Preeclampsia Prediction by Machine Learning in Twin Pregnancies

Area of Science:

• Maternal-fetal medicine
• Machine learning in healthcare
• Biomarker discovery

Background:

• Preeclampsia prediction models and aspirin prophylaxis are established for singleton pregnancies but lacking for twin gestations.
• Twin pregnancies have a higher risk of preeclampsia and adverse outcomes.
• Accurate prediction and effective prevention strategies are crucial for twin pregnancies.

Purpose of the Study:

• To develop and validate machine learning algorithms for predicting preeclampsia in twin pregnancies.
• To identify key maternal factors and biomarkers for preeclampsia prediction across all trimesters.
• To evaluate the efficacy of aspirin prophylaxis in preventing preeclampsia in twin pregnancies.

Main Methods:

• Prospective enrollment of 596 women with twin pregnancies.
• Development of machine learning models using maternal factors and biomarkers from first, second, and third trimesters.
• Evaluation of screening performance using receiver operating characteristic (ROC) curves and aspirin intervention based on risk stratification.

Main Results:

• An optimal multi-trimester machine learning model achieved an area under the ROC curve of 0.97 (91% detection rate at 10% false positive rate).
• Key predictors included maternal factors, mean arterial pressure, cell-free fetal DNA, placental growth factor, and soluble fms-like tyrosine kinase-1.
• Aspirin treatment, administered to 43.1% of participants, did not significantly reduce preeclampsia rates.

Conclusions:

• Machine learning models incorporating multi-trimester data demonstrate high efficacy in predicting preeclampsia in twin pregnancies.
• The effectiveness of current aspirin prophylaxis regimens in preventing preeclampsia in twin pregnancies requires further investigation.
• An accessible tool for predicting preeclampsia risk in twins is available.