TCT-607 Machine Learning for Predicting Major Adverse Cardiac Events in Percutaneous Coronary Intervention of Coronary Artery Chronic Total Occlusion

Document Type

Conference Proceeding

Publication Date



Background: Predicting major adverse cardiac events (MACE) in chronic total occlusion (CTO) percutaneous coronary intervention (PCI) can assist with decision making and procedural planning. Methods: We analyzed 4,681 CTO PCIs from the PROGRESS-CTO (Prospective Global Registry for the Study of CTO intervention) registry performed between 2012 and 2022 at 42 centers. 8 machine learning (ML) methods were applied based on 32 parameters: multilayer perceptron (MLP) with decay model; support vector machine model; generalized additive model; MLP; extreme gradient boosting model; random forest model; Bayes GLM: Bayesian generalized linear model; and AvNNet: Neural Networks Using Model Averaging. The intervals were estimated using a bootstrap approach. The performance of the models was assessed using the receiver-operating characteristic curve. Results: The overall MACE rate was 1.9%. The median age of the patients was 65 years (57, 71 years). The data were divided into testing (n = 936) and training (n = 3,745) sets. The performance of the ML models on the testing set was as follows: MLP with decay model AUC 0.60; support vector machine model: 0.67; generalized additive model: 0.71; MLP: 0.65; extreme gradient boosting model: 0.44; random forest model: 0.60; Bayes GLM 0.72; AvNNet: 0.70. The AUC of the PROGRESS-CTO Complication logistic regression model was 0.675. The AUC of ML models and the Bayes-GLM method was found to have 50% of AUCs between 0.72 and 0.81, whereas the model built using linear logistic method had 50% of AUCs between 0.66 and 0.74 (Figure 1). [Formula presented] Conclusion: A machine learning-based model may improve the prediction of MACE in CTO PCI. Categories: CORONARY: Complex and Higher Risk Procedures for indicated Patients (CHIP)

First Page


This document is currently not available here.