An externally validated machine learning model to predict cesarean surgical site infections

Area of Science:

• Obstetrics and Gynecology
• Infectious Disease Epidemiology
• Health Informatics

Background:

• Cesarean surgical site infections (cSSI) pose a significant risk to maternal health.
• Accurate risk stratification is crucial for effective prevention and management strategies.
• Existing models may not fully leverage data available at hospital discharge.

Purpose of the Study:

• To develop and validate a machine learning (ML) model for predicting cSSI risk.
• To identify key predictors of cSSI using data available at hospital discharge.
• To enable early identification of high-risk patients for targeted interventions.

Main Methods:

• Retrospective cohort study utilizing data from two academic medical centers.
• Model development and internal validation on 2336 cesarean deliveries.
• External validation and generalization testing on 2936 cesarean deliveries.
• XGBoost classifier employed, optimized via random search and cross-validation.

Main Results:

• The XGBoost model achieved an AUC of 0.72 (internal) and 0.69 (external validation).
• Primary predictors identified: length of hospital stay, indication for cesarean, and estimated blood loss.
• The model stratified risk, with a 22.7% cSSI risk in the highest-risk group versus 0.8% in the lowest.

Conclusions:

• A validated ML model can predict cSSI risk using discharge information.
• Length of hospital stay, indication for cesarean, and blood loss are significant predictors.
• This tool can guide clinical surveillance and interventions for high-risk patients.