Construction of a preeclampsia prediction model based on machine learning

Objective To construct a preeclampsia predictive model based on CatBoost algorithm and logistic regression (LR) algorithm, in order to provide reference for the early prevention and treatment of people at high risk of preeclampsia.Methods A total of 1 325 pregnant women who were examined and delivered in Xuzhou Central Hospital from January 2012 to December 2021 were selected, including 461 preeclampsia women (a research group) and 864 normal pregnant women (a control group). Their general physical examination data, demographic characteristics, the results of blood routine test, urine routine test and biochemical indexes were collected for retrospective analysis. Through statistical analysis, the independent influencing factors for the development of preeclampsia were screened out. Then, a prediction model was constructed using the optimal parameters of CatBoost algorithm and LR algorithm by grid search, and the prediction effect of the model was evaluated.Results When C=100, penalty="l2", solver="liblinear", the LR model achieved the optimal effect, AUC=0.976 9, accuracy=0.944 7, precision=0.959 0, recall=0.873 1, and F1=0.914 1. When depth=5, iterations=500, l2_leaf_reg=1, learning_rate=0.1, and rsm=0.5, the CatBoost model achieved the optimal effect, AUC=0.983 0, accuracy=0.952 3, precision=0.967 5, recall=0.888 1, and F1=0.926 1.Conclusions These two risk prediction models have good performance in predictive performance, which can effectively predict the occurrence of preeclampsia, and provide potential application value for the early identification of preeclampsia in clinical practice.