Reassurance‐Focused First‐Trimester Euploidy Screening With Machine Learning

Objective To develop and internally validate a machine‐learning (ML) model for first‐trimester screening that prioritizes fetal euploidy reassurance and to compare its performance with the conventional Combined Screening Test (CST). Methods This retrospective diagnostic‐accuracy study included 13,755 singleton pregnancies with known cytogenetic outcomes screened between 2019 and 2023. A calibrated extreme‐gradient‐boosting model was developed using a stratified 60–20–20 split into training, validation, and independent test sets. Predictors included routine maternal, biochemical, and ultrasound variables. Performance was compared with CST using accuracy, discrimination, and stratified analyses by maternal age and CST risk category. Results In the independent test set, ML achieved higher accuracy than CST (89.9% vs. 82.2%) by reducing false alarms among euploid pregnancies (10.2% vs. 18.0%), a 43% relative reduction. Discrimination was excellent for both methods (ML AUC 0.97; CST AUC 0.93), without a significant difference. Both methods missed two aneuploid pregnancies. Gains were greatest in pregnancies aged ≥ 35 years and in the CST high‐risk and intermediate‐risk categories. Conclusion A reassurance‐oriented ML approach preserved aneuploidy detection while substantially reducing false alarms. External validation is required before clinical implementation.