Hospital Drug Burden and Post-Discharge Health Care Utilization in Older Adults Prescribed Central Nervous System-Active Medications

Abstract

Background:

The Drug Burden Index (DBI) quantifies cumulative exposure to anticholinergic and sedative medications, and can be used to measure medication burden during hospitalization. This study evaluates using automated DBI calculations from electronic health records to track DBI changes during hospitalization and identify high-risk older adults, and assess associations with post-discharge health utilization outcomes.

Methods:

This retrospective cohort study used electronic health records of patients aged 65+ from three academic health system hospitals (2018–2020), and prescribed central nervous system (CNS)-acting medications around hospitalization (90 days before and after). Logistic regression assessed the associations between DBI at admission, discharge, and changes in DBI during hospitalization with 30-day readmission, and discharge destination. Linear regression assessed the relationship between DBI at these time points and 90-day post-discharge outpatient visits.

Results:

A total of 3,450 eligible patients had complete DBI measurements, with 80% having no change during hospitalization. DBI at admission, discharge, or changes during hospitalization were not associated with 30-day readmission. Any change in DBI, compared to no DBI change, was associated with reduced odds of being discharged home (DBI increase (worsening drug burden) OR 0.29 (95% CI 0.21, 0.40)); DBI decrease OR 0.35, (0.28, 0.43)). High DBIs at admission or discharge were linked to more clinical visits post-discharge, while a DBI increase during hospitalization was associated with fewer visits (RR 0.86, (0.73–1.00)).

Conclusion:

DBI changes during hospitalization impact discharge disposition and follow-up care needs but not 30-day readmission, highlighting opportunities to optimize medication management during hospitalization.

INTRODUCTION

Older adults commonly present to the hospital with central nervous system (CNS) medications, which are linked to poor health outcomes such as increased falls, delirium, and mortality.^1–4 CNS medications, particularly antipsychotics, benzodiazepines, and opioids, significantly heighten risks due to older adults' increased sensitivity to these medications and their side effects.⁵ Studies consistently show that CNS medications in this population contribute to adverse drug events, longer hospital stays, and higher readmission rates.^6–8

Despite these risks, discontinuation of CNS medications during hospitalization remains infrequent, ⁹ and clinicians lack scalable, objective tools to identify older adults most vulnerable to medication-related harm during care transitions. Hospitalization is a dynamic period of clinical instability in which medication regimens are frequently modified, and these changes may signal heightened vulnerability and influence discharge destination and post-discharge follow-up. Emerging evidence suggests that targeted deprescribing during hospitalization can reduce total medication burden and improve cognition, function, and mobility,^10–20 but identifying patients most likely to benefit remains challenging, as current approaches rely on medication counts or clinician judgment that may not capture cumulative pharmacologic burden. The Drug Burden Index (DBI) offers a scalable, objective measure to quantify CNS medication burden and identify patients at elevated risk.

The DBI quantifies a patient’s cumulative exposure to anticholinergic and sedative CNS-acting medications, providing a measure of drug burden that has been linked to functional decline, cognitive impairment, and increased hospitalization risk.⁶ Unlike simple medication counts, the DBI captures dose changes and can be automated using electronic health record (EHR) medication order data,^21–24 enabling scalable real-time monitoring of changes during hospitalization.

This study examines whether automated DBI monitoring could identify patients at risk for adverse post-discharge outcomes, focusing on 30-day readmission or emergency department (ED) return as the primary outcome, and discharge destination and 90-day outpatient utilization as secondary outcomes.We also compared dynamic DBI changes with static admission and discharge values to assess whether changing versus stable medication burden better predicted post-hospital outcomes.²⁵

METHODS

Study Design and Setting:

This is a retrospective study using EHR data of patients aged 65 or older, hospitalized on general medicine services at three hospitals within the Duke University Health System (1 tertiary academic- and 2 community-based hospitals) between 2018–2020, who were alive 90 days after discharge, and who had at least two visits to the health system in the year prior to hospitalization. Inclusion required at least one outpatient CNS medication prescription within 90 days before or after hospitalization and at least one outpatient visit within 90 days on each side; survival to 90 days was required to ensure complete post-discharge utilization ascertainment. The first hospitalization during the study period was defined as the index hospitalization.

