Theoretical Framework for Implementing Precision Dosing Using Machine Learning in HIV-Positive Patients in Nigerian Clinical Settings

Antiretroviral therapy response in HIV-positive Nigerian patients is governed by complex interactions between pharmacokinetics, pharmacogenomics, co-morbidities, and adherence behaviours that traditional dosing algorithms inadequately capture. This study developed a theoretical framework for implementing machine learning-based precision dosing of antiretroviral therapy in Nigerian clinical settings. The framework development employed a multi-phase approach combining systematic review of machine learning applications in ART dosing optimisation, primary data collection from a cohort of 200 HIV-positive patients at ART clinics in Rivers and Enugu States, and theoretical modelling and validation. Data collected encompassed ART regimen, plasma drug concentrations (efavirenz, lopinavir), pharmacogenomic variants, co-infections, adherence measures, and viral load outcomes. Multiple machine learning algorithms were evaluated including gradient boosting machines, random forests, and neural network architectures for predictive performance. The gradient boosting model achieved the highest accuracy for predicting subtherapeutic drug exposure (AUC-ROC 0.89), substantially outperforming population pharmacokinetic Bayesian models (AUC-ROC 0.74). The original Precision ART Dosing Framework (PADF) integrates machine learning prediction, clinical decision support, pharmacist-physician dosing recommendation dialogue, and routine outcome monitoring into a structured clinical workflow. Framework components are validated through a clinical simulation study involving 25 pharmacists and physicians. The PADF constitutes an original contribution to precision medicine in HIV pharmacotherapy and provides a scalable implementation blueprint for resource-limited Nigerian ART clinics.

Keywords: precision dosing, machine learning, antiretroviral therapy, HIV, Nigerian clinical settings