Brief Reports

Chronic kidney disease (CKD) affects over 13% of the US population and is associated with increased morbidity, mortality, and healthcare costs.[1] However, only 10% of individuals with CKD are aware of their diagnoses.[2] Even in those with stage 5 CKD, only 60% of individuals are aware of their CKD.[3] To our knowledge, no work has examined CKD awareness in a hospitalized patient population.

Patient awareness of their CKD diagnosis is important because progression of kidney disease can be slowed by patient self‐management of diabetes and hypertension.[4] Patient awareness of CKD may also increase acceptance of preend‐stage renal disease (ESRD) patient education and nephrology referral, which have been shown to delay CKD progression and improve clinical status at dialysis initiation.[5] However, only 60% of patients with advanced CKD have visited a nephrologist in the past year or have seen a nephrologist prior to dialysis initiation.[1]

The hospital is an important site for patient education and linkage to outpatient care for patients with CKD.[6] The hospital serves high‐risk patients who may not be well connected to outpatient care or who have lesswell‐controlled disease.[6, 7] Thus, hospitalization represents an opportunity to identify existing CKD and to use a multidisciplinary approach to preventative care, patient education, and patient‐provider planning for renal replacement therapy needs. In our cross‐sectional study in an urban, minority‐serving hospital, we sought to determine what patient factors were associated with hospitalized patients correctly self‐identifying as having CKD.

METHODS

RESULTS

DISCUSSION

Although prior work has examined CKD awareness in the general population and in high‐risk cohorts,[2, 3, 13] this is the first study examining CKD awareness in an urban, underserved hospitalized population. We found that overall patients' CKD awareness was low (32%), but increased as high as 63% for CKD stage 5, even after controlling for patient demographic, clinical characteristics, and healthcare use. Our overall rate of CKD awareness was higher than prior studies overall and at lower CKD stages.[2, 3, 13] Our work is consistent with prior literature that shows increasing CKD awareness with advancing CKD stage.[2, 3, 13]

Older patients (>80 years) had lower awareness of CKD. Older patients are more likely to have a near normal creatinine, despite a markedly reduced eGFR, so their CKD may go unnoticed.[14] Even with appropriate recognition, providers may also feel like their CKD is unlikely to progress to ESRD, given its stability and/or their competing risk of death.[15] Finally, older hospitalized patients may also be less likely to report a CKD diagnosis due to difficulty in recall due to denial, dementia, or delirium.

One limitation is that our case‐finding for CKD was physician ICD‐9 coding, which is highly specific but not sensitive.[9] The majority of patients with physician‐identified CKD were CKD unspecified, perhaps due to poor coding, physician underdocumentation, or physician under‐recognition of CKD stage. Although only 27% of the CKD unspecified group correctly self‐identified as having CKD, over 75% were found to be CKD stage 3 or higher, which should trigger additional monitoring or care based on guidelines.[10] In addition, despite statistical significance, we may not be able to make meaningful inferences about our small other group (nonwhite, non‐African American). Our sample was from 1 hospitalan urban, academic, tertiary care center with a large proportion of African American patientswhich may limit generalizability. The multivariable model will need to be tested in other populations for reproducibility.

Our study significantly contributes to the literature by examining patient awareness of CKD in a high‐risk, urban, hospitalized minority population. Other study strengths include use of basic demographic information, as well as survey and laboratory data for a richer examination of the associations between patient factors and CKD awareness.

CONCLUSION

Hospitalized patients with CKD have a low CKD awareness. Patient awareness of their CKD is increased with physician documentation of CKD severity. Patient awareness of their CKD must be coupled with provider awareness and CKD documentation to link patients to multidisciplinary CKD education and care to slow CKD progression and reduce associated cardiovascular and metabolic complications. Further work is needed across hospitals to determineand improveCKD awareness among both patients and providers.

Chronic kidney disease (CKD) affects over 13% of the US population and is associated with increased morbidity, mortality, and healthcare costs.[1] However, only 10% of individuals with CKD are aware of their diagnoses.[2] Even in those with stage 5 CKD, only 60% of individuals are aware of their CKD.[3] To our knowledge, no work has examined CKD awareness in a hospitalized patient population.

Patient awareness of their CKD diagnosis is important because progression of kidney disease can be slowed by patient self‐management of diabetes and hypertension.[4] Patient awareness of CKD may also increase acceptance of preend‐stage renal disease (ESRD) patient education and nephrology referral, which have been shown to delay CKD progression and improve clinical status at dialysis initiation.[5] However, only 60% of patients with advanced CKD have visited a nephrologist in the past year or have seen a nephrologist prior to dialysis initiation.[1]

The hospital is an important site for patient education and linkage to outpatient care for patients with CKD.[6] The hospital serves high‐risk patients who may not be well connected to outpatient care or who have lesswell‐controlled disease.[6, 7] Thus, hospitalization represents an opportunity to identify existing CKD and to use a multidisciplinary approach to preventative care, patient education, and patient‐provider planning for renal replacement therapy needs. In our cross‐sectional study in an urban, minority‐serving hospital, we sought to determine what patient factors were associated with hospitalized patients correctly self‐identifying as having CKD.

METHODS

RESULTS

DISCUSSION

Although prior work has examined CKD awareness in the general population and in high‐risk cohorts,[2, 3, 13] this is the first study examining CKD awareness in an urban, underserved hospitalized population. We found that overall patients' CKD awareness was low (32%), but increased as high as 63% for CKD stage 5, even after controlling for patient demographic, clinical characteristics, and healthcare use. Our overall rate of CKD awareness was higher than prior studies overall and at lower CKD stages.[2, 3, 13] Our work is consistent with prior literature that shows increasing CKD awareness with advancing CKD stage.[2, 3, 13]

Older patients (>80 years) had lower awareness of CKD. Older patients are more likely to have a near normal creatinine, despite a markedly reduced eGFR, so their CKD may go unnoticed.[14] Even with appropriate recognition, providers may also feel like their CKD is unlikely to progress to ESRD, given its stability and/or their competing risk of death.[15] Finally, older hospitalized patients may also be less likely to report a CKD diagnosis due to difficulty in recall due to denial, dementia, or delirium.

One limitation is that our case‐finding for CKD was physician ICD‐9 coding, which is highly specific but not sensitive.[9] The majority of patients with physician‐identified CKD were CKD unspecified, perhaps due to poor coding, physician underdocumentation, or physician under‐recognition of CKD stage. Although only 27% of the CKD unspecified group correctly self‐identified as having CKD, over 75% were found to be CKD stage 3 or higher, which should trigger additional monitoring or care based on guidelines.[10] In addition, despite statistical significance, we may not be able to make meaningful inferences about our small other group (nonwhite, non‐African American). Our sample was from 1 hospitalan urban, academic, tertiary care center with a large proportion of African American patientswhich may limit generalizability. The multivariable model will need to be tested in other populations for reproducibility.

Our study significantly contributes to the literature by examining patient awareness of CKD in a high‐risk, urban, hospitalized minority population. Other study strengths include use of basic demographic information, as well as survey and laboratory data for a richer examination of the associations between patient factors and CKD awareness.

CONCLUSION

Hospitalized patients with CKD have a low CKD awareness. Patient awareness of their CKD is increased with physician documentation of CKD severity. Patient awareness of their CKD must be coupled with provider awareness and CKD documentation to link patients to multidisciplinary CKD education and care to slow CKD progression and reduce associated cardiovascular and metabolic complications. Further work is needed across hospitals to determineand improveCKD awareness among both patients and providers.

Chronic kidney disease (CKD) affects over 13% of the US population and is associated with increased morbidity, mortality, and healthcare costs.[1] However, only 10% of individuals with CKD are aware of their diagnoses.[2] Even in those with stage 5 CKD, only 60% of individuals are aware of their CKD.[3] To our knowledge, no work has examined CKD awareness in a hospitalized patient population.

Patient awareness of their CKD diagnosis is important because progression of kidney disease can be slowed by patient self‐management of diabetes and hypertension.[4] Patient awareness of CKD may also increase acceptance of preend‐stage renal disease (ESRD) patient education and nephrology referral, which have been shown to delay CKD progression and improve clinical status at dialysis initiation.[5] However, only 60% of patients with advanced CKD have visited a nephrologist in the past year or have seen a nephrologist prior to dialysis initiation.[1]

The hospital is an important site for patient education and linkage to outpatient care for patients with CKD.[6] The hospital serves high‐risk patients who may not be well connected to outpatient care or who have lesswell‐controlled disease.[6, 7] Thus, hospitalization represents an opportunity to identify existing CKD and to use a multidisciplinary approach to preventative care, patient education, and patient‐provider planning for renal replacement therapy needs. In our cross‐sectional study in an urban, minority‐serving hospital, we sought to determine what patient factors were associated with hospitalized patients correctly self‐identifying as having CKD.

METHODS

RESULTS

DISCUSSION

Although prior work has examined CKD awareness in the general population and in high‐risk cohorts,[2, 3, 13] this is the first study examining CKD awareness in an urban, underserved hospitalized population. We found that overall patients' CKD awareness was low (32%), but increased as high as 63% for CKD stage 5, even after controlling for patient demographic, clinical characteristics, and healthcare use. Our overall rate of CKD awareness was higher than prior studies overall and at lower CKD stages.[2, 3, 13] Our work is consistent with prior literature that shows increasing CKD awareness with advancing CKD stage.[2, 3, 13]

Older patients (>80 years) had lower awareness of CKD. Older patients are more likely to have a near normal creatinine, despite a markedly reduced eGFR, so their CKD may go unnoticed.[14] Even with appropriate recognition, providers may also feel like their CKD is unlikely to progress to ESRD, given its stability and/or their competing risk of death.[15] Finally, older hospitalized patients may also be less likely to report a CKD diagnosis due to difficulty in recall due to denial, dementia, or delirium.

One limitation is that our case‐finding for CKD was physician ICD‐9 coding, which is highly specific but not sensitive.[9] The majority of patients with physician‐identified CKD were CKD unspecified, perhaps due to poor coding, physician underdocumentation, or physician under‐recognition of CKD stage. Although only 27% of the CKD unspecified group correctly self‐identified as having CKD, over 75% were found to be CKD stage 3 or higher, which should trigger additional monitoring or care based on guidelines.[10] In addition, despite statistical significance, we may not be able to make meaningful inferences about our small other group (nonwhite, non‐African American). Our sample was from 1 hospitalan urban, academic, tertiary care center with a large proportion of African American patientswhich may limit generalizability. The multivariable model will need to be tested in other populations for reproducibility.

Our study significantly contributes to the literature by examining patient awareness of CKD in a high‐risk, urban, hospitalized minority population. Other study strengths include use of basic demographic information, as well as survey and laboratory data for a richer examination of the associations between patient factors and CKD awareness.

CONCLUSION

Hospitalized patients with CKD have a low CKD awareness. Patient awareness of their CKD is increased with physician documentation of CKD severity. Patient awareness of their CKD must be coupled with provider awareness and CKD documentation to link patients to multidisciplinary CKD education and care to slow CKD progression and reduce associated cardiovascular and metabolic complications. Further work is needed across hospitals to determineand improveCKD awareness among both patients and providers.

Cirrhosis is a leading cause of death in the United States. In 2010, cirrhosis resulted in an estimated 49,500 deaths, which represented a significant increase from 35,500 deaths 2 decades ago.[1] Cirrhotic patients are susceptible to numerous disease‐specific complications including ascites, esophageal varices, hepatic encephalopathy (HE), and hepatorenal syndrome (HRS).[2]

Esophageal varices develop in approximately 50% of patient with cirrhosis, and their presence correlates with the severity of liver disease.[3] In cirrhotic patients, esophageal variceal bleeding (EVB) occurs at an annual rate of 5% to 15% and results in substantial morbidity and mortality.[3] Utilizing US national data, Jamal et al. reported a decline in the rate of hospitalizations related to EVB from 1988 to 2002.[4] However, recent large‐scale studies relating to the epidemiology of EVB are lacking. We conducted a retrospective analysis using a national US database to study the differences in demographic characteristics, rate of complications, outcomes, and temporal trends in hospitalized cirrhotic patients with and without EVB.

