Identifying adherent patients to newly initiated statins using previous adherence to chronic medications

BACKGROUND: Statins are one of the most frequently prescribed medications in the United States. Despite well-documented benefits in managing hyperlipidemia and reducing cardiovascular risks, statin adherence remains suboptimal. Several effective interventions have been implemented to improve adherence to statins. However, identifying patients who are at risk for developing poor medication adherence at the time of treatment initiation could assist in planning early targeted interventions. Studies have suggested that previous adherence to chronic medications is a strong predictor of future adherence to newly initiated medications. OBJECTIVE: To investigate patients’ adherence to newly initiated statins by measuring previous adherence to angiotensin-converting enzyme inhibitors (ACEIs), angiotensin II receptor blockers (ARBs), and oral antidiabetic drugs (OADs). METHODS: A retrospective cohort study was conducted using administrative claims data from January 2016 to May 2018. New statin initiators were identified and included in the study if they were continuously enrolled in the health plan and had at least 1 prescription for ACEIs, ARBs, or OADs 1 year before statin initiation (pre-index period). Baseline adherence to ACEIs/ARBs, OADs, or both was calculated during a 1-year pre-index period using proportion of days covered (PDC) and defined as PDC ≥ 0.80. Adherence to statins was assessed 1 year after statin initiation and was the primary outcome, with a PDC ≥ 0.80 considered as adherent. Patient demographics were measured during the pre-index period. Multivariable logistic regression was conducted for each cohort separately to determine an association between baseline adherence and future statin adherence controlling for various demographic and clinical characteristics. RESULTS: 1,223 ACEI/ARB users, 714 OAD users, and 452 concomitant ACEI/ARB and OAD users were identified. In the regression model, adherence to baseline medications was significantly associated with 1-year adherence to statins (ACEI/ARB users: OR = 1.75, 95% CI = 1.37-2.25; OAD users: OR = 2.02, 95% CI = 1.46-2.79; concomitant ACEI/ARB and OAD users: OR = 1.73, 95% CI = 1.16-2.58). CONCLUSIONS: Past adherence to baseline medications may predict future adherence to newly initiated statins. Identifying patients likely to become nonadherent during treatment initiation could enable health care providers in recognizing individuals at risk of nonadherence and intervene earlier to enhance future adherence.


METHODS:
A retrospective cohort study was conducted using administrative claims data from January 2016 to May 2018. New statin initiators were identified and included in the study if they were continuously enrolled in the health plan and had at least 1 prescription for ACEIs, ARBs, or OADs 1 year before statin initiation (pre-index period). Baseline adherence to ACEIs/ARBs, OADs, or both was calculated during a 1-year pre-index period using proportion of days covered (PDC) and defined as PDC ≥ 0.80. Adherence to statins was assessed 1 year after statin initiation and was the primary outcome, with a PDC ≥ 0.80 considered as adherent. Patient demographics were measured during the pre-index period. Multivariable logistic regression was conducted for each cohort separately to determine an association between baseline adherence and future statin adherence controlling for various demographic and clinical characteristics. RESULTS: 1,223 ACEI/ARB users, 714 OAD users, and 452 concomitant ACEI/ARB and OAD users were identified. In the regression model, adherence to baseline medications was significantly associated with 1-year adherence to statins (ACEI/ARB users: OR = 1.75, 95% CI = 1.37-2.25; OAD users: OR = 2.02, 95% CI = 1.46-2.79; concomitant ACEI/ARB and OAD users: OR = 1.73, 95% CI = 1. 16-2.58).

CONCLUSIONS:
Past adherence to baseline medications may predict future adherence to newly initiated statins. Identifying patients likely to become nonadherent during treatment initiation could enable health care providers in recognizing individuals at risk of nonadherence and intervene earlier to enhance future adherence.
What is already known about this subject • Despite well-established benefits, statin adherence rates are suboptimal.
• Targeting medication adherence interventions at potential nonadherent patients will increase their efficiency.
• Adding information from past medication adherence to administrative claims data improves predictability of the models identifying nonadherent patients.

