Hospital-associated disability due to avoidable hospitalizations among older adults

Abstract

Background:

Hospital-associated disability (HAD) is a common complication during the course of acute care hospitalizations in older adults. Many admissions are for ambulatory care sensitive conditions (ACSCs), considered potentially avoidable hospitalizations—conditions that might be treated in outpatient settings to prevent hospitalization and HAD. We compared the incidence of HAD between older adults hospitalized for ACSCs versus those hospitalized for other diagnoses.

Methods:

We conducted a retrospective cohort study in inpatient (non-ICU) medical and surgical units of a large southeastern regional academic medical center. Participants were 38,960 older adults ≥ 65 years of age admitted from January 1, 2015, to December 31, 2019. The primary outcome was HAD, defined as decline on the Katz Activities of Daily Living (ADL) scale from hospital admission to discharge. We used generalized linear mixed models to examine differences in HAD between hospitalizations with a primary diagnosis for an ACSC using standard definitions versus primary diagnosis for other conditions, adjusting for covariates and repeated observations for individuals with multiple hospitalizations.

Results:

We found that 10% of older adults were admitted for an ACSC, with rates of HAD in those admitted for ACSCs lower than those admitted for other conditions (16% vs. 20.7%, p < 0.001). Age, comorbidity, admission functional status, and admission cognitive impairment were significant predictors for development of HAD. ACSC admissions to medical and medical/surgical services had lower odds of HAD compared with admissions for other conditions, with no significant differences between ACSC and non-ACSC admissions to surgical services.

Conclusions:

Rates of HAD among older adults hospitalized for ACSCs are substantial, though lower than rates of HAD with hospitalization for other conditions, reflecting that acute care hospitalization is not a benign event in this population. Treatment of ACSCs in the outpatient setting could be an important component of efforts to reduce HAD.

INTRODUCTION

It is now widely recognized that acute care hospitalization among older adults accelerates functional decline, which can lead to chronic physical impairment and ultimately mortality.^1-5 This hospital-associated disability (HAD), defined as a new loss of ability to complete one or more activities of daily living (ADLs) independently after hospitalization, is prevalent in nearly 30% of acute care admissions among older adults.⁶ Interventions such as the Hospital Elder Life Program (HELP) and the Acute Care for Elders (ACE) model have been shown to reduce functional disability among hospitalized older patents.^7-9 However, a recent meta-analysis demonstrated that the incidence rate of HAD has not significantly decreased over the last 30 years, despite the development of programs such as HELP and ACE.¹⁰ The identification of novel modifiable risk factors for HAD is a potential way to develop additional approaches to prevent or mitigate HAD.

Ambulatory care sensitive conditions (ACSCs) are those conditions that can be treated with outpatient primary care management, which provides an intervenable opportunity to avoid hospital admission.¹¹ ACSCs encompass both conditions where acute decline can be avoided entirely with better preventive care, and conditions where early stages of acute decline can be managed on an outpatient basis to avoid hospitalization. ACSCs may be acute conditions (e.g., pneumonia, urinary tract infections) or exacerbations of chronic conditions (e.g., congestive heart failure, chronic obstructive pulmonary disease),¹² and are gaining increased attention in acute care as a measure of healthcare quality and accessibility.¹³ Although originally conceptualized as a measure of healthcare access in younger disadvantaged populations,¹⁴ ACSCs are particularly relevant to the older adult population, who experience the highest rate of hospitalizations¹⁵ and hospitalizations for ACSCs.¹⁶ By identifying ACSCs, hospitalizations could be avoided with early intervention in the primary care setting, thus potentially decreasing the incidence of HAD and subsequent loss of independence. Therefore, the main objective of this study is to compare the incidence of HAD between older adults hospitalized for ACSCs versus those hospitalized for other diagnoses.

METHODS

Study design

This study was a retrospective cohort analysis of inpatient hospital admissions to examine rates of HAD for ACSC-related hospitalizations versus hospitalizations for other conditions.

Study site and patient population

Participant data were collected from the electronic health records (EHR) of the University of Alabama at Birmingham (UAB) Hospital, a large academic medical center in the Southeastern US. Data were collected for the period from January 1, 2015, to December 31, 2019.