METHODS

We utilized biennial data (20022012) from the Healthcare Cost and Utilization Project Nationwide Inpatient Sample using methods described earlier.[5] Initially, we extracted all entries with any discharge diagnosis of cirrhosis (International Classification of Diseases, Ninth Revision, Clinical Modification [ICD‐9‐CM] codes: 571.2, 571.5, 571.6) in adult patients ages 18 years and older.[6] Within this cirrhotic population, we next extracted all entries with any discharge diagnosis of EVB (ICD‐9‐CM codes: 456.0., 456.20).[6] Population‐based rates relating to hospital discharges were reported as per 100,000 population/year.

The outcome variables of interest were in‐hospital mortality, total charges (rounded to the nearest $1000) and length of stay (LOS). Demographic details and hospital characteristics were also extracted. Cases were queried for complications well recognized in cirrhotic patients. These included urinary tract infection (UTI) (ICD‐9‐CM codes: 1122, 59010‐11, 5902‐03, 59080‐81, 5950, 5970, 5990), skin and subcutaneous tissue infections (SSCI) (ICD‐9‐CM codes: 680‐82, 684, 686), spontaneous bacterial peritonitis (SBP) (ICD‐9‐CM codes: 56723, 5672), Clostridium difficile infection (ICD‐9‐CM code: 00845), or pneumonia (ICD‐9‐CM codes: 480‐83, 487).[6] Also queried were HE (ICD‐9‐CM code: 572.2)[7] and HRS (ICD‐9‐CM code: 572.4).[8] Comorbid conditions were assessed using the Elixhauser comorbidity index minus the presence of liver disorders but including alcohol abuse.[9]

Statistical analyses were performed using SAS version 9.3 (SAS Institute, Cary, NC). To determine the independent association of EVB on outcome variables, we performed case‐control matching (EVB vs no EVB). We used high‐dimensional propensity scores in a 1:5 matching ratio with a greedy matching algorithm generated by regression analysis of patients with EVB based on demographics details (age, gender, insurance status), comorbid conditions, alcohol abuse, infections as detailed above, HE, and HRS. The ² test and the Mann‐Whitney U test compared categorical and continuous variables. For trend analysis, we used the Cochrane‐Armitage test. The threshold for significance for all analyses was P<0.01.

RESULTS

In 2012, there were 570,020 hospital discharges related to cirrhosis in patients 18 years of age and older. Within this cohort, EVB occurred in 32,945 discharges (5.78%). Table 1 details differences between cirrhotic patients with and without EVB. Comparatively, patients with EVB were younger (median age 55 years, interquartile range [IQR] 13 years vs median age 58 years, IQR 15 years; P<0.01), more likely to be male (70.1% vs 60.4%; P<0.01), and without health insurance (21.0% vs 12.50%; P<0.01). Minor differences between the 2 groups were observed in respect to hospital region, location, teaching status, and household income quartile. There was no difference in the number of comorbid conditions (median 4 comorbid conditions in each group).

Patients with EVB suffered a significantly higher rate of alcohol abuse (63.90% vs 48.80%; P<0.01). EVB was also associated with an overall lower incidence of infection (13.50% vs 24.10%; P<0.01). Specifically, the greatest difference in rates of infection were observed for UTI (6.90% vs 13.10%; P<0.01) and SSCI (1.70% vs 6.30%; P<0.01). Also, patients with EVB demonstrated a small, yet significant increased incidence of HE (18.80% vs 17.70%; P<0.01) and HRS (4.30% vs 3.70%; P<0.01).

Cirrhotic patients with EVB demonstrated worse overall outcomes compared to their counterparts without EVB. This manifested in an unadjusted higher mortality rate (9.90% vs 5.80%; P<0.01) and increased hospital charges (median $41,000 [IQR $49,000] vs $28,000 [IQR $39,000]; P<0.01). LOS between the 2 groups did not differ (median 4 days). After adjusting for demographic differences, complications, and comorbid conditions, EVB in patients with cirrhosis continued to be independently associated with a higher mortality rate (10.00% vs 5.00%; P<0.01) and increased hospital charges (median $41,000 [IQR $49,000] vs $26,000 [IQR $34,000]; P<0.01). Again, LOS was similar for the 2 groups (median 4 days).

Between the years 2002 and 2012, the number of hospital discharges related to cirrhosis increased from 337,956 to 570,220 (P<0.01). Concurrently, the incidence of EVB in this population declined from 8.60% to 5.78% (Figure 1), representing an overall decrease of 33.0% with a significant decreased trend (P<0.01).

Figure 1

We also calculated population‐adjusted hospitalization rates for discharges related to cirrhosis and EVB. The rate of cirrhosis‐related discharges continued to demonstrate an increased trend from 157.42/100,000 population in 2002 to 237.43/100,000 population in 2012 (P<0.01). However, no significant trend was observed for EVB‐related hospital discharges in the same period of time (13.60/100,000 population in 2002 to 13.72/100,000 population in 2012; P=0.91).

DISCUSSION

Our results indicated a significantly higher rate of alcohol abuse in cirrhotic patients with EVB. Alcohol consumption is an independent risk factor for esophageal variceal bleeding.[10, 11] Continued alcohol consumption not only increases the risk for development of varices but may also precipitate variceal rupture.[10] Other risk factors associated with EVB in this study (younger age, male, lower economic status) are likely related to a higher incidence of alcohol abuse in this demographic.[12]

Patients with EVB were also noted to have a lower overall incidence of infection, especially UTI and SSCI. The use of broad‐spectrum antibiotics decreases mortality from secondary infection and improves the prognosis of cirrhotic patients with EVB.[13, 14] The American Association for the Study of Liver Diseases recommends the use of third‐generation cephalosporins in the setting of EVB.[3] The widespread adoption of this in clinical practice may have contributed to a decreased rate of infection in patients with EVB. The difference in the incidence rates of HE and HRS, although statistically significant, were small, and likely the consequence of the large numbers involved in our study.

Our results also indicate that cirrhotic patients with EVB were twice as likely to die compared to matched counterparts without EVB. The increased mortality associated with EVB could be related to hemorrhagic/hypovolemic shock and cardiovascular collapse, aspiration into airway, multiorgan dysfunction due to poor perfusion, infections including SBP, and HE. Although prior studies have demonstrated the relationship between EVB and increased mortality, typically they have been restricted to small single‐center studies involving fewer than 200 patients.[6, 7, 8, 9] Cirrhotic patients with EVB also incurred significantly higher hospital charges compared to matched counterparts. Interestingly, the hospital LOS did not differ between the 2 groups. Intensive care and procedural costs were likely a major contributor to the higher charges; cirrhotic patients with EVB underwent a median of 3 procedures (IQR 2) during their hospital stay compared to a median of 1 procedure (IQR 3) for cirrhotic patients without EVB (P<0.01; data not shown).

In contrast to trends from earlier decades,[4] the population‐adjusted rate of EVB‐related hospital discharges did not change significantly from 2002 to 2012. However, these data are confounded in their interpretation by a substantial increase in the prevalence of cirrhosis in the United States during the same time period.[15] Therefore, it may be more meaningful to state that there was a contemporaneous decline in EVB‐related hospital discharges when considered in the context of a complicating rate in hospitalized cirrhotic patients. These results are consistent with a recent single‐center study[16] and are very likely the fruition of intensive screening programs with primary and secondary prophylaxis for EVB involving esophageal variceal ligation and pharmacotherapy (‐blockers) as well as the increased acceptance of transjugular intrahepatic portosystemic shunt placement.[17, 18, 19]

There are limitations to our study. First, we relied exclusively on ICD‐9‐CM codes for case identification. Second, there is a nonavailability of data pertaining to Model for End‐Stage Liver Disease score calculations, medication, and antibiotic usage. Third, the Nationwide Inpatient Sample database does not allow for distinguishing individual patients with repeat admissions. Finally, our results represent a weighted estimate of national data.

CONCLUSION

EVB in cirrhotic patients was associated with significantly higher mortality and increased hospital charges. Also, the rate of EVB‐related hospital discharges as a complicating factor in patients with cirrhosis declined significantly during the decade 2002 to 2012. This likely reflects the ongoing effectiveness of primary and secondary prophylaxis.

Cirrhosis is a leading cause of death in the United States. In 2010, cirrhosis resulted in an estimated 49,500 deaths, which represented a significant increase from 35,500 deaths 2 decades ago.[1] Cirrhotic patients are susceptible to numerous disease‐specific complications including ascites, esophageal varices, hepatic encephalopathy (HE), and hepatorenal syndrome (HRS).[2]

Esophageal varices develop in approximately 50% of patient with cirrhosis, and their presence correlates with the severity of liver disease.[3] In cirrhotic patients, esophageal variceal bleeding (EVB) occurs at an annual rate of 5% to 15% and results in substantial morbidity and mortality.[3] Utilizing US national data, Jamal et al. reported a decline in the rate of hospitalizations related to EVB from 1988 to 2002.[4] However, recent large‐scale studies relating to the epidemiology of EVB are lacking. We conducted a retrospective analysis using a national US database to study the differences in demographic characteristics, rate of complications, outcomes, and temporal trends in hospitalized cirrhotic patients with and without EVB.

METHODS

We utilized biennial data (20022012) from the Healthcare Cost and Utilization Project Nationwide Inpatient Sample using methods described earlier.[5] Initially, we extracted all entries with any discharge diagnosis of cirrhosis (International Classification of Diseases, Ninth Revision, Clinical Modification [ICD‐9‐CM] codes: 571.2, 571.5, 571.6) in adult patients ages 18 years and older.[6] Within this cirrhotic population, we next extracted all entries with any discharge diagnosis of EVB (ICD‐9‐CM codes: 456.0., 456.20).[6] Population‐based rates relating to hospital discharges were reported as per 100,000 population/year.

The outcome variables of interest were in‐hospital mortality, total charges (rounded to the nearest $1000) and length of stay (LOS). Demographic details and hospital characteristics were also extracted. Cases were queried for complications well recognized in cirrhotic patients. These included urinary tract infection (UTI) (ICD‐9‐CM codes: 1122, 59010‐11, 5902‐03, 59080‐81, 5950, 5970, 5990), skin and subcutaneous tissue infections (SSCI) (ICD‐9‐CM codes: 680‐82, 684, 686), spontaneous bacterial peritonitis (SBP) (ICD‐9‐CM codes: 56723, 5672), Clostridium difficile infection (ICD‐9‐CM code: 00845), or pneumonia (ICD‐9‐CM codes: 480‐83, 487).[6] Also queried were HE (ICD‐9‐CM code: 572.2)[7] and HRS (ICD‐9‐CM code: 572.4).[8] Comorbid conditions were assessed using the Elixhauser comorbidity index minus the presence of liver disorders but including alcohol abuse.[9]

Statistical analyses were performed using SAS version 9.3 (SAS Institute, Cary, NC). To determine the independent association of EVB on outcome variables, we performed case‐control matching (EVB vs no EVB). We used high‐dimensional propensity scores in a 1:5 matching ratio with a greedy matching algorithm generated by regression analysis of patients with EVB based on demographics details (age, gender, insurance status), comorbid conditions, alcohol abuse, infections as detailed above, HE, and HRS. The ² test and the Mann‐Whitney U test compared categorical and continuous variables. For trend analysis, we used the Cochrane‐Armitage test. The threshold for significance for all analyses was P<0.01.

RESULTS

In 2012, there were 570,020 hospital discharges related to cirrhosis in patients 18 years of age and older. Within this cohort, EVB occurred in 32,945 discharges (5.78%). Table 1 details differences between cirrhotic patients with and without EVB. Comparatively, patients with EVB were younger (median age 55 years, interquartile range [IQR] 13 years vs median age 58 years, IQR 15 years; P<0.01), more likely to be male (70.1% vs 60.4%; P<0.01), and without health insurance (21.0% vs 12.50%; P<0.01). Minor differences between the 2 groups were observed in respect to hospital region, location, teaching status, and household income quartile. There was no difference in the number of comorbid conditions (median 4 comorbid conditions in each group).

Patients with EVB suffered a significantly higher rate of alcohol abuse (63.90% vs 48.80%; P<0.01). EVB was also associated with an overall lower incidence of infection (13.50% vs 24.10%; P<0.01). Specifically, the greatest difference in rates of infection were observed for UTI (6.90% vs 13.10%; P<0.01) and SSCI (1.70% vs 6.30%; P<0.01). Also, patients with EVB demonstrated a small, yet significant increased incidence of HE (18.80% vs 17.70%; P<0.01) and HRS (4.30% vs 3.70%; P<0.01).