This study used the same EHR-based cohort as a prior analysis of CNS medication use.⁹ The original study characterized patterns of CNS medication prescribing across the peri-hospital period. The current study is a secondary analysis focused on operationalizing the DBI, restricted to patients with complete medication data across all four time points. The Duke University institutional review board approved the study, which was reported in accordance with Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) guidelines (Supplemental Table 1).

CNS Medication Exposure at Four Landmark Time Points Around Hospitalization:

The CNS-acting medication classes examined were antianxiety, anticholinergics, anticonvulsants, antidepressants, antipsychotics, hypnotics, muscle relaxants, and opioids. Medication exposure was determined using outpatient prescription records to better reflect medications taken at admission and avoid misclassification from transient inpatient prescribing. Automated algorithms extracted medication orders from the EHR to identify active outpatient prescriptions at four time points: 90 days prior to admission, on admission, at discharge, and 90 days post-discharge. These time points captured baseline use and early post-hospital changes while minimizing bias from short-term inpatient medication use. An outpatient medication was considered active if the start date occurred before and the end date after the time point. This approach has been previously described and validated in the same parent cohort,¹ with automated abstraction showing high accuracy against manual review (sensitivity 91–97%, specificity 92–100%). Over-the-counter CNS medications (e.g., diphenhydramine) were included if reported by patients and documented in the EHR with an order start date. Medication orders were categorized by therapeutic class to analyze use patterns at each time point

Drug Burden Index:

The Drug Burden Index (DBI) quantifies cumulative exposure to anticholinergic and sedative medications and captures dose changes and medication discontinuations.⁶ DBI was calculated for each eligible medication using the formula $\text{Daily}\text{Dose}/(\text{Daily}\text{Dose}+\text{Minimum}\text{Effective}\text{Daily}\text{Dose}$), with minimum effective doses obtained from Lexicomp; individual drug scores (range 0–1) were summed to generate total DBI. DBI was calculated at four time points (90 days pre-admission, admission, discharge, and 90 days post-discharge) and included both scheduled and as-needed (PRN) medications, using the prescribed dose to estimate PRN exposure. We defined DBI change as the difference between discharge DBI and admission DBI (discharge minus admission), the primary exposure of interest. A clinically meaningful change in DBI from admission to discharge was set at ≥ 0.5. This threshold was selected as it approximates the DBI value of a single medication (ranging from 0 to 1) and reflects either the increase, dose tapering, or discontinuation of at least 1 DBI-related medication, consistent with prior research.^26,27 DBI at admission and discharge were examined in secondary analyses because they represent key clinical decision points: admission DBI reflects baseline medication-related risk, while discharge DBI reflects residual burden entering the post-acute period, when vulnerability to adverse outcomes and care fragmentation is high.

Outcomes:

The primary outcome was 30-day hospital readmission or emergency department (ED) return, analyzed as a composite and separately, defined as any hospital admission or ED visit within the Duke University Health System within 30 days of index discharge. Secondary outcomes included discharge to home (home with self-care, home health, hospice, or self-directed discharge) versus facility (skilled nursing, inpatient rehabilitation, long-term acute care, intermediate care, other acute hospital, inpatient hospice, or other facility), and the number of outpatient clinic visits within 90 days among patients discharged home. Consistent with the inclusion criteria, all patients had at least one outpatient visit within 90 days post-discharge, therefore none had zero visits in this analysis. These outpatient visits were restricted to those occurring within the Duke University Health System and included in-person and virtual (telehealth) visits captured in the EHR.

Patient Characteristics:

We examined demographics (age, sex, white or black race) and comorbidities (cognitive impairment, chronic kidney disease) that could influence prescribing patterns. These characteristics were obtained through automated EHR abstraction. Cognitive impairment, including dementia, and chronic kidney disease, including end-stage renal disease, were identified using ICD-10 codes (Supplemental Table 1). These comorbidities were specifically collected for separate studies on CNS medication use in patients with cognitive impairment and chronic kidney disease. Other comorbidities were not extracted from the EHR and, therefore, not included in the analytic dataset.

