Original Research

The Veterans Health Administration (VHA) is the largest US integrated health system, providing care to veterans through VHA and non-VHA practitioners and facilities.^1,2 Providing high-quality, timely, and veteran-centric care remains a priority for the VHA. Legislative efforts have expanded opportunities for eligible veterans to receive care in the community purchased by VHA, known as community care (CC).¹ The Veterans Access, Choice, and Accountability Act of 2014 came in response to reports of long wait times and drive times for patients.^3-5 The MISSION Act of 2018 expanded access to CC by streamlining it and broadening eligibility criteria, especially for veterans in rural communities who often experience more barriers in accessing care than veterans living in urban communities.^1,6-10 Since the implementation of the Choice and MISSION Acts, > 2.7 million veterans have received care through community practitioners within the VHA CC network.¹¹

Background

Increased access to CC could benefit veterans living in rural communities by increasing care options and circumventing challenges to accessing VHA care (ie, geographic, transportation, and distance barriers, practitioner and specialist shortages, and hospital closures). ^5,9,10,12,13 However, health care system deficits in rural areas could also limit CC effectiveness for veterans living in those communities. ³ Other challenges posed by using CC include care coordination, information sharing, care continuity, delayed payments to CC practitioners, and mixed findings regarding CC quality.^5,8,13,14 VHA practitioners are specifically trained to meet the multifaceted needs unique to veterans’ health and subculture, training CC practitioners may not receive.^5,15

CC offers services for primary care and a broad range of specialties, including rehabilitation services such as physical therapy (PT).⁶ PT is used for the effective treatment of various conditions veterans experience and promote wellbeing and independence.¹⁶ US Department of Veterans Affairs (VA) databases reveal a high prevalence of veterans receiving PT services through CC; PT is one of the most frequently used CC outpatient specialty services by veterans living in rural communities.^14,17

Telerehabitltation Enterprisewide Initiative

VHA has greatly invested in delivering care virtually, especially for veterans living in rural communities.¹⁸ In 2017, the VHA Office of Rural Health funded the Telerehabilitation Enterprise-Wide Initiative (TR-EWI) in partnership with the Physical Medicine and Rehabilitation Services national program office to increase access to specialized rehabilitation services for veterans living in rural communities by leveraging telehealth technologies.^18-21 This alternative mode of health care delivery allows clinicians to overcome access barriers by delivering rehabilitation therapies directly to veterans' homes or nearby community-based outpatient clinics. TR-EWI was conceived as a hub-and-spoke model, where rehabilitation expertise at the hub was virtually delivered to spoke sites that did not have in-house expertise. In subsequent years, the TR-EWI also evolved to provide targeted telerehabilitation programs within rural-serving community-based outpatient clinics, including PT as a predominant service.^19,20

As TR-EWI progressed—and in conjunction with the uptake of telehealth across VHA during the COVID-19 pandemic—there has been increased focus on PT telerehabilitation, especially for the 4.6 million veterans in rural communities.^18,22,23 Because health care delivery system deficits in rural areas could limit the effective use of CC, many TR-EWI sites hope to reduce their CC referrals by providing telehealth PT services to veterans who might otherwise need to be referred to CC. This strategy aligns with VHA goals of providing high-quality and timely care. To better understand opportunities for programs like TR-EWI to provide rehabilitation services for veterans and reduce care sent to the community, research that examines CC referral trends for PT over time is warranted.

This study examines CC from a rehabilitation perspective with a focus on CC referral trends for PT, specifically for Veterans Integrated Service Networks (VISNs) where TREWI sites are located. The study’s objectives were to describe rehabilitation PT services being referred to CC and examine associated CC costs for PT services. Two research questions guided the study. First, what are the utilization trends for CC PT referrals from fiscal year (FY) 2019 to FY 2022? Secondly, what is the cost breakdown of CC for PT referrals from FY 2020 to FY 2022?

Methods

This study was conducted by a multidisciplinary team comprised of public health, disability, rehabilitation counseling, and PT professionals. It was deemed a quality improvement project under VA guidance and followed the SQUIRE guidelines for quality improvement reporting.^24,25 The study used the VA Common Operating Platform (Palantir) to obtain individual-level CC referral data from the HealthShare Referral Manager (HSRM) database and consult data from the Computerized Patient Record System. Palantir is used to store and integrate VA data derived from the VA Corporate Data Warehouse and VHA Support Service Center. Referrals are authorizations for care to be delivered by a CC practitioner.

TR-EWI is comprised of 7 sites: VISN 2, VISN 4, VISN 8, VISN 12, VISN 15, VISN 19, and VISN 22. Each site provides telerehabilitation services with an emphasis on reaching veterans living in rural communities. We joined the referrals and consults cubes in Palantir to extract PT referrals for FY 2019 to FY 2022 for the 7 VISNs with TR-EWI sites and obtain referral-specific information and demographic characteristics. ²⁶ Data were extracted in October 2022.

The VHA Community Care Referral Dashboard (CC Dashboard) provided nonindividual level CC cost data.²⁷ The CC Dashboard provides insights into the costs of CC services for VHA enrollees by category of care, standardized episode of care, and eligibility. Data are based on nationallevel HSRM referrals that are not suspended or linked to a canceled or discontinued consult. Data were aggregated by VISN. The dashboard only includes referrals dating back to FY 2020; therefore, PT data from FY 2020 through FY 2022 for VISNs with TR-EWI sites were collected. Data were extracted in December 2022.

This study examined CC referrals, station name, eligibility types, clinical diagnoses (International Classification of Diseases, Tenth Revision codes), and demographic information in the Palantir dataset. Six eligibility criteria can qualify a veteran to receive CC.²⁸ Within clinical diagnoses, the variable of interest was the provisional diagnosis. Patient demographics included age, gender, and rurality of residence, as determined by the Rural-Urban Commuting Area system.^29,30 Rural and highly rural categories were combined for analysis. For the CC cost dataset, this study examined CC referrals, referral cost, and eligibility type.

Analysis

For the first research question, we examined referral data from FY 2019 to FY 2022 using the Palantir dataset, performed descriptive statistical analysis for all variables, and analyzed data to identify trends. Descriptive statistics were completed using IBM SPSS Statistics for Windows Version 29.0.0.0.

A qualitative analysis of provisional diagnosis data revealed what is being referred to CC for PT. A preliminary overview of provisional diagnosis data was conducted to familiarize coders with the data. We developed a coding framework to categorize diagnoses based on anatomical location, body structure, and clinical areas of interest. Data were reviewed individually and grouped into categories within the coding framework before meeting as a team to achieve group consensus on categorization. We then totaled the frequency of occurrence for provisional diagnoses within each category. Qualitative analyses were completed using Microsoft Excel.

For the second research question, the study used the CC cost dataset to examine the cost breakdown of CC PT referrals from FY 2020 to FY 2022. We calculated the number and cost of PT referrals across eligibility groups for each FY and VISN. Data were analyzed using SPSS to identify cost trends.

Results

There were 344,406 referrals to CC for PT from FY 2019 to FY 2022 for the 7 VISNs analyzed (Table 1). Of these, 22.5% were from FY 2019, 19.1% from FY 2020, 28.2% from FY 2021, and 30.3% from FY 2022. VISN 8 and VISN 22 reported the most overall PT referrals, with VISN 8 comprising 22.2% and VISN 22 comprising 18.1% of all referrals. VISN 2 reported the least overall referrals (3.7%). VISN 4 and VISN 12 had decreases in referrals over time. VISN 2 and VISN 15 had decreases in referrals from FY 2019 to FY 2021 and slight increases from FY 2021 to FY 2022. VISN 19 and VISN 22 both saw slight increases from FY 2019 to FY 2020 and substantial increases from FY 2020 to FY 2022, with FY 2022 accounting for 40.0% and 42.3% of all referrals for VISN 19 and VISN 20, respectively (Figure 1).

For FY 2019 and FY 2020, VISN 8 had the highest percentage of referrals (26.7% and 23.2%, respectively), whereas VISN 22 was among the lowest (7.3% and 11.4%, respectively). However, for FY 2021 and FY 2022, VISN 22 reported the highest percentage of referrals (23.5% and 25.3%, respectively) compared to all other VISNs. VISN 2 consistently reported the lowest percentage of referrals across all years.

There were 56 stations analyzed across the 7 VISNs (Appendix 1). Nine stations each accounted for ≥ 3.0% of the total PT referrals and only 2 stations accounted for > 5.0% of referrals. Orlando, Florida (6.0%), Philadelphia, Pennsylvania (5.2%), Tampa, Florida (4.9%), Aurora, Colorado (4.9%), and Gainesville, Florida (4.4%) reported the top 5 highest referrals, with 3 being from VISN 8 (Orlando, Tampa, Gainesville). Stations with the lowest reported referrals were all in VISN 2 in New York: The Bronx, (0%), New York Harbor (0%), Hudson Valley (0.1%) and Finger Lakes (0.2%).

Rurality

Urban stations comprised 56.2% and rural stations comprised 39.8% of PT CC referrals, while 0.2% of referrals were from insular isle US territories: Guam, American Samoa, Northern Marianas, and the Virgin Islands. The sample had missing or unknown data for 3.8% of referrals. FY 2022 had the largest difference in rural and urban referrals. Additionally, there was an overall trend of more referrals over time for rural and urban, with a large increase in rural (+40.0%) and urban (+62.7%) referrals from FY 2020 to FY 2021 and a modest increase from FY 2021 to FY 2022 (+5.2% for rural and +9.1% for urban). There was a decrease in rural (-7.0%) and urban (-3.5%) referrals from FY 2019 to FY 2020 (Figure 2).

There were differences in referrals by rurality and VISN (Table 2). VISN 12, VISN 15, and VISN 19 reported more rural than urban referrals, whereas VISN 4, VISN 8, and VISN 22 reported more urban than rural referrals. VISN 2 reported similar numbers for both, with slightly more urban than rural referrals. When reviewing trends over time for each FY, VISN 12, VISN 15, and VISN 19 reported more rural than urban referrals and VISN 4, VISN 8, and VISN 22 had more urban than rural referrals. In FY 2019 and FY 2020, VISN 2 reported slightly more urban than rural referrals but almost the same number of referrals in FY 2021 and FY 2022 (Appendix 2).

Demographics

The mean (SD) age was 61.2 (15.8) years (range, 20-105). Most PT CC referrals were for veterans aged 70 to 79 years (26.9%), followed by 60 to 69 years (20.7%), and 50 to 59 years (16.4%) (Appendix 3). Trends were consistent across VISNs. There was less of a difference between rural and urban referral percentages as the population aged. Veterans aged < 49 years residing in more urban areas accounted for more referrals to CC compared to their rural counterparts. This difference was less apparent in the 70 to 79 years and 80 to 89 years age brackets.

Most PT CC referrals (81.2%) were male and 14.8% were female. About 3.6% of referral data were missing sex information, and there was a smaller difference between male veterans living in rural communities and male veterans living in urban communities compared with female veterans. A total of 42.9% of male veterans resided in rural areas compared to 56.8% in urban areas; 32.7% of female veterans resided in rural areas compared to 66.9% in urban areas (Appendix 3).

Other Criteria

Of the 334,406 referrals, 114,983 (34.4%) had eligibility data, mostly from FY 2021 and FY 2022 (Table 3). Available eligibility data were likely affected by the MISSION Act and new regulations for reporting CC eligibility. Distance (33.4%) was the most common eligibility criteria, followed by timeliness of care (28.8%), and best medical interest (19.8%); 40.4% were rural and 59.5% were urban. Distance (55.4%) was most common for rural veterans, while timeliness of care (39.7%) was most common for urban veterans. For both groups, the second most common eligibility reason was best medical interest (Appendix 4).

Bone, joint, or soft tissue disorders were common diagnoses, with 25.2% located in the lower back, 14.7% in the shoulder, and 12.8% in the knee (Appendix 5). Amputations of the upper and lower limbs, fractures, cancer-related diagnoses, integumentary system disorders, thoracic and abdominal injuries and disorders, and other medical and mental health conditions each accounted for < 1% of the total diagnoses.

Costs

At time of analysis, the CC Dashboard had cost data available for 200,204 CC PT referrals from FY 2020 to FY 2022. The difference in referral numbers for the 2 datasets is likely attributed to several factors: CC cost data is exclusively from the HSRM, whereas Palantir includes other data sources; how VA cleans data pulled into Palantir; how the CC Dashboard algorithm populates data; and variances based on timing of reporting and/or if referrals are eventually canceled.

The total cost of PT CC referrals from FY 2020 to FY 2022 in selected VISNs was about $220,615,399 (Appendix 6). Appendix 7 details the methodology for determining the average standardized episode- of-care cost by VISN and how referral costs are calculated. Data show a continuous increase in total estimated cost from $46.8 million in FY 2020 to $92.1 million in FY 2022. From FY 2020 to FY 2022, aggregate costs ranged from $6,758,053 in VISN 2 to $47,209,162 in VISN 8 (Figure 3). The total referral cost for PT was highest at VISN 4 in FY 2020 ($10,447,140) and highest at VISN 22 in FY 2021 ($18,835,657) and FY 2022 ($22,962,438) (Figure 4). For referral costs from FY 2020 to FY 2022, distance accounted for $75,561,948 (34.3%), timeliness of care accounted for $60,413,496 (27.3%), and best medical interest accounted for $46,291,390 (21.0%) (Table 4).

Overall costs were primarily driven by specific VISNs within each eligibility type (Appendix 8; Figure 5). VISN 19, VISN 22, and VISN 15 accounted for the highest referral costs for distance; VISN 22, VISN 8, and VISN 19 accounted for the secondhighest referral cost, timeliness of care; and VISN 4, VISN 8, and VISN 12 accounted for the third-highest referral cost, best medical interest (Figure 5). VISN 2, VISN 4, VISN 12, VISN 15, and VISN 22 had service unavailable as an eligibility type with 1 of the top 3 associated referral costs, which was higher in cost than timeliness of care for VISN 2, VISN 4, VISN 12, and VISN 15.

Discussion

This study examines the referral of rehabilitation PT services to CC, evaluates CC costs for PT services, and analyzes utilization and cost trends among veterans within the VHA. Utilization data demonstrated a decrease in referrals from FY 2019 to FY 2020 and increases in referrals from FY 2020 to FY 2022 for most variables of interest, with cost data exhibiting similar trends. Results highlight the need for further investigation to address variations in PT referrals and costs across VISNs and eligibility reasons for CC referral.

Results demonstrated a noteworthy increase in PT CC referrals over time. The largest increase occurred from FY 2020 to FY 2021, with a smaller increase from FY 2021 to FY 2022. During this period, total enrollee numbers decreased by 3.0% across the 7 VISNs included in this analysis and by 1.6% across all VISNs, a trend that illustrates an overall decrease in enrollees as CC use increased. Results align with the implementation of the MISSION Act of 2018, which further expanded veterans’ options to use CC.^1,6,7 Results also align with the onset of the COVID-19 pandemic, which disrupted care access for many veterans, placed a larger emphasis on the use of telehealth, and increased opportunities to stay within the VA for care by rapidly shifting to telehealth and leveraging telerehabilitation investments and initiatives (such as TR-EWI).^20,31

VISN 8, VISN 19, and VISN 22, accounted for more than half of PT referrals. These VISNs had higher enrollee counts compared to the other VISNs.³² VISN 8 consistently had high levels of referrals, whereas VISN 19 and VISN 22 saw dramatic increases in FY 2021 and FY 2022. In contrast, VISN 4 and VISN 12 gradually decreased referrals during the study. VISN 2 had the lowest referral numbers during the study period, and all stations with the lowest individual referral numbers were located within VISN 2. Of the VISNs included in this study, VISN 2 had the second lowest number of enrollees (324,042).³² Reasons for increases and decreases over time could not be determined based on data collected in this study.

There were more urban than rural PT CC referrals; however, both exhibited an increase in referrals over time. This is consistent with population trends showing that most VHA patients (62.6%) and veterans (75.9%) reside in urban areas, which could explain some of the trends in this study.³³ Some VISNs have larger urban catchment areas (eg, VISN 8 and VISN 22), and some have larger rural catchment areas (eg, VISN 15 and VISN 19), which could partially explain the rural-urban differences by VISN.³² Rural-urban referral trends might also reflect existing health care delivery system deficits in rural areas and known challenges associated with accessing health care for veterans living in rural communities.^8,9

This study found larger differences in rural and urban PT CC referrals for younger age groups, with more than twice as many urban referrals in veterans aged 20 to 29 years and aged 30 to 39 years, and roughly 1.8 times as many urban referrals in veterans aged 40 to 49 years. However, there were similar numbers of rural and urban referrals in those aged 70 to 79 years and aged 80 to 89 years. These trends are consistent with data showing veterans residing in rural communities are older than their urban counterparts.^23,34 Data suggest that older veteran populations might seek PT at higher rates than younger veteran populations. Moreover, data suggest there could be differences in PT-seeking rates for younger veteran populations who reside in rural vs urban areas. Additional research is needed to understand these trends.

Distance and timeliness of care were the predominant reasons for referral among eligibility groups, which is consistent with the MISSION Act goals.^1,6,7 The most common eligibility reason for rural referrals was distance; timeliness of care was most common for urban referrals. This finding is expected, as veterans living in rural communities are farther away from VHA facilities and have longer drive times, whereas veterans living in urban communities might live closer, yet experience longer wait times due to services and/or appointment availability. Best medical interest accounted for almost 20% of referrals, which does not provide detailed insights into why those veterans were referred to CC.

The top PT diagnoses referred to CC were related to bone, joint, or soft tissue disorders of the lower back, shoulder, and knee. This suggests that musculoskeletal-related issues are prevalent among veterans seeking PT care, which is consistent with research that found > 50% of veterans receiving VHA care have musculoskeletal disorders.³⁵ The probability of experiencing musculoskeletal problems increases with age, as does the need for PT services. Amputations and fractures accounted for < 1% of CC referrals, which is consistent with the historic provision of VHA clinical specialized care to conditions prevalent among veterans. It may also represent VHA efforts to internally provide care for complex conditions requiring more extensive interdisciplinary coordination.

The total cost of referrals over time was about $221 million. VISN 8 accounted for the highest overall cost; VISN 2 had the lowest, mirroring referral utilization trends and aligning with VISN enrollee numbers. VISN 19 and VISN 22 reported large cost increases from FY 2020 to FY 2021. Total referral costs increased by $34.9 million from FY 2020 to FY 2021, which may be due to health care inflation (2.9% during FY 2019 to FY 2022), increased awareness of CC services, or increased VHA wait times.³⁶ Additionally, there were limitations in care provided across health care systems during the COVID-19 pandemic, including the VA.⁵ The increase from FY 2020 to FY 2021 may reflect a rebound from restrictions in appointments across VA, CC, and the private sector.

While the increase in total referral cost may be partly attributed to inflation, the cost effectiveness and efficiency of referring veterans to CC vs keeping veterans within VHA care is an ongoing debate.⁵ Examining and addressing cost drivers within the top eligibility types and their respective VISNs is necessary to determine resource allocation and improve quality of care. This study found that best medical interest and unavailable services accounted for 33.4% of the total cost of CC referrals, highlighting the need for policies that strengthen in-house competencies and recruit personnel to provide PT services currently unavailable within the VA.

Future Directions

The VHA should explore opportunities for in-house care, especially for services appropriate for telehealth.^18,20,37 Data indicated a smaller cost increase from FY 2021 to FY 2022 compared to the relatively large increase from FY 2020 to FY 2021. The increased telehealth usage across VHA by TR-EWI and non—TR-EWI sites within selected VISNs may have contributed to limiting the increase in CC costs. Future studies should investigate contextual factors of increased telehealth usage, which would offer guidance for implementation to optimize the integration of telehealth with PT rehabilitation provided in-house. Additionally, future studies can examine potential limitations experienced during PT telehealth visits, such as the inability to conduct hands-on assessments, challenges in viewing the quality of patient movement, ensuring patient safety in the remote environment, and the lack of PT equipment in homes for telehealth visits, and how these challenges are being addressed.^38,39 Research is also needed to understand tradeoffs of CC vs VHA care and the potential and cost benefits of keeping veterans within VHA using programs like TR-EWI.⁵ Veterans living in rural communities may especially benefit from this as expanding telehealth options can provide access to PT care that may not be readily available, enabling them to stay connected and engaged in their care.^18,40

Future studies could examine contributory factors to rising costs, such as demographic shifts, changes in PT service utilization, and policy. Researchers might also consider qualitative studies with clinicians and veterans within each VISN, which may provide insights into how local factors impact PT referral to the community.

Limitations

Due to its descriptive nature, this study can only speculate about factors influencing trends. Limitations include the inability to link the Palantir and CC Dashboard datasets for cost comparisons and potential data change over time on Palantir due to platform updates. The focus on VISNs with TREWI sites limited generalizability and this study did not compare CC PT vs VHA PT. Finally, there may have been cost drivers not identified in this study.

Conclusions

This descriptive study provides insights into the utilization and cost of PT CC referrals for selected VISNs. Cost trends underscore the financial commitment to providing PT services to veterans. Understanding what factors are driving this cost is necessary for VHA to optimally provide and manage the rehabilitation resources needed to serve veterans through traditional in-person care, telehealth, and CC options while ensuring timely, highquality care.

The Veterans Health Administration (VHA) is the largest US integrated health system, providing care to veterans through VHA and non-VHA practitioners and facilities.^1,2 Providing high-quality, timely, and veteran-centric care remains a priority for the VHA. Legislative efforts have expanded opportunities for eligible veterans to receive care in the community purchased by VHA, known as community care (CC).¹ The Veterans Access, Choice, and Accountability Act of 2014 came in response to reports of long wait times and drive times for patients.^3-5 The MISSION Act of 2018 expanded access to CC by streamlining it and broadening eligibility criteria, especially for veterans in rural communities who often experience more barriers in accessing care than veterans living in urban communities.^1,6-10 Since the implementation of the Choice and MISSION Acts, > 2.7 million veterans have received care through community practitioners within the VHA CC network.¹¹

Background

Increased access to CC could benefit veterans living in rural communities by increasing care options and circumventing challenges to accessing VHA care (ie, geographic, transportation, and distance barriers, practitioner and specialist shortages, and hospital closures). ^5,9,10,12,13 However, health care system deficits in rural areas could also limit CC effectiveness for veterans living in those communities. ³ Other challenges posed by using CC include care coordination, information sharing, care continuity, delayed payments to CC practitioners, and mixed findings regarding CC quality.^5,8,13,14 VHA practitioners are specifically trained to meet the multifaceted needs unique to veterans’ health and subculture, training CC practitioners may not receive.^5,15

CC offers services for primary care and a broad range of specialties, including rehabilitation services such as physical therapy (PT).⁶ PT is used for the effective treatment of various conditions veterans experience and promote wellbeing and independence.¹⁶ US Department of Veterans Affairs (VA) databases reveal a high prevalence of veterans receiving PT services through CC; PT is one of the most frequently used CC outpatient specialty services by veterans living in rural communities.^14,17

Telerehabitltation Enterprisewide Initiative

VHA has greatly invested in delivering care virtually, especially for veterans living in rural communities.¹⁸ In 2017, the VHA Office of Rural Health funded the Telerehabilitation Enterprise-Wide Initiative (TR-EWI) in partnership with the Physical Medicine and Rehabilitation Services national program office to increase access to specialized rehabilitation services for veterans living in rural communities by leveraging telehealth technologies.^18-21 This alternative mode of health care delivery allows clinicians to overcome access barriers by delivering rehabilitation therapies directly to veterans' homes or nearby community-based outpatient clinics. TR-EWI was conceived as a hub-and-spoke model, where rehabilitation expertise at the hub was virtually delivered to spoke sites that did not have in-house expertise. In subsequent years, the TR-EWI also evolved to provide targeted telerehabilitation programs within rural-serving community-based outpatient clinics, including PT as a predominant service.^19,20

As TR-EWI progressed—and in conjunction with the uptake of telehealth across VHA during the COVID-19 pandemic—there has been increased focus on PT telerehabilitation, especially for the 4.6 million veterans in rural communities.^18,22,23 Because health care delivery system deficits in rural areas could limit the effective use of CC, many TR-EWI sites hope to reduce their CC referrals by providing telehealth PT services to veterans who might otherwise need to be referred to CC. This strategy aligns with VHA goals of providing high-quality and timely care. To better understand opportunities for programs like TR-EWI to provide rehabilitation services for veterans and reduce care sent to the community, research that examines CC referral trends for PT over time is warranted.

This study examines CC from a rehabilitation perspective with a focus on CC referral trends for PT, specifically for Veterans Integrated Service Networks (VISNs) where TREWI sites are located. The study’s objectives were to describe rehabilitation PT services being referred to CC and examine associated CC costs for PT services. Two research questions guided the study. First, what are the utilization trends for CC PT referrals from fiscal year (FY) 2019 to FY 2022? Secondly, what is the cost breakdown of CC for PT referrals from FY 2020 to FY 2022?

Methods

This study was conducted by a multidisciplinary team comprised of public health, disability, rehabilitation counseling, and PT professionals. It was deemed a quality improvement project under VA guidance and followed the SQUIRE guidelines for quality improvement reporting.^24,25 The study used the VA Common Operating Platform (Palantir) to obtain individual-level CC referral data from the HealthShare Referral Manager (HSRM) database and consult data from the Computerized Patient Record System. Palantir is used to store and integrate VA data derived from the VA Corporate Data Warehouse and VHA Support Service Center. Referrals are authorizations for care to be delivered by a CC practitioner.

