Potentially inappropriate medication use is associated with increased risk of incident disability in healthy older adults

Efforts to minimize medication risks among older adults include avoidance of potentially inappropriate medications (PIMs). However, most PIMs research has focused on older people in aged or inpatient care, creating an evidence gap for community‐dwelling older adults. To address this gap, we investigated the impact of PIMs use in the ASPirin in Reducing Events in the Elderly (ASPREE) clinical trial cohort.


INTRODUCTION
Medications play a vital role in primary prevention and health maintenance, but any use of medication requires careful consideration of the balance between risk and benefit. This balance is particularly important in older adults, in whom physiological changes associated with aging alter the pharmacokinetic and pharmacodynamic response to medications leading to increased vulnerability to their adverse effects. 1,2 Medications shown to be associated with excess morbidity (relative to potential benefit) in older adults are identified as "potentially inappropriate medications" (PIMs). 3 Several explicit criteria, including the AGS Beers Criteria ® and the STOPP criteria, have been developed to identify PIMs and guide prescribing decisions in older adults. PIMs have been associated with increased risk of hospitalization, worsening of physical function, and death in vulnerable populations, such as those with cognitive impairment, or in aged care and inpatient care. [4][5][6][7][8] This research focuses on PIMs in vulnerable populations creates an evidence gap. Life expectancy at older ages is increasing 9 and, at the same time, the proportion of older people reporting fair or poor health status and significant functional limitations is decreasing. 10 Many individuals now reach old age in relatively good health and live in the community without the confounding underlying morbidity and frailty that are prevalent in the vulnerable populations, where PIM risks have been established. We have previously reported that community-dwelling, "healthy" older people have similar prevalence of PIMs to the general older population, 11 but it is not known whether these medications carry an increased risk of harm for healthy older adults as has been reported in those with poorer health status. Furthermore, previous research into PIMs has focused on Key points • Among community-dwelling, healthy older adults, potentially inappropriate medications (PIMs) use was associated with significantly increased risk of incident disability (defined as persistent loss of at least one activity of daily living) and hospitalization. • Increased risk of disability with PIMs use was evident in those for whom disability preceded hospitalization, as well as vice versa. • Polypharmacy did not modify the associations, and PIMs users without concurrent polypharmacy were at increased risk of disability and hospitalization.
Why does this paper matter?
Our findings suggest that PIMs use may start the disability cascade among healthy communitydwelling older adults and confirm that caution is warranted when prescribing PIMs to older adults regardless of their health, function, and polypharmacy status.
definitive outcomes, such as mortality or hospitalization, but not functional outcomes, such as the ability to perform daily tasks, which are key to ongoing independence. Understanding the risk profile of PIMs in healthy older adults is necessary to determine how to balance the potential risks of certain medications against the potential benefits in this growing population group. ASPirin in Reducing Events in the Elderly (ASPREE) was a randomized, placebo-controlled primary prevention trial of 100 mg daily aspirin in community-dwelling older adults in Australia and the US. 12 Participants were required to be in good health, free of preexisting major cardiovascular disease, cognitively intact, and able to independently perform basic activities of daily living (ADL). In this analysis, we aimed to determine if baseline PIMs use in community-dwelling older adults was (a) associated with poorer functional outcomes (disability-free survival, death, incident dementia, and incident persistent disability) and hospitalization, and (b) whether any risks observed were attributable to specific classes of PIMs.

ASPREE clinical trial
This is a secondary analysis of data from ASPREE. Briefly, 19,114 healthy people aged 70 years or older (65 years or older for US minorities) were randomized in Australia (n = 16,703) and the US (n = 2411). 12 Recruitment commenced in March 2010 and the trial concluded in June 2017, with a median 4.7 years of follow-up involving annual in-person study visits conducted between 2011 and 2017. At baseline, participants had a lower prevalence of diabetes mellitus, osteoarthritis, chronic kidney disease, obesity, dyslipidemia, and smoking, and hence were generally healthier than the broader population of a similar age. 12 Furthermore, participants reported higher health-related quality of life (HRQoL) than the general older population. 13 Detailed methods and results of ASPREE are described elsewhere. 12,14 Collection of medication from participants Participants were asked to bring their medications, or a current medication list, to their baseline data collection visit. Research staff reviewed each medication and confirmed whether the medication was prescribed by the participant's doctor. All prescription medications were coded according to the World Health Organization Anatomical and Therapeutic Chemical coding system (https://whocc.no/atc_ddd_ index/). Detailed methods of the medication collection and coding process have been described elsewhere. 15

