This study retrospectively analyzed 2015-2017 Medicare data linked to Area Health Resources Files (AHRF) [19]. Medicare data included the Master Beneficiary Summary File (MBSF), Parts A/B claims, the Part D Event (PDE) File, and the Part D MTM Data File [20]. From these Medicare data files, beneficiary-level data were obtained. Specifically, the MBSF provided demographic and enrollment information for Medicare beneficiaries [21], and diagnosis information was obtained from Parts A/B data [20, 22]. Prescription utilization information, such as the drug name, service dates, and days supply, was provided by the PDE File [23], while MTM services information and CMR receipt were provided by the Part D MTM File [24]. Additionally, county-level information, such as healthcare capacity, income per capita, and education levels within the population, was obtained from AHRF [19]. The CMS Research Data Assistant Center facilitated access to the Medicare data utilized for this study.
The study sample included Medicare beneficiaries who met the following criteria in a study year: (1) aged 65 years or older; (2) were alive at the end of the study year; (3) had continuous Medicare Parts A, B, and D coverage; (4) had a diagnosis of AD based on the International Classification of Diseases version 9 (ICD-9) and version 10 (ICD-10) codes identified in medical claims from 2010 to 2017; and (5) met the inclusion/exclusion criteria that Pharmacy Quality Alliance (PQA) developed for calculating its adherence measure for statin medications. Based on the PQA criteria, individuals were included if they received at least two fills for statin medications on separate dates during the study period, with the first fill occurring 91 days prior to the end of the study period; patients were excluded if they had a diagnosis of end-stage renal disease or a record of hospice care [25, 26].
To examine the effects of CMRs, a Difference-in-Differences (DID) approach was used to compare the outcome differences across racial/ethnic groups between a treatment group (CMR recipients) and a control group (CMR non-recipients). The treatment group was composed of MTM enrollees who received a CMR. The control group included non-MTM enrollees who met MTM eligibility criteria but did not receive a CMR. Propensity score matching was utilized to ensure that the treatment and control groups contained patients with comparable characteristics [27, 28]. The propensity score represented the predicted probability of each individual receiving a CMR and was estimated by a logistic regression which accounted for all patient and community characteristics. Individuals in the control and treatment groups were then matched in a 3:1 ratio using the nearest neighbor propensity score without replacement [27, 28]. Finally, the propensity-score-matched treatment and control group members from each study year were pooled to form the final study sample.
The MTM eligibility criteria used by this study were based on the CMS guidelines and MTM program practices [11, 29, 30]. In general, Medicare Part D enrollees were deemed eligible for MTM services by Part D plans if the following three conditions were met: (1) had at least two to three chronic conditions; (2) had at least two to eight Part D covered medications; and (3) were likely to have minimum medications costs of $3138 in 2015, $3507 in 2016, and $3919 in 2017 [11, 29,30,31]. To account for the representative MTM eligibility thresholds in the analysis, the mode values of three chronic conditions and eight covered medications were analyzed for each study year [16, 32, 33]. A list of 25 chronic conditions was used to identify the number of chronic conditions a patient had [31].
A binary outcome variable was constructed to measure nonadherence to statin medications (nonadherent = 1; adherent = 0). While there are several hyperlipidemia medications available, statins were analyzed for this study since statins are the most widely used medications for hyperlipidemia treatment. Nonadherence was measured in terms of the proportion of days covered (PDC), in the same manner as the adherence measure for statin medications developed by the PQA and adopted by the CMS Star Ratings [34]. If the PDC was less than 80% for the statin medication received, the individual was considered nonadherent. Since the service date of the CMR receipt was available, the PDC was measured based on all prescriptions received after the CMR receipt date in the treatment group. For CMR non-recipients, the prescription records for the entire year were used to measure the outcome.
The conceptual framework for this study was the Gelberg-Andersen’s Behavioral Model for Vulnerable Populations [35]. The individual- and community-level characteristics were classified as predisposing, enabling, and need factors based on their relationship to prescription utilization [35]. Predisposing factors refer to patient characteristics that predetermine patients’ utilization of medications. For this study, individual-level predisposing factors were age, gender, and race/ethnicity. The community-level predisposing factors were the proportion of married-couple families, the proportion of the population with high school or higher education, income per capita, and the proportion of the uninsured population. The racial/ethnic groups included were Whites, Blacks, Hispanics, Asians/Pacific Islanders (Asians), and Others. The enabling factors in this study were community-level characteristics that represent the accessibility of healthcare services. These included metropolitan statistical area (MSA), health professional shortage area (HPSA), and census regions. Finally, the need factors are characteristics that represent an individual’s perceived or actual health status. For this study, the need factor was a risk adjustment summary score that indicates the expected healthcare expenses of the individual in relation to the average Medicare beneficiary [36].
Differences in characteristics were compared between CMR recipients and non-recipients by analyzing continuous variables with t-tests and categorical variables with Chi-squared tests. Chi-squared tests were conducted to examine the differences in the proportions of statin nonadherence across racial/ethnic groups between CMR recipients and non-recipients.
Multivariate logistic regression analyses were carried out in two stages. First, an adjusted regression was run separately for each study group to examine the factors affecting the likelihood of nonadherence, particularly the association between the outcome and each minority race dummy variable (Blacks, Hispanics, Asians, and Others) in comparison to Whites. Then, a DID analysis was performed using the same adjusted regression model and interaction terms between CMR receipt and race/ethnicity dummy variables with Whites as the reference group. The odds ratio (OR) of the interaction terms represented the effect of receiving a CMR on racial/ethnic disparity between minority groups and Whites. Specifically, receiving a CMR would be associated with reduced racial/ethnic disparities in the likelihood of statin medication nonadherence if the OR is negative.
Because multiple years of data were pooled for the analyses, some Medicare beneficiaries were likely to appear in more than one year if they met the inclusion criteria for multiple years. Robust standard errors were therefore used to account for potential correlations between the outcomes of a single beneficiary across different years. In addition, because community-level covariates were used, standard errors were clustered at the county level to account for possible correlations within a county. All statistical analyses were performed using SAS®9.4. This study was approved by the Institutional Review Board (approval number: #20-07197-XM) at the corresponding author’s institution.