We deliberately kept our inclusion criteria broad to allow maximal generalization to the hospitalized older adult population. Inclusion criteria for this study were (1) age ≥ 65; (2) admission to one of the medical or surgical units on inpatient or observation status during the study period; (3) no admission or transfer to a designated intensive care unit (ICU) at any point during acute hospitalization, as procedures common to the ICU (such as sedation and paralysis) would likely increase the risk of HAD; (4) assessment of admission (pre-hospitalization) Katz ADL.

Outcome

The primary outcome for this study is the development of HAD, defined as a decrease in the Katz Index¹⁷ of ADLs between hospital admission and discharge, consistent with the minimal important change (MIC) of 0.5 point on this scale.¹⁸ The six activities included in the Katz Index are: bathing, dressing, toileting, transferring, continence, and feeding. Each item is scored as 0 = completely dependent, 1 = partially dependent, and 2 = completely independent, yielding a total score of 0–12 with higher scores indicating greater independence. The Katz Index is collected by nursing staff at admission, assessing the patient's functional status immediately before hospitalization, and every 3 days thereafter under our Virtual ACE quality improvement project,¹⁹ and is part of the UAB EHR. Nursing staff collecting data for Virtual ACE undergo rigorous training through the hospital's geriatric quality improvement program to ensure reliability and validity.

Individuals having only an admission Katz Index score were defined as not having HAD as change in Katz Index scores could not be documented. Based on the schedule for administration of the Katz Index, a primary reason for only a single Katz Index would be shorter length of stay, and exclusion of such individuals would preferentially select more severely ill individuals. However, to examine potential biases by including individuals with a single Katz Index where change (or lack of change) in ADLs could not be documented, we conducted sensitivity analyses requiring two or more Katz Index scores for determination of HAD. We also conducted sensitivity analyses excluding hospitalizations with the lowest 10% of admission Katz scores to exclude potential floor effects where ADLs were too low to decline further. Inhospital death was classified as discharge from the hospital. Thus, if ADLs declined before death, the participant was classified as developing HAD; if ADLs did not decline before death, or only one assessment was performed, the participant was classified as not developing HAD.

In a random sample of 49 older patients admitted to the general medical and surgical wards, we blindly assessed Katz ADL scores within 24 h of admission. The total scores obtained by a trained research assistant (RA) were compared with the nurses' scores documented in Virtual ACE during the same period. We found a weighted kappa of 0.87 (95% CI 0.81, 0.93) between nurses and the RA, indicating the Katz ADL total scores provided by the nurses are reliable.

Predictor

We used the primary diagnosis responsible for hospitalization to categorize the stay as ACSC (+) or (−) based on the ACSC definition and associated condition list by Purdy and colleagues (Tables S1 and S2).²⁰ This commonly used definition focuses on 19 conditions that are frequently seen in the National Health Service in England and could be used to measure the effectiveness of primary care for the general population in that system. Because varying definitions of ACSC have been proposed,^21-23 we also performed sensitivity analyses to determine if different published ACSC definitions would alter our conclusions (see Supplemental Methods (Data S1) for other definitions considered).

Covariates

Demographics

Demographic characteristics of age in years, gender, and race were based on self-reported data recorded in the EHR.

Dementia

Dementia is a risk factor for HAD,²⁴ and was included as a predictor in all models. All-cause dementia was determined using ICD-10 diagnostic codes, based on the ICD-9 codes of Goodman and colleagues²⁵ and classified as present or absent.

Comorbidity

As comorbidity is a recognized risk factor for HAD,²⁴ we included the Elixhauser comorbidity index²⁶ as a predictor in our models. The Elixhauser comorbidity index was calculated using ICD-10 diagnostic codes, based on the algorithm of Quan and colleagues.²⁷ The Elixhauser specifically excludes dementia from its codes, allowing us to examine the effects of dementia separately. The Elixhauser score is a count of the number of comorbidities and ranges from 0 to 30, with higher scores representing greater comorbidity. As this range of scores is much larger than the range of other predictors, which can cause problems with numerical computations in model fitting,²⁸ the Elixhauser score was scaled by dividing by 10 for all statistical analyses.

Admission functional status

Because the change in functional status defining HAD may be influenced by the starting level of functional status (e.g., floor effects on measures of ADLs), admission functional status was included in all models. Admission functional status was defined as the Katz ADL score collected at the time of hospital admission.

