HIV remains a significant health issue, especially in sub-Saharan Africa. There are 39 million people living with HIV (PLWH) globally. Treatment with ART improves patient outcomes by suppressing the HIV RNA viral load. However, not all patients treated with ART suppress the HIV RNA viral load. This research paper explores the potential predictors of VL suppression in ART-treated PLWH. We used retrospective data from the 4820 ART-treated participants enrolled through population-based surveys conducted in Zambia and Malawi. We applied several machine learning (ML) classifiers and used the top classifiers to identify the predictors of VL suppression. The age of participants ranged from 15 to 64 years, with a majority being females. The predictive performance of the various ML classifiers ranged from 64% to 92%. In our data from both countries, the logistic classifier was among the top classifiers and was as follows: Malawi (AUC = 0.9255) and Zambia (AUC = 0.8095). Thus, logistic regression was used to identify the predictors of viral suppression. Our findings indicated that besides ART treatment status, older age, higher CD4 T-cell count, and longer duration of ART were identified as significant predictors of viral suppression. Though not statistically significant, ART initiation 12 months or more before the survey, urban residence, and wealth index were also associated with VL suppression. Our findings indicate that HIV prevention programs in the region should integrate education on early ART initiation and adherence in PLWH.
Keywords: HIV; HIV-1 viral load; machine learning; sub-Saharan Africa.