Prevalence and Persistence of Cost-Related Medication Nonadherence Among Medicare Beneficiaries at High Risk of Hospitalization

IMPORTANCE The unaffordability of drugs has been a persistent and elusive challenge in the US health care system. Little is known about the prevalence and persistence of cost-related medication nonadherence (CRN) in a population with high-cost, high-need resource utilization. OBJECTIVE To evaluate the prevalence and persistence of CRN among Medicare beneficiaries at high risk of hospitalization as well as the characteristics associated with CRN in this population. DESIGN, SETTING, AND PARTICIPANTS This cohort study used survey data from Medicare patients at high risk of hospitalization and with a life expectancy greater than 12 months at an urban academic medical center from November 6, 2012, to January 30, 2018. Patients were followed up for 12 months at 3-month intervals from baseline, for a total of 5 surveys. Data were analyzed from September 1, 2020, to January 5, 2021. MAIN OUTCOMES AND MEASURES Self-reported CRN, using a metric of persistence and transiency. Based on the results of the 5 surveys, CRN was categorized as persistent (3 or more surveys), intermittent (2), transient (1), and any (1 or more). Multiple logistic regression analyses were used to evaluate factors associated with persistent and transient CRN. RESULTS Of the 1655 Medicare beneficiaries followed up during the 15-month study period, 1036 (62.6%) were women and 1452 (87.7%) were Black or African American; 769 (46.5%) were younger than 65 years, and 886 (53.5%) were 65 years or older (mean [SD] age, 62.4 [15.9] years). A total of 374 patients (22.6%) reported CRN at baseline, 810 (48.9%) reported any CRN, and 230 (13.9%) reported persistent CRN (148 [19.2%] of those younger than 65 years and 82 [9.3%] of those 65 years or older). The 230 patients who had persistent CRN accounted for 28% of those who reported CRN at least once during the 15-month study period. Younger age (eg, <50 years vs 75 years: adjusted odds ratio [AOR], 3.07; 95% CI, 1.61-5.86; P = .001), worse self-reported health (AOR, 1.59; 95% CI, 1.10-2.31; P = .01), and depression (AOR, 1.58; 95% CI, 1.11-2.24; P = .01) were associated with greater likelihood of persistent CRN. The population-adjusted prevalence of CRN was 53.6% (887 patients). CONCLUSIONS AND RELEVANCE The findings suggest that CRN is prevalent, moderately persistent, and variable in the Medicare population at high risk of hospitalization despite coverage by insurance. Longitudinal follow-up and refined predictive modeling of CRN appear to be needed to identify and target more precisely those with persistent CRN and to develop effective interventions. JAMA Network Open. 2021;4(3):e210498. doi:10.1001/jamanetworkopen.2021.0498 Key Points Question What are the prevalence and persistence of cost-related medication nonadherence (CRN) among Medicare beneficiaries at high risk of hospitalization, a population with highcost, high-need resource utilization, and what is the longitudinal pattern of CRN over time? Findings In this cohort study of 1655 Medicare beneficiaries, the populationadjusted prevalence of CRN was 53.6%, and 28.4% of those who reported CRN at least once had persistent CRN during the 15-month study period. Younger age, worse self-reported health, and depression were associated with greater likelihood of persistent CRN. Meaning The findings suggest that CRN is prevalent, moderately persistent, and variable in the Medicare population at high risk of hospitalization despite coverage by insurance. Author affiliations and article information are listed at the end of this article. Open Access. This is an open access article distributed under the terms of the CC-BY License. JAMA Network Open. 2021;4(3):e210498. doi:10.1001/jamanetworkopen.2021.0498 (Reprinted) March 3, 2021 1/11 Downloaded From: https://jamanetwork.com/ by a Non-Human Traffic (NHT) User on 08/29/2021


Introduction
High pharmaceutical drug prices have been a persistent and elusive challenge for the US health care system. 1 One in 4 adults in the US has a difficult time affording their medications, 2

