Effect of Machine Learning on Risk Stratification for Antiretroviral Treatment Failure in People Living with HIV

Abstract

Objective: Despite the widespread use of antiretroviral therapy (ART), HIV virologic failure remains a significant global public health challenge. This study aims to develop and validate a nomogram-based scoring system to predict the incidence and determinants of virologic failure in people living with HIV (PLWH), facilitating timely interventions and reducing unnecessary transitions to second-line regimens. Methods: A total of 9879 patients with HIV/AIDS were included. The predictive model was developed using a training cohort (N = 5,189) and validated internally (N = 2,228) and externally (N = 2,462) with independent cohorts. Multivariable logistic regression, with variables selected through least absolute shrinkage and selection operator (LASSO) regression, was employed. The final model was presented as a nomogram and transformed into a user-friendly scoring system. Results: Key predictors in the scoring system included delayed ART initiation (6 points), poor adherence (7 points), ART discontinuation (6 points), side effects (9 points), CD4+ T cell count (10 points), and follow-up safety index (FSI) (9 points). With a cutoff of 15.5 points, the area under the curve (AUC) for the training and validation sets was 0.807, 0.784, and 0.745, respectively. The scoring system demonstrated robust diagnostic performance across cohorts. Conclusion: This novel model provides an accurate, well-calibrated tool for predicting virologic failure at the individual level, offering valuable clinical utility in optimizing HIV management.