Machine Learning Across the HIV Care Cascade: A Study Protocol on Clinical Prediction, Symptom-Based Risk Stratification, and Epidemiological Forecasting Using Open African and Global Data

This protocol describes a three-arm computational study examining machine learning across the HIV care cascade using exclusively free, open-access datasets. Arm A uses the ACTG175 clinical trial dataset to demonstrate rigorous, fully cross-validated outcome prediction, directly addressing a documented pattern in the published literature where reported AUCs for viral load suppression prediction range from 0.65 (larger, rigorously validated studies) to 0.999 (smaller, likely overfit studies). Arm B builds a standalone symptom-based clinical risk model using an independently collected Nigerian patient cohort. Arm C quantifies whether epidemiological forecasting models trained on global country-year HIV surveillance data perform differently for African versus non-African countries. All datasets require no registration or data-use agreement. This is a pre-results study protocol; a follow-up manuscript reporting experimental findings will be linked upon completion.