Calculators / Predictive Models
CranioSTRIP Live
Craniosynostosis Surgery Transfusion Requirement Inference Platform. Estimates the risk that a child needs a perioperative transfusion and, if so, how much. Results update as you type.
Model Card
- Intended use
- Research and education: estimating perioperative transfusion risk and volume for children under 36 months undergoing craniosynostosis surgery, to inform discussion and planning. Not a substitute for clinical judgment, crossmatch protocols, or intraoperative monitoring.
- Outcomes predicted
- Stage 1: any perioperative blood transfusion (probability). Stage 2: transfusion volume in mL/kg among children who are transfused, reported as a median with an interquartile range. Volume estimates are unreliable for children who do not end up needing transfusion.
- Development data
- American College of Surgeons National Surgical Quality Improvement Program Pediatric (ACS NSQIP-P) data. Stage 1 is a LightGBM gradient-boosted tree classifier with 26 preoperative inputs. Stage 2 is an NGBoost model with a log-normal output and 22 inputs, trained on transfused patients. Both were tuned and evaluated with 5-fold nested cross-validation.
- Validation
-
Stage 1: transfusion risk Nested CV mean (95% CI) AUC-ROC 0.77 (0.75–0.78) AUC-PR 0.74 (0.73–0.75) Brier score 0.20 (0.19–0.20) Calibration slope 1.05 (0.94–1.16) Sensitivity / specificity, 43% threshold 0.81 / 0.60 Sensitivity / specificity, 50% threshold 0.74 / 0.68 Stage 2: transfusion volume Nested CV mean (95% CI) Median absolute error 6.9 mL/kg (6.6–7.1) Mean absolute error 10.4 mL/kg (10.0–10.8) Interquartile range coverage (target 50%) 53% (52–53%) Calibration slope (median) 1.35 (1.03–1.66) - Limitations
-
- Developed on registry data from participating US hospitals; performance may differ at other centers and with different blood management protocols.
- Inputs are limited to preoperative variables recorded in ACS NSQIP-P. Operative details such as duration, use of tranexamic acid, and intraoperative blood loss are not included.
- Validated for ages 0 to 36 months, weights 1 to 25 kg, gestational age 24 to 40 weeks, hematocrit 15 to 60%, and platelets 20 to 1,000 ×10⁹/L.
- How to cite
- Deng J, Gharavi A, Heybati K, Haile D, Ahn ES, Mardini S, Gibreel W. Development and nested cross-validation of machine learning models for predicting transfusion requirements in pediatric craniosynostosis surgery. J Plast Reconstr Aesthet Surg. 2026;118:209–220. doi:10.1016/j.bjps.2026.05.033. PubMed 42214978.
The American College of Surgeons National Surgical Quality Improvement Program and the hospitals participating in the ACS NSQIP are the source of the data used herein; they have not verified and are not responsible for the statistical validity of the data analysis or the conclusions derived by the authors.
For research and education. Not an FDA-cleared device. Predictions describe populations similar to the development cohort and should not be the sole basis for any clinical decision. Calculations run in your browser; nothing you enter is sent anywhere.