Polypharmacy and mortality association by chronic kidney disease status: The REasons for Geographic And Racial Differences in Stroke Study

Abstract Many Americans take multiple medications simultaneously (polypharmacy). Polypharmacy's effects on mortality are uncertain. We endeavored to assess the association between polypharmacy and mortality in a large U.S. cohort and examine potential effect modification by chronic kidney disease (CKD) status. The REasons for Geographic And Racial Differences in Stroke cohort data (n = 29 627, comprised of U.S. black and white adults) were used. During a baseline home visit, pill bottle inspections ascertained medications used in the previous 2 weeks. Polypharmacy status (major [≥8 ingredients], minor [6–7 ingredients], and none [0–5 ingredients]) was determined by counting the total number of generic ingredients. Cox models (time‐on‐study and age‐time‐scale methods) assessed the association between polypharmacy and mortality. Alternative models examined confounding by indication and possible effect modification by CKD. Over 4.9 years median follow‐up, 2538 deaths were observed. Major polypharmacy was associated with increased mortality in all models, with hazard ratios and 95% confidence intervals ranging from 1.22 (1.07–1.40) to 2.35 (2.15–2.56), with weaker associations in more adjusted models. Minor polypharmacy was associated with mortality in some, but not all, models. The polypharmacy–mortality association did not differ by CKD status. While residual confounding by indication cannot be excluded, in this large American cohort, major polypharmacy was consistently associated with mortality.


| INTRODUC TI ON
Americans consume many prescription and over-the-counter (OTC) medications. 1,2 With over 300 000 marketed OTC products 3 and approximately 5 billion OTC products purchased annually, 4 an estimated 70%-90% of illnesses involve at least some self-treatment. 5 While medications' health benefits are indisputable, approximately half of all prescriptions may be used improperly. 6 Additionally, drugs' side effects are often treated with more medication, leading to a "prescribing cascade." 7 Drug allergies, drug-drug and drugdisease interactions, and direct toxicity are all hazards. If categorized as a disease, adverse drug reactions are estimated to be the fourth leading cause of death. 8 Polypharmacy, or high medication use, 9 can exert polytherapeutic effects and/or polytoxicities. 10 The term "polypharmacy" sometimes has negative connotations, suggesting inappropriate/ excessive medication use; however, the simultaneous administration of many drugs can be the standard of care. Polypharmacy is often defined in two ways: using more drugs than clinically warranted or taking more than a threshold drug count, for example, five. 11 Polypharmacy is a known risk factor for adverse health events, including cognitive decline, 12,13 falls, 14,15 and adverse drug reactions. 16 Based on possible drug-drug interactions 17 and adverse drug reactions, 16 polypharmacy poses plausible risks; however, the relation of polypharmacy with mortality among general, communitydwelling Americans remains largely unexplored. Individuals with chronic kidney disease (CKD) may be especially vulnerable to any adverse effects of polypharmacy because kidney function is critical for drug excretion; however, data are very limited on CKD's role in the polypharmacy-mortality association. Addressing these knowledge gaps, we analyzed the large, national REasons for Geographic And Racial Differences in Stroke cohort.

| Study design
REasons for Geographic And Racial Differences in Stroke cohort is a nationwide, longitudinal study that began in 2003 and was described previously. 18 Briefly, the analytic sample consisted of 29 627 (Data S1) community-dwelling black and white Americans age ≥45 years with at least one follow-up mortality assessment. The cohort recruitment occurred throughout the continental United States using the Genesys commercial database, 19

| Data
A computer-assisted telephone interview collected information on demographic, socioeconomic status (SES), medical, and lifestyle variables. Examination Management Services Inc. scheduled a home visit and instructed the participant to collect all medicines used in the previous 2 weeks. During the home visit, signed informed consent was obtained, and anthropomorphic measurements and blood samples were collected and sent to a central laboratory. The company's personnel examined each medicine present ("pill bottle" inspection including creams/eye drops/injectables) and cataloged its name (generic/brand), but neither dose nor use frequency, on a standardized form. Medications given outside the home, such as at an infusion center, were not included. These records were processed into an electronic database of 34 776 distinct recorded medication names. For prescriptions/OTCs, a generic name and medication class were assigned (e.g., acetaminophen, miscellaneous analgesic) by a research pharmacist and project staff using Drugs.com. 22 For 1.62% of medications, the generic name could not be identified (e.g., "amocardone" or "tylewok") and were assigned the generic name "unknown". Each unknown was assumed to correspond to one generic ingredient.
When assessing polypharmacy, supplements (vitamins/minerals/herbals/nutraceuticals) were excluded due to their heterogeneity, lack of universal nomenclature, and limited US Food & Drug Administration oversight. 23,24 Some vitamins/minerals are available both in supplemental and prescription varieties; we tried to distinguish the prescription forms which counted toward polypharmacy (e.g., isotretinoin) from the OTC-available forms (e.g., vitamin A) that were considered supplements. Many drugs come in combination form; the combination pill generic ingredient count was the total number of active ingredients. Some participants reported taking the same generic drug multiple times, whether from different formulations (e.g., long-, medium-, and short-acting insulin) or using the same medicine twice (e.g., two different acetaminophen-containing, multicomponent analgesics); in such cases, the total ingredient sum included that agent multiple times.
Polypharmacy was characterized using three categories of total prescription/OTC medication ingredient counts (excluding supplements), as suggested elsewhere 25 : no polypharmacy (≤5 total ingredients), minor polypharmacy (6-7 ingredients), and major polypharmacy (≥8 ingredients). Presence of CKD was defined as selfreported dialysis or glomerular filtration rate <60 mL/min/1.73 m 2 using the modified diet in renal disease equation applied to serum creatinine collected with baseline laboratories (albuminuria was not considered). 2,26 Cohort members were called approximately every 6 months to ascertain vital status. Additionally, deaths were identified through