Admission cognitive function

To examine the impact of cognitive impairment not captured by dementia diagnoses, we included cognitive function assessed using the six-item screener (SIS)²⁹ at the time of admission. The SIS is a brief and psychometrically proven screening tool for cognitive status. The SIS is scored as the number of correct items, yielding a score of 0–6 with higher scores indicating better cognitive function. This measure is a component of the UAB Virtual ACE program and is collected at the time of hospital admission.

Length of stay

As longer hospitalization could lead to longer periods of immobility and greater HAD, we included length of stay as a predictor in all models. Length of stay (LOS) was calculated as the number of days between hospital admission and discharge, as defined in the EHR.

Admission status

We classified admissions as inpatient or observation status based on Medicare definitions, and conducted stratified analyses by admission type, as observation stays may be selected for less severely ill patients.

Service type

We classified the types of providers (medical or surgical) based on the specialty of the treating physician. We classified the type of service as medical only, surgical only, or medical/surgical (for patients who transferred from medical to surgical services, or vice versa, during hospitalization) based on the type(s) of provider(s) caring for the patient during the entire hospitalization. We conducted stratified analyses based on service type to examine potential differences between medical and surgical patients in the development of HAD.

Statistical analysis

We used generalized linear mixed-effects models (GLMMs; covariance pattern models)³⁰ to examine whether risk of HAD was different between hospitalizations for an ACSC versus hospitalizations for other conditions, adjusting for covariates. The mixed-effects model was used as it accounts for the correlation between observations for subjects with repeated hospitalizations and better controls for Type I error in the presence of missing data.³¹ Analyses were conducted using R,³² version 3.6.1 and the lmer package,³³ version 1.1–23.

Study oversight

All study procedures were exempted by the local Institutional Review Board.

RESULTS

Across the 5-year study period, there were a total of 84,648 hospitalizations among 48,758 participants, with 62,154 hospitalizations among 38,960 participants meeting inclusion/exclusion criteria (Figure 1). There was a median of 1 hospitalization per participant (range 1–37, interquartile range 1–2). Participant characteristics for all hospitalizations are shown in Table 1, while participant characteristics at first hospitalization are shown in Table S6. Participants admitted for an ACSC were older, more likely to be non-White and female, had slightly more comorbidities, have greater prevalence of all-cause dementia, and have slightly better admission functional status. Less than 3% of ACSC and non-ACSC hospitalizations were on observation status. The majority of ACSC admissions were to medical services, while non-ACSC admissions were more evenly split between medical and surgical services. Transfers from medical to surgical services, or vice versa, occurred in 7% of ACSC admissions and 11% of non-ACSC admissions. Approximately 10% of all primary diagnoses were for an ACSC using the Purdy definition, which was used in our main analysis. HAD occurred in 20.2% of admissions, with primary diagnoses for ACSCs (using the Purdy definition) accounting for 7.9% of hospitalizations resulting in HAD. Rates of HAD were lower for hospitalizations with a primary diagnosis for an ACSC versus hospitalizations with a primary diagnosis for other conditions (16% versus 20.7%, p < 0.001).

FIGURE 1.

Flow diagram for participant selection. The number of hospitalizations does not equal the number of participants due to multiple hospitalizations for some participants during the study period.

TABLE 1.

Participant characteristics by ambulatory care sensitive condition status

	ACSC hospitalization N = 6221	Other hospitalization N = 55,933	p value	N
Age, years	76.2 (8.77)	75.3 (7.98)	<0.001	62,154
Race:			<0.001	62,154
White	3707 (59.6%)	39,325 (70.3%)
Black or African American	2252 (36.2%)	14,632 (26.2%)
Other	262 (4.2%)	1976 (3.5%)
Gender: Female	3458 (55.6%)	28,631 (51.2%)	<0.001	62,154
Admission status:			0.015	62,154
Inpatient	6091 (97.9%)	54,473 (97.4%)
Observation	130 (2.1%)	1460 (2.6%)
Service type:			<0.001	62,154
Medical	5425 (87.2%)	30,591 (54.7%)
Surgical	364 (5.9%)	19,211 (34.3%)
Medical/Surgical	432 (6.9%)	6129 (11.0%)
Missing	0 (0.0%)	2 (0.0%)
Elixhauser score	5.17 (2.22)	4.32 (2.34)	<0.001	62,154
Dementia: Yes	674 (10.8%)	5388 (9.6%)	0.003	62,154
Hospital-associated disability: Yes	994 (16.0%)	11,551 (20.7%)	<0.001	62,154
Length of stay, Days	5.53 (6.42)	5.69 (8.50)	0.082	62,154
Admission SIS	5.11 (1.69)	5.19 (1.68)	<0.001	62,154
Admission Katz score	9.15 (3.86)	9.15 (4.08)	0.918	62,154