Analysis:

Univariate and bivariate descriptive statistics were used to characterize the study population, presented as means (standard deviation) for continuous variables and counts (percentages) for categorical variables. The primary outcome was 30-day hospital readmission or return ED visit. Change in DBI from admission to discharge was the primary exposure, categorized as a clinically meaningful decrease (≥0.5), no change, or increase (≥0.5), reflecting hospitalization as a dynamic period of medication modification. DBI at admission and discharge were examined as continuous secondary exposures to contextualize baseline and residual medication burden and compare associations of static versus dynamic medication burden with post-discharge outcomes. All models were adjusted for age, sex, race, cognitive impairment, and chronic kidney disease. Multivariate models were adjusted for covariates, including age, sex, race, dementia/cognitive impairment, and chronic kidney disease. However, formal interaction terms (e.g. hospital site, cognitive impairment) were not evaluated due to the post hoc nature of the analysis and limited power for effect modification testing. Two regression approaches were used to assess the association of DBI with different outcomes. First, logistic regression models were used to determine the relationships between DBI at admission, discharge, and changes of ≥0.5 during hospitalization with outcomes of 30-day readmission/return ED visits and discharge destination. Second, Poisson models were constructed to examine the association between DBI and the number of post-discharge outpatient visits within 90 days. Results are presented as Odds Ratios or incident Rate Ratios with 95% confidence intervals (CI). With our sample size, we anticipated sufficient power to detect moderate to large effects between the DBI categories with an alpha level of p < 0.05, and an expected readmission rate of 20%. All statistics, including data management and DBI calculations, were performed using SAS v9.4 software (SAS Institute Inc. 2020. SAS/STAT^® 15.2 User’s Guide. Cary, NC: SAS Institute Inc.).

RESULTS

A total of 4,666 patients met the inclusion criteria,⁹ with 3,450 (74%) having complete DBI data at all four time points. Patients with missing doses, frequencies, or minimum effective doses (26%) were excluded. The analytic sample (n=3,450) had a mean age of 74.9 ± 9.4 years and a mean hospital length of stay of 4.7 ± 5.9 days (median 3.0 interquartile range [IQR] 2.0–6.0) (Table 1). The mean DBI ranged from 0.99 to 1.19 across the four time points, with the highest DBI 90 days post-discharge and the lowest at discharge (Table 1). During hospitalization, 1 in 5 patients experienced a clinically meaningful change in DBI of ≥0.05: 15% (n=523) had a decrease (mean −0.99, SD 1.3), and 5% (n=173) had an increase (mean 0.87, SD 0.5). Most patients (80%, n=2,754) had no clinically meaningful change in DBI during hospitalization (Table 1).

Table 1.

Patient Characteristics

N=3450	Mean (SD) or N (%)
Demographics
Age	74.9 (9.4)
Female	2122 (62%)
White race	2401 (70%)
Comorbidities
Mild cognitive impairment/dementia	453 (13%)
Chronic kidney disease/End stage renal disease	840 (24%)
Drug Burden Index (DBI)
DBI mean around hospitalization
90 days before	1.06 (1.04)
Admission	1.10 (1.03)
Discharge	0.99 (0.96)
90 days after	1.19 (1.03)
Absolute change in DBI (discharge – admission)
Decrease	−0.99 (1.3)
No change	0.00 (0.7)
Increase	0.87 (0.5)
Meaningful change in DBI
Decrease by ≥0.5	523 (15%)
No Change	2754 (80%)
Increase by ≥0.5	173 (5%)
Health Care Utilization
Hospital
Duke University Hospital (academic, tertiary)	1759 (51%)
Duke Regional Hospital (community)	759 (22%)
Duke Raleigh Hospital (community)	932 (27%)
Index hospital length of stay, days
mean (SD)	4.7 (5.9)
median (IQR)	3.0 (2.0, 6.0)
Readmissions, hospital/ED return composite, overall	531 (15%)
Decrease DBI by 0.5 (n= 523)	87 (17%)
No change (n=2754)	420 (15%)
Increase DBI by 0.5 (n=173)	24 (14%)
Discharged home, overall	2660 (77%)
Decrease DBI (n=523)	327 (63%)
No change (n=2754)	2236 (81%)
Increase DBI (n=173)	97 (56%)
Pre-hospital outpatient visits in the past year, number	7.5 (6.2)
Post-hospital outpatient visits, within 90days of discharge, number	1.9 (2.5)

