MedAdhereAI: An Interpretable Machine Learning Pipeline for Predicting Medication Non-Adherence in Chronic Disease Patients Using Real-World Refill Data

Medication non-adherence remains a significant challenge in managing chronic conditions like diabetes and hypertension, leading to increased morbidity, preventable hospitalizations, and over $300 billion in annual healthcare costs. This burden is particularly pronounced in resource-limited settings, where fragmented data and limited resources hinder early risk identification. This study introduces MedAdhereAI, an interpretable machine learning pipeline designed to predict medication non-adherence using real-world refill and claims data. The pipeline encompasses data exploration, temporal feature engineering, model development, and SHAP explainability. Logistic regression and random forest models were selected for their balance of predictive performance and interpretability, making them suitable for clinical deployment. Evaluated on a publicly available dataset of anonymized refill records for patients with diabetes and hypertension, the logistic regression model achieved an AUC of 0.82 and a Brier Score of 0.1749, while the random forest model achieved an AUC of 0.77. SHAP analysis identified total_visits, AGE, and refill gaps as key predictors. These results highlight the potential of MedAdhereAI as a decision-support tool for identifying patients at risk of medication non-adherence, facilitating targeted interventions and improved resource allocation in healthcare systems.