Assessing medication-related burden and medication adherence among older patients from Central Nepal: A machine learning approach

BackgroundNepal is experiencing a rapid demographic shift toward an aging population, with concurrent increase in morbidity and medication-related problems. Despite this, the multidimensional experience of medication-related burden (MRB) and refill adherence remain under-studied, particularly through the lens of socio-demographic, clinical and medication-related predictive features. This study aimed to assess MRB and medication adherence, and utilize machine learning (ML) architectures to identify complex factors influencing both. MethodsA cross-sectional study conducted among 390 ambulatory older patients (aged[≥]65 years) at Bharatpur Hospital, Nepal. MRB and medication adherence was assessed using Living with Medications Questionnaire (LMQ-3) and Adherence to Refills and Medication Scale (ARMS). Six ML architectures (Ordinary Least Square, LightGBM, Random Forest, XGBoost, SVM, and Penalized linear regression) were employed to predict ARMS and LMQ scores using various socio-demographic, clinical and medication-related predictive features. Model explainability was provided through SHAP (Shapley Additive exPlanations). All the analysis were performed using R. ResultsThe median LMQ-3 score was 110.0 (IQR 14.0), reflecting a moderate medication-related burden, while the median ARMS score of 21.0 (IQR 6.0) indicated moderate non-adherence. Random forest was the superior predictive model for both MRB and adherence. SHAP analysis revealed requiring assistance for medication and polypharmacy as the most significant drivers of both increased burden and poor adherence. Interaction analysis revealed that while polypharmacy typically worsens adherence, the risk is partially mitigated when patients receive physical or cognitive assistance. Additional, financial factors and employment status emerged as significant predictors. ConclusionOlder patients in Nepal face a significant medication-related burden and non-adherence, driven largely by regimen complexity and the need for support. The high predictive accuracy of ML models suggests that clinical interventions should prioritize simplified regimens and patient-centered counseling for those with high dependency. These findings provide a data-driven rational for policy-level medication optimization strategies in Nepals evolving healthcare system. Author SummaryAs ageing occurs, there is high chance of presence of the chronic conditions. For management of these conditions, older people are often prescribed with multiple medications and often are vulnerable to those medications whose risk outweighs benefits. As a result there is high chance of occurrence of adverse effects; these effects have caused substantial degradation of the health related quality of life. Although pharmacotherapy is the mainstay of the chronic disease management, older people often feel medication burden. Medication burden is practical experience arising from the practical and psychological challenge while managing medications. Research shows high burden causes the non-adherence, a significant problem among older adults, causing significant problem in pharmacotherapy. Hence, we used validated Questionnaire for assessing the medication burden and medication adherence among older ambulatory adults attending the central hospital of Nepal. We used machine learning approach for the high prediction of the predictors influencing the medication related burden and mediation adherence. Moderate burden was observed among older adults and moderate non-adherence was also observed. We found needing assistance for medication management and multiple medications were the strongest predictors for both Medication burden and non-adherence. Our Study provides new insights and area for the implementation of clinical intervention for the medication optimization.