weight at time of surgery ± 2 weeks; weight 30 days postsurgery ± 2 weeks; weight 60 days postsurgery ± 2 weeks; weight 1 year postsurgery ± 4 months. Mean percent change in weight was calculated from recorded weights between each allocated time point. A weight loss of ≥ 3% was found to be clinically relevant and was chosen as the minimal cutoff value when analyzing outcomes associated with weight trends.

Nutrition consultations were abstracted as follows: Preoperative nutrition consultations were defined as occurring between time of cancer diagnosis and surgery in either the inpatient or outpatient setting; inpatient postoperative nutrition consultations occurred during admission for surgery; readmission nutrition consultations occurred on readmission in inpatient setting, if applicable; outpatient postoperative nutrition consultations were defined as occurring up to 2 months postdischarge in the outpatient setting.

Albumin values were extracted as follows: Preoperative albumin levels were defined as up to 4 months prior to diagnosis, and postoperative albumin levels were defined as 2 to 6 months after surgery.

 

Analysis

The data were described using mean (SD) for continuous variables and number and percentages for categorical variables. Where appropriate, Fisher exact test, Pearson χ² test, Spearman ρ, and Mann-Whitney U test were used for tests of significance. SAS (SAS Institute) was utilized for multivariable analysis. The significance level was P = .05 for all tests.

Results

There were 115 patients in the CC cohort and 33 in the RC cohort. The mean (SD) age at diagnosis was 70 (9.1) for CC group and 59 (1.4) for RC group (Table 1). 

For the CC and RC cohorts, the mean (SD) pathologic tumor stage was 2.3 (0.8) and 1.6 (1.2), respectively; the mean (SD) nodal stages was 0.4 (0.7) for both groups. Most patients (84 [73%] in the CC cohort and 29 [88%] in the RC cohort) had at least 1 recorded comorbidity (ie, diabetes mellitus, peripheral arterial disease, coronary artery disease, or history of cerebral vascular accident/transient ischemic attack). Malnutrition as determined by BMI criteria was not highly prevalent in the patient cohort. At time of diagnosis, none of the patients were underweight (BMI < 18.5); 24 patients with CC (27%) and 5 patients with RC (15%) were overweight (BMI 25.0-29.9); and 43 patients with CC (48%) and 15 patients with RC (45%) were obese (BMI ≥ 30). CC and RC cohorts had a mean preoperative albumin of 3.71 and 3.60, respectively. Low preoperative albumin (< 3.5), was present in 25 patients with CC (22%) and in 11 patients with RC (33%).

Weight Trends

At time of diagnosis, the mean (SD) BMI was 29.9 (7.1) for the CC group and 30.9 (7.4) for the RC group. The mean (SD) time in days from diagnosis to the date of surgery for the CC group was 43.9 (26.8) and 172.1 (39.1) for the RC group. Mean changes in BMI ranged from -7.0% to +4.9% (Table 2).

From 1 year to 6 months before diagnosis, 40 of 80 patients lost weight in the CC cohort (mean change, +1.9%) and 6 of 22 patients lost weight in the RC cohort (mean change, + 0.5%). From 6 months before diagnosis to time of diagnosis, 47 of 74 patients lost weight in the CC cohort (mean change, -1.5%) and 14 of 21 patients lost weight in the RC cohort (mean change, -2.5%). From time of diagnosis to time of surgery, 36 of 104 patients with CC and 14 of 32 patients with RC lost weight with a mean weight change of and +0.1% and -0.3%, respectively. In the 6 months before surgery, any amount of weight loss was observed in 58 patients (66%) in the CC group and in 12 patients (57%) in the RC group. In this time frame, in the CC cohort, 32 patients (36%) were observed to have at least 3% weight loss, and 23 (26%) were observed to have at least 5% weight loss (Table 3). 

In the 6 months before surgery, 20 of 88 patients (23%) in the CC cohort lost ≥ 10 lb.

In patients who completed adjuvant chemotherapy in the CC group, mean (SD) BMI at the beginning and end of chemotherapy was 32.6 (8.6) and 33.1 (8.7), respectively, and a -0.3% mean change in weight was observed. In the RC group, mean (SD) BMI was 28.2 (5.0) at the initiation of adjuvant chemotherapy and 28.4 (5.0) at its completion, with a +2.6% mean change in weight.

In the immediate postoperative period, most patients were losing weight in both the CC and RC groups (mean, -3.5% and -7.0% at 1 month postoperative, respectively). At 1-year after surgery, patients had modest mean increases in weight: +1.3% for patients with CC and +4.9% for patients with RC.

A relatively large proportion of patients had missing data on weights at various data points (Table 4). 

Preoperative weight trends were hampered by this limitation. Of the 115 patients, only 62 (54%) had data available to evaluate weight dynamics from 1 year prior to time of diagnosis, 74 (64%) from 6 months prior to diagnosis, and 73 (63%) for 1 year prior to time of surgery. The trend from 6 months pre-diagnosis until the time of surgery allowed for the most complete analysis: Data were available for 88 of 115 patients (77%). Data were missing in 33 patients with RC as well; thus, data analysis is focused on the CC cohort.

 

Nutrition Consultations

In the CC group, preoperative nutrition consultations (either inpatient or outpatient) occurred in 17 patients (15%). Inpatient postoperative nutrition evaluations occurred in 110 patients (96%) (Table 5). 

  Of these inpatient postoperative consultations, either oral or enteral nutritional supplements were prescribed 26 times (24%). Patients had a postoperative outpatient nutrition follow-up within 2 months postdischarge in only 14 cases (12%). Of 15 patients who were readmitted to the hospital, 11 (69%) had a nutrition reconsultation on readmission.

In the RC group, preoperative inpatient or outpatient nutrition consultations occurred in 12 patients (36%). Eight of those occurred before initiation of neoadjuvant chemoradiotherapy. All 33 patients received an inpatient postoperative nutrition evaluation during admission. Oral or enteral nutrition supplements were prescribed 19 times (58%). Postoperative outpatient nutrition consultations occurred for 24 patients (73%). Of the 19 patients who were readmitted to the hospital, 3 (16%) had a nutrition reconsultation on readmission.

Outcomes

The primary outcomes observed were delayed recovery, hospital readmission and LOS, and completion of adjuvant chemotherapy as indicated. Delayed recovery was observed in 35 patients with CC (40%) and 21 patients with RC (64%). Multivariable analysis in the CC cohort demonstrated that weight change was significantly associated with delayed recovery. Among those with ≥ 3% weight loss in the 6-month preoperative period (the weight measurement 6 months prior to diagnosis to date of surgery), 20 patients (63%) had delayed recovery compared with 15 patients (27%) without ≥ 3% weight loss who experienced delayed recovery (χ² = 10.84; P < .001).

Weight loss of ≥ 3% in the 6-month preoperative period also was significantly associated with complications. Of patients with at least 3% preoperative weight loss, 16 (50%) experienced complications, while 8 (14%) with < 3% preoperative weight loss experienced complications (χ² = 11.20; P < .001). Notably, ≥ 3% weight loss in the 1-year preoperative period before surgery was not significantly associated with delayed recovery. Any degree of 30-day postoperative weight loss was not correlated with delayed recovery. Finally, low preoperative albumin also was not correlated with delayed recovery (Fisher exact; P = .13). Table 3 displays differences based on presence of delayed recovery in the 88 patients with CC 6 months before surgery. Of note, ≥ 10-lb weight loss in the 6 months preceding surgery also correlated with delayed recovery (P = .01).In our cohort, 3% weight loss over 6 months had a sensitivity of 57%, specificity of 77%, positive predictive value 63%, and negative predictive value 73% for delayed recovery. By comparison, a 10-lb weight loss in 6 months per ASPEN criteria had a sensitivity of 40%, specificity of 85%, positive predictive value 64%, and negative predictive value 68% for delayed recovery.

Hospital Readmissions and LOS

Hospital readmissions occurred within the first 30 days in 11 patients (10%) in the CC cohort and 12 patients (36%) in the RC cohort. Readmissions occurred between 31 and 60 days in 4 (3%) and 7 (21%) of CC and RC cohorts, respectively. The presence of ≥ 3% weight loss in the 6-month preoperative period was significantly associated with 60-day readmissions. Among those with ≥ 3% preoperative weight loss, 8 patients (25%) had readmissions within 60 days vs 3 patients (5%) without 3% preoperative weight loss who had readmissions within 60 days (Fisher exact; P = .02).

 

Mean (SD) LOS was 6.4 (4.7) days (range, 1-28) for patients with CC and 8.8 (5.1) days (range, 3-23) for patients with RC. Mean (SD) LOS increased to 10.2 (4.3) days and 9.7 (6.0) days in patients with delayed recovery in the CC and RC cohorts, respectively. The mean (SD) LOS was 5.2 (2.8) days and 6.3 (2.2) days in patients without delayed recovery in the CC and RC cohorts, respectively. There was no significant difference when examining association between percent weight change and LOS for either initial admission (r_s = -0.1409; 2-tailed P = .19) or for initial and readmission combined (r_s = -0.13532; 2-tailed P = .21) within the CC cohort.

Chemotherapy

Within the CC cohort, 31 patients (27%) had an indication for adjuvant chemotherapy. Of these, 25 of 31 (81%) started chemotherapy within 12 weeks of surgical resection, and of these, 17 of 25 patients (68%) completed chemotherapy as indicated. Within the RC cohort all 33 patients had an indication for adjuvant chemotherapy, of these 18 of 33 patients (55%) began within 12 weeks of surgical resection, and 10 of 18 (56%) completed chemotherapy as indicated.

Among the CC cohort who began but did not complete adjuvant chemotherapy, there was no significant association between completion of chemotherapy and preoperative weight loss. Of patients with ≥ 3% preoperative weight loss, 3 (43%) did not complete chemotherapy vs 3 patients (27%) without preoperative weight loss who did not complete chemotherapy as indicated (Fisher exact; P = .63). Finally, low preoperative albumin did not significantly correlate with lack of completion of adjuvant chemotherapy (Fisher exact; P = .99).

Discussion

This study highlights several important findings. There were no patients in our cohort that met ASPEN malnourishment criteria with a BMI < 18.5. Twenty percent of patients lost at least 10 lb in 6 months before the operation. Notably, patients had significant associations with adverse outcomes with less pronounced weight loss than previously noted. As has been established previously, malnourishment can be difficult to screen for, and BMI also is often an imprecise tool.¹² In the CC cohort, weight loss ≥ 3% from 6 months before surgery was significantly associated with delayed recovery, complications, and hospital readmissions. We did not identify a statistically significant effect on chemotherapy completion. However, the numerical difference was suggestive of a difference, and a type 2 error is possible due to our limited sample size.

Our findings imply that the effects of even mild malnutrition are even more profound than previously thought. Significantly, this applies to overweight and obese patients as well, as these constituted a significant fraction of our cohort. A finding of ≥ 3% weight loss at the time of CC diagnosis may provide an opportunity for a focused nutrition intervention up to the time of surgery. Second, although nutrition consultation was frequent in the inpatient setting during the hospital admission (96%-100%), rates of nutrition evaluation were as low as 15% before surgery and 12% after surgery, representing a key area for improvement and focused intervention. An optimal time for intervention and nutrition prehabilitation would be at time of diagnosis before surgery with plans for continued aggressive monitoring and subsequent follow-up. Our finding seems to provide a more sensitive tool to identify patients at risk for delayed recovery compared with the ASPEN-driven assessment. Given the simplicity and the clinical significance, our test consisting of 3% weight loss over 6 months, with its sensitivity of 57%, may be superior to the ASPEN 10-lb weight loss, with its sensitivity of 40% in our cohort.

