Machine Learning Prediction of Short Cervix in Mid-Pregnancy Based on Multimodal Data from the First-Trimester...

Area of Science:

• Obstetrics and Gynecology
• Medical Artificial Intelligence
• Maternal-Fetal Medicine

Background:

• A short cervix in the second trimester is a significant risk factor for preterm birth.
• Current methods for predicting a short cervix lack reliability in the first trimester.
• There is a need for accurate and cost-effective early pregnancy prediction of mid-trimester short cervix.

Purpose of the Study:

• To develop and validate a machine learning model for predicting mid-trimester short cervix using first-trimester clinical data.
• To identify key clinical predictors of mid-trimester short cervix.

Main Methods:

• 1480 pregnant women with preterm birth risk factors were recruited.
• Cervical length was assessed at 20-24 weeks; short cervix defined as <25 mm.
• Seven machine learning models were trained, with XGBoost selected for its performance and analyzed using SHAP values.

Main Results:

• 25.4% of participants developed mid-trimester short cervix.
• The XGBoost model showed high predictive accuracy across training, test, and independent datasets.
• Key predictors included pre-pregnancy BMI, history of pregnancy loss, leukocyte count, and vaginal microbiome status.

Conclusions:

• The XGBoost model accurately predicts mid-trimester short cervix from first-trimester data.
• This provides a 6-week window for intervention before standard assessment.
• Early prediction can guide preventive measures to reduce spontaneous preterm birth risk.