Application of machine learning in predicting medication adherence of patients with cardiovascular diseases: a systematic review of the literature

Abstract

Background: Cardiovascular disease (CVD) is among the most common chronic diseases in the US. Adequate controlling CVD risk factors with medications can have a significant impact on patients’ long-term outcome. Early identification of patients with low adherence to medications using predictive models through machine learning (ML) may enhance outcomes of patients with CVD. The objective of the current review was to systematically identify and summarize published predictive models that used ML to assess medication adherence (MA) among patients with chronic CVD [heart failure (HF) and coronary artery disease (CAD)] or their main risk factors [hypertension (HTN) and hypercholesterolemia (HCL)]. Methods: A targeted review of English literature was undertaken in PubMed and Google Scholar from January 1, 2000 to August 9, 2021. All studies that used ML to predict MA among patients with chronic CVD or their main risk factors were eligible for this review. Risk of bias was assessed based on the studies’ sample size, data analysis methods, variables in each model, and validation methods used. Characteristics and outcomes of each study were summarized in tables. Results: A total of 12 studies met the eligibility criteria. Selected studies evaluated MA among patients with HF (n=2), CAD (n=3), HTN (n=5), and HCL (n=2). The most used ML algorithms used were random forest (RF) (n=5), support vector machine (SVM) (n=4), and neural network (NN) (n=4). The accuracy of the models ranged from 0.53 to 0.97. Discussion: Most studies used cross-validation to evaluate the internal validation of the model. However, none of the models were externally validated using a different dataset. Using ML to predict MA among patients with CVD and their risk factors is relatively new with only a few studies identified. Compared to conventional statistical methods, fewer restrictions for inclusion of variables in ML models, may enhance the model performance. More research is required to predict MA with higher accuracy and external validity.