Note: Values shown are for 38,960 unique participants; participants with multiple hospitalizations contributed data at each hospitalization. Service type of medical/surgical indicates participants who were transferred from medical to surgical services, or vice versa, during their hospitalization.

Abbreviations: ACSC, ambulatory care sensitive condition; SIS, six-item screener.

The results of the GLMM analysis are shown in Table 2. After adjusting for other predictors, the risk of HAD was lower among hospitalizations with a primary diagnosis for ACSCs versus hospitalizations with a primary diagnosis for other conditions (odds ratio [OR] = 0.58; 95% CI 0.53, 0.64; Figure 2). Sensitivity analyses using alternative definitions of ACSCs did not change the results of the analysis (data not shown). LOS was the biggest risk factor for HAD, increasing from an odds of 5.74 (95% CI 5.23–6.29) for LOS of 3–4 days to an odds of 32.69 (95% CI 29.32–36.46) for LOS of 8 or more days, relative to an LOS of 0–2 days. Age also greatly increased the risk of HAD, going from an odds of 1.85 (95% CI 1.73–1.97) for age 75–84 to an odds of 3.59 (95% CI 2.93, 4.39) for age 95 and above, relative to age 65–74. The presence of dementia and higher levels of comorbidity were also associated with increased risk of HAD (OR 1.52, 95% CI 1.37–1.68, and OR 1.08, 95% CI 1.06–1.09, respectively). Interestingly, higher admission Katz scores were associated with higher risk of HAD (OR 1.23, 95% CI 1.22–1.24), perhaps because individuals with lower admission Katz scores had already lost ADLs and did not have as much room for decline during hospitalization. However, only 1.0%–5.2% of hospitalizations had an admission Katz score of 0 (Figure S1), and sensitivity analyses excluding the lowest 10% of admission Katz scores did not alter our findings (Table S13). Sex was associated with a significant risk for HAD, with an odds of 1.19 (95% CI 1.13–1.26) for females relative to males.

TABLE 2.

Predictors of HAD

	HAD
Age (65–74)	Reference
Age (75–84)	1.75 [1.66, 1.86]
Age (85–94)	2.15 [1.99, 2.33]
Age (95+)	3.18 [2.66, 3.80]
Race, White	Reference
Race, Black or African American	0.90 [0.85, 0.96]
Race, Other	0.91 [0.79, 1.04]
Gender, Female	1.16 [1.10, 1.22]
Elixhauser score	1.88 [1.68, 2.10]
Dementia	1.45 [1.33, 1.59]
Admission SIS	0.97 [0.95, 0.99]
Admission Katz score	1.20 [1.19, 1.21]
Length of stay, 1–2 days	Reference
Length of stay, 3–4 days	5.22 [4.78, 5.69]
Length of stay, 5–7 days	13.46 [12.28, 14.76]
Length of stay, 8+ days	24.95 [22.61, 27.54]
ACSC (Purdy)	0.61 [0.56, 0.66]
Observations	62,154
Subjects	38,960
Conditional R2	0.42
Marginal R2	0.31
AIC	52196.50

Note: HAD was defined as a decline on the Katz Index between admission and discharge; individuals with a single Katz Index were classified as not having HAD. Generalized linear mixed models were used to examine the association of ACSCs with HAD for each hospitalization, adjusting for covariates and for correlations among multiple hospitalizations for each participant. Model diagnostics in the bottom of the table show the number of hospitalizations (observations) and participants, as well as the proportion of variance explained by both the fixed and random effects (conditional R²) and by only the fixed effects (marginal R²). AIC provides a relative measure of model fit that balances goodness of fit to the data versus model simplicity. Abbreviations: ACSC, ambulatory care sensitive condition (using the Purdy definition²⁰); AIC, Akaike's information criterion; CI, confidence interval; HAD, hospital-associated disability; OR, odds ratio.

