Impact of Hospitalizations on Problematic Medication Use Among Community-Dwelling Persons With Dementia

Abstract

Background

Hospitalizations are frequently disruptive for persons with dementia (PWD) in part due to the use of potentially problematic medications for complications such as delirium, pain, and insomnia. We sought to determine the impact of hospitalizations on problematic medication prescribing in the months following hospitalization.

Methods

We included community-dwelling PWD in the Health and Retirement Study aged ≥66 with a hospitalization from 2008 to 2018. We characterized problematic medications as medications that negatively affect cognition (strongly anticholinergics/sedative-hypnotics), medications from the 2019 Beers criteria, and medications from STOPP-V2. To capture durable changes, we compared problematic medications 4 weeks prehospitalization (baseline) to 4 months posthospitalization period. We used a generalized linear mixed model with Poisson distribution adjusting for age, sex, comorbidity count, prehospital chronic medications, and timepoint.

Results

Among 1 475 PWD, 504 had a qualifying hospitalization (median age 84 (IQR = 79–90), 66% female, 17% Black). There was a small increase in problematic medications from the baseline to posthospitalization timepoint that did not reach statistical significance (adjusted mean 1.28 vs 1.40, difference 0.12 (95% CI −0.03, 0.26), p = .12). Results were consistent across medication domains and certain subgroups. In one prespecified subgroup, individuals on <5 prehospital chronic medications showed a greater increase in posthospital problematic medications compared with those on ≥5 medications (p = .04 for interaction, mean increase from baseline to posthospitalization of 0.25 for those with <5 medications (95% CI 0.05, 0.44) vs. 0.06 (95% CI −0.12, 0.25) for those with ≥5 medications).

Conclusions

Hospitalizations had a small, nonstatistically significant effect on longer-term problematic medication use among PWD.

Medication changes are common among older adults in the setting of a hospitalization, with many of the newly added medications considered potentially inappropriate (1–3). For persons with dementia who are more likely to be frail, hospitalizations frequently represent a critical transition period in their health trajectory (4–6). This is often due to high rates of complications, such as delirium and hospital-associated disability, that can have long-lasting effects on an individual’s cognitive and physical functioning (7,8). Hospitalizations also commonly exacerbate neuropsychiatric symptoms (eg, aggression, agitation, and hallucinations) and increase symptom burdens (eg, difficulty sleeping due to disrupted sleep-wake cycle, pain due to acute medical conditions, and functional incontinence due to limited hospital mobility) (9–11).

Therefore, events during a hospitalization may prompt well-meaning clinicians to prescribe medications that are potentially problematic for persons with dementia. Examples include antipsychotics for the treatment of delirium, nonbenzodiazepine sedative hypnotics (eg, zolpidem) for the treatment of insomnia, or opioids for pain (12,13). Additionally, inpatient clinicians frequently intensify an individual’s outpatient antihypertensive and antidiabetic regimens at hospital discharge despite potentially controlled blood pressures and glucose values at home (14,15). This may be driven by a reflexive response to elevated inpatient blood pressure readings and glucose values in the setting of acute illness and pain. Addition of these potentially problematic medications at hospital discharge has been associated with hospital readmissions and adverse drug events (16,17). Many of these medications may be continued during subsequent outpatient follow-up appointments leading to prolonged use of problematic medications.

On the other hand, hospitalizations may serve as an opportunity for clinicians to review a patient’s medication list and deprescribe potentially harmful medications (18–21). This is particularly true if the hospitalization is related to an adverse drug event, such as delirium or falls. Multiple studies involving medication review interventions (eg, pharmacist led reviews or computerized decision support tools) have been conducted with the goal of reducing hospital readmissions and emergency department visits (18,19,22–24). Some of these have shown promising results although they are not widely implemented.

Little is known about how hospitalization affects longer-term problematic medication use among persons with dementia (ie, several months after the acute care episode) with few studies taking a comprehensive approach to characterize a broad spectrum of potentially problematic medications. Assessing problematic medication use several months after a hospitalization is important as it more closely reflects an individual’s updated medication list after they have had an opportunity to follow up with longitudinal prescribers such as primary care clinicians. If hospitalizations are a primary driver of long-term problematic medication use among persons with dementia, this would argue for increasing resources to hospital-based interventions that may reduce problematic prescribing. Therefore, we sought to determine the impact of hospitalization on longer-term problematic medication use across multiple domains using a nationally representative sample of community-dwelling persons with dementia.