and among
Medicare beneficiaries in 2006 11.5% reported medication nonadherence owing to financial barriers. 3 Numerous behavioral, social, economic, medical, and policy-related factors are associated with medication nonadherence, [4][5][6][7][8][9][10][11][12] and medication nonadherence is associated with increased hospitalization rates and emergency department visits, higher mortality rates, worse patient outcomes, and increased downstream costs, all of which impose avoidable and substantial health care costs on society. [13][14][15][16][17][18] With use of cross-sectional data sets over time, researchers have found that although Medicare Part D has provided outpatient prescription drug coverage for Medicare beneficiaries since its implementation in 2006, the prevalence of cost-related medication nonadherence (CRN) has actually increased among the sickest Medicare beneficiaries, including older patients with complex medical needs and people with disabilities. [19][20][21][22] Although these crosssectional estimates showed the aggregate trend in CRN for the populations studied, because the individuals were not longitudinally linked, the level of CRN behaviors, such as persistence and transiency, are unclear. Without knowledge of the key characteristics associated with CRN behaviors, targeting individuals who report CRN occasionally for intervention may be economically inefficient and fail to focus on those who are persistently unable to pay for medications. In addition, the lack of longitudinal follow-up may lead to underestimation of the prevalence rate of CRN and to distorted risk profiles for patients at high risk of CRN.
It is estimated that in the US, patients with high-need, high-cost resource utilization (approximately 5% of the population) disproportionately account for 50% of all healthcare spending, 23 and the Medicare population at high risk of hospitalization well represents this group because many are older adults, have multiple chronic conditions, or experience extreme functional limitations. Although this group has been the topic of many policy discussions, little is known about their behaviors with regard to CRN because physicians and patients infrequently discuss CRN and, when they do, both parties are often frustrated by the lack of a clear solution. 24 In addition, most of the literature 3,10,12 on CRN anchors its measure of prevalence on a 1-time cross-sectional survey, implicitly assuming that a 1-time measure would be persistent or stable during the recall period, such as 1 year. However, if CRN is not static or binary, a 1-time cross-sectional measure may focus an intervention on individuals with transient CRN and fail to target those who have persistent difficulty paying for medications and therefore require structured financial assistance. This is a particularly important distinction for practice and policy in the Medicare population because all patients have public insurance, with some having both Medicare and Medicaid to help cover their coinsurance and premium. 25 A high prevalence of CRN in this population would suggest inadequacy of insurance coverage, poor benefit design for drugs, and/or high price sensitivity by the patients, all of which would require policy action to improve insurance coverage and reduce downstream costs associated with CRN behaviors. Therefore, we evaluated the prevalence, persistency, and transiency of CRN over time in a sample of Medicare beneficiaries at high risk of hospitalization. Understanding the CRN behaviors in this population over time is important because nonadherence to medication may be associated with worse health outcomes and higher downstream costs compared with those in other populations. We also assessed potential factors associated with persistent CRN by analyzing the patients' sociodemographic and health characteristics and how these factors may be associated with the protective effect of Medicare coverage. medical center. 26 The enrollment criterion was hospitalization at least once in the past year or emergency department care at the time of enrollment. The study was approved by The University of Chicago institutional review board, with written informed consent provided by participants. This study followed the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) reporting guideline.
Our internal analysis indicated that the annual health costs for patients fulfilling the enrollment criterion were 300% to 400% of the mean annual health cost for Medicare beneficiaries during the follow-up period. The CRN rate in this population at baseline was likely higher than that in the general population because the patients fulfilling the enrollment criterion likely had higher resource utilization and severity of illness. This criterion allowed us to study the persistency of CRN in a population at increased risk for CRN. The enrollment and study period was from November 6, 2012, to January 30, 2018, during which the US economy steadily recovered from the Great Recession in 2008; thus, data could be analyzed for patient CRN behaviors without major economic volatility.
Although the study on the Comprehensive Care Physician model was a randomized clinical trial, for the present study, we pooled data from the study and control arms because our interest was primarily tracking the pattern of CRN rather than the effect of the model on CRN over time. We controlled for the study arm in the multiple regression analysis to reflect the population-mean CRN adjusted for other confounding factors. We also conducted a stratified analysis of the trajectory of CRN rates in the 2 study arms over time to ascertain whether the CRN rates were similar during this study period.
Our study included 5 surveys of CRN: a baseline survey of CRN in the 3 months before study enrollment and 4 follow-up surveys at 3-month intervals in the subsequent 12 months. Thus, the total study period covered by the CRN questionnaires was 15 months. To fully capture the CRN trajectory for the complete study sample, we excluded 54 patients without baseline data, 7 who eventually withdrew from the study, 61 who did not complete all 4 follow-up surveys, and 223 who died during the 12-month period, resulting in a total of 1655 participants in this study. Patients who died within 12 months after enrollment had a short life expectancy, and we did not have enough observation points to construct a comparable measure of prevalence and persistence over time. In addition, resource utilization among Medicare patients in their last year of life accounts for one-quarter of Medicare spending. [27][28][29] Such heterogeneity in resource utilization patterns (and thus costs to patients) and its association with patients' behaviors requires a separate investigation of CRN to avoid confounding of research.
Patients who did not answer CRN questionnaires intermittently during the 12-month follow-up period after the baseline survey were considered as not reporting CRN for that observation. The overall response rate for the follow-up surveys was 93%. Among the respondents, 2.9% of responses were "don't know" or "refusal" to 1 or more of the 4 CRN questions on the survey.
The CRN was self-reported based on 4 questions that were adopted from the Medicare Current Beneficiary Survey, 3 with a recall timeframe set at 3 months instead of 1 year. The survey included the following question: "during the past 3 months, have you ever done the following due to cost: (1) not fill or refill a prescription, (2) delay filling a prescription, (3) skip doses, or (4) take smaller doses to make medication last longer." The CRN was categorized as 1 if the patients reported any of these 4 options and 0 if they did not report any option. Conceptually, the 3-month timeframe may have reduced the recall bias and provided more detailed information about CRN behavior than the 12-month recall questionnaires. A metric was developed to measure the CRN longitudinally. Any CRN was defined as reporting CRN on at least 1 survey during the study period; baseline CRN, on the enrollment survey; transient CRN, on 1 survey; intermittent CRN, on 2 surveys; and persistent CRN, on Demographic and health characteristics collected at baseline were included to examine variations in CRN between subgroups in the study population and to adjust for confounders to evaluate the population-mean factors associated with CRN. These factors included age, gender, race/ ethnicity, educational level, health literacy, 30 income, insurance, self-reported health, health conditions (cancer; cardiovascular disease, including angina, congestive heart failure, and coronary artery disease; depression; kidney disease; and diabetes), hospitalizations in the past 12 months, and study group (standard of care vs intervention). These factors have been reported to be associated with CRN in cross-sectional studies, 4-12 but how they are associated with CRN longitudinally in a population with high-cost, high-resource utilization is unknown. Of note, although the surveys asked for reports of sex, because such reports encompass a cultural indicator of a person's personal social and cultural identity rather than the biological characteristics of males and females, we use gender instead of sex in the context of CRN behaviors; such behaviors are beyond the biological difference owing to differential cultural roles by men and women. There has been robust discussion in the literature [31][32][33][34] regarding sex differences in the use of health care resources across the life span and the role of gender in contributing to such differences when socioeconomic contexts are incorporated in medical care. Although the literature on gender difference in CRN has increased, 34 to our knowledge, no study has specifically focused on the association of gender with persistence of CRN.
Demographic and health characteristics and CRN were stratified by age (<65 years and Ն65 years) because patients younger than 65 years who were covered by Medicare were mostly those with disability or end-stage kidney disease and had disease profiles and health care needs that were different from those 65 years or older. 12 Research has shown that this age cutoff may be associated with a difference in CRN behaviors. 12