FIGURE 2.

Adjusted odds of HAD for ACSC versus non-ACSC admissions by service type. ACSC, ambulatory care sensitive condition; HAD, hospital-associated disability. Odds ratios show the odds of developing HAD, stratified by service type, and adjusted for age, race, gender, Elixhauser comorbidity score, all-cause dementia, admission six-item screener score, admission Katz Index score, and length of stay.

2+ Katz scores

Analyses requiring 2 or more Katz scores for determining HAD showed similar results to analyses requiring only 1 or more Katz scores (Table 3), reinforcing the robustness of our primary results. The risk of HAD among admissions for ACSCs was lower (OR 0.48, 95% CI 0.48–0.53). The association of LOS with HAD was substantially lower (for LOS of 8 or more days, OR 4.75, 95% CI 4.18–5.41), while the associations for age (for age 95 and above, OR 4.64, 95% CI 3.58–6.03) and dementia (OR 1.69, 95% CI 1.49–1.92) were higher.

TABLE 3.

Predictors of hospital-associated disability (2+ Katz scores)

	OR [95% CI] for HAD
Age (65–74)	Reference
Age (85–94)	2.16 [1.96, 2.38]
Age (95+)	3.73 [2.99, 4.65]
Race, Black or African American	0.91 [0.85, 0.98]
Race, White	Reference
Race, Other	0.90 [0.76, 1.06]
Gender, Female	1.18 [1.11, 1.26]
Elixhauser score	2.30 [2.01, 2.63]
Dementia	1.57 [1.41, 1.76]
Admission SIS	0.96 [0.94, 0.98]
Admission Katz score	1.26 [1.25, 1.28]
Length of stay, 3–4 days	1.61 [1.44, 1.80]
Length of stay, 1–2 days	Reference
Length of stay, 5–7 days	2.54 [2.28, 2.84]
Length of stay, 8+ days	3.73 [3.32, 4.18]
ACSC (Purdy)	0.52 [0.47, 0.57]
Hospitalizations	32,226
Participants	22,897
Conditional R2	0.38
Marginal R2	0.18
AIC	38944.89

Note: HAD was defined as a decline in the Katz Index between admission and discharge; individuals with a single Katz Index were excluded from the analysis. Generalized linear mixed models were used to examine the association of ACSCs with HAD for each hospitalization, adjusting for covariates and for correlations among multiple hospitalizations for each participant. Model diagnostics in the bottom of the table show the number of hospitalizations (observations) and participants, as well as the proportion of variance explained by both the fixed and random effects (conditional R²) and by only the fixed effects (marginal R²). AIC provides a relative measure of model fit that balances goodness of fit to the data versus model simplicity. Abbreviations: ACSC, ambulatory care sensitive condition (using the Purdy definition²⁰); AIC, Akaike's information criterion; CI, confidence interval; HAD, hospital-associated disability; OR, odds ratio.

Admission status and service

Analyses of only inpatient admissions showed minimal changes from analyses of all participants (Table S7), which would be expected as greater than 97% of hospitalizations were inpatient admissions. Analyses of only observation status participants showed that most of the associations were not statistically significant (Table S8); the only association that remained significant was for an LOS of 3–4 days, which did show an increased odds of HAD.

Analyses of admissions to medical services only were generally consistent with analyses of all participants (Table S9), although the associations were reduced, and race was no longer statistically significant. Admissions to medical services for ACSCs did show a reduced odds of developing HAD relative to admissions to medical services for other conditions (OR 0.75, 95% CI 0.69–0.83; Figure 2).

Analyses of admissions to surgical services only were generally consistent with the analyses of all participants (Table S10), but the association of ACSCs with HAD was no longer statistically significant (OR 0.82, 95% CI 0.60–1.11; Figure 2), and the association with the Elixhauser comorbidity score was slightly higher (OR 2.22, 95% CI 1.80–2.75). Thus, there was no detectable difference in the odds of developing HAD with admissions for ACSCs versus admissions for other conditions in this population.