Method

Study Cohort

We included participants from the Health and Retirement Study (HRS) between 2008 and 2018. The HRS is a nationally representative survey of U.S. adults with interviews conducted every 2 years (25). HRS participants were classified as having dementia based on a validated algorithm by Hurd et al., which includes predictors such as age, sex, cognitive tests, and physical functioning (26,27). This algorithm was developed using participants from the Aging, Demographics, and Memory Study who underwent extensive clinical evaluation and neuropsychological testing to establish dementia diagnoses. It has shown high accuracy in subsequent validation studies (27). We included community-dwelling persons with dementia aged ≥66 with at least one hospitalization during the follow-up period. For individuals with multiple hospitalizations during the study period, we assessed problematic medication use around the first hospitalization.

Identification of Medications

We obtained information on prescriptions, prescription dates, and days supplied through Medicare Part D claims data. In our study cohort, we included individuals who had at least one Medicare Part D prescription fill at 1 of the 3 timepoints surrounding the hospitalization, as described later. Prescriptions were linked to the Medi-Span database to collect information on drug names and drug classes using the Generic Product Identifier classification system. To assess problematic medication use at various timepoints throughout the study, we used a medication-on-hand approach with a 30-day grace period (28). Briefly, the medication-on-hand approach classifies an individual as being on a medication if the most recent fill provides enough supply to last through a specific timepoint. A grace period is used to account for transient nonadherence or stockpiling of medications. The medication-on-hand approach with a 30-day grace period has shown high accuracy in a previous study for identifying prevalent medication use (28).

Defining Potentially Problematic Medication Use

We defined 3 broad categories of potentially problematic medications: (a) medications that negatively affect cognition, (b) medications to avoid based on 2019 American Geriatrics Society Beers Criteria for Potentially Inappropriate Medication Use in Older Adults, and (c) medications to avoid based on the Screening Tool of Older Persons’ Prescriptions Version 2 (STOPP-V2) (29,30). Of note, there is some overlap between these categories in the types of problematic medications identified. Additional details on how we defined the specific criterion from these 3 categories can be found in a previous publication outlining potentially problematic medication use among persons with dementia (31).

Briefly, medications that negatively affect cognition included strongly anticholinergic medications (Table 7 of 2019 Beers criteria) and sedative-hypnotics based on previous studies and the Sedative Load Model (Supplementary Methods, Supplementary Tables 1 and 2) (29,32,33). Medications included in the 2019 Beers and STOPP-V2 criteria are categorized broadly as medications to avoid in older adults in general (eg, long-acting sulfonylureas, megestrol, benzodiazepines), medications to avoid given comorbidities (eg, thiazolidinediones in heart failure or cholinesterase inhibitors in syncope), and medications that may be harmful when used in combination (eg, opioids and benzodiazepines). We used International Classification of Diseases (ICD)-9 and ICD-10 diagnosis codes from Medicare outpatient, inpatient, and Carrier files to identify comorbidities. We excluded problematic medication criteria which were based on creatinine clearance due to a lack of laboratory values. For criteria involving overlapping medications or chronic use (eg, >3 months continuous use), we used fill dates and days of supply. We omitted certain criteria for which it was challenging to determine inappropriateness (Supplementary Methods).

Timepoints for Assessing Problematic Medication Use

We used a self-controlled longitudinal design in which we compared problematic medication use for each individual at 3 distinct timepoints before and after a hospitalization. This allowed us to use a repeated measures design to account for underlying individual time trends in problematic medication prescribing given that medications are added or removed irrespective of a hospitalization. The 3 timepoints included: (a) a few months before a hospitalization (prebaseline), (b) 4 weeks before a hospitalization (baseline), and (c) 4 months after the hospitalization period (posthospitalization longer-term prescribing; Supplementary Figure 1).

