Personalized Surgical Transfusion Risk Prediction Using Machine Learning to Guide Preoperative Type and Screen Orders

Area of Science:

• Anesthesiology
• Medical Informatics
• Health Services Research

Background:

• Accurate surgical transfusion risk estimation is crucial for blood bank resource management and anesthetic planning.
• Traditional methods rely solely on procedure-specific information, potentially leading to inefficient allocation of preoperative type and screen orders.

Purpose of the Study:

• To develop and evaluate a machine learning model for predicting the likelihood of red blood cell transfusion.
• To compare the performance of this model against a traditional, procedure-centric approach.

Main Methods:

• Utilized the American College of Surgeons National Surgical Quality Improvement Program Participant Use File for model training.
• Developed four machine learning models incorporating surgery- and patient-specific variables.
• Validated models internally and externally, comparing a gradient boosting machine against a baseline model.

Main Results:

• The gradient boosting machine model demonstrated superior performance over the baseline model.
• At 96% sensitivity, the best model recommended type and screens for 30-36% of patients, reducing unnecessary orders compared to the baseline (45-57%).
• Key predictors included procedure-specific transfusion rates and preoperative hematocrit.

Conclusions:

• A personalized transfusion risk prediction model using surgery- and patient-specific variables was successfully created.
• This model offers improved performance for guiding preoperative type and screen orders compared to traditional methods.