TR-EWI is comprised of 7 sites: VISN 2, VISN 4, VISN 8, VISN 12, VISN 15, VISN 19, and VISN 22. Each site provides telerehabilitation services with an emphasis on reaching veterans living in rural communities. We joined the referrals and consults cubes in Palantir to extract PT referrals for FY 2019 to FY 2022 for the 7 VISNs with TR-EWI sites and obtain referral-specific information and demographic characteristics. ²⁶ Data were extracted in October 2022.

The VHA Community Care Referral Dashboard (CC Dashboard) provided nonindividual level CC cost data.²⁷ The CC Dashboard provides insights into the costs of CC services for VHA enrollees by category of care, standardized episode of care, and eligibility. Data are based on nationallevel HSRM referrals that are not suspended or linked to a canceled or discontinued consult. Data were aggregated by VISN. The dashboard only includes referrals dating back to FY 2020; therefore, PT data from FY 2020 through FY 2022 for VISNs with TR-EWI sites were collected. Data were extracted in December 2022.

This study examined CC referrals, station name, eligibility types, clinical diagnoses (International Classification of Diseases, Tenth Revision codes), and demographic information in the Palantir dataset. Six eligibility criteria can qualify a veteran to receive CC.²⁸ Within clinical diagnoses, the variable of interest was the provisional diagnosis. Patient demographics included age, gender, and rurality of residence, as determined by the Rural-Urban Commuting Area system.^29,30 Rural and highly rural categories were combined for analysis. For the CC cost dataset, this study examined CC referrals, referral cost, and eligibility type.

Analysis

For the first research question, we examined referral data from FY 2019 to FY 2022 using the Palantir dataset, performed descriptive statistical analysis for all variables, and analyzed data to identify trends. Descriptive statistics were completed using IBM SPSS Statistics for Windows Version 29.0.0.0.

A qualitative analysis of provisional diagnosis data revealed what is being referred to CC for PT. A preliminary overview of provisional diagnosis data was conducted to familiarize coders with the data. We developed a coding framework to categorize diagnoses based on anatomical location, body structure, and clinical areas of interest. Data were reviewed individually and grouped into categories within the coding framework before meeting as a team to achieve group consensus on categorization. We then totaled the frequency of occurrence for provisional diagnoses within each category. Qualitative analyses were completed using Microsoft Excel.

For the second research question, the study used the CC cost dataset to examine the cost breakdown of CC PT referrals from FY 2020 to FY 2022. We calculated the number and cost of PT referrals across eligibility groups for each FY and VISN. Data were analyzed using SPSS to identify cost trends.

Results

There were 344,406 referrals to CC for PT from FY 2019 to FY 2022 for the 7 VISNs analyzed (Table 1). Of these, 22.5% were from FY 2019, 19.1% from FY 2020, 28.2% from FY 2021, and 30.3% from FY 2022. VISN 8 and VISN 22 reported the most overall PT referrals, with VISN 8 comprising 22.2% and VISN 22 comprising 18.1% of all referrals. VISN 2 reported the least overall referrals (3.7%). VISN 4 and VISN 12 had decreases in referrals over time. VISN 2 and VISN 15 had decreases in referrals from FY 2019 to FY 2021 and slight increases from FY 2021 to FY 2022. VISN 19 and VISN 22 both saw slight increases from FY 2019 to FY 2020 and substantial increases from FY 2020 to FY 2022, with FY 2022 accounting for 40.0% and 42.3% of all referrals for VISN 19 and VISN 20, respectively (Figure 1).

For FY 2019 and FY 2020, VISN 8 had the highest percentage of referrals (26.7% and 23.2%, respectively), whereas VISN 22 was among the lowest (7.3% and 11.4%, respectively). However, for FY 2021 and FY 2022, VISN 22 reported the highest percentage of referrals (23.5% and 25.3%, respectively) compared to all other VISNs. VISN 2 consistently reported the lowest percentage of referrals across all years.

There were 56 stations analyzed across the 7 VISNs (Appendix 1). Nine stations each accounted for ≥ 3.0% of the total PT referrals and only 2 stations accounted for > 5.0% of referrals. Orlando, Florida (6.0%), Philadelphia, Pennsylvania (5.2%), Tampa, Florida (4.9%), Aurora, Colorado (4.9%), and Gainesville, Florida (4.4%) reported the top 5 highest referrals, with 3 being from VISN 8 (Orlando, Tampa, Gainesville). Stations with the lowest reported referrals were all in VISN 2 in New York: The Bronx, (0%), New York Harbor (0%), Hudson Valley (0.1%) and Finger Lakes (0.2%).

Rurality

Urban stations comprised 56.2% and rural stations comprised 39.8% of PT CC referrals, while 0.2% of referrals were from insular isle US territories: Guam, American Samoa, Northern Marianas, and the Virgin Islands. The sample had missing or unknown data for 3.8% of referrals. FY 2022 had the largest difference in rural and urban referrals. Additionally, there was an overall trend of more referrals over time for rural and urban, with a large increase in rural (+40.0%) and urban (+62.7%) referrals from FY 2020 to FY 2021 and a modest increase from FY 2021 to FY 2022 (+5.2% for rural and +9.1% for urban). There was a decrease in rural (-7.0%) and urban (-3.5%) referrals from FY 2019 to FY 2020 (Figure 2).

There were differences in referrals by rurality and VISN (Table 2). VISN 12, VISN 15, and VISN 19 reported more rural than urban referrals, whereas VISN 4, VISN 8, and VISN 22 reported more urban than rural referrals. VISN 2 reported similar numbers for both, with slightly more urban than rural referrals. When reviewing trends over time for each FY, VISN 12, VISN 15, and VISN 19 reported more rural than urban referrals and VISN 4, VISN 8, and VISN 22 had more urban than rural referrals. In FY 2019 and FY 2020, VISN 2 reported slightly more urban than rural referrals but almost the same number of referrals in FY 2021 and FY 2022 (Appendix 2).

Demographics

The mean (SD) age was 61.2 (15.8) years (range, 20-105). Most PT CC referrals were for veterans aged 70 to 79 years (26.9%), followed by 60 to 69 years (20.7%), and 50 to 59 years (16.4%) (Appendix 3). Trends were consistent across VISNs. There was less of a difference between rural and urban referral percentages as the population aged. Veterans aged < 49 years residing in more urban areas accounted for more referrals to CC compared to their rural counterparts. This difference was less apparent in the 70 to 79 years and 80 to 89 years age brackets.

Most PT CC referrals (81.2%) were male and 14.8% were female. About 3.6% of referral data were missing sex information, and there was a smaller difference between male veterans living in rural communities and male veterans living in urban communities compared with female veterans. A total of 42.9% of male veterans resided in rural areas compared to 56.8% in urban areas; 32.7% of female veterans resided in rural areas compared to 66.9% in urban areas (Appendix 3).

Other Criteria

Of the 334,406 referrals, 114,983 (34.4%) had eligibility data, mostly from FY 2021 and FY 2022 (Table 3). Available eligibility data were likely affected by the MISSION Act and new regulations for reporting CC eligibility. Distance (33.4%) was the most common eligibility criteria, followed by timeliness of care (28.8%), and best medical interest (19.8%); 40.4% were rural and 59.5% were urban. Distance (55.4%) was most common for rural veterans, while timeliness of care (39.7%) was most common for urban veterans. For both groups, the second most common eligibility reason was best medical interest (Appendix 4).

Bone, joint, or soft tissue disorders were common diagnoses, with 25.2% located in the lower back, 14.7% in the shoulder, and 12.8% in the knee (Appendix 5). Amputations of the upper and lower limbs, fractures, cancer-related diagnoses, integumentary system disorders, thoracic and abdominal injuries and disorders, and other medical and mental health conditions each accounted for < 1% of the total diagnoses.

Costs

At time of analysis, the CC Dashboard had cost data available for 200,204 CC PT referrals from FY 2020 to FY 2022. The difference in referral numbers for the 2 datasets is likely attributed to several factors: CC cost data is exclusively from the HSRM, whereas Palantir includes other data sources; how VA cleans data pulled into Palantir; how the CC Dashboard algorithm populates data; and variances based on timing of reporting and/or if referrals are eventually canceled.

The total cost of PT CC referrals from FY 2020 to FY 2022 in selected VISNs was about $220,615,399 (Appendix 6). Appendix 7 details the methodology for determining the average standardized episode- of-care cost by VISN and how referral costs are calculated. Data show a continuous increase in total estimated cost from $46.8 million in FY 2020 to $92.1 million in FY 2022. From FY 2020 to FY 2022, aggregate costs ranged from $6,758,053 in VISN 2 to $47,209,162 in VISN 8 (Figure 3). The total referral cost for PT was highest at VISN 4 in FY 2020 ($10,447,140) and highest at VISN 22 in FY 2021 ($18,835,657) and FY 2022 ($22,962,438) (Figure 4). For referral costs from FY 2020 to FY 2022, distance accounted for $75,561,948 (34.3%), timeliness of care accounted for $60,413,496 (27.3%), and best medical interest accounted for $46,291,390 (21.0%) (Table 4).

Overall costs were primarily driven by specific VISNs within each eligibility type (Appendix 8; Figure 5). VISN 19, VISN 22, and VISN 15 accounted for the highest referral costs for distance; VISN 22, VISN 8, and VISN 19 accounted for the secondhighest referral cost, timeliness of care; and VISN 4, VISN 8, and VISN 12 accounted for the third-highest referral cost, best medical interest (Figure 5). VISN 2, VISN 4, VISN 12, VISN 15, and VISN 22 had service unavailable as an eligibility type with 1 of the top 3 associated referral costs, which was higher in cost than timeliness of care for VISN 2, VISN 4, VISN 12, and VISN 15.

Discussion

This study examines the referral of rehabilitation PT services to CC, evaluates CC costs for PT services, and analyzes utilization and cost trends among veterans within the VHA. Utilization data demonstrated a decrease in referrals from FY 2019 to FY 2020 and increases in referrals from FY 2020 to FY 2022 for most variables of interest, with cost data exhibiting similar trends. Results highlight the need for further investigation to address variations in PT referrals and costs across VISNs and eligibility reasons for CC referral.

Results demonstrated a noteworthy increase in PT CC referrals over time. The largest increase occurred from FY 2020 to FY 2021, with a smaller increase from FY 2021 to FY 2022. During this period, total enrollee numbers decreased by 3.0% across the 7 VISNs included in this analysis and by 1.6% across all VISNs, a trend that illustrates an overall decrease in enrollees as CC use increased. Results align with the implementation of the MISSION Act of 2018, which further expanded veterans’ options to use CC.^1,6,7 Results also align with the onset of the COVID-19 pandemic, which disrupted care access for many veterans, placed a larger emphasis on the use of telehealth, and increased opportunities to stay within the VA for care by rapidly shifting to telehealth and leveraging telerehabilitation investments and initiatives (such as TR-EWI).^20,31

VISN 8, VISN 19, and VISN 22, accounted for more than half of PT referrals. These VISNs had higher enrollee counts compared to the other VISNs.³² VISN 8 consistently had high levels of referrals, whereas VISN 19 and VISN 22 saw dramatic increases in FY 2021 and FY 2022. In contrast, VISN 4 and VISN 12 gradually decreased referrals during the study. VISN 2 had the lowest referral numbers during the study period, and all stations with the lowest individual referral numbers were located within VISN 2. Of the VISNs included in this study, VISN 2 had the second lowest number of enrollees (324,042).³² Reasons for increases and decreases over time could not be determined based on data collected in this study.

There were more urban than rural PT CC referrals; however, both exhibited an increase in referrals over time. This is consistent with population trends showing that most VHA patients (62.6%) and veterans (75.9%) reside in urban areas, which could explain some of the trends in this study.³³ Some VISNs have larger urban catchment areas (eg, VISN 8 and VISN 22), and some have larger rural catchment areas (eg, VISN 15 and VISN 19), which could partially explain the rural-urban differences by VISN.³² Rural-urban referral trends might also reflect existing health care delivery system deficits in rural areas and known challenges associated with accessing health care for veterans living in rural communities.^8,9

This study found larger differences in rural and urban PT CC referrals for younger age groups, with more than twice as many urban referrals in veterans aged 20 to 29 years and aged 30 to 39 years, and roughly 1.8 times as many urban referrals in veterans aged 40 to 49 years. However, there were similar numbers of rural and urban referrals in those aged 70 to 79 years and aged 80 to 89 years. These trends are consistent with data showing veterans residing in rural communities are older than their urban counterparts.^23,34 Data suggest that older veteran populations might seek PT at higher rates than younger veteran populations. Moreover, data suggest there could be differences in PT-seeking rates for younger veteran populations who reside in rural vs urban areas. Additional research is needed to understand these trends.

Distance and timeliness of care were the predominant reasons for referral among eligibility groups, which is consistent with the MISSION Act goals.^1,6,7 The most common eligibility reason for rural referrals was distance; timeliness of care was most common for urban referrals. This finding is expected, as veterans living in rural communities are farther away from VHA facilities and have longer drive times, whereas veterans living in urban communities might live closer, yet experience longer wait times due to services and/or appointment availability. Best medical interest accounted for almost 20% of referrals, which does not provide detailed insights into why those veterans were referred to CC.

The top PT diagnoses referred to CC were related to bone, joint, or soft tissue disorders of the lower back, shoulder, and knee. This suggests that musculoskeletal-related issues are prevalent among veterans seeking PT care, which is consistent with research that found > 50% of veterans receiving VHA care have musculoskeletal disorders.³⁵ The probability of experiencing musculoskeletal problems increases with age, as does the need for PT services. Amputations and fractures accounted for < 1% of CC referrals, which is consistent with the historic provision of VHA clinical specialized care to conditions prevalent among veterans. It may also represent VHA efforts to internally provide care for complex conditions requiring more extensive interdisciplinary coordination.

The total cost of referrals over time was about $221 million. VISN 8 accounted for the highest overall cost; VISN 2 had the lowest, mirroring referral utilization trends and aligning with VISN enrollee numbers. VISN 19 and VISN 22 reported large cost increases from FY 2020 to FY 2021. Total referral costs increased by $34.9 million from FY 2020 to FY 2021, which may be due to health care inflation (2.9% during FY 2019 to FY 2022), increased awareness of CC services, or increased VHA wait times.³⁶ Additionally, there were limitations in care provided across health care systems during the COVID-19 pandemic, including the VA.⁵ The increase from FY 2020 to FY 2021 may reflect a rebound from restrictions in appointments across VA, CC, and the private sector.

While the increase in total referral cost may be partly attributed to inflation, the cost effectiveness and efficiency of referring veterans to CC vs keeping veterans within VHA care is an ongoing debate.⁵ Examining and addressing cost drivers within the top eligibility types and their respective VISNs is necessary to determine resource allocation and improve quality of care. This study found that best medical interest and unavailable services accounted for 33.4% of the total cost of CC referrals, highlighting the need for policies that strengthen in-house competencies and recruit personnel to provide PT services currently unavailable within the VA.

Future Directions

The VHA should explore opportunities for in-house care, especially for services appropriate for telehealth.^18,20,37 Data indicated a smaller cost increase from FY 2021 to FY 2022 compared to the relatively large increase from FY 2020 to FY 2021. The increased telehealth usage across VHA by TR-EWI and non—TR-EWI sites within selected VISNs may have contributed to limiting the increase in CC costs. Future studies should investigate contextual factors of increased telehealth usage, which would offer guidance for implementation to optimize the integration of telehealth with PT rehabilitation provided in-house. Additionally, future studies can examine potential limitations experienced during PT telehealth visits, such as the inability to conduct hands-on assessments, challenges in viewing the quality of patient movement, ensuring patient safety in the remote environment, and the lack of PT equipment in homes for telehealth visits, and how these challenges are being addressed.^38,39 Research is also needed to understand tradeoffs of CC vs VHA care and the potential and cost benefits of keeping veterans within VHA using programs like TR-EWI.⁵ Veterans living in rural communities may especially benefit from this as expanding telehealth options can provide access to PT care that may not be readily available, enabling them to stay connected and engaged in their care.^18,40

Future studies could examine contributory factors to rising costs, such as demographic shifts, changes in PT service utilization, and policy. Researchers might also consider qualitative studies with clinicians and veterans within each VISN, which may provide insights into how local factors impact PT referral to the community.

Limitations

Due to its descriptive nature, this study can only speculate about factors influencing trends. Limitations include the inability to link the Palantir and CC Dashboard datasets for cost comparisons and potential data change over time on Palantir due to platform updates. The focus on VISNs with TREWI sites limited generalizability and this study did not compare CC PT vs VHA PT. Finally, there may have been cost drivers not identified in this study.

Conclusions

This descriptive study provides insights into the utilization and cost of PT CC referrals for selected VISNs. Cost trends underscore the financial commitment to providing PT services to veterans. Understanding what factors are driving this cost is necessary for VHA to optimally provide and manage the rehabilitation resources needed to serve veterans through traditional in-person care, telehealth, and CC options while ensuring timely, highquality care.

The Veterans Health Administration (VHA) is the largest US integrated health system, providing care to veterans through VHA and non-VHA practitioners and facilities.^1,2 Providing high-quality, timely, and veteran-centric care remains a priority for the VHA. Legislative efforts have expanded opportunities for eligible veterans to receive care in the community purchased by VHA, known as community care (CC).¹ The Veterans Access, Choice, and Accountability Act of 2014 came in response to reports of long wait times and drive times for patients.^3-5 The MISSION Act of 2018 expanded access to CC by streamlining it and broadening eligibility criteria, especially for veterans in rural communities who often experience more barriers in accessing care than veterans living in urban communities.^1,6-10 Since the implementation of the Choice and MISSION Acts, > 2.7 million veterans have received care through community practitioners within the VHA CC network.¹¹

Background

Increased access to CC could benefit veterans living in rural communities by increasing care options and circumventing challenges to accessing VHA care (ie, geographic, transportation, and distance barriers, practitioner and specialist shortages, and hospital closures). ^5,9,10,12,13 However, health care system deficits in rural areas could also limit CC effectiveness for veterans living in those communities. ³ Other challenges posed by using CC include care coordination, information sharing, care continuity, delayed payments to CC practitioners, and mixed findings regarding CC quality.^5,8,13,14 VHA practitioners are specifically trained to meet the multifaceted needs unique to veterans’ health and subculture, training CC practitioners may not receive.^5,15

CC offers services for primary care and a broad range of specialties, including rehabilitation services such as physical therapy (PT).⁶ PT is used for the effective treatment of various conditions veterans experience and promote wellbeing and independence.¹⁶ US Department of Veterans Affairs (VA) databases reveal a high prevalence of veterans receiving PT services through CC; PT is one of the most frequently used CC outpatient specialty services by veterans living in rural communities.^14,17

Telerehabitltation Enterprisewide Initiative

VHA has greatly invested in delivering care virtually, especially for veterans living in rural communities.¹⁸ In 2017, the VHA Office of Rural Health funded the Telerehabilitation Enterprise-Wide Initiative (TR-EWI) in partnership with the Physical Medicine and Rehabilitation Services national program office to increase access to specialized rehabilitation services for veterans living in rural communities by leveraging telehealth technologies.^18-21 This alternative mode of health care delivery allows clinicians to overcome access barriers by delivering rehabilitation therapies directly to veterans' homes or nearby community-based outpatient clinics. TR-EWI was conceived as a hub-and-spoke model, where rehabilitation expertise at the hub was virtually delivered to spoke sites that did not have in-house expertise. In subsequent years, the TR-EWI also evolved to provide targeted telerehabilitation programs within rural-serving community-based outpatient clinics, including PT as a predominant service.^19,20

As TR-EWI progressed—and in conjunction with the uptake of telehealth across VHA during the COVID-19 pandemic—there has been increased focus on PT telerehabilitation, especially for the 4.6 million veterans in rural communities.^18,22,23 Because health care delivery system deficits in rural areas could limit the effective use of CC, many TR-EWI sites hope to reduce their CC referrals by providing telehealth PT services to veterans who might otherwise need to be referred to CC. This strategy aligns with VHA goals of providing high-quality and timely care. To better understand opportunities for programs like TR-EWI to provide rehabilitation services for veterans and reduce care sent to the community, research that examines CC referral trends for PT over time is warranted.

This study examines CC from a rehabilitation perspective with a focus on CC referral trends for PT, specifically for Veterans Integrated Service Networks (VISNs) where TREWI sites are located. The study’s objectives were to describe rehabilitation PT services being referred to CC and examine associated CC costs for PT services. Two research questions guided the study. First, what are the utilization trends for CC PT referrals from fiscal year (FY) 2019 to FY 2022? Secondly, what is the cost breakdown of CC for PT referrals from FY 2020 to FY 2022?

Methods

This study was conducted by a multidisciplinary team comprised of public health, disability, rehabilitation counseling, and PT professionals. It was deemed a quality improvement project under VA guidance and followed the SQUIRE guidelines for quality improvement reporting.^24,25 The study used the VA Common Operating Platform (Palantir) to obtain individual-level CC referral data from the HealthShare Referral Manager (HSRM) database and consult data from the Computerized Patient Record System. Palantir is used to store and integrate VA data derived from the VA Corporate Data Warehouse and VHA Support Service Center. Referrals are authorizations for care to be delivered by a CC practitioner.

TR-EWI is comprised of 7 sites: VISN 2, VISN 4, VISN 8, VISN 12, VISN 15, VISN 19, and VISN 22. Each site provides telerehabilitation services with an emphasis on reaching veterans living in rural communities. We joined the referrals and consults cubes in Palantir to extract PT referrals for FY 2019 to FY 2022 for the 7 VISNs with TR-EWI sites and obtain referral-specific information and demographic characteristics. ²⁶ Data were extracted in October 2022.

The VHA Community Care Referral Dashboard (CC Dashboard) provided nonindividual level CC cost data.²⁷ The CC Dashboard provides insights into the costs of CC services for VHA enrollees by category of care, standardized episode of care, and eligibility. Data are based on nationallevel HSRM referrals that are not suspended or linked to a canceled or discontinued consult. Data were aggregated by VISN. The dashboard only includes referrals dating back to FY 2020; therefore, PT data from FY 2020 through FY 2022 for VISNs with TR-EWI sites were collected. Data were extracted in December 2022.

This study examined CC referrals, station name, eligibility types, clinical diagnoses (International Classification of Diseases, Tenth Revision codes), and demographic information in the Palantir dataset. Six eligibility criteria can qualify a veteran to receive CC.²⁸ Within clinical diagnoses, the variable of interest was the provisional diagnosis. Patient demographics included age, gender, and rurality of residence, as determined by the Rural-Urban Commuting Area system.^29,30 Rural and highly rural categories were combined for analysis. For the CC cost dataset, this study examined CC referrals, referral cost, and eligibility type.

Analysis

For the first research question, we examined referral data from FY 2019 to FY 2022 using the Palantir dataset, performed descriptive statistical analysis for all variables, and analyzed data to identify trends. Descriptive statistics were completed using IBM SPSS Statistics for Windows Version 29.0.0.0.

A qualitative analysis of provisional diagnosis data revealed what is being referred to CC for PT. A preliminary overview of provisional diagnosis data was conducted to familiarize coders with the data. We developed a coding framework to categorize diagnoses based on anatomical location, body structure, and clinical areas of interest. Data were reviewed individually and grouped into categories within the coding framework before meeting as a team to achieve group consensus on categorization. We then totaled the frequency of occurrence for provisional diagnoses within each category. Qualitative analyses were completed using Microsoft Excel.

For the second research question, the study used the CC cost dataset to examine the cost breakdown of CC PT referrals from FY 2020 to FY 2022. We calculated the number and cost of PT referrals across eligibility groups for each FY and VISN. Data were analyzed using SPSS to identify cost trends.

Results

There were 344,406 referrals to CC for PT from FY 2019 to FY 2022 for the 7 VISNs analyzed (Table 1). Of these, 22.5% were from FY 2019, 19.1% from FY 2020, 28.2% from FY 2021, and 30.3% from FY 2022. VISN 8 and VISN 22 reported the most overall PT referrals, with VISN 8 comprising 22.2% and VISN 22 comprising 18.1% of all referrals. VISN 2 reported the least overall referrals (3.7%). VISN 4 and VISN 12 had decreases in referrals over time. VISN 2 and VISN 15 had decreases in referrals from FY 2019 to FY 2021 and slight increases from FY 2021 to FY 2022. VISN 19 and VISN 22 both saw slight increases from FY 2019 to FY 2020 and substantial increases from FY 2020 to FY 2022, with FY 2022 accounting for 40.0% and 42.3% of all referrals for VISN 19 and VISN 20, respectively (Figure 1).

For FY 2019 and FY 2020, VISN 8 had the highest percentage of referrals (26.7% and 23.2%, respectively), whereas VISN 22 was among the lowest (7.3% and 11.4%, respectively). However, for FY 2021 and FY 2022, VISN 22 reported the highest percentage of referrals (23.5% and 25.3%, respectively) compared to all other VISNs. VISN 2 consistently reported the lowest percentage of referrals across all years.