Our primary goal was to assess longer-term problematic medication use in a steady-state period a few months after the hospitalization period. The purpose of the 1st timepoint (prebaseline) was to assess problematic medication use several months prior to the hospitalization to serve as a basis for establishing a trend in problematic medication prescribing for each individual in the months leading up to the hospitalization. The 2nd timepoint (baseline) was 4 weeks prior to the hospital admission date for all individuals to establish baseline prehospital problematic medication use. For the 3rd timepoint, we assessed problematic medication use 4 months after the acute care episode, which included the hospitalization period itself and any hospital readmissions or skilled nursing facility (SNF) admissions (Supplementary Figure 1). This means that the time between the 2nd and 3rd timepoint varied for each individual depending on whether they were readmitted and whether they were admitted to an SNF. Because each individual served as his or her own control between the prehospitalization and posthospitalization time points, we ensured that the duration of time between the 1st and 2nd timepoint (prebaseline to baseline) equaled the duration of time between the 2nd and 3rd timepoint (baseline to posthospital; Supplementary Figure 1). This approach allowed us to more robustly model changes in problematic medication use over similar time periods for each individual patient. In our sample, the median (interquartile range) duration for the hospitalization period (which includes the index hospital admission and any hospital readmissions or SNF stays if applicable) was 6 (3–62) days. Individuals who died or were enrolled in hospice without a period of 4 months to assess their medications at the 3rd timepoint (posthospitalization period) were excluded.

Outcome

The primary outcome was the number of potentially problematic medications before and after a hospitalization overall and across individual categories. For medications that were flagged under multiple categories (eg, lorazepam under medications that negatively affect cognition, 2019 Beers, and STOPP-V2), we counted each medication only once when reporting the overall results.

Statistical Analysis

We used a generalized linear mixed model with Poisson distribution where the outcome was modeled as a count of the number of problematic medications. The model accounted for the intra-individual trend in problematic medications at the 3 timepoints discussed above (prebaseline, baseline, and posthospitalization). HRS survey weights were used to account for the complex survey design. We adjusted for age, sex, comorbidity count (sum score ranging from 0 to 7 based on self-reported diagnoses of hypertension, diabetes, cancer, heart disease, lung disease, stroke, and arthritis), and number of prehospital chronic medications. We defined number of prehospital chronic medications as the number of overall medications (not just problematic medications) with at least 28-day supply that an individual was receiving at baseline (4 weeks prior to hospitalization). This definition allowed us to capture an individual’s degree of polypharmacy based on chronic medications. We used the margins command in Stata to obtain the adjusted marginal predicted mean number of problematic medications at the 3 timepoints.

We prespecified 4 subgroup analyses based on factors likely to influence problematic medication use among persons with dementia (34). We stratified individuals by comorbidity count (<3 vs ≥3 using the definition above), index hospitalization length of stay (<7 days vs ≥7 days), whether the individual experienced a hospital readmission or SNF stay following the index hospitalization (yes vs no), and number of prehospital chronic overall medications as defined above (<5 medications vs ≥5 medications). We additionally performed 5 exploratory subgroup analyses based on sociodemographic and hospitalization factors: region of the country (Northeast, Midwest, South, and West), geographic category (metropolitan, micropolitan, and small town/rural), total household wealth (less than or greater than median), reason for hospital admission (categorized as cardiac, gastrointestinal, infectious, musculoskeletal, neurologic, pulmonary, and other), and type of hospital admission (elective vs emergent/urgent).

Analyses were conducted using SAS 9.4 (SAS Institute Inc.) and STATA 17.0 (Stata Corp). The statistical significance threshold was a 2-sided p-value < .05. The study was reviewed and approved by the University of California, San Francisco Committee on Human Research.

Results

Supplementary Figure 2 shows the study flow chart. A total of 1 475 participants in the HRS between 2008 and –2018 were classified as having dementia of which 795 had at least 1 qualifying hospitalization during the study period. We excluded 291 participants who died or were enrolled in hospice in the 4-month posthospitalization period. These individuals shared similar characteristics to those included in the final cohort, including similar levels of polypharmacy and baseline chronic medications (Supplementary Table 3). The final cohort size was 504 hospitalized persons with dementia (which represents 1.8 million individuals after applying national-level survey weights).

Table 1 shows the baseline characteristics of the cohort. Median age was 84 years (interquartile range = 79–90), 66% were female, 17% identified as non-Hispanic Black, and 8% identified as Hispanic (Table 1). There was a high prevalence of chronic conditions and functional impairment (54% had at least 1 dependency in activities of daily living). Additionally, 71% were on at least 5 chronic medications at baseline.

Table 1.

