Author + information
- Takahiro Tsushima,
- Sadeer Al-Kindi,
- Fahd Nadeem,
- Joshua Clevenger,
- Guilherme Attizzani,
- Yakov Elgudin,
- Alan Markowitz,
- Marco Costa,
- Daniel I. Simon,
- Mauricio Arruda,
- Judith Mackall and
- Sergio Thal
Whereas several preoperative characteristics have been associated with the risk of Cardiac Implantable Electronic Device (CIED) implantation after transcatheter aortic valve replacement (TAVR), an accurate risk prediction is not established yet.
This is a single center, retrospective study of consecutive patients who underwent TAVR from March 2011 to October 2018. Using 34 pre-TAVR clinical variables, we evaluated the predictive role of various machine learning techniques (J48 Tree, Random Forest, Naïve Bayes, Multilevel Perceptron, K nearest Neighbor, Locally-Weighted Learning, Instance Based Learner, Sequential Minimal Optimization, and Logistic Regression). Model performance was assessed using area under the curve (AUC) for receiver operating characteristics (ROC) and precision recall curve (PRC), using 10-fold cross-validation. Analysis was performed using WEKA Explorer version 3.8.3.
A total of 888 patients were included and 184 cases required CIED. Among the examined models, logistic regression (AUC=0.737) and Naïve Bayes (AUC=0.721) models had the best AUC result under ROC (Figure). Logistic regression (AUC=0.801), Naïve Bayes (AUC=0.795), random forest tree (AUC=0.788), and locally weighted learning models (AUC=0.798) had the higher AUC under the PRC (Figure).
Machine learning techniques can predict post-TAVR CIED implantation with good performance. Logistic regression and Naïve Bayes appear to have the best model performance.
Posters Hall_Hall A
Saturday, March 28, 2020, 3:45 p.m.-4:30 p.m.
Session Title: Arrhythmias and Clinical EP: Devices 3
Abstract Category: 05. Arrhythmias and Clinical EP: Devices
Presentation Number: 1217-245
- 2020 American College of Cardiology Foundation