There were 56 stations analyzed across the 7 VISNs (Appendix 1). Nine stations each accounted for ≥ 3.0% of the total PT referrals and only 2 stations accounted for > 5.0% of referrals. Orlando, Florida (6.0%), Philadelphia, Pennsylvania (5.2%), Tampa, Florida (4.9%), Aurora, Colorado (4.9%), and Gainesville, Florida (4.4%) reported the top 5 highest referrals, with 3 being from VISN 8 (Orlando, Tampa, Gainesville). Stations with the lowest reported referrals were all in VISN 2 in New York: The Bronx, (0%), New York Harbor (0%), Hudson Valley (0.1%) and Finger Lakes (0.2%).

Rurality

Urban stations comprised 56.2% and rural stations comprised 39.8% of PT CC referrals, while 0.2% of referrals were from insular isle US territories: Guam, American Samoa, Northern Marianas, and the Virgin Islands. The sample had missing or unknown data for 3.8% of referrals. FY 2022 had the largest difference in rural and urban referrals. Additionally, there was an overall trend of more referrals over time for rural and urban, with a large increase in rural (+40.0%) and urban (+62.7%) referrals from FY 2020 to FY 2021 and a modest increase from FY 2021 to FY 2022 (+5.2% for rural and +9.1% for urban). There was a decrease in rural (-7.0%) and urban (-3.5%) referrals from FY 2019 to FY 2020 (Figure 2).

There were differences in referrals by rurality and VISN (Table 2). VISN 12, VISN 15, and VISN 19 reported more rural than urban referrals, whereas VISN 4, VISN 8, and VISN 22 reported more urban than rural referrals. VISN 2 reported similar numbers for both, with slightly more urban than rural referrals. When reviewing trends over time for each FY, VISN 12, VISN 15, and VISN 19 reported more rural than urban referrals and VISN 4, VISN 8, and VISN 22 had more urban than rural referrals. In FY 2019 and FY 2020, VISN 2 reported slightly more urban than rural referrals but almost the same number of referrals in FY 2021 and FY 2022 (Appendix 2).

Demographics

The mean (SD) age was 61.2 (15.8) years (range, 20-105). Most PT CC referrals were for veterans aged 70 to 79 years (26.9%), followed by 60 to 69 years (20.7%), and 50 to 59 years (16.4%) (Appendix 3). Trends were consistent across VISNs. There was less of a difference between rural and urban referral percentages as the population aged. Veterans aged < 49 years residing in more urban areas accounted for more referrals to CC compared to their rural counterparts. This difference was less apparent in the 70 to 79 years and 80 to 89 years age brackets.

Most PT CC referrals (81.2%) were male and 14.8% were female. About 3.6% of referral data were missing sex information, and there was a smaller difference between male veterans living in rural communities and male veterans living in urban communities compared with female veterans. A total of 42.9% of male veterans resided in rural areas compared to 56.8% in urban areas; 32.7% of female veterans resided in rural areas compared to 66.9% in urban areas (Appendix 3).

Other Criteria

Of the 334,406 referrals, 114,983 (34.4%) had eligibility data, mostly from FY 2021 and FY 2022 (Table 3). Available eligibility data were likely affected by the MISSION Act and new regulations for reporting CC eligibility. Distance (33.4%) was the most common eligibility criteria, followed by timeliness of care (28.8%), and best medical interest (19.8%); 40.4% were rural and 59.5% were urban. Distance (55.4%) was most common for rural veterans, while timeliness of care (39.7%) was most common for urban veterans. For both groups, the second most common eligibility reason was best medical interest (Appendix 4).

Bone, joint, or soft tissue disorders were common diagnoses, with 25.2% located in the lower back, 14.7% in the shoulder, and 12.8% in the knee (Appendix 5). Amputations of the upper and lower limbs, fractures, cancer-related diagnoses, integumentary system disorders, thoracic and abdominal injuries and disorders, and other medical and mental health conditions each accounted for < 1% of the total diagnoses.

Costs

At time of analysis, the CC Dashboard had cost data available for 200,204 CC PT referrals from FY 2020 to FY 2022. The difference in referral numbers for the 2 datasets is likely attributed to several factors: CC cost data is exclusively from the HSRM, whereas Palantir includes other data sources; how VA cleans data pulled into Palantir; how the CC Dashboard algorithm populates data; and variances based on timing of reporting and/or if referrals are eventually canceled.

The total cost of PT CC referrals from FY 2020 to FY 2022 in selected VISNs was about $220,615,399 (Appendix 6). Appendix 7 details the methodology for determining the average standardized episode- of-care cost by VISN and how referral costs are calculated. Data show a continuous increase in total estimated cost from $46.8 million in FY 2020 to $92.1 million in FY 2022. From FY 2020 to FY 2022, aggregate costs ranged from $6,758,053 in VISN 2 to $47,209,162 in VISN 8 (Figure 3). The total referral cost for PT was highest at VISN 4 in FY 2020 ($10,447,140) and highest at VISN 22 in FY 2021 ($18,835,657) and FY 2022 ($22,962,438) (Figure 4). For referral costs from FY 2020 to FY 2022, distance accounted for $75,561,948 (34.3%), timeliness of care accounted for $60,413,496 (27.3%), and best medical interest accounted for $46,291,390 (21.0%) (Table 4).

Overall costs were primarily driven by specific VISNs within each eligibility type (Appendix 8; Figure 5). VISN 19, VISN 22, and VISN 15 accounted for the highest referral costs for distance; VISN 22, VISN 8, and VISN 19 accounted for the secondhighest referral cost, timeliness of care; and VISN 4, VISN 8, and VISN 12 accounted for the third-highest referral cost, best medical interest (Figure 5). VISN 2, VISN 4, VISN 12, VISN 15, and VISN 22 had service unavailable as an eligibility type with 1 of the top 3 associated referral costs, which was higher in cost than timeliness of care for VISN 2, VISN 4, VISN 12, and VISN 15.

Discussion

This study examines the referral of rehabilitation PT services to CC, evaluates CC costs for PT services, and analyzes utilization and cost trends among veterans within the VHA. Utilization data demonstrated a decrease in referrals from FY 2019 to FY 2020 and increases in referrals from FY 2020 to FY 2022 for most variables of interest, with cost data exhibiting similar trends. Results highlight the need for further investigation to address variations in PT referrals and costs across VISNs and eligibility reasons for CC referral.

Results demonstrated a noteworthy increase in PT CC referrals over time. The largest increase occurred from FY 2020 to FY 2021, with a smaller increase from FY 2021 to FY 2022. During this period, total enrollee numbers decreased by 3.0% across the 7 VISNs included in this analysis and by 1.6% across all VISNs, a trend that illustrates an overall decrease in enrollees as CC use increased. Results align with the implementation of the MISSION Act of 2018, which further expanded veterans’ options to use CC.^1,6,7 Results also align with the onset of the COVID-19 pandemic, which disrupted care access for many veterans, placed a larger emphasis on the use of telehealth, and increased opportunities to stay within the VA for care by rapidly shifting to telehealth and leveraging telerehabilitation investments and initiatives (such as TR-EWI).^20,31

VISN 8, VISN 19, and VISN 22, accounted for more than half of PT referrals. These VISNs had higher enrollee counts compared to the other VISNs.³² VISN 8 consistently had high levels of referrals, whereas VISN 19 and VISN 22 saw dramatic increases in FY 2021 and FY 2022. In contrast, VISN 4 and VISN 12 gradually decreased referrals during the study. VISN 2 had the lowest referral numbers during the study period, and all stations with the lowest individual referral numbers were located within VISN 2. Of the VISNs included in this study, VISN 2 had the second lowest number of enrollees (324,042).³² Reasons for increases and decreases over time could not be determined based on data collected in this study.

There were more urban than rural PT CC referrals; however, both exhibited an increase in referrals over time. This is consistent with population trends showing that most VHA patients (62.6%) and veterans (75.9%) reside in urban areas, which could explain some of the trends in this study.³³ Some VISNs have larger urban catchment areas (eg, VISN 8 and VISN 22), and some have larger rural catchment areas (eg, VISN 15 and VISN 19), which could partially explain the rural-urban differences by VISN.³² Rural-urban referral trends might also reflect existing health care delivery system deficits in rural areas and known challenges associated with accessing health care for veterans living in rural communities.^8,9

This study found larger differences in rural and urban PT CC referrals for younger age groups, with more than twice as many urban referrals in veterans aged 20 to 29 years and aged 30 to 39 years, and roughly 1.8 times as many urban referrals in veterans aged 40 to 49 years. However, there were similar numbers of rural and urban referrals in those aged 70 to 79 years and aged 80 to 89 years. These trends are consistent with data showing veterans residing in rural communities are older than their urban counterparts.^23,34 Data suggest that older veteran populations might seek PT at higher rates than younger veteran populations. Moreover, data suggest there could be differences in PT-seeking rates for younger veteran populations who reside in rural vs urban areas. Additional research is needed to understand these trends.

Distance and timeliness of care were the predominant reasons for referral among eligibility groups, which is consistent with the MISSION Act goals.^1,6,7 The most common eligibility reason for rural referrals was distance; timeliness of care was most common for urban referrals. This finding is expected, as veterans living in rural communities are farther away from VHA facilities and have longer drive times, whereas veterans living in urban communities might live closer, yet experience longer wait times due to services and/or appointment availability. Best medical interest accounted for almost 20% of referrals, which does not provide detailed insights into why those veterans were referred to CC.

The top PT diagnoses referred to CC were related to bone, joint, or soft tissue disorders of the lower back, shoulder, and knee. This suggests that musculoskeletal-related issues are prevalent among veterans seeking PT care, which is consistent with research that found > 50% of veterans receiving VHA care have musculoskeletal disorders.³⁵ The probability of experiencing musculoskeletal problems increases with age, as does the need for PT services. Amputations and fractures accounted for < 1% of CC referrals, which is consistent with the historic provision of VHA clinical specialized care to conditions prevalent among veterans. It may also represent VHA efforts to internally provide care for complex conditions requiring more extensive interdisciplinary coordination.

The total cost of referrals over time was about $221 million. VISN 8 accounted for the highest overall cost; VISN 2 had the lowest, mirroring referral utilization trends and aligning with VISN enrollee numbers. VISN 19 and VISN 22 reported large cost increases from FY 2020 to FY 2021. Total referral costs increased by $34.9 million from FY 2020 to FY 2021, which may be due to health care inflation (2.9% during FY 2019 to FY 2022), increased awareness of CC services, or increased VHA wait times.³⁶ Additionally, there were limitations in care provided across health care systems during the COVID-19 pandemic, including the VA.⁵ The increase from FY 2020 to FY 2021 may reflect a rebound from restrictions in appointments across VA, CC, and the private sector.

While the increase in total referral cost may be partly attributed to inflation, the cost effectiveness and efficiency of referring veterans to CC vs keeping veterans within VHA care is an ongoing debate.⁵ Examining and addressing cost drivers within the top eligibility types and their respective VISNs is necessary to determine resource allocation and improve quality of care. This study found that best medical interest and unavailable services accounted for 33.4% of the total cost of CC referrals, highlighting the need for policies that strengthen in-house competencies and recruit personnel to provide PT services currently unavailable within the VA.

Future Directions

The VHA should explore opportunities for in-house care, especially for services appropriate for telehealth.^18,20,37 Data indicated a smaller cost increase from FY 2021 to FY 2022 compared to the relatively large increase from FY 2020 to FY 2021. The increased telehealth usage across VHA by TR-EWI and non—TR-EWI sites within selected VISNs may have contributed to limiting the increase in CC costs. Future studies should investigate contextual factors of increased telehealth usage, which would offer guidance for implementation to optimize the integration of telehealth with PT rehabilitation provided in-house. Additionally, future studies can examine potential limitations experienced during PT telehealth visits, such as the inability to conduct hands-on assessments, challenges in viewing the quality of patient movement, ensuring patient safety in the remote environment, and the lack of PT equipment in homes for telehealth visits, and how these challenges are being addressed.^38,39 Research is also needed to understand tradeoffs of CC vs VHA care and the potential and cost benefits of keeping veterans within VHA using programs like TR-EWI.⁵ Veterans living in rural communities may especially benefit from this as expanding telehealth options can provide access to PT care that may not be readily available, enabling them to stay connected and engaged in their care.^18,40

Future studies could examine contributory factors to rising costs, such as demographic shifts, changes in PT service utilization, and policy. Researchers might also consider qualitative studies with clinicians and veterans within each VISN, which may provide insights into how local factors impact PT referral to the community.

Limitations

Due to its descriptive nature, this study can only speculate about factors influencing trends. Limitations include the inability to link the Palantir and CC Dashboard datasets for cost comparisons and potential data change over time on Palantir due to platform updates. The focus on VISNs with TREWI sites limited generalizability and this study did not compare CC PT vs VHA PT. Finally, there may have been cost drivers not identified in this study.

Conclusions

This descriptive study provides insights into the utilization and cost of PT CC referrals for selected VISNs. Cost trends underscore the financial commitment to providing PT services to veterans. Understanding what factors are driving this cost is necessary for VHA to optimally provide and manage the rehabilitation resources needed to serve veterans through traditional in-person care, telehealth, and CC options while ensuring timely, highquality care.

The US Department of Veterans Affairs (VA) has been integral in resident training. Resident surgical training requires a balance of supervision and autonomy, along with procedure repetition and appropriate feedback.^1-3 Non-VA research has found that resident participation across various otolaryngology procedures, including thyroidectomy, neck dissection, and laryngectomy, does not increase patient morbidity.^4-7 However, resident involvement in private and academic settings that included nonhead and neck procedures was linked to increased operative time and reduced productivity, as determined by work relative value units (wRVUs).^7-13 This has also been identified in other specialties, including general surgery, orthopedics, and ophthalmology.^14-16

Unlike the private sector, surgeon compensation at the VA is not as closely linked to operative productivity, offering a unique setting for resident training. While VA integration in otolaryngology residency programs increases resident case numbers, particularly in head and neck cases, the impact on VA patient outcomes and productivity is unknown.¹⁷ The use of larynxpreserving treatment modalities for laryngeal cancer has led to a decline in the number of total laryngectomies performed, which could potentially impact resident operative training for laryngectomies.^18-20

This study sought to determine the impact of resident participation on operative time, wRVUs, and patient outcomes in veterans who underwent a total laryngectomy. This study was reviewed and approved by the MedStar Georgetown University Hospital Institutional Review Board and Research and Development Committee (#1595672).

Methods

A retrospective cohort of veterans nationwide who underwent total laryngectomy between 2001 and 2021, with or without neck dissection, was identified from the Veterans Affairs Surgical Quality Improvement Program (VASQIP). Data were extracted via the VA Informatics and Computing Infrastructure and patients were included based on Current Procedural Terminology codes for total laryngectomy, with or without neck dissection (31320, 31360, 31365). Laryngopharyngectomies, partial laryngectomies, and minimally invasive laryngectomies were excluded. VASQIP nurse data managers reviewed patient data for operative data, postoperative outcomes (including 30- day morbidity and mortality), and preoperative risk factors (Appendix).²¹

The VASQIP data provide the highest resident or postgraduate year (PGY) per surgery. PGY 1, 2, and 3 were considered junior residents and PGY ≥4, surgical fellows, and individuals who took research years during residency were considered senior residents. Cases performed by attending physicians alone were compared with those involving junior or senior residents.

Patient demographic data included age, body mass index, smoking and alcohol use, weight loss, and functional status. Consumption of any tobacco products within 12 months of surgery was considered tobacco use. Drinking on average ≥2 alcoholic beverages daily was considered alcohol use. Weight loss was defined as a 10% reduction in body weight within the 6 months before surgery, excluding patients enrolled in a weight loss program. Functional status was categorized as independent, partially dependent, totally dependent, and unknown.

Primary outcomes included operative time, wRVUs generated, and wRVUs generated per hour of operative time. Postoperative complications were recorded both as a continuous variable and as a binary variable for presence or absence of a complication. Additional outcome variables included length of postoperative hospital stay, return to the operating room (OR), and death within 30 days of surgery.

Statistical Analysis

Data were summarized using frequency and percentage for categorical variables and median with IQR for continuous variables. Data were also summarized based on resident involvement in the surgery and the PGY level of the residents involved. The occurrence of total laryngectomy, rate of complications, and patient return to the OR were summarized by year.

Univariate associations between resident involvement and surgical outcomes were analyzed using the Kruskal-Wallis test for continuous variables and the ÷2 test for categorical variables. A Fisher exact test was used when the cell count in the contingency table was < 5. The univariate associations between surgical outcomes and demographic/preoperative variables were examined using 2-sided Wilcoxon ranksum tests or Kruskal-Wallis tests between continuous variables and categorical variables, X² or Fisher exact test between 2 categorical variables, and 2-sided Spearman correlation test between 2 continuous variables. A false-discovery rate approach was used for simultaneous posthoc tests to determine the adjusted P values for wRVUs generated/operative time for attending physicians alone vs with junior residents and for attending physicians alone vs with senior residents. Models were used to evaluate the effects of resident involvement on surgical outcomes, adjusting for variables that showed significant univariate associations. Linear regression models were used for operative time, wRVUs generated, wRVUs generated/operative time, and length of postoperative stay. A logistic regression model was used for death within 30 days. Models were not built for postoperative complications or patient return to the OR, as these were only statistically significantly associated with the patient’s preoperative functional status. A finding was considered significant if P < .05. All analyses were performed using statistical software RStudio Version 2023.03.0.

Results

Between 2001 and 2021, 1857 patients who underwent total laryngectomy were identified from the VASQIP database nationwide. Most of the total laryngectomies were staffed by an attending physician with a senior resident (n = 1190, 64%), 446 (24%) were conducted by the attending physician alone, and 221 (12%) by an attending physician with a junior resident (Table 1). The mean operating time for an attending physician alone was 378 minutes, 384 minutes for an attending physician with a senior resident, and 432 minutes for an attending physician with a junior resident (Table 2). There was a statistically significant increase in operating time for laryngectomies with resident participation compared to attending physicians operating alone (P < .001).

When the wRVUs generated/operative time was analyzed, there was a statistically significant difference between comparison groups. Total laryngectomies performed by attending physicians alone had the highest wRVUs generated/operative time (5.5), followed by laryngectomies performed by attending physicians with senior residents and laryngectomies performed by attending physicians with junior residents (5.2 and 4.8, respectively; P = .002). Table 3 describes adjusted P values for wRVUs generated/ operative time for total laryngectomies performed by attending physicians alone vs with junior residents (P = .003) and for attending physicians alone vs with senior residents (P = .02). Resident participation in total laryngectomies did not significantly impact the development or number of postoperative complications or the rate of return to the OR.

The number of laryngectomies performed in a single fiscal year peaked in 2010 at 170 cases (Figure 1). Between 2001 and 2021, the mean rates of postoperative complications (21.3%) and patient return to the OR (14.6%) did not significantly change. Resident participation in total laryngectomies also peaked in 2010 at 89.0% but has significantly declined, falling to a low of 43.6% in 2021 (Figure 2). From 2001 to 2011, the mean resident participation rate in total laryngectomies was 80.6%, compared with 68.3% from 2012 to 2021 (P < .001).

The effect of various demographic and preoperative characteristics on surgical outcomes was also analyzed. A linear regression model accounted for each variable significantly associated with operative time. On multivariable analysis, when all other variables were held constant, Table 4 shows the estimated change in operative time based on certain criteria. For instance, the operative time for attendings with junior residents surgeries was 40 minutes longer (95% CI, 16 to 64) than that of attending alone surgeries (P = .001). Furthermore, operative time decreased by 1.1 minutes (95% CI, 0.30 to 2.04) for each 1-year increase in patient age (P = .009).

A multivariable logistic regression model evaluated the effect of resident involvement on 30-day mortality rates. Senior resident involvement (P = .02), partially dependent functional status (P = .01), totally dependent functional status (P < .001), and advanced age (P = .02) all were significantly associated with 30-day mortality (Table 5). When other variables remained constant, the odds of death for totally dependent patients were 10.4 times higher than that of patients with independent functional status. Thus, totally dependent functional status appeared to have a greater impact on this outcome than resident participation. The linear regression model for postoperative length of stay demonstrated that senior resident involvement (P = .04), functional status (partially dependent vs independent P < .001), and age (P = .03) were significantly associated with prolonged length of stay.

Discussion

Otolaryngology residency training is designed to educate future otolaryngologists through hands-on learning, adequate feedback, and supervision.¹ Although this exposure is paramount for resident education, balancing appropriate supervision and autonomy while mitigating patient risk has been difficult. Numerous non-VA studies have reviewed the impact of resident participation on patient outcomes in various specialties, ranging from a single institution to the National Surgical Quality Improvement Program (NSQIP).^4,5,7,22 This study is the first to describe the nationwide impact of resident participation on outcomes in veterans undergoing total laryngectomy.

This study found that resident participation increases operative time and decreases wRVUs generated/operative time without impacting complication rates or patient return to the OR. This reinforces the notion that under close supervision, resident participation does not negatively impact patient outcomes. Resident operative training requires time and dedication by the attending physician and surgical team, thereby increasing operative time. Because VA physician compensation is not linked with productivity as closely as it is in other private and academic settings, surgeons can dedicate more time to operative teaching. This study found that a total laryngectomy involving a junior resident took about 45 minutes longer than an attending physician working alone.

As expected, with longer operative times, the wRVUs generated/operative time ratio was lower in cases with resident participation. Even though resident participation leads to lower OR efficiency, their participation may not significantly impact ancillary costs.²³ However, a recent study from NSQIP found an opportunity cost of $60.44 per hour for surgeons operating with a resident in head and neck cases.¹³

Postoperative complications and mortality are key measures of surgical outcomes in addition to operative time and efficiency. This study found that neither junior nor senior resident participation significantly increased complication rates or patient return to the OR. Despite declining resident involvement and the number of total laryngectomy surgeries in the VA, the complication rate has remained steady. The 30-day mortality rate was significantly higher in cases involving senior residents compared to cases with attending physicians alone. This could be a result of senior resident participation in more challenging cases, such as laryngectomies performed as salvage surgery following radiation. Residents are more often involved in cases with greater complexity at teaching institutions.^24-26 Therefore, the higher mortality seen among laryngectomies with senior resident involvement is likely due to the higher complexity of those cases.

The proportion of resident involvement in laryngectomies at VA medical centers has been decreasing over time. Due to the single payer nature of the VA health care system and the number of complex and comorbid patients, the VA offers an invaluable space for resident education in the OR. The fact that less than half of laryngectomies in 2021 involved resident participation is noteworthy for residency training programs. As wRVU compensation models evolve, VA attending surgeons may face less pressure to move the case along, leading to a high potential for operative teaching. Therefore, complex cases, such as laryngectomies, are often ideal for resident participation in the VA.

The steady decline in total laryngectomies at the VA parallels the recent decrease seen in non-VA settings.²⁰ This is due in part to the use of larynx-preserving treatment modalities for laryngeal cancer as well as decreases in the incidence of laryngeal cancer due to population level changes in smoking behaviors. ^18,19 Although a laryngectomy is not a key indicator case as determined by the Accreditation Council for Graduate Medical Education, it is important for otolaryngology residents to be exposed to these cases and have a thorough understanding of the operative technique.²⁷ Total laryngectomy was selected for this study because it is a complex and time-consuming surgery with somewhat standardized surgical steps. Unlike microvascular surgery that is very rarely performed by an attending physician alone, laryngectomies can be performed by attending physicians alone or with a resident.²⁸

Limitations

Since this was a retrospective study, it was susceptible to errors in data entry and data extraction from the VASQIP database. Another limitation is the lack of preoperative treatment data on tumor stage and prior nonoperative treatment. For example, a salvage laryngectomy after treatment with radiation and/or chemoradiation is a higher risk procedure than an upfront laryngectomy. Senior resident involvement may be more common in patients undergoing salvage laryngectomy due to the high risk of postoperative fistula and other complications. This may have contributed to the association identified between senior resident participation and 30-day mortality.

Since we could not account for residents who took research years or were fellows, a senior resident may have been mislabeled as a junior resident or vice versa. However, because most research years occur following the third year of residency. We are confident that PGY-1, PGY-2, and PGY-3 is likely to capture junior residents. Other factors, such as coattending surgeon cases, medical student assistance, and fellow involvement may have also impacted the results of this study.

Conclusions

This study is the first to investigate the impact of resident participation on operative time, wRVUs generated, and complication rates in head and neck surgery at VA medical centers. It found that resident participation in total laryngectomies among veterans increased operative time and reduced wRVUs generated per hour but did not impact complication rate or patient return to the OR. The VA offers a unique and invaluable space for resident education and operative training, and the recent decline in resident participation among laryngectomies is important for residency programs to acknowledge and potentially address moving forward.

In contrast to oral cavity resections which can vary from partial glossectomies to composite resections, laryngectomy represents a homogenous procedure from which to draw meaningful conclusions about complication rates, operative time, and outcome. Future directions should include studying other types of head and neck surgery in the VA to determine whether the impact of resident participation mirrors the findings of this study.