Baseline Characteristics of Community-Dwelling Older Adults With Dementia Enrolled in the Health and Retirement Study From 2008 to 2018 With a Hospitalization

Characteristic	Persons With Dementia (N = 504)a
Age in years, median (IQR)	84 (79–90)
Female sex	331 (66%)
Race/ethnicity
Non-Hispanic White	353 (75%)
Non-Hispanic Black	117 (17%)
Hispanic	34 (8%)
Lives alone	148 (36%)
Self-reported comorbidities
Cancer	98 (21%)
Diabetes	163 (32%)
Heart disease	226 (48%)
Hypertension	381 (80%)
Lung disease	82 (17%)
Stroke	163 (31.8%)
Arthritis	378 (74.5%)
≥1 IADL dependencyb	364 (74%)
≥1 ADL dependencyb	251 (54%)
Polypharmacy at baseline (prescription for ≥5 chronic medications)c	364 (71%)
Median (IQR) number of baseline chronic medicationsc	7 (5–10)

Notes: ADL = activities of daily living; IADL = instrumental activities of daily living; IQR = interquartile range.

^aThe numbers in each column represent the raw number of individuals with a particular characteristic. The survey-weighted percentages in each column represent the column percent based on the weighted sample size after using national survey weights from the Health and Retirement Study. Therefore, the percentages in the table are not the same as the raw percentage.

^bADLs include needing help with bathing, dressing, eating, toileting, getting in and out of bed, and walking across the room. IADLs include needing help with managing medications, shopping, meal preparation, using the phone, and managing finances.

^cThe number of medications was assessed by looking at the number of unique medications with at least a 28-day supply at baseline (4 weeks before a hospitalization).

Table 2 shows the adjusted mean number of problematic medications before and after a hospitalization overall, by individual medication domains, and across subgroups. Overall, there was a small increase in potentially problematic medications from the baseline (4 weeks before hospitalization) to posthospitalization period that was not statistically significant (adjusted mean problematic medications 1.28 vs 1.40, adjusted difference = 0.12 (95% CI −0.03, 0.26), p = .12; Table 2, Figure 1, Supplementary Table 4). Comparing problematic medication use a few months prior to the hospitalization (prebaseline) to 4 weeks prior to hospitalization (baseline) to assess for underlying time trends in prescribing unrelated to the hospitalization, there was no change (adjusted mean problematic medications 1.25 vs 1.28, difference = 0.03 (95% CI −0.11, 0.17), p = .68). Overall, the survey-weighted percentage of individuals receiving at least 1 potentially problematic medication was 59% at the prebaseline timepoint, 60% at the baseline timepoint, and 61% at the posthospitalization timepoint. A histogram of the distribution of potentially problematic medications at the 3 timepoints is shown in Supplementary Figure 3.

Table 2.

Adjusted Mean Number of Problematic Medications Before and After a Hospitalization Among Community-Dwelling Older Adults With Dementia

Category	N	Adjusted Mean Number Of Problematic Medicationsa			Adjusted Difference for Baseline vs Posthospital timepoint (95% CI)	P-value for Baseline vs Posthospital Comparisonb	Interaction P-value
		Prebaseline (~5 months before hospital)	Baseline (4 weeks before hospital)	Posthospital (4 months after hospital)
Overall	504	1.25	1.28	1.40	0.12 (−0.03, 0.26)	0.12	N/A
Problematic medication domains
Medications that negatively affect cognition	504	0.38	0.38	0.44	0.05 (−0.03, 0.14)	0.23	N/A
2019 Beers criteria	504	0.95	0.96	1.09	0.12 (−0.01, 0.26)	0.06	N/A
STOPP Version 2	504	0.62	0.62	0.71	0.09 (−0.01, 0.20)	0.08	N/A
Subgroups
Number of chronic medications at baselinec
<5 medications	140	0.47	0.48	0.73	0.25 (0.05, 0.44)	0.01	0.04
≥5 medications	364	1.55	1.58	1.64	0.06 (−0.12, 0.25)	0.51
Comorbidity countd
<3	158	0.95	1.06	1.14	0.09 (−0.15, 0.32)	0.47	0.71
≥3	346	1.39	1.38	1.51	0.13 (−0.06, 0.31)	0.17
Hospital length of stay
<7 days	398	1.23	1.26	1.37	0.10 (−0.06, 0.26)	0.22	0.96
≥7 days	106	1.32	1.34	1.51	0.17 (−0.16, 0.50)	0.31
Hospital readmission or skilled nursing facility stay following index hospitalization
No	102	1.08	1.15	1.22	0.07 (−0.24, 0.37)	0.66	0.93
Yes	402	1.29	1.31	1.44	0.13 (−0.04, 0.29)	0.13

Notes: CI = confidence interval; STOPP = Screening Tool of Older Persons’ Prescriptions.

