Construction and evaluation of machine learning-based predictive models for early-onset preeclampsia.

Area of Science:

• Obstetrics and Gynecology
• Medical Informatics
• Computational Biology

Background:

• Early-onset preeclampsia (EOPE) poses significant risks to maternal and fetal health.
• Identifying predictive factors for EOPE is crucial for timely intervention and improved outcomes.

Purpose of the Study:

• To identify key influencing factors of early-onset preeclampsia (EOPE).
• To develop and validate a machine learning-based prediction model for EOPE.

Main Methods:

• Utilized Python for data analysis on 1040 pregnant women, splitting data into training (80%) and testing (20%) sets.
• Applied logistic regression, XGBoost, random forest, support vector machine, and artificial neural network algorithms.
• Validated models using resampling, evaluating accuracy, sensitivity, specificity, F1 score, and AUC.

Main Results:

• Identified pre-pregnancy BMI, number of pregnancies, mean arterial pressure, smoking, alpha-fetoprotein, and conception method as significant EOPE predictors.
• The XGBoost model achieved the highest performance, with an AUC of 0.963 in the training set and 0.936 in the test set.
• The XGBoost model yielded an F1 score of 0.554 in the training set and 0.488 in the test set.

Conclusions:

• The XGBoost-based prediction model demonstrates strong predictive capabilities for EOPE.
• This model can serve as a valuable tool for assessing EOPE risk in pregnant women.