Potentially inappropriate medications
PIMs were defined as any medication where the overall risk associated with their use may outweigh any possible benefit. For this analysis, any medication listed under Table 2 of the 2019 AGS Beers Criteria ® for PIM use in older adults was included. 3 Rather than using all the medications from the criteria, this subgroup of medications was chosen because of the strong recommendation to avoid them, as opposed to other subgroups of medications that should be used with caution or avoided only for certain disease states. Where the data necessary to determine if the medication met the AGS Beers criteria ® (e.g., lack of dose, dosing regimen, or indication data) were not collected, the medication was not considered to be PIM (e.g., insulin sliding scale). Proton pump inhibitors (PPIs) were considered PIMs if they were not co-prescribed with a non-steroidal antiinflammatory drug (NSAID). This analysis pertains to baseline medications that were prescribed prior to the commencement of study medication (i.e., aspirin or placebo) for the trial. Further rationale for the choice of this criteria has been published elsewhere, 11 and a full list of PIMs used in this analysis and medications that were excluded from the PIMs analysis is included in Table S1.

Covariates
Hypertension was defined as a systolic blood pressure of ≥140 mmHg or diastolic blood pressure ≥90 mmHg or pharmaceutical treatment for high blood pressure. Diabetes was defined as self-report of diabetes or fasting blood glucose of ≥126 mg/dL or on pharmaceutical treatment for diabetes. Frailty was categorized on the basis of adapted Fried frailty criteria, which included body weight, strength, exhaustion, walking speed, and physical activity. 16 The category of prefrail included participants who met one or two of these five criteria, and the category of frail included those who met three or more criteria. Polypharmacy was defined as presence of five or more prescription medications.

Outcomes
Loss of disability-free survival (hereafter DFS) was assessed using a composite of the first occurrence of death from any cause, dementia, or persistent physical disability. Persistent physical disability (hereafter disability) was defined as experiencing "a lot of difficulty" or requiring assistance for any one of six basic ADLs for a period of at least six months. 14 Dementia was confirmed using DSM-IV criteria by an independent adjudication committee. Details of the adjudication process have been published elsewhere. 14 Incident disability and incident dementia analysis included the first occurrence of the event of interest regardless of whether it was preceded by or followed by another event of interest. We also examined the association between PIMs and hospitalizations. In this analysis, incident hospitalization was defined as first admission to the hospital for a period of 24 h or more, for any reason.

Statistical analysis
Descriptive statistics (frequencies, percentages, means) were used to summarize PIMs' prevalence data. Unadjusted odds ratios (ORs) were used to describe associations between baseline PIMs' exposure and each of age, study drug treatment group, polypharmacy, and physical function (the ability to walk up one flight of stairs). Cox proportional-hazards regression models were used to estimate the association between baseline exposure to PIMs and DFS and each of death, incident dementia, disability, and hospitalization. Models were then adjusted for previously identified confounders of PIMs within the ASPREE cohort 11 : sex, age, country, years of education, frailty, average gait speed (usual walking speed over 3 m), hypertension, diabetes, polypharmacy, and the presence of depression (using the Center for Epidemiological Studies Depression-10 questions score of ≥8) 12 at baseline. Confidence intervals (CIs) were not adjusted for multiple comparisons. To disentangle the impact of PIMs from polypharmacy, a test for interaction between polypharmacy and PIM was conducted, and an analysis stratified by baseline polypharmacy is presented. To explore the relationship between disability and hospitalization, we conducted sensitivity analysis in those who experienced both events.

