Machine Learning Prediction for Spinal Deformity Surgery Blood Transfusion

Area of Science:

• Orthopedic Surgery
• Medical Informatics
• Transfusion Medicine

Background:

• Spinal deformity surgery (SDS) carries significant risks associated with intraoperative blood loss and transfusion.
• Blood transfusions can lead to adverse events, including transfusion reactions, infection transmission, and immunosuppression.
• Accurate prediction of intraoperative blood transfusion (IBT) needs is vital for patient safety and resource management.

Purpose of the Study:

• To develop and evaluate machine learning (ML) models for predicting IBT risk in patients undergoing SDS.
• To identify key factors influencing IBT requirements in spinal deformity surgery.
• To create a tool for clinicians to assess IBT risk.

Main Methods:

• A retrospective study involving 162 patients undergoing SDS across 11 centers in China.
• Utilized 39 candidate factors and employed Lasso regression for feature selection.
• Evaluated ten ML algorithms, including Random Forest (RF), with performance assessed via ROC, precision-recall, and calibration curves.
• Applied SHapley Additive exPlanations for model interpretability and developed a web calculator.

Main Results:

• The Random Forest (RF) model demonstrated superior predictive performance (AUC of ROC: 0.8716).
• Seven key predictors for IBT were identified: age, BMI, preoperative hematocrit, fibrinogen, prefunction, bone graft, and fusion levels.
• Level 4 vertebral fusion surgery was associated with the highest IBT risk (OR = 20.78).

Conclusions:

• Machine learning algorithms can effectively predict IBT risk in spinal deformity surgery.
• These predictive models aid in optimizing blood resource allocation and transfusion strategies.
• The developed web calculator provides a practical tool for clinical risk assessment.