Machine Learning Approach to Predict Emergency Cesarean Sections Among Nulliparous Women.

Introduction The obstetrical team's efforts are consistently focused on minimizing the number of cesarean sections, particularly in nulliparous women. One of the most crucial steps is to understand the risk factors that predispose the woman to a cesarean section. This study aimed to identify the predictors of emergency cesarean sections in nulliparous women using a machine learning approach. Methods A retrospective cohort study was carried out at a maternal tertiary center in Iran among nulliparous women with a single cephalic pregnancy, ≥37 weeks of gestation, and induced or spontaneous labor, who gave birth between January 2020 and December 2022. The exclusion criteria were maternal request for cesarean section or those who delivered via cesarean section before the onset of labor. The rate of emergency cesarean section and the performance of machine learning in predicting emergency cesarean section were the outcome measures. Twenty-three factors potentially linked to the method of childbirth were initially identified, and included age, educational level, place of residence, medical insurance, nationality, attending prenatal education course, gestational age, the onset of labor, having a doula during the labor process, analgesia during labor, history of infertility, history of abortion, maternal anemia, cardiovascular disease, diabetes, maternal obesity, preeclampsia, prolonged rupture of membrane, placenta abruption, meconium amniotic fluid, intrauterine growth retardation, newborn weight, and newborn sex. The input data were fed into seven machine learning models: linear regression, logistic regression, decision tree classification, random forest classification, XGBoost classification, permutation classification (KNN), and deep learning. Results During the study period, 1916 (71.8%) of the 2668 births were vaginal, while 752 (28.2%) were by cesarean section. Cesarean sections were more common in mothers of advanced age and with a higher level of education. Attending a prenatal education course was also linked to the method of childbirth. Induced labor was more common in women who had a cesarean section. Those who had a doula were more likely to give birth vaginally. Maternal diabetes, obesity, preeclampsia, thyroid disease, placental abruption, meconium amniotic fluid, and fetal macrosomia were all linked to the method of childbirth. The area under the curve (AUC) for each model turned out to be: linear regression (0.86), XGBoost classification (0.83), logistic regression (0.79), deep learning (0.78), permutation classification (K-Nearest Neighbors or KNN) (0.77), decision tree classification (0.76), and random forest classification (0.72). Linear regression had a better diagnostic performance than other models with the area under the ROC curve (AUROC): 0.86, accuracy: 0.82, precision: 0.79, recall: 0.85, and F1-Score: 0.79). The linear regression model showed that advanced maternal age, advanced maternal education, diabetes, preeclampsia, placenta abruption, hypothyroidism, meconium amniotic fluid, late-term pregnancy, doula support, and attending prenatal courses were predictors of emergency cesarean section in nulliparous women. Conclusions Utilizing a clinical database and various machine learning algorithms showed potential in predicting emergency cesarean section. Additional prospective research, including intrapartum clinical characteristics, is essential for improving the accuracy of prediction accuracy.