Using Machine Learning to Predict Adverse Events in Patients Undergoing Transcatheter Left Atrial Appendage...

Area of Science:

• Cardiovascular Medicine
• Medical Informatics
• Machine Learning in Healthcare

Background:

• Standard cardiovascular risk models exhibit limitations in predicting rare events.
• Machine learning (ML) offers potential to enhance predictive accuracy.

Purpose of the Study:

• To compare the predictive performance of standard and ML methods in the National Cardiovascular Data Registry Left Atrial Appendage Occlusion (LAAO) Registry.
• To evaluate prediction models for in-hospital major adverse events (MAEs) and individual events in LAAO patients.

Main Methods:

• Utilized logistic regression (LR), LASSO, and eXtreme Gradient Boosting (XGBoost) for prediction.
• Employed a 70% development and 30% validation cohort split.
• Assessed models using 16 and expanded 51 variables from the LAAO Registry.

Main Results:

• XGBoost demonstrated superior performance for predicting composite MAEs compared to LR and LASSO.
• With expanded variables, XGBoost (AUC 0.653) marginally outperformed LASSO (AUC 0.644) for MAEs.
• Predictive performance decreased for infrequent events across all methods; XGBoost showed incremental improvement for mortality prediction.

Conclusions:

• XGBoost enhanced the discrimination of composite MAEs and several individual events in a nationwide LAAO cohort.
• While XGBoost improved prediction of rare mortality events, this was not consistent for all rare outcomes.