Predicting medication wastage using machine learning based on patient beliefs.

Objectives: Medication wastage is a critical issue impacting the sustainability of subsidised healthcare systems in Southeast Asia due to financial and resource constraints. This study aimed to develop a machine learning (ML) model to predict medication wastage by analysing patient demographics, health conditions and beliefs about medicines, using Malaysia as a case study. Methods: A cross-sectional survey was conducted involving 734 patients across six public healthcare facilities in Malaysia. Data on demographics, medication history and beliefs about medicines were collected using validated questionnaires. Multiple ML regression models were evaluated to predict medication wastage, with performance assessed based on root mean squared error (RMSE). Results: The XGBoost model achieved the best performance with the lowest RMSE of 4.67, outperforming other models (RMSE range:4.68-5.10). It also performed best using only seven features selected by sequential backward elimination method using LR, making it practical for clinical implementation. Key predictors of medication wastage included beliefs about medicines, age, ethnicity, region and monthly income. Conclusion: This study is the first to apply ML to address medication wastage in a Southeast Asian context, filling a critical research gap. The proposed model provides a foundation for developing targeted interventions to reduce medication wastage and supports policymakers and healthcare providers in optimising the allocation of subsidised medications. The insights are broadly applicable to other countries with similar healthcare resource challenges.