Predicting Quality of Life in People Living with HIV: A Machine Learning Model Integrating Multidimensional Determinants.

Abstract

Objective: With survival steadily improving among people living with HIV(PLWH), quality of life (QoL) has emerged as the ultimate benchmark of therapeutic success. We therefore aimed to develop and validate machine learning models that predict QoL trend in PLWH, identifying key determinants to inform personalized interventions and optimize long-term well-being.

Methods: In this longitudinal observational study, PLWH were recruited from March 2024 to December 2024. Sociodemographic and clinical variables were collected, and the 31-item WHOQOL-HIV BREF was adopted as the QoL measure. The symptom experience was assessed using the Self-Report Symptom Scale (SRSS). All variables were incorporated into machine learning models to develop predictive algorithms.

Results: This study included 676 eligible participants with HIV in the cohort. The Gaussian Process (GP) model demonstrated the highest testing AUC of 0.811 and 0.815 in the training dataset. The GP model excels in metrics such as accuracy, AUC, recall, precision, F1 score, Kappa, MCC, Log Loss, and Brier score. In the decision curve analysis (DCA), the five machine learning models exhibited similar net benefits over a broad range of threshold probabilities (ranging from 0.2 to 0.8) compared to the Random Forest (RF) model. The GP and the MLP showed enhanced net benefits at intermediate to high threshold probabilities (30 ~ 60%). The SHAP technique identified the top four predictors of QoL, ranked by importance, with symptom burden being highlighted as the most critical predictor variable.

Conclusions: The machine-learning model, predominantly a GP model, demonstrated the better predictive performance among the six models evaluated, for detecting the QoL predictor in PLWH, indicating that symptom burden influences QoL level. Our findings highlight a non-linear relationship between ART duration and QoL, with diminished well-being during mid-treatment (6 ~ 10 years) linked to treatment fatigue and cumulative toxicities, emphasizing the necessity of dynamic psychosocial support and tailored interventions to sustain long-term QoL in HIV care.