Association of hospital DBI with 30-day hospital readmissions or return to ED visits

We observed that 15% (n=531/3450) of the patients were readmitted or returned to the ED within 30 days. We examined both static DBI at admission and discharge and clinically meaningful dynamic changes during hospitalization in relation to 30-day readmissions and return ED visits. We found no significant association between DBI scores at admission or discharge, nor the raw change in DBI from admission to discharge, with 30-day hospital readmissions, 30-day return to ED visits, or composite hospital readmissions/ED returns (Figure 1 and Supplemental Table 2). Specifically, the composite outcome was not affected by the DBI score at admission (OR 1.04, (95% CI 0.95, 1.14)) or discharge (OR 1.01, (0.92, 1.12)). Additionally, neither a clinically meaningful decrease nor increase in DBI by 0.5 from admission to discharge showed a significant impact on composite hospital readmissions or return to ED visit compared to no DBI change (Decrease DBI: OR 1.10, (0.85, 1.42); Increase DBI: OR 0.91, (0.58, 1.43)).

Figure 1. Forest Plots of Health Care Utilization Outcomes according to Drug Burden during Hospitalization.

Forest plots visualizing the odds ratios (OR) and 95% confidence intervals (CI) for A) 30-day readmission/return ED visits, and B) Discharge Destination, and C) the incident Risk Ratio (RR) for post-discharge clinic visits. The reference category is “No change” in DBI score.

Association of hospital DBI with discharge destination to home.

Overall, 77% (n=2660/3450) of patients were discharged to home. For descriptive context, discharge to home was most common among patients with no change in DBI during hospitalization (81%; 2,236/2,754), compared with those with a DBI decrease (63%; 327/523) or increase (56%; 97/173) (Table 1). In adjusted analysis, admission DBI was not associated with discharge to home (OR 1.05, (0.96, 1.15)), whereas higher DBI at discharge was associated with greater odds of discharge to home (OR 1.25, (1.13, 1.39)). In the analysis examining the primary exposure change in DBI, both a 0.5-unit decrease (OR 0.35, (0.28, 0.43)) and increase (OR 0.29, (0.21, 0.40)) from admission to discharge were linked to lower odds of discharge to home (Figure 1 and Supplemental Table 3). Cognitive impairment (OR 0.54 (0.43, 0.67),) and age (OR 0.94, [0.93,0.95]) also reduced the likelihood of home discharge.

Association of hospital DBI with post-discharge outpatient clinic visits among those discharged home.

We similarly evaluated DBI at admission and discharge in relation to post-discharge outpatient utilization, followed by analyses examining DBI change during hospitalization. In adjusted analyses among patients discharged home, higher DBI at admission (RR 1.06, 95% CI 1.03–1.09) and discharge (RR 1.06, 95% CI 1.03–1.09) was associated with more outpatient visits within 90 days (Supplemental Table 4), corresponding to a 6% increase in visit rates per unit higher DBI. A clinically meaningful DBI decrease was not associated with post-discharge visit frequency (RR 1.00, 95% CI 0.93–1.09), whereas DBI increases during hospitalization were associated with fewer visits (RR 0.86, 95% CI 0.73–1.00), representing a 14% reduction in follow-up visits.

DISCUSSION

In this study of hospitalized older adults prescribed CNS-acting medications, DBI at admission, DBI at discharge, and clinically meaningful changes in DBI during hospitalization were not associated with 30-day hospital readmission or return ED visits, our prespecified primary outcome. Although approximately one in five patients experienced a DBI change of ≥0.5 during hospitalization, this null finding suggests that DBI change alone does not capture the drivers of short-term acute care utilization in this population. Rather than indicating a lack of clinical relevance, these findings clarify the limits of DBI as a short-term risk stratification tool. Three factors likely contributed to this result, informed by our findings and prior literature.

