A Machine Learning Approach to Predict Weight Change in ART-Experienced People Living With HIV.

Area of Science:

• Biomedical informatics
• HIV medicine
• Machine learning applications

Background:

• Antiretroviral therapy (ART) can cause weight changes in people with HIV.
• Predicting weight fluctuations is crucial for managing metabolic health in this population.

Purpose of the Study:

• To develop machine learning (ML) models for predicting percentage weight change over time in ART-experienced individuals with HIV.
• To assess the performance of ML models using various clinical and body composition variables.

Main Methods:

• An observational study involving 3,321 ART-experienced people with HIV.
• Development of 10 predictive ML models using an 80/20 training/test split.
• Utilized SHapley Additive exPlanations to interpret model predictions.

Main Results:

• The top-performing model, incorporating dual-energy x-ray absorptiometry (DXA) for body composition, exceeded 90% accuracy.
• This model predicted weight at the next visit with less than 5% error.
• Weight gain was defined as a >5% change at the subsequent visit.

Conclusions:

• Machine learning models incorporating body composition and metabolic variables demonstrate excellent predictive performance.
• Simpler models lacking detailed data are insufficient for precise prediction but can identify individuals not at risk for weight gain.