Machine Learning to Predict Outcomes of Fetal Cardiac Disease: A Pilot Study

Area of Science:

• Cardiology
• Medical Informatics
• Genetics

Background:

• Prenatal diagnosis of congenital heart disease (CHD) presents significant challenges in predicting postnatal outcomes.
• Machine learning (ML) offers a promising approach to enhance prognostic accuracy and reduce clinical uncertainty.

Purpose of the Study:

• To develop and evaluate ML algorithms for predicting in utero/neonatal death, high-acuity neonatal care, and favorable outcomes in fetuses with CHD.
• To identify key predictors influencing these outcomes.

Main Methods:

• A pilot study trained ML models using the XgBoost algorithm with fivefold cross-validation on clinical data from 150 fetuses diagnosed with cardiac disease.
• Outcomes predicted included fetal/neonatal death, high-acuity neonatal care, and favorable outcomes.

Main Results:

• Prediction models achieved AUCs of 0.76 for death, 0.84 for high-acuity care, and 0.73 for favorable outcomes.
• Key predictors for death included non-cardiac abnormalities and CHD severity.
• Anti-Ro antibody and CHD severity predicted high-acuity care; absence of right heart disease, genetic abnormalities, and maternal medications predicted favorable outcomes.

Conclusions:

• ML models demonstrate good discrimination in predicting key prenatal and postnatal outcomes for fetuses with CHD.
• These models can aid clinicians in managing uncertainty associated with prenatal CHD diagnoses.
• Further research can refine ML applications for personalized CHD management.