What this study adds
• Past adherence to angiotensinconverting enzyme inhibitors, angiotensin II receptor blockers, and/ or oral antidiabetic drugs is strongly associated with future adherence to statins within 12 months of treatment initiation.
• Using previous adherence to chronic medications, health care providers and payers could identify statin new users likely to be nonadherent at the time of treatment initiation and develop interventions to enhance future adherence.
Statins are one of the highest prescribed class of drugs in the United States. A nationally representative data analysis reported that more than 25% of U.S. individuals aged over 40 years were prescribed statins from 2012 to 2013. 1,2 According to recent recommendations by the American College of Cardiology/American Heart Association (AHA), the number of patients indicated for statin therapy is expected to rise. 3 AHA recommends statins as a cornerstone medication for preventing cardiovascular events such as stroke, myocardial infarction, and coronary artery disease. 4 Statins have been the mainstay of lipid-lowering therapy over the past decades. 3 Despite well-established benefits in managing dyslipidemia and reducing risks for cardiovascular mortality and morbidity at a low cost, 5-11 statin adherence rates are still suboptimal. Literature indicates that about 50% of patients discontinue statin therapy within a year of treatment initiation, and this number grows over time. [12][13][14][15][16][17] This trend in poor adherence is in line with key findings of the Centers for Disease Control and Prevention, which reports that approximately 50% of the patients are nonadherent to their medications at any given time. 18 Poor medication adherence is a significant and challenging public health issue that has been identified as the leading cause of preventable morbidity, mortality, and health care costs. 19 Although some successful interventions have been designed to improve adherence, [20][21][22][23][24] it is widely acknowledged that these interventions should be targeted to appropriate patients and prioritize those who may benefit from it. 25 According to a recent systematic review, almost all the clinical trials assessing the effect of interventions on statin adherence recruited patients without previous adherence problems, which may not be a feasible and cost-effective approach in larger scales. 26 Given the limited time and resources for all stakeholders involved in the development and implementation of adherence programs, it is crucial to identify patients who are likely to fail treatment due to poor medication adherence in the future. This will further help clinicians, pharmacists, and other health care providers implement targeted interventions, preventing early discontinuation and maximizing efficacy of the treatment. 27 However, identifying individuals who require such interventions has remained challenging.
Several approaches were implemented by earlier research to predict medication-taking behavior for statins, using nonmodifiable and modifiable factors. While age, gender, ethnicity, and comorbidity are nonmodifiable patient factors associated with adherence, patients' beliefs, attitudes, and behaviors are considered potential modifiable predictors for adherence. 28 First, previous observational studies used patients' demographics, socioeconomic status, and regimen complexity to predict future adherence. These factors were successful in predicting patients' adherence. 29,30 However, they could not strongly discriminate between the nonadherent and adherent cohort. 31,32 Second, past prescription refill behavior was suggested as a better predictor for statin adherence compared with patient medication-taking health beliefs among patients enrolled in a national managed care plan. 33 Third, Franklin et al. (2015) used group-based trajectory modeling to predict a 12-month trajectory pattern by incorporating initial patterns of refill among statin initiators. 34 In this study, Franklin et al. reported that statinfilling behavior during the 2-4 months after treatment initiation strongly predicted future adherence trajectory in Medicare beneficiaries aged 65 years and older. 34 Finally, previous adherence to medications prescribed for other chronic conditions, including antihypertensives, antidiabetics, anticoagulants, and antidepressants, was measured to predict future adherence to statins using administrative claims data. [35][36][37] Results from these studies suggested that adding measures of past adherence to chronic medications could enhance predictability of the models compared with other models using baseline clinical factors; therefore, it may help identify patients at a higher risk of nonadherence.
The purpose of this study was to investigate patients' adherence to newly initiated statins by measuring previous adherence to common medications used for chronic conditions, including angiotensin-converting enzyme inhibitors (ACEIs), angiotensin II receptor blockers (ARBs), and oral antidiabetic drugs (OADs). The hypothesis was that patients who are adherent to their past chronic medication use will show better adherence to newly initiated statins over the first 12 months. In addition to the frequent use of ACEIs/ ARBs and OADs in patients who are prescribed statins, these medications have been selected for the baseline adherence measures, since adherence to them is part of the Centers for Medicare & Medicaid Services (CMS) Star Rating Program to measure health plan performance. 38,39

STUDY DESIGN AND DATA SOURCES
A retrospective cohort study was conducted using administrative claims data from patients enrolled in a Texas-based Medicare Advantage plan between January 2016 and May 2018. Several data files from the health plan were used in the analysis, including membership and member summary, medical claims, and pharmacy claim files. Membership files contained demographic information and CMS risk scores for patients. Medical claims included diagnostic information in the form of International Classification of Diseases, Tenth Revision, Clinical Modification (ICD-10-CM) codes and into 3 separate cohorts based on their baseline medication use: cohort I-patients who had any prescriptions for ACEI or ARB medications; cohort II-patients who had any prescriptions for OADs; and cohort III-patients who had any prescriptions for ACEIs/ARBs and OADs. The study design is illustrated in Figure 1.