First, DBI change derived from automated medication orders may not capture clinical intent or appropriateness. Medication changes during hospitalization often reflect appropriate management of acute symptoms (e.g., pain, agitation, insomnia, delirium) rather than inappropriate prescribing. Because DBI assigns equivalent weight to anticholinergic and sedative medications and does not account for indication, symptom burden, or short-term therapeutic intent, both increases and decreases in DBI may reflect clinical complexity rather than directional changes in risk, particularly when calculated from outpatient orders rather than medication administration data.^6,28

Second, DBI-related risk may manifest over longer time horizons or through outcomes other than 30-day readmission. Prior studies most consistently link DBI exposure to functional decline, falls, mortality, and institutionalization rather than short-term reutilization.^6,27,29,30 A recent systematic review demonstrated substantial heterogeneity in DBI–outcome associations and inconsistent evidence linking DBI to readmission risk.^31–34 Given the multifactorial nature of readmissions, medication burden alone may be insufficient to detect risk in the short term, particularly in a cohort with uniformly high baseline CNS medication exposure

Third, unmeasured clinical and system-level factors, including functional status, illness severity, symptom burden, and care received outside the health system, may have attenuated observable associations with readmission. As a result, DBI change during hospitalization may function more as a marker of patient complexity or instability than as a direct determinant of short-term utilization, limiting its predictive utility in isolation.

Differences from prior studies further underscore the importance of population and methodological context. While some studies have reported associations between DBI and readmission risk, such as higher DBI among readmitted patients³³ or increased readmissions following DBI increases in older adults hospitalized for falls,³⁴ our study focused on older adults already prescribed CNS-acting medications, a population with high baseline DBI exposure and limited variability in medication burden. Additionally, prior studies examined selected high-risk subgroups or used manual abstraction or dispensing data,³⁵ whereas our study evaluated scalable, automated EHR-derived DBI calculations, which may not consistently capture actionable risk without added clinical context. These differences in population, exposure measurement, and outcome selection likely explain the divergent findings.

Although DBI was not associated with short-term readmissions, hypothesis-generating secondary analyses showed that any clinically meaningful DBI change during hospitalization, whether increase or decrease, was associated with lower odds of discharge to home, while stable DBI was associated with home discharge. This finding aligns with prior work linking medication burden and anticholinergic or sedative exposure to post-discharge institutionalization.²⁹ One interpretation is that medication changes are more common in patients with complex or unstable hospital courses who are more likely to require post-acute care. Our findings extend prior work by suggesting that DBI change during hospitalization, rather than absolute DBI alone, may serve as a marker of clinical instability or complexity. Similar patterns have been reported among patients hospitalized for falls or delirium, in whom DBI increased or decreased, respectively, and discharge to skilled nursing facilities was common.^28,34 An alternative explanation is that DBI changes, including deprescribing, may transiently disrupt prior medication equilibrium and increase short-term care needs. While we cannot distinguish among mechanisms, these findings should not be interpreted as evidence that DBI change causally determines discharge destination. Rather, they underscore the importance of careful monitoring and coordinated discharge planning when CNS-active medications are modified during hospitalization. A subset of patients discharged home received home health services, representing an opportunity for ongoing medication monitoring.

Additional secondary analysis showed that DBI level and change were also associated with post-discharge outpatient follow-up. Among patients discharged home, higher DBI at admission and discharge was associated with more outpatient visits within 90 days, consistent with prior studies linking DBI exposure to increased healthcare utilization, including greater general practitioner visits.³⁰ In contrast, DBI increases during hospitalization were associated with fewer post-discharge visits, a paradoxical finding that may reflect unmeasured transitions to hospice, care outside the health system, or gaps in follow-up among patients with newly increased CNS medication burden. Importantly, this pattern raises concern that patients at potentially higher medication-related risk may be under-monitored during a vulnerable transition period, underscoring the importance of effective care coordination during hospital-to-community transitions.^36,37

Taken together, our findings suggest that automated DBI monitoring may be better suited for identifying patients at risk for suboptimal care transitions and follow-up gaps than for predicting short-term readmissions. Tracking DBI changes over time, rather than a single value, may help clinicians identify patients with medication instability who could benefit from targeted pharmacy review, enhanced discharge planning, or structured follow-up. However, our results also highlight important limitations of DBI when used as a standalone automated signal. The non-linear scaling of the DBI formula, reliance on medication orders rather than administration, and lack of contextual information about medication intent constrain its interpretability in real-world EHR implementations.^{6,25, 27}

Limitations

This study has several limitations. First, because the data were derived from a single health system, findings may not be fully generalizable. We partially mitigated this by including both tertiary and community hospitals and by using the widely adopted Epic EHR platform; however, restricting the cohort to patients with documented CNS medication exposure within 90 days before and after hospitalization may limit applicability to broader older adult populations.

