Prediction of Emergency Cesarean Delivery in Chile using Machine Learning

Abstract

Abstract Background: Emergency cesarean section (EmCS) is associated with a higher risk of intraoperative and postoperative maternal complications for both the mother and the offspring. Identifying which women who deliver by EmCS without indication for elective C-section should be a concern for health systems. Objective: To examine predictors related to EmCS in women with a medium-low socioeconomic status from the southeast area of ​​Santiago de Chile. Methods: This study involves a secondary analysis of all single birth records at Dr. Sótero del Río Hospital in the southeast public health district of Santiago, Chile, from 2002 to 2018 (n = 83,936). In total, fifty-nine potential predictors of EmCS were studied, 28 variables related to the pregnancy period, and the other 31 variables were related to the delivery period. Fivemachine learning (ML) algorithms were applied: Logistic regression, Random forest, AdaBoost, XGBoost, and Optimal classification tree. Results: The prevalence of EmCS was 18.6%, with an increase of 48.8% in the study period. Women's profiles were identified using eight factors that predicted EmCS (parity, previous cesarean section, labor already initiated, maternal age, gestational age, maternal height, pregestational body mass index, and the appearance of amniotic fluid). The Optimal classification tree was the algorithm with the highest sensitivity (0.74). The highest probability of EmCS (46%) occurred in multiparous women with one previous cesarean section. Conclusions: Most of the EmCS predictors are easily identifiable before delivery (age, parity, previous cesarean section, and maternal anthropometry). ML techniques are useful tools for predicting the risk of EmCS, potentially guiding the clinical decisions of health professionals.