Preeclampsia Prediction by Machine Learning in Twin Pregnancies

Résumé fourni par la source

INTRODUCTION: While machine learning models successfully predict preeclampsia in singleton pregnancies and aspirin prophylaxis prevents preterm preeclampsia, no parallel models exist for twin pregnancies. This study developed machine learning algorithms to predict preeclampsia in twins using maternal factors and biomarkers from all three trimesters. METHODS: We prospectively enrolled 596 pregnant women with twin pregnancies at 11+0 to 13+6 weeks' gestation. Machine learning models assessed the efficacy of maternal factors and biomarkers for preeclampsia prediction across all trimesters. Screening performance was evaluated using area under the receiver operating characteristic (ROC) curves. Women with first-trimester risk >1/100 received aspirin treatment (150-160 mg/day) based on Fetal Medicine Foundation (FMF) twin-specific algorithms, while others received 80-100 mg/day or no treatment according to local guidelines. RESULTS: Sixty-seven women (11.2%) developed preeclampsia, including 40 (6.7%) with preterm preeclampsia. Key first-trimester markers included maternal factors, mean arterial pressure, cell-free fetal DNA, placental growth factor, and blood group B. Second- and third-trimester predictors comprised placental growth factor, soluble fms-like tyrosine kinase-1, and mean arterial pressure. The optimal machine learning model incorporating all three trimesters achieved an area under the ROC curve of 0.97 with 91% detection rate at 10% false-positive rate. CONCLUSION: Despite aspirin treatment in 257 (43.1%) women, logistic regression showed no significant reduction in preeclampsia rates. These findings suggest that while multi-trimester biomarkers effectively predict preeclampsia in twins, the effect of aspirin prophylaxis in twin pregnancies has yet to be proven. An app to predict this score is available at: twin-pe.math.biu.ac.il or by contact with the corresponding author.

Ce résumé expose les affirmations des auteurs. BNTIC ne l’interprète pas comme une validation indépendante des résultats.