Background: Virological failure remains a major barrier to achieving sustained viral suppression among people living with Human Immunodeficiency Virus (HIV) in sub-Saharan Africa. Delayed identification of patients at risk is associated with poor clinical outcomes and continued transmission. This study aimed to develop and evaluate predictive models to identify patients at increased risk of virological failure during antiretroviral therapy (ART) in Cameroon.
Methods: A retrospective analysis was conducted using clinical patient records from HIV treatment centers in Cameroon between 2023 and 2024. Adult patients on ART for at least six months were included. Virological failure was defined as a viral load exceeding 1000 copies per milliliter. Multiple predictors, including demographic, clinical, and treatment variables, were used to train the predictive models. The models implemented where logistic regression (LR), Random Forest algorithms (RF), Support Vector Machine (SVM) and XG Boosting (XGB). They were evaluated using cross-validation to avoid overfitting. Model performance was assessed using sensitivity (senc), precision (pre), specificity (spec), F1 Score and area under the receiver operating characteristic curve (AUC).
Results: Of 1,968 patients included, 11.7% experienced virological failure. Amongst the models random, RF model demonstrated the highest predictive performance, achieving a high detection rate for virological failure (sens – 97.8%, pre – 100%, f1 score – 98.9%). Across models, feature importance analysis identified poor treatment adherence, shorter duration on ART and residential region, which emerged as the most influential predictors.
Conclusions: Predictive modeling using routinely collected clinical data can support early identification of HIV patients at risk of virological failure. Integrating such tools into routine HIV can enhance clinical decision-making, enable targeted interventions, and improve treatment outcomes in a resource-limited setting like ours. By prioritizing patients with elevated risk, these approaches can facilitate timely adherence support, closer virological monitoring, and optimized use of limited healthcare resources.