OUTCOME MEASUREMENT
Adherence to statins was the outcome of interest for this study and was measured using proportion of days covered (PDC) during the 1-year follow-up period. One-year PDC was calculated as the total number of follow-up days covered with dispensed medications divided by the total number of days in the follow-up period (360 days). 40,41 PDC was dichotomized to categorize patients as "statin adherent" (≥ 0.80) and "statin nonadherent" (< 0.80) groups.

BASELINE MEASUREMENTS
Baseline adherence was the primary explanatory variable of the study and was measured separately for patients in each cohort during the 1-year pre-index period using PDC. Adherence was defined as 1-year PDC ≥ 0.80 for ACEI/ARB use in cohort I and for OAD use in cohort II. In the third cohort, patients were considered adherent if they had 1-year PDC ≥ 0.80 for both ACEI/ARB use and OAD use. 42 Since adherence was measured in accordance to pharmacologic classes and not individually, switching was allowed within drug categories, as well as between ACEIs and ARBs.
Patient demographics and clinical characteristics were measured during the pre-index period and conceptualized procedure information in the form of Current Procedural Terminology codes for outpatient encounters. Pharmacy claims included patient and drug identifying information, fill dates, dosage information, and days supplied for each prescription.
This study was approved by the Institutional Review Board at the University of Houston.

STUDY POPULATION
Using pharmacy claims, all patients who newly initiated statins from January 2017 until the end of May 2017 were identified (identification period). The date of statin treatment initiation during the identification period was determined as the index date. One year before the index date was defined as the pre-index period. Patients were followed for 1 year after the index date, which was considered as the follow-up period. New statin users were identified if they did not have any previous prescription for statins within 180 days before the index date. 34 Statin initiators were included in the study if (a) they had at least 1 pharmacy claim for ACEIs, ARBs, or OADs during the 3-month period between 455 and 365 days before the index date and (b) they had continuous enrollment over the entire study period. Patients were excluded if they had contraindications for ACEI/ARB use, which included angioedema, hyperkalemia, and renal artery stenosis, during the pre-index period or for statin use that included cirrhosis, rhabdomyolysis, and end-stage renal disease during the follow-up period, identified by ICD-10-CM codes. The study categorized patients

FIGURE 1
Study Design

SENSITIVITY ANALYSIS
In the sensitivity analysis, patients in all 3 cohorts were required to have at least 2 pharmacy claims for statins, as well as the baseline medications (ACEI/ARBs or OADs or both) to calculate PDC. We ran similar logistic regression models on the cohorts with stricter inclusion criteria. All statistical analyses were performed using SAS version 9.4 (SAS Institute, Cary, NC) at a priori significance level of 0.05.

Results
In total, 22,925 patients enrolled in the Texas-based Medicare Advantage plan received at least 1 statin prescription during the identification period. Of these, 3,549 were incident statin using Andersen's behavioral model (ABM). Based on ABM, determinants of an individual's health service utilization are categorized as predisposing, enabling, and need characteristics. 43 In this study, predisposing characteristics included age and gender, enabling characteristics included low-income subsidy (given to eligible Medicare beneficiaries with limited income and resources and helps them with prescription drug costs), and need characteristics included comorbidities identified by ICD-10-CM codes, number of unique medications on the index date (apart from statin), CMS risk score, and previous hospitalizations. 44 The CMS risk score is an indicator for medication burden and disease severity and consists of 189 disease classifications that are provided in Medicare Advantage plan data. 12,45,46

STATISTICAL ANALYSIS
Descriptive analysis was performed to summarize patient demographics and clinical characteristics. Group differences were assessed using chi-square tests for categorical variables and t-tests for continuous variables. A multivariable logistic regression was conducted to investigate the association between adherence to the baseline medication and future statin adherence within each cohort, controlling for various demographic and clinical characteristics.