Cirrhotic patients with EVB demonstrated worse overall outcomes compared to their counterparts without EVB. This manifested in an unadjusted higher mortality rate (9.90% vs 5.80%; P<0.01) and increased hospital charges (median $41,000 [IQR $49,000] vs $28,000 [IQR $39,000]; P<0.01). LOS between the 2 groups did not differ (median 4 days). After adjusting for demographic differences, complications, and comorbid conditions, EVB in patients with cirrhosis continued to be independently associated with a higher mortality rate (10.00% vs 5.00%; P<0.01) and increased hospital charges (median $41,000 [IQR $49,000] vs $26,000 [IQR $34,000]; P<0.01). Again, LOS was similar for the 2 groups (median 4 days).

Between the years 2002 and 2012, the number of hospital discharges related to cirrhosis increased from 337,956 to 570,220 (P<0.01). Concurrently, the incidence of EVB in this population declined from 8.60% to 5.78% (Figure 1), representing an overall decrease of 33.0% with a significant decreased trend (P<0.01).

Figure 1

We also calculated population‐adjusted hospitalization rates for discharges related to cirrhosis and EVB. The rate of cirrhosis‐related discharges continued to demonstrate an increased trend from 157.42/100,000 population in 2002 to 237.43/100,000 population in 2012 (P<0.01). However, no significant trend was observed for EVB‐related hospital discharges in the same period of time (13.60/100,000 population in 2002 to 13.72/100,000 population in 2012; P=0.91).

DISCUSSION

Our results indicated a significantly higher rate of alcohol abuse in cirrhotic patients with EVB. Alcohol consumption is an independent risk factor for esophageal variceal bleeding.[10, 11] Continued alcohol consumption not only increases the risk for development of varices but may also precipitate variceal rupture.[10] Other risk factors associated with EVB in this study (younger age, male, lower economic status) are likely related to a higher incidence of alcohol abuse in this demographic.[12]

Patients with EVB were also noted to have a lower overall incidence of infection, especially UTI and SSCI. The use of broad‐spectrum antibiotics decreases mortality from secondary infection and improves the prognosis of cirrhotic patients with EVB.[13, 14] The American Association for the Study of Liver Diseases recommends the use of third‐generation cephalosporins in the setting of EVB.[3] The widespread adoption of this in clinical practice may have contributed to a decreased rate of infection in patients with EVB. The difference in the incidence rates of HE and HRS, although statistically significant, were small, and likely the consequence of the large numbers involved in our study.

Our results also indicate that cirrhotic patients with EVB were twice as likely to die compared to matched counterparts without EVB. The increased mortality associated with EVB could be related to hemorrhagic/hypovolemic shock and cardiovascular collapse, aspiration into airway, multiorgan dysfunction due to poor perfusion, infections including SBP, and HE. Although prior studies have demonstrated the relationship between EVB and increased mortality, typically they have been restricted to small single‐center studies involving fewer than 200 patients.[6, 7, 8, 9] Cirrhotic patients with EVB also incurred significantly higher hospital charges compared to matched counterparts. Interestingly, the hospital LOS did not differ between the 2 groups. Intensive care and procedural costs were likely a major contributor to the higher charges; cirrhotic patients with EVB underwent a median of 3 procedures (IQR 2) during their hospital stay compared to a median of 1 procedure (IQR 3) for cirrhotic patients without EVB (P<0.01; data not shown).

In contrast to trends from earlier decades,[4] the population‐adjusted rate of EVB‐related hospital discharges did not change significantly from 2002 to 2012. However, these data are confounded in their interpretation by a substantial increase in the prevalence of cirrhosis in the United States during the same time period.[15] Therefore, it may be more meaningful to state that there was a contemporaneous decline in EVB‐related hospital discharges when considered in the context of a complicating rate in hospitalized cirrhotic patients. These results are consistent with a recent single‐center study[16] and are very likely the fruition of intensive screening programs with primary and secondary prophylaxis for EVB involving esophageal variceal ligation and pharmacotherapy (‐blockers) as well as the increased acceptance of transjugular intrahepatic portosystemic shunt placement.[17, 18, 19]

There are limitations to our study. First, we relied exclusively on ICD‐9‐CM codes for case identification. Second, there is a nonavailability of data pertaining to Model for End‐Stage Liver Disease score calculations, medication, and antibiotic usage. Third, the Nationwide Inpatient Sample database does not allow for distinguishing individual patients with repeat admissions. Finally, our results represent a weighted estimate of national data.

CONCLUSION

EVB in cirrhotic patients was associated with significantly higher mortality and increased hospital charges. Also, the rate of EVB‐related hospital discharges as a complicating factor in patients with cirrhosis declined significantly during the decade 2002 to 2012. This likely reflects the ongoing effectiveness of primary and secondary prophylaxis.

Cirrhosis is a leading cause of death in the United States. In 2010, cirrhosis resulted in an estimated 49,500 deaths, which represented a significant increase from 35,500 deaths 2 decades ago.[1] Cirrhotic patients are susceptible to numerous disease‐specific complications including ascites, esophageal varices, hepatic encephalopathy (HE), and hepatorenal syndrome (HRS).[2]

Esophageal varices develop in approximately 50% of patient with cirrhosis, and their presence correlates with the severity of liver disease.[3] In cirrhotic patients, esophageal variceal bleeding (EVB) occurs at an annual rate of 5% to 15% and results in substantial morbidity and mortality.[3] Utilizing US national data, Jamal et al. reported a decline in the rate of hospitalizations related to EVB from 1988 to 2002.[4] However, recent large‐scale studies relating to the epidemiology of EVB are lacking. We conducted a retrospective analysis using a national US database to study the differences in demographic characteristics, rate of complications, outcomes, and temporal trends in hospitalized cirrhotic patients with and without EVB.

METHODS

We utilized biennial data (20022012) from the Healthcare Cost and Utilization Project Nationwide Inpatient Sample using methods described earlier.[5] Initially, we extracted all entries with any discharge diagnosis of cirrhosis (International Classification of Diseases, Ninth Revision, Clinical Modification [ICD‐9‐CM] codes: 571.2, 571.5, 571.6) in adult patients ages 18 years and older.[6] Within this cirrhotic population, we next extracted all entries with any discharge diagnosis of EVB (ICD‐9‐CM codes: 456.0., 456.20).[6] Population‐based rates relating to hospital discharges were reported as per 100,000 population/year.

The outcome variables of interest were in‐hospital mortality, total charges (rounded to the nearest $1000) and length of stay (LOS). Demographic details and hospital characteristics were also extracted. Cases were queried for complications well recognized in cirrhotic patients. These included urinary tract infection (UTI) (ICD‐9‐CM codes: 1122, 59010‐11, 5902‐03, 59080‐81, 5950, 5970, 5990), skin and subcutaneous tissue infections (SSCI) (ICD‐9‐CM codes: 680‐82, 684, 686), spontaneous bacterial peritonitis (SBP) (ICD‐9‐CM codes: 56723, 5672), Clostridium difficile infection (ICD‐9‐CM code: 00845), or pneumonia (ICD‐9‐CM codes: 480‐83, 487).[6] Also queried were HE (ICD‐9‐CM code: 572.2)[7] and HRS (ICD‐9‐CM code: 572.4).[8] Comorbid conditions were assessed using the Elixhauser comorbidity index minus the presence of liver disorders but including alcohol abuse.[9]

Statistical analyses were performed using SAS version 9.3 (SAS Institute, Cary, NC). To determine the independent association of EVB on outcome variables, we performed case‐control matching (EVB vs no EVB). We used high‐dimensional propensity scores in a 1:5 matching ratio with a greedy matching algorithm generated by regression analysis of patients with EVB based on demographics details (age, gender, insurance status), comorbid conditions, alcohol abuse, infections as detailed above, HE, and HRS. The ² test and the Mann‐Whitney U test compared categorical and continuous variables. For trend analysis, we used the Cochrane‐Armitage test. The threshold for significance for all analyses was P<0.01.

RESULTS

In 2012, there were 570,020 hospital discharges related to cirrhosis in patients 18 years of age and older. Within this cohort, EVB occurred in 32,945 discharges (5.78%). Table 1 details differences between cirrhotic patients with and without EVB. Comparatively, patients with EVB were younger (median age 55 years, interquartile range [IQR] 13 years vs median age 58 years, IQR 15 years; P<0.01), more likely to be male (70.1% vs 60.4%; P<0.01), and without health insurance (21.0% vs 12.50%; P<0.01). Minor differences between the 2 groups were observed in respect to hospital region, location, teaching status, and household income quartile. There was no difference in the number of comorbid conditions (median 4 comorbid conditions in each group).

Patients with EVB suffered a significantly higher rate of alcohol abuse (63.90% vs 48.80%; P<0.01). EVB was also associated with an overall lower incidence of infection (13.50% vs 24.10%; P<0.01). Specifically, the greatest difference in rates of infection were observed for UTI (6.90% vs 13.10%; P<0.01) and SSCI (1.70% vs 6.30%; P<0.01). Also, patients with EVB demonstrated a small, yet significant increased incidence of HE (18.80% vs 17.70%; P<0.01) and HRS (4.30% vs 3.70%; P<0.01).

Cirrhotic patients with EVB demonstrated worse overall outcomes compared to their counterparts without EVB. This manifested in an unadjusted higher mortality rate (9.90% vs 5.80%; P<0.01) and increased hospital charges (median $41,000 [IQR $49,000] vs $28,000 [IQR $39,000]; P<0.01). LOS between the 2 groups did not differ (median 4 days). After adjusting for demographic differences, complications, and comorbid conditions, EVB in patients with cirrhosis continued to be independently associated with a higher mortality rate (10.00% vs 5.00%; P<0.01) and increased hospital charges (median $41,000 [IQR $49,000] vs $26,000 [IQR $34,000]; P<0.01). Again, LOS was similar for the 2 groups (median 4 days).

Between the years 2002 and 2012, the number of hospital discharges related to cirrhosis increased from 337,956 to 570,220 (P<0.01). Concurrently, the incidence of EVB in this population declined from 8.60% to 5.78% (Figure 1), representing an overall decrease of 33.0% with a significant decreased trend (P<0.01).

Figure 1

We also calculated population‐adjusted hospitalization rates for discharges related to cirrhosis and EVB. The rate of cirrhosis‐related discharges continued to demonstrate an increased trend from 157.42/100,000 population in 2002 to 237.43/100,000 population in 2012 (P<0.01). However, no significant trend was observed for EVB‐related hospital discharges in the same period of time (13.60/100,000 population in 2002 to 13.72/100,000 population in 2012; P=0.91).

DISCUSSION

Our results indicated a significantly higher rate of alcohol abuse in cirrhotic patients with EVB. Alcohol consumption is an independent risk factor for esophageal variceal bleeding.[10, 11] Continued alcohol consumption not only increases the risk for development of varices but may also precipitate variceal rupture.[10] Other risk factors associated with EVB in this study (younger age, male, lower economic status) are likely related to a higher incidence of alcohol abuse in this demographic.[12]

Patients with EVB were also noted to have a lower overall incidence of infection, especially UTI and SSCI. The use of broad‐spectrum antibiotics decreases mortality from secondary infection and improves the prognosis of cirrhotic patients with EVB.[13, 14] The American Association for the Study of Liver Diseases recommends the use of third‐generation cephalosporins in the setting of EVB.[3] The widespread adoption of this in clinical practice may have contributed to a decreased rate of infection in patients with EVB. The difference in the incidence rates of HE and HRS, although statistically significant, were small, and likely the consequence of the large numbers involved in our study.

Our results also indicate that cirrhotic patients with EVB were twice as likely to die compared to matched counterparts without EVB. The increased mortality associated with EVB could be related to hemorrhagic/hypovolemic shock and cardiovascular collapse, aspiration into airway, multiorgan dysfunction due to poor perfusion, infections including SBP, and HE. Although prior studies have demonstrated the relationship between EVB and increased mortality, typically they have been restricted to small single‐center studies involving fewer than 200 patients.[6, 7, 8, 9] Cirrhotic patients with EVB also incurred significantly higher hospital charges compared to matched counterparts. Interestingly, the hospital LOS did not differ between the 2 groups. Intensive care and procedural costs were likely a major contributor to the higher charges; cirrhotic patients with EVB underwent a median of 3 procedures (IQR 2) during their hospital stay compared to a median of 1 procedure (IQR 3) for cirrhotic patients without EVB (P<0.01; data not shown).

