Medication adherence among people with cardiovascular disease: A multivariable predictive model development and validation

Abstract

<p><b>Background</b>: Medication non-adherence is common among people with cardiovascular disease (CVD) due to its chronic nature that requires long-term therapy and multiple medications. Medication adherence support is required for people with CVD. Supporting all patients may not be possible in health settings with a high patient burden, busy workflow or limited resources. Consequently, people with high-risk of non-adherence need to be identified using a predictive model to prioritize them for medication adherence support. However, a medication adherence predictive model is lacking for people with CVD. This research project aimed to identify and validate outcome measure (medication adherence), identify candidate predictors for medication adherence, and then develop and internally validate a multivariable medication adherence predictive model for people with CVD.</p> <p><b>Methods</b>: A COSMIN systematic review was conducted following a published protocol and COSMIN guidelines to identify studies that reported on the psychometric properties of medication adherence patient-reported outcome measures (MA-PROMs) available for people with CVD and select the most suitable MA-PROM for people with CVD. Candidate medication adherence predictors were identified using existing literature and qualitative exploration of hospital pharmacists' views on and experiences with medication adherence. Psychometric testing was conducted on the selected MA-PROM from the COSMIN review in the Amharic language to demonstrate its accuracy and reliability for the study site. Structural validity using exploratory and confirmatory factor analysis, internal consistency, convergent validity, known-group validity (KGV) and clinical validity were confirmed before using it to measure medication adherence in this research project. The full dataset with candidate predictors for medication adherence and a validated measure of medication adherence was randomly split into development and validation cohorts. The development cohort was used to develop the multivariable predictive model that was named Medication Adherence Risk Assessment Tool (MA-RAT). The performance of MA-RAT was evaluated for discrimination (using the concordance index (C-index)) and calibration (using the HosmerLemeshow test (HLT) and slope on the calibration plot) that provide information for its internal validation.</p> <p><b>Results and discussion</b>: Of the 84 studies included in the COSMIN systematic review, 40 MA-PROMs were identified for people with CVD. Adherence to Refills and Medication Scale (ARMS) was selected as the most suitable MA-PROM for people with CVD as ARMS is comprehensive and has moderate to high-quality evidence for sufficient results on 4 psychometric properties. Hospital pharmacists were interviewed about barriers, enablers, perceived roles of pharmacists, and strategies to support medication adherence. Prioritizing patients at high risk for medication non-adherence was one of the strategies identified by hospital pharmacists for medication adherence support. A total of 23 candidate predictors related to sociodemographic, patient, therapy, medical condition, and healthcare factors were obtained from the existing literature and the qualitative study involving hospital pharmacists. The 9-item Amharic version ARMS (ARMS-9Am) was obtained from structural validity analysis and found to be a unidimensional scale with adequate internal consistency (α =0.74). ARMS-9Am has exhibited moderate convergent validity with pill count (ρ =-0.42), along with good KGV for blood pressure (BP), cholesterol level and heart failure symptom control groups. Based on BP control, ARMS-9Am (AUC = 0.81) had excellent discriminatory power that showed good clinical validity with 87.8% specificity and 70.4% sensitivity at a score of less than 10 for medication adherence. The independent candidate predictors for medication adherence measured using ARMS-9Am were polypharmacy, perceived stress, patientprovider relationship, worrying about side effects, comorbidity and age. These independent predictors formed the final predictive model (MA-RAT) with a point score ranging from 0 to 12. In the validation cohort, MA-RAT showed good calibration (HLT=3.68" p=0.88, slope=0.99" R2=0.96) and discrimination (AUC=0.75, p<0.001, 95%CI (0.68-0.81)).</p> <p><b>Conclusion</b>: This research project has identified ARMS as the most suitable MA-PROM for people with CVD, evaluated the psychometric properties of ARMS in the Amharic language, identified contextual medication adherence predictors, and then constructed a multivariable predictive model for medication adherence (MA-RAT). MA-RAT has 6 independent predictors (polypharmacy, perceived stress, patient-provider relationship, worrying about side effects, comorbidity and age) with a score range of 0 to 12. A MA-RAT score of ≤8 can be used to identify people with CVD at high risk for medication non-adherence. After external validation, MA-RAT could be used to assist pharmacists to provide a proactive medication adherence support.</p>