Second, this secondary, post hoc analysis used data not originally designed for DBI calculation; consequently, 26% of otherwise eligible patients were excluded due to missing dose, frequency, or minimum effective dose data, potentially introducing selection bias. Medication exposure was inferred from EHR order dates rather than dispensing or claims data, which may have misclassified true exposure, particularly for as-needed medications, and failed to capture prescriptions outside the health system. Although automated abstraction methods were validated, some misclassification of admission and discharge medication lists is possible. Future studies incorporating inpatient dispensing data may improve DBI measurement accuracy.

Third, readmissions occurring outside the health system were not captured, likely underestimating readmission rates. In addition, multiple statistical comparisons raise the possibility of type I error.

Fourth, the analysis did not account for key patient- and hospitalization-level factors (e.g., functional status, illness severity, symptom burden, conditions affecting CNS medication metabolism such as hepatic dysfunction, length of stay, psychiatric or overall comorbidity burden). These unmeasured factors may contribute to reverse causality, whereby clinical instability or discharge planning drives medication changes rather than the reverse, and longer stays or greater clinical complexity may further increase opportunities for both medication modification and non-home discharge. Because standard comorbidity indices (e.g., Charlson or Elixhauser) could not be calculated, residual confounding is likely and may explain why both DBI increases and decreases were associated with lower odds of discharge home.

Fifth, restricting the cohort to patients alive 90 days post-discharge introduces survivor bias; those excluded may have had higher DBI or greater increases during hospitalization, potentially biasing associations toward the null for short-term outcomes. Consequently, findings may underestimate associations between medication burden and adverse outcomes in the highest-risk patients.

Finally, several findings, particularly the association of higher DBI at discharge with home discharge and DBI increases with fewer post-discharge clinic visits, were unexpected and warrant cautious interpretation. These patterns may reflect methodological features of DBI itself. The non-linear DBI formula reduces sensitivity to dose increases at higher exposure levels, potentially dampening associations with adverse outcomes at the elevated DBI values observed in this cohort (mean DBI 1.1).^25,27 In addition, DBI assumes equivalency across anticholinergic and sedative medications and does not account for medication intent, appropriateness, or adherence, particularly when derived from medication orders rather than administration data. Together, these factors suggest that automated, EHR-derived DBI may not consistently provide actionable signals during care transitions without added clinical context, underscoring the need for refinement and complementary approaches that incorporate medication intent, patient complexity, and care context.

Conclusion:

In summary, among hospitalized older adults prescribed CNS-acting medications, DBI changes during hospitalization were not associated with 30-day readmission or ED return, underscoring the limits of DBI for short-term risk prediction. In contrast, DBI levels and changes were associated with discharge destination and post-discharge follow-up, suggesting greater utility for care transition planning. Future work should refine how DBI is operationalized and integrated into EHR-based workflows to support clinically meaningful decision-making rather than isolated risk prediction.

Key Points.

• Automated Drug Burden Index (DBI) calculation was feasible using electronic health record data and revealed that most older adults using central nervous system (CNS) medications experienced no clinically meaningful change in DBI during hospitalization (80%).

• DBI changes during hospitalization did not predict 30-day readmission but were significantly associated with an increased likelihood of discharge to a skilled nursing facility for patients with either a DBI increase or decrease compared to no change.

• Higher DBI scores at admission and discharge were associated with more post-discharge clinic visits, but a DBI increase during hospitalization was paradoxically linked to fewer post-discharge clinic visits, suggesting a potential for insufficient follow-up in high-risk patients.

Why does this paper matter?

This study highlights the utility of automated DBI monitoring in the electronic health record to identify high-risk older adults who need focused medication management before discharge, particularly for ensuring appropriate discharge disposition and closing potential gaps in follow-up care.