 

Previous Studies

Our findings are consistent with previous studies that have demonstrated that perioperative weight loss and malnutrition are correlated with delayed recovery and complications, such as wound healing, in patients with GI cancer.^2,4,5,8 In a retrospective study of more than 7000 patients with CC, those who were overweight or obese were found to have an improved overall survival compared with other BMI categories, and those who were underweight had an increased 30-day mortality and postoperative complications.¹⁶

In another retrospective study of 3799 patients with CC, those who were overweight and obese had an improved 5-year survival rate compared with patients whose weight was normal or underweight. Outcomes were found to be stage dependent.¹⁷ In this study cohort, all patients were either overweight or obese and remained in that category even with weight loss. This may have contributed to overall improved outcomes.

Implications and Next Steps

Our study has several implications. One is that BMI criteria < 18.5 may not be a good measure for malnutrition given that about 75% of the patients in our cohort were overweight or obese and none were underweight. We also show a concrete, easily identifiable finding of percent weight change that could be addressed as an automated electronic notification and potentially identify a patient at risk and serve as a trigger for both timely and early nutrition intervention. It seems to be more sensitive than the ASPEN criterion of 10-lb weight loss in 6 months before surgery. Sensitivity is especially appealing given the ease and potential of embedding this tool in an electronic health record and the clinical importance of the consequent intervention. Preoperative as opposed to perioperative nutrition optimization at time of CC diagnosis is essential, as it may help improve postsurgical outcomes as well as oncologic outcomes, including completion of adjuvant chemotherapy. Finally, although our study found that rates of inpatient postoperative nutrition consultation were high, rates of outpatient nutrition consultation in the preoperative period were low. This represents a missed opportunity for intervention before surgery. Similarly, rates of postoperative nutrition follow-up period were low, which points to an area for improvement in longitudinal and holistic care.

We suggest modifications to nutrition intervention protocols, such as ERAS, which should start at the time of GI malignancy diagnosis.¹⁸ Other suggestions include standard involvement of nutritionists in inpatient and outpatient settings with longitudinal follow-up in the preoperative and postoperative periods and patient enrollment in a nutrition program with monitoring at time of diagnosis at the VHA. Our findings as well as previous literature suggest that the preoperative period is the most important time to intervene with regard to nutrition optimization and represents an opportunity for intensive prehabilitation. Future areas of research include incorporating other important measures of malnourishment independent of BMI into future study designs, such as sarcopenia and adipose tissue density, to better assess body composition and predict prognostic risk in CC.^18,19

Strengths and Limitations

This study is limited by its single-center, retrospective design and small sample sizes, and we acknowledge the limitations of our data set. However, the strength of this VHA-based study is that the single-payer system allows for complete capture of perioperative data as well as the opportunity for focused preoperative interventions to improve outcomes. To our knowledge, there is no currently existing literature on improving nutrition protocols at the VHA for patients with a GI malignancy. These retrospective data will help inform current gaps in quality improvement and supportive oncology as it relates to optimizing malnourishment in veterans undergoing surgical resection for their cancer.

 

Conclusions

In the CC cohort, weight loss of ≥ 3% from 6 months prior to time of surgery was significantly associated with delayed recovery, complications, and hospital readmissions. Our findings suggest that patients with CC undergoing surgery may benefit from an intensive, early nutrition prehabilitation. Preoperative nutrition optimization may help improve postsurgical outcomes as well as oncologic outcomes, including completion of adjuvant chemotherapy. Further research would be able to clarify these hypotheses.

To the Editor:

Melanoma is an aggressive form of skin cancer with a high rate of metastasis and poor prognosis.¹ Historically, Hispanic and/or Latino patients have presented with more advanced-stage melanomas and have lower survival rates compared with non-Hispanic and/or non-Latino White patients.² In this study, we evaluated recent data from the last decade to investigate if disparities in melanoma tumor stage at diagnosis and risk for metastases continue to exist in the Hispanic and/or Latino population.

We conducted a retrospective review of melanoma patients at 2 major medical centers in Los Angeles, California—Keck Medicine of USC and Los Angeles County-USC Medical Center—from January 2010 to January 2020. The data collected from electronic medical records included age at melanoma diagnosis, sex, race and ethnicity, insurance type, Breslow depth of lesion, presence of ulceration, and presence of lymph node or distant metastases. Melanoma tumor stage was determined using the American Joint Committee on Cancer classification. Patients who self-reported their ethnicity as not Hispanic and/or Latino were designated to this group regardless of their reported race. Those patients who reported their ethnicity as not Hispanic and/or Latino and reported their race as White were designated as non-Hispanic and/or non-Latino White. This study was approved by the institutional review board of the University of Southern California (Los Angeles). Data analysis was performed using the Pearson χ² test, Fisher exact test, and Wilcoxon rank sum test. Statistical significance was determined at P<.05.

The final cohort of patients included 79 Hispanic and/or Latino patients and 402 non-Hispanic and/or non-Latino White patients. The median age for the Hispanic and/or Latino group was 54 years and 64 years for the non-Hispanic and/or non-Latino White group (P<.001). There was a greater percentage of females in the Hispanic and/or Latino group compared with the non-Hispanic and/or non-Latino White group (53.2% vs 34.6%)(P=.002). Hispanic and/or Latino patients presented with more advanced tumor stage melanomas (T3: 15.2%; T4: 21.5%) compared with non-Hispanic and/or non-Latino White patients (T3: 8.0%; T4: 10.7%)(P=.004). Furthermore, Hispanic and/or Latino patients had higher rates of lymph node metastases compared with non-Hispanic and/or non-Latino White patients (20.3% vs 7.7% [P<.001]) and higher rates of distant metastases (12.7% vs 5.2% [P=.014])(Table 1). The majority of Hispanic and/or Latino patients had Medicaid (39.2%), while most non-Hispanic and/or non-Latino White patients had a preferred provider organization insurance plan (37.3%) or Medicare (34.3%)(P<.001)(Table 2).

This retrospective study analyzing nearly 10 years of recent melanoma data found that disparities in melanoma diagnosis and treatment continue to exist among Hispanic and/or Latino patients. Compared to non-Hispanic and/or non-Latino White patients, Hispanic and/or Latino patients were diagnosed with melanoma at a younger age and the proportion of females with melanoma was higher. Cormier et al² also reported that Hispanic patients were younger at melanoma diagnosis, and females represented a larger majority of patients in the Hispanic population compared with the White population. Hispanic and/or Latino patients in our study had more advanced melanoma tumor stage at diagnosis and a higher risk of lymph node and distant metastases, similar to findings reported by Koblinksi et al.³

Our retrospective cohort study demonstrated that the demographics of Hispanic and/or Latino patients with melanoma differ from non-Hispanic and/or non-Latino White patients, specifically with a greater proportion of younger and female patients in the Hispanic and/or Latino population. We also found that Hispanic and/or Latino patients continue to experience worse melanoma outcomes compared with non-Hispanic and/or non-Latino White patients. Further studies are needed to investigate the etiologies behind these health care disparities and potential interventions to address them. In addition, there needs to be increased awareness of the risk for melanoma in Hispanic and/or Latino patients among both health care providers and patients.

Limitations of this study included a smaller sample size of patients from one geographic region. The retrospective design of this study also increased the risk for selection bias, as some of the patients may have had incomplete records or were lost to follow-up. Therefore, the study cohort may not be representative of the general population. Additionally, patients’ skin types could not be determined using standardized tools such as the Fitzpatrick scale, thus we could not assess how patient skin type may have affected melanoma outcomes.

To the Editor:

Melanoma is an aggressive form of skin cancer with a high rate of metastasis and poor prognosis.¹ Historically, Hispanic and/or Latino patients have presented with more advanced-stage melanomas and have lower survival rates compared with non-Hispanic and/or non-Latino White patients.² In this study, we evaluated recent data from the last decade to investigate if disparities in melanoma tumor stage at diagnosis and risk for metastases continue to exist in the Hispanic and/or Latino population.

We conducted a retrospective review of melanoma patients at 2 major medical centers in Los Angeles, California—Keck Medicine of USC and Los Angeles County-USC Medical Center—from January 2010 to January 2020. The data collected from electronic medical records included age at melanoma diagnosis, sex, race and ethnicity, insurance type, Breslow depth of lesion, presence of ulceration, and presence of lymph node or distant metastases. Melanoma tumor stage was determined using the American Joint Committee on Cancer classification. Patients who self-reported their ethnicity as not Hispanic and/or Latino were designated to this group regardless of their reported race. Those patients who reported their ethnicity as not Hispanic and/or Latino and reported their race as White were designated as non-Hispanic and/or non-Latino White. This study was approved by the institutional review board of the University of Southern California (Los Angeles). Data analysis was performed using the Pearson χ² test, Fisher exact test, and Wilcoxon rank sum test. Statistical significance was determined at P<.05.

The final cohort of patients included 79 Hispanic and/or Latino patients and 402 non-Hispanic and/or non-Latino White patients. The median age for the Hispanic and/or Latino group was 54 years and 64 years for the non-Hispanic and/or non-Latino White group (P<.001). There was a greater percentage of females in the Hispanic and/or Latino group compared with the non-Hispanic and/or non-Latino White group (53.2% vs 34.6%)(P=.002). Hispanic and/or Latino patients presented with more advanced tumor stage melanomas (T3: 15.2%; T4: 21.5%) compared with non-Hispanic and/or non-Latino White patients (T3: 8.0%; T4: 10.7%)(P=.004). Furthermore, Hispanic and/or Latino patients had higher rates of lymph node metastases compared with non-Hispanic and/or non-Latino White patients (20.3% vs 7.7% [P<.001]) and higher rates of distant metastases (12.7% vs 5.2% [P=.014])(Table 1). The majority of Hispanic and/or Latino patients had Medicaid (39.2%), while most non-Hispanic and/or non-Latino White patients had a preferred provider organization insurance plan (37.3%) or Medicare (34.3%)(P<.001)(Table 2).

This retrospective study analyzing nearly 10 years of recent melanoma data found that disparities in melanoma diagnosis and treatment continue to exist among Hispanic and/or Latino patients. Compared to non-Hispanic and/or non-Latino White patients, Hispanic and/or Latino patients were diagnosed with melanoma at a younger age and the proportion of females with melanoma was higher. Cormier et al² also reported that Hispanic patients were younger at melanoma diagnosis, and females represented a larger majority of patients in the Hispanic population compared with the White population. Hispanic and/or Latino patients in our study had more advanced melanoma tumor stage at diagnosis and a higher risk of lymph node and distant metastases, similar to findings reported by Koblinksi et al.³

Our retrospective cohort study demonstrated that the demographics of Hispanic and/or Latino patients with melanoma differ from non-Hispanic and/or non-Latino White patients, specifically with a greater proportion of younger and female patients in the Hispanic and/or Latino population. We also found that Hispanic and/or Latino patients continue to experience worse melanoma outcomes compared with non-Hispanic and/or non-Latino White patients. Further studies are needed to investigate the etiologies behind these health care disparities and potential interventions to address them. In addition, there needs to be increased awareness of the risk for melanoma in Hispanic and/or Latino patients among both health care providers and patients.

Limitations of this study included a smaller sample size of patients from one geographic region. The retrospective design of this study also increased the risk for selection bias, as some of the patients may have had incomplete records or were lost to follow-up. Therefore, the study cohort may not be representative of the general population. Additionally, patients’ skin types could not be determined using standardized tools such as the Fitzpatrick scale, thus we could not assess how patient skin type may have affected melanoma outcomes.

To the Editor:

Melanoma is an aggressive form of skin cancer with a high rate of metastasis and poor prognosis.¹ Historically, Hispanic and/or Latino patients have presented with more advanced-stage melanomas and have lower survival rates compared with non-Hispanic and/or non-Latino White patients.² In this study, we evaluated recent data from the last decade to investigate if disparities in melanoma tumor stage at diagnosis and risk for metastases continue to exist in the Hispanic and/or Latino population.