The US Department of Veterans Affairs (VA) has been integral in resident training. Resident surgical training requires a balance of supervision and autonomy, along with procedure repetition and appropriate feedback.^1-3 Non-VA research has found that resident participation across various otolaryngology procedures, including thyroidectomy, neck dissection, and laryngectomy, does not increase patient morbidity.^4-7 However, resident involvement in private and academic settings that included nonhead and neck procedures was linked to increased operative time and reduced productivity, as determined by work relative value units (wRVUs).^7-13 This has also been identified in other specialties, including general surgery, orthopedics, and ophthalmology.^14-16

Unlike the private sector, surgeon compensation at the VA is not as closely linked to operative productivity, offering a unique setting for resident training. While VA integration in otolaryngology residency programs increases resident case numbers, particularly in head and neck cases, the impact on VA patient outcomes and productivity is unknown.¹⁷ The use of larynxpreserving treatment modalities for laryngeal cancer has led to a decline in the number of total laryngectomies performed, which could potentially impact resident operative training for laryngectomies.^18-20

This study sought to determine the impact of resident participation on operative time, wRVUs, and patient outcomes in veterans who underwent a total laryngectomy. This study was reviewed and approved by the MedStar Georgetown University Hospital Institutional Review Board and Research and Development Committee (#1595672).

Methods

A retrospective cohort of veterans nationwide who underwent total laryngectomy between 2001 and 2021, with or without neck dissection, was identified from the Veterans Affairs Surgical Quality Improvement Program (VASQIP). Data were extracted via the VA Informatics and Computing Infrastructure and patients were included based on Current Procedural Terminology codes for total laryngectomy, with or without neck dissection (31320, 31360, 31365). Laryngopharyngectomies, partial laryngectomies, and minimally invasive laryngectomies were excluded. VASQIP nurse data managers reviewed patient data for operative data, postoperative outcomes (including 30- day morbidity and mortality), and preoperative risk factors (Appendix).²¹

The VASQIP data provide the highest resident or postgraduate year (PGY) per surgery. PGY 1, 2, and 3 were considered junior residents and PGY ≥4, surgical fellows, and individuals who took research years during residency were considered senior residents. Cases performed by attending physicians alone were compared with those involving junior or senior residents.

Patient demographic data included age, body mass index, smoking and alcohol use, weight loss, and functional status. Consumption of any tobacco products within 12 months of surgery was considered tobacco use. Drinking on average ≥2 alcoholic beverages daily was considered alcohol use. Weight loss was defined as a 10% reduction in body weight within the 6 months before surgery, excluding patients enrolled in a weight loss program. Functional status was categorized as independent, partially dependent, totally dependent, and unknown.

Primary outcomes included operative time, wRVUs generated, and wRVUs generated per hour of operative time. Postoperative complications were recorded both as a continuous variable and as a binary variable for presence or absence of a complication. Additional outcome variables included length of postoperative hospital stay, return to the operating room (OR), and death within 30 days of surgery.

Statistical Analysis

Data were summarized using frequency and percentage for categorical variables and median with IQR for continuous variables. Data were also summarized based on resident involvement in the surgery and the PGY level of the residents involved. The occurrence of total laryngectomy, rate of complications, and patient return to the OR were summarized by year.

Univariate associations between resident involvement and surgical outcomes were analyzed using the Kruskal-Wallis test for continuous variables and the ÷2 test for categorical variables. A Fisher exact test was used when the cell count in the contingency table was < 5. The univariate associations between surgical outcomes and demographic/preoperative variables were examined using 2-sided Wilcoxon ranksum tests or Kruskal-Wallis tests between continuous variables and categorical variables, X² or Fisher exact test between 2 categorical variables, and 2-sided Spearman correlation test between 2 continuous variables. A false-discovery rate approach was used for simultaneous posthoc tests to determine the adjusted P values for wRVUs generated/operative time for attending physicians alone vs with junior residents and for attending physicians alone vs with senior residents. Models were used to evaluate the effects of resident involvement on surgical outcomes, adjusting for variables that showed significant univariate associations. Linear regression models were used for operative time, wRVUs generated, wRVUs generated/operative time, and length of postoperative stay. A logistic regression model was used for death within 30 days. Models were not built for postoperative complications or patient return to the OR, as these were only statistically significantly associated with the patient’s preoperative functional status. A finding was considered significant if P < .05. All analyses were performed using statistical software RStudio Version 2023.03.0.

Results

Between 2001 and 2021, 1857 patients who underwent total laryngectomy were identified from the VASQIP database nationwide. Most of the total laryngectomies were staffed by an attending physician with a senior resident (n = 1190, 64%), 446 (24%) were conducted by the attending physician alone, and 221 (12%) by an attending physician with a junior resident (Table 1). The mean operating time for an attending physician alone was 378 minutes, 384 minutes for an attending physician with a senior resident, and 432 minutes for an attending physician with a junior resident (Table 2). There was a statistically significant increase in operating time for laryngectomies with resident participation compared to attending physicians operating alone (P < .001).

When the wRVUs generated/operative time was analyzed, there was a statistically significant difference between comparison groups. Total laryngectomies performed by attending physicians alone had the highest wRVUs generated/operative time (5.5), followed by laryngectomies performed by attending physicians with senior residents and laryngectomies performed by attending physicians with junior residents (5.2 and 4.8, respectively; P = .002). Table 3 describes adjusted P values for wRVUs generated/ operative time for total laryngectomies performed by attending physicians alone vs with junior residents (P = .003) and for attending physicians alone vs with senior residents (P = .02). Resident participation in total laryngectomies did not significantly impact the development or number of postoperative complications or the rate of return to the OR.

The number of laryngectomies performed in a single fiscal year peaked in 2010 at 170 cases (Figure 1). Between 2001 and 2021, the mean rates of postoperative complications (21.3%) and patient return to the OR (14.6%) did not significantly change. Resident participation in total laryngectomies also peaked in 2010 at 89.0% but has significantly declined, falling to a low of 43.6% in 2021 (Figure 2). From 2001 to 2011, the mean resident participation rate in total laryngectomies was 80.6%, compared with 68.3% from 2012 to 2021 (P < .001).

The effect of various demographic and preoperative characteristics on surgical outcomes was also analyzed. A linear regression model accounted for each variable significantly associated with operative time. On multivariable analysis, when all other variables were held constant, Table 4 shows the estimated change in operative time based on certain criteria. For instance, the operative time for attendings with junior residents surgeries was 40 minutes longer (95% CI, 16 to 64) than that of attending alone surgeries (P = .001). Furthermore, operative time decreased by 1.1 minutes (95% CI, 0.30 to 2.04) for each 1-year increase in patient age (P = .009).

A multivariable logistic regression model evaluated the effect of resident involvement on 30-day mortality rates. Senior resident involvement (P = .02), partially dependent functional status (P = .01), totally dependent functional status (P < .001), and advanced age (P = .02) all were significantly associated with 30-day mortality (Table 5). When other variables remained constant, the odds of death for totally dependent patients were 10.4 times higher than that of patients with independent functional status. Thus, totally dependent functional status appeared to have a greater impact on this outcome than resident participation. The linear regression model for postoperative length of stay demonstrated that senior resident involvement (P = .04), functional status (partially dependent vs independent P < .001), and age (P = .03) were significantly associated with prolonged length of stay.

Discussion

Otolaryngology residency training is designed to educate future otolaryngologists through hands-on learning, adequate feedback, and supervision.¹ Although this exposure is paramount for resident education, balancing appropriate supervision and autonomy while mitigating patient risk has been difficult. Numerous non-VA studies have reviewed the impact of resident participation on patient outcomes in various specialties, ranging from a single institution to the National Surgical Quality Improvement Program (NSQIP).^4,5,7,22 This study is the first to describe the nationwide impact of resident participation on outcomes in veterans undergoing total laryngectomy.

This study found that resident participation increases operative time and decreases wRVUs generated/operative time without impacting complication rates or patient return to the OR. This reinforces the notion that under close supervision, resident participation does not negatively impact patient outcomes. Resident operative training requires time and dedication by the attending physician and surgical team, thereby increasing operative time. Because VA physician compensation is not linked with productivity as closely as it is in other private and academic settings, surgeons can dedicate more time to operative teaching. This study found that a total laryngectomy involving a junior resident took about 45 minutes longer than an attending physician working alone.

As expected, with longer operative times, the wRVUs generated/operative time ratio was lower in cases with resident participation. Even though resident participation leads to lower OR efficiency, their participation may not significantly impact ancillary costs.²³ However, a recent study from NSQIP found an opportunity cost of $60.44 per hour for surgeons operating with a resident in head and neck cases.¹³

Postoperative complications and mortality are key measures of surgical outcomes in addition to operative time and efficiency. This study found that neither junior nor senior resident participation significantly increased complication rates or patient return to the OR. Despite declining resident involvement and the number of total laryngectomy surgeries in the VA, the complication rate has remained steady. The 30-day mortality rate was significantly higher in cases involving senior residents compared to cases with attending physicians alone. This could be a result of senior resident participation in more challenging cases, such as laryngectomies performed as salvage surgery following radiation. Residents are more often involved in cases with greater complexity at teaching institutions.^24-26 Therefore, the higher mortality seen among laryngectomies with senior resident involvement is likely due to the higher complexity of those cases.

The proportion of resident involvement in laryngectomies at VA medical centers has been decreasing over time. Due to the single payer nature of the VA health care system and the number of complex and comorbid patients, the VA offers an invaluable space for resident education in the OR. The fact that less than half of laryngectomies in 2021 involved resident participation is noteworthy for residency training programs. As wRVU compensation models evolve, VA attending surgeons may face less pressure to move the case along, leading to a high potential for operative teaching. Therefore, complex cases, such as laryngectomies, are often ideal for resident participation in the VA.

The steady decline in total laryngectomies at the VA parallels the recent decrease seen in non-VA settings.²⁰ This is due in part to the use of larynx-preserving treatment modalities for laryngeal cancer as well as decreases in the incidence of laryngeal cancer due to population level changes in smoking behaviors. ^18,19 Although a laryngectomy is not a key indicator case as determined by the Accreditation Council for Graduate Medical Education, it is important for otolaryngology residents to be exposed to these cases and have a thorough understanding of the operative technique.²⁷ Total laryngectomy was selected for this study because it is a complex and time-consuming surgery with somewhat standardized surgical steps. Unlike microvascular surgery that is very rarely performed by an attending physician alone, laryngectomies can be performed by attending physicians alone or with a resident.²⁸

Limitations

Since this was a retrospective study, it was susceptible to errors in data entry and data extraction from the VASQIP database. Another limitation is the lack of preoperative treatment data on tumor stage and prior nonoperative treatment. For example, a salvage laryngectomy after treatment with radiation and/or chemoradiation is a higher risk procedure than an upfront laryngectomy. Senior resident involvement may be more common in patients undergoing salvage laryngectomy due to the high risk of postoperative fistula and other complications. This may have contributed to the association identified between senior resident participation and 30-day mortality.

Since we could not account for residents who took research years or were fellows, a senior resident may have been mislabeled as a junior resident or vice versa. However, because most research years occur following the third year of residency. We are confident that PGY-1, PGY-2, and PGY-3 is likely to capture junior residents. Other factors, such as coattending surgeon cases, medical student assistance, and fellow involvement may have also impacted the results of this study.

Conclusions

This study is the first to investigate the impact of resident participation on operative time, wRVUs generated, and complication rates in head and neck surgery at VA medical centers. It found that resident participation in total laryngectomies among veterans increased operative time and reduced wRVUs generated per hour but did not impact complication rate or patient return to the OR. The VA offers a unique and invaluable space for resident education and operative training, and the recent decline in resident participation among laryngectomies is important for residency programs to acknowledge and potentially address moving forward.

In contrast to oral cavity resections which can vary from partial glossectomies to composite resections, laryngectomy represents a homogenous procedure from which to draw meaningful conclusions about complication rates, operative time, and outcome. Future directions should include studying other types of head and neck surgery in the VA to determine whether the impact of resident participation mirrors the findings of this study.

The US Department of Veterans Affairs (VA) has been integral in resident training. Resident surgical training requires a balance of supervision and autonomy, along with procedure repetition and appropriate feedback.^1-3 Non-VA research has found that resident participation across various otolaryngology procedures, including thyroidectomy, neck dissection, and laryngectomy, does not increase patient morbidity.^4-7 However, resident involvement in private and academic settings that included nonhead and neck procedures was linked to increased operative time and reduced productivity, as determined by work relative value units (wRVUs).^7-13 This has also been identified in other specialties, including general surgery, orthopedics, and ophthalmology.^14-16

Unlike the private sector, surgeon compensation at the VA is not as closely linked to operative productivity, offering a unique setting for resident training. While VA integration in otolaryngology residency programs increases resident case numbers, particularly in head and neck cases, the impact on VA patient outcomes and productivity is unknown.¹⁷ The use of larynxpreserving treatment modalities for laryngeal cancer has led to a decline in the number of total laryngectomies performed, which could potentially impact resident operative training for laryngectomies.^18-20

This study sought to determine the impact of resident participation on operative time, wRVUs, and patient outcomes in veterans who underwent a total laryngectomy. This study was reviewed and approved by the MedStar Georgetown University Hospital Institutional Review Board and Research and Development Committee (#1595672).

Methods

A retrospective cohort of veterans nationwide who underwent total laryngectomy between 2001 and 2021, with or without neck dissection, was identified from the Veterans Affairs Surgical Quality Improvement Program (VASQIP). Data were extracted via the VA Informatics and Computing Infrastructure and patients were included based on Current Procedural Terminology codes for total laryngectomy, with or without neck dissection (31320, 31360, 31365). Laryngopharyngectomies, partial laryngectomies, and minimally invasive laryngectomies were excluded. VASQIP nurse data managers reviewed patient data for operative data, postoperative outcomes (including 30- day morbidity and mortality), and preoperative risk factors (Appendix).²¹

The VASQIP data provide the highest resident or postgraduate year (PGY) per surgery. PGY 1, 2, and 3 were considered junior residents and PGY ≥4, surgical fellows, and individuals who took research years during residency were considered senior residents. Cases performed by attending physicians alone were compared with those involving junior or senior residents.

Patient demographic data included age, body mass index, smoking and alcohol use, weight loss, and functional status. Consumption of any tobacco products within 12 months of surgery was considered tobacco use. Drinking on average ≥2 alcoholic beverages daily was considered alcohol use. Weight loss was defined as a 10% reduction in body weight within the 6 months before surgery, excluding patients enrolled in a weight loss program. Functional status was categorized as independent, partially dependent, totally dependent, and unknown.

Primary outcomes included operative time, wRVUs generated, and wRVUs generated per hour of operative time. Postoperative complications were recorded both as a continuous variable and as a binary variable for presence or absence of a complication. Additional outcome variables included length of postoperative hospital stay, return to the operating room (OR), and death within 30 days of surgery.

Statistical Analysis

Data were summarized using frequency and percentage for categorical variables and median with IQR for continuous variables. Data were also summarized based on resident involvement in the surgery and the PGY level of the residents involved. The occurrence of total laryngectomy, rate of complications, and patient return to the OR were summarized by year.

Univariate associations between resident involvement and surgical outcomes were analyzed using the Kruskal-Wallis test for continuous variables and the ÷2 test for categorical variables. A Fisher exact test was used when the cell count in the contingency table was < 5. The univariate associations between surgical outcomes and demographic/preoperative variables were examined using 2-sided Wilcoxon ranksum tests or Kruskal-Wallis tests between continuous variables and categorical variables, X² or Fisher exact test between 2 categorical variables, and 2-sided Spearman correlation test between 2 continuous variables. A false-discovery rate approach was used for simultaneous posthoc tests to determine the adjusted P values for wRVUs generated/operative time for attending physicians alone vs with junior residents and for attending physicians alone vs with senior residents. Models were used to evaluate the effects of resident involvement on surgical outcomes, adjusting for variables that showed significant univariate associations. Linear regression models were used for operative time, wRVUs generated, wRVUs generated/operative time, and length of postoperative stay. A logistic regression model was used for death within 30 days. Models were not built for postoperative complications or patient return to the OR, as these were only statistically significantly associated with the patient’s preoperative functional status. A finding was considered significant if P < .05. All analyses were performed using statistical software RStudio Version 2023.03.0.

Results

Between 2001 and 2021, 1857 patients who underwent total laryngectomy were identified from the VASQIP database nationwide. Most of the total laryngectomies were staffed by an attending physician with a senior resident (n = 1190, 64%), 446 (24%) were conducted by the attending physician alone, and 221 (12%) by an attending physician with a junior resident (Table 1). The mean operating time for an attending physician alone was 378 minutes, 384 minutes for an attending physician with a senior resident, and 432 minutes for an attending physician with a junior resident (Table 2). There was a statistically significant increase in operating time for laryngectomies with resident participation compared to attending physicians operating alone (P < .001).

When the wRVUs generated/operative time was analyzed, there was a statistically significant difference between comparison groups. Total laryngectomies performed by attending physicians alone had the highest wRVUs generated/operative time (5.5), followed by laryngectomies performed by attending physicians with senior residents and laryngectomies performed by attending physicians with junior residents (5.2 and 4.8, respectively; P = .002). Table 3 describes adjusted P values for wRVUs generated/ operative time for total laryngectomies performed by attending physicians alone vs with junior residents (P = .003) and for attending physicians alone vs with senior residents (P = .02). Resident participation in total laryngectomies did not significantly impact the development or number of postoperative complications or the rate of return to the OR.

The number of laryngectomies performed in a single fiscal year peaked in 2010 at 170 cases (Figure 1). Between 2001 and 2021, the mean rates of postoperative complications (21.3%) and patient return to the OR (14.6%) did not significantly change. Resident participation in total laryngectomies also peaked in 2010 at 89.0% but has significantly declined, falling to a low of 43.6% in 2021 (Figure 2). From 2001 to 2011, the mean resident participation rate in total laryngectomies was 80.6%, compared with 68.3% from 2012 to 2021 (P < .001).

The effect of various demographic and preoperative characteristics on surgical outcomes was also analyzed. A linear regression model accounted for each variable significantly associated with operative time. On multivariable analysis, when all other variables were held constant, Table 4 shows the estimated change in operative time based on certain criteria. For instance, the operative time for attendings with junior residents surgeries was 40 minutes longer (95% CI, 16 to 64) than that of attending alone surgeries (P = .001). Furthermore, operative time decreased by 1.1 minutes (95% CI, 0.30 to 2.04) for each 1-year increase in patient age (P = .009).

A multivariable logistic regression model evaluated the effect of resident involvement on 30-day mortality rates. Senior resident involvement (P = .02), partially dependent functional status (P = .01), totally dependent functional status (P < .001), and advanced age (P = .02) all were significantly associated with 30-day mortality (Table 5). When other variables remained constant, the odds of death for totally dependent patients were 10.4 times higher than that of patients with independent functional status. Thus, totally dependent functional status appeared to have a greater impact on this outcome than resident participation. The linear regression model for postoperative length of stay demonstrated that senior resident involvement (P = .04), functional status (partially dependent vs independent P < .001), and age (P = .03) were significantly associated with prolonged length of stay.

Discussion

Otolaryngology residency training is designed to educate future otolaryngologists through hands-on learning, adequate feedback, and supervision.¹ Although this exposure is paramount for resident education, balancing appropriate supervision and autonomy while mitigating patient risk has been difficult. Numerous non-VA studies have reviewed the impact of resident participation on patient outcomes in various specialties, ranging from a single institution to the National Surgical Quality Improvement Program (NSQIP).^4,5,7,22 This study is the first to describe the nationwide impact of resident participation on outcomes in veterans undergoing total laryngectomy.

This study found that resident participation increases operative time and decreases wRVUs generated/operative time without impacting complication rates or patient return to the OR. This reinforces the notion that under close supervision, resident participation does not negatively impact patient outcomes. Resident operative training requires time and dedication by the attending physician and surgical team, thereby increasing operative time. Because VA physician compensation is not linked with productivity as closely as it is in other private and academic settings, surgeons can dedicate more time to operative teaching. This study found that a total laryngectomy involving a junior resident took about 45 minutes longer than an attending physician working alone.

As expected, with longer operative times, the wRVUs generated/operative time ratio was lower in cases with resident participation. Even though resident participation leads to lower OR efficiency, their participation may not significantly impact ancillary costs.²³ However, a recent study from NSQIP found an opportunity cost of $60.44 per hour for surgeons operating with a resident in head and neck cases.¹³

Postoperative complications and mortality are key measures of surgical outcomes in addition to operative time and efficiency. This study found that neither junior nor senior resident participation significantly increased complication rates or patient return to the OR. Despite declining resident involvement and the number of total laryngectomy surgeries in the VA, the complication rate has remained steady. The 30-day mortality rate was significantly higher in cases involving senior residents compared to cases with attending physicians alone. This could be a result of senior resident participation in more challenging cases, such as laryngectomies performed as salvage surgery following radiation. Residents are more often involved in cases with greater complexity at teaching institutions.^24-26 Therefore, the higher mortality seen among laryngectomies with senior resident involvement is likely due to the higher complexity of those cases.

The proportion of resident involvement in laryngectomies at VA medical centers has been decreasing over time. Due to the single payer nature of the VA health care system and the number of complex and comorbid patients, the VA offers an invaluable space for resident education in the OR. The fact that less than half of laryngectomies in 2021 involved resident participation is noteworthy for residency training programs. As wRVU compensation models evolve, VA attending surgeons may face less pressure to move the case along, leading to a high potential for operative teaching. Therefore, complex cases, such as laryngectomies, are often ideal for resident participation in the VA.

The steady decline in total laryngectomies at the VA parallels the recent decrease seen in non-VA settings.²⁰ This is due in part to the use of larynx-preserving treatment modalities for laryngeal cancer as well as decreases in the incidence of laryngeal cancer due to population level changes in smoking behaviors. ^18,19 Although a laryngectomy is not a key indicator case as determined by the Accreditation Council for Graduate Medical Education, it is important for otolaryngology residents to be exposed to these cases and have a thorough understanding of the operative technique.²⁷ Total laryngectomy was selected for this study because it is a complex and time-consuming surgery with somewhat standardized surgical steps. Unlike microvascular surgery that is very rarely performed by an attending physician alone, laryngectomies can be performed by attending physicians alone or with a resident.²⁸

Limitations

Since this was a retrospective study, it was susceptible to errors in data entry and data extraction from the VASQIP database. Another limitation is the lack of preoperative treatment data on tumor stage and prior nonoperative treatment. For example, a salvage laryngectomy after treatment with radiation and/or chemoradiation is a higher risk procedure than an upfront laryngectomy. Senior resident involvement may be more common in patients undergoing salvage laryngectomy due to the high risk of postoperative fistula and other complications. This may have contributed to the association identified between senior resident participation and 30-day mortality.

Since we could not account for residents who took research years or were fellows, a senior resident may have been mislabeled as a junior resident or vice versa. However, because most research years occur following the third year of residency. We are confident that PGY-1, PGY-2, and PGY-3 is likely to capture junior residents. Other factors, such as coattending surgeon cases, medical student assistance, and fellow involvement may have also impacted the results of this study.

Conclusions

This study is the first to investigate the impact of resident participation on operative time, wRVUs generated, and complication rates in head and neck surgery at VA medical centers. It found that resident participation in total laryngectomies among veterans increased operative time and reduced wRVUs generated per hour but did not impact complication rate or patient return to the OR. The VA offers a unique and invaluable space for resident education and operative training, and the recent decline in resident participation among laryngectomies is important for residency programs to acknowledge and potentially address moving forward.

In contrast to oral cavity resections which can vary from partial glossectomies to composite resections, laryngectomy represents a homogenous procedure from which to draw meaningful conclusions about complication rates, operative time, and outcome. Future directions should include studying other types of head and neck surgery in the VA to determine whether the impact of resident participation mirrors the findings of this study.

Low-dose aspirin commonly is used for the prevention of cardiovascular disease (CVD) but is associated with an increased risk of major bleeding.¹ The use of aspirin for primary prevention is largely extrapolated from clinical trials showing benefit in the secondary prevention of myocardial infarction and ischemic stroke. However, results from the Aspirin in Reducing Events in the Elderly (ASPREE) trial challenged this practice.² The ASPREE trial, conducted in the United States and Australia from 2010 to 2014, sought to determine whether daily 100 mg aspirin, was superior to placebo in promoting disability-free survival among older adults. Participants were aged ≥ 70 years (≥ 65 years for Hispanic and Black US participants), living in the community, and were free from preexisting CVD, cerebrovascular disease, or any chronic condition likely to limit survival to < 5 years. The study found no significant difference in the primary endpoints of death, dementia, or persistent physical disability, but there was a significantly higher risk of major hemorrhage in the aspirin group (3.8% vs 2.8%; hazard ratio, 1.38; 95% CI, 1.18-1.62; P < .001).

Several medical societies have updated their guideline recommendations for aspirin for primary prevention of CVD. The 2022 United States Public Service Task Force (USPSTF) provides a grade C recommendation (at least moderate certainty that the net benefit is small) to consider low-dose aspirin for the primary prevention of CVD on an individual patient basis for adults aged 40 to 59 years who have a ≥ 10% 10-year CVD risk. For adults aged ≥ 60 years, the USPSTF recommendation is grade D (moderate or high certainty that the practice has no net benefit or that harms outweigh the benefits) for low-dose aspirin use.^1,3 The American College of Cardiology and American Heart Association (ACC/AHA) recommend considering low-dose aspirin for primary prevention of atherosclerotic cardiovascular disease (ASCVD) among select adults aged 40 to 70 years at higher CVD risk but not at increased risk of bleeding.⁴ The American Diabetes Association (ADA) recommends low-dose aspirin for primary prevention of CVD in patients with diabetes and additional risk factors such as family history of premature ASCVD, hypertension, dyslipidemia, smoking, or chronic kidney disease, and who are not at higher risk of bleeding.⁵ The ADA standards also caution against the use of aspirin as primary prevention in patients aged > 70 years. Low-dose aspirin use is not recommended for the primary prevention of CVD in older adults or adults of any age who are at increased risk of bleeding.

