Three predictive models for transthoracic device closure of atrial septal defect: a comprehensive comparative analysis

Abstract

Background: Atrial Septal Defect (ASD) is a common congenital heart disease. Although transthoracic occlusion of ASD is increasingly seen as a less invasive alternative to traditional surgery, the choice of procedure often relies on the surgeon's experience and the insights gained from transesophageal echocardiography (TEE) images, lacking concrete, quantifiable standards. This study aims to guide clinical surgical decisions by developing and comparing predictive models. Methods: We analyzed 312 patients with secundum atrial septal defects who underwent either transthoracic device closure or open surgical repair. By measuring and analyzing preoperative TEE data and utilizing statistical and programming software, we developed three predictive models: logistic regression, random forest, and multiple linear regression. The original data were divided into a training dataset and a testing dataset, with each model trained on the training dataset and validated on the testing dataset. The predictive performance of each model was then evaluated on both adult and pediatric data. Results: Each model exhibited remarkable predictive power. The prediction accuracies in the training dataset for the logistic regression model, random forest model, and multiple linear regression model were 0.892, 0.888, and 0.868, respectively. In the testing dataset, the prediction accuracies were 0.9194, 0.871, and 0.9194, respectively. Conclusions: This study developed three predictive models to guide clinical surgical decisions for ASD closure surgery. Among them, the multivariate linear regression model demonstrated significant predictive accuracy, offering the potential to streamline the clinical decision-making process. This model does not require complex computations and is capable of providing quick, real-time predictions.