Validation and optimization of a blood transfusion prediction model for low transfusion rate adult cardiac surgery

Abstract

Introduction

Blood transfusion is associated with adverse events and should be prevented. Preoperative identification of patients at risk is necessary and makes subsequent preventive intervention possible. Many risk models have been developed of which the Transfusion Risk and Clinical Knowledge (TRACK) model includes criteria reflecting daily practice. The aim of this study is to validate and update the TRACK model in a low-transfusion-rate adult cardiac-surgery population.

Methods

External validation of the TRACK model was performed using a database of 4072 adult patients receiving cardiac surgery between 2015 and 2022 (original TRACK model). Subsequently, the original TRACK model coefficients were updated by cross-validation (uTRACK model). Preoperative antiplatelet therapy was added as an extra variable to the updated TRACK model (uTRACK + APT model).

Results

In our population, 26% of patients received red blood cell transfusions. The original TRACK model demonstrated good discrimination (AUC-ROC of 0.76; 95% CI 0.74 – 0.78) but inadequate calibration (p < .001). Updating TRACK coefficients resulted in improved discrimination (AUC-ROC of 0.78; 95% CI 0.75 – 0.81), calibration (p = .19), and reclassification (Categorical NRI: 0.21; 95% CI [0.17 – 0.24]; p < .001). Adding preoperative antiplatelet therapy did not significantly improve net reclassification improvement (Categorical NRI: 0.01; 95% CI -0.001 