Biomathematical Analysis of Machine Learning Models for Predicting TLD (Tenofovir, Lamivudine, and Dolutegravir) Treatment Response in Digital Health Systems

Abstract

The application of machine learning in digital health systems has created new opportunities for improving treatment monitoring among individuals receiving antiretroviral therapy. This study conducted a biomathematical analysis of machine learning models designed to predict treatment response to the Tenofovir–Lamivudine–Dolutegravir (TLD) regimen among people living with human immunodeficiency virus (HIV) in a selected province in the Philippines. A retrospective dataset obtained from a provincial digital health information system was analyzed, consisting of anonymized demographic, clinical, and laboratory data from patients undergoing TLD therapy. Predictive models including logistic regression, random forest, and gradient boosting were developed and evaluated using biomathematical modeling techniques. Model performance was assessed using accuracy, sensitivity, specificity, and area under the receiver operating characteristic curve (AUC). Results indicated that machine learning algorithms effectively predicted virologic suppression outcomes. Among the models tested, the gradient boosting algorithm achieved the highest predictive performance. The biomathematical analysis revealed nonlinear interactions among baseline viral load, adherence indicators, immune status, and treatment duration. Integrating predictive analytics into digital health platforms may enable earlier identification of patients at risk for treatment failure and support data-driven clinical decision-making. These findings highlight the potential of computational methods in strengthening HIV treatment monitoring systems in resource-limited healthcare settings.