In contrast to trends from earlier decades,[4] the population‐adjusted rate of EVB‐related hospital discharges did not change significantly from 2002 to 2012. However, these data are confounded in their interpretation by a substantial increase in the prevalence of cirrhosis in the United States during the same time period.[15] Therefore, it may be more meaningful to state that there was a contemporaneous decline in EVB‐related hospital discharges when considered in the context of a complicating rate in hospitalized cirrhotic patients. These results are consistent with a recent single‐center study[16] and are very likely the fruition of intensive screening programs with primary and secondary prophylaxis for EVB involving esophageal variceal ligation and pharmacotherapy (‐blockers) as well as the increased acceptance of transjugular intrahepatic portosystemic shunt placement.[17, 18, 19]

There are limitations to our study. First, we relied exclusively on ICD‐9‐CM codes for case identification. Second, there is a nonavailability of data pertaining to Model for End‐Stage Liver Disease score calculations, medication, and antibiotic usage. Third, the Nationwide Inpatient Sample database does not allow for distinguishing individual patients with repeat admissions. Finally, our results represent a weighted estimate of national data.

CONCLUSION

EVB in cirrhotic patients was associated with significantly higher mortality and increased hospital charges. Also, the rate of EVB‐related hospital discharges as a complicating factor in patients with cirrhosis declined significantly during the decade 2002 to 2012. This likely reflects the ongoing effectiveness of primary and secondary prophylaxis.

National guidelines for the management of childhood pneumonia highlight the need for the development of objective outcome measures to inform clinical decision making, establish benchmarks of care, and compare treatments and interventions.[1] Time to clinical stability (TCS) is a measure reported in adult pneumonia studies that incorporates vital signs, ability to eat, and mental status to objectively assess readiness for discharge.[2, 3, 4] TCS has not been validated among children as it has in adults,[5, 6, 7, 8] although such measures could prove useful for assessing discharge readiness with applications in both clinical and research settings. The objective of our study was to test the performance of pediatric TCS measures among children hospitalized with pneumonia.

METHODS

RESULTS

Combination TCS Measures

Within each age group, the percentage of children achieving stability was relatively consistent across the 4 combined TCS measures (Table 2); however, more children were considered unstable at discharge (and fewer classified as stable on admission) as the number of included parameters increased. More children <5 years of age reached stability (range, 80.0%85.6%) compared to children 5 years of age (range, 68.3%72.3%). We also noted increasing median TCS with increasing disease severity (Figure 1, P<0.01) (see Supporting Fig. 1AC in the online version of this article); TCS was only slightly shorter than LOS across all 3 levels of the severity scale.

Table 2.Progression to Stability for Four TCS Measures Among Children Hospitalized With Community‐Acquired Pneumonia

TCS Measures	<2 Years, n=130		24 Years, n=90		517 Years, n=101		P Value
	No. (%)*	Median (IQR) TCS, h	No. (%)	Median (IQR) TCS, h	No. (%)	Median (IQR) TCS, h
NOTE: For each measure, time to clinical stability (TCS) was calculated by subtracting the time and date of the last abnormal value for the included parameters from admission time and date to determine time to stability for each parameter in hours; children stable on admission for all 4 parameters not included (n=11). Abbreviations: HR, heart rate; IQR, interquartile range; O2, supplemental oxygen; RR, respiratory rate; T, temperature; *Number (%) of children who reached stability more than 6 hours prior to discharge. P value comparing median TCS by age group, estimated using nonparametric test of trend.
RR+O2	108 (83.1)	40.5 (20.175.0)	72 (80.0)	39.6 (15.679.2)	69 (68.3)	30.4 (14.759.2)	0.08
RR+O2+HR	109 (83.8)	40.2 (19.573.9)	73 (81.1)	35.9 (15.977.6)	68 (67.3)	29.8 (17.256.6)	0.11
RR+O2+T	110 (84.6)	40.5 (20.770.1)	77 (85.6)	39.1 (18.477.6)	73 (72.3)	28.2 (14.744.7)	0.03
RR+O2+HR+T	110 (84.6)	40.5 (20.770.1)	72 (80.0)	39.7 (20.177.5)	71 (70.3)	29.2 (18.254)	0.05

Figure 1

DISCUSSION

Our study demonstrates that longitudinal TCS measures consisting of routinely collected physiologic parameters may be useful for objectively assessing disease recovery and clinical readiness for discharge among children hospitalized with pneumonia. A simple TCS measure incorporating respiratory rate and oxygen requirement performed similarly to the more complex combinations and classified fewer children as unstable at discharge. However, we also note several challenges that deserve additional study prior to the application of a pediatric TCS measure in clinical and research settings.

Vital signs and supplemental oxygen use are used clinically to assess disease severity and response to therapy among children with acute respiratory illness. Because these objective parameters are routinely collected among hospitalized children, the systematization of these data could inform clinical decision making around hospital discharge. Similar to early warning scores used to detect impending clinical deterioration,[11] TCS measures, by signaling normalization of stability parameters in a consistent and objective manner, could serve as an early signal of readiness for discharge. However, maximizing the clinical utility of TCS would require embedding the process within the electronic health record, a tool that could also have implications for the Centers for Medicare and Medicaid Services' meaningful use regulations.[12]

TCS could also serve as an outcome measure in research and quality efforts. Increased disease severity was associated with longer TCS for the 4 combined measures; TCS also demonstrated strong agreement with LOS. Furthermore, TCS minimizes the influence of factors unrelated to disease that may impact LOS (eg, frequency of hospital rounds, transportation difficulties, or social impediments to discharge), an advantage when studying outcomes for research and quality benchmarking.

The percentage of children reaching stability and the median TCS for the combined measures demonstrated little variation within each age group, likely because respiratory rate and need for supplemental oxygen, 2 of the parameters with the longest individual time to stability, were also included in each of the combination measures. This suggests that less‐complex measures incorporating only respiratory rate and need for supplemental oxygen may be sufficient to assess clinical stability, particularly because these parameters are objectively measured and possess a direct physiological link to pneumonia. In contrast, the other parameters may be more often influenced by factors unrelated to disease severity.

Our study also highlights several shortcomings of the pediatric TCS measures. Despite use of published, age‐based reference values,[13] we noted wide variation in the achievement of stability across individual parameters, especially for children 5 years old. Overall, 21% of children had 1 abnormal parameter at discharge. Even the simplest combined measure classified 13.4% of children as unstable at discharge. Discharge with unstable parameters was not associated with 7‐day readmission, although our study was underpowered to detect small differences. Additional study is therefore needed to evaluate less restrictive cutoff values on calculated TCS and the impact of hospital discharge prior to reaching stability. In particular, relaxing the upper limit for normal respiratory rate in adolescents (16 breaths per minute) to more closely approximate the adult TCS parameter (24 breaths per minute) should be explored. Refinement and standardization of age‐based vital sign reference values specific to hospitalized children may also improve the performance of these measures.[14]

Several limitations deserve discussion. TCS parameters and readmission data were abstracted retrospectively from a single institution, and our findings may not be generalizable. Although clinical staff routinely measured these data, measurement variation likely exists. Nevertheless, such variation is likely systematic, limiting the impact of potential misclassification. TCS was calculated based on the last abnormal value for each parameter; prior fluctuations between normal and abnormal periods of stability were not captured. We were unable to assess room air oxygen saturations. Instead, supplemental oxygen use served as a surrogate for hypoxia. At our institution, oxygen therapy is provided for children with pneumonia to maintain oxygen saturations of 90% to 92%. We did not assess work of breathing (a marker of severe pneumonia) or ability to eat (a component of adult TCS measures). We initially considered the evaluation of intravenous fluids as a proxy for ability to eat (addition of this parameter to the 4 parameter TCS resulted in a modest increase in median time to stability, data not shown); however, we felt the lack of institutional policy and subjective nature of this parameter detracted from our study's objectives. Finally, we were not able to determine clinical readiness for discharge beyond the measurement of vital sign parameters. Therefore, prospective evaluation of the proposed pediatric TCS measures in broader populations will be important to build upon our findings, refine stability parameters, and test the utility of new parameters (eg, ability to eat, work of breathing) prior to use in clinical settings.

Our study provides an initial evaluation of TCS measures for assessing severity and recovery among children hospitalized with pneumonia. Similar to adults, such validated TCS measures may ultimately prove useful for improving the quality of both clinical care and research, although additional study to more clearly define stability criteria is needed prior to implementation.

National guidelines for the management of childhood pneumonia highlight the need for the development of objective outcome measures to inform clinical decision making, establish benchmarks of care, and compare treatments and interventions.[1] Time to clinical stability (TCS) is a measure reported in adult pneumonia studies that incorporates vital signs, ability to eat, and mental status to objectively assess readiness for discharge.[2, 3, 4] TCS has not been validated among children as it has in adults,[5, 6, 7, 8] although such measures could prove useful for assessing discharge readiness with applications in both clinical and research settings. The objective of our study was to test the performance of pediatric TCS measures among children hospitalized with pneumonia.

METHODS

RESULTS

Combination TCS Measures

Within each age group, the percentage of children achieving stability was relatively consistent across the 4 combined TCS measures (Table 2); however, more children were considered unstable at discharge (and fewer classified as stable on admission) as the number of included parameters increased. More children <5 years of age reached stability (range, 80.0%85.6%) compared to children 5 years of age (range, 68.3%72.3%). We also noted increasing median TCS with increasing disease severity (Figure 1, P<0.01) (see Supporting Fig. 1AC in the online version of this article); TCS was only slightly shorter than LOS across all 3 levels of the severity scale.

Table 2.Progression to Stability for Four TCS Measures Among Children Hospitalized With Community‐Acquired Pneumonia

TCS Measures	<2 Years, n=130		24 Years, n=90		517 Years, n=101		P Value
	No. (%)*	Median (IQR) TCS, h	No. (%)	Median (IQR) TCS, h	No. (%)	Median (IQR) TCS, h
NOTE: For each measure, time to clinical stability (TCS) was calculated by subtracting the time and date of the last abnormal value for the included parameters from admission time and date to determine time to stability for each parameter in hours; children stable on admission for all 4 parameters not included (n=11). Abbreviations: HR, heart rate; IQR, interquartile range; O2, supplemental oxygen; RR, respiratory rate; T, temperature; *Number (%) of children who reached stability more than 6 hours prior to discharge. P value comparing median TCS by age group, estimated using nonparametric test of trend.
RR+O2	108 (83.1)	40.5 (20.175.0)	72 (80.0)	39.6 (15.679.2)	69 (68.3)	30.4 (14.759.2)	0.08
RR+O2+HR	109 (83.8)	40.2 (19.573.9)	73 (81.1)	35.9 (15.977.6)	68 (67.3)	29.8 (17.256.6)	0.11
RR+O2+T	110 (84.6)	40.5 (20.770.1)	77 (85.6)	39.1 (18.477.6)	73 (72.3)	28.2 (14.744.7)	0.03
RR+O2+HR+T	110 (84.6)	40.5 (20.770.1)	72 (80.0)	39.7 (20.177.5)	71 (70.3)	29.2 (18.254)	0.05

Figure 1

DISCUSSION

Our study demonstrates that longitudinal TCS measures consisting of routinely collected physiologic parameters may be useful for objectively assessing disease recovery and clinical readiness for discharge among children hospitalized with pneumonia. A simple TCS measure incorporating respiratory rate and oxygen requirement performed similarly to the more complex combinations and classified fewer children as unstable at discharge. However, we also note several challenges that deserve additional study prior to the application of a pediatric TCS measure in clinical and research settings.

Vital signs and supplemental oxygen use are used clinically to assess disease severity and response to therapy among children with acute respiratory illness. Because these objective parameters are routinely collected among hospitalized children, the systematization of these data could inform clinical decision making around hospital discharge. Similar to early warning scores used to detect impending clinical deterioration,[11] TCS measures, by signaling normalization of stability parameters in a consistent and objective manner, could serve as an early signal of readiness for discharge. However, maximizing the clinical utility of TCS would require embedding the process within the electronic health record, a tool that could also have implications for the Centers for Medicare and Medicaid Services' meaningful use regulations.[12]

TCS could also serve as an outcome measure in research and quality efforts. Increased disease severity was associated with longer TCS for the 4 combined measures; TCS also demonstrated strong agreement with LOS. Furthermore, TCS minimizes the influence of factors unrelated to disease that may impact LOS (eg, frequency of hospital rounds, transportation difficulties, or social impediments to discharge), an advantage when studying outcomes for research and quality benchmarking.