We conducted a retrospective review of melanoma patients at 2 major medical centers in Los Angeles, California—Keck Medicine of USC and Los Angeles County-USC Medical Center—from January 2010 to January 2020. The data collected from electronic medical records included age at melanoma diagnosis, sex, race and ethnicity, insurance type, Breslow depth of lesion, presence of ulceration, and presence of lymph node or distant metastases. Melanoma tumor stage was determined using the American Joint Committee on Cancer classification. Patients who self-reported their ethnicity as not Hispanic and/or Latino were designated to this group regardless of their reported race. Those patients who reported their ethnicity as not Hispanic and/or Latino and reported their race as White were designated as non-Hispanic and/or non-Latino White. This study was approved by the institutional review board of the University of Southern California (Los Angeles). Data analysis was performed using the Pearson χ² test, Fisher exact test, and Wilcoxon rank sum test. Statistical significance was determined at P<.05.

The final cohort of patients included 79 Hispanic and/or Latino patients and 402 non-Hispanic and/or non-Latino White patients. The median age for the Hispanic and/or Latino group was 54 years and 64 years for the non-Hispanic and/or non-Latino White group (P<.001). There was a greater percentage of females in the Hispanic and/or Latino group compared with the non-Hispanic and/or non-Latino White group (53.2% vs 34.6%)(P=.002). Hispanic and/or Latino patients presented with more advanced tumor stage melanomas (T3: 15.2%; T4: 21.5%) compared with non-Hispanic and/or non-Latino White patients (T3: 8.0%; T4: 10.7%)(P=.004). Furthermore, Hispanic and/or Latino patients had higher rates of lymph node metastases compared with non-Hispanic and/or non-Latino White patients (20.3% vs 7.7% [P<.001]) and higher rates of distant metastases (12.7% vs 5.2% [P=.014])(Table 1). The majority of Hispanic and/or Latino patients had Medicaid (39.2%), while most non-Hispanic and/or non-Latino White patients had a preferred provider organization insurance plan (37.3%) or Medicare (34.3%)(P<.001)(Table 2).

This retrospective study analyzing nearly 10 years of recent melanoma data found that disparities in melanoma diagnosis and treatment continue to exist among Hispanic and/or Latino patients. Compared to non-Hispanic and/or non-Latino White patients, Hispanic and/or Latino patients were diagnosed with melanoma at a younger age and the proportion of females with melanoma was higher. Cormier et al² also reported that Hispanic patients were younger at melanoma diagnosis, and females represented a larger majority of patients in the Hispanic population compared with the White population. Hispanic and/or Latino patients in our study had more advanced melanoma tumor stage at diagnosis and a higher risk of lymph node and distant metastases, similar to findings reported by Koblinksi et al.³

Our retrospective cohort study demonstrated that the demographics of Hispanic and/or Latino patients with melanoma differ from non-Hispanic and/or non-Latino White patients, specifically with a greater proportion of younger and female patients in the Hispanic and/or Latino population. We also found that Hispanic and/or Latino patients continue to experience worse melanoma outcomes compared with non-Hispanic and/or non-Latino White patients. Further studies are needed to investigate the etiologies behind these health care disparities and potential interventions to address them. In addition, there needs to be increased awareness of the risk for melanoma in Hispanic and/or Latino patients among both health care providers and patients.

Limitations of this study included a smaller sample size of patients from one geographic region. The retrospective design of this study also increased the risk for selection bias, as some of the patients may have had incomplete records or were lost to follow-up. Therefore, the study cohort may not be representative of the general population. Additionally, patients’ skin types could not be determined using standardized tools such as the Fitzpatrick scale, thus we could not assess how patient skin type may have affected melanoma outcomes.

The incidence of skin cancer continues to increase, and it is by far the most common malignancy in the United States. Based on the sheer incidence and prevalence of skin cancer, early detection and treatment are critical. Looking at melanoma alone, the 5-year survival rate is greater than 99% when detected early but falls to 71% when the disease reaches the lymph nodes and 32% with metastasis to distant organs.¹ Furthermore, a 2018 study found stage I melanoma patients who were treated 4 months after biopsy had a 41% increased risk of death compared with those treated within the first month.² However, many patients are not seen by a dermatologist first for examination of suspicious skin lesions and instead are referred by a general practitioner or primary care mid-level provider. Therefore, many patients experience a longer time to diagnosis or treatment, which directly correlates with survival rate.

Dermoscopy is a noninvasive diagnostic tool for skin lesions, including melanoma. Using a handheld dermoscope (or dermatoscope), a transilluminating light source magnifies skin lesions and allows for the visualization of subsurface skin structures within the epidermis, dermoepidermal junction, and papillary dermis.³ Dermoscopy has been shown to improve a dermatologist’s accuracy in diagnosing malignant melanoma vs clinical evaluation with the unaided eye.^4,5 More recently, dermoscopy has been digitized, allowing for the collection and documentation of case photographs. Dermoscopy also has expanded past the scope of dermatologists and has become increasingly useful in primary care.⁶ Among family physicians, dermoscopy also has been shown to have a higher sensitivity for melanoma detection compared to gross examination.⁷ Therefore, both the increased diagnostic performance of malignant melanoma using a dermoscope and the expanded use of dermoscopy in medical care validate the evaluation of an artificial intelligence (AI) algorithm in diagnosing malignant melanoma using dermoscopic images.

Triage (Triage Technologies Inc) is an AI application that uses a web interface and combines a pretrained convolutional neural network (CNN) with a reinforcement learning agent as a question-answering model. The CNN algorithm can classify 133 different skin diseases, 7 of which it is able to classify using dermoscopic images. This study sought to evaluate the performance of Triage’s dermoscopic classifier in identifying lesions as benign or malignant to determine whether AI could assist in the triage of skin cancer cases to shorten time to diagnosis.

Materials and Methods

The MClass-D test set from the International Skin Imaging Collaboration was assessed by both AI and practicing medical providers. The set was composed of 80 benign nevi and 20 biopsy-verified malignant melanomas. Board-certified US dermatologists (n=23), family physicians (n=7), and primary care mid-level providers (n=12)(ie, nurse practitioners, physician assistants) were asked to label the images as benign or malignant. The results from the medical providers were then compared to the performance of the AI application by looking at the sensitivity, specificity, accuracy, positive predictive value (PPV), and negative predictive value (NPV). Statistical significance was determined with a 1 sample t test run through RStudio (Posit Software, PBC), and P<.05 was considered significant.

Results

The AI application performed extremely well in differentiating between benign nevi and malignant melanomas, with a sensitivity of 80%, specificity of 95%, accuracy of 92%, PPV of 80%, and NPV of 95% (Table 1). When compared with practicing medical providers, the AI performed significantly better in almost all categories (P<.05)(Figure 1). With all medical providers combined, the AI had significantly higher accuracy, sensitivity, and specificity (P<.05). The accuracy of the individual medical providers ranged from 32% to 78%.

Compared with dermatologists, the AI was significantly more specific and accurate and demonstrated a higher PPV and NPV (P<.05). There was no significant difference between the AI and dermatologists in sensitivity or labeling the true malignant lesions as malignant. The dermatologists who participated had been practicing from 1.5 years to 44 years, with an average of 16 years of dermatologic experience. There was no correlation between years practicing and performance in determining the malignancy of lesions. Of 14 dermatologists, dermoscopy was used daily by 10 and occasionally by 3, but only 6 dermatologists had any formal training. Dermatologists who used dermoscopy averaged 11 years of use.

The AI also performed significantly better than the primary care providers, including both family physicians and mid-level providers (P<.05). With the family physicians and mid-level provider scores combined, the AI showed a statistically significantly better performance in all categories examined, including sensitivity, specificity, accuracy, PPV, and NPV (P<.05). However, when compared with family physicians alone, the AI did not demonstrate a statistically significant difference in sensitivity.

 

 

Comment

Automatic Visual Recognition Development—The AI application we studied was developed by dermatologists as a tool to assist in the screening of skin lesions suspicious for melanoma or a benign neoplasm.⁸ Developing AI applications that can reliably recognize objects in photographs has been the subject of considerable research. Notable progress in automatic visual recognition was shown in 2012 when a deep learning model won the ImageNet object recognition challenge and outperformed competing approaches by a large margin.^9,10 The ImageNet competition, which has been held annually since 2010, required participants to build a visual classification system that distinguished among 1000 object categories using 1.2 million labeled images as training data. In 2017, participants developed automated visual systems that surpassed the estimated human performance.¹¹ Given this success, the organization decided to deliver a more challenging competition involving 3D imaging—Medical ImageNet, a petabyte-scale, cloud-based, open repository project—with goals including image classification and annotation.¹²

Convolutional Neural Networks—Convolutional neural networks are computer system architectures commonly employed for making predictions from images.¹³ Convolutional neural networks are based on a set of layers of learned filters that perform convolution, a mathematical operation that reflects the relationship between the 2 functions. The main algorithm that makes the learning possible is called backpropagation, wherein an error is computed at the output and distributed backward through the neural network’s layers.¹⁴ Although CNNs and backpropagation methods have existed since 1989, recent technologic advances have allowed for deep learning–based algorithms to be widely integrated with everyday applications.¹⁵ Advances in computational power in the form of graphics processing units and parallelization, the existence of large data sets such as the ImageNet database, and the rise of software frameworks have allowed for quick prototyping and deployment of deep learning models.^16,17

Convolutional neural networks have demonstrated potential to excel at a wide range of visual tasks. In dermatology, visual recognition methods often rely on using either a pretrained CNN as a feature extractor for further classification or fine-tuning a pretrained network on dermoscopic images.^18-20 In 2017, a model was trained on 130,000 clinical images of benign and malignant skin lesions. Its performance was found to be in line with that of 21 US board-certified dermatology experts when diagnosing skin cancers from clinical images confirmed by biopsy.²¹

Triage—The AI application Triage is composed of several components contained in a web interface (Figure 2). To use the interface, the user must sign up and upload a photograph to the website. The image first passes through a gated-logic visual classifier that rejects any images that do not contain a visible skin condition. If the image contains a skin condition, the image is passed to a skin classifier that predicts the probability of the image containing 1 of 133 classes of skin conditions, 7 of which the application can diagnose with a dermoscopic image.

Image courtesy of Triage Technologies Inc and Izhaar Tejani, BA (Toronto, Ontario, Canada).

The AI application uses several techniques when training a CNN model. To address skin condition class imbalances (when more examples exist for 1 class than the others) in the training data, additional weights are applied to mistakes made on underrepresented classes, which encourages the model to better detect cases with low prevalence in the data set. Data augmentation techniques such as rotating, zooming, and flipping the training images are applied to allow the model to become more familiar with variability in the input images. Convolutional neural networks are trained using a well-known neural network optimization method called Stochastic gradient descent with momentum.²²

The final predictions are refined by a question-and-answer system that encodes dermatology knowledge and is currently under active development. Finally, the top k most probable conditions are displayed to the user, where k≤5. An initial prototype of the system was described in a published research paper in the 2019 medical imaging workshop of the Neural Information Systems conference.²³

The prototype demonstrated that combining a pretrained CNN with a reinforcement learning agent as a question-answering model increased the classification confidence and accuracy of its visual symptom checker and decreased the average number of questions asked to narrow down the differential diagnosis. The reinforcement learning approach increases the accuracy more than 20% compared with the CNN-only approach, which only uses visual information to predict the condition.²³

 

This application’s current visual question-answering system is trained on a diverse set of data that includes more than 20 years of clinical encounters and user-uploaded cases submitted by more than 150,000 patients and 10,000 clinicians in more than 150 countries. All crowdsourced images used for training the dermoscopy classifier are biopsy-verified images contributed by dermatologists. These data are made up of case photographs that are tagged with metadata around the patient’s age, sex, symptoms, and diagnoses. The CNN algorithm used covers 133 skin disease classes, representing 588 clinical conditions. It also can automatically detect 7 malignant, premalignant, and benign dermoscopic categories, which is the focus of this study (Table 2). Diagnoses are verified by patient response to treatment, biopsy results, and dermatologist consensus.

In addition to having improved performance, supporting more than 130 disease classes, and having a diverse data set, the application used has beat competing technologies.^20,24 The application currently is available on the internet in more than 30 countries after it received Health Canada Class I medical device approval and the CE mark in Europe.

