Application of machine learning in predicting postoperative arrhythmia following transcatheter closure of perimembranous ventricular septal defects.

Background: Arrhythmia is a frequent complication following transcatheter device closure of perimembranous ventricular septal defects (pmVSD). However, there is currently a lack of a convenient tool for predicting postoperative arrhythmia.

Aims: This research aimed to use machine learning algorithms to predict the risk of postoperative arrhythmia in pmVSD patients.

Material and methods: A retrospective study was conducted on 1384 children with pmVSD who underwent successful transcatheter device closure at a single-center hospital from March 2002 to March 2024. Subjects were assigned to a training set (n = 970) and a validation set (n = 414) in a 7:3 ratio. Four machine learning methods - SVM, LR, RF, and XGBoost - were used to develop models for predicting postoperative arrhythmia based on preoperative and intraoperative baseline information with clinical significance, as well as relevant content mentioned in previously published journals. The models' performance was evaluated using the area under the receiver operating characteristic curve (AUC), sensitivity, specificity, accuracy, negative predictive value, and positive predictive value. The optimal model was used to create a nomogram and further calibrated with calibration curves.

Results: In the prediction of postoperative arrhythmias, the LR model outperformed the XGBoost, SVM, and RF models, achieving an AUC of 0.863 (95% CI, 0.827-0.900). Consequently, we utilized the LR model to construct a nomogram based on 5 variables: weight, procedure time, defect diameter, pre-interventional arrhythmia, and the difference in the diameter between the occluder and defect exceeding 2 mm. The calibration curves illustrated a strong agreement between the actual and predicted outcomes.

Conclusions: The machine learning model accurately predicts postoperative arrhythmias, aiding in risk stratification of pmVSD patients and guiding clinical decisions.