Cardiovascular risk assessment enhanced by automated machine learning in a multi-phase study

Cardiovascular diseases (CVDs) are the leading cause of death worldwide, and current predictors such as lipoprotein (a) [Lp(a)] and risk scores have limitations. Automated machine learning (AutoML) offers the potential to improve CVD risk prediction by processing large datasets and developing tailored models without the need for extensive data science expertise. Using clinical datasets from the LURIC (n = 3316) and UMC/M (n = 423) studies, we built AutoML models to predict Lp(a), specific CVDs and CVD-related mortality in three phases. Phase 1 identified key CVD determinants such as age, Lp(a), troponin T, BMI and cholesterol with good accuracy (AUC 0.6249 to 0.9101). Phase 2 validated models in the UMC/M dataset and showed robust performance (AUC 0.7224 to 0.8417), with SHAP analysis highlighting predictors like statin therapy, age and NTproBNP. Phase 3 focused on cardiovascular mortality prediction, achieving high AUC values (0.74 to 0.85) and showed data drift, highlighting the need for model adjustment.