The percentage of children reaching stability and the median TCS for the combined measures demonstrated little variation within each age group, likely because respiratory rate and need for supplemental oxygen, 2 of the parameters with the longest individual time to stability, were also included in each of the combination measures. This suggests that less‐complex measures incorporating only respiratory rate and need for supplemental oxygen may be sufficient to assess clinical stability, particularly because these parameters are objectively measured and possess a direct physiological link to pneumonia. In contrast, the other parameters may be more often influenced by factors unrelated to disease severity.

Our study also highlights several shortcomings of the pediatric TCS measures. Despite use of published, age‐based reference values,[13] we noted wide variation in the achievement of stability across individual parameters, especially for children 5 years old. Overall, 21% of children had 1 abnormal parameter at discharge. Even the simplest combined measure classified 13.4% of children as unstable at discharge. Discharge with unstable parameters was not associated with 7‐day readmission, although our study was underpowered to detect small differences. Additional study is therefore needed to evaluate less restrictive cutoff values on calculated TCS and the impact of hospital discharge prior to reaching stability. In particular, relaxing the upper limit for normal respiratory rate in adolescents (16 breaths per minute) to more closely approximate the adult TCS parameter (24 breaths per minute) should be explored. Refinement and standardization of age‐based vital sign reference values specific to hospitalized children may also improve the performance of these measures.[14]

Several limitations deserve discussion. TCS parameters and readmission data were abstracted retrospectively from a single institution, and our findings may not be generalizable. Although clinical staff routinely measured these data, measurement variation likely exists. Nevertheless, such variation is likely systematic, limiting the impact of potential misclassification. TCS was calculated based on the last abnormal value for each parameter; prior fluctuations between normal and abnormal periods of stability were not captured. We were unable to assess room air oxygen saturations. Instead, supplemental oxygen use served as a surrogate for hypoxia. At our institution, oxygen therapy is provided for children with pneumonia to maintain oxygen saturations of 90% to 92%. We did not assess work of breathing (a marker of severe pneumonia) or ability to eat (a component of adult TCS measures). We initially considered the evaluation of intravenous fluids as a proxy for ability to eat (addition of this parameter to the 4 parameter TCS resulted in a modest increase in median time to stability, data not shown); however, we felt the lack of institutional policy and subjective nature of this parameter detracted from our study's objectives. Finally, we were not able to determine clinical readiness for discharge beyond the measurement of vital sign parameters. Therefore, prospective evaluation of the proposed pediatric TCS measures in broader populations will be important to build upon our findings, refine stability parameters, and test the utility of new parameters (eg, ability to eat, work of breathing) prior to use in clinical settings.

Our study provides an initial evaluation of TCS measures for assessing severity and recovery among children hospitalized with pneumonia. Similar to adults, such validated TCS measures may ultimately prove useful for improving the quality of both clinical care and research, although additional study to more clearly define stability criteria is needed prior to implementation.

National guidelines for the management of childhood pneumonia highlight the need for the development of objective outcome measures to inform clinical decision making, establish benchmarks of care, and compare treatments and interventions.[1] Time to clinical stability (TCS) is a measure reported in adult pneumonia studies that incorporates vital signs, ability to eat, and mental status to objectively assess readiness for discharge.[2, 3, 4] TCS has not been validated among children as it has in adults,[5, 6, 7, 8] although such measures could prove useful for assessing discharge readiness with applications in both clinical and research settings. The objective of our study was to test the performance of pediatric TCS measures among children hospitalized with pneumonia.

METHODS

RESULTS

Combination TCS Measures

Within each age group, the percentage of children achieving stability was relatively consistent across the 4 combined TCS measures (Table 2); however, more children were considered unstable at discharge (and fewer classified as stable on admission) as the number of included parameters increased. More children <5 years of age reached stability (range, 80.0%85.6%) compared to children 5 years of age (range, 68.3%72.3%). We also noted increasing median TCS with increasing disease severity (Figure 1, P<0.01) (see Supporting Fig. 1AC in the online version of this article); TCS was only slightly shorter than LOS across all 3 levels of the severity scale.

Table 2.Progression to Stability for Four TCS Measures Among Children Hospitalized With Community‐Acquired Pneumonia

TCS Measures	<2 Years, n=130		24 Years, n=90		517 Years, n=101		P Value
	No. (%)*	Median (IQR) TCS, h	No. (%)	Median (IQR) TCS, h	No. (%)	Median (IQR) TCS, h
NOTE: For each measure, time to clinical stability (TCS) was calculated by subtracting the time and date of the last abnormal value for the included parameters from admission time and date to determine time to stability for each parameter in hours; children stable on admission for all 4 parameters not included (n=11). Abbreviations: HR, heart rate; IQR, interquartile range; O2, supplemental oxygen; RR, respiratory rate; T, temperature; *Number (%) of children who reached stability more than 6 hours prior to discharge. P value comparing median TCS by age group, estimated using nonparametric test of trend.
RR+O2	108 (83.1)	40.5 (20.175.0)	72 (80.0)	39.6 (15.679.2)	69 (68.3)	30.4 (14.759.2)	0.08
RR+O2+HR	109 (83.8)	40.2 (19.573.9)	73 (81.1)	35.9 (15.977.6)	68 (67.3)	29.8 (17.256.6)	0.11
RR+O2+T	110 (84.6)	40.5 (20.770.1)	77 (85.6)	39.1 (18.477.6)	73 (72.3)	28.2 (14.744.7)	0.03
RR+O2+HR+T	110 (84.6)	40.5 (20.770.1)	72 (80.0)	39.7 (20.177.5)	71 (70.3)	29.2 (18.254)	0.05

Figure 1

DISCUSSION

Our study demonstrates that longitudinal TCS measures consisting of routinely collected physiologic parameters may be useful for objectively assessing disease recovery and clinical readiness for discharge among children hospitalized with pneumonia. A simple TCS measure incorporating respiratory rate and oxygen requirement performed similarly to the more complex combinations and classified fewer children as unstable at discharge. However, we also note several challenges that deserve additional study prior to the application of a pediatric TCS measure in clinical and research settings.

Vital signs and supplemental oxygen use are used clinically to assess disease severity and response to therapy among children with acute respiratory illness. Because these objective parameters are routinely collected among hospitalized children, the systematization of these data could inform clinical decision making around hospital discharge. Similar to early warning scores used to detect impending clinical deterioration,[11] TCS measures, by signaling normalization of stability parameters in a consistent and objective manner, could serve as an early signal of readiness for discharge. However, maximizing the clinical utility of TCS would require embedding the process within the electronic health record, a tool that could also have implications for the Centers for Medicare and Medicaid Services' meaningful use regulations.[12]

TCS could also serve as an outcome measure in research and quality efforts. Increased disease severity was associated with longer TCS for the 4 combined measures; TCS also demonstrated strong agreement with LOS. Furthermore, TCS minimizes the influence of factors unrelated to disease that may impact LOS (eg, frequency of hospital rounds, transportation difficulties, or social impediments to discharge), an advantage when studying outcomes for research and quality benchmarking.

The percentage of children reaching stability and the median TCS for the combined measures demonstrated little variation within each age group, likely because respiratory rate and need for supplemental oxygen, 2 of the parameters with the longest individual time to stability, were also included in each of the combination measures. This suggests that less‐complex measures incorporating only respiratory rate and need for supplemental oxygen may be sufficient to assess clinical stability, particularly because these parameters are objectively measured and possess a direct physiological link to pneumonia. In contrast, the other parameters may be more often influenced by factors unrelated to disease severity.

Our study also highlights several shortcomings of the pediatric TCS measures. Despite use of published, age‐based reference values,[13] we noted wide variation in the achievement of stability across individual parameters, especially for children 5 years old. Overall, 21% of children had 1 abnormal parameter at discharge. Even the simplest combined measure classified 13.4% of children as unstable at discharge. Discharge with unstable parameters was not associated with 7‐day readmission, although our study was underpowered to detect small differences. Additional study is therefore needed to evaluate less restrictive cutoff values on calculated TCS and the impact of hospital discharge prior to reaching stability. In particular, relaxing the upper limit for normal respiratory rate in adolescents (16 breaths per minute) to more closely approximate the adult TCS parameter (24 breaths per minute) should be explored. Refinement and standardization of age‐based vital sign reference values specific to hospitalized children may also improve the performance of these measures.[14]

Several limitations deserve discussion. TCS parameters and readmission data were abstracted retrospectively from a single institution, and our findings may not be generalizable. Although clinical staff routinely measured these data, measurement variation likely exists. Nevertheless, such variation is likely systematic, limiting the impact of potential misclassification. TCS was calculated based on the last abnormal value for each parameter; prior fluctuations between normal and abnormal periods of stability were not captured. We were unable to assess room air oxygen saturations. Instead, supplemental oxygen use served as a surrogate for hypoxia. At our institution, oxygen therapy is provided for children with pneumonia to maintain oxygen saturations of 90% to 92%. We did not assess work of breathing (a marker of severe pneumonia) or ability to eat (a component of adult TCS measures). We initially considered the evaluation of intravenous fluids as a proxy for ability to eat (addition of this parameter to the 4 parameter TCS resulted in a modest increase in median time to stability, data not shown); however, we felt the lack of institutional policy and subjective nature of this parameter detracted from our study's objectives. Finally, we were not able to determine clinical readiness for discharge beyond the measurement of vital sign parameters. Therefore, prospective evaluation of the proposed pediatric TCS measures in broader populations will be important to build upon our findings, refine stability parameters, and test the utility of new parameters (eg, ability to eat, work of breathing) prior to use in clinical settings.

Our study provides an initial evaluation of TCS measures for assessing severity and recovery among children hospitalized with pneumonia. Similar to adults, such validated TCS measures may ultimately prove useful for improving the quality of both clinical care and research, although additional study to more clearly define stability criteria is needed prior to implementation.

A number of observational studies have documented the association between prolonged bed rest during hospitalization with adverse short‐ and long‐term functional impairments and disability in older patients.[1, 2, 3, 4] However, the body of evidence on the benefits of early mobilization on functional outcomes in both critically ill patients and more stable patients on medical‐surgical floors remains inconclusive.[5, 6, 7, 8, 9] Despite the increased emphasis on mobilizing patients early and often in the inpatient setting, there is surprisingly little information available regarding how typically active adult patients are during their hospital stay. The few published studies that are available are limited by small samples and types of patients who were monitored.[10, 11, 12, 13, 14] Therefore, the purpose of this real‐world study was to describe the level of ambulation in a large sample of hospitalized adult patients using a validated consumer‐grade wireless accelerometer.

METHODS

This was a prospective cohort study of ambulatory patients from 3 medical‐surgical units of a community hospital from March 2014 through July 2014. The study was approved by the Kaiser Permanente Southern California Institutional Review Board. All ambulatory medical and surgical adult patients were eligible for the study except for those with isolation precautions. Patients wore an accelerometer (Tractivity; Kineteks Corp., Vancouver, BC, Canada) on the ankle from soon after admission to the unit until discharge home. The sensors were only removed for bathing and medical procedures, at which time the devices were secured to the patient's bed and reworn upon their return to the room. The nursing staff was trained to use the vendor application to register the sensor to the patient, secure the sensor to the patient's ankle, transfer the sensor data to the vendor server, review the step counts on the web application, and manually key the step count into the electronic medical records (EMRs) as part of routine nursing workflow. The staff otherwise continued with usual patient mobilization practices.

We previously validated the Tractivity device in a field study of 20 hospitalized patients using a research‐grade accelerometer, Stepwatch, as the gold standard (unpublished data). We found that the inter‐Tractivity device reliability was near perfect (intraclass correlation=0.99), and that the Tractivity step counts correlated highly with the nurses' documentation on a paper log of distance walked measured in feet (r=0.76). A small number of steps (<100) were recorded over 24 hours when the device was worn by 2 bed bound patients. The 24‐hour Tractivity step count had acceptable limits of agreement with the Stepwatch (+284 [standard deviation: 314] steps; 95% limits of agreement 911‐343). In addition, for the current study, when we examined the step counts between patients who were classified by the nursing team as being able to walk <50 feet (n=320) compared to patients who were able to walk >50 feet (n=434), we found a significant difference in the median number of steps over a 24‐hour period (854 vs 1697, P<0.0001).

