Early Warning of Structural Complications Related to Perimembranous Ventricular Septal Defect Using Unstructured Echocardiographic Text Mining and Machine Learning

The goal of this observational study is to develop and test a AI-based prediction model for children with perimembranous ventricular septal defect, also called PMVSD. PMVSD is a type of hole between the lower chambers of the heart. Some children with PMVSD may later develop heart structure problems, such as aortic valve prolapse or leakage, subaortic fibrous ridge, left ventricle-to-right atrium shunt, or right ventricular outflow tract obstruction.

This study will use past medical records and echocardiography reports from children who received care at six hospitals in China between 2004 and 2022. The main questions it aims to answer are:

Can information written in echocardiography reports help predict which children with PMVSD are more likely to develop heart structure problems? Can natural language processing and machine learning improve early risk prediction when used together with routine clinical information? Researchers will review existing, de-identified medical data. They will use natural language processing to turn written descriptions in echocardiography reports into data that a computer model can analyze. These descriptions may include details about the edge of the heart defect, the direction of blood flow, and the relationship between the defect and nearby heart valves.

Participants will not receive any study treatment or extra tests. The study will only use information already collected during routine medical care. The prediction models will be trained and tested using data from different hospitals to see how well they work across medical centers.