Recent literature using the US Department of Veterans Affairs (VA) Corporate Data Warehouse database confirms 86,555 of 1.8 million veterans aged > 70 years (5%) were taking low-dose aspirin for primary prevention of ASCVD despite guideline recommendations.⁶ Higher risk of gastrointestinal and other major bleeding from low-dose aspirin has been reported in the literature.¹ Major bleeds represent a significant burden to the health care system with an estimated mean $13,093 cost for gastrointestinal bleed hospitalization.⁷

Considering the large scale aspirin use without appropriate indication within the veteran population, the risk of adverse effects, and the significant cost to patients and the health care system, it is imperative to determine the best approach to efficiently deprescribe aspirin for primary prevention among geriatric patients. Deprescribing refers to the systematic and supervised process of dose reduction or drug discontinuation with the goal of improving health and/or reducing the risk of adverse effects.⁸ During patient visits, primary care practitioners (PCPs) have opportunities to discontinue aspirin, but these encounters are time-limited and deprescribing might be secondary to more acute primary care needs. The shortage of PCPs is expected to worsen in coming years, which could further reduce their availability to assess inappropriate aspirin use.⁹

VA clinical pharmacist practitioners (CPPs) serve as medication experts and work autonomously under a broad scope of practice as part of the patient aligned care team.^10-12 CPPs can free up time for PCPs and facilitate deprescribing efforts, especially for older adults. One retrospective cohort study conducted at a VA medical center found that CPPs deprescribed more potentially inappropriate medications among individuals aged ≥ 80 years compared with usual care with PCPs (26.8% vs 16.1%; P < .001).^12,13 An aspirin deprescribing protocol conducted in 2022 resulted in nearly half of veterans aged ≥ 70 years contacted by phone agreeing to stop aspirin. Although this study supports the role pharmacists can play in reducing aspirin use in accordance with guidelines, the authors acknowledge that their interventions had a mean time of 12 minutes per patient and would require workflow changes.¹⁴ The purpose of this study is to evaluate the efficiency of aspirin deprescribing through 2 approaches: direct deprescribing by pharmacists using populationlevel review compared with clinicians following a pharmacist-led education.

Methods

This was a single-center quality improvement cohort study at the Durham VA Health Care System (DVAHCS) in North Carolina. Patients included were aged ≥ 70 years without known ASCVD who received care at any of 3 DVAHCS community-based outpatient clinics and prescribed aspirin. Patient data was obtained using the VIONE (Deprescribing Dashboard called Vital, Important, Optional, Not indicated, and Every medication has a specific indication or diagnosis) dashboard.¹⁵ VIONE was developed to identify potentially inappropriate medications (PIMs) that are eligible to deprescribe based on Beers Criteria, Screening Tool of Older Personsf Prescriptions criteria, and common clinical scenarios when clinicians determine the risk outweighs the benefit to continue a specific medication. ^16,17 VIONE is used to reduce polypharmacy and improve patient safety, comfort, and medication adherence. Aspirin for patients aged ≥ 70 years without a history of ASCVD is a PIM identified by VIONE. Patients aged ≥ 70 years were chosen as an inclusion criteria in this study to match the ASPREE trial inclusion criteria and age inclusion criteria in the VIONE dashboard for aspirin deprescribing.² Patient lists were generated for these potentially inappropriate aspirin prescriptions for 3 months before clinician staff education presentations, the day of the presentations, and 3 months after.

The primary endpoint was the number of veterans with aspirin deprescribed directly by 2 pharmacists over 12 weeks, divided by total patient care time spent, compared with the change in number of veterans with aspirin deprescribed by any DVAHCS physician, nurse practitioner, physician assistant, or CPP over 12 weeks, divided by the total pharmacist time spent on PCP education. Secondary endpoints were the number of aspirin orders discontinued by pharmacists and CPPs, the number of aspirin orders discontinued 12 weeks before pharmacist-led education compared with the number of aspirin orders discontinued 12 weeks after CPP-led education, average and median pharmacist time spent per patient encounter, and time of direct patient encounters vs time spent on PCP education.

Pharmacists reviewed each patient who met the inclusion criteria from the list generated by VIONE on December 1, 2022, for aspirin appropriateness according to the ACC/AHA and USPSTF guidelines, with the goal to discontinue aspirin for primary prevention of ASCVD and no other indications.^1,4 Pharmacists documented their visits using VIONE methodology in the Computerized Patient Record System (CPRS) using a polypharmacy review note. CPPs contacted patients who were taking aspirin for primary prevention by unscheduled telephone call to assess for aspirin adherence, undocumented history of ASCVD, cardiovascular risk factors, and history of bleeding. Aspirin was discontinued if patients met guideline criteria recommendations and agreed to discontinuation. Risk-benefit discussions were completed when patients without known ASCVD were considered high risk because the ACC/AHA guidelines mention there is insufficient evidence of safety and efficacy of aspirin for primary prevention for patients with other known ASCVD risk factors (eg, strong family history of premature myocardial infarction, inability to achieve lipid, blood pressure, or glucose targets, or significant elevation in coronary artery calcium score).

High risk was defined as family history of premature ASCVD (in a male first-degree relative aged < 55 years or a female first-degree relative aged < 65 years), most recent blood pressure or 2 blood pressure results in the last 12 months > 160/100 mm Hg, recent hemoglobin A_1c > 9%, and/or low-density lipoprotein > 190 mg/dL or not prescribed an indicated statin.³ Aspirin was continued or discontinued according to patient preference after the personalized risk-benefit discussion.

For patients with a clinical indication for aspirin use other than ASCVD (eg, atrial fibrillation not on anticoagulation, venous thromboembolism prophylaxis, carotid artery disease), CPPs documented their assessment and when appropriate deferred to the PCP for consideration of stopping aspirin. For patients with undocumented ASCVD, CPPs added their ASCVD history to their problem list and aspirin was continued. PCPs were notified by alert when aspirin was discontinued and when patients could not be reached by telephone.

presented a review of recent guideline updates and supporting literature at 2 online staff meetings. The education sessions lasted about 10 minutes and were presented to PCPs across 3 community-based outpatient clinics. An estimated 40 minutes were spent creating the PowerPoint education materials, seeking feedback, making edits, and answering questions or emails from PCPs after the presentation. During the presentation, pharmacists encouraged PCPs to discontinue aspirin (active VA prescriptions and reported over-the-counter use) for primary prevention of ASCVD in patients aged ≥ 70 years during their upcoming appointments and consider risk factors recommended by the ACC/AHA guidelines when applicable. PCPs were notified that CPPs planned to start a population review for discontinuing active VA aspirin prescriptions on December 1, 2022. The primary endpoint and secondary endpoints were analyzed using descriptive statistics. All data were analyzed using Microsoft Excel.

Results

A total of 868 patients aged ≥ 70 years with active prescriptions for aspirin were identified on December 1, 2022. After applying inclusion and exclusion criteria for the pharmacist population review, 224 patients were included for cohort final analysis (Figure). All 868 patients were eligible for the CPP intervention. Primary reasons for exclusion from the CPP population included over-thecounter aspirin and a history of ASCVD in the patient’s problem list. All patients were male, with a mean (SD) age of 75 (4.4) years (Table 1). Most patients were prescribed aspirin, 81 mg daily (n = 220; 98%).

The direct CPP deprescribing intervention resulted in 2 aspirin prescriptions discontinued per hour of pharmacist time and 67 aspirin prescriptions discontinued per hour of pharmacist time via the PCP education intervention. CPPs discontinued 66 aspirin orders in the 12 weeks before the PCP education sessions. A total of 230 aspirin prescriptions were discontinued in the 12 weeks following the PCP education sessions, with 97 discontinued directly by CPPs and 133 discontinued by PCPs. The PCP education session yielded an additional 67 discontinued aspirin orders compared with the 12 weeks before the education sessions (Table 2).

The CPP direct deprescribing intervention took about 48.3 hours, accounting for health record review and time interacting with patients. The PCP education intervention took about 60 minutes, which included time for preparing and delivering education materials (Table 3). CPP deprescribing encounter types, interventions, and related subcategories, and other identified indications to continue aspirin are listed in Table 4.

Discussion

Compared with direct deprescribing by pharmacists, the PCP education intervention was more efficient based on number of aspirin orders discontinued by pharmacist time. PCPs discontinued twice as many aspirin prescriptions in the 12 weeks after pharmacist-led education compared with the 12 weeks before.

Patients were primarily contacted by telephone (73%) for deprescribing. Among the 163 patients reached by phone and encouraged to discontinue aspirin, 97 patients (60%) accepted the recommendation, which was similar to the acceptance rates found in the literature (48% to 55%).^14,18 Although many veterans continued taking aspirin (78%), most had indications for its continued use, such as a history of ASCVD, atrial fibrillation without anticoagulation, and carotid artery stenosis, and complex comorbidities that required further discussion with their PCP. Less common uses for aspirin were identified through CPRS review or patient reports included cerebral small vessel disease without history of ASCVD, subclavian artery stenosis, thrombocytosis, bioprosthetic valve replacement, giant cell arteritis, rheumatoid arthritis, and prevention of second eye involvement of ischemic optic neuropathy.

to describe the benefit of clinical pharmacy services for deprescribing aspirin for primary prevention of ASCVD through PCP education. Previously published literature has assessed alternative ways to identify or discontinue PIMs—including aspirin—among geriatric patients. One study evaluated the use of marking inappropriate aspirin prescriptions in the electronic health database, leading to a significant reduction in incidence of inappropriate aspirin prescribing; however, it did not assess changes in discontinuation rates of existing aspirin prescriptions.¹⁹ The previous VA pharmacist aspirin deprescribing protocol demonstrated pharmacists’ aptitude at discontinuing aspirin for primary prevention but only used direct patient contact and did not compare efficiency with other methods, including PCP education.¹⁴

This quality improvement project contributes new data to the existing literature to support the use of clinical pharmacists to discontinue aspirin for primary prevention and suggests a strong role for pharmacists as educators on clinical guidelines, in addition to their roles directly deprescribing PIMs in clinical practice. This study is further strengthened by its use of VIONE, which previously has demonstrated effectiveness in deprescribing a variety of PIMs in primary care settings.²⁰

Despite using VIONE for generating a list of patients eligible for deprescription, our CPRS review found that this list was frequently inaccurate. For example, a small portion of patients were on the VIONE generated list indicating they had no ASCVD history, but had transient ischemic attack listed in their problem lists. Patient problem lists often were missing documented ASCVD history that was revealed by patient interview or CPRS review. It is possible that patients interviewed might have omitted relevant ASCVD history because of low health literacy, conditions affecting memory, or use of health care services outside the VA system.

There were several instances of aspirin used for other non-ASCVD indications, such as primary stroke prevention in atrial fibrillation. The ACC/AHA atrial fibrillation guidelines previously provided a Class IIb recommendation (benefit is greater than risk but additional studies are needed) for considering no antithrombic therapy or treatment with oral anticoagulant or aspirin for nonvalvular atrial fibrillation with CHA₂DS₂-VASc (Congestive heart failure, Hypertension, Age [> 65 y, 1 point; > 75 y, 2 points], Diabetes, previous Stroke/transient ischemic attack [2 points]) score of 1.²¹ The ACC/ AHA guidelines were updated in 2023 to recommend against antiplatelet therapy as an alternative to anticoagulation for reducing cardioembolic stroke risk among patients with atrial fibrillation with no indication for antiplatelet therapy because of risk of harm.²² If a patient has no risk factors for stroke, aspirin is not recommended to prevent thromboembolic events because of a lack of benefit. Interventions from this quality improvement study were completed before the 2023 atrial fibrillation guideline was published and therefore in this study aspirin was not discontinued when used for atrial fibrillation. Aspirin use for atrial fibrillation might benefit from similar discontinuation efforts analyzed within this study. Beyond atrial fibrillation, major guidelines do not comment on the use of aspirin for any other indications in the absence of clinical ASCVD.

Limitations

This study is limited by the lack of clinical consensus for complex patients and demonstrates the importance of individualized patient assessment when considering discontinuing aspirin. Because of the project’s relatively short intervention period, aspirin deprescribing rates could decrease over time and repeated education efforts might be necessary to see lasting impact. Health care professionals from services outside of primary care also might have discontinued aspirin during the study period unrelated to the education and these discontinued aspirin prescriptions could contribute to the higher rate observed among PCPs. This study included a specific population cohort of male, US veterans and might not reflect other populations where these interventions could be implemented.

The measurement of time spent by pharmacists and PCPs is an additional limitation. Although it is expected that PCPs attempt to discontinue aspirin during their existing patient care appointments, the time spent during visits was not measured or documented. Direct deprescribing by pharmacist CPRS review required a significant amount of time and could be a barrier to successful intervention by CPPs in patient aligned care teams.

To reduce the time pharmacists spent completing CPRS reviews, an aspirin deprescribing clinical reminder tool could be used to assess use and appropriate indication quickly during any primary care visit led by a PCP or CPP. In addition, it is recommended that pharmacists regularly educate health care professionals on guideline recommendations for aspirin use among geriatric patients. Future studies of the incidence of major cardiovascular events after aspirin deprescribing among geriatric patients and a longitudinal cost/benefit analysis could support these initiatives.

Conclusions

In this study, pharmacists successfully deprescribed inappropriate medications, such as aspirin. However, pharmacist-led PCP education is more efficient compared with direct deprescribing using a population-level review. PCP education requires less time and could allow ambulatory care pharmacists to spend more time on other direct patient care interventions to improve quality and access to care in primary care clinics. This study’s results further support the role of pharmacists in deprescribing PIMs for older adults and the use of a deprescribing tool, such as VIONE, in a primary care setting.

Low-dose aspirin commonly is used for the prevention of cardiovascular disease (CVD) but is associated with an increased risk of major bleeding.¹ The use of aspirin for primary prevention is largely extrapolated from clinical trials showing benefit in the secondary prevention of myocardial infarction and ischemic stroke. However, results from the Aspirin in Reducing Events in the Elderly (ASPREE) trial challenged this practice.² The ASPREE trial, conducted in the United States and Australia from 2010 to 2014, sought to determine whether daily 100 mg aspirin, was superior to placebo in promoting disability-free survival among older adults. Participants were aged ≥ 70 years (≥ 65 years for Hispanic and Black US participants), living in the community, and were free from preexisting CVD, cerebrovascular disease, or any chronic condition likely to limit survival to < 5 years. The study found no significant difference in the primary endpoints of death, dementia, or persistent physical disability, but there was a significantly higher risk of major hemorrhage in the aspirin group (3.8% vs 2.8%; hazard ratio, 1.38; 95% CI, 1.18-1.62; P < .001).

Several medical societies have updated their guideline recommendations for aspirin for primary prevention of CVD. The 2022 United States Public Service Task Force (USPSTF) provides a grade C recommendation (at least moderate certainty that the net benefit is small) to consider low-dose aspirin for the primary prevention of CVD on an individual patient basis for adults aged 40 to 59 years who have a ≥ 10% 10-year CVD risk. For adults aged ≥ 60 years, the USPSTF recommendation is grade D (moderate or high certainty that the practice has no net benefit or that harms outweigh the benefits) for low-dose aspirin use.^1,3 The American College of Cardiology and American Heart Association (ACC/AHA) recommend considering low-dose aspirin for primary prevention of atherosclerotic cardiovascular disease (ASCVD) among select adults aged 40 to 70 years at higher CVD risk but not at increased risk of bleeding.⁴ The American Diabetes Association (ADA) recommends low-dose aspirin for primary prevention of CVD in patients with diabetes and additional risk factors such as family history of premature ASCVD, hypertension, dyslipidemia, smoking, or chronic kidney disease, and who are not at higher risk of bleeding.⁵ The ADA standards also caution against the use of aspirin as primary prevention in patients aged > 70 years. Low-dose aspirin use is not recommended for the primary prevention of CVD in older adults or adults of any age who are at increased risk of bleeding.

Recent literature using the US Department of Veterans Affairs (VA) Corporate Data Warehouse database confirms 86,555 of 1.8 million veterans aged > 70 years (5%) were taking low-dose aspirin for primary prevention of ASCVD despite guideline recommendations.⁶ Higher risk of gastrointestinal and other major bleeding from low-dose aspirin has been reported in the literature.¹ Major bleeds represent a significant burden to the health care system with an estimated mean $13,093 cost for gastrointestinal bleed hospitalization.⁷

Considering the large scale aspirin use without appropriate indication within the veteran population, the risk of adverse effects, and the significant cost to patients and the health care system, it is imperative to determine the best approach to efficiently deprescribe aspirin for primary prevention among geriatric patients. Deprescribing refers to the systematic and supervised process of dose reduction or drug discontinuation with the goal of improving health and/or reducing the risk of adverse effects.⁸ During patient visits, primary care practitioners (PCPs) have opportunities to discontinue aspirin, but these encounters are time-limited and deprescribing might be secondary to more acute primary care needs. The shortage of PCPs is expected to worsen in coming years, which could further reduce their availability to assess inappropriate aspirin use.⁹

VA clinical pharmacist practitioners (CPPs) serve as medication experts and work autonomously under a broad scope of practice as part of the patient aligned care team.^10-12 CPPs can free up time for PCPs and facilitate deprescribing efforts, especially for older adults. One retrospective cohort study conducted at a VA medical center found that CPPs deprescribed more potentially inappropriate medications among individuals aged ≥ 80 years compared with usual care with PCPs (26.8% vs 16.1%; P < .001).^12,13 An aspirin deprescribing protocol conducted in 2022 resulted in nearly half of veterans aged ≥ 70 years contacted by phone agreeing to stop aspirin. Although this study supports the role pharmacists can play in reducing aspirin use in accordance with guidelines, the authors acknowledge that their interventions had a mean time of 12 minutes per patient and would require workflow changes.¹⁴ The purpose of this study is to evaluate the efficiency of aspirin deprescribing through 2 approaches: direct deprescribing by pharmacists using populationlevel review compared with clinicians following a pharmacist-led education.

Methods

This was a single-center quality improvement cohort study at the Durham VA Health Care System (DVAHCS) in North Carolina. Patients included were aged ≥ 70 years without known ASCVD who received care at any of 3 DVAHCS community-based outpatient clinics and prescribed aspirin. Patient data was obtained using the VIONE (Deprescribing Dashboard called Vital, Important, Optional, Not indicated, and Every medication has a specific indication or diagnosis) dashboard.¹⁵ VIONE was developed to identify potentially inappropriate medications (PIMs) that are eligible to deprescribe based on Beers Criteria, Screening Tool of Older Personsf Prescriptions criteria, and common clinical scenarios when clinicians determine the risk outweighs the benefit to continue a specific medication. ^16,17 VIONE is used to reduce polypharmacy and improve patient safety, comfort, and medication adherence. Aspirin for patients aged ≥ 70 years without a history of ASCVD is a PIM identified by VIONE. Patients aged ≥ 70 years were chosen as an inclusion criteria in this study to match the ASPREE trial inclusion criteria and age inclusion criteria in the VIONE dashboard for aspirin deprescribing.² Patient lists were generated for these potentially inappropriate aspirin prescriptions for 3 months before clinician staff education presentations, the day of the presentations, and 3 months after.

The primary endpoint was the number of veterans with aspirin deprescribed directly by 2 pharmacists over 12 weeks, divided by total patient care time spent, compared with the change in number of veterans with aspirin deprescribed by any DVAHCS physician, nurse practitioner, physician assistant, or CPP over 12 weeks, divided by the total pharmacist time spent on PCP education. Secondary endpoints were the number of aspirin orders discontinued by pharmacists and CPPs, the number of aspirin orders discontinued 12 weeks before pharmacist-led education compared with the number of aspirin orders discontinued 12 weeks after CPP-led education, average and median pharmacist time spent per patient encounter, and time of direct patient encounters vs time spent on PCP education.

Pharmacists reviewed each patient who met the inclusion criteria from the list generated by VIONE on December 1, 2022, for aspirin appropriateness according to the ACC/AHA and USPSTF guidelines, with the goal to discontinue aspirin for primary prevention of ASCVD and no other indications.^1,4 Pharmacists documented their visits using VIONE methodology in the Computerized Patient Record System (CPRS) using a polypharmacy review note. CPPs contacted patients who were taking aspirin for primary prevention by unscheduled telephone call to assess for aspirin adherence, undocumented history of ASCVD, cardiovascular risk factors, and history of bleeding. Aspirin was discontinued if patients met guideline criteria recommendations and agreed to discontinuation. Risk-benefit discussions were completed when patients without known ASCVD were considered high risk because the ACC/AHA guidelines mention there is insufficient evidence of safety and efficacy of aspirin for primary prevention for patients with other known ASCVD risk factors (eg, strong family history of premature myocardial infarction, inability to achieve lipid, blood pressure, or glucose targets, or significant elevation in coronary artery calcium score).

High risk was defined as family history of premature ASCVD (in a male first-degree relative aged < 55 years or a female first-degree relative aged < 65 years), most recent blood pressure or 2 blood pressure results in the last 12 months > 160/100 mm Hg, recent hemoglobin A_1c > 9%, and/or low-density lipoprotein > 190 mg/dL or not prescribed an indicated statin.³ Aspirin was continued or discontinued according to patient preference after the personalized risk-benefit discussion.

For patients with a clinical indication for aspirin use other than ASCVD (eg, atrial fibrillation not on anticoagulation, venous thromboembolism prophylaxis, carotid artery disease), CPPs documented their assessment and when appropriate deferred to the PCP for consideration of stopping aspirin. For patients with undocumented ASCVD, CPPs added their ASCVD history to their problem list and aspirin was continued. PCPs were notified by alert when aspirin was discontinued and when patients could not be reached by telephone.

presented a review of recent guideline updates and supporting literature at 2 online staff meetings. The education sessions lasted about 10 minutes and were presented to PCPs across 3 community-based outpatient clinics. An estimated 40 minutes were spent creating the PowerPoint education materials, seeking feedback, making edits, and answering questions or emails from PCPs after the presentation. During the presentation, pharmacists encouraged PCPs to discontinue aspirin (active VA prescriptions and reported over-the-counter use) for primary prevention of ASCVD in patients aged ≥ 70 years during their upcoming appointments and consider risk factors recommended by the ACC/AHA guidelines when applicable. PCPs were notified that CPPs planned to start a population review for discontinuing active VA aspirin prescriptions on December 1, 2022. The primary endpoint and secondary endpoints were analyzed using descriptive statistics. All data were analyzed using Microsoft Excel.

Results

A total of 868 patients aged ≥ 70 years with active prescriptions for aspirin were identified on December 1, 2022. After applying inclusion and exclusion criteria for the pharmacist population review, 224 patients were included for cohort final analysis (Figure). All 868 patients were eligible for the CPP intervention. Primary reasons for exclusion from the CPP population included over-thecounter aspirin and a history of ASCVD in the patient’s problem list. All patients were male, with a mean (SD) age of 75 (4.4) years (Table 1). Most patients were prescribed aspirin, 81 mg daily (n = 220; 98%).

The direct CPP deprescribing intervention resulted in 2 aspirin prescriptions discontinued per hour of pharmacist time and 67 aspirin prescriptions discontinued per hour of pharmacist time via the PCP education intervention. CPPs discontinued 66 aspirin orders in the 12 weeks before the PCP education sessions. A total of 230 aspirin prescriptions were discontinued in the 12 weeks following the PCP education sessions, with 97 discontinued directly by CPPs and 133 discontinued by PCPs. The PCP education session yielded an additional 67 discontinued aspirin orders compared with the 12 weeks before the education sessions (Table 2).

The CPP direct deprescribing intervention took about 48.3 hours, accounting for health record review and time interacting with patients. The PCP education intervention took about 60 minutes, which included time for preparing and delivering education materials (Table 3). CPP deprescribing encounter types, interventions, and related subcategories, and other identified indications to continue aspirin are listed in Table 4.

Discussion

Compared with direct deprescribing by pharmacists, the PCP education intervention was more efficient based on number of aspirin orders discontinued by pharmacist time. PCPs discontinued twice as many aspirin prescriptions in the 12 weeks after pharmacist-led education compared with the 12 weeks before.

Patients were primarily contacted by telephone (73%) for deprescribing. Among the 163 patients reached by phone and encouraged to discontinue aspirin, 97 patients (60%) accepted the recommendation, which was similar to the acceptance rates found in the literature (48% to 55%).^14,18 Although many veterans continued taking aspirin (78%), most had indications for its continued use, such as a history of ASCVD, atrial fibrillation without anticoagulation, and carotid artery stenosis, and complex comorbidities that required further discussion with their PCP. Less common uses for aspirin were identified through CPRS review or patient reports included cerebral small vessel disease without history of ASCVD, subclavian artery stenosis, thrombocytosis, bioprosthetic valve replacement, giant cell arteritis, rheumatoid arthritis, and prevention of second eye involvement of ischemic optic neuropathy.

to describe the benefit of clinical pharmacy services for deprescribing aspirin for primary prevention of ASCVD through PCP education. Previously published literature has assessed alternative ways to identify or discontinue PIMs—including aspirin—among geriatric patients. One study evaluated the use of marking inappropriate aspirin prescriptions in the electronic health database, leading to a significant reduction in incidence of inappropriate aspirin prescribing; however, it did not assess changes in discontinuation rates of existing aspirin prescriptions.¹⁹ The previous VA pharmacist aspirin deprescribing protocol demonstrated pharmacists’ aptitude at discontinuing aspirin for primary prevention but only used direct patient contact and did not compare efficiency with other methods, including PCP education.¹⁴

This quality improvement project contributes new data to the existing literature to support the use of clinical pharmacists to discontinue aspirin for primary prevention and suggests a strong role for pharmacists as educators on clinical guidelines, in addition to their roles directly deprescribing PIMs in clinical practice. This study is further strengthened by its use of VIONE, which previously has demonstrated effectiveness in deprescribing a variety of PIMs in primary care settings.²⁰

Despite using VIONE for generating a list of patients eligible for deprescription, our CPRS review found that this list was frequently inaccurate. For example, a small portion of patients were on the VIONE generated list indicating they had no ASCVD history, but had transient ischemic attack listed in their problem lists. Patient problem lists often were missing documented ASCVD history that was revealed by patient interview or CPRS review. It is possible that patients interviewed might have omitted relevant ASCVD history because of low health literacy, conditions affecting memory, or use of health care services outside the VA system.

