New Prediction Model May May Help Improve Blood Usage Practices

Researchers from Montefiore Einstein Medical Center (MMC) have developed and validated a clinical prediction model that may help hospitals improve blood utilization by identifying surgical patients most likely to require intraoperative packed red blood cell (pRBC) transfusions. The team published findings from the prognostic study in JAMA Network Open. 

Researchers developed the model, called Transfusion Forecast Utility for Surgical Events (TRANSFUSE), using data from more than 800,000 adult surgical patients across two large hospital systems. Through stepwise backward regression, they identified 24 independent predictors of intraoperative transfusion from an initial pool of 30 candidate variables. These include surgical complexity and urgency, American Society of Anesthesiologists (ASA) physical status, anemia severity, platelet count, liver function and hypoalbuminemia, among others. Certain procedure types, such as vascular, cardiac, orthopedic/trauma and transplant surgeries, also contributed to risk. Assigned weights range from 1 to 16, with the highest values linked to severe anemia and complex or emergent procedures. Total scores translate to transfusion probabilities ranging from less than 0.01% to more than 96%.

In the training cohort at MMC, TRANSFUSE achieved an area under the receiver operating characteristic curve (AUC) of 0.93. Internal validation showed a similar performance (AUC = 0.92), and results were consistent in an external validation cohort at Beth Israel Deaconess Medical Center (AUC = 0.90). Using a predefined score threshold of 30 to classify risk, the model correctly identified 86.8% of transfusion recipients in the training cohort, with a negative predictive value of 99.7%.