Reassurance-Focused First-Trimester Euploidy Screening With Machine Learning.

Area of Science:

• Prenatal diagnostics
• Medical artificial intelligence
• Genetics

Background:

• First-trimester screening is crucial for assessing fetal aneuploidy risk.
• The Combined Screening Test (CST) is a conventional method with limitations in false positive rates.
• Machine learning (ML) offers potential for improved accuracy in medical diagnostics.

Purpose of the Study:

• To develop and validate a machine-learning (ML) model for first-trimester screening.
• To enhance fetal euploidy reassurance by minimizing false positives.
• To compare the performance of the ML model against the conventional Combined Screening Test (CST).

Main Methods:

• A retrospective study of 13,755 singleton pregnancies with known cytogenetic outcomes.
• Development of a calibrated extreme-gradient-boosting ML model using maternal, biochemical, and ultrasound data.
• Comparison of ML model performance with CST using accuracy, discrimination (AUC), and stratified analyses.

Main Results:

• The ML model demonstrated higher accuracy (89.9% vs. 82.2%) than CST in an independent test set.
• ML significantly reduced false alarms in euploid pregnancies by 43% (10.2% vs. 18.0%).
• Both methods had excellent discrimination (ML AUC 0.97; CST AUC 0.93) and missed two aneuploid cases each, with greatest gains in older mothers and higher-risk CST categories.

Conclusions:

• An ML-based approach for first-trimester screening effectively reassures fetal euploidy while substantially reducing false alarms.
• The ML model shows promise for improving upon existing screening methods.
• External validation is recommended prior to widespread clinical adoption.