There were several instances of aspirin used for other non-ASCVD indications, such as primary stroke prevention in atrial fibrillation. The ACC/AHA atrial fibrillation guidelines previously provided a Class IIb recommendation (benefit is greater than risk but additional studies are needed) for considering no antithrombic therapy or treatment with oral anticoagulant or aspirin for nonvalvular atrial fibrillation with CHA₂DS₂-VASc (Congestive heart failure, Hypertension, Age [> 65 y, 1 point; > 75 y, 2 points], Diabetes, previous Stroke/transient ischemic attack [2 points]) score of 1.²¹ The ACC/ AHA guidelines were updated in 2023 to recommend against antiplatelet therapy as an alternative to anticoagulation for reducing cardioembolic stroke risk among patients with atrial fibrillation with no indication for antiplatelet therapy because of risk of harm.²² If a patient has no risk factors for stroke, aspirin is not recommended to prevent thromboembolic events because of a lack of benefit. Interventions from this quality improvement study were completed before the 2023 atrial fibrillation guideline was published and therefore in this study aspirin was not discontinued when used for atrial fibrillation. Aspirin use for atrial fibrillation might benefit from similar discontinuation efforts analyzed within this study. Beyond atrial fibrillation, major guidelines do not comment on the use of aspirin for any other indications in the absence of clinical ASCVD.

Limitations

This study is limited by the lack of clinical consensus for complex patients and demonstrates the importance of individualized patient assessment when considering discontinuing aspirin. Because of the project’s relatively short intervention period, aspirin deprescribing rates could decrease over time and repeated education efforts might be necessary to see lasting impact. Health care professionals from services outside of primary care also might have discontinued aspirin during the study period unrelated to the education and these discontinued aspirin prescriptions could contribute to the higher rate observed among PCPs. This study included a specific population cohort of male, US veterans and might not reflect other populations where these interventions could be implemented.

The measurement of time spent by pharmacists and PCPs is an additional limitation. Although it is expected that PCPs attempt to discontinue aspirin during their existing patient care appointments, the time spent during visits was not measured or documented. Direct deprescribing by pharmacist CPRS review required a significant amount of time and could be a barrier to successful intervention by CPPs in patient aligned care teams.

To reduce the time pharmacists spent completing CPRS reviews, an aspirin deprescribing clinical reminder tool could be used to assess use and appropriate indication quickly during any primary care visit led by a PCP or CPP. In addition, it is recommended that pharmacists regularly educate health care professionals on guideline recommendations for aspirin use among geriatric patients. Future studies of the incidence of major cardiovascular events after aspirin deprescribing among geriatric patients and a longitudinal cost/benefit analysis could support these initiatives.

Conclusions

In this study, pharmacists successfully deprescribed inappropriate medications, such as aspirin. However, pharmacist-led PCP education is more efficient compared with direct deprescribing using a population-level review. PCP education requires less time and could allow ambulatory care pharmacists to spend more time on other direct patient care interventions to improve quality and access to care in primary care clinics. This study’s results further support the role of pharmacists in deprescribing PIMs for older adults and the use of a deprescribing tool, such as VIONE, in a primary care setting.

Low-dose aspirin commonly is used for the prevention of cardiovascular disease (CVD) but is associated with an increased risk of major bleeding.¹ The use of aspirin for primary prevention is largely extrapolated from clinical trials showing benefit in the secondary prevention of myocardial infarction and ischemic stroke. However, results from the Aspirin in Reducing Events in the Elderly (ASPREE) trial challenged this practice.² The ASPREE trial, conducted in the United States and Australia from 2010 to 2014, sought to determine whether daily 100 mg aspirin, was superior to placebo in promoting disability-free survival among older adults. Participants were aged ≥ 70 years (≥ 65 years for Hispanic and Black US participants), living in the community, and were free from preexisting CVD, cerebrovascular disease, or any chronic condition likely to limit survival to < 5 years. The study found no significant difference in the primary endpoints of death, dementia, or persistent physical disability, but there was a significantly higher risk of major hemorrhage in the aspirin group (3.8% vs 2.8%; hazard ratio, 1.38; 95% CI, 1.18-1.62; P < .001).

Several medical societies have updated their guideline recommendations for aspirin for primary prevention of CVD. The 2022 United States Public Service Task Force (USPSTF) provides a grade C recommendation (at least moderate certainty that the net benefit is small) to consider low-dose aspirin for the primary prevention of CVD on an individual patient basis for adults aged 40 to 59 years who have a ≥ 10% 10-year CVD risk. For adults aged ≥ 60 years, the USPSTF recommendation is grade D (moderate or high certainty that the practice has no net benefit or that harms outweigh the benefits) for low-dose aspirin use.^1,3 The American College of Cardiology and American Heart Association (ACC/AHA) recommend considering low-dose aspirin for primary prevention of atherosclerotic cardiovascular disease (ASCVD) among select adults aged 40 to 70 years at higher CVD risk but not at increased risk of bleeding.⁴ The American Diabetes Association (ADA) recommends low-dose aspirin for primary prevention of CVD in patients with diabetes and additional risk factors such as family history of premature ASCVD, hypertension, dyslipidemia, smoking, or chronic kidney disease, and who are not at higher risk of bleeding.⁵ The ADA standards also caution against the use of aspirin as primary prevention in patients aged > 70 years. Low-dose aspirin use is not recommended for the primary prevention of CVD in older adults or adults of any age who are at increased risk of bleeding.

Recent literature using the US Department of Veterans Affairs (VA) Corporate Data Warehouse database confirms 86,555 of 1.8 million veterans aged > 70 years (5%) were taking low-dose aspirin for primary prevention of ASCVD despite guideline recommendations.⁶ Higher risk of gastrointestinal and other major bleeding from low-dose aspirin has been reported in the literature.¹ Major bleeds represent a significant burden to the health care system with an estimated mean $13,093 cost for gastrointestinal bleed hospitalization.⁷

Considering the large scale aspirin use without appropriate indication within the veteran population, the risk of adverse effects, and the significant cost to patients and the health care system, it is imperative to determine the best approach to efficiently deprescribe aspirin for primary prevention among geriatric patients. Deprescribing refers to the systematic and supervised process of dose reduction or drug discontinuation with the goal of improving health and/or reducing the risk of adverse effects.⁸ During patient visits, primary care practitioners (PCPs) have opportunities to discontinue aspirin, but these encounters are time-limited and deprescribing might be secondary to more acute primary care needs. The shortage of PCPs is expected to worsen in coming years, which could further reduce their availability to assess inappropriate aspirin use.⁹

VA clinical pharmacist practitioners (CPPs) serve as medication experts and work autonomously under a broad scope of practice as part of the patient aligned care team.^10-12 CPPs can free up time for PCPs and facilitate deprescribing efforts, especially for older adults. One retrospective cohort study conducted at a VA medical center found that CPPs deprescribed more potentially inappropriate medications among individuals aged ≥ 80 years compared with usual care with PCPs (26.8% vs 16.1%; P < .001).^12,13 An aspirin deprescribing protocol conducted in 2022 resulted in nearly half of veterans aged ≥ 70 years contacted by phone agreeing to stop aspirin. Although this study supports the role pharmacists can play in reducing aspirin use in accordance with guidelines, the authors acknowledge that their interventions had a mean time of 12 minutes per patient and would require workflow changes.¹⁴ The purpose of this study is to evaluate the efficiency of aspirin deprescribing through 2 approaches: direct deprescribing by pharmacists using populationlevel review compared with clinicians following a pharmacist-led education.

Methods

This was a single-center quality improvement cohort study at the Durham VA Health Care System (DVAHCS) in North Carolina. Patients included were aged ≥ 70 years without known ASCVD who received care at any of 3 DVAHCS community-based outpatient clinics and prescribed aspirin. Patient data was obtained using the VIONE (Deprescribing Dashboard called Vital, Important, Optional, Not indicated, and Every medication has a specific indication or diagnosis) dashboard.¹⁵ VIONE was developed to identify potentially inappropriate medications (PIMs) that are eligible to deprescribe based on Beers Criteria, Screening Tool of Older Personsf Prescriptions criteria, and common clinical scenarios when clinicians determine the risk outweighs the benefit to continue a specific medication. ^16,17 VIONE is used to reduce polypharmacy and improve patient safety, comfort, and medication adherence. Aspirin for patients aged ≥ 70 years without a history of ASCVD is a PIM identified by VIONE. Patients aged ≥ 70 years were chosen as an inclusion criteria in this study to match the ASPREE trial inclusion criteria and age inclusion criteria in the VIONE dashboard for aspirin deprescribing.² Patient lists were generated for these potentially inappropriate aspirin prescriptions for 3 months before clinician staff education presentations, the day of the presentations, and 3 months after.

The primary endpoint was the number of veterans with aspirin deprescribed directly by 2 pharmacists over 12 weeks, divided by total patient care time spent, compared with the change in number of veterans with aspirin deprescribed by any DVAHCS physician, nurse practitioner, physician assistant, or CPP over 12 weeks, divided by the total pharmacist time spent on PCP education. Secondary endpoints were the number of aspirin orders discontinued by pharmacists and CPPs, the number of aspirin orders discontinued 12 weeks before pharmacist-led education compared with the number of aspirin orders discontinued 12 weeks after CPP-led education, average and median pharmacist time spent per patient encounter, and time of direct patient encounters vs time spent on PCP education.

Pharmacists reviewed each patient who met the inclusion criteria from the list generated by VIONE on December 1, 2022, for aspirin appropriateness according to the ACC/AHA and USPSTF guidelines, with the goal to discontinue aspirin for primary prevention of ASCVD and no other indications.^1,4 Pharmacists documented their visits using VIONE methodology in the Computerized Patient Record System (CPRS) using a polypharmacy review note. CPPs contacted patients who were taking aspirin for primary prevention by unscheduled telephone call to assess for aspirin adherence, undocumented history of ASCVD, cardiovascular risk factors, and history of bleeding. Aspirin was discontinued if patients met guideline criteria recommendations and agreed to discontinuation. Risk-benefit discussions were completed when patients without known ASCVD were considered high risk because the ACC/AHA guidelines mention there is insufficient evidence of safety and efficacy of aspirin for primary prevention for patients with other known ASCVD risk factors (eg, strong family history of premature myocardial infarction, inability to achieve lipid, blood pressure, or glucose targets, or significant elevation in coronary artery calcium score).

High risk was defined as family history of premature ASCVD (in a male first-degree relative aged < 55 years or a female first-degree relative aged < 65 years), most recent blood pressure or 2 blood pressure results in the last 12 months > 160/100 mm Hg, recent hemoglobin A_1c > 9%, and/or low-density lipoprotein > 190 mg/dL or not prescribed an indicated statin.³ Aspirin was continued or discontinued according to patient preference after the personalized risk-benefit discussion.

For patients with a clinical indication for aspirin use other than ASCVD (eg, atrial fibrillation not on anticoagulation, venous thromboembolism prophylaxis, carotid artery disease), CPPs documented their assessment and when appropriate deferred to the PCP for consideration of stopping aspirin. For patients with undocumented ASCVD, CPPs added their ASCVD history to their problem list and aspirin was continued. PCPs were notified by alert when aspirin was discontinued and when patients could not be reached by telephone.

presented a review of recent guideline updates and supporting literature at 2 online staff meetings. The education sessions lasted about 10 minutes and were presented to PCPs across 3 community-based outpatient clinics. An estimated 40 minutes were spent creating the PowerPoint education materials, seeking feedback, making edits, and answering questions or emails from PCPs after the presentation. During the presentation, pharmacists encouraged PCPs to discontinue aspirin (active VA prescriptions and reported over-the-counter use) for primary prevention of ASCVD in patients aged ≥ 70 years during their upcoming appointments and consider risk factors recommended by the ACC/AHA guidelines when applicable. PCPs were notified that CPPs planned to start a population review for discontinuing active VA aspirin prescriptions on December 1, 2022. The primary endpoint and secondary endpoints were analyzed using descriptive statistics. All data were analyzed using Microsoft Excel.

Results

A total of 868 patients aged ≥ 70 years with active prescriptions for aspirin were identified on December 1, 2022. After applying inclusion and exclusion criteria for the pharmacist population review, 224 patients were included for cohort final analysis (Figure). All 868 patients were eligible for the CPP intervention. Primary reasons for exclusion from the CPP population included over-thecounter aspirin and a history of ASCVD in the patient’s problem list. All patients were male, with a mean (SD) age of 75 (4.4) years (Table 1). Most patients were prescribed aspirin, 81 mg daily (n = 220; 98%).

The direct CPP deprescribing intervention resulted in 2 aspirin prescriptions discontinued per hour of pharmacist time and 67 aspirin prescriptions discontinued per hour of pharmacist time via the PCP education intervention. CPPs discontinued 66 aspirin orders in the 12 weeks before the PCP education sessions. A total of 230 aspirin prescriptions were discontinued in the 12 weeks following the PCP education sessions, with 97 discontinued directly by CPPs and 133 discontinued by PCPs. The PCP education session yielded an additional 67 discontinued aspirin orders compared with the 12 weeks before the education sessions (Table 2).

The CPP direct deprescribing intervention took about 48.3 hours, accounting for health record review and time interacting with patients. The PCP education intervention took about 60 minutes, which included time for preparing and delivering education materials (Table 3). CPP deprescribing encounter types, interventions, and related subcategories, and other identified indications to continue aspirin are listed in Table 4.

Discussion

Compared with direct deprescribing by pharmacists, the PCP education intervention was more efficient based on number of aspirin orders discontinued by pharmacist time. PCPs discontinued twice as many aspirin prescriptions in the 12 weeks after pharmacist-led education compared with the 12 weeks before.

Patients were primarily contacted by telephone (73%) for deprescribing. Among the 163 patients reached by phone and encouraged to discontinue aspirin, 97 patients (60%) accepted the recommendation, which was similar to the acceptance rates found in the literature (48% to 55%).^14,18 Although many veterans continued taking aspirin (78%), most had indications for its continued use, such as a history of ASCVD, atrial fibrillation without anticoagulation, and carotid artery stenosis, and complex comorbidities that required further discussion with their PCP. Less common uses for aspirin were identified through CPRS review or patient reports included cerebral small vessel disease without history of ASCVD, subclavian artery stenosis, thrombocytosis, bioprosthetic valve replacement, giant cell arteritis, rheumatoid arthritis, and prevention of second eye involvement of ischemic optic neuropathy.

to describe the benefit of clinical pharmacy services for deprescribing aspirin for primary prevention of ASCVD through PCP education. Previously published literature has assessed alternative ways to identify or discontinue PIMs—including aspirin—among geriatric patients. One study evaluated the use of marking inappropriate aspirin prescriptions in the electronic health database, leading to a significant reduction in incidence of inappropriate aspirin prescribing; however, it did not assess changes in discontinuation rates of existing aspirin prescriptions.¹⁹ The previous VA pharmacist aspirin deprescribing protocol demonstrated pharmacists’ aptitude at discontinuing aspirin for primary prevention but only used direct patient contact and did not compare efficiency with other methods, including PCP education.¹⁴

This quality improvement project contributes new data to the existing literature to support the use of clinical pharmacists to discontinue aspirin for primary prevention and suggests a strong role for pharmacists as educators on clinical guidelines, in addition to their roles directly deprescribing PIMs in clinical practice. This study is further strengthened by its use of VIONE, which previously has demonstrated effectiveness in deprescribing a variety of PIMs in primary care settings.²⁰

Despite using VIONE for generating a list of patients eligible for deprescription, our CPRS review found that this list was frequently inaccurate. For example, a small portion of patients were on the VIONE generated list indicating they had no ASCVD history, but had transient ischemic attack listed in their problem lists. Patient problem lists often were missing documented ASCVD history that was revealed by patient interview or CPRS review. It is possible that patients interviewed might have omitted relevant ASCVD history because of low health literacy, conditions affecting memory, or use of health care services outside the VA system.

There were several instances of aspirin used for other non-ASCVD indications, such as primary stroke prevention in atrial fibrillation. The ACC/AHA atrial fibrillation guidelines previously provided a Class IIb recommendation (benefit is greater than risk but additional studies are needed) for considering no antithrombic therapy or treatment with oral anticoagulant or aspirin for nonvalvular atrial fibrillation with CHA₂DS₂-VASc (Congestive heart failure, Hypertension, Age [> 65 y, 1 point; > 75 y, 2 points], Diabetes, previous Stroke/transient ischemic attack [2 points]) score of 1.²¹ The ACC/ AHA guidelines were updated in 2023 to recommend against antiplatelet therapy as an alternative to anticoagulation for reducing cardioembolic stroke risk among patients with atrial fibrillation with no indication for antiplatelet therapy because of risk of harm.²² If a patient has no risk factors for stroke, aspirin is not recommended to prevent thromboembolic events because of a lack of benefit. Interventions from this quality improvement study were completed before the 2023 atrial fibrillation guideline was published and therefore in this study aspirin was not discontinued when used for atrial fibrillation. Aspirin use for atrial fibrillation might benefit from similar discontinuation efforts analyzed within this study. Beyond atrial fibrillation, major guidelines do not comment on the use of aspirin for any other indications in the absence of clinical ASCVD.

Limitations

This study is limited by the lack of clinical consensus for complex patients and demonstrates the importance of individualized patient assessment when considering discontinuing aspirin. Because of the project’s relatively short intervention period, aspirin deprescribing rates could decrease over time and repeated education efforts might be necessary to see lasting impact. Health care professionals from services outside of primary care also might have discontinued aspirin during the study period unrelated to the education and these discontinued aspirin prescriptions could contribute to the higher rate observed among PCPs. This study included a specific population cohort of male, US veterans and might not reflect other populations where these interventions could be implemented.

The measurement of time spent by pharmacists and PCPs is an additional limitation. Although it is expected that PCPs attempt to discontinue aspirin during their existing patient care appointments, the time spent during visits was not measured or documented. Direct deprescribing by pharmacist CPRS review required a significant amount of time and could be a barrier to successful intervention by CPPs in patient aligned care teams.

To reduce the time pharmacists spent completing CPRS reviews, an aspirin deprescribing clinical reminder tool could be used to assess use and appropriate indication quickly during any primary care visit led by a PCP or CPP. In addition, it is recommended that pharmacists regularly educate health care professionals on guideline recommendations for aspirin use among geriatric patients. Future studies of the incidence of major cardiovascular events after aspirin deprescribing among geriatric patients and a longitudinal cost/benefit analysis could support these initiatives.

Conclusions

In this study, pharmacists successfully deprescribed inappropriate medications, such as aspirin. However, pharmacist-led PCP education is more efficient compared with direct deprescribing using a population-level review. PCP education requires less time and could allow ambulatory care pharmacists to spend more time on other direct patient care interventions to improve quality and access to care in primary care clinics. This study’s results further support the role of pharmacists in deprescribing PIMs for older adults and the use of a deprescribing tool, such as VIONE, in a primary care setting.

Clinical practice has shifted from vitamin K antagonists to direct oral anticoagulants (DOACs) for atrial fibrillation treatment due to their more favorable risk-benefit profile and less lifestyle modification required.^1,2 However, the advantage of a lower bleeding risk with DOACs could be compromised by potentially problematic pharmacokinetic interactions like those conferred by antiplatelets or nonsteroidal anti-inflammatory drugs (NSAIDs).^3,4 Treating a patient needing anticoagulation with a DOAC who has comorbidities may introduce unavoidable drug-drug interactions. This particularly happens with over-the-counter and prescription NSAIDs used for the management of pain and inflammatory conditions.⁵

NSAIDs primarily affect 2 cyclooxygenase (COX) enzyme isomers, COX-1 and COX-2.⁶ COX-1 helps maintain gastrointestinal (GI) mucosa integrity and platelet aggregation processes, whereas COX-2 is engaged in pain signaling and inflammation mediation. COX-1 inhibition is associated with more bleeding-related adverse events (AEs), especially in the GI tract. COX-2 inhibition is thought to provide analgesia and anti-inflammatory properties without elevating bleeding risk. This premise is responsible for the preferential use of celecoxib, a COX-2 selective NSAID, which should confer a lower bleeding risk compared to nonselective NSAIDs such as ibuprofen and naproxen.⁷ NSAIDs have been documented as independent risk factors for bleeding. NSAID users are about 3 times as likely to develop GI AEs compared to nonNSAID users.⁸

Many clinicians aim to further mitigate NSAID-associated bleeding risk by coprescribing a proton pump inhibitor (PPI). PPIs provide gastroprotection against NSAID-induced mucosal injury and sequential complication of GI bleeding. In a multicenter randomized control trial, patients who received concomitant PPI therapy while undergoing chronic NSAID therapy—including nonselective and COX-2 selective NSAIDs—had a significantly lower risk of GI ulcer development (placebo, 17.0%; 20 mg esomeprazole, 5.2%; 40 mg esomeprazole, 4.6%).⁹ Current clinical guidelines for preventing NSAIDassociated bleeding complications recommend using a COX-2 selective NSAID in combination with PPI therapy for patients at high risk for GI-related bleeding, including the concomitant use of anticoagulants.¹⁰

There is evidence suggesting an increased bleeding risk with NSAIDs when used in combination with vitamin K antagonists such as warfarin.^11,12 A systematic review of warfarin and concomitant NSAID use found an increased risk of overall bleeding with NSAID use in combination with warfarin (odds ratio 1.58; 95% CI, 1.18-2.12), compared to warfarin alone.¹²

Posthoc analyses of randomized clinical trials have also demonstrated an increased bleeding risk with oral anticoagulation and concomitant NSAID use.^13,14 In the RE-LY trial, NSAID users on warfarin or dabigatran had a statistically significant increased risk of major bleeding compared to non-NSAID users (hazard ratio [HR] 1.68; 95% CI, 1.40- 2.02; P < .001).¹³ In the ARISTOTLE trial, patients on warfarin or apixaban who were incident NSAID users were found to have an increased risk of major bleeding (HR 1.61; 95% CI, 1.11-2.33) and clinically relevant nonmajor bleeding (HR 1.70; 95% CI, 1.16- 2.48).¹⁴ These trials found a statistically significant increased bleeding risk associated with NSAID use, though the populations evaluated included patients taking warfarin and patients taking DOACs. These trials did not evaluate the bleeding risk of concomitant NSAID use among DOACs alone.

Evidence on NSAID-associated bleeding risk with DOACs is lacking in settings where the patient population, prescribing practices, and monitoring levels are variable. Within the Veterans Health Administration, clinical pharmacist practitioners (CPPs) in anticoagulation clinics oversee DOAC therapy management. CPPs monitor safety and efficacy of DOAC therapies through a population health management tool, the DOAC Dashboard.¹⁵ The DOAC Dashboard creates alerts for patients who may require an intervention based on certain clinical parameters, such as drug-drug interactions.¹⁶ Whenever a patient on a DOAC is prescribed an NSAID, an alert is generated on the DOAC Dashboard to flag the CPPs for the potential need for an intervention. If NSAID therapy remains clinically indicated, CPPs may recommend risk reduction strategies such as a COX-2 selective NSAID or coprescribing a PPI.¹⁰

The DOAC Dashboard provides an ideal setting for investigating the effects of NSAID use, NSAID selectivity, and PPI coprescribing on DOAC bleeding rates. With an increasing population of patients receiving anticoagulation therapy with a DOAC, more guidance regarding the bleeding risk of concomitant NSAID use with DOACs is needed. Studies evaluating the bleeding risk with concomitant NSAID use in patients on a DOAC alone are limited. This is the first study to date to compare bleeding risk with concomitant NSAID use between DOACs. This study provides information on bleeding risk with NSAID use among commonly prescribed DOACs, rivaroxaban and apixaban, and the potential impacts of current risk reduction strategies.

METHODS

This single-center retrospective cohort review was performed using the electronic health records (EHRs) of patients enrolled in the US Department of Veterans Affairs (VA) Mountain Home Healthcare System who received rivaroxaban or apixaban from December 2020 to December 2022. This study received approval from the East Tennessee State University/VA Institutional Review Board committee.

