Predictive modeling of emergency cesarean delivery

Area of Science:

• Obstetrics and Gynecology
• Medical Informatics
• Clinical Decision Support

Background:

• Emergency cesarean sections (ECSs) are critical surgical procedures with significant implications for maternal and infant outcomes.
• Accurate prediction of ECS is essential for optimizing resource allocation and improving patient care.
• Existing methods for identifying risk factors (RFs) for ECS have limitations in discriminatory accuracy (DA).

Purpose of the Study:

• To enhance the discriminatory accuracy (DA) of predicting emergency cesarean sections (ECSs).
• To identify key risk factors (RFs) and their interactions contributing to ECS indications.
• To evaluate the performance of different predictive models, including logistic regression and random forest models (RFMs).

Main Methods:

• Prospective data collection from 6,157 births across four public hospitals in Spain.
• Utilized likelihood ratios, logistic regression, classification trees (CTREEs), and random forest models (RFMs) to analyze risk factors.
• Assessed model discriminatory accuracy (DA) using areas under the receiver-operating-characteristic curves (AUCs).

Main Results:

• Individual risk factors (RFs) demonstrated low to moderate predictive value for ECS.
• Logistic regression models achieved a DA ranging from 0.74 to 0.81.
• The random forest model (RFM) achieved a significantly higher DA (AUC = 0.94), indicating superior predictive performance.

Conclusions:

• Fetal, maternal, and contextual risk factors alone are insufficient for accurate ECS prediction.
• The combination and interaction of risk factors, particularly within specific hospital contexts, are crucial for improving ECS prediction.
• Advanced modeling techniques like RFMs offer a promising approach to enhance the appropriateness of ECS indications.