Interpretable machine learning for predicting placental abruption in early-onset preeclampsia: model development and...

Area of Science:

• Obstetrics and Gynecology
• Machine Learning in Healthcare
• Perinatal Medicine

Background:

• Early-onset preeclampsia (EOPE) is a severe form of preeclampsia associated with significant maternal and fetal risks, including placental abruption.
• Placental abruption in EOPE cases can lead to severe complications, necessitating improved predictive strategies.

Purpose of the Study:

• To develop and validate an interpretable machine learning (IML) model for predicting placental abruption in patients diagnosed with EOPE.
• To identify key clinical predictors for placental abruption in the EOPE cohort.

Main Methods:

• A retrospective analysis of 580 EOPE patients was conducted, with data randomly split into training (70%) and validation (30%) sets.
• Feature selection using LASSO regression and the Boruta algorithm identified significant predictors.
• Six machine learning algorithms were trained and evaluated using AUC, F1 score, calibration curves, and decision curve analysis (DCA). SHapley Additive exPlanations (SHAP) were used for interpretability.

Main Results:

• Eight optimal predictors were identified: urinary protein, placental growth factor (PlGF), diastolic blood pressure (DBP), age, fibrinogen (FIB), prepregnancy BMI, disease severity, and smoking during pregnancy.
• The Random Forest (RF) model demonstrated superior performance with a validation AUC of 0.894.
• SHAP analysis highlighted urinary protein, PlGF, FIB, and DBP as dominant predictors, with specific levels correlating to increased abruption risk.

Conclusions:

• The developed SHAP-based RF model offers high predictive accuracy and interpretability for placental abruption in EOPE.
• This interpretable, data-driven approach facilitates individualized risk assessment and may enhance early detection and personalized management strategies in clinical settings.