Construction of a prediction model of histologic chorioamnionitis based on machine learning

Abstract

Objective To develop a machine learning prediction model for histologic chorioamnionitis (HCA) in patients with preterm premature rupture of membranes (PPROM) after delivery. Methods A total of 512 pregnant women who were diagnosed with PPROM at 28~36+6 weeks in our hospital were enrolled in this study. After 31 clinical items consisting of general information, maternal history, laboratory data, and imaging parameters were collected for each subject, they were randomly divided into a training set (n=358) and a testing set (n=154) in a ratio of 7:3. Least absolute shrinkage and selection operator (LASSO) algorithm was used to identify independent predictors in the training set. Based on these factors (reproductive tract infection, ultrasound fetal weight, gestational age at PPROM, maximum body temperature before antibiotics, number of previous uterine operations, and interval between rupture of membranes and antibiotic use), 6 machine learning models, including support vector machine (SVM), logistic regression (LR), multilayer perceptron (MLP), random forest (RF), extreme gradient boosting (XGB) and decision tree (DT) were developed. Area under the curve (AUCs) value was used to compare the prediction efficiency of the built models. Then prediction efficiency and clinical applicability of optimal model were evaluated to calculate the importance of risk factors. Results Among 6 machine learning algorithms, the SVM model performed best with an accuracy of 0.862, sensitivity of 0.840, and specificity of 0.750 in the testing set. The predicted values derived from the model were highly consistent with the actual situations in discrimination of calibration curve and decision curve analysis. SHapley Additive exPlanations (SHAP) plot prioritized the predictive weights of 6 risk factors, and reproductive tract infection was of most significant importance. Conclusion An SVM prediction model of HCA occurrence in PPROM patients is successfully constructed. Its excellent predictive ability contributes to guiding clinical treatment, improving adverse pregnancy outcomes, and promoting maternal and infant health.

Keywords

• preterm premature rupture of membranes
• chorioamnionitis
• machine learning
• prediction model