A Dolutegravir-Associated Hyperglycemia Computational Prediction Tool for People Living with HIV in Uganda

Dolutegravir-based antiretroviral therapy (ART) is the recommended treatment for persons living with HIV (PLWH). While incidence and prevalence rates are unclear among PLWH, clinical research has shown that the use of Dolutegravir (DTG) results in momentous hyperglycemia. Identification of patients at risk of DTG-associated hyperglycemia prior to switching to DTG regimens would lower morbidity and mortality in this group. A machine learning (ML) prediction tool was developed and evaluated for this purpose. ML methods were used to process and model a longitudinal cohort secondary dataset of 9077 treatment-experienced participants. DTG-associated hyperglycemia risk factors were used as model features. The data was split into training and testing datasets in a ratio of 2:1. A total of 6807 records were used to train eight models. Among others, the study found the XG-Boost model with the best metrics of; 0.87 probability of classifying positives, 0.67 a precision to positives, 0.86 area under the precision-recall curve, 0.76 F1 score, and 0.72 Cohen Kappa. ML techniques can be harnessed to build DTG-associated hyperglycemia prediction tools for screening PLWH before being switched to DTG and avoid unintended hyperglycemia.