Can AI Reliably Detect Melanoma?—In our study, of the lesions labeled benign, the higher PPV and NPV of the AI algorithm means that the lesions were more reliably true benign lesions, and the lesions labeled as malignant were more likely to be true malignant lesions. Therefore, the diagnosis given by the AI compared with the medical provider was significantly more likely to be correct. These findings demonstrate that this AI application can reliably detect malignant melanoma using dermoscopic images. However, this study was limited by the small sample size of medical providers. Further studies are necessary to assess whether the high diagnostic accuracy of the application translates to expedited referrals and a decrease in unnecessary biopsies.

Dermoscopy Training—This study looked at dermoscopic images instead of gross examination, as is often done in clinic, which draws into question the dermoscopic training dermatologists receive. The diagnostic accuracy using dermoscopic images has been shown to be higher than evaluation with the naked eye.^5,6 However, there currently is no standard for dermoscopic training in dermatology residencies, and education varies widely.²⁵ These data suggest that there may be a lack of dermoscopic training among dermatologists, which could accentuate the difference in performance between dermatologists and AI. Most primary care providers also lack formal dermoscopy training. Although dermoscopy has been shown to increase the diagnostic efficacy of primary care providers, this increase does not become apparent until the medical provider has had years of formal training in addition to clinical experience, which is not commonly provided in the medical training that primary care providers receive.^8,26

Conclusion

It is anticipated that AI will shape the future of medicine and become incorporated into daily practice.²⁷ Artificial intelligence will not replace physicians but rather assist clinicians and help to streamline medical care. Clinicians will take on the role of interpreting AI output and integrate it into patient care. With this advancement, it is important to highlight that for AI to improve the quality, efficiency, and accessibility of health care, clinicians must be equipped with the right training.^27-29

The incidence of skin cancer continues to increase, and it is by far the most common malignancy in the United States. Based on the sheer incidence and prevalence of skin cancer, early detection and treatment are critical. Looking at melanoma alone, the 5-year survival rate is greater than 99% when detected early but falls to 71% when the disease reaches the lymph nodes and 32% with metastasis to distant organs.¹ Furthermore, a 2018 study found stage I melanoma patients who were treated 4 months after biopsy had a 41% increased risk of death compared with those treated within the first month.² However, many patients are not seen by a dermatologist first for examination of suspicious skin lesions and instead are referred by a general practitioner or primary care mid-level provider. Therefore, many patients experience a longer time to diagnosis or treatment, which directly correlates with survival rate.

Dermoscopy is a noninvasive diagnostic tool for skin lesions, including melanoma. Using a handheld dermoscope (or dermatoscope), a transilluminating light source magnifies skin lesions and allows for the visualization of subsurface skin structures within the epidermis, dermoepidermal junction, and papillary dermis.³ Dermoscopy has been shown to improve a dermatologist’s accuracy in diagnosing malignant melanoma vs clinical evaluation with the unaided eye.^4,5 More recently, dermoscopy has been digitized, allowing for the collection and documentation of case photographs. Dermoscopy also has expanded past the scope of dermatologists and has become increasingly useful in primary care.⁶ Among family physicians, dermoscopy also has been shown to have a higher sensitivity for melanoma detection compared to gross examination.⁷ Therefore, both the increased diagnostic performance of malignant melanoma using a dermoscope and the expanded use of dermoscopy in medical care validate the evaluation of an artificial intelligence (AI) algorithm in diagnosing malignant melanoma using dermoscopic images.

Triage (Triage Technologies Inc) is an AI application that uses a web interface and combines a pretrained convolutional neural network (CNN) with a reinforcement learning agent as a question-answering model. The CNN algorithm can classify 133 different skin diseases, 7 of which it is able to classify using dermoscopic images. This study sought to evaluate the performance of Triage’s dermoscopic classifier in identifying lesions as benign or malignant to determine whether AI could assist in the triage of skin cancer cases to shorten time to diagnosis.

Materials and Methods

The MClass-D test set from the International Skin Imaging Collaboration was assessed by both AI and practicing medical providers. The set was composed of 80 benign nevi and 20 biopsy-verified malignant melanomas. Board-certified US dermatologists (n=23), family physicians (n=7), and primary care mid-level providers (n=12)(ie, nurse practitioners, physician assistants) were asked to label the images as benign or malignant. The results from the medical providers were then compared to the performance of the AI application by looking at the sensitivity, specificity, accuracy, positive predictive value (PPV), and negative predictive value (NPV). Statistical significance was determined with a 1 sample t test run through RStudio (Posit Software, PBC), and P<.05 was considered significant.

Results

The AI application performed extremely well in differentiating between benign nevi and malignant melanomas, with a sensitivity of 80%, specificity of 95%, accuracy of 92%, PPV of 80%, and NPV of 95% (Table 1). When compared with practicing medical providers, the AI performed significantly better in almost all categories (P<.05)(Figure 1). With all medical providers combined, the AI had significantly higher accuracy, sensitivity, and specificity (P<.05). The accuracy of the individual medical providers ranged from 32% to 78%.

Compared with dermatologists, the AI was significantly more specific and accurate and demonstrated a higher PPV and NPV (P<.05). There was no significant difference between the AI and dermatologists in sensitivity or labeling the true malignant lesions as malignant. The dermatologists who participated had been practicing from 1.5 years to 44 years, with an average of 16 years of dermatologic experience. There was no correlation between years practicing and performance in determining the malignancy of lesions. Of 14 dermatologists, dermoscopy was used daily by 10 and occasionally by 3, but only 6 dermatologists had any formal training. Dermatologists who used dermoscopy averaged 11 years of use.

The AI also performed significantly better than the primary care providers, including both family physicians and mid-level providers (P<.05). With the family physicians and mid-level provider scores combined, the AI showed a statistically significantly better performance in all categories examined, including sensitivity, specificity, accuracy, PPV, and NPV (P<.05). However, when compared with family physicians alone, the AI did not demonstrate a statistically significant difference in sensitivity.

 

 

Comment

Automatic Visual Recognition Development—The AI application we studied was developed by dermatologists as a tool to assist in the screening of skin lesions suspicious for melanoma or a benign neoplasm.⁸ Developing AI applications that can reliably recognize objects in photographs has been the subject of considerable research. Notable progress in automatic visual recognition was shown in 2012 when a deep learning model won the ImageNet object recognition challenge and outperformed competing approaches by a large margin.^9,10 The ImageNet competition, which has been held annually since 2010, required participants to build a visual classification system that distinguished among 1000 object categories using 1.2 million labeled images as training data. In 2017, participants developed automated visual systems that surpassed the estimated human performance.¹¹ Given this success, the organization decided to deliver a more challenging competition involving 3D imaging—Medical ImageNet, a petabyte-scale, cloud-based, open repository project—with goals including image classification and annotation.¹²

Convolutional Neural Networks—Convolutional neural networks are computer system architectures commonly employed for making predictions from images.¹³ Convolutional neural networks are based on a set of layers of learned filters that perform convolution, a mathematical operation that reflects the relationship between the 2 functions. The main algorithm that makes the learning possible is called backpropagation, wherein an error is computed at the output and distributed backward through the neural network’s layers.¹⁴ Although CNNs and backpropagation methods have existed since 1989, recent technologic advances have allowed for deep learning–based algorithms to be widely integrated with everyday applications.¹⁵ Advances in computational power in the form of graphics processing units and parallelization, the existence of large data sets such as the ImageNet database, and the rise of software frameworks have allowed for quick prototyping and deployment of deep learning models.^16,17

Convolutional neural networks have demonstrated potential to excel at a wide range of visual tasks. In dermatology, visual recognition methods often rely on using either a pretrained CNN as a feature extractor for further classification or fine-tuning a pretrained network on dermoscopic images.^18-20 In 2017, a model was trained on 130,000 clinical images of benign and malignant skin lesions. Its performance was found to be in line with that of 21 US board-certified dermatology experts when diagnosing skin cancers from clinical images confirmed by biopsy.²¹

Triage—The AI application Triage is composed of several components contained in a web interface (Figure 2). To use the interface, the user must sign up and upload a photograph to the website. The image first passes through a gated-logic visual classifier that rejects any images that do not contain a visible skin condition. If the image contains a skin condition, the image is passed to a skin classifier that predicts the probability of the image containing 1 of 133 classes of skin conditions, 7 of which the application can diagnose with a dermoscopic image.

Image courtesy of Triage Technologies Inc and Izhaar Tejani, BA (Toronto, Ontario, Canada).

The AI application uses several techniques when training a CNN model. To address skin condition class imbalances (when more examples exist for 1 class than the others) in the training data, additional weights are applied to mistakes made on underrepresented classes, which encourages the model to better detect cases with low prevalence in the data set. Data augmentation techniques such as rotating, zooming, and flipping the training images are applied to allow the model to become more familiar with variability in the input images. Convolutional neural networks are trained using a well-known neural network optimization method called Stochastic gradient descent with momentum.²²

The final predictions are refined by a question-and-answer system that encodes dermatology knowledge and is currently under active development. Finally, the top k most probable conditions are displayed to the user, where k≤5. An initial prototype of the system was described in a published research paper in the 2019 medical imaging workshop of the Neural Information Systems conference.²³

The prototype demonstrated that combining a pretrained CNN with a reinforcement learning agent as a question-answering model increased the classification confidence and accuracy of its visual symptom checker and decreased the average number of questions asked to narrow down the differential diagnosis. The reinforcement learning approach increases the accuracy more than 20% compared with the CNN-only approach, which only uses visual information to predict the condition.²³

 

This application’s current visual question-answering system is trained on a diverse set of data that includes more than 20 years of clinical encounters and user-uploaded cases submitted by more than 150,000 patients and 10,000 clinicians in more than 150 countries. All crowdsourced images used for training the dermoscopy classifier are biopsy-verified images contributed by dermatologists. These data are made up of case photographs that are tagged with metadata around the patient’s age, sex, symptoms, and diagnoses. The CNN algorithm used covers 133 skin disease classes, representing 588 clinical conditions. It also can automatically detect 7 malignant, premalignant, and benign dermoscopic categories, which is the focus of this study (Table 2). Diagnoses are verified by patient response to treatment, biopsy results, and dermatologist consensus.

In addition to having improved performance, supporting more than 130 disease classes, and having a diverse data set, the application used has beat competing technologies.^20,24 The application currently is available on the internet in more than 30 countries after it received Health Canada Class I medical device approval and the CE mark in Europe.

Can AI Reliably Detect Melanoma?—In our study, of the lesions labeled benign, the higher PPV and NPV of the AI algorithm means that the lesions were more reliably true benign lesions, and the lesions labeled as malignant were more likely to be true malignant lesions. Therefore, the diagnosis given by the AI compared with the medical provider was significantly more likely to be correct. These findings demonstrate that this AI application can reliably detect malignant melanoma using dermoscopic images. However, this study was limited by the small sample size of medical providers. Further studies are necessary to assess whether the high diagnostic accuracy of the application translates to expedited referrals and a decrease in unnecessary biopsies.

