An early screening model for preeclampsia: utilizing zero-cost maternal predictors exclusively

Area of Science:

• Obstetrics and Gynecology
• Medical Informatics
• Machine Learning in Healthcare

Background:

• Preeclampsia poses significant risks to maternal and fetal health.
• Existing screening methods often lack sufficient sensitivity or cost-effectiveness.
• There is a need for reliable, low-cost early screening models for preeclampsia.

Purpose of the Study:

• To develop and validate a reliable, low-cost early screening model for preeclampsia.
• To identify novel clinical predictors for preeclampsia risk assessment.
• To compare the performance of the developed model against existing approaches.

Main Methods:

• A retrospective cohort of 25,709 pregnancies was used for model development.
• A data augmentation technique (α-inverse weighted-GMM + RUS) was applied.
• Ten machine learning models were trained, with AdaBoost selected based on sensitivity at a 10% false positive rate.
• The optimal model utilized 16 clinical predictors, including previously unreported factors.

Main Results:

• The AdaBoost model achieved an Area Under the ROC Curve of 0.8008 and a sensitivity of 0.5190.
• The model demonstrated a >50% improvement in sensitivity over checklist-based approaches.
• It showed at least a 28% increase in performance compared to multivariable models using only maternal predictors.
• Validation in an independent cohort confirmed the model's effectiveness.

Conclusions:

• An effective, low-cost early screening model for preeclampsia was developed using machine learning and zero-cost clinical predictors.
• The model exhibits superior performance compared to existing methods, offering significant improvements in sensitivity.
• This approach holds potential to increase screening participation rates when combined with other high-performance methods.