Objective: This study aimed to develop a comprehensive machine learning (ML)-based prediction model for intrahepatic cholestasis of pregnancy (IHCP) by integrating multi-modal data including demographic characteristics, laboratory biochemical indicators, and ultrasonic echocardiographic parameters. The model was designed to stratify ICP severity and remain applicable in settings lacking total bile acid (TBA) testing, which addresses current diagnostic gaps and may support the reduction of adverse perinatal outcomes. Methods: A retrospective cohort of 750 pregnant women (525 in training, 225 in testing) between July 2020 and October 2023 from the Central Hospital of Enshi Tujia and Miao Autonomous Prefecture was recruited for the study. Seven ML algorithms (Logistic regression, Decision Tree, Random Forest RF, Extreme Gradient Boosting XGBoost, Regularized Support Vector Machine RSVM, Multilayer Perceptron MLP, and Elastic Net ENET). Results: The RF model exhibited superior performance, achieving ROC-AUC of 0.90 (training) and 0.86 (testing), with sensitivity and specificity both ≥ 0.93 in the testing cohort. Key predictors included pruritus, TBA, glycocholic acid, alkaline phosphatase, and ultrasonic indicators (ventricular wall mean thickness, myocardial echogenicity). Notably, the model retained efficacy without TBA, maintaining precision ≥ 0.75 across recall values of 0.6– 0.9. Conclusion: The multi-modal RF model effectively predicts IHCP, enables severity stratification, and enhances accessibility in resource-limited settings, providing valuable support for targeted clinical interventions and may support the reduction of adverse perinatal outcomes. Keywords: intrahepatic cholestasis of pregnancy, machine learning, random forest, prediction model, ultrasonic radiomics, biochemical indicators
Liang et al. (Sun,) studied this question.