Among admissions utilizing both medical and surgical services (i.e., those who transferred from medical to surgical services, or vice versa), analyses showed several similarities to analyses of all participants (Table S11), though race, dementia diagnosis, and admission SIS scores were no longer significant. There was still a significantly reduced odds of developing HAD for those admitted for an ACSC relative to admissions for other conditions (OR 0.69, 95% CI 0.52–0.90; Figure 2).

Excluding initial year of virtual ACE implementation

Analyses excluding the first year in which data were collected as part of Virtual ACE were similar to those using all participants (Table S12). Although there was an attenuation of most of the predictors in the sensitivity analysis, all predictors that were significant in the latter remained significant in the former.

DISCUSSION

This study represents the first analysis of the contribution of inpatient hospitalizations for ACSCs to HAD. Consistent with previous studies,¹⁰ the rate of HAD across all hospitalizations was 20.2%, with 7.9% of hospitalizations resulting in HAD having a primary diagnosis for ACSCs. We also found that 10% of hospitalizations were for ACSCs, with 16% of hospitalizations for ACSCs developing HAD. These findings indicate that hospitalizations for ACSCs are not benign events but are associated with potentially significant consequences for the individual patient and for the health care of older adults.

Stratified analyses by service type (medical only, surgical only, or medical/surgical) showed that the lower OR of HAD for ACSC admissions was largely driven by admissions to medical or medical/surgical services. Though there was no statistically significant difference in the odds of HAD for ACSC versus non-ACSC admissions to surgical services, the OR did favor a lower rate of HAD among those with ACSCs. The small number of ACSC admissions to surgical services may have kept this result from being significant.

This study adds to the growing literature on the pervasive impact of HAD⁶ by examining potentially modifiable pre-hospitalization factors contributing to the development of HAD. Several inhospital risk factors for HAD have been identified,²⁴ particularly low mobility. A recent meta-analysis found that hospitalized patients spend 87%–100% of their time sitting or lying in bed³⁴; this low level of mobility occurs even among individuals who were functionally independent before admission.³⁵ Providers may not encourage ambulation due to perceived lack of patient motivation and fear of patient falls.³⁶ Thus, patients admitted for ACSCs may have low mobility and development of HAD, even if their condition is not severe enough to restrict ambulation. This is particularly concerning as approximately 28% of individuals developing HAD do not recover to pre-admission function a year after hospitalization.⁵

The effects of low mobility may be reflected in our models by the sizeable contribution of LOS, as prolonged hospitalization may result in longer periods of immobility, leading to loss of muscle strength and mass that would predispose a patient to functional decline in ADLs and development of HAD. Longer LOS would also lead to more opportunities for other iatrogenic events, such as adverse drug effects and hospital-acquired infections, that may affect the development of HAD. However, we also recognize that the development of HAD, or development of hospital-acquired conditions contributing to HAD, may also lead to longer LOS to address this complication and changes in discharge needs. While this does not invalidate the results of our analysis, our models capture only a part of the complex relationships in HAD. In addition to LOS, other included predictors may also have a role as an outcome representing the consequences of HAD.

This study also adds a new dimension to the study of ACSCs. To date, studies of ACSCs have largely focused on the economic impact of increased healthcare utilization from potentially avoidable hospitalizations. However, hospitalization is not a benign event, particularly among older adults.³⁷ Although programs such as ACE units have demonstrated reductions in HAD, such programs are not widespread enough to meet the needs of all hospitalized older adults, and additional strategies for preventing HAD are needed. Shifting care for ACSCs from the inpatient to the outpatient setting would presumably avoid the development of HAD. The benefits of this shift may be greater among groups at higher risk of hospitalization for ACSCs, such as those with dementia,³⁸ who are also at higher risk for development of HAD.²⁴ In addition, our results do show that reductions in hospitalizations for conditions other than ACSCs would be beneficial in reducing rates of HAD. ACSCs have been proposed as a potentially intervenable opportunity due to the association of improved primary care delivery with reduced hospitalizations for ACSCs, but the focus on ACSCs should not detract from efforts related to other causes of hospitalizations.