RESULTS
At baseline, 7396 (39% of the total) participants had a prescription for at least one PIM ( Figure 1). The proportion of participants who became lost to follow-up or died was similar between those with and without a PIM, as was median follow-up time. Of those prescribed a PIM at baseline, 79% remained on at least one PIM throughout follow-up. Of those who were not prescribed a PIM at baseline, 46% were prescribed a PIM at some stage during follow-up. Participants who were aged 65-69 years were less likely (OR 0.72; 95% CI 0.59, 0.86) to be prescribed a PIM compared with those aged 70-74, and those aged 80-84 were more likely to be on a PIM (OR 1.22; 95%CI 1.11, 1.34), as were participants with polypharmacy or who reported any difficulty walking up a flight of stairs (Table 1). PPIs prescribed without concurrent NSAID use (54.8% of participants with PIMs use) were the most common PIMs, followed by NSAIDs without concurrent PPI use (17.3%), benzodiazepines (17%), androgens and estrogens (11.8%), and drugs with anti-cholinergic properties (11.6%) (see Table S2). With regard to PPI use, 4714 participants were prescribed PPIs at baseline. 17 Of these, 4054 (86%) were considered PIMs based on the 2019 AGS Beers Criteria ® .
As shown in  Table S4 reports hazard ratios for comparison of PIM between those with and without polypharmacy, including an interaction between PIM and polypharmacy. The interaction term is significant only for hospitalization, and the relevant combination of coefficients suggests that, compared to those with PIM and not polypharmacy, those with PIM and polypharmacy do not exhibit elevated risk, all other variables held constant. Table S5 shows the relationship between the individual classes of PIM and clinical outcomes. There was an increased risk of disability associated with use of antipsychotics (adjusted HR 1.96, 95% CI 1.17, 3.28), PPIs (adjusted HR 1.34, 95% CI 1.07, 1.67), and benzodiazepines (adjusted HR 1.38, 95% CI 1.01, 1.89). PPI use that met the criteria for PIM was associated with a lower risk of death following adjustment for confounders (12.1 events for 1000 person years no PIM vs. 11.0 PIM, adjusted HR 0.85, 95% CI 0.73, 0.99), and Table S6 shows cause of death within this group. No difference in rates was detected for any of the outcomes for cardiovascular PIMs or pain medications, excluding NSAIDs.

DISCUSSION
Prescribing medications that have been identified as potentially inappropriate for older people requires T A B L E 3 Risk of disability-free survival, death, dementia, disability, and hospitalization outcomes by baseline potentially inappropriate medications (PIMs) exposure, based on modified 2019 AGS Beers Criteria ® , stratified by polypharmacy status. Incident includes the first occurrence of disability or dementia regardless of whether the event was preceded by another event of interest (e.g., incident disability includes the first occurrence of disability regardless of whether that disability was preceded by dementia).
consideration of the individual circumstances. In clinical practice, it may be deemed that the benefits outweigh the risks for a given individual, especially for adults who have reached an older age free of significant life-limiting illness or disability and therefore may be more robust. However, after adjustment for baseline comorbidities, including frailty, we have found that PIM exposure in healthy older adults without major cardiovascular disease or baseline disability is associated with higher rates of incident physical disability and hospitalization. We found no difference in risk for death or dementia, both of which occur more commonly than disability, and this resulted in a null finding for the broader category of disability-free survivial.

Potential clinical impact of findings related to PIMs and physical disability
To our knowledge, this is the first study to explicitly explore associations between PIMs exposure and disability (defined as the persistent loss of at least one ADL) in community-dwelling older adults. Although there is a well-established association between specific PIMs and fractures [18][19][20] and fractures that have been associated with disability, 21  hospitalization and disability may be components on a disease continuum with a hospitalization for PIMs initiating a cascade of events that results in functional decline. This raises the question of whether hospitalization, which is a well-established risk associated with PIMs use, may be driving the increased risk we observed for disability. However, this does not appear to be the case. Almost half of those who developed disabilities did not have a prior hospitalization. Sensitivity analysis found that PIMs exposure was associated with an increased risk of disability, specifically in those for whom disability preceded hospitalization. This suggests that PIMs may increase the risk of disability through pathways that are not linked with prior hospitalizations, and may even initiate the disability cascade in healthy older adults. Persistent loss of an ADL in older people has been associated with substantial reduction in HRQoL. 23 The ability to live independently is a critical factor in aging and loss of independence is a known fear among the general older population. 24 Bearing this in mind, our findings of an almost 50% increase in the rate of persistent physical disability associated with exposure to PIMs suggest that caution is warranted when prescribing PIMs to older adults, regardless of their health and function.

