Predicting Successful Chronic Total Occlusion Crossing With Primary Antegrade Wiring Using Machine Learning
Athanasios Rempakos1, Michaella Alexandrou1, Deniz Mutlu1
1Minneapolis Heart Institute and Minneapolis Heart Institute Foundation, Abbott Northwestern Hospital, Minneapolis, Minnesota, USA.
JACC. Cardiovascular Interventions
|July 6, 2024
Summary
Predicting successful chronic total occlusion crossing with primary antegrade wiring (AW) is crucial. A new machine learning model accurately identifies factors influencing success, aiding interventional cardiologists.
Area of Science:
- Cardiovascular Interventions
- Medical Machine Learning
- Interventional Cardiology
Background:
- Limited data exists for predicting successful chronic total occlusion (CTO) crossing using primary antegrade wiring (AW).
- Accurate prediction is vital for optimizing percutaneous coronary intervention (PCI) strategies.
Purpose of the Study:
- To develop and validate a machine learning (ML) prognostic model for predicting successful CTO crossing with primary AW.
- To identify key predictors of successful primary AW in CTO interventions.
Main Methods:
- Utilized data from 12,136 primary AW cases from the PROGRESS CTO registry (2012-2023).
- Developed and compared five ML models, with extreme gradient boosting showing the best performance.
- Employed SHAP explainer for feature importance analysis and hyperparameter tuning.
Main Results:
- Primary AW success rate was 57.4%.
- The optimized extreme gradient boosting model achieved an AUC of 0.780 in the testing set.
- Occlusion length, stump characteristics, and collaterals were significant predictors; aorto-ostial location was least impactful.
Conclusions:
- A highly predictive ML model with 14 features was developed for successful primary AW in CTO PCI.
- The model can aid in decision-making for complex CTO interventions.
- A web-based prediction tool is available online.
