Utilizing machine learning to predict the risk factors of episiotomy in parturient women.

BACKGROUND

Episiotomy has specific indications that, if properly followed, can effectively prevent women from experiencing severe lacerations that may result in significant complications like anal incontinence. However, the risk factors related to episiotomy has been the center of much debate in the medical field in the past few years.

OBJECTIVE

The present study used a machine learning model to predict the factors that put women at the risk of having episiotomy using intrapartum data.

STUDY DESIGN

This was a retrospective cohort study design. Factors such as age, educational level, residency place, medical insurance, nationality, attendance at prenatal education courses, parity, gestational age, onset of labor, presence of a doula during labor, maternal health conditions like anemia, diabetes, preeclampsia, prolonged rupture of membrane, placenta abruption, presence of meconium in amniotic fluid, intrauterine growth retardation, intrauterine fetal death, maternal body mass index, and fetal distress were extracted from the electronic health record system of a tertiary-care medical center in Iran, from January 2022 to January 2023. The criteria for inclusion were vaginal delivery of a single pregnancy. Deliveries done through scheduled/emergency cesarean section or at the mother's request were excluded. The participants were divided into two groups: those who had vaginal deliveries with episiotomy and those who had vaginal deliveries without episiotomy. The significant variables, as determined by their -values, were selected as features for the eight machine-learning models. The evaluation of performance included area under the curve (AUC), accuracy, precision, recall, and F1-Score.

RESULTS

During the study period, out of 1775 vaginal deliveries, 629 (35.4%) required an episiotomy. Each model had an AUC value assigned to it: linear regression (0.85), deep learning (0.82), support vector machine (0.79), light gradient-boosting (0.79), logistic regression (0.78), XGBoost classification (0.77), random forest classification (0.76), decision tree classification (0.75), and permutation classification-knn (0.70). Linear regression had a better diagnostic performance among all the models with the area under the ROC curve (AUC): 0.85, accuracy: 0.80, precision: 0.74, recall: 0.86, and F_1 score: 0.79). Parity, labor onset, gestational age, body mass index, and doula support were the leading clinical factors related to episiotomy, according to their importance rankings.

CONCLUSIONS

Utilizing a clinical dataset and various machine learning models to assess the risk factors of episiotomy resulted in promising results. Further research, focusing on intrapartum clinical data and perspectives of the birth attendant, is necessary to enhance the accuracy of predictions.