Statistical Analysis
Data analysis was performed from September 1, 2020, to January 5, 2021. We conducted multiple logistic regression analyses to evaluate the potential factors associated with persistent and transient CRN. We conducted a bivariate analysis on the association of gender with transient, intermittent, persistent, and no CRN using χ 2 tests. A multiple regression analysis of the association of gender with the persistence and transiency of CRN was performed, adjusting for other sociodemographic and health variables.
To obtain population-adjusted estimates, we conducted multiple logistic regression analyses for overall CRN, controlling for demographic and health characteristics in the full sample. P = .05 was considered statistically significant for 2-sided tests. The analyses were conducted using Stata, version 15.0 (StataCorp LLC).

Results
Of the 1655 Figure 1 shows the prevalence of CRN behaviors at baseline and during the study period.  These variables were not significantly associated with transient CRN.

Discussion
Cost-related medication nonadherence was widely reported by patients in this Medicare population at high risk of hospitalization despite coverage by Medicare insurance, with a prevalence rate 366%     37 In addition, patients with low income who were not dually eligible could receive extra help elsewhere. 38 Many Medicare drug insurance programs have formularies that require high out-ofpocket payments for drugs that are not preferred or are outside formularies. 39 High out-of-pocket

Limitations
This study has limitations. First, self-reporting is always affected by recall bias and may also be affected by a sense of shame when reporting CRN; thus, the CRN rates reported in this study may be underestimates. We aimed to reduce recall bias in this study by setting the recall time to 3 months instead of the 12-month period used in most studies. 3

Conclusions
In this cohort study, Medicare beneficiaries at high risk of hospitalization had higher rates of CRN than the general Medicare population despite having insurance coverage. A significant proportion of individuals reported persistent CRN, and the number of patients with transient and persistent CRN differed between those younger than 65 years and those 65 or older. More research appears to be needed to understand CRN patterns in this population to formulate health and social policies to identify and target those with persistent CRN and improve the efficiency of interventions.