Context and interpretation of hospitalization and death findings
Findings of increased risk of hospitalization associated with PIM exposure in our study were similar to those previously published, [25][26][27] adding strength to the evidence that PIM use is associated with increased hospitalization not only in more vulnerable populations but also in healthier older adults. Overall, we did not observe any difference in the risk of death in those with PIM exposure compared to those without PIMs. While this result is consistent with recent studies in modestly sized cohorts (500-600 people) of older people discharged from hospital or residents in nursing homes who were likely to be more unwell, 28,29 it contrasts with results of a larger (n = 1606) study in Brazil that showed a 44% increase in the risk of death for PIMs using the 2012 AGS Beers Criteria ® . 30 Our study utilized the updated 2019 AGS Beers Criteria ® and therefore included PPIs, which may explain the differences in results as PPIs were associated with a decreased risk of death in our cohort (see further discussion below). Notably, two recent studies have shown an association between increased risk of death and new initiation of PIMs but not longer term use. 31,32 If ASPREE participants represent longer term PIM users, then our findings would be in line with these two studies. However, ASPREE did not collect data on historical medication use, and hence it was not possible to determine when those on PIMs at baseline were initially prescribed them.

Impact of polypharmacy on PIM findings
In analysis stratified by polypharmacy, the raw rates of disability and hospitalization were higher in those with polypharmacy compared to those without polypharmacy regardless of PIM status. This is consistent with the wellknown risks of polypharmacy as an independent predictor of hospitalization and death in older people. 33 However, we found that participants without polypharmacy who were exposed to PIMs had an increased risk of disability or hospitalization. This suggests that PIMs should still be prescribed, but with caution for older people without polypharmacy.

Risk profile of individual PIMs
No single PIM class appeared to be responsible for the increased risks we observed, and anti-psychotics, benzodiazepines, and PPIs were all independently associated with increased risk of disability in our cohort. Given their established adverse effects, both antipsychotics and benzodiazepines have been consistently included on the list of PIMs from the initial publication by Beers 34 until the most recently released AGS Beers criteria (in 2019). 3 However, PPIs were only recently included. PPI use without appropriate clinical indication or concurrent NSAID use was added to the 2019 criteria based on evidence of increased risk of Clostridium difficile infection and bone loss and fractures. 3 Given that fractures are a risk factor for disability, 35,36 our findings of an increased risk of disability with PPI use are consistent with this rationale. However, we also observed an association between PPI use and a decreased risk of death. This beneficial association is in conflict with previous studies conducted in the general population and in more vulnerable older populations, where PPIs have been linked with increased risk of both mortality. 37,38 The raw difference in mortality rates between those exposed to PPIs that met the criteria and those not exposed was small, and therefore it is possible that our mortality finding is an erroneous result that is attributable to adjustment for confounders. Additionally, our analysis was limited to medications that met the PIMs criteria according to Table 2 of the 2019 AGS Beers Criteria ® . This meant that not all PPIs were considered to be PIMs, and examination of broader PPI use was beyond the scope of our analysis. Although efforts have been made in recent years to reduce the number of older adults on high-dose PPIs, longer term use of high-dose PPIs remains common in Australia, the US, France, and Iceland. [39][40][41][42][43] Therefore, we believe further investigation of the associations between PPI use and functional outcomes, and analysis of PPI use more broadly may be warranted given the substantial proportion of older people taking this class of medication.
Potential impact of changes in prescription medication use during follow-up Older people may change medications frequently. We chose to look at the impact of PIMs exposure at a certain point in time (baseline in the ASPREE study) rather than ongoing or time-varying exposure in order to explore the simple clinical question of whether an older person who is currently on a PIM (for any reason) is at increased risk of poor functional outcomes. While the majority of participants (79%) who were on a PIM at baseline remained on a PIM throughout the entire follow-up period, many of those not prescribed PIMs at baseline commenced on a PIM at some point during follow-up (46%). If anything, initiation of PIMs in the No PIM group likely diluted the true difference between the groups and resulted in an underestimation of the risk of disability associated with PIMs use. Therefore, we do not believe our findings are undermined by our chosen methodology but instead reflect the real-world prescribing environment.

