Utilizing Machine Learning to Predict Perioperative Blood Transfusion in Pediatric Craniosynostosis Patients Undergoing Cranial Vault Remodeling.

BACKGROUND

Cranial vault remodeling (CVR) for craniosynostosis is associated with high transfusion rates and related complications, yet factors contributing to transfusion risk remain incompletely defined. This study aimed to identify patient-level and procedure-level predictors of transfusion and assess their association with postoperative outcomes.

METHODS

This retrospective cohort study used the 2012 to 2023 ACS NSQIP-Pediatric database to identify patients aged 2 and under who underwent cranial vault remodeling for craniosynostosis using CPT and ICD codes. The authors examined demographic, comorbidity, preoperative laboratory, intraoperative, and postoperative data. Perioperative blood transfusion was predicted using various supervised machine learning algorithms: logistic regression, naive Bayes, K-nearest neighbors, decision trees, random forests, and extreme gradient boosting. Model performance was assessed in training and testing phases for discrimination, calibration, and clinical utility.

RESULTS

A total of 10,732 patients were identified, of which 5781 (53.9%) received a perioperative blood transfusion. The mean age of the blood transfusion group was 8.53 months compared with 5.71 months for the no transfusion group (P<0.001). Multivariable logistic regression revealed that longer anesthesia duration (aOR: 2.07, 95% CI: 1.76-2.44) was significantly associated with increased odds of transfusion. Conversely, factors associated with lower odds of transfusion included higher preoperative hematocrit (aOR: 0.94, 95% CI: 0.89-0.99), along with minimally invasive surgery (aOR: 0.34, 95% CI: 0.13-0.87) and combination approaches (aOR: 0.34, 95% CI: 0.16-0.70) compared with open procedures. The logistic regression model demonstrated the highest discriminative performance (AUC=0.787), with comparable results observed for the random forest model (AUC=0.777). Key predictors identified included anesthesia duration, age, weight, preoperative albumin, platelet count, serum creatinine, hematocrit, and white blood cell count.

CONCLUSION

Patients undergoing craniosynostosis repair are at increased risk for transfusion based on age, comorbidities, surgical approach, and ethnicity. These findings highlight the need for early intervention and strategies to reduce transfusion risk in vulnerable populations.