Patients were identified through the DOAC Dashboard, aged 21 to 100 years, and received rivaroxaban or apixaban at a therapeutic dose: rivaroxaban 10 to 20 mg daily or apixaban 2.5 to 5 mg twice daily. Patients were excluded if they were prescribed dual antiplatelet therapy, received rivaroxaban at dosing indicated for peripheral vascular disease, were undergoing dialysis, had evidence of moderate to severe hepatic impairment or any hepatic disease with coagulopathy, were undergoing chemotherapy or radiation, or had hematological conditions with predisposed bleeding risk. These patients were excluded to mitigate the potential confounding impact from nontherapeutic DOAC dosing strategies and conditions associated with an increased bleeding risk.

Eligible patients were stratified based on NSAID use. NSAID users were defined as patients prescribed an oral NSAID, including both acute and chronic courses, at any point during the study time frame while actively on a DOAC. Bleeding events were reviewed to evaluate rates between rivaroxaban and apixaban among NSAID and nonNSAID users. Identified NSAID users were further assessed for NSAID selectivity and PPI coprescribing as a subgroup analysis for the secondary assessment.

Data Collection

Baseline data were collected, including age, body mass index, anticoagulation indication, DOAC agent, DOAC dose, and DOAC total daily dose. Baseline serum creatinine levels, liver function tests, hemoglobin levels, and platelet counts were collected from the most recent data available immediately prior to the bleeding event, if applicable.

The DOAC Dashboard was reviewed for active and dismissed drug interaction alerts to identify patients taking rivaroxaban or apixaban who were prescribed an NSAID. Patients were categorized in the NSAID group if an interacting drug alert with an NSAID was reported during the study time frame. Data available through the interacting drug alerts on NSAID use were limited to the interacting drug name and date of the reported flag. Manual EHR review was required to confirm dates of NSAID therapy initiation and NSAID discontinuation, if applicable.

Data regarding concomitant antiplatelet use were obtained through review of the active and dismissed drug interaction alerts on the DOAC Dashboard. Concomitant antiplatelet use was defined as the prescribing of a single antiplatelet agent at any point while receiving DOAC therapy. Data on concomitant antiplatelets were collected regardless of NSAID status.

Data on coprescribed PPI therapy were obtained through manual EHR review of identified NSAID users. Coprescribed PPI therapy was defined as the prescribing of a PPI at any point during NSAID therapy. Data regarding PPI use among non-NSAID users were not collected because the secondary endpoint was designed to assess PPI use only among patients coprescribed a DOAC and NSAID.

Outcomes

Bleeding events were identified through an outcomes report generated by the DOAC Dashboard based on International Classification of Diseases, Tenth Revision diagnosis codes associated with a bleeding event. The outcomes report captures diagnoses from the outpatient and inpatient care settings. Reported bleeding events were limited to patients who received a DOAC at any point in the 6 months prior to the event and excluded patients with recent DOAC initiation within 7 days of the event, as these patients are not captured on the DOAC Dashboard.

All reported bleeding events were manually reviewed in the EHR and categorized as a major or clinically relevant nonmajor bleed, according to International Society of Thrombosis and Haemostasis criteria. Validated bleeding events were then crossreferenced with the interacting drug alerts report to identify events with potentially overlapping NSAID therapy at the time of the event. Overlapping NSAID therapy was defined as the prescribing of an NSAID at any point in the 6 months prior to the event. All events with potential overlapping NSAID therapies were manually reviewed for confirmation of NSAID status at the time of the event.

The primary endpoint was a composite of any bleeding event per International Society of Thrombosis and Haemostasis criteria. The secondary endpoint evaluated the potential impact of NSAID selectivity or PPI coprescribing on the bleeding rate among the NSAID user groups.

Statistical Analysis

Analyses were performed consistent with the methods used in the ARISTOTLE and RE-LY trials. It was determined that a sample size of 504 patients, with ≥ 168 patients in each group, would provide 80% power using a 2-sided a of 0.05. HRs with 95% CIs and respective P values were calculated using a SPSS-adapted online calculator.

RESULTS

The DOAC Dashboard identified 681 patients on rivaroxaban and 3225 patients on apixaban; 72 patients on rivaroxaban (10.6%) and 300 patients on apixaban (9.3%) were NSAID users. The mean age of NSAID users was 66.9 years in the rivaroxaban group and 72.4 years in the apixaban group. The mean age of non-NSAID users was 71.5 years in the rivaroxaban group and 75.6 years in the apixaban group. No appreciable differences were observed among subgroups in body mass index, renal function, hepatic function, hemoglobin, or platelet counts, and no statistically significant differences were identified (Table 1). Antiplatelet agents identified included aspirin, clopidogrel, prasugrel, and ticagrelor. Fifteen patients (20.3%) in the rivaroxaban group and 87 patients (28.7%) in the apixaban group had concomitant antiplatelet and NSAID use. Forty-five patients on rivaroxaban (60.8%) and 170 (55.9%) on apixaban were prescribed concomitant PPI and NSAID at baseline. Among non-NSAID users, there was concomitant antiplatelet use for 265 patients (43.6%) in the rivaroxaban group and 1401 patients (47.9%) in the apixaban group. Concomitant PPI use was identified among 63 patients (60.0%) taking selective NSAIDs and 182 (57.2%) taking nonselective NSAIDs.

A total of 423 courses of NSAIDs were identified: 85 NSAID courses in the rivaroxaban group and 338 NSAID courses in the apixaban group. Most NSAID courses involved a nonselective NSAID in the rivaroxaban and apixaban NSAID user groups: 75.2% (n = 318) aggregately compared to 71.8% (n = 61) and 76.0% (n = 257) in the rivaroxaban and apixaban groups, respectively. The most frequent NSAID courses identified were meloxicam (26.7%; n = 113), celecoxib (24.8%; n = 105), ibuprofen (19.1%; n = 81), and naproxen (13.5%; n = 57). Data regarding NSAID therapy initiation and discontinuation dates were not readily available. As a result, the duration of NSAID courses was not captured.

There was no statistically significant difference in bleeding rates between rivaroxaban and apixaban among NSAID users (HR 1.04; 95% CI, 0.98-1.12) or non-NSAID users (HR 1.15; 95% CI, 0.80-1.66) (Table 2). Apixaban non-NSAID users had a higher rate of major bleeds (HR 0.32; 95% CI, 0.17-0.61) while rivaroxaban non-NSAID users had a higher rate of clinically relevant nonmajor bleeds (HR 1.63; 95% CI, 1.10-2.54).

The sample size for the secondary endpoint consisted of bleeding events that were confirmed to have had an overlapping NSAID prescribed at the time of the event. For this secondary assessment, there was 1 rivaroxaban NSAID user bleeding event and 4 apixaban NSAID user bleeding events. For the rivaroxaban NSAID user bleeding event, the NSAID was nonselective and a PPI was not coprescribed. For the apixaban NSAID user bleeding events, 2 NSAIDs were nonselective and 2 were selective. All patients with apixaban and NSAID bleeding events had a coprescribed PPI. There was no clinically significant difference in the bleeding rates observed for NSAID selectivity or PPI coprescribing among the NSAID user subgroups.

DISCUSSION

This study found that there was no statistically significant difference for bleeding rates of major and nonmajor bleeding events between rivaroxaban and apixaban among NSAID users and non-NSAID users. This study did not identify a clinically significant impact on bleeding rates from NSAID selectivity or PPI coprescribing among the NSAID users.

There were notable but not statistically significant differences in baseline characteristics observed between the NSAID and non-NSAID user groups. On average, the rivaroxaban and apixaban NSAID users were younger compared with those not taking NSAIDs. NSAIDs, specifically nonselective NSAIDs, are recognized as potentially inappropriate medications for older adults given that this population is at an increased risk for GI ulcer development and/or GI bleeding.¹⁷ The non-NSAID user group likely consisted of older patients compared to the NSAID user group as clinicians may avoid prescribing NSAIDs to older adults regardless of concomitant DOAC therapy.

In addition to having an older patient population, non-NSAID users were more frequently prescribed a concomitant antiplatelet when compared with NSAID users. This prescribing pattern may be due to clinicians avoiding the use of NSAIDs in patients receiving DOAC therapy in combination with antiplatelet therapy, as these patients have been found to have an increased bleeding rate compared to DOAC therapy alone.¹⁸

Non-NSAID users had an overall higher bleeding rate for both major and nonmajor bleeding events. Based on this observation, it could be hypothesized that antiplatelet agents have a higher risk of bleeding in comparison to NSAIDs. In a subanalysis of the EXPAND study evaluating risk factors of major bleeding in patients receiving rivaroxaban, concomitant use of antiplatelet agents demonstrated a statistically significant increased risk of bleeding (HR 1.6; 95% CI, 1.2-2.3; P = .003) while concomitant use of NSAIDs did not (HR 0.8; 95% CI, 0.3-2.2; P = .67).¹⁹

In assessing PPI status at baseline, a majority of both rivaroxaban and apixaban NSAID users were coprescribed a PPI. This trend aligns with current clinical guideline recommendations for the prescribing of PPI therapy for GI protection in high-risk patients, such as those on DOAC therapy and concomitant NSAID therapy.¹⁰ Given the high proportion of NSAID users coprescribed a PPI at baseline, it may be possible that the true incidence of NSAID-associated bleeding events was higher than what this study found. This observation may reflect the impact from timely implementation of risk mitigation strategies by CPPs in the anticoagulation clinic. However, this study was not constructed to assess the efficacy of PPI use in this manner.

It is important to note the patients included in this study were followed by a pharmacist in an anticoagulation clinic using the DOAC Dashboard.¹⁵ This population management tool allows CPPs to make proactive interventions when a patient taking a DOAC receives an NSAID prescription, such as recommending the coprescribing of a PPI or use of a selective NSAID.^10,16 These standards of care may have contributed to an overall reduced bleeding rate among the NSAID user group and may not be reflective of private practice.

The planned analysis of this study was modeled after the posthoc analysis of the RE-LY and ARISTOTLE trials. Both trials demonstrated an increased risk of bleeding with oral anticoagulation, including DOAC and warfarin, in combination with NSAID use. However, both trials found that NSAID use in patients treated with a DOAC was not independently associated with increased bleeding events compared with warfarin.^13,14 The results of this study are comparable to the RE-LY and ARISTOTLE findings that NSAID use among patients treated with rivaroxaban or apixaban did not demonstrate a statistically significant increased bleeding risk.

Studies of NSAID use in combination with DOAC therapy have been limited to patient populations consisting of both DOAC and warfarin. Evidence from these trials outlines the increased bleeding risk associated with NSAID use in combination with oral anticoagulation; however, these patient populations include those on a DOAC and warfarin.^13,14,19,20 Given the limited evidence on NSAID use among DOACs alone, it is assumed NSAID use in combination with DOACs has a similar risk of bleeding as warfarin use. This may cause clinicians to automatically exclude NSAID therapy as a treatment option for patients on a DOAC who are otherwise clinically appropriate candidates, such as those with underlying inflammatory conditions. Avoiding NSAID therapy in this patient population may lead to suboptimal pain management and increase the risk of patient harm from methods such as inappropriate opioid therapy prescribing.

DOAC therapy should not be a universal limitation to the use of NSAIDs. Although the risk of bleeding with NSAID therapy is always present, deliberate NSAID prescribing in addition to the timely implementation of risk mitigation strategies may provide an avenue for safe NSAID prescribing in patients receiving a DOAC. A population health-based approach to DOAC management, such as the DOAC Dashboard, appears to be effective at preventing patient harm when NSAIDs are prescribed in conjunction with DOACs.

Limitations

The DOAC Dashboard has been shown to be effective and efficient at monitoring DOAC therapy from a population-based approach.¹⁶ Reports generated through the DOAC Dashboard provide convenient access to patient data which allows for timely interventions; however, there are limits to its use for data collection. All the data elements necessary to properly assess bleeding risk with validated tools, such as HAS-BLED (hypertension, abnormal renal/liver function, stroke, bleeding history or predisposition, labile international normalized ratio, elderly, drugs/ alcohol concomitantly), are not available on DOAC Dashboard reports. Due to this constraint, bleeding risk assessments were not conducted at baseline and this study was unable to include risk modeling. Additionally, data elements like initiation and discontinuation dates and duration of therapies were not readily available. As a result, this study was unable to incorporate time as a data point.

This was a retrospective study that relied on manual review of chart documentation to verify bleeding events, but data obtained through the DOAC Dashboard were transferred directly from the EHR.¹⁵ Bleeding events available for evaluation were restricted to those that occurred at a VA facility. Additionally, the sample size within the rivaroxaban NSAID user group did not reach the predefined sample size required to reach power and may have been too small to detect a difference if one did exist. The secondary assessment had a low sample size of NSAID user bleeding events, making it difficult to fully assess its impact on NSAID selectivity and PPI coprescribing on bleeding rates. All courses of NSAIDs were equally valued regardless of the dose or therapy duration; however, this is consistent with how NSAID use was defined in the RE-LY and ARISTOTLE trials.

CONCLUSIONS

This retrospective cohort review found no statistically significant difference in the composite bleeding rates between rivaroxaban and apixaban among NSAID users and non-NSAID users. Moreover, there was no clinically significant impact observed for bleeding rates in regard to NSAID selectivity and PPI coprescribing among NSAID users. However, coprescribing of PPI therapy to patients on a DOAC who are clinically indicated for an NSAID may reduce the risk of bleeding. Population health management tools, such as the DOAC Dashboard, may also allow clinicians to safely prescribe NSAIDs to patients on a DOAC. Further large-scale observational studies are needed to quantify the real-world risk of bleeding with concomitant NSAID use among DOACs alone and to evaluate the impact from NSAID selectivity or PPI coprescribing.

Clinical practice has shifted from vitamin K antagonists to direct oral anticoagulants (DOACs) for atrial fibrillation treatment due to their more favorable risk-benefit profile and less lifestyle modification required.^1,2 However, the advantage of a lower bleeding risk with DOACs could be compromised by potentially problematic pharmacokinetic interactions like those conferred by antiplatelets or nonsteroidal anti-inflammatory drugs (NSAIDs).^3,4 Treating a patient needing anticoagulation with a DOAC who has comorbidities may introduce unavoidable drug-drug interactions. This particularly happens with over-the-counter and prescription NSAIDs used for the management of pain and inflammatory conditions.⁵

NSAIDs primarily affect 2 cyclooxygenase (COX) enzyme isomers, COX-1 and COX-2.⁶ COX-1 helps maintain gastrointestinal (GI) mucosa integrity and platelet aggregation processes, whereas COX-2 is engaged in pain signaling and inflammation mediation. COX-1 inhibition is associated with more bleeding-related adverse events (AEs), especially in the GI tract. COX-2 inhibition is thought to provide analgesia and anti-inflammatory properties without elevating bleeding risk. This premise is responsible for the preferential use of celecoxib, a COX-2 selective NSAID, which should confer a lower bleeding risk compared to nonselective NSAIDs such as ibuprofen and naproxen.⁷ NSAIDs have been documented as independent risk factors for bleeding. NSAID users are about 3 times as likely to develop GI AEs compared to nonNSAID users.⁸

Many clinicians aim to further mitigate NSAID-associated bleeding risk by coprescribing a proton pump inhibitor (PPI). PPIs provide gastroprotection against NSAID-induced mucosal injury and sequential complication of GI bleeding. In a multicenter randomized control trial, patients who received concomitant PPI therapy while undergoing chronic NSAID therapy—including nonselective and COX-2 selective NSAIDs—had a significantly lower risk of GI ulcer development (placebo, 17.0%; 20 mg esomeprazole, 5.2%; 40 mg esomeprazole, 4.6%).⁹ Current clinical guidelines for preventing NSAIDassociated bleeding complications recommend using a COX-2 selective NSAID in combination with PPI therapy for patients at high risk for GI-related bleeding, including the concomitant use of anticoagulants.¹⁰

There is evidence suggesting an increased bleeding risk with NSAIDs when used in combination with vitamin K antagonists such as warfarin.^11,12 A systematic review of warfarin and concomitant NSAID use found an increased risk of overall bleeding with NSAID use in combination with warfarin (odds ratio 1.58; 95% CI, 1.18-2.12), compared to warfarin alone.¹²

Posthoc analyses of randomized clinical trials have also demonstrated an increased bleeding risk with oral anticoagulation and concomitant NSAID use.^13,14 In the RE-LY trial, NSAID users on warfarin or dabigatran had a statistically significant increased risk of major bleeding compared to non-NSAID users (hazard ratio [HR] 1.68; 95% CI, 1.40- 2.02; P < .001).¹³ In the ARISTOTLE trial, patients on warfarin or apixaban who were incident NSAID users were found to have an increased risk of major bleeding (HR 1.61; 95% CI, 1.11-2.33) and clinically relevant nonmajor bleeding (HR 1.70; 95% CI, 1.16- 2.48).¹⁴ These trials found a statistically significant increased bleeding risk associated with NSAID use, though the populations evaluated included patients taking warfarin and patients taking DOACs. These trials did not evaluate the bleeding risk of concomitant NSAID use among DOACs alone.

Evidence on NSAID-associated bleeding risk with DOACs is lacking in settings where the patient population, prescribing practices, and monitoring levels are variable. Within the Veterans Health Administration, clinical pharmacist practitioners (CPPs) in anticoagulation clinics oversee DOAC therapy management. CPPs monitor safety and efficacy of DOAC therapies through a population health management tool, the DOAC Dashboard.¹⁵ The DOAC Dashboard creates alerts for patients who may require an intervention based on certain clinical parameters, such as drug-drug interactions.¹⁶ Whenever a patient on a DOAC is prescribed an NSAID, an alert is generated on the DOAC Dashboard to flag the CPPs for the potential need for an intervention. If NSAID therapy remains clinically indicated, CPPs may recommend risk reduction strategies such as a COX-2 selective NSAID or coprescribing a PPI.¹⁰

The DOAC Dashboard provides an ideal setting for investigating the effects of NSAID use, NSAID selectivity, and PPI coprescribing on DOAC bleeding rates. With an increasing population of patients receiving anticoagulation therapy with a DOAC, more guidance regarding the bleeding risk of concomitant NSAID use with DOACs is needed. Studies evaluating the bleeding risk with concomitant NSAID use in patients on a DOAC alone are limited. This is the first study to date to compare bleeding risk with concomitant NSAID use between DOACs. This study provides information on bleeding risk with NSAID use among commonly prescribed DOACs, rivaroxaban and apixaban, and the potential impacts of current risk reduction strategies.

METHODS

This single-center retrospective cohort review was performed using the electronic health records (EHRs) of patients enrolled in the US Department of Veterans Affairs (VA) Mountain Home Healthcare System who received rivaroxaban or apixaban from December 2020 to December 2022. This study received approval from the East Tennessee State University/VA Institutional Review Board committee.

Patients were identified through the DOAC Dashboard, aged 21 to 100 years, and received rivaroxaban or apixaban at a therapeutic dose: rivaroxaban 10 to 20 mg daily or apixaban 2.5 to 5 mg twice daily. Patients were excluded if they were prescribed dual antiplatelet therapy, received rivaroxaban at dosing indicated for peripheral vascular disease, were undergoing dialysis, had evidence of moderate to severe hepatic impairment or any hepatic disease with coagulopathy, were undergoing chemotherapy or radiation, or had hematological conditions with predisposed bleeding risk. These patients were excluded to mitigate the potential confounding impact from nontherapeutic DOAC dosing strategies and conditions associated with an increased bleeding risk.

Eligible patients were stratified based on NSAID use. NSAID users were defined as patients prescribed an oral NSAID, including both acute and chronic courses, at any point during the study time frame while actively on a DOAC. Bleeding events were reviewed to evaluate rates between rivaroxaban and apixaban among NSAID and nonNSAID users. Identified NSAID users were further assessed for NSAID selectivity and PPI coprescribing as a subgroup analysis for the secondary assessment.

Data Collection

Baseline data were collected, including age, body mass index, anticoagulation indication, DOAC agent, DOAC dose, and DOAC total daily dose. Baseline serum creatinine levels, liver function tests, hemoglobin levels, and platelet counts were collected from the most recent data available immediately prior to the bleeding event, if applicable.

The DOAC Dashboard was reviewed for active and dismissed drug interaction alerts to identify patients taking rivaroxaban or apixaban who were prescribed an NSAID. Patients were categorized in the NSAID group if an interacting drug alert with an NSAID was reported during the study time frame. Data available through the interacting drug alerts on NSAID use were limited to the interacting drug name and date of the reported flag. Manual EHR review was required to confirm dates of NSAID therapy initiation and NSAID discontinuation, if applicable.

Data regarding concomitant antiplatelet use were obtained through review of the active and dismissed drug interaction alerts on the DOAC Dashboard. Concomitant antiplatelet use was defined as the prescribing of a single antiplatelet agent at any point while receiving DOAC therapy. Data on concomitant antiplatelets were collected regardless of NSAID status.

Data on coprescribed PPI therapy were obtained through manual EHR review of identified NSAID users. Coprescribed PPI therapy was defined as the prescribing of a PPI at any point during NSAID therapy. Data regarding PPI use among non-NSAID users were not collected because the secondary endpoint was designed to assess PPI use only among patients coprescribed a DOAC and NSAID.

Outcomes

Bleeding events were identified through an outcomes report generated by the DOAC Dashboard based on International Classification of Diseases, Tenth Revision diagnosis codes associated with a bleeding event. The outcomes report captures diagnoses from the outpatient and inpatient care settings. Reported bleeding events were limited to patients who received a DOAC at any point in the 6 months prior to the event and excluded patients with recent DOAC initiation within 7 days of the event, as these patients are not captured on the DOAC Dashboard.

All reported bleeding events were manually reviewed in the EHR and categorized as a major or clinically relevant nonmajor bleed, according to International Society of Thrombosis and Haemostasis criteria. Validated bleeding events were then crossreferenced with the interacting drug alerts report to identify events with potentially overlapping NSAID therapy at the time of the event. Overlapping NSAID therapy was defined as the prescribing of an NSAID at any point in the 6 months prior to the event. All events with potential overlapping NSAID therapies were manually reviewed for confirmation of NSAID status at the time of the event.

The primary endpoint was a composite of any bleeding event per International Society of Thrombosis and Haemostasis criteria. The secondary endpoint evaluated the potential impact of NSAID selectivity or PPI coprescribing on the bleeding rate among the NSAID user groups.

Statistical Analysis

Analyses were performed consistent with the methods used in the ARISTOTLE and RE-LY trials. It was determined that a sample size of 504 patients, with ≥ 168 patients in each group, would provide 80% power using a 2-sided a of 0.05. HRs with 95% CIs and respective P values were calculated using a SPSS-adapted online calculator.

RESULTS

The DOAC Dashboard identified 681 patients on rivaroxaban and 3225 patients on apixaban; 72 patients on rivaroxaban (10.6%) and 300 patients on apixaban (9.3%) were NSAID users. The mean age of NSAID users was 66.9 years in the rivaroxaban group and 72.4 years in the apixaban group. The mean age of non-NSAID users was 71.5 years in the rivaroxaban group and 75.6 years in the apixaban group. No appreciable differences were observed among subgroups in body mass index, renal function, hepatic function, hemoglobin, or platelet counts, and no statistically significant differences were identified (Table 1). Antiplatelet agents identified included aspirin, clopidogrel, prasugrel, and ticagrelor. Fifteen patients (20.3%) in the rivaroxaban group and 87 patients (28.7%) in the apixaban group had concomitant antiplatelet and NSAID use. Forty-five patients on rivaroxaban (60.8%) and 170 (55.9%) on apixaban were prescribed concomitant PPI and NSAID at baseline. Among non-NSAID users, there was concomitant antiplatelet use for 265 patients (43.6%) in the rivaroxaban group and 1401 patients (47.9%) in the apixaban group. Concomitant PPI use was identified among 63 patients (60.0%) taking selective NSAIDs and 182 (57.2%) taking nonselective NSAIDs.

A total of 423 courses of NSAIDs were identified: 85 NSAID courses in the rivaroxaban group and 338 NSAID courses in the apixaban group. Most NSAID courses involved a nonselective NSAID in the rivaroxaban and apixaban NSAID user groups: 75.2% (n = 318) aggregately compared to 71.8% (n = 61) and 76.0% (n = 257) in the rivaroxaban and apixaban groups, respectively. The most frequent NSAID courses identified were meloxicam (26.7%; n = 113), celecoxib (24.8%; n = 105), ibuprofen (19.1%; n = 81), and naproxen (13.5%; n = 57). Data regarding NSAID therapy initiation and discontinuation dates were not readily available. As a result, the duration of NSAID courses was not captured.

There was no statistically significant difference in bleeding rates between rivaroxaban and apixaban among NSAID users (HR 1.04; 95% CI, 0.98-1.12) or non-NSAID users (HR 1.15; 95% CI, 0.80-1.66) (Table 2). Apixaban non-NSAID users had a higher rate of major bleeds (HR 0.32; 95% CI, 0.17-0.61) while rivaroxaban non-NSAID users had a higher rate of clinically relevant nonmajor bleeds (HR 1.63; 95% CI, 1.10-2.54).

The sample size for the secondary endpoint consisted of bleeding events that were confirmed to have had an overlapping NSAID prescribed at the time of the event. For this secondary assessment, there was 1 rivaroxaban NSAID user bleeding event and 4 apixaban NSAID user bleeding events. For the rivaroxaban NSAID user bleeding event, the NSAID was nonselective and a PPI was not coprescribed. For the apixaban NSAID user bleeding events, 2 NSAIDs were nonselective and 2 were selective. All patients with apixaban and NSAID bleeding events had a coprescribed PPI. There was no clinically significant difference in the bleeding rates observed for NSAID selectivity or PPI coprescribing among the NSAID user subgroups.