However, there are some limitations that must be acknowledged. Foremost, complications occurring during the hospitalization, rather than the ACSC itself, may have led to the development of HAD. However, it seems likely that avoiding hospitalization for ACSCs would also avoid complications occurring during hospitalization, in turn leading to an indirect reduction in HAD. More detailed studies of the course of HAD during hospitalization are clearly warranted to address this question, and to examine potential interventions to prevent HAD among individuals whose hospitalizations could not be avoided. Second, while nurses received coaching on assessment using the Katz Index, our Virtual ACE program does not include a formal evaluation of nurses' performance over time, which may result in drift in scores. Third, our study examined only one academic medical center in the southeastern United States. Although we expect our findings to be generalizable to other groups and settings, further studies are needed to confirm this. Fourth, we restricted our analysis to individuals who did not have an ICU admission during their hospital stay, because procedures common to the ICU (such as sedation and paralysis) would likely increase the risk of HAD. Approximately 4.9% of hospitalizations with an ICU admission in our data had a primary diagnosis of an ACSC. Separate studies specific to the ICU are needed to understand the unique risks for development of HAD in this population. Fifth, although the ICD-10 codes for dementia in our study are based on published lists, the sensitivity and specificity of these codes have not been formally evaluated.³⁹ Similarly, the performance of the SIS has not been formally evaluated in surgical settings. Sixth, examining hospitalizations for ACSCs focuses on only a single condition as the reason for hospitalizations. Most older adults have multiple comorbid conditions⁴⁰ and comorbidity increases the risk of hospitalizations for ACSCs.⁴¹ While such results do not invalidate our findings, interventions to reduce rates of hospitalizations for ACSCs in older adults will likely need to address the complexity of care not present in younger populations. Seventh, our definition of HAD is based on loss of ADLs, which is a commonly used approach. While loss of ADLs is a clinically meaningful outcome, the Katz Index does not capture changes in physical and cognitive performance that may be of concern to both patients and providers, such as difficulty walking. Eighth, our analysis is in the context of our Virtual ACE quality improvement program, which includes regular assessments of ADLs and mobility protocols to promote maintenance of independence. These activities could potentially alter the rates of HAD and the relationship of HAD with risk factors, which we are unable to disentangle with our study design.

In summary, our analysis demonstrated that rates of HAD among older adults hospitalized for ACSCs were lower than rates of HAD during hospitalization for other conditions but were still substantial. Improvements in healthcare delivery to reduce avoidable hospitalizations due to ACSCs offer a novel approach to potentially prevent HAD, and complement inhospital mobility programs to reduce the incidence of HAD. Given the lack of progress in reducing the incidence of HAD over the last 30 years, further research into this area is clearly warranted to reduce the burden of HAD in the older adult population.

Supplementary Material

Supplement

Data S1. Supporting Information.

Table S1. Comparison of ACSC definitions used in analyses

Table S2. Ambulatory care senstitive conditions in the Purdy classification

Table S3. Ambulatory care sensitive conditions in the Freund classification

Table S4. Ambulatory care sensitive conditions in the Sundmacher classification

Table S5. Ambulatory care sensitive conditions in the Naumann classification

Table S6. Participant characteristics by ambulatory care sensitive condition status at first hospitalization

Table S7. Predictors of hospital-associated disability (inpatients only)

Table S8. Predictors of hospital-associated disability (observation patients only)

Table S9. Predictors of hospital-associated disability (medical admissions only)

Table S10. Predictors of hospital-associated disability (surgical admissions only)

Table S11.. Predictors of hospital-associated disability (medical/surgical admissions)

Table S12. Predictors of hospital-associated disability, excluding inital year of virtual ACE implementation

Table S13. Predictors of hospital-associated disability, excluding individuals with the lowest 10% of baseline Katz ADL scores

Figure S1. Distribution of admissions Katz scores for all hospitalizations (N=62, 152) by service type

Key points.

• Among older adults hospitalized for ambulatory care sensitive conditions (ACSCs; avoidable hospitalizations), 16% developed hospital-associated disability (HAD).

• HAD was less likely to occur in hospitalizations for ACSCs compared with hospitalization for non-ACSC conditions, but rates of HAD were still substantial.

Why does this paper matter?

Rates of HAD have remained unchanged over the past 30 years despite the development of programs to reduce the incidence of HAD. Improved recognition and treatment of ACSCs on an outpatient basis offers another potential way to prevent the occurrence of HAD.