The step count data were exported from the vendor's server, examined for irregularities, and merged with administrative and clinical data for analysis. Data extracted from the EMR system included sociodemographic (age, gender, marital status, and race/ethnicity) and clinical characteristics (LACE score [readmission risk score based on length of stay (L); acuity of the admission (A); comorbidity of the patient (measured with the Charlson comorbidity index score) (C); and emergency department use (measured as the number of visits in the six months before admission) (E),[15] Charlson Comorbidity Index, length of stay, principal discharge diagnosis, and body mass index), and nursing documentation of functional status (bed bound, sit up in bed, stand next to bed, walk <50 feet, and walk >50 feet).

Descriptive statistics and nonparametric tests (Kruskal‐Wallis and Wilcoxon signed rank) were used to analyze the non‐normally distributed step count data. Quantile regression[16] was used to determine the association between the frequency of the care team's review and documentation of steps, with median total step count adjusting for age, gender, LACE score, and medicine/surgical service line. Whereas linear regression allows one to describe how the mean of a given outcome changes with respect to some set of covariates in circumstances where data are normally distributed, quantile regression allows one to assess how a set of covariates are related to a prespecified quantile (eg, 50% percentile median) of an outcome distribution. This modeling is especially appropriate here, because step count data are not normally distributed. Because step counts can vary with a number of factors, such as age and principal admitting and discharge diagnoses, we stratified our analyses by age (<65 or 65 years) and service lines (medical or surgical) due to the relatively small numbers of patients in each of the diagnostic groupings. Statistical analyses were performed using SAS version 9.3 (SAS Institute Inc., Cary, NC); P values <0.05 were considered statistically significant.

RESULTS

A total of 1667 patients wore the activity sensor during their hospital stay. We included 777 patients in our analysis who had lengths of stay long enough for 24 hours of continuous monitoring, and almost half of these patients had at least 48 hours of monitoring (n=378). The demographic and clinical characteristics of the sample are detailed in Table 1. The sample included mostly medical patients (77%), with a mean age of 6017 years, 57% females, and 55% nonwhites. Nearly all patients (97%) were classified as ambulatory at discharge based on the EMR data. Approximately 44% of the sensors were lost, mostly due to nursing staff forgetting to remove the devices at discharge; device failure was minimal (n=10).

Patients accrued a median of 1158 (interquartile range: 6362238) steps over the 24 hours prior to discharge to home (Table 2). Approximately 13 (2%) patients registered zero steps in the last 24 hours; this may have been due to patients truly not accruing any steps, device failure, or the device was registered but never worn by the patient. Patients who were 65 years and older on both the medicine and surgical services accrued fewer steps compared to younger patients (962 vs 1294, P<0.0001). For patients who had at least 48 hours of continuous monitoring (n=378), there was a median increase of 377 steps from the first 24 hours from admission to the unit to the final 24 hours prior to discharge (811 steps to 1188 steps, P<0.0001) (Table 3 and Figure 1). The average length of stay for these patients was 5.74.9 days. Despite the longer length of stay, the level of ambulation at discharge was similar to patients with shorter stays. This is further illustrated in Figure 2 in the spaghetti plots of total steps over 4, 24‐hour monitoring increments. Ignoring the outliers, the plots suggest the following: (1) step counts tended to increase or stay about the same over the course of a hospitalization; and (2) for the medicine service line, step counts in the final 24 hours prior to discharge for patients with longer lengths of stay (72 or 96 hours) did not appear to be substantially different from patients with shorter lengths of stay. The data for the surgical patients are either too sparse or erratic to make any firm conclusions. Patients accrued steps throughout the day with the highest percentage of steps logged at approximately 6 _am and 6 _pm; these data are based on time stamps from the device, not the time of data transfer or documentation in the EMR (Figure 3).

Figure 1

Figure 2

Figure 3

More frequent documentation of step counts in the EMR (proxy for step count data retrieval and review from the vendor web site) by the care team was associated with higher total step counts after adjustments for relevant covariates (P0.001); 3 or more documentations over a 24‐hour period appears to be a minimal frequency to achieving approximately 200 steps more than the median value (Table 4).

DISCUSSION

We found that ambulatory medical‐surgical patients accrued a median of 1158 total steps in the 24 hours prior to their discharge home, which translates to walking approximately 500 meters; older patients accrued fewer steps compared to younger patients. In patients with longer length of stay, the level of ambulation at discharge was similar to patients with shorter stays, suggesting there may be an ambulation threshold (1100 steps) that patients achieve regardless of the length of stay before they are discharged home. In addition, patients whose care team reviewed and documented step counts at least 3 times over a 24‐hour period accrued significantly more steps than patients whose care team made fewer documentations.

The median step counts accrued by surgical patients in our study are similar to that found in Cook and colleagues'[14] report of patients after elective cardiac surgery using another popular consumer‐grade accelerometer. The providers in that study also had access to the data via a dashboard, but it was not clear how this information was used. Brown et al.[12] conducted the first study to objectively monitor mobility using 2 accelerometers in 45 older male veterans who had no prior mobility impairment, and found that patients spent 83% of their hospitalization lying in bed. The veterans spent about 3% of the time (43 minutes per day) standing or walking over a mean length of stay of 5 days. In a similar study with 43 older Dutch patients who had an average length of stay of 7 days, Pedersen et al.[10] found that patients spent 71% of their time lying, 21% sitting, and 4% standing or walking. Unfortunately, neither the Brown et al. nor Pedersen et al. studies were able to distinguish between standing and ambulatory activities. In a more recent study of 47 patients on medical‐surgical units at 2 hospitals that relied on time and motion observation methods, the mean duration for ambulation was <2 minutes during an 8‐hour period.[13]

We took advantage of the variability in the nursing documentation of step counts in the EMR to determine if there was a dose‐response relationship between the frequency of nursing documentation in a 24‐hour period and number of steps patients accrued. We hypothesized that if nurses make an effort to retrieve data from the vendor website and manually key in the step counts in the EMR, they are more likely to incorporate this information in their nursing care, share the information with patients and other clinicians, and therefore create a positive feedback loop for greater ambulation. Although our findings suggest a positive association between more frequent documentation and increased step counts, we cannot exclude the possibility that nurses naturally modulate the frequency with which they review and document step counts based on their overall judgment of the patients' mobility status (ie, patients who are more functionally impaired are assumed to accrue fewer steps over a shift, and therefore, nurses are less inclined to retrieve and document the information frequently). Future studies could prospectively examine what the optimal frequency for review and feedback of step counts is during a typical 8‐ or 12‐hour nursing shift for both patients and the nursing care team to promote ambulation.

A major strength of our study is the collection of objective ambulation data on a large inpatient sample by clinical staff as part of routine nursing care. This strength is balanced with several limitations. Due to the temporal pattern associated with ambulation, we were only able to analyze data for patients who had at least 24 hours of continuous monitoring. This could affect the generalizability of our findings, though we believe there is limited pragmatic value in closely tracking ambulation in patients who have such short stays. There was substantial variability in the step counts, reflecting the mix of medical versus surgical patients and their age, with very small samples available for meaningful subgroup analyses other than what we have presented. We were not able to measure other dimensions of mobility such as transfers or sitting in a chair, because the sensor is designed to only measure steps. In addition, we lost a large number of devices, mostly due to staff forgetting to remove the devices from patients' ankles at discharge. Finally, because we did not blind the nurses and patients to the step count data, the preliminary normative step counts that we present in this article may be higher than expected in patients cared for on medical‐surgical units.

In summary, we found that it is possible to measure ambulation objectively and reliably in hospitalized patients, and have provided preliminary normative step counts for a representative but heterogeneous medical‐surgical population. We also found that most patients who were discharged were ambulating at least 1100 steps over the 24 hours prior to leaving the hospital, regardless of their length of stay. This might suggest that step counts could be a useful parameter in determining readiness for hospital discharge. Our data also suggest that more frequent, objective monitoring of step counts by the nursing care team was associated with patients ambulating more. Both of these findings deserve further exploration. Future studies will need to be conducted on larger samples of medical and surgical hospitalized patients to adequately establish more refined step count norms for specific clinical populations, but especially for older patients, because this age group is at a particularly higher risk of poor functional outcomes with hospitalization. Having accurate and reliable information on ambulation is fundamental to any effort to improve ambulation in hospitalized patients. Moreover, knowing the normative range for step counts in the last 24 hours prior to discharge across specific clinical and age subgroups, could assist with discharge planning and provision of appropriate rehabilitative services in the home or community for safe transitions out of the hospital.[17]

A number of observational studies have documented the association between prolonged bed rest during hospitalization with adverse short‐ and long‐term functional impairments and disability in older patients.[1, 2, 3, 4] However, the body of evidence on the benefits of early mobilization on functional outcomes in both critically ill patients and more stable patients on medical‐surgical floors remains inconclusive.[5, 6, 7, 8, 9] Despite the increased emphasis on mobilizing patients early and often in the inpatient setting, there is surprisingly little information available regarding how typically active adult patients are during their hospital stay. The few published studies that are available are limited by small samples and types of patients who were monitored.[10, 11, 12, 13, 14] Therefore, the purpose of this real‐world study was to describe the level of ambulation in a large sample of hospitalized adult patients using a validated consumer‐grade wireless accelerometer.

METHODS

This was a prospective cohort study of ambulatory patients from 3 medical‐surgical units of a community hospital from March 2014 through July 2014. The study was approved by the Kaiser Permanente Southern California Institutional Review Board. All ambulatory medical and surgical adult patients were eligible for the study except for those with isolation precautions. Patients wore an accelerometer (Tractivity; Kineteks Corp., Vancouver, BC, Canada) on the ankle from soon after admission to the unit until discharge home. The sensors were only removed for bathing and medical procedures, at which time the devices were secured to the patient's bed and reworn upon their return to the room. The nursing staff was trained to use the vendor application to register the sensor to the patient, secure the sensor to the patient's ankle, transfer the sensor data to the vendor server, review the step counts on the web application, and manually key the step count into the electronic medical records (EMRs) as part of routine nursing workflow. The staff otherwise continued with usual patient mobilization practices.

We previously validated the Tractivity device in a field study of 20 hospitalized patients using a research‐grade accelerometer, Stepwatch, as the gold standard (unpublished data). We found that the inter‐Tractivity device reliability was near perfect (intraclass correlation=0.99), and that the Tractivity step counts correlated highly with the nurses' documentation on a paper log of distance walked measured in feet (r=0.76). A small number of steps (<100) were recorded over 24 hours when the device was worn by 2 bed bound patients. The 24‐hour Tractivity step count had acceptable limits of agreement with the Stepwatch (+284 [standard deviation: 314] steps; 95% limits of agreement 911‐343). In addition, for the current study, when we examined the step counts between patients who were classified by the nursing team as being able to walk <50 feet (n=320) compared to patients who were able to walk >50 feet (n=434), we found a significant difference in the median number of steps over a 24‐hour period (854 vs 1697, P<0.0001).

The step count data were exported from the vendor's server, examined for irregularities, and merged with administrative and clinical data for analysis. Data extracted from the EMR system included sociodemographic (age, gender, marital status, and race/ethnicity) and clinical characteristics (LACE score [readmission risk score based on length of stay (L); acuity of the admission (A); comorbidity of the patient (measured with the Charlson comorbidity index score) (C); and emergency department use (measured as the number of visits in the six months before admission) (E),[15] Charlson Comorbidity Index, length of stay, principal discharge diagnosis, and body mass index), and nursing documentation of functional status (bed bound, sit up in bed, stand next to bed, walk <50 feet, and walk >50 feet).

Descriptive statistics and nonparametric tests (Kruskal‐Wallis and Wilcoxon signed rank) were used to analyze the non‐normally distributed step count data. Quantile regression[16] was used to determine the association between the frequency of the care team's review and documentation of steps, with median total step count adjusting for age, gender, LACE score, and medicine/surgical service line. Whereas linear regression allows one to describe how the mean of a given outcome changes with respect to some set of covariates in circumstances where data are normally distributed, quantile regression allows one to assess how a set of covariates are related to a prespecified quantile (eg, 50% percentile median) of an outcome distribution. This modeling is especially appropriate here, because step count data are not normally distributed. Because step counts can vary with a number of factors, such as age and principal admitting and discharge diagnoses, we stratified our analyses by age (<65 or 65 years) and service lines (medical or surgical) due to the relatively small numbers of patients in each of the diagnostic groupings. Statistical analyses were performed using SAS version 9.3 (SAS Institute Inc., Cary, NC); P values <0.05 were considered statistically significant.