| Statistical analysis
Cox proportional hazards models with the time-on-study outcome (or attained age outcome 31,32 ) until death or censoring examined the polypharmacy-mortality association. CKD was evaluated a priori as a potential effect modifier of polypharmacy on mortality. The agetime-scale models included the same covariates, except attained age was instead the outcome of interest (conditioning on study entry age, with birth cohort stratification). Models 1-7 (Table 1) are subsets of the "full" model 8. Multiple models were utilized because the causal pathway for polypharmacy and mortality is not established, particularly given this cohort's heterogeneity. Aside from models 1-7, no other "reduced" models were considered.
Two propensity-adjusted models addressed confounding by indication. 33 In these models, all candidate confounders from Table 2 were included in a multiple logistic regression (propensity) analyses that used binary polypharmacy status (defined as ≥8 total ingredients) as the dependent variable. Each participant's polypharmacy propensity was estimated, and participants' propensities (irrespective of actual polypharmacy status) were divided into quintiles or deciles. After stratifying on propensity quintiles or deciles, a stratified, no-interaction (hazard ratio assumed constant for all propensity quintiles/deciles) Cox proportional hazard regression used only major/minor polypharmacy as mortality predictors.
Collinearity was assessed for the time-on-study models using a Statistical Analysis Software (SAS) macro. 34 SAS 9.2 was used. The proportional hazards assumption for the time-on-study models was checked by constructing univariable log-log survival plots and by examining univariable model Schoenfeld residuals 35 failure time correlations. 36 For the age-time-scale models, the proportional hazards assumption was assessed with Schoenfeld residuals.

| RE SULTS
Overall, 171 573 individual medications were transcribed during in-home visits. The most common generics and medication classes are shown in Tables S3 and S4, respectively. Among all 30 181 participants, 21.1%, 15.8%, and 63.2% were categorized as receiving major, minor, and no polypharmacy, respectively. The cohort characteristics comparing the major polypharmacy group (polypharmacy+) to all others (polypharmacy−) are presented in Table 2. In the analytic sample, the mean age was 64.9 years, 45% were male, 41% black, 56% stroke-belt residents, 24% with normal BMI, 11% with CKD, and 16% and 31% were in "excellent" and "very good" selfreported health, respectively. Relative to the polypharmacy group, those with major polypharmacy (polypharmacy+) included a greater proportion of females and stroke-belt residents, and those with less education, lower income, higher BMI, more comorbidities (CKD, hypertension, dyslipidemia, diabetes, coronary artery disease), and lower self-reported health ( Table 2). In crude analyses, older adults, blacks, males, individuals with less education or income, smokers, those with poorer self-reported health, and those with comorbidities showed higher mortality.
Median follow-up was 4.9 years; 2538 deaths occurred. As seen in the Kaplan-Meier plot (Figure 1), major polypharmacy had the highest mortality, followed by minor polypharmacy, and the no-polypharmacy group (log-rank p < .0001). In all time-on-study (  The two methods of modeling time-to-event (age-time-scale and time-on-study) gave similar results with <3% difference across model-specific hazard ratio estimates. The models that controlled for propensity scores using stratification gave results consistent in magnitude with models including covariates as separate terms (Table S2).

