Using machine learning for analyzing medication adherence behavior among multimorbidity patients

Abstract

This study explores how clustering or unsupervised machine learning can be used, for discovering hidden relationships in Morisky Medical Adherence Scale (MMAS-8) questionnaire behavior data of multimorbidity patients. The clustering results of this study can help clinicians identify the unique characteristics of each cluster in patients to provide targeted interventions. Thus the study objectives are two-fold i) cluster patient’s medication adherence behavior with multimorbidity, and ii) describe the characteristics of the patients in each cluster based on their demographic characteristics. This study is based on 201 outdoor patients having mean age of 43.65 years (range 18…87 years), median age of 45 years at three government hospitals located in Islamabad region. First the patients were asked to answer a 13-item questionnaire in Urdu concerning the factors that could have affected adherence to the recommended medical treatment. Subsequently, each patient answered the Urdu language version of the Morisky Medical Adherence Scale (MMAS-8) questionnaire. After data preprocessing, the response of each participant was scored to ascertain the adherence to the recommended medical treatment and finally all variables were statistically analyzed and all records clustered. The overall mean MMAS-8 score was 4.55. Two variables were found to be statistically considerably associated with low adherence i.e. i) number of dependent children (p = 0.005) and ii) age (p = 0.02), however, clustering revealed mostly unmarried patients obviously without Maternal/Child issues who, never or rarely forgot to take medication.