UCLA

These studies aimed to build prediction models for cardiovascular-related outcomes among people living with HIV and compared model performance with previous standard models. The three outcomes were new subclinical atherosclerotic plaques, all-cause mortality, and non-HIV mortality. The two standard models for comparison were the pooled cohort equations (PCE) and the Veteran Aging Cohort Study (VACS) Index. We used the MACS/WIHS Combined Cohort Study to build each model in the whole data and separately as sex-specific models. During the model building process, we incorporated machine learning techniques, including the elastic net regularization and the boosting ensemble approach, for better prediction. The results from the three studies point in the same direction that the new model offers better predictive performance than the previous benchmarking models. However, only the 10-year sex-specific models and the 10-year MWCCS model for all-cause mortality satisfied all the performance measures and would be recommended to further pursue external validation.