^aAdjusted for age, sex, comorbidity count, and number of baseline chronic medications.

^bThis p-value represents the statistical test for the difference between the adjusted mean number of problematic medications in the posthospitalization period and baseline (4 weeks before hospitalization) using the contrast command in Stata from the adjusted model.

^cThe number of baseline chronic medications was assessed by looking at the number of unique medications with at least a 28-day supply at the baseline (4 weeks before hospitalization) timepoint. This includes all medications not only potentially problematic medications.

^dSum score ranging from 0 to 7 based on self-reported diagnoses of hypertension, diabetes, cancer, heart disease, lung disease, stroke, and arthritis.

Figure 1.

Changes in problematic medication use before and after hospitalization among community-dwelling persons with dementia.^a STOPP Version 2 = Screening Tool of Older Persons’ Prescriptions Version 2.

^aNumbers in the figure represent the marginal mean number of potentially problematic medications from a generalized linear mixed model with Poisson distribution adjusted for age, sex, comorbidity count, and number of prehospital chronic medications. See the Methods section and Supplementary Figure 1 for details on how the timepoints were defined.

Results were similar by individual medication domains, including medications that negatively affect cognition and those in 2019 Beers and STOPP V2. For example, the adjusted mean number of problematic medications based on 2019 Beers criteria for the prebaseline (several months before hospitalization), baseline (4 weeks before hospitalization), and posthospitalization timepoints were 0.95, 0.96, and 1.09, respectively (p = .06 for baseline vs posthospitalization comparison).

Results were also similar across prespecified subgroups (comorbidity count, hospital length of stay, and whether the individual was readmitted to the hospital or had an SNF stay) and exploratory subgroups (region of the country, geographic category, household wealth, reason for hospital admission, and type of hospital admission; Table 2 and Supplementary Table 5). For example, the adjusted mean number of problematic medications from the baseline to posthospitalization timepoints increased slightly from 1.31 to 1.44 for those who had a hospital readmission or SNF stay (N = 402; p = .13) and from 1.15 to 1.22 for those who did not have a hospital readmission or SNF stay (N = 102; p = .66; p = .93 for interaction).

For the subgroup based on number of chronic medications at baseline (4 weeks before hospitalization), individuals with <5 baseline chronic medications (N = 140) showed a greater increase in posthospital problematic medications than those with ≥5 baseline chronic medications (N = 364; p = .04 for interaction). In this subgroup analysis, the adjusted mean difference for baseline vs posthospital problematic medications was 0.25 (95% CI 0.05, 0.44) for those with <5 medications (increase from 0.48 to 0.73 problematic medications, p = .01) vs 0.06 (95% CI −0.12, 0.25) for those with ≥5 medications (increase from 1.58 to 1.64 problematic medications, p = .51; Table 2, Supplementary Figure 4).

Figure 2 shows the sum of potentially problematic medications across all individuals stratified by drug class after applying person-level survey weights to make the numbers nationally representative. Potentially problematic medications increased slightly across primarily psychotropic medication classes (eg, antipsychotics, benzodiazepines, antidepressants) before and after a hospitalization. Figure 3 and Supplementary Figure 5 show the weighted percentage of individuals whose total number of problematic medications stayed the same, decreased, or increased at the prebaseline to baseline (4 weeks before hospitalization) timepoints and baseline (4 weeks before hospitalization) to posthospitalization timepoints (4 months after the hospitalization period). Overall, 17% of individuals had an increase in the number of problematic medications in the prebaseline to baseline timepoints compared with 23% in the baseline to posthospital timepoints. Additionally, 16% of individuals had a reduction in the number of problematic medications in the prebaseline to baseline timepoints compared with 20% in the baseline to posthospital timepoints.

Figure 2.