DISCUSSION

This study found that there was no statistically significant difference for bleeding rates of major and nonmajor bleeding events between rivaroxaban and apixaban among NSAID users and non-NSAID users. This study did not identify a clinically significant impact on bleeding rates from NSAID selectivity or PPI coprescribing among the NSAID users.

There were notable but not statistically significant differences in baseline characteristics observed between the NSAID and non-NSAID user groups. On average, the rivaroxaban and apixaban NSAID users were younger compared with those not taking NSAIDs. NSAIDs, specifically nonselective NSAIDs, are recognized as potentially inappropriate medications for older adults given that this population is at an increased risk for GI ulcer development and/or GI bleeding.¹⁷ The non-NSAID user group likely consisted of older patients compared to the NSAID user group as clinicians may avoid prescribing NSAIDs to older adults regardless of concomitant DOAC therapy.

In addition to having an older patient population, non-NSAID users were more frequently prescribed a concomitant antiplatelet when compared with NSAID users. This prescribing pattern may be due to clinicians avoiding the use of NSAIDs in patients receiving DOAC therapy in combination with antiplatelet therapy, as these patients have been found to have an increased bleeding rate compared to DOAC therapy alone.¹⁸

Non-NSAID users had an overall higher bleeding rate for both major and nonmajor bleeding events. Based on this observation, it could be hypothesized that antiplatelet agents have a higher risk of bleeding in comparison to NSAIDs. In a subanalysis of the EXPAND study evaluating risk factors of major bleeding in patients receiving rivaroxaban, concomitant use of antiplatelet agents demonstrated a statistically significant increased risk of bleeding (HR 1.6; 95% CI, 1.2-2.3; P = .003) while concomitant use of NSAIDs did not (HR 0.8; 95% CI, 0.3-2.2; P = .67).¹⁹

In assessing PPI status at baseline, a majority of both rivaroxaban and apixaban NSAID users were coprescribed a PPI. This trend aligns with current clinical guideline recommendations for the prescribing of PPI therapy for GI protection in high-risk patients, such as those on DOAC therapy and concomitant NSAID therapy.¹⁰ Given the high proportion of NSAID users coprescribed a PPI at baseline, it may be possible that the true incidence of NSAID-associated bleeding events was higher than what this study found. This observation may reflect the impact from timely implementation of risk mitigation strategies by CPPs in the anticoagulation clinic. However, this study was not constructed to assess the efficacy of PPI use in this manner.

It is important to note the patients included in this study were followed by a pharmacist in an anticoagulation clinic using the DOAC Dashboard.¹⁵ This population management tool allows CPPs to make proactive interventions when a patient taking a DOAC receives an NSAID prescription, such as recommending the coprescribing of a PPI or use of a selective NSAID.^10,16 These standards of care may have contributed to an overall reduced bleeding rate among the NSAID user group and may not be reflective of private practice.

The planned analysis of this study was modeled after the posthoc analysis of the RE-LY and ARISTOTLE trials. Both trials demonstrated an increased risk of bleeding with oral anticoagulation, including DOAC and warfarin, in combination with NSAID use. However, both trials found that NSAID use in patients treated with a DOAC was not independently associated with increased bleeding events compared with warfarin.^13,14 The results of this study are comparable to the RE-LY and ARISTOTLE findings that NSAID use among patients treated with rivaroxaban or apixaban did not demonstrate a statistically significant increased bleeding risk.

Studies of NSAID use in combination with DOAC therapy have been limited to patient populations consisting of both DOAC and warfarin. Evidence from these trials outlines the increased bleeding risk associated with NSAID use in combination with oral anticoagulation; however, these patient populations include those on a DOAC and warfarin.^13,14,19,20 Given the limited evidence on NSAID use among DOACs alone, it is assumed NSAID use in combination with DOACs has a similar risk of bleeding as warfarin use. This may cause clinicians to automatically exclude NSAID therapy as a treatment option for patients on a DOAC who are otherwise clinically appropriate candidates, such as those with underlying inflammatory conditions. Avoiding NSAID therapy in this patient population may lead to suboptimal pain management and increase the risk of patient harm from methods such as inappropriate opioid therapy prescribing.

DOAC therapy should not be a universal limitation to the use of NSAIDs. Although the risk of bleeding with NSAID therapy is always present, deliberate NSAID prescribing in addition to the timely implementation of risk mitigation strategies may provide an avenue for safe NSAID prescribing in patients receiving a DOAC. A population health-based approach to DOAC management, such as the DOAC Dashboard, appears to be effective at preventing patient harm when NSAIDs are prescribed in conjunction with DOACs.

Limitations

The DOAC Dashboard has been shown to be effective and efficient at monitoring DOAC therapy from a population-based approach.¹⁶ Reports generated through the DOAC Dashboard provide convenient access to patient data which allows for timely interventions; however, there are limits to its use for data collection. All the data elements necessary to properly assess bleeding risk with validated tools, such as HAS-BLED (hypertension, abnormal renal/liver function, stroke, bleeding history or predisposition, labile international normalized ratio, elderly, drugs/ alcohol concomitantly), are not available on DOAC Dashboard reports. Due to this constraint, bleeding risk assessments were not conducted at baseline and this study was unable to include risk modeling. Additionally, data elements like initiation and discontinuation dates and duration of therapies were not readily available. As a result, this study was unable to incorporate time as a data point.

This was a retrospective study that relied on manual review of chart documentation to verify bleeding events, but data obtained through the DOAC Dashboard were transferred directly from the EHR.¹⁵ Bleeding events available for evaluation were restricted to those that occurred at a VA facility. Additionally, the sample size within the rivaroxaban NSAID user group did not reach the predefined sample size required to reach power and may have been too small to detect a difference if one did exist. The secondary assessment had a low sample size of NSAID user bleeding events, making it difficult to fully assess its impact on NSAID selectivity and PPI coprescribing on bleeding rates. All courses of NSAIDs were equally valued regardless of the dose or therapy duration; however, this is consistent with how NSAID use was defined in the RE-LY and ARISTOTLE trials.

CONCLUSIONS

This retrospective cohort review found no statistically significant difference in the composite bleeding rates between rivaroxaban and apixaban among NSAID users and non-NSAID users. Moreover, there was no clinically significant impact observed for bleeding rates in regard to NSAID selectivity and PPI coprescribing among NSAID users. However, coprescribing of PPI therapy to patients on a DOAC who are clinically indicated for an NSAID may reduce the risk of bleeding. Population health management tools, such as the DOAC Dashboard, may also allow clinicians to safely prescribe NSAIDs to patients on a DOAC. Further large-scale observational studies are needed to quantify the real-world risk of bleeding with concomitant NSAID use among DOACs alone and to evaluate the impact from NSAID selectivity or PPI coprescribing.

Clinical practice has shifted from vitamin K antagonists to direct oral anticoagulants (DOACs) for atrial fibrillation treatment due to their more favorable risk-benefit profile and less lifestyle modification required.^1,2 However, the advantage of a lower bleeding risk with DOACs could be compromised by potentially problematic pharmacokinetic interactions like those conferred by antiplatelets or nonsteroidal anti-inflammatory drugs (NSAIDs).^3,4 Treating a patient needing anticoagulation with a DOAC who has comorbidities may introduce unavoidable drug-drug interactions. This particularly happens with over-the-counter and prescription NSAIDs used for the management of pain and inflammatory conditions.⁵

NSAIDs primarily affect 2 cyclooxygenase (COX) enzyme isomers, COX-1 and COX-2.⁶ COX-1 helps maintain gastrointestinal (GI) mucosa integrity and platelet aggregation processes, whereas COX-2 is engaged in pain signaling and inflammation mediation. COX-1 inhibition is associated with more bleeding-related adverse events (AEs), especially in the GI tract. COX-2 inhibition is thought to provide analgesia and anti-inflammatory properties without elevating bleeding risk. This premise is responsible for the preferential use of celecoxib, a COX-2 selective NSAID, which should confer a lower bleeding risk compared to nonselective NSAIDs such as ibuprofen and naproxen.⁷ NSAIDs have been documented as independent risk factors for bleeding. NSAID users are about 3 times as likely to develop GI AEs compared to nonNSAID users.⁸

Many clinicians aim to further mitigate NSAID-associated bleeding risk by coprescribing a proton pump inhibitor (PPI). PPIs provide gastroprotection against NSAID-induced mucosal injury and sequential complication of GI bleeding. In a multicenter randomized control trial, patients who received concomitant PPI therapy while undergoing chronic NSAID therapy—including nonselective and COX-2 selective NSAIDs—had a significantly lower risk of GI ulcer development (placebo, 17.0%; 20 mg esomeprazole, 5.2%; 40 mg esomeprazole, 4.6%).⁹ Current clinical guidelines for preventing NSAIDassociated bleeding complications recommend using a COX-2 selective NSAID in combination with PPI therapy for patients at high risk for GI-related bleeding, including the concomitant use of anticoagulants.¹⁰

There is evidence suggesting an increased bleeding risk with NSAIDs when used in combination with vitamin K antagonists such as warfarin.^11,12 A systematic review of warfarin and concomitant NSAID use found an increased risk of overall bleeding with NSAID use in combination with warfarin (odds ratio 1.58; 95% CI, 1.18-2.12), compared to warfarin alone.¹²

Posthoc analyses of randomized clinical trials have also demonstrated an increased bleeding risk with oral anticoagulation and concomitant NSAID use.^13,14 In the RE-LY trial, NSAID users on warfarin or dabigatran had a statistically significant increased risk of major bleeding compared to non-NSAID users (hazard ratio [HR] 1.68; 95% CI, 1.40- 2.02; P < .001).¹³ In the ARISTOTLE trial, patients on warfarin or apixaban who were incident NSAID users were found to have an increased risk of major bleeding (HR 1.61; 95% CI, 1.11-2.33) and clinically relevant nonmajor bleeding (HR 1.70; 95% CI, 1.16- 2.48).¹⁴ These trials found a statistically significant increased bleeding risk associated with NSAID use, though the populations evaluated included patients taking warfarin and patients taking DOACs. These trials did not evaluate the bleeding risk of concomitant NSAID use among DOACs alone.

Evidence on NSAID-associated bleeding risk with DOACs is lacking in settings where the patient population, prescribing practices, and monitoring levels are variable. Within the Veterans Health Administration, clinical pharmacist practitioners (CPPs) in anticoagulation clinics oversee DOAC therapy management. CPPs monitor safety and efficacy of DOAC therapies through a population health management tool, the DOAC Dashboard.¹⁵ The DOAC Dashboard creates alerts for patients who may require an intervention based on certain clinical parameters, such as drug-drug interactions.¹⁶ Whenever a patient on a DOAC is prescribed an NSAID, an alert is generated on the DOAC Dashboard to flag the CPPs for the potential need for an intervention. If NSAID therapy remains clinically indicated, CPPs may recommend risk reduction strategies such as a COX-2 selective NSAID or coprescribing a PPI.¹⁰

The DOAC Dashboard provides an ideal setting for investigating the effects of NSAID use, NSAID selectivity, and PPI coprescribing on DOAC bleeding rates. With an increasing population of patients receiving anticoagulation therapy with a DOAC, more guidance regarding the bleeding risk of concomitant NSAID use with DOACs is needed. Studies evaluating the bleeding risk with concomitant NSAID use in patients on a DOAC alone are limited. This is the first study to date to compare bleeding risk with concomitant NSAID use between DOACs. This study provides information on bleeding risk with NSAID use among commonly prescribed DOACs, rivaroxaban and apixaban, and the potential impacts of current risk reduction strategies.

METHODS

This single-center retrospective cohort review was performed using the electronic health records (EHRs) of patients enrolled in the US Department of Veterans Affairs (VA) Mountain Home Healthcare System who received rivaroxaban or apixaban from December 2020 to December 2022. This study received approval from the East Tennessee State University/VA Institutional Review Board committee.

Patients were identified through the DOAC Dashboard, aged 21 to 100 years, and received rivaroxaban or apixaban at a therapeutic dose: rivaroxaban 10 to 20 mg daily or apixaban 2.5 to 5 mg twice daily. Patients were excluded if they were prescribed dual antiplatelet therapy, received rivaroxaban at dosing indicated for peripheral vascular disease, were undergoing dialysis, had evidence of moderate to severe hepatic impairment or any hepatic disease with coagulopathy, were undergoing chemotherapy or radiation, or had hematological conditions with predisposed bleeding risk. These patients were excluded to mitigate the potential confounding impact from nontherapeutic DOAC dosing strategies and conditions associated with an increased bleeding risk.

Eligible patients were stratified based on NSAID use. NSAID users were defined as patients prescribed an oral NSAID, including both acute and chronic courses, at any point during the study time frame while actively on a DOAC. Bleeding events were reviewed to evaluate rates between rivaroxaban and apixaban among NSAID and nonNSAID users. Identified NSAID users were further assessed for NSAID selectivity and PPI coprescribing as a subgroup analysis for the secondary assessment.

Data Collection

Baseline data were collected, including age, body mass index, anticoagulation indication, DOAC agent, DOAC dose, and DOAC total daily dose. Baseline serum creatinine levels, liver function tests, hemoglobin levels, and platelet counts were collected from the most recent data available immediately prior to the bleeding event, if applicable.

The DOAC Dashboard was reviewed for active and dismissed drug interaction alerts to identify patients taking rivaroxaban or apixaban who were prescribed an NSAID. Patients were categorized in the NSAID group if an interacting drug alert with an NSAID was reported during the study time frame. Data available through the interacting drug alerts on NSAID use were limited to the interacting drug name and date of the reported flag. Manual EHR review was required to confirm dates of NSAID therapy initiation and NSAID discontinuation, if applicable.

Data regarding concomitant antiplatelet use were obtained through review of the active and dismissed drug interaction alerts on the DOAC Dashboard. Concomitant antiplatelet use was defined as the prescribing of a single antiplatelet agent at any point while receiving DOAC therapy. Data on concomitant antiplatelets were collected regardless of NSAID status.

Data on coprescribed PPI therapy were obtained through manual EHR review of identified NSAID users. Coprescribed PPI therapy was defined as the prescribing of a PPI at any point during NSAID therapy. Data regarding PPI use among non-NSAID users were not collected because the secondary endpoint was designed to assess PPI use only among patients coprescribed a DOAC and NSAID.

Outcomes

Bleeding events were identified through an outcomes report generated by the DOAC Dashboard based on International Classification of Diseases, Tenth Revision diagnosis codes associated with a bleeding event. The outcomes report captures diagnoses from the outpatient and inpatient care settings. Reported bleeding events were limited to patients who received a DOAC at any point in the 6 months prior to the event and excluded patients with recent DOAC initiation within 7 days of the event, as these patients are not captured on the DOAC Dashboard.

All reported bleeding events were manually reviewed in the EHR and categorized as a major or clinically relevant nonmajor bleed, according to International Society of Thrombosis and Haemostasis criteria. Validated bleeding events were then crossreferenced with the interacting drug alerts report to identify events with potentially overlapping NSAID therapy at the time of the event. Overlapping NSAID therapy was defined as the prescribing of an NSAID at any point in the 6 months prior to the event. All events with potential overlapping NSAID therapies were manually reviewed for confirmation of NSAID status at the time of the event.

The primary endpoint was a composite of any bleeding event per International Society of Thrombosis and Haemostasis criteria. The secondary endpoint evaluated the potential impact of NSAID selectivity or PPI coprescribing on the bleeding rate among the NSAID user groups.

Statistical Analysis

Analyses were performed consistent with the methods used in the ARISTOTLE and RE-LY trials. It was determined that a sample size of 504 patients, with ≥ 168 patients in each group, would provide 80% power using a 2-sided a of 0.05. HRs with 95% CIs and respective P values were calculated using a SPSS-adapted online calculator.

RESULTS

The DOAC Dashboard identified 681 patients on rivaroxaban and 3225 patients on apixaban; 72 patients on rivaroxaban (10.6%) and 300 patients on apixaban (9.3%) were NSAID users. The mean age of NSAID users was 66.9 years in the rivaroxaban group and 72.4 years in the apixaban group. The mean age of non-NSAID users was 71.5 years in the rivaroxaban group and 75.6 years in the apixaban group. No appreciable differences were observed among subgroups in body mass index, renal function, hepatic function, hemoglobin, or platelet counts, and no statistically significant differences were identified (Table 1). Antiplatelet agents identified included aspirin, clopidogrel, prasugrel, and ticagrelor. Fifteen patients (20.3%) in the rivaroxaban group and 87 patients (28.7%) in the apixaban group had concomitant antiplatelet and NSAID use. Forty-five patients on rivaroxaban (60.8%) and 170 (55.9%) on apixaban were prescribed concomitant PPI and NSAID at baseline. Among non-NSAID users, there was concomitant antiplatelet use for 265 patients (43.6%) in the rivaroxaban group and 1401 patients (47.9%) in the apixaban group. Concomitant PPI use was identified among 63 patients (60.0%) taking selective NSAIDs and 182 (57.2%) taking nonselective NSAIDs.

A total of 423 courses of NSAIDs were identified: 85 NSAID courses in the rivaroxaban group and 338 NSAID courses in the apixaban group. Most NSAID courses involved a nonselective NSAID in the rivaroxaban and apixaban NSAID user groups: 75.2% (n = 318) aggregately compared to 71.8% (n = 61) and 76.0% (n = 257) in the rivaroxaban and apixaban groups, respectively. The most frequent NSAID courses identified were meloxicam (26.7%; n = 113), celecoxib (24.8%; n = 105), ibuprofen (19.1%; n = 81), and naproxen (13.5%; n = 57). Data regarding NSAID therapy initiation and discontinuation dates were not readily available. As a result, the duration of NSAID courses was not captured.

There was no statistically significant difference in bleeding rates between rivaroxaban and apixaban among NSAID users (HR 1.04; 95% CI, 0.98-1.12) or non-NSAID users (HR 1.15; 95% CI, 0.80-1.66) (Table 2). Apixaban non-NSAID users had a higher rate of major bleeds (HR 0.32; 95% CI, 0.17-0.61) while rivaroxaban non-NSAID users had a higher rate of clinically relevant nonmajor bleeds (HR 1.63; 95% CI, 1.10-2.54).

The sample size for the secondary endpoint consisted of bleeding events that were confirmed to have had an overlapping NSAID prescribed at the time of the event. For this secondary assessment, there was 1 rivaroxaban NSAID user bleeding event and 4 apixaban NSAID user bleeding events. For the rivaroxaban NSAID user bleeding event, the NSAID was nonselective and a PPI was not coprescribed. For the apixaban NSAID user bleeding events, 2 NSAIDs were nonselective and 2 were selective. All patients with apixaban and NSAID bleeding events had a coprescribed PPI. There was no clinically significant difference in the bleeding rates observed for NSAID selectivity or PPI coprescribing among the NSAID user subgroups.

DISCUSSION

This study found that there was no statistically significant difference for bleeding rates of major and nonmajor bleeding events between rivaroxaban and apixaban among NSAID users and non-NSAID users. This study did not identify a clinically significant impact on bleeding rates from NSAID selectivity or PPI coprescribing among the NSAID users.

There were notable but not statistically significant differences in baseline characteristics observed between the NSAID and non-NSAID user groups. On average, the rivaroxaban and apixaban NSAID users were younger compared with those not taking NSAIDs. NSAIDs, specifically nonselective NSAIDs, are recognized as potentially inappropriate medications for older adults given that this population is at an increased risk for GI ulcer development and/or GI bleeding.¹⁷ The non-NSAID user group likely consisted of older patients compared to the NSAID user group as clinicians may avoid prescribing NSAIDs to older adults regardless of concomitant DOAC therapy.

In addition to having an older patient population, non-NSAID users were more frequently prescribed a concomitant antiplatelet when compared with NSAID users. This prescribing pattern may be due to clinicians avoiding the use of NSAIDs in patients receiving DOAC therapy in combination with antiplatelet therapy, as these patients have been found to have an increased bleeding rate compared to DOAC therapy alone.¹⁸

Non-NSAID users had an overall higher bleeding rate for both major and nonmajor bleeding events. Based on this observation, it could be hypothesized that antiplatelet agents have a higher risk of bleeding in comparison to NSAIDs. In a subanalysis of the EXPAND study evaluating risk factors of major bleeding in patients receiving rivaroxaban, concomitant use of antiplatelet agents demonstrated a statistically significant increased risk of bleeding (HR 1.6; 95% CI, 1.2-2.3; P = .003) while concomitant use of NSAIDs did not (HR 0.8; 95% CI, 0.3-2.2; P = .67).¹⁹

In assessing PPI status at baseline, a majority of both rivaroxaban and apixaban NSAID users were coprescribed a PPI. This trend aligns with current clinical guideline recommendations for the prescribing of PPI therapy for GI protection in high-risk patients, such as those on DOAC therapy and concomitant NSAID therapy.¹⁰ Given the high proportion of NSAID users coprescribed a PPI at baseline, it may be possible that the true incidence of NSAID-associated bleeding events was higher than what this study found. This observation may reflect the impact from timely implementation of risk mitigation strategies by CPPs in the anticoagulation clinic. However, this study was not constructed to assess the efficacy of PPI use in this manner.

It is important to note the patients included in this study were followed by a pharmacist in an anticoagulation clinic using the DOAC Dashboard.¹⁵ This population management tool allows CPPs to make proactive interventions when a patient taking a DOAC receives an NSAID prescription, such as recommending the coprescribing of a PPI or use of a selective NSAID.^10,16 These standards of care may have contributed to an overall reduced bleeding rate among the NSAID user group and may not be reflective of private practice.

The planned analysis of this study was modeled after the posthoc analysis of the RE-LY and ARISTOTLE trials. Both trials demonstrated an increased risk of bleeding with oral anticoagulation, including DOAC and warfarin, in combination with NSAID use. However, both trials found that NSAID use in patients treated with a DOAC was not independently associated with increased bleeding events compared with warfarin.^13,14 The results of this study are comparable to the RE-LY and ARISTOTLE findings that NSAID use among patients treated with rivaroxaban or apixaban did not demonstrate a statistically significant increased bleeding risk.

Studies of NSAID use in combination with DOAC therapy have been limited to patient populations consisting of both DOAC and warfarin. Evidence from these trials outlines the increased bleeding risk associated with NSAID use in combination with oral anticoagulation; however, these patient populations include those on a DOAC and warfarin.^13,14,19,20 Given the limited evidence on NSAID use among DOACs alone, it is assumed NSAID use in combination with DOACs has a similar risk of bleeding as warfarin use. This may cause clinicians to automatically exclude NSAID therapy as a treatment option for patients on a DOAC who are otherwise clinically appropriate candidates, such as those with underlying inflammatory conditions. Avoiding NSAID therapy in this patient population may lead to suboptimal pain management and increase the risk of patient harm from methods such as inappropriate opioid therapy prescribing.

DOAC therapy should not be a universal limitation to the use of NSAIDs. Although the risk of bleeding with NSAID therapy is always present, deliberate NSAID prescribing in addition to the timely implementation of risk mitigation strategies may provide an avenue for safe NSAID prescribing in patients receiving a DOAC. A population health-based approach to DOAC management, such as the DOAC Dashboard, appears to be effective at preventing patient harm when NSAIDs are prescribed in conjunction with DOACs.

Limitations

The DOAC Dashboard has been shown to be effective and efficient at monitoring DOAC therapy from a population-based approach.¹⁶ Reports generated through the DOAC Dashboard provide convenient access to patient data which allows for timely interventions; however, there are limits to its use for data collection. All the data elements necessary to properly assess bleeding risk with validated tools, such as HAS-BLED (hypertension, abnormal renal/liver function, stroke, bleeding history or predisposition, labile international normalized ratio, elderly, drugs/ alcohol concomitantly), are not available on DOAC Dashboard reports. Due to this constraint, bleeding risk assessments were not conducted at baseline and this study was unable to include risk modeling. Additionally, data elements like initiation and discontinuation dates and duration of therapies were not readily available. As a result, this study was unable to incorporate time as a data point.

This was a retrospective study that relied on manual review of chart documentation to verify bleeding events, but data obtained through the DOAC Dashboard were transferred directly from the EHR.¹⁵ Bleeding events available for evaluation were restricted to those that occurred at a VA facility. Additionally, the sample size within the rivaroxaban NSAID user group did not reach the predefined sample size required to reach power and may have been too small to detect a difference if one did exist. The secondary assessment had a low sample size of NSAID user bleeding events, making it difficult to fully assess its impact on NSAID selectivity and PPI coprescribing on bleeding rates. All courses of NSAIDs were equally valued regardless of the dose or therapy duration; however, this is consistent with how NSAID use was defined in the RE-LY and ARISTOTLE trials.

CONCLUSIONS

This retrospective cohort review found no statistically significant difference in the composite bleeding rates between rivaroxaban and apixaban among NSAID users and non-NSAID users. Moreover, there was no clinically significant impact observed for bleeding rates in regard to NSAID selectivity and PPI coprescribing among NSAID users. However, coprescribing of PPI therapy to patients on a DOAC who are clinically indicated for an NSAID may reduce the risk of bleeding. Population health management tools, such as the DOAC Dashboard, may also allow clinicians to safely prescribe NSAIDs to patients on a DOAC. Further large-scale observational studies are needed to quantify the real-world risk of bleeding with concomitant NSAID use among DOACs alone and to evaluate the impact from NSAID selectivity or PPI coprescribing.