Machine learning based-prediction of Hepatitis B Virus co-infection among HIV infected individuals in Uganda

Abstract

Background: Hepatitis B virus (HBV) and human immunodeficiency virus (HIV) co-infection was reported to be a major clinical challenge in sub-Saharan Africa, particularly in Uganda, where both infections were highly prevalent. Evidence indicated that co-infected individuals faced an elevated risk of rapid liver disease progression, morbidity, and mortality. Early detection of HBV among people living with HIV was noted to be limited due to resource constraints, absence of routine screening, and the asymptomatic nature of early-stage HBV. Objectives: The study aimed to develop and evaluate a machine learning-based model for predicting HBV co-infection among HIV-infected individuals in Uganda. Study design: A retrospective study design was applied. Methods: Data was obtained from the Joint Clinical Research Centre (JCRC) electronic medical records (EMR) system including socio-demographic, clinical, and laboratory variables. Multiple supervised machine learning models were trained and evaluated using standard performance metrics, with cross-validation applied to assess model robustness. Results: The feature importance and interpretability analysis showed that Age, ALB, BILT and CD4 and ART duration were significant predictors of HBV co-infection. The best-performing model achieved accuracy of 92%, sensitivity of 80%, and specificity of 100%. Conclusion: It was concluded that a machine learning-based approach could accurately predict HBV co-infection in HIV-infected individuals, offering a cost-effective and scalable screening aid in resource-limited settings. Way forward: The integration of such predictive models into routine HIV care was recommended to improve early HBV detection and guide timely interventions, with further validation in larger and more diverse cohort.