Nationally representative counts of potentially problematic medications among community-dwelling persons with dementia before and after hospitalization by drug class.^a STOPP Version 2 = Screening Tool of Older Persons’ Prescriptions Version 2.

^aThe numbers in the figure represent the survey-weighted counts of potentially problematic medications summed across all persons with dementia included in the study. We applied person-level survey weights to the raw counts of problematic medications to make the results nationally representative. When providing the sum of medications, we only counted each medication once given that each medication could be classified within multiple domains (eg, 2019 Beers and STOPP-V2). The numbers in the figure are scaled such that each number equals the number of medications in 10 000s. For example, among the 504 persons with dementia in the cohort, there were 51 antipsychotics prescribed at the baseline timepoint which, after applying national survey weights, represents roughly 200 000 antipsychotic prescriptions and appears as 20 in the figure. Medications were grouped roughly by drug class. For example, urinary antispasmodics included medications such as oxybutynin, darifenacin, fesoterodine, solifenacin, and tolterodine. Pain medications include opioids, non-steroidal anti-inflammatory drugs, muscle relaxants, gabapentin, and pregabalin. Gastrointestinal medications include proton pump inhibitors, anti-emetics, and others.

Figure 3.

Percentage of persons with dementia who experienced either no change, an increase, or a reduction in the total number of problematic medications in the time periods before and after hospitalization.^a

^aThe 504 persons with dementia included in the cohort were divided into 4 categories based on a comparison of their number of problematic medications at the 3 timepoints (prebaseline, baseline, and posthospital). These categories included: individuals not on any problematic medications, those who continued on the same number of problematic medications (eg, going from 2 problematic medications at baseline to 2 problematic medications at the posthospital timepoint), those who stopped all or reduced the number of problematic medications (eg, going from 3 problematic medications at baseline to 2 problematic medications at the posthospital timepoint), and individuals who increased the number of problematic medications (eg, going from 2 problematic medications at baseline to 3 problematic medications at the posthospital timepoint). Survey weights were used to make these percentages nationally representative.

Discussion

In this nationally representative study, a hospitalization for community-dwelling persons with dementia led to a small, nonstatistically significant increase in problematic medication use several months after the hospitalization period. Results were consistent across multiple domains of potentially problematic medications, including those that negatively affect cognition (eg, strongly anticholinergics and sedative hypnotics) and medications to avoid based on the 2019 Beers criteria and STOPP Version 2. The increase in longer-term problematic medications following a hospitalization was statistically significant for individuals on <5 prehospital chronic medications.

Previous studies have generally focused on the immediate hospitalization period for assessing problematic medication use (also termed potentially inappropriate medications (PIM)) (35–43). Among older adults in general, Weir et al. identified 2 402 older adults admitted at tertiary care hospitals in Canada (35). At discharge, 66% were prescribed at least one PIM at discharge with 49% continuing a PIM from prior to admission and 31% prescribed at least one new PIM. Increasing PIM use was associated with increased risk of adverse drug events, emergency department visits, and hospital readmissions. Hsu et al. identified 3 460 older adults hospitalized at a tertiary teaching hospital in Taiwan and found that 65% had increasing PIMs during hospitalization (36). Increasing PIM use was associated with greater functional decline, prolonged length of stay, and higher mortality. Among persons with dementia, Reinold et al. and Hook et al. found a higher anticholinergic burden at discharge (40,41). Psychotropics (eg, antipsychotics and antidepressants) were a major driver.

Our study adds to this literature by taking a more comprehensive approach to identifying problematic medications across several domains. Many of our measures of problematic medications based on 2019 Beers and STOPP V2 criteria identify problematic medications within certain contexts rather than just the presence of the medication (eg, thiazolidinediones in heart failure). We also examined medication changes over a longer time-period following the hospitalization period unlike previous studies. Our primary goal using this time horizon was to characterize an individual’s medication list after follow-up with longitudinal prescribers and to identify problematic medication exposures that have been propagated for months. Using our measures of problematic medications, we show that hospitalizations had a relatively small and statistically insignificant effect on the number of problematic medications among persons with dementia several months after the initial hospitalization (adjusted mean difference = 0.12, p = .12). Although the 95% CI for this difference of −0.03 to 0.26 excludes any large effects, more modest but clinically meaningful effects could still be present. This may be especially true given that certain medications are potentially more problematic than others (eg, benzodiazepine prescription compared with long-term use of a proton pump inhibitor).