RESULTS

A total of 1667 patients wore the activity sensor during their hospital stay. We included 777 patients in our analysis who had lengths of stay long enough for 24 hours of continuous monitoring, and almost half of these patients had at least 48 hours of monitoring (n=378). The demographic and clinical characteristics of the sample are detailed in Table 1. The sample included mostly medical patients (77%), with a mean age of 6017 years, 57% females, and 55% nonwhites. Nearly all patients (97%) were classified as ambulatory at discharge based on the EMR data. Approximately 44% of the sensors were lost, mostly due to nursing staff forgetting to remove the devices at discharge; device failure was minimal (n=10).

Patients accrued a median of 1158 (interquartile range: 6362238) steps over the 24 hours prior to discharge to home (Table 2). Approximately 13 (2%) patients registered zero steps in the last 24 hours; this may have been due to patients truly not accruing any steps, device failure, or the device was registered but never worn by the patient. Patients who were 65 years and older on both the medicine and surgical services accrued fewer steps compared to younger patients (962 vs 1294, P<0.0001). For patients who had at least 48 hours of continuous monitoring (n=378), there was a median increase of 377 steps from the first 24 hours from admission to the unit to the final 24 hours prior to discharge (811 steps to 1188 steps, P<0.0001) (Table 3 and Figure 1). The average length of stay for these patients was 5.74.9 days. Despite the longer length of stay, the level of ambulation at discharge was similar to patients with shorter stays. This is further illustrated in Figure 2 in the spaghetti plots of total steps over 4, 24‐hour monitoring increments. Ignoring the outliers, the plots suggest the following: (1) step counts tended to increase or stay about the same over the course of a hospitalization; and (2) for the medicine service line, step counts in the final 24 hours prior to discharge for patients with longer lengths of stay (72 or 96 hours) did not appear to be substantially different from patients with shorter lengths of stay. The data for the surgical patients are either too sparse or erratic to make any firm conclusions. Patients accrued steps throughout the day with the highest percentage of steps logged at approximately 6 _am and 6 _pm; these data are based on time stamps from the device, not the time of data transfer or documentation in the EMR (Figure 3).

Figure 1

Figure 2

Figure 3

More frequent documentation of step counts in the EMR (proxy for step count data retrieval and review from the vendor web site) by the care team was associated with higher total step counts after adjustments for relevant covariates (P0.001); 3 or more documentations over a 24‐hour period appears to be a minimal frequency to achieving approximately 200 steps more than the median value (Table 4).

DISCUSSION

We found that ambulatory medical‐surgical patients accrued a median of 1158 total steps in the 24 hours prior to their discharge home, which translates to walking approximately 500 meters; older patients accrued fewer steps compared to younger patients. In patients with longer length of stay, the level of ambulation at discharge was similar to patients with shorter stays, suggesting there may be an ambulation threshold (1100 steps) that patients achieve regardless of the length of stay before they are discharged home. In addition, patients whose care team reviewed and documented step counts at least 3 times over a 24‐hour period accrued significantly more steps than patients whose care team made fewer documentations.

The median step counts accrued by surgical patients in our study are similar to that found in Cook and colleagues'[14] report of patients after elective cardiac surgery using another popular consumer‐grade accelerometer. The providers in that study also had access to the data via a dashboard, but it was not clear how this information was used. Brown et al.[12] conducted the first study to objectively monitor mobility using 2 accelerometers in 45 older male veterans who had no prior mobility impairment, and found that patients spent 83% of their hospitalization lying in bed. The veterans spent about 3% of the time (43 minutes per day) standing or walking over a mean length of stay of 5 days. In a similar study with 43 older Dutch patients who had an average length of stay of 7 days, Pedersen et al.[10] found that patients spent 71% of their time lying, 21% sitting, and 4% standing or walking. Unfortunately, neither the Brown et al. nor Pedersen et al. studies were able to distinguish between standing and ambulatory activities. In a more recent study of 47 patients on medical‐surgical units at 2 hospitals that relied on time and motion observation methods, the mean duration for ambulation was <2 minutes during an 8‐hour period.[13]

We took advantage of the variability in the nursing documentation of step counts in the EMR to determine if there was a dose‐response relationship between the frequency of nursing documentation in a 24‐hour period and number of steps patients accrued. We hypothesized that if nurses make an effort to retrieve data from the vendor website and manually key in the step counts in the EMR, they are more likely to incorporate this information in their nursing care, share the information with patients and other clinicians, and therefore create a positive feedback loop for greater ambulation. Although our findings suggest a positive association between more frequent documentation and increased step counts, we cannot exclude the possibility that nurses naturally modulate the frequency with which they review and document step counts based on their overall judgment of the patients' mobility status (ie, patients who are more functionally impaired are assumed to accrue fewer steps over a shift, and therefore, nurses are less inclined to retrieve and document the information frequently). Future studies could prospectively examine what the optimal frequency for review and feedback of step counts is during a typical 8‐ or 12‐hour nursing shift for both patients and the nursing care team to promote ambulation.

A major strength of our study is the collection of objective ambulation data on a large inpatient sample by clinical staff as part of routine nursing care. This strength is balanced with several limitations. Due to the temporal pattern associated with ambulation, we were only able to analyze data for patients who had at least 24 hours of continuous monitoring. This could affect the generalizability of our findings, though we believe there is limited pragmatic value in closely tracking ambulation in patients who have such short stays. There was substantial variability in the step counts, reflecting the mix of medical versus surgical patients and their age, with very small samples available for meaningful subgroup analyses other than what we have presented. We were not able to measure other dimensions of mobility such as transfers or sitting in a chair, because the sensor is designed to only measure steps. In addition, we lost a large number of devices, mostly due to staff forgetting to remove the devices from patients' ankles at discharge. Finally, because we did not blind the nurses and patients to the step count data, the preliminary normative step counts that we present in this article may be higher than expected in patients cared for on medical‐surgical units.

In summary, we found that it is possible to measure ambulation objectively and reliably in hospitalized patients, and have provided preliminary normative step counts for a representative but heterogeneous medical‐surgical population. We also found that most patients who were discharged were ambulating at least 1100 steps over the 24 hours prior to leaving the hospital, regardless of their length of stay. This might suggest that step counts could be a useful parameter in determining readiness for hospital discharge. Our data also suggest that more frequent, objective monitoring of step counts by the nursing care team was associated with patients ambulating more. Both of these findings deserve further exploration. Future studies will need to be conducted on larger samples of medical and surgical hospitalized patients to adequately establish more refined step count norms for specific clinical populations, but especially for older patients, because this age group is at a particularly higher risk of poor functional outcomes with hospitalization. Having accurate and reliable information on ambulation is fundamental to any effort to improve ambulation in hospitalized patients. Moreover, knowing the normative range for step counts in the last 24 hours prior to discharge across specific clinical and age subgroups, could assist with discharge planning and provision of appropriate rehabilitative services in the home or community for safe transitions out of the hospital.[17]

A number of observational studies have documented the association between prolonged bed rest during hospitalization with adverse short‐ and long‐term functional impairments and disability in older patients.[1, 2, 3, 4] However, the body of evidence on the benefits of early mobilization on functional outcomes in both critically ill patients and more stable patients on medical‐surgical floors remains inconclusive.[5, 6, 7, 8, 9] Despite the increased emphasis on mobilizing patients early and often in the inpatient setting, there is surprisingly little information available regarding how typically active adult patients are during their hospital stay. The few published studies that are available are limited by small samples and types of patients who were monitored.[10, 11, 12, 13, 14] Therefore, the purpose of this real‐world study was to describe the level of ambulation in a large sample of hospitalized adult patients using a validated consumer‐grade wireless accelerometer.

METHODS

This was a prospective cohort study of ambulatory patients from 3 medical‐surgical units of a community hospital from March 2014 through July 2014. The study was approved by the Kaiser Permanente Southern California Institutional Review Board. All ambulatory medical and surgical adult patients were eligible for the study except for those with isolation precautions. Patients wore an accelerometer (Tractivity; Kineteks Corp., Vancouver, BC, Canada) on the ankle from soon after admission to the unit until discharge home. The sensors were only removed for bathing and medical procedures, at which time the devices were secured to the patient's bed and reworn upon their return to the room. The nursing staff was trained to use the vendor application to register the sensor to the patient, secure the sensor to the patient's ankle, transfer the sensor data to the vendor server, review the step counts on the web application, and manually key the step count into the electronic medical records (EMRs) as part of routine nursing workflow. The staff otherwise continued with usual patient mobilization practices.

We previously validated the Tractivity device in a field study of 20 hospitalized patients using a research‐grade accelerometer, Stepwatch, as the gold standard (unpublished data). We found that the inter‐Tractivity device reliability was near perfect (intraclass correlation=0.99), and that the Tractivity step counts correlated highly with the nurses' documentation on a paper log of distance walked measured in feet (r=0.76). A small number of steps (<100) were recorded over 24 hours when the device was worn by 2 bed bound patients. The 24‐hour Tractivity step count had acceptable limits of agreement with the Stepwatch (+284 [standard deviation: 314] steps; 95% limits of agreement 911‐343). In addition, for the current study, when we examined the step counts between patients who were classified by the nursing team as being able to walk <50 feet (n=320) compared to patients who were able to walk >50 feet (n=434), we found a significant difference in the median number of steps over a 24‐hour period (854 vs 1697, P<0.0001).

The step count data were exported from the vendor's server, examined for irregularities, and merged with administrative and clinical data for analysis. Data extracted from the EMR system included sociodemographic (age, gender, marital status, and race/ethnicity) and clinical characteristics (LACE score [readmission risk score based on length of stay (L); acuity of the admission (A); comorbidity of the patient (measured with the Charlson comorbidity index score) (C); and emergency department use (measured as the number of visits in the six months before admission) (E),[15] Charlson Comorbidity Index, length of stay, principal discharge diagnosis, and body mass index), and nursing documentation of functional status (bed bound, sit up in bed, stand next to bed, walk <50 feet, and walk >50 feet).

Descriptive statistics and nonparametric tests (Kruskal‐Wallis and Wilcoxon signed rank) were used to analyze the non‐normally distributed step count data. Quantile regression[16] was used to determine the association between the frequency of the care team's review and documentation of steps, with median total step count adjusting for age, gender, LACE score, and medicine/surgical service line. Whereas linear regression allows one to describe how the mean of a given outcome changes with respect to some set of covariates in circumstances where data are normally distributed, quantile regression allows one to assess how a set of covariates are related to a prespecified quantile (eg, 50% percentile median) of an outcome distribution. This modeling is especially appropriate here, because step count data are not normally distributed. Because step counts can vary with a number of factors, such as age and principal admitting and discharge diagnoses, we stratified our analyses by age (<65 or 65 years) and service lines (medical or surgical) due to the relatively small numbers of patients in each of the diagnostic groupings. Statistical analyses were performed using SAS version 9.3 (SAS Institute Inc., Cary, NC); P values <0.05 were considered statistically significant.

RESULTS

A total of 1667 patients wore the activity sensor during their hospital stay. We included 777 patients in our analysis who had lengths of stay long enough for 24 hours of continuous monitoring, and almost half of these patients had at least 48 hours of monitoring (n=378). The demographic and clinical characteristics of the sample are detailed in Table 1. The sample included mostly medical patients (77%), with a mean age of 6017 years, 57% females, and 55% nonwhites. Nearly all patients (97%) were classified as ambulatory at discharge based on the EMR data. Approximately 44% of the sensors were lost, mostly due to nursing staff forgetting to remove the devices at discharge; device failure was minimal (n=10).