FIGURE 2
Cohort Formation   Table 2 represents the results from logistic regression analysis.

SENSITIVITY ANALYSIS
The results of sensitivity analysis showed that the primary explanatory variables among all 3 cohorts remained significant in the regression models using at least 2 pharmacy claims requirement for measuring adherence to statins and baseline medications (cohort I: OR =   35 By grouping patients into 3 separate cohorts using specific drug classes for baseline adherence measurements, our study attempted to reduce some of the potential biases caused by different safety and efficacy profiles affecting patient's adherence and determine the association between past adherence and future statin adherence in a more homogeneous group. Another study by Franklin et al. (2015) predicted adherence trajectories for 1 year after statin initiation using the initial 2-4 months filling indicator. 34 Although they found that initial adherence is a strong and useful tool to predict future adherence pattern over baseline clinical characteristics alone, they did not take into account the effect of past adherence to other chronic medications on future adherence to newly initiated statins. 34 According to our study, patients aged ≥ 80 years were less likely to be in the statin adherent group as compared with those aged < 70 years. The association between age and adherence is still controversial; however, our results are consistent with studies concluding that medication adherence is adversely affected by age. 29  impairments, multiple comorbid conditions, complex treatment regimens, and higher adverse effects over time. 49 Another finding from the present study was that patients with depression had a higher likelihood of being adherent to statins. Existing literature provides varied findings regarding the association between adherence and depression. [50][51][52][53][54] A potential reason is that medication adherence is influenced by the severity of depression and depressive symptoms experienced by patients. Patients diagnosed with depression who are in remission due to antidepressant use may have improved medication adherence. However, we did not measure antidepressant use and depression severity, since the association was beyond the scope of the current study.
The last significant finding from our analysis was that patients receiving low-income subsidies were more likely to be adherent to statins compared with those with no income subsidy. This association is in accordance with a previous study that revealed that low-income subsidy beneficiaries from a Medicare Part D plan had lower out-of-pocket costs, higher medication fills, and better adherence. 55 However, we only found this significant association in the cohort of ACEI/ARB users.
The variables of gender; comorbidities (e.g., dementia, CVD, hypertension, T2D, hyperlipidemia, and CKD); previous hospitalization; number of other medications on the index date; and CMS risk score were not found to be significantly associated with the study outcome variable in any of the cohorts. According to a systematic review, females are less likely to be adherent to statins. 56 However, the current study did not find a significant difference in statin adherence across gender, which may be due to the new statin user population, as well as the limited follow-up time in this study.

Conclusions
This study demonstrated that, among patients enrolled in a Medicare Advantage plan, adherence to past medication used for chronic conditions such as ACEIs, ARBs, and OADs is associated with future adherence to newly initiated statins. Given that medication adherence is a modifiable health behavior, identifying patients at a higher risk of becoming nonadherent at the time of treatment initiation could enable health care providers and payers to identify barriers early on, develop targeted intervention plans, and enhance future adherence to statin therapy.

LIMITATIONS
Our findings should be interpreted in light of some limitations. First, the data available for the present study lacked information for some sociodemographic variables such as race or education, which could be potential confounders. Second, adherence was calculated from pharmacy claims that only reflect prescription-filling behavior and not actual medication-taking patterns.
Third, pharmacy claim data did not allow for the assessment of the reasons for medication nonadherence or discontinuation. Statin therapy may be discontinued appropriately by physicians due to adverse effects, frailty, comorbidity, and polypharmacy, especially in the very elderly population, considering the limited benefit to and life expectancy of these patients. 59 However, by excluding patients with potential precautions or contraindicated use, we expected to have a low number of patients with poor medication adherence due to serious adverse effects.
Fourth, although 0.80 PDC is a commonly accepted cut-off point for measuring adherence, we may have lost some information by creating a binary variable.
Finally, the generalizability of the results of this study is limited, since it was conducted among patients enrolled in a Medicare Advantage plan based in Texas. Given that the majority of patients enrolled in Medicare Advantage plans are from the elderly population, our findings could not be generalized to younger adults. Therefore, future studies should evaluate the generalizability of the results in patients with different age groups, insurance providers, and from different geographical areas.