This primary finding may be due to several factors. For one, many problematic medications that were initially prescribed prior to the hospitalization (eg, by outpatient clinicians) were continued throughout the hospitalization period and during longitudinal follow-up. Second, new problematic medications started during the hospitalization (eg, antipsychotics for delirium, opioids for pain, or nonbenzodiazepine sedative-hypnotics for insomnia) may have been discontinued prior to hospital discharge. Finally, new problematic medications prescribed upon discharge for hospital-associated temporary reasons may have been deprescribed or not refilled in the outpatient setting.

Our findings complement a 2024 study by Pavon et al. which examined central nervous system (CNS) medication use among older adults around the hospitalization period (44). CNS medication use was generally stable across a hospitalization episode with 74% of individuals exhibiting persistent use of any CNS medication across 4 timepoints (90 days before a hospitalization, at admission, at discharge, and 90 days after a hospitalization). Notably, only 7% of individuals had a CNS medication discontinued during the hospitalization, and 64% of these individuals had a CNS medication started or restarted within 90 days after hospitalization. This study and our findings showing that psychotropic medications were the primary medication class that increased following a hospitalization highlight both the high prevalence of potentially problematic CNS-active medications among older adults and the need to extend hospital-based deprescribing efforts into the postdischarge period.

We found that persons with dementia on fewer prehospital chronic medications demonstrated a larger increase in longer-term problematic medications following a hospitalization. This represents a potentially clinically meaningful increase. One explanation could be that these individuals were generally healthier prior to the hospitalization. An acute crisis event that led to the hospitalization may have precipitated a decline in their health and the use of problematic medications. Clinicians also may feel more comfortable prescribing potentially problematic medications to individuals on fewer medications (eg, due to less concern about compounding side effects and/or drug-drug interactions). Additionally, individuals on a greater number of medications at baseline may have relatively less room to increase the number of problematic medications. This mirrors the work of Growdon et al. who examined psychotropic medication use following a hospitalization (45). In this study, hospitalized persons with dementia showed a high prevalence of preadmission psychotropic medication use (63% using at least 1 psychotropic medication) with only a modest incidence of new psychotropic medication prescribing in the range of 2%–3% per medication class. Thus, there may have been little room for the addition of new psychotropic medications given the high prevalence prehospitalization.

We found that about half of individuals with dementia who were admitted to the hospital experienced either no change or an increase in their potentially problematic medications 4 months after the posthospitalization period. Given the high prevalence of potentially problematic medications among this population, this further highlights the significant barriers that are present to deprescribing. We also found that a large proportion of individuals experienced either a reduction or increase in the total number of problematic medications in both the months preceding the hospitalization (16% reduced, 17% increased) and around the hospitalization period (20% reduced, 23% increased). This suggests that problematic medication prescribing is often dynamic, particularly as many of these medications are for potentially limited durations (eg, antipsychotics for acutely worsening neuropsychiatric symptoms). Although our overall results showed a minimal effect of hospitalization on changes in the mean number of problematic medications, the percentage of individuals who had either a reduction or increase in their problematic medications was somewhat larger for individuals around the hospitalization period compared with the months preceding the hospitalization. Future work examining problematic medication turnover around the hospitalization period in more granular detail using a comprehensive set of measures is needed.

Multiple randomized trials have been conducted involving interventions to reduce problematic medication use among hospitalized older adults (18,19). These have generally involved medication reviews by pharmacists often using criteria such as STOPP or computerized decision support software. Some have shown benefits in reducing hospital readmissions and emergency department visits, particularly those combining medication review in combination with medication reconciliation and patient education (18). There are few trials specifically involving hospitalized persons with dementia. One study of 130 hospitalized older adults with advanced frailty transferring to long-term nursing care (where ~75% had a diagnosis of dementia) found that a STOPPFrail-guided deprescribing plan presented to attending physicians reduced polypharmacy (46). In the postacute setting, the Shed-MEDS trial found that older adults with polypharmacy who received a deprescribing intervention (eg, comprehensive medication review, deprescribing recommendations) took a mean of 14% fewer medications and reduced PIMs at the 90-day follow-up (47).

