Abstract
Background Early detection of cephalic dystocia is challenging, and current clinical assessment tools are limited. Machine learning offers unique advantages, enabling the generation of predictive models using various types of clinical data. Our model aims to integrate objective ultrasound data with psychological and sociological characteristics and obstetric treatment data to predict the individual probability of cephalic dystocia in pregnant women.Methods We collected data from 302 pregnant women who underwent examinations and deliveries at Southern Medical University's Nanfang Hospital from January 2022 to December 2023. We utilized basic patient characteristics, foetal ultrasound parameters, maternal anthropometric data, maternal psychological measurements, and obstetric medical records to train and test the machine learning models. Our study analysed the effectiveness of three machine learning models: least absolute shrinkage and selection operator (LASSO) regression, decision tree, and random forest. The precision, accuracy, recall, and area under the receiver operating characteristic (ROC) Curve (AUC) were used to evaluate the performance of the models.Results Among the three machine learning models, the LASSO-based logistic regression model demonstrated the best predictive performance, with an AUC value of 0.833. We found that maternal ischial spine diameter, fetal biparietal diameter, fetal biophysical profile score, artificial rupture of membranes, labor analgesia, childbirth self-efficacy, and other variables were predictive factors for cephalic dystocia.Conclusions This study constructed and validated a prediction model for cephalic dystocia via three machine learning methods, which can help clinicians improve the probability of identifying pregnant women at risk for cephalic dystocia.