| DISCUSS ION
Drugs play vital and irreplaceable roles in medicine. While polypharmacy may sometimes be the standard of care, polypharmacy can occur unnecessarily and inappropriately, exposing patients to potentially serious risks and inspiring the call for "deprescribing." 12,13,37 In this longitudinal study of a racially diverse, nationwide sample of the general U.S. adult population, we found that research involves European geriatrics. One factor that likely contributes to the literature's incongruous findings is that most research cannot distinguish rational, indicated as polypharmacy (e.g., using aspirin, statin, beta blocker, and angiotensin receptor blocker following a myocardial infarction), from illogical, "haphazard" polypharmacy (a type 1 diabetic who is prescribed metformin and glipizide in addition to insulin). Assuming confounding by indication could be fully controlled, then evidence-based polypharmacy would be expected to decrease mortality (assuming the medications contributing to the polypharmacy were for high-risk pathologies such as cardiovascular disease and diabetes and not symptomatic relief such as acetaminophen for osteoarthritis). As such, depending on the proportion of cohort members for which polypharmacy resulted from medication accumulation and not thoughtful prescribing, a positive polypharmacy-mortality association would be anticipated.
Conversely, if cohort polypharmacy reflects the implementation of evidence-based clinical guidelines, then a negative polypharmacymortality hazard ratio is expected. Finally, a null association would be predicted if both rational, beneficial polypharmacy and disorganized, TA B L E 1 Multiple models considered to assess polypharmacy-mortality association (Continues) deleterious polypharmacy were found in roughly equal proportions in the cohort.
To briefly summarize the largely international literature on polypharmacy-mortality: Jyrkka reported mixed-polypharmacy mortality results among Finns, 38 and Espino found a positive association among Mexican-Americans. 39 Iwata reported higher 1-year mortality among Japanese elderly polypharmacy users following hospital discharge. 40 Incalzi reported higher in-hospital mortality among Italian polypharmacy patients. 41 Richardson reported higher 2-year mortality in older UK polypharmacy users. 42 Spanish, 43 French, 44 Italian, 45 Chinese, 46 Brazilian, 47 and New Zealand 48 geriatric research also reported increased mortality among polypharmacy patients. Conversely, Wauters found no polypharmacy-mortality association in a small geriatric Belgian cohort 49 and, furthermore, report an association between geriatric medication underuse and mortality. 50  Regarding the limited prior exploration of the polypharmacymortality relation in Americans, Secora used the Atherosclerosis Risk in Communities (ARIC) cohort and found an overall similar polypharmacy dose-response association with mortality, along with a lack-of-effect modification by CKD status. 56 However, the REGARDS cohort is much larger than ARIC and has a national scope, we defined polypharmacy differently, and we used a broader range of models. Finally, the consistency of results with our propensity-matched analyses contrasts with Schöttker's analyses of polypharmacy and mortality in German adults where their original multivariate-adjusted association was lost after also controlling for propensity score. 57 The finding of significant hazard ratios for major polyphar- F I G U R E 2 Kaplan-Meier all-causemortality plot for polypharmacy*CKD status (log rank p-value < .0001). Green = CKD −, no polypharm; Red = CKD −, minor polypharm; Blue = CKD −, major polypharm; Yellow = CKD +, no polypharm; Pink = CKD +, minor polypharm; Brown = CKD +, major polypharm numerous biological processes converging in death, it is difficult a priori to distinguish confounders from mediators.
As such, given the heterogeneous biological nature of both exposure and outcome, selecting an "optimal" model that accounts for the underlying pharmacology is challenging; the results are conditional on the models. We addressed this problem by conducting analyses comparing the "full" model (with many possible confounders) to a series of reduced models.
Additional important limitations include that no information on medication indication, dose, or use frequency/use duration was collected. Also, it is implicitly assumed that one baseline medica-

| CON CLUS IONS
We found a polypharmacy and all-cause mortality association. As hypothesized, mortality was related to polypharmacy degree; however, unexpectedly, no CKD effect modification was observed.
Further research is warranted to understand the impact of drug dosages and the relative contributions of different drug classes to the observed polypharmacy-mortality relationship. The specificity of the biological pathway(s) (e.g., refined pharmacological exposure beyond simple medication count) and exploration of potential CKDbased polypharmacy vulnerability (or therapeutic opportunity) merit further investigation.

D I SCLOS RE
The authors have no conflicts of interest to declare.

DATA V E R I FI C ATI O N
WC had full access to all study data and takes responsibility for the integrity of the data and the accuracy of the data analysis. All authors confirm a role in the manuscript.

ACK N OWLED G M ENTS
This research project is supported by cooperative agreement U01 Additional funding for WC was provided by MSTP 2T32GM008169-28 and T32 DK061296. Representatives from the NIH did not have any role in the design and conduct of the study, the collection, management, analysis, and interpretation of the data, or the preparation or approval of the manuscript.

DATA AVA I L A B I L I T Y S TAT E M E N T
The data that support the findings of this study are available on request from the corresponding author. The data are not publicly available due to privacy or ethnical restrictions.