In comparison, less effort has been devoted to interventions that target outpatient prescribing, particularly among persons with dementia. Although interventions based in the hospital and postacute care setting are important and can play a crucial role in reducing existing problematic medications, our study indicates that efforts to improve longer-term medication appropriateness for persons with dementia may show higher yield by targeting nonhospital prescribing. The OPTIMIZE pragmatic cluster randomized trial was one example of a deprescribing education intervention (eg, educational brochure and deprescribing tip sheets) targeting older adults with cognitive impairment taking ≥5 medications and their primary care clinicians (48). Although the overall results suggested little effect on long-term medications and PIMs, subgroup results suggested that the intervention may be more effective in individuals with higher levels of polypharmacy.

A notable strength of our study is the utilization of a nationally representative sample of community-dwelling persons with dementia linked to detailed medication data. Additionally, we captured longer-term problematic medication prescribing using a comprehensive classification system that spanned several medication domains. Our study does have a few limitations. First, the dementia algorithm we used may be subject to misclassification. Although this algorithm has shown high accuracy in validation studies, its accuracy is reduced in racial and ethnic minorities and less-educated individuals (27). Second, to assess medication use at the posthospitalization timepoint, we excluded individuals who died or enrolled in hospice within 4 months of the hospitalization period. This was necessary given our focus on longer-term problematic medication prescribing several months after the acute hospitalization period, and we felt that decision-making around medication use for these individuals is likely different from our intended population. Third, our measures may not capture the entirety of potentially problematic medication use. For example, we did not have information on medications commonly obtained over the counter, such as aspirin, ibuprofen, iron, and vitamins. Additionally, our measures did not necessarily capture some medications that represent the intensification and likely over-treatment of chronic conditions such as diabetes and hypertension. Treatment intensification has been linked to adverse outcomes in previous studies (14–17). Fourth, our focus on longer-term prescribing limits our ability to identify short-term problematic medication prescriptions that were filled in the immediate postacute period. Finally, Although we only found heterogeneity in problematic medication changes among those with fewer baseline chronic medications, other unmeasured factors may also contribute to variation (eg, acute care for elders unit).

Conclusion

Among community-dwelling persons with dementia, hospitalization had a small, nonstatistically significant effect on changes in longer-term problematic medication use. This finding was consistent across different categories of problematic medications. For individuals on few prehospital chronic medications, hospitalization may play a relatively larger role in longer-term problematic medication use. Given the high prevalence of individuals receiving potentially problematic medications both before and after a hospitalization, our results suggest that nonhospital prescribing plays an important role in longer-term problematic medication use among persons with dementia. Reducing problematic medication prescribing for older adults with dementia requires a system-wide approach that targets the continuum of healthcare settings, including hospitals, postacute care facilities, and outpatient clinics.

Supplementary Material

Supplementary data are available at The Journals of Gerontology, Series A: Biological Sciences and Medical Sciences online.

Funding

This work was supported by the following grants from the National Institute on Aging (NIA): Dr. Deardorff (grant numbers T32AG000212, R03AG082859, P30AG044281), Dr. Growdon (grant number R03AG078804), Dr. Blank (grant number K23AG080163), Dr. Bongiovanni (grant number K23AG073523), Dr. Yaffe (grant number R35AG071916), Dr. Boscardin (grant numbers P01AG066605, P30AG044281), Dr. Boockvar (grant number P01AG066605), Dr. Steinman (grant numbers P01AG066605, P30AG044281, K24AG049057). Dr. Deardorff was supported by the National Center for Advancing Translational Sciences of the National Institutes of Health (grant number KL2TR001870). Dr. Growdon was supported by the National Institutes of Health/Agency for Healthcare Research and Quality (grant number K12HS026383). Dr. Bongiovanni was supported by the Robert Wood Johnson Foundation (grant number P0553126). Dr. Boockvar was supported by the Veterans Affairs Patient Safety Centers of Inquiry.

Conflict of Interest

The authors report no conflicts of interest. The content is solely the responsibility of the authors and does not necessarily represent the official views of the National Institutes of Health, Robert Wood Johnson Foundation, Veterans Affairs, or United States government.

Author Contributions

All authors participated in the study conception and design, interpretation of data, and provided revisions to the manuscript. B.J., W.J.B., and W.J.D. contributed to the statistical analysis. W.J.D. was involved in the preparation of the manuscript.