D-4 | Advanced Predictive Analytics in Cardiac Care: Unveiling Key Determinants of Rehospitalization in Mechanical Circulatory Support Patients

Mechanical circulatory support (MCS) has markedly improved the management of advanced heart failure. However, rehospitalization (RH) remains a significant challenge in MCS patients. This study leverages machine learning to identify critical predictors of RH, aiding in risk stratification and treatment optimization. Utilizing the MedaMACS dataset, which includes demographic and clinical data on 171 MCS patients, we deployed four machine learning models: logistic regression, random forest, support vector machine, and multi-layer perceptron. The models were trained and evaluated using MedaMACS data. Feature importance scores were assigned to each variable in the random forest model, determining their predictive power for RH. Logistic regression and support vector machine models demonstrated the highest accuracy in RH prediction, with an Area Under the Curve (AUC) of 0.85 each, outperforming the random forest (AUC = 0.71) and multi-layer perceptron (AUC = 0.65) models. The random forest analysis identified key RH predictors, such as systemic blood pressure, ventricular assist device (VAD) related factors, and aortic insufficiency. Our findings underscore the efficacy of logistic regression and support vector machine models in predicting RH risk in MCS patients. The identification of crucial predictive factors like systemic blood pressure and VAD-specific issues provides valuable insights for clinicians. This study paves the way for more effective, personalized treatment strategies, potentially reducing RH rates in this patient population.