Proactive recognition and early detection in communities through targeted HIV screening.

Area of Science:

• Public Health
• Epidemiology
• Machine Learning in Healthcare

Background:

• Human Immunodeficiency Virus (HIV) poses a significant global health challenge, especially in resource-limited settings.
• Early HIV detection is crucial for effective intervention but is often hindered by stigma and limited access to testing.
• Predictive models can enhance HIV screening by identifying individuals at higher risk.

Purpose of the Study:

• To develop and validate a machine learning model for predicting HIV risk using demographic and lifestyle variables.
• To improve early HIV detection in resource-limited settings.
• To create a publicly accessible tool for HIV risk assessment.

Main Methods:

• Analysis of data from 39,295 individuals in Shiraz, Iran.
• Development and validation of an Extreme Gradient Boosting (XGBoost) model.
• Stratified five-fold cross-validation and performance evaluation using accuracy, AUC, and Cohen's Kappa.

Main Results:

• Key predictors of HIV risk identified include drug injection, age, spouse's HIV status, occupation, and prison record.
• The XGBoost model (PREDICT-HIV) achieved high accuracy (0.89) and good discriminatory ability (AUC=0.84).
• Model performance was consistent across validation folds, indicating robustness.

Conclusions:

• Socio-demographic and behavioral factors are important for HIV risk prediction.
• The PREDICT-HIV model shows potential for practical implementation in resource-limited settings for early identification.
• Further validation in diverse populations and inclusion of socioeconomic variables can enhance global HIV prevention efforts.