Strengths and limitations
A key strength of our study is the prospective design with regular physical disability screening and robust clinical event adjudication, which minimized ascertainment bias. We used a validated PIMs tool. We used a large sample of healthy older adults and were able to control for a wide range of demographic, lifestyle, and known risk factors. Given the observational design, it is not possible to evaluate causality. Reverse causality is a major factor in measuring the association between medications and health outcomes. However, our participants were clinically free of dementia, myocardial infarction, stroke, and other cardiovascular diseases, such as transient ischemic attack or angina, at baseline and were independent with all ADLs.
Overall, data quality was high with limited missing data. 44 This analysis accounted for a wide range of potential confounding variables, but potential for residual confounding remains. Our analysis was based on the presence of PIMs at study entry, and we did not collect medication dose, nor did we collect non-prescription medications other than NSAID use. Length of exposure prior to randomization was not collected, and therefore we cannot rule out a selection bias caused by an impact of PIMs prior to enrollment. We could not account for changes in PIM exposure during follow-up. There was also a lack of information regarding the clinical indications surrounding the PIM, which limited the level of detail to which the 2019 AGS Beers Criteria ® could be definitively applied. Previous research has shown that the use of explicit criteria to evaluate prescribing may account for only a small portion of drugs deemed inappropriate by implicit review. 45 Ascertainment of medications relied on self-report and the checking of packaging and prescriptions by study staff. However, wherever possible, medical records were used to prompt participants about medications they may have omitted, mitigating the limitations of self-report.

CONCLUSION
PIMs exposure at baseline was associated with increased rates of incident physical disability and/or incident hospitalization in healthy older people without significant baseline cardiovascular, cognitive, or physical impairment. Our findings emphasize the importance of caution when prescribing PIMs to older adults, including those in otherwise good health.
AUTHOR CONTRIBUTIONS Jessica E. Lockery was responsible for the concept and design of work, some data analysis and interpretation, and preparation of the first draft of the manuscript. Taya A. Collyer contributed to the data analysis, drafting, and critical revision of the manuscript. Michael E. Ernst contributed to the design of the work and interpretation of the results. Michael E. Ernst, Suzanne G. Orchard, Anne Murray, Mark R. Nelson, Nigel P. Stocks, Chris Moran, Rory Wolfe, and Robyn L. Woods critically reviewed and revised the manuscript. All authors have approved the final version of this manuscript ACKNOWLEDGMENTS A. G. Bayer provided aspirin and matching placebo. The authors acknowledge the dedicated and skilled staff in Australia and the United States for the conduct of the trial. The authors are also very grateful to the ASPREE participants, who so willingly volunteered for this study, and to the general practitioners and medical clinics that supported the participants in the ASPREE study. Trial Registration: International Standard Randomized Controlled Trial Number Register (ISRCTN83772183) and clinicaltrials.gov (NCT01038583). Open access publishing facilitated by Monash University, as part of the Wiley -Monash University agreement via the Council of Australian University Librarians.

SUPPORTING INFORMATION
Additional supporting information can be found online in the Supporting Information section at the end of this article.
Data S1. Supplementary analysis and tables Table S1. Medications included and excluded as PIMs for this analysis based on modified AGS Beers Criteria. Table S2. Baseline PIMs prevalence by medication class. Table S3. Risk of disability and hospitalization outcomes by baseline PIMs exposure, in participants who experienced both events. Table S4. Assessment of interaction between Cox models for PIMs and Polypharmacy. Table S5. Risk of primary endpoint, death, dementia, disability, and hospitalization outcomes by baseline PIMs medication category exposure.