Dermoscopy Training—This study looked at dermoscopic images instead of gross examination, as is often done in clinic, which draws into question the dermoscopic training dermatologists receive. The diagnostic accuracy using dermoscopic images has been shown to be higher than evaluation with the naked eye.^5,6 However, there currently is no standard for dermoscopic training in dermatology residencies, and education varies widely.²⁵ These data suggest that there may be a lack of dermoscopic training among dermatologists, which could accentuate the difference in performance between dermatologists and AI. Most primary care providers also lack formal dermoscopy training. Although dermoscopy has been shown to increase the diagnostic efficacy of primary care providers, this increase does not become apparent until the medical provider has had years of formal training in addition to clinical experience, which is not commonly provided in the medical training that primary care providers receive.^8,26

Conclusion

It is anticipated that AI will shape the future of medicine and become incorporated into daily practice.²⁷ Artificial intelligence will not replace physicians but rather assist clinicians and help to streamline medical care. Clinicians will take on the role of interpreting AI output and integrate it into patient care. With this advancement, it is important to highlight that for AI to improve the quality, efficiency, and accessibility of health care, clinicians must be equipped with the right training.^27-29

The incidence of skin cancer continues to increase, and it is by far the most common malignancy in the United States. Based on the sheer incidence and prevalence of skin cancer, early detection and treatment are critical. Looking at melanoma alone, the 5-year survival rate is greater than 99% when detected early but falls to 71% when the disease reaches the lymph nodes and 32% with metastasis to distant organs.¹ Furthermore, a 2018 study found stage I melanoma patients who were treated 4 months after biopsy had a 41% increased risk of death compared with those treated within the first month.² However, many patients are not seen by a dermatologist first for examination of suspicious skin lesions and instead are referred by a general practitioner or primary care mid-level provider. Therefore, many patients experience a longer time to diagnosis or treatment, which directly correlates with survival rate.

Dermoscopy is a noninvasive diagnostic tool for skin lesions, including melanoma. Using a handheld dermoscope (or dermatoscope), a transilluminating light source magnifies skin lesions and allows for the visualization of subsurface skin structures within the epidermis, dermoepidermal junction, and papillary dermis.³ Dermoscopy has been shown to improve a dermatologist’s accuracy in diagnosing malignant melanoma vs clinical evaluation with the unaided eye.^4,5 More recently, dermoscopy has been digitized, allowing for the collection and documentation of case photographs. Dermoscopy also has expanded past the scope of dermatologists and has become increasingly useful in primary care.⁶ Among family physicians, dermoscopy also has been shown to have a higher sensitivity for melanoma detection compared to gross examination.⁷ Therefore, both the increased diagnostic performance of malignant melanoma using a dermoscope and the expanded use of dermoscopy in medical care validate the evaluation of an artificial intelligence (AI) algorithm in diagnosing malignant melanoma using dermoscopic images.

Triage (Triage Technologies Inc) is an AI application that uses a web interface and combines a pretrained convolutional neural network (CNN) with a reinforcement learning agent as a question-answering model. The CNN algorithm can classify 133 different skin diseases, 7 of which it is able to classify using dermoscopic images. This study sought to evaluate the performance of Triage’s dermoscopic classifier in identifying lesions as benign or malignant to determine whether AI could assist in the triage of skin cancer cases to shorten time to diagnosis.

Materials and Methods

The MClass-D test set from the International Skin Imaging Collaboration was assessed by both AI and practicing medical providers. The set was composed of 80 benign nevi and 20 biopsy-verified malignant melanomas. Board-certified US dermatologists (n=23), family physicians (n=7), and primary care mid-level providers (n=12)(ie, nurse practitioners, physician assistants) were asked to label the images as benign or malignant. The results from the medical providers were then compared to the performance of the AI application by looking at the sensitivity, specificity, accuracy, positive predictive value (PPV), and negative predictive value (NPV). Statistical significance was determined with a 1 sample t test run through RStudio (Posit Software, PBC), and P<.05 was considered significant.

Results

The AI application performed extremely well in differentiating between benign nevi and malignant melanomas, with a sensitivity of 80%, specificity of 95%, accuracy of 92%, PPV of 80%, and NPV of 95% (Table 1). When compared with practicing medical providers, the AI performed significantly better in almost all categories (P<.05)(Figure 1). With all medical providers combined, the AI had significantly higher accuracy, sensitivity, and specificity (P<.05). The accuracy of the individual medical providers ranged from 32% to 78%.

Compared with dermatologists, the AI was significantly more specific and accurate and demonstrated a higher PPV and NPV (P<.05). There was no significant difference between the AI and dermatologists in sensitivity or labeling the true malignant lesions as malignant. The dermatologists who participated had been practicing from 1.5 years to 44 years, with an average of 16 years of dermatologic experience. There was no correlation between years practicing and performance in determining the malignancy of lesions. Of 14 dermatologists, dermoscopy was used daily by 10 and occasionally by 3, but only 6 dermatologists had any formal training. Dermatologists who used dermoscopy averaged 11 years of use.

The AI also performed significantly better than the primary care providers, including both family physicians and mid-level providers (P<.05). With the family physicians and mid-level provider scores combined, the AI showed a statistically significantly better performance in all categories examined, including sensitivity, specificity, accuracy, PPV, and NPV (P<.05). However, when compared with family physicians alone, the AI did not demonstrate a statistically significant difference in sensitivity.

 

 

Comment

Automatic Visual Recognition Development—The AI application we studied was developed by dermatologists as a tool to assist in the screening of skin lesions suspicious for melanoma or a benign neoplasm.⁸ Developing AI applications that can reliably recognize objects in photographs has been the subject of considerable research. Notable progress in automatic visual recognition was shown in 2012 when a deep learning model won the ImageNet object recognition challenge and outperformed competing approaches by a large margin.^9,10 The ImageNet competition, which has been held annually since 2010, required participants to build a visual classification system that distinguished among 1000 object categories using 1.2 million labeled images as training data. In 2017, participants developed automated visual systems that surpassed the estimated human performance.¹¹ Given this success, the organization decided to deliver a more challenging competition involving 3D imaging—Medical ImageNet, a petabyte-scale, cloud-based, open repository project—with goals including image classification and annotation.¹²

Convolutional Neural Networks—Convolutional neural networks are computer system architectures commonly employed for making predictions from images.¹³ Convolutional neural networks are based on a set of layers of learned filters that perform convolution, a mathematical operation that reflects the relationship between the 2 functions. The main algorithm that makes the learning possible is called backpropagation, wherein an error is computed at the output and distributed backward through the neural network’s layers.¹⁴ Although CNNs and backpropagation methods have existed since 1989, recent technologic advances have allowed for deep learning–based algorithms to be widely integrated with everyday applications.¹⁵ Advances in computational power in the form of graphics processing units and parallelization, the existence of large data sets such as the ImageNet database, and the rise of software frameworks have allowed for quick prototyping and deployment of deep learning models.^16,17

Convolutional neural networks have demonstrated potential to excel at a wide range of visual tasks. In dermatology, visual recognition methods often rely on using either a pretrained CNN as a feature extractor for further classification or fine-tuning a pretrained network on dermoscopic images.^18-20 In 2017, a model was trained on 130,000 clinical images of benign and malignant skin lesions. Its performance was found to be in line with that of 21 US board-certified dermatology experts when diagnosing skin cancers from clinical images confirmed by biopsy.²¹

Triage—The AI application Triage is composed of several components contained in a web interface (Figure 2). To use the interface, the user must sign up and upload a photograph to the website. The image first passes through a gated-logic visual classifier that rejects any images that do not contain a visible skin condition. If the image contains a skin condition, the image is passed to a skin classifier that predicts the probability of the image containing 1 of 133 classes of skin conditions, 7 of which the application can diagnose with a dermoscopic image.

Image courtesy of Triage Technologies Inc and Izhaar Tejani, BA (Toronto, Ontario, Canada).

The AI application uses several techniques when training a CNN model. To address skin condition class imbalances (when more examples exist for 1 class than the others) in the training data, additional weights are applied to mistakes made on underrepresented classes, which encourages the model to better detect cases with low prevalence in the data set. Data augmentation techniques such as rotating, zooming, and flipping the training images are applied to allow the model to become more familiar with variability in the input images. Convolutional neural networks are trained using a well-known neural network optimization method called Stochastic gradient descent with momentum.²²

The final predictions are refined by a question-and-answer system that encodes dermatology knowledge and is currently under active development. Finally, the top k most probable conditions are displayed to the user, where k≤5. An initial prototype of the system was described in a published research paper in the 2019 medical imaging workshop of the Neural Information Systems conference.²³

The prototype demonstrated that combining a pretrained CNN with a reinforcement learning agent as a question-answering model increased the classification confidence and accuracy of its visual symptom checker and decreased the average number of questions asked to narrow down the differential diagnosis. The reinforcement learning approach increases the accuracy more than 20% compared with the CNN-only approach, which only uses visual information to predict the condition.²³

 

This application’s current visual question-answering system is trained on a diverse set of data that includes more than 20 years of clinical encounters and user-uploaded cases submitted by more than 150,000 patients and 10,000 clinicians in more than 150 countries. All crowdsourced images used for training the dermoscopy classifier are biopsy-verified images contributed by dermatologists. These data are made up of case photographs that are tagged with metadata around the patient’s age, sex, symptoms, and diagnoses. The CNN algorithm used covers 133 skin disease classes, representing 588 clinical conditions. It also can automatically detect 7 malignant, premalignant, and benign dermoscopic categories, which is the focus of this study (Table 2). Diagnoses are verified by patient response to treatment, biopsy results, and dermatologist consensus.

In addition to having improved performance, supporting more than 130 disease classes, and having a diverse data set, the application used has beat competing technologies.^20,24 The application currently is available on the internet in more than 30 countries after it received Health Canada Class I medical device approval and the CE mark in Europe.

Can AI Reliably Detect Melanoma?—In our study, of the lesions labeled benign, the higher PPV and NPV of the AI algorithm means that the lesions were more reliably true benign lesions, and the lesions labeled as malignant were more likely to be true malignant lesions. Therefore, the diagnosis given by the AI compared with the medical provider was significantly more likely to be correct. These findings demonstrate that this AI application can reliably detect malignant melanoma using dermoscopic images. However, this study was limited by the small sample size of medical providers. Further studies are necessary to assess whether the high diagnostic accuracy of the application translates to expedited referrals and a decrease in unnecessary biopsies.

Dermoscopy Training—This study looked at dermoscopic images instead of gross examination, as is often done in clinic, which draws into question the dermoscopic training dermatologists receive. The diagnostic accuracy using dermoscopic images has been shown to be higher than evaluation with the naked eye.^5,6 However, there currently is no standard for dermoscopic training in dermatology residencies, and education varies widely.²⁵ These data suggest that there may be a lack of dermoscopic training among dermatologists, which could accentuate the difference in performance between dermatologists and AI. Most primary care providers also lack formal dermoscopy training. Although dermoscopy has been shown to increase the diagnostic efficacy of primary care providers, this increase does not become apparent until the medical provider has had years of formal training in addition to clinical experience, which is not commonly provided in the medical training that primary care providers receive.^8,26

Conclusion

It is anticipated that AI will shape the future of medicine and become incorporated into daily practice.²⁷ Artificial intelligence will not replace physicians but rather assist clinicians and help to streamline medical care. Clinicians will take on the role of interpreting AI output and integrate it into patient care. With this advancement, it is important to highlight that for AI to improve the quality, efficiency, and accessibility of health care, clinicians must be equipped with the right training.^27-29

Amyotrophic lateral sclerosis (ALS) is an incurable neurodegenerative disorder that results in progressive deterioration of motor neurons in the ventral horn of the spinal cord, which results in loss of voluntary muscle movements.¹ Eventually, typical daily tasks become difficult to perform, and as the disease progresses, the ability to eat and breathe is impaired.² Reports from 2015 show the annual incidence of ALS is 5 cases per 100,000 people, with the total number of cases reported at more than 16,000 in the United States.³ In clinical practice, disease progression is routinely assessed by the Revised Amyotrophic Lateral Sclerosis Functional Rating Scale (ALSFRS-R). Typical decline is 1 point per month.⁴

Unfortunately, at this time, ALS care focuses on symptom management, including prevention of weight loss; implementation of communication strategies; and management of pain, constipation, excess secretions, cramping, and breathing. Despite copious research into treatment options, few exist. Riluzole is an oral medication administered twice daily and has been on the market since 1995.^5-7 Efficacy was demonstrated in a study showing statistically significant survival at 12 months compared with controls (74% vs 58%, respectively; P = .014).⁶ Since its approval, riluzole has become part of standard-of-care ALS management.