Patients accrued a median of 1158 (interquartile range: 6362238) steps over the 24 hours prior to discharge to home (Table 2). Approximately 13 (2%) patients registered zero steps in the last 24 hours; this may have been due to patients truly not accruing any steps, device failure, or the device was registered but never worn by the patient. Patients who were 65 years and older on both the medicine and surgical services accrued fewer steps compared to younger patients (962 vs 1294, P<0.0001). For patients who had at least 48 hours of continuous monitoring (n=378), there was a median increase of 377 steps from the first 24 hours from admission to the unit to the final 24 hours prior to discharge (811 steps to 1188 steps, P<0.0001) (Table 3 and Figure 1). The average length of stay for these patients was 5.74.9 days. Despite the longer length of stay, the level of ambulation at discharge was similar to patients with shorter stays. This is further illustrated in Figure 2 in the spaghetti plots of total steps over 4, 24‐hour monitoring increments. Ignoring the outliers, the plots suggest the following: (1) step counts tended to increase or stay about the same over the course of a hospitalization; and (2) for the medicine service line, step counts in the final 24 hours prior to discharge for patients with longer lengths of stay (72 or 96 hours) did not appear to be substantially different from patients with shorter lengths of stay. The data for the surgical patients are either too sparse or erratic to make any firm conclusions. Patients accrued steps throughout the day with the highest percentage of steps logged at approximately 6 _am and 6 _pm; these data are based on time stamps from the device, not the time of data transfer or documentation in the EMR (Figure 3).

Figure 1

Figure 2

Figure 3

More frequent documentation of step counts in the EMR (proxy for step count data retrieval and review from the vendor web site) by the care team was associated with higher total step counts after adjustments for relevant covariates (P0.001); 3 or more documentations over a 24‐hour period appears to be a minimal frequency to achieving approximately 200 steps more than the median value (Table 4).

DISCUSSION

We found that ambulatory medical‐surgical patients accrued a median of 1158 total steps in the 24 hours prior to their discharge home, which translates to walking approximately 500 meters; older patients accrued fewer steps compared to younger patients. In patients with longer length of stay, the level of ambulation at discharge was similar to patients with shorter stays, suggesting there may be an ambulation threshold (1100 steps) that patients achieve regardless of the length of stay before they are discharged home. In addition, patients whose care team reviewed and documented step counts at least 3 times over a 24‐hour period accrued significantly more steps than patients whose care team made fewer documentations.

The median step counts accrued by surgical patients in our study are similar to that found in Cook and colleagues'[14] report of patients after elective cardiac surgery using another popular consumer‐grade accelerometer. The providers in that study also had access to the data via a dashboard, but it was not clear how this information was used. Brown et al.[12] conducted the first study to objectively monitor mobility using 2 accelerometers in 45 older male veterans who had no prior mobility impairment, and found that patients spent 83% of their hospitalization lying in bed. The veterans spent about 3% of the time (43 minutes per day) standing or walking over a mean length of stay of 5 days. In a similar study with 43 older Dutch patients who had an average length of stay of 7 days, Pedersen et al.[10] found that patients spent 71% of their time lying, 21% sitting, and 4% standing or walking. Unfortunately, neither the Brown et al. nor Pedersen et al. studies were able to distinguish between standing and ambulatory activities. In a more recent study of 47 patients on medical‐surgical units at 2 hospitals that relied on time and motion observation methods, the mean duration for ambulation was <2 minutes during an 8‐hour period.[13]

We took advantage of the variability in the nursing documentation of step counts in the EMR to determine if there was a dose‐response relationship between the frequency of nursing documentation in a 24‐hour period and number of steps patients accrued. We hypothesized that if nurses make an effort to retrieve data from the vendor website and manually key in the step counts in the EMR, they are more likely to incorporate this information in their nursing care, share the information with patients and other clinicians, and therefore create a positive feedback loop for greater ambulation. Although our findings suggest a positive association between more frequent documentation and increased step counts, we cannot exclude the possibility that nurses naturally modulate the frequency with which they review and document step counts based on their overall judgment of the patients' mobility status (ie, patients who are more functionally impaired are assumed to accrue fewer steps over a shift, and therefore, nurses are less inclined to retrieve and document the information frequently). Future studies could prospectively examine what the optimal frequency for review and feedback of step counts is during a typical 8‐ or 12‐hour nursing shift for both patients and the nursing care team to promote ambulation.

A major strength of our study is the collection of objective ambulation data on a large inpatient sample by clinical staff as part of routine nursing care. This strength is balanced with several limitations. Due to the temporal pattern associated with ambulation, we were only able to analyze data for patients who had at least 24 hours of continuous monitoring. This could affect the generalizability of our findings, though we believe there is limited pragmatic value in closely tracking ambulation in patients who have such short stays. There was substantial variability in the step counts, reflecting the mix of medical versus surgical patients and their age, with very small samples available for meaningful subgroup analyses other than what we have presented. We were not able to measure other dimensions of mobility such as transfers or sitting in a chair, because the sensor is designed to only measure steps. In addition, we lost a large number of devices, mostly due to staff forgetting to remove the devices from patients' ankles at discharge. Finally, because we did not blind the nurses and patients to the step count data, the preliminary normative step counts that we present in this article may be higher than expected in patients cared for on medical‐surgical units.

In summary, we found that it is possible to measure ambulation objectively and reliably in hospitalized patients, and have provided preliminary normative step counts for a representative but heterogeneous medical‐surgical population. We also found that most patients who were discharged were ambulating at least 1100 steps over the 24 hours prior to leaving the hospital, regardless of their length of stay. This might suggest that step counts could be a useful parameter in determining readiness for hospital discharge. Our data also suggest that more frequent, objective monitoring of step counts by the nursing care team was associated with patients ambulating more. Both of these findings deserve further exploration. Future studies will need to be conducted on larger samples of medical and surgical hospitalized patients to adequately establish more refined step count norms for specific clinical populations, but especially for older patients, because this age group is at a particularly higher risk of poor functional outcomes with hospitalization. Having accurate and reliable information on ambulation is fundamental to any effort to improve ambulation in hospitalized patients. Moreover, knowing the normative range for step counts in the last 24 hours prior to discharge across specific clinical and age subgroups, could assist with discharge planning and provision of appropriate rehabilitative services in the home or community for safe transitions out of the hospital.[17]

Transitions of care from hospital to home are high‐risk times for patients.[1, 2] Increasing complexity of hospital admissions and shorter lengths of stay demand more effective coordination of care between hospitalists and outpatient clinicians.[3, 4, 5] Inaccurate, delayed, or incomplete clinical handoversthat is, transfer of information and professional responsibility and accountability[6]can lead to patient harm, and has been recognized as a key cause of preventable morbidity by the World Health Organization and The Joint Commission.[6, 7, 8] Conversely, when done effectively, transitions can result in improved patient health outcomes, reduced readmission rates, and higher patient and provider satisfaction.³

Previous studies note deficits in communication at discharge and primary care provider (PCP) dissatisfaction with discharge practices.[4, 9, 10, 11, 12, 13] In studies at academic medical centers, there were low rates of direct communication between inpatient and outpatient providers, mainly because of providers' belief that the discharge summary was adequate and the presence of significant barriers to direct communication.[14, 15] However, studies have shown that discharge summaries often omit critical information, and often are not available to PCPs in a timely manner.[10, 11, 12, 16] In response, the Society of Hospital Medicine developed a discharge checklist to aide in standardization of safe discharge practices.[1, 5] Discharge summary templates further attempt to improve documentation of patients' hospital courses. An electronic medical record (EMR) system shared by both inpatient and outpatient clinicians should impart several advantages: (1) automated alerts provide timely notification to PCPs regarding admission and discharge, (2) discharge summaries are available to the PCP as soon as they are written, and (3) all patient information pertaining to the hospitalization is available to the PCP.

Although it is plausible that shared EMRs should facilitate transitions of care by streamlining communication between hospitalists and PCPs, guidelines on format and content of PCP communication at discharge in the era of a shared EMR have yet to be defined. In this study, we sought to understand current discharge communication practices and PCP satisfaction within a shared EMR at our institution, and to identify key areas in which communication can be improved.

METHODS

RESULTS

Seventy‐five of 124 (60%) clinicians (46% residents, 54% attendings) completed the survey. Thirty‐nine (52%) PCPs were satisfied or very satisfied with communication at patient discharge. Although most reported receiving automated discharge notifications (71%), only 39% felt that the notifications plus the discharge summaries were adequate communication for safe transition of care from hospital to community. Fifty‐one percent desired direct contact beyond a discharge summary. There were no differences in preferences on discharge communication between resident and attending PCPs (P>0.05).

Over three‐fourths of PCPs surveyed preferred that for patients with complex hospitalizations (multiple readmissions, multiple active comorbidities, goals of care changes, high‐risk medication changes, time‐sensitive follow‐up needs), an additional e‐mail or verbal communication was needed to augment the information in the discharge summary (Figure 1). Only 31% reported receiving such communication.

Figure 1

When asked about important items to communicate for safer transitions of care, PCPs reported finding the following elements most critical: (1) medication changes (93%), (2) follow‐up actions for the PCP (88%), and (3) active medical issues (84%) (Figure 2).

Figure 2

CONCLUSIONS

In the era of shared EMRs, real‐time access to medication lists, pending test results, and discharge summaries should facilitate care transitions at discharge.[18, 19] We conducted a study to determine PCP perceptions of discharge communication after implementation of a shared EMR. We found that although PCPs largely acknowledged timely receipt of automated discharge notifications and discharge summaries, the majority of PCPs felt that most discharges required additional communication to ensure safe transition of care.

Guidelines for discharge communication emphasize timely communication with the PCP, primarily through discharge summaries containing key safety elements.[1, 5, 10] At our institution, we have improved the timeliness and quality of discharge summaries according to guideline recommendations,[17] and conducted quality improvement projects to improve rates of direct communication with PCPs.[9] In addition, the shared EMR provides automated notifications to PCPs when their patients are discharged. Despite these interventions, our survey shows that PCP satisfaction with discharge communication is still inadequate. PCPs desired direct communication that highlights active medical issues, medication changes, and specific follow‐up actions. Although all of these topics are included in our discharge summary template (see Supporting Information, Appendix 1, in the online version of this article), it is possible that the templated discharge summaries lend themselves to longer documents and information overload, as prior studies have documented the desire for more succinct discharge summaries.[18] We also found that automated notifications of discharge were less reliable and useful for PCPs than anticipated. There were several reasons for this: (1) discharge summaries sometimes were sent to PCPs uncoupled from the discharge notification, (2) there were errors with the generation and delivery of automated messages at the rollout of the new system, and (3) PCPs received many other automated system messages, meaning that discharge notifications could be easily missed. These factors all likely contribute to PCPs' desire for high‐touch communication that highlights the most salient aspects of each patient's hospitalization. It is also possible that automated notifications and depersonalized discharge summaries create distance and a less‐collaborative feeling to patient care. PCPs want more direct communication, and desire to play a more active role in inpatient management, especially for complex hospitalizations.[18] This emphasis on direct communication resonates with previous studies conducted before shared EMRs existed.[9, 12, 19]

Our study had several limitations. First, because this is a single‐institution study at a tertiary care academic setting, the results may not be generalizable to all shared EMR settings, and may not reflect all the challenges of communication with the wider community of outpatient providers. One can postulate that inpatient and outpatient clinician relationships are stronger in an academic setting than in other more disparate environments, where direct communication may happen even less frequently. Of note, our low rates of direct communication are consistent with other single‐ and multi‐institution studies, suggesting that our findings are generalizable.[14, 15] Second, our survey is limited in its ability to distinguish those patients who require high‐touch communication and those who do not. Third, although we have used the survey to assess PCP satisfaction in previous studies, it is not a validated instrument, and therefore we cannot reliably say that increasing direct PCP communication would increase their satisfaction around discharge. Last, the PCP‐reported rates of discharge communication are subjective and may be influenced by recall bias. We did not have a systematic way to confirm the actual rates of communication at discharge, which could have occurred through EMR messages, e‐mails, phone calls, or pages.

Although a shared EMR allows for real‐time access to patient data, it does not eliminate PCPs' desire for direct 2‐way dialogue at discharge, especially for complex patients. Key information desired in such communication should include active medical issues, medication changes, and follow‐up needs, which is consistent with prior studies. Standardizing this direct communication process in an efficient way can be challenging. Further elucidation of PCP preferences around which patients necessitate higher‐level communication and preferred methods and timing of communication is needed, as well as determining the most efficient and effective method for hospitalists to provide such communication. Improving communication between hospitalists and PCPs requires not just the presence of a shared EMR, but additional, systematic efforts to engage both inpatient and outpatient clinicians in collaborative care.