Building Artificial Intelligence, Machine Learning, and Causal Models to Improve Cardiac Health

Abstract The prediction of myocardial infarction (MI) outcomes remains pivotal for advancing treatment strategies in cardiology. This study proposes to compare the variable selection efficacy and predictive power of LASSO logistic regression and XGBoost, incorporating a causal mediation approach to elaborate on the underlying mechanisms influencing MI risks. Leveraging the GUSTO-I dataset on MI patients, this research has employed LASSO logistic regression to facilitate variable selection, aiming to identify critical predictors with substantial impacts on MI outcomes. Alternatively, the XGBoost algorithm has been utilized to assess its approach to variable selection and prediction. Both methods have been evaluated based on their Area Under the Receiver Operating Characteristic Curve (AUC) to determine their predictive accuracy. Furthermore, causal mediation analysis has been integrated to explore the putative effects of key predictors, identified in previous studies. The analysis aimed to reveal which method offers superior predictive performance and how well each can be interpreted in a clinical context. This study can contribute valuable insights into the comparative advantages of using LASSO logistic regression versus XGBoost in the field of cardiovascular disease research. By integrating causal mediation analysis, we aim to extend beyond mere prediction to offer a deeper understanding of the causal relationships and mechanisms at play.