In 2017, the US Food and Drug Administration (FDA) approved edaravone, an IV medication that was found to slow the progression of ALS in some patients.^8-12 Oxidative stress caused by free radicals is hypothesized to increase the progression of ALS by motor neuron degradation.¹³ Edaravone works as a free radical and peroxynitrite scavenger and has been shown to eliminate lipid peroxides and hydroxyl radicals known to damage endothelial and neuronal cells.¹²

Given the mechanism of action of edaravone, it seemed to be a promising option to slow the progression of ALS. A 2019 systematic review analyzed 3 randomized studies with 367 patients and found a statistically significant difference in change in ALSFRS-R scores between patients treated with edaravone for 24 weeks compared with patients treated with the placebo (mean difference, 1.63; 95% CI, 0.26-3.00; P = .02).¹² Secondary endpoints evaluated included percent forced vital capacity (%FVC), grip strength, and pinch strength: All showing no significant difference when comparing IV edaravone with placebo.

A 2022 postmarketing study of 324 patients with ALS evaluated the safety and efficacy of long-term edaravone treatment. IV edaravone therapy for > 24 weeks was well tolerated, although it was not associated with any disease-modifying benefit when comparing ALSFRS-R scores with patients not receiving edaravone over a median 13.9 months (ALSFRS-R points/month, -0.91 vs -0.85; P = .37).¹³ A third ALS treatment medication, sodium phenylbutyrate/taurursodiol was approved in 2022 but not available during our study period and not included here.^14,15

Studies have shown an increased incidence of ALS in the veteran population. Veterans serving in the Gulf War were nearly twice as likely to develop ALS as those not serving in the Gulf.¹⁶ However, existing literature regarding the effectiveness of edaravone does not specifically examine the effect on this unique population. The objective of this study was to assess the effect of IV edaravone on ALS progression in veterans compared with veterans who received standard of care.

 

Methods

This study was conducted at a large, academic US Department of Veterans Affairs (VA) medical center. Patients with ALS are followed by a multidisciplinary clinic composed of a neurologist, pulmonologist, clinical pharmacist, social worker, speech therapist, physical therapist, occupational therapist, dietician, clinical psychologist, wheelchair clinic representative, and benefits representative. Patients are typically seen for a half-day appointment about every 3 months. During these visits, a comprehensive review of disease progression is performed. This review entails completion of the ALSFRS-R, physical examination, and pulmonary function testing. Speech intelligibility stage (SIS) is assessed by a speech therapist as well. SIS is scored from 1 (no detectable speech disorder) to 5 (no functional speech). All patients followed in this multidisciplinary ALS clinic receive standard-of-care treatment. This includes the discussion of treatment options that if appropriate are provided to help manage a wide range of complications associated with this disease (eg, pain, cramping, constipation, excessive secretions, weight loss, dysphagia). As a part of these personal discussions, treatment with riluzole is also offered as a standard-of-care pharmacologic option.

Study Design

This retrospective case-control study was conducted using electronic health record data to compare ALS progression in patients on IV edaravone therapy with standard of care. The Indiana University/Purdue University, Indianapolis Institutional Review Board and the VA Research and Development Committee approved the study. The control cohort received the standard of care. Patients in the case cohort received standard of care and edaravone 60 mg infusions daily for an initial cycle of 14 days on treatment, followed by 14 days off. All subsequent cycles were 10 of 14 days on treatment followed by 14 days off. The initial 2 doses were administered in the outpatient infusion clinic to monitor for a hypersensitivity reaction. Patients then had a peripherally inserted central catheter line placed and received doses on days 3 through 14 at home. A port was placed for subsequent cycles, which were also completed at home. Appropriateness of edaravone therapy was assessed by the neurologist at each follow-up appointment. Therapy was then discontinued if warranted based on disease progression or patient preference.

Study Population

Patients included were aged 18 to 75 years with diagnosed ALS. Patients with complications that might influence evaluation of medication efficacy (eg, Parkinson disease, schizophrenia, significant dementia, other major medical morbidity) were excluded. Patients were also excluded if they were on continuous bilevel positive airway pressure and/or had a total score of ≤ 3 points on ALSFRS-R items for dyspnea, orthopnea, or respiratory insufficiency. Due to our small sample size, patients were excluded if treatment was < 6 months, which is the gold standard of therapy duration established by clinical trials.^9,11,12

The standard-of-care cohort included patients enrolled in the multidisciplinary clinic September 1, 2014 to August 31, 2017. These patients were compared in a 2:1 ratio with patients who received IV edaravone. The edaravone cohort included patients who initiated treatment with IV edaravone between September 1, 2017, and August 31, 2020. This date range prior to the approval of edaravone was chosen to compare patients at similar stages of disease progression and to have the largest sample size possible.

Data Collection

Data were obtained for eligible patients using the VA Computerized Patient Record System. Demographic data gathered for each patient included age, sex, weight, height, body mass index (BMI), race, and riluzole use.

The primary endpoint was the change in ALSFRS-R score after 6 months of IV edaravone compared with standard-of-care ALS management. Secondary outcomes included change in ALSFRS-R scores 3, 12, 18, and 24 months after therapy initiation, change in %FVC and SIS 3, 6, 12, 18, and 24 months after therapy initiation, duration of edaravone completed (months), time to death (months), and adverse events.

 

Statistical Analysis

Comparisons between the edaravone and control groups for differences in patient characteristics were made using χ² and 2-sample t tests for categorical and continuous variables, respectively. Comparisons between the 2 groups for differences in study outcomes (ALSFRS-R scores, %FVC, SIS) at each time point were evaluated using 2-sample t tests. Adverse events and adverse drug reactions were compared between groups using χ² tests. Statistical significance was set at 0.05.

We estimated that a sample size of 21 subjects in the edaravone (case) group and 42 in the standard-of-care (control) group would be needed to achieve 80% power to detect a difference of 6.5 between the 2 groups for the change in ALSFRS-R scores. This 80% power was calculated based on a 2-sample t test, and assuming a 2-sided 5% significance level and a within-group SD of 8.5.⁹ Statistical analysis was conducted using Microsoft Excel.

Results

A total of 96 unique patients were seen in our multidisciplinary ALS clinic between September 1, 2014, and August 31, 2017 (Figure).

Of the 96 patients, 10 met exclusion criteria. From the remaining 86, 42 were randomly selected for the standard-of-care group. A total of 27 patients seen in multidisciplinary ALS clinic between September 1, 2017, and August 31, 2020, received at least 1 dose of IV edaravone. Of the 27 edaravone patients, 6 were excluded for not completing a total of 6 months of edaravone. Two of the 6 excluded developed a rash, which resolved within 1 week after discontinuing edaravone. The other 4 discontinued edaravone before 6 months because of disease progression.

Baseline Characteristics

Baseline demographics were similar between the groups (Table 1). Most patients were White men with a mean age of 60 years. Baseline %FVC was about 68%. Fewer patients in the standard-of-care group were taking riluzole than in the edaravone group (67% vs 95%, respectively; P = .002). Mean (SD) baseline SIS scores were slightly higher in the standard-of-care group vs the edaravone group (2.0 [1.0] vs 1.4 [0.6], respectively; P = .01).

Efficacy

No difference was found in the ALSFRS-R scores at 6 months between the IV edaravone and standard-of-care groups (P = .84) (Table 2). Our study did not meet power to calculate statistical analysis at 12, 18, and 24 months due to its size. No difference was found in change from baseline %FVC at 6 months between the 2 groups (P = .30) (Table 3). Change between the 2 groups in baseline SIS at 6 months also was not different (P = .69) (Table 4). Sample size was insufficient to calculate %FVC and SIS at the 12, 18, and 24 month intervals.

No difference was noted in time to death between groups (P = .93) (Table 5), and no adverse events were reported in either group.

 

Discussion

This 24-month, case-control retrospective study assessed efficacy and safety of IV edaravone for the management of ALS. Although the landmark edaravone study showed slowed progression of ALS at 6 and 12 months, the effectiveness of edaravone outside the clinical trial setting has been less compelling.^9-11,13 A later study showed no difference in change in ALSFRS-R score at 6 months compared with that of the placebo group.⁷ In our study, no statistically significant difference was found for change in ALSFRS-R scores at 6 months.

Our study was unique given we evaluated a veteran population. The link between the military and ALS is largely unknown, although studies have shown increased incidence of ALS in people with a military history compared with that of the general population.^16-18 Our study was also unique because it was single-centered in design and allowed for outcome assessments, including ALSFRS-R scores, SIS, and %FVC measurements, to all be conducted by the same practitioner to limit variability. Unfortunately, our sample size resulted in a cohort that was underpowered at 12, 18, and 24 months. In addition, there was a lack of data on chart review for SIS and %FVC measurements at 24 months. As ALS progresses toward end stage, SIS and %FVC measurements can become difficult and burdensome on the patient to obtain, and the ALS multidisciplinary team may decide not to gather these data points as ALS progresses. As a result, change in SIS and %FVC measurements were unable to be reported due to lack of gathering this information at the 24-month mark in the edaravone group. Due to the cost and administration burden associated with edaravone, it is important that assessment of disease progression is performed regularly to assess benefit and appropriateness of continued therapy. The oral formulation of edaravone was approved in 2022, shortly after the completion of data collection for this study.^19,20 Although our study did not analyze oral edaravone, the administration burden of treatment would be reduced with the oral formulation, and we hypothesize there will be increased patient interest in ALS management with oral vs IV edaravone. Evaluation of long-term treatment for efficacy and safety beyond 24 months has not been evaluated. Future studies should continue to evaluate edaravone use in a larger veteran population.

Limitations

One limitation for our study alluded to earlier in the discussion was sample size. Although this study met power at the 6-month mark, it was limited by the number of patients who received more than 6 months of edaravone (n = 21). As a result, statistical analyses between treatment groups were underpowered at 12, 18, and 24 months. Our study had 80% power to detect a difference of 6.5 between the groups for the change in ALSFRS-R scores. Previous studies detected a statistically significant difference in ALSFRS-R scores, with a difference in ALSFRS-R scores of 2.49 between groups.⁸ Future studies should evaluate a larger sample size of patients who are prescribed edaravone.

Another limitation was that the edaravone and standard-of-care group data were gathered from different time periods. Two different time frames were selected to increase sample size by gathering data over a longer period and to account for patients who may have qualified for IV edaravone but could not receive it as it was not yet available on the market. There were no known changes to the standard of care between the time periods that would affect results. As noted previously, the standard-of-care group had fewer patients taking riluzole compared with the edaravone group, which may have confounded our results. We concluded patients opting for edaravone were more likely to trial riluzole, taken by mouth twice daily, before starting edaravone, a once-daily IV infusion.

Conclusions

No difference in the rate of ALS progression was noted between patients who received IV edaravone vs standard of care at 6 months. In addition, no difference was noted in other objective measures of disease progression, including %FVC, SIS, and time to death. As a result, the decision to initiate and continue edaravone therapy should be made on an individualized basis according to a prescriber’s clinical judgment and a patient’s goals. Edaravone therapy should be discontinued when disease progression occurs or when medication administration becomes a burden.

Acknowledgments

This material is the result of work supported with resources and the use of facilities at Veteran Health Indiana.

Amyotrophic lateral sclerosis (ALS) is an incurable neurodegenerative disorder that results in progressive deterioration of motor neurons in the ventral horn of the spinal cord, which results in loss of voluntary muscle movements.¹ Eventually, typical daily tasks become difficult to perform, and as the disease progresses, the ability to eat and breathe is impaired.² Reports from 2015 show the annual incidence of ALS is 5 cases per 100,000 people, with the total number of cases reported at more than 16,000 in the United States.³ In clinical practice, disease progression is routinely assessed by the Revised Amyotrophic Lateral Sclerosis Functional Rating Scale (ALSFRS-R). Typical decline is 1 point per month.⁴

Unfortunately, at this time, ALS care focuses on symptom management, including prevention of weight loss; implementation of communication strategies; and management of pain, constipation, excess secretions, cramping, and breathing. Despite copious research into treatment options, few exist. Riluzole is an oral medication administered twice daily and has been on the market since 1995.^5-7 Efficacy was demonstrated in a study showing statistically significant survival at 12 months compared with controls (74% vs 58%, respectively; P = .014).⁶ Since its approval, riluzole has become part of standard-of-care ALS management.

