Development of machine learning-based models to predict congenital heart disease: A matched case-control study

Area of Science:

• Medical Informatics
• Public Health
• Genetics

Background:

• Current congenital heart disease (CHD) prediction tools lack interpretability and convenience.
• Personalized CHD management strategies are hindered by inadequate prediction tools.

Purpose of the Study:

• Develop and validate a machine learning-based risk stratification model for CHD prediction.
• Improve the accuracy and convenience of CHD risk assessment.

Main Methods:

• Utilized data from 1,759 participants in a case-control study (2014-2016) in Northwest China.
• Employed Least Absolute Shrinkage and Selection Operator (LASSO) for predictor selection from 47 variables.
• Built and evaluated five machine learning algorithms, including eXtreme Gradient Boosting (XGB), using metrics like AUROC, F1 score, and Brier score.

Main Results:

• The XGB model achieved superior performance with an AUROC of 0.772 in the testing dataset and 0.738 in external validation.
• Key predictors identified were rural living, low wealth index, and short-term folic acid supplementation (<90 days).
• The developed risk score effectively stratified participants into low, moderate, and high-risk categories, showing significant risk variations.

Conclusions:

• Machine learning offers a feasible and effective approach for CHD prediction.
• The risk scores can identify pregnant women at high risk for fetal CHD.
• This tool provides valuable insights for primary prevention and CHD management strategies.