Dynamic prediction of HIV-related incomplete immune reconstitution: A multicenter, large cohort study using advanced joint modeling

Abstract

Incomplete immune reconstitution (IIR) is a serious complication affecting 10 to 40% of people living with HIV (PLWH) despite effective antiretroviral therapy, leading to increased morbidity and mortality. Current risk prediction models rely on single–time point measurements and lack dynamic assessment capabilities. We developed a dynamic joint prediction system for IIR risk (DJPSIIR) using Bayesian joint modeling to analyze longitudinal data from 21,862 PLWH across 31 Chinese provinces (2003–2024). The system integrates continuous CD4+ T cell counts and CD4/CD8 ratios with clinical parameters to generate real-time risk predictions. DJPSIIR demonstrated strong discriminatory performance with area under receiver operating characteristic curves of 0.890 to 0.912 for 5- to 7-year predictions, consistently outperforming expert assessments and 19 machine learning algorithms across multiple validation cohorts. Our dynamic prediction system enables precise identification of high-risk individuals and could transform clinical decision-making by facilitating timely interventions to prevent IIR progression in HIV care.