In 2017, the US Food and Drug Administration (FDA) approved edaravone, an IV medication that was found to slow the progression of ALS in some patients.^8-12 Oxidative stress caused by free radicals is hypothesized to increase the progression of ALS by motor neuron degradation.¹³ Edaravone works as a free radical and peroxynitrite scavenger and has been shown to eliminate lipid peroxides and hydroxyl radicals known to damage endothelial and neuronal cells.¹²

Given the mechanism of action of edaravone, it seemed to be a promising option to slow the progression of ALS. A 2019 systematic review analyzed 3 randomized studies with 367 patients and found a statistically significant difference in change in ALSFRS-R scores between patients treated with edaravone for 24 weeks compared with patients treated with the placebo (mean difference, 1.63; 95% CI, 0.26-3.00; P = .02).¹² Secondary endpoints evaluated included percent forced vital capacity (%FVC), grip strength, and pinch strength: All showing no significant difference when comparing IV edaravone with placebo.

A 2022 postmarketing study of 324 patients with ALS evaluated the safety and efficacy of long-term edaravone treatment. IV edaravone therapy for > 24 weeks was well tolerated, although it was not associated with any disease-modifying benefit when comparing ALSFRS-R scores with patients not receiving edaravone over a median 13.9 months (ALSFRS-R points/month, -0.91 vs -0.85; P = .37).¹³ A third ALS treatment medication, sodium phenylbutyrate/taurursodiol was approved in 2022 but not available during our study period and not included here.^14,15

Studies have shown an increased incidence of ALS in the veteran population. Veterans serving in the Gulf War were nearly twice as likely to develop ALS as those not serving in the Gulf.¹⁶ However, existing literature regarding the effectiveness of edaravone does not specifically examine the effect on this unique population. The objective of this study was to assess the effect of IV edaravone on ALS progression in veterans compared with veterans who received standard of care.

 

Methods

This study was conducted at a large, academic US Department of Veterans Affairs (VA) medical center. Patients with ALS are followed by a multidisciplinary clinic composed of a neurologist, pulmonologist, clinical pharmacist, social worker, speech therapist, physical therapist, occupational therapist, dietician, clinical psychologist, wheelchair clinic representative, and benefits representative. Patients are typically seen for a half-day appointment about every 3 months. During these visits, a comprehensive review of disease progression is performed. This review entails completion of the ALSFRS-R, physical examination, and pulmonary function testing. Speech intelligibility stage (SIS) is assessed by a speech therapist as well. SIS is scored from 1 (no detectable speech disorder) to 5 (no functional speech). All patients followed in this multidisciplinary ALS clinic receive standard-of-care treatment. This includes the discussion of treatment options that if appropriate are provided to help manage a wide range of complications associated with this disease (eg, pain, cramping, constipation, excessive secretions, weight loss, dysphagia). As a part of these personal discussions, treatment with riluzole is also offered as a standard-of-care pharmacologic option.

Study Design

This retrospective case-control study was conducted using electronic health record data to compare ALS progression in patients on IV edaravone therapy with standard of care. The Indiana University/Purdue University, Indianapolis Institutional Review Board and the VA Research and Development Committee approved the study. The control cohort received the standard of care. Patients in the case cohort received standard of care and edaravone 60 mg infusions daily for an initial cycle of 14 days on treatment, followed by 14 days off. All subsequent cycles were 10 of 14 days on treatment followed by 14 days off. The initial 2 doses were administered in the outpatient infusion clinic to monitor for a hypersensitivity reaction. Patients then had a peripherally inserted central catheter line placed and received doses on days 3 through 14 at home. A port was placed for subsequent cycles, which were also completed at home. Appropriateness of edaravone therapy was assessed by the neurologist at each follow-up appointment. Therapy was then discontinued if warranted based on disease progression or patient preference.

Study Population

Patients included were aged 18 to 75 years with diagnosed ALS. Patients with complications that might influence evaluation of medication efficacy (eg, Parkinson disease, schizophrenia, significant dementia, other major medical morbidity) were excluded. Patients were also excluded if they were on continuous bilevel positive airway pressure and/or had a total score of ≤ 3 points on ALSFRS-R items for dyspnea, orthopnea, or respiratory insufficiency. Due to our small sample size, patients were excluded if treatment was < 6 months, which is the gold standard of therapy duration established by clinical trials.^9,11,12

The standard-of-care cohort included patients enrolled in the multidisciplinary clinic September 1, 2014 to August 31, 2017. These patients were compared in a 2:1 ratio with patients who received IV edaravone. The edaravone cohort included patients who initiated treatment with IV edaravone between September 1, 2017, and August 31, 2020. This date range prior to the approval of edaravone was chosen to compare patients at similar stages of disease progression and to have the largest sample size possible.

Data Collection

Data were obtained for eligible patients using the VA Computerized Patient Record System. Demographic data gathered for each patient included age, sex, weight, height, body mass index (BMI), race, and riluzole use.

The primary endpoint was the change in ALSFRS-R score after 6 months of IV edaravone compared with standard-of-care ALS management. Secondary outcomes included change in ALSFRS-R scores 3, 12, 18, and 24 months after therapy initiation, change in %FVC and SIS 3, 6, 12, 18, and 24 months after therapy initiation, duration of edaravone completed (months), time to death (months), and adverse events.

 

Statistical Analysis

Comparisons between the edaravone and control groups for differences in patient characteristics were made using χ² and 2-sample t tests for categorical and continuous variables, respectively. Comparisons between the 2 groups for differences in study outcomes (ALSFRS-R scores, %FVC, SIS) at each time point were evaluated using 2-sample t tests. Adverse events and adverse drug reactions were compared between groups using χ² tests. Statistical significance was set at 0.05.

We estimated that a sample size of 21 subjects in the edaravone (case) group and 42 in the standard-of-care (control) group would be needed to achieve 80% power to detect a difference of 6.5 between the 2 groups for the change in ALSFRS-R scores. This 80% power was calculated based on a 2-sample t test, and assuming a 2-sided 5% significance level and a within-group SD of 8.5.⁹ Statistical analysis was conducted using Microsoft Excel.

Results

A total of 96 unique patients were seen in our multidisciplinary ALS clinic between September 1, 2014, and August 31, 2017 (Figure).

Of the 96 patients, 10 met exclusion criteria. From the remaining 86, 42 were randomly selected for the standard-of-care group. A total of 27 patients seen in multidisciplinary ALS clinic between September 1, 2017, and August 31, 2020, received at least 1 dose of IV edaravone. Of the 27 edaravone patients, 6 were excluded for not completing a total of 6 months of edaravone. Two of the 6 excluded developed a rash, which resolved within 1 week after discontinuing edaravone. The other 4 discontinued edaravone before 6 months because of disease progression.

Baseline Characteristics

Baseline demographics were similar between the groups (Table 1). Most patients were White men with a mean age of 60 years. Baseline %FVC was about 68%. Fewer patients in the standard-of-care group were taking riluzole than in the edaravone group (67% vs 95%, respectively; P = .002). Mean (SD) baseline SIS scores were slightly higher in the standard-of-care group vs the edaravone group (2.0 [1.0] vs 1.4 [0.6], respectively; P = .01).

Efficacy

No difference was found in the ALSFRS-R scores at 6 months between the IV edaravone and standard-of-care groups (P = .84) (Table 2). Our study did not meet power to calculate statistical analysis at 12, 18, and 24 months due to its size. No difference was found in change from baseline %FVC at 6 months between the 2 groups (P = .30) (Table 3). Change between the 2 groups in baseline SIS at 6 months also was not different (P = .69) (Table 4). Sample size was insufficient to calculate %FVC and SIS at the 12, 18, and 24 month intervals.

No difference was noted in time to death between groups (P = .93) (Table 5), and no adverse events were reported in either group.

 

Discussion

This 24-month, case-control retrospective study assessed efficacy and safety of IV edaravone for the management of ALS. Although the landmark edaravone study showed slowed progression of ALS at 6 and 12 months, the effectiveness of edaravone outside the clinical trial setting has been less compelling.^9-11,13 A later study showed no difference in change in ALSFRS-R score at 6 months compared with that of the placebo group.⁷ In our study, no statistically significant difference was found for change in ALSFRS-R scores at 6 months.

Our study was unique given we evaluated a veteran population. The link between the military and ALS is largely unknown, although studies have shown increased incidence of ALS in people with a military history compared with that of the general population.^16-18 Our study was also unique because it was single-centered in design and allowed for outcome assessments, including ALSFRS-R scores, SIS, and %FVC measurements, to all be conducted by the same practitioner to limit variability. Unfortunately, our sample size resulted in a cohort that was underpowered at 12, 18, and 24 months. In addition, there was a lack of data on chart review for SIS and %FVC measurements at 24 months. As ALS progresses toward end stage, SIS and %FVC measurements can become difficult and burdensome on the patient to obtain, and the ALS multidisciplinary team may decide not to gather these data points as ALS progresses. As a result, change in SIS and %FVC measurements were unable to be reported due to lack of gathering this information at the 24-month mark in the edaravone group. Due to the cost and administration burden associated with edaravone, it is important that assessment of disease progression is performed regularly to assess benefit and appropriateness of continued therapy. The oral formulation of edaravone was approved in 2022, shortly after the completion of data collection for this study.^19,20 Although our study did not analyze oral edaravone, the administration burden of treatment would be reduced with the oral formulation, and we hypothesize there will be increased patient interest in ALS management with oral vs IV edaravone. Evaluation of long-term treatment for efficacy and safety beyond 24 months has not been evaluated. Future studies should continue to evaluate edaravone use in a larger veteran population.

Limitations

One limitation for our study alluded to earlier in the discussion was sample size. Although this study met power at the 6-month mark, it was limited by the number of patients who received more than 6 months of edaravone (n = 21). As a result, statistical analyses between treatment groups were underpowered at 12, 18, and 24 months. Our study had 80% power to detect a difference of 6.5 between the groups for the change in ALSFRS-R scores. Previous studies detected a statistically significant difference in ALSFRS-R scores, with a difference in ALSFRS-R scores of 2.49 between groups.⁸ Future studies should evaluate a larger sample size of patients who are prescribed edaravone.

Another limitation was that the edaravone and standard-of-care group data were gathered from different time periods. Two different time frames were selected to increase sample size by gathering data over a longer period and to account for patients who may have qualified for IV edaravone but could not receive it as it was not yet available on the market. There were no known changes to the standard of care between the time periods that would affect results. As noted previously, the standard-of-care group had fewer patients taking riluzole compared with the edaravone group, which may have confounded our results. We concluded patients opting for edaravone were more likely to trial riluzole, taken by mouth twice daily, before starting edaravone, a once-daily IV infusion.

Conclusions

No difference in the rate of ALS progression was noted between patients who received IV edaravone vs standard of care at 6 months. In addition, no difference was noted in other objective measures of disease progression, including %FVC, SIS, and time to death. As a result, the decision to initiate and continue edaravone therapy should be made on an individualized basis according to a prescriber’s clinical judgment and a patient’s goals. Edaravone therapy should be discontinued when disease progression occurs or when medication administration becomes a burden.

Acknowledgments

This material is the result of work supported with resources and the use of facilities at Veteran Health Indiana.

Amyotrophic lateral sclerosis (ALS) is an incurable neurodegenerative disorder that results in progressive deterioration of motor neurons in the ventral horn of the spinal cord, which results in loss of voluntary muscle movements.¹ Eventually, typical daily tasks become difficult to perform, and as the disease progresses, the ability to eat and breathe is impaired.² Reports from 2015 show the annual incidence of ALS is 5 cases per 100,000 people, with the total number of cases reported at more than 16,000 in the United States.³ In clinical practice, disease progression is routinely assessed by the Revised Amyotrophic Lateral Sclerosis Functional Rating Scale (ALSFRS-R). Typical decline is 1 point per month.⁴

Unfortunately, at this time, ALS care focuses on symptom management, including prevention of weight loss; implementation of communication strategies; and management of pain, constipation, excess secretions, cramping, and breathing. Despite copious research into treatment options, few exist. Riluzole is an oral medication administered twice daily and has been on the market since 1995.^5-7 Efficacy was demonstrated in a study showing statistically significant survival at 12 months compared with controls (74% vs 58%, respectively; P = .014).⁶ Since its approval, riluzole has become part of standard-of-care ALS management.

In 2017, the US Food and Drug Administration (FDA) approved edaravone, an IV medication that was found to slow the progression of ALS in some patients.^8-12 Oxidative stress caused by free radicals is hypothesized to increase the progression of ALS by motor neuron degradation.¹³ Edaravone works as a free radical and peroxynitrite scavenger and has been shown to eliminate lipid peroxides and hydroxyl radicals known to damage endothelial and neuronal cells.¹²

Given the mechanism of action of edaravone, it seemed to be a promising option to slow the progression of ALS. A 2019 systematic review analyzed 3 randomized studies with 367 patients and found a statistically significant difference in change in ALSFRS-R scores between patients treated with edaravone for 24 weeks compared with patients treated with the placebo (mean difference, 1.63; 95% CI, 0.26-3.00; P = .02).¹² Secondary endpoints evaluated included percent forced vital capacity (%FVC), grip strength, and pinch strength: All showing no significant difference when comparing IV edaravone with placebo.

A 2022 postmarketing study of 324 patients with ALS evaluated the safety and efficacy of long-term edaravone treatment. IV edaravone therapy for > 24 weeks was well tolerated, although it was not associated with any disease-modifying benefit when comparing ALSFRS-R scores with patients not receiving edaravone over a median 13.9 months (ALSFRS-R points/month, -0.91 vs -0.85; P = .37).¹³ A third ALS treatment medication, sodium phenylbutyrate/taurursodiol was approved in 2022 but not available during our study period and not included here.^14,15

Studies have shown an increased incidence of ALS in the veteran population. Veterans serving in the Gulf War were nearly twice as likely to develop ALS as those not serving in the Gulf.¹⁶ However, existing literature regarding the effectiveness of edaravone does not specifically examine the effect on this unique population. The objective of this study was to assess the effect of IV edaravone on ALS progression in veterans compared with veterans who received standard of care.

 

Methods

This study was conducted at a large, academic US Department of Veterans Affairs (VA) medical center. Patients with ALS are followed by a multidisciplinary clinic composed of a neurologist, pulmonologist, clinical pharmacist, social worker, speech therapist, physical therapist, occupational therapist, dietician, clinical psychologist, wheelchair clinic representative, and benefits representative. Patients are typically seen for a half-day appointment about every 3 months. During these visits, a comprehensive review of disease progression is performed. This review entails completion of the ALSFRS-R, physical examination, and pulmonary function testing. Speech intelligibility stage (SIS) is assessed by a speech therapist as well. SIS is scored from 1 (no detectable speech disorder) to 5 (no functional speech). All patients followed in this multidisciplinary ALS clinic receive standard-of-care treatment. This includes the discussion of treatment options that if appropriate are provided to help manage a wide range of complications associated with this disease (eg, pain, cramping, constipation, excessive secretions, weight loss, dysphagia). As a part of these personal discussions, treatment with riluzole is also offered as a standard-of-care pharmacologic option.

Study Design

This retrospective case-control study was conducted using electronic health record data to compare ALS progression in patients on IV edaravone therapy with standard of care. The Indiana University/Purdue University, Indianapolis Institutional Review Board and the VA Research and Development Committee approved the study. The control cohort received the standard of care. Patients in the case cohort received standard of care and edaravone 60 mg infusions daily for an initial cycle of 14 days on treatment, followed by 14 days off. All subsequent cycles were 10 of 14 days on treatment followed by 14 days off. The initial 2 doses were administered in the outpatient infusion clinic to monitor for a hypersensitivity reaction. Patients then had a peripherally inserted central catheter line placed and received doses on days 3 through 14 at home. A port was placed for subsequent cycles, which were also completed at home. Appropriateness of edaravone therapy was assessed by the neurologist at each follow-up appointment. Therapy was then discontinued if warranted based on disease progression or patient preference.

Study Population

Patients included were aged 18 to 75 years with diagnosed ALS. Patients with complications that might influence evaluation of medication efficacy (eg, Parkinson disease, schizophrenia, significant dementia, other major medical morbidity) were excluded. Patients were also excluded if they were on continuous bilevel positive airway pressure and/or had a total score of ≤ 3 points on ALSFRS-R items for dyspnea, orthopnea, or respiratory insufficiency. Due to our small sample size, patients were excluded if treatment was < 6 months, which is the gold standard of therapy duration established by clinical trials.^9,11,12

The standard-of-care cohort included patients enrolled in the multidisciplinary clinic September 1, 2014 to August 31, 2017. These patients were compared in a 2:1 ratio with patients who received IV edaravone. The edaravone cohort included patients who initiated treatment with IV edaravone between September 1, 2017, and August 31, 2020. This date range prior to the approval of edaravone was chosen to compare patients at similar stages of disease progression and to have the largest sample size possible.

Data Collection

Data were obtained for eligible patients using the VA Computerized Patient Record System. Demographic data gathered for each patient included age, sex, weight, height, body mass index (BMI), race, and riluzole use.

The primary endpoint was the change in ALSFRS-R score after 6 months of IV edaravone compared with standard-of-care ALS management. Secondary outcomes included change in ALSFRS-R scores 3, 12, 18, and 24 months after therapy initiation, change in %FVC and SIS 3, 6, 12, 18, and 24 months after therapy initiation, duration of edaravone completed (months), time to death (months), and adverse events.

 

Statistical Analysis

Comparisons between the edaravone and control groups for differences in patient characteristics were made using χ² and 2-sample t tests for categorical and continuous variables, respectively. Comparisons between the 2 groups for differences in study outcomes (ALSFRS-R scores, %FVC, SIS) at each time point were evaluated using 2-sample t tests. Adverse events and adverse drug reactions were compared between groups using χ² tests. Statistical significance was set at 0.05.

We estimated that a sample size of 21 subjects in the edaravone (case) group and 42 in the standard-of-care (control) group would be needed to achieve 80% power to detect a difference of 6.5 between the 2 groups for the change in ALSFRS-R scores. This 80% power was calculated based on a 2-sample t test, and assuming a 2-sided 5% significance level and a within-group SD of 8.5.⁹ Statistical analysis was conducted using Microsoft Excel.

Results

A total of 96 unique patients were seen in our multidisciplinary ALS clinic between September 1, 2014, and August 31, 2017 (Figure).

Of the 96 patients, 10 met exclusion criteria. From the remaining 86, 42 were randomly selected for the standard-of-care group. A total of 27 patients seen in multidisciplinary ALS clinic between September 1, 2017, and August 31, 2020, received at least 1 dose of IV edaravone. Of the 27 edaravone patients, 6 were excluded for not completing a total of 6 months of edaravone. Two of the 6 excluded developed a rash, which resolved within 1 week after discontinuing edaravone. The other 4 discontinued edaravone before 6 months because of disease progression.

Baseline Characteristics

Baseline demographics were similar between the groups (Table 1). Most patients were White men with a mean age of 60 years. Baseline %FVC was about 68%. Fewer patients in the standard-of-care group were taking riluzole than in the edaravone group (67% vs 95%, respectively; P = .002). Mean (SD) baseline SIS scores were slightly higher in the standard-of-care group vs the edaravone group (2.0 [1.0] vs 1.4 [0.6], respectively; P = .01).

Efficacy

No difference was found in the ALSFRS-R scores at 6 months between the IV edaravone and standard-of-care groups (P = .84) (Table 2). Our study did not meet power to calculate statistical analysis at 12, 18, and 24 months due to its size. No difference was found in change from baseline %FVC at 6 months between the 2 groups (P = .30) (Table 3). Change between the 2 groups in baseline SIS at 6 months also was not different (P = .69) (Table 4). Sample size was insufficient to calculate %FVC and SIS at the 12, 18, and 24 month intervals.

No difference was noted in time to death between groups (P = .93) (Table 5), and no adverse events were reported in either group.

 

Discussion

This 24-month, case-control retrospective study assessed efficacy and safety of IV edaravone for the management of ALS. Although the landmark edaravone study showed slowed progression of ALS at 6 and 12 months, the effectiveness of edaravone outside the clinical trial setting has been less compelling.^9-11,13 A later study showed no difference in change in ALSFRS-R score at 6 months compared with that of the placebo group.⁷ In our study, no statistically significant difference was found for change in ALSFRS-R scores at 6 months.

Our study was unique given we evaluated a veteran population. The link between the military and ALS is largely unknown, although studies have shown increased incidence of ALS in people with a military history compared with that of the general population.^16-18 Our study was also unique because it was single-centered in design and allowed for outcome assessments, including ALSFRS-R scores, SIS, and %FVC measurements, to all be conducted by the same practitioner to limit variability. Unfortunately, our sample size resulted in a cohort that was underpowered at 12, 18, and 24 months. In addition, there was a lack of data on chart review for SIS and %FVC measurements at 24 months. As ALS progresses toward end stage, SIS and %FVC measurements can become difficult and burdensome on the patient to obtain, and the ALS multidisciplinary team may decide not to gather these data points as ALS progresses. As a result, change in SIS and %FVC measurements were unable to be reported due to lack of gathering this information at the 24-month mark in the edaravone group. Due to the cost and administration burden associated with edaravone, it is important that assessment of disease progression is performed regularly to assess benefit and appropriateness of continued therapy. The oral formulation of edaravone was approved in 2022, shortly after the completion of data collection for this study.^19,20 Although our study did not analyze oral edaravone, the administration burden of treatment would be reduced with the oral formulation, and we hypothesize there will be increased patient interest in ALS management with oral vs IV edaravone. Evaluation of long-term treatment for efficacy and safety beyond 24 months has not been evaluated. Future studies should continue to evaluate edaravone use in a larger veteran population.

Limitations

One limitation for our study alluded to earlier in the discussion was sample size. Although this study met power at the 6-month mark, it was limited by the number of patients who received more than 6 months of edaravone (n = 21). As a result, statistical analyses between treatment groups were underpowered at 12, 18, and 24 months. Our study had 80% power to detect a difference of 6.5 between the groups for the change in ALSFRS-R scores. Previous studies detected a statistically significant difference in ALSFRS-R scores, with a difference in ALSFRS-R scores of 2.49 between groups.⁸ Future studies should evaluate a larger sample size of patients who are prescribed edaravone.

Another limitation was that the edaravone and standard-of-care group data were gathered from different time periods. Two different time frames were selected to increase sample size by gathering data over a longer period and to account for patients who may have qualified for IV edaravone but could not receive it as it was not yet available on the market. There were no known changes to the standard of care between the time periods that would affect results. As noted previously, the standard-of-care group had fewer patients taking riluzole compared with the edaravone group, which may have confounded our results. We concluded patients opting for edaravone were more likely to trial riluzole, taken by mouth twice daily, before starting edaravone, a once-daily IV infusion.

Conclusions

No difference in the rate of ALS progression was noted between patients who received IV edaravone vs standard of care at 6 months. In addition, no difference was noted in other objective measures of disease progression, including %FVC, SIS, and time to death. As a result, the decision to initiate and continue edaravone therapy should be made on an individualized basis according to a prescriber’s clinical judgment and a patient’s goals. Edaravone therapy should be discontinued when disease progression occurs or when medication administration becomes a burden.

Acknowledgments

This material is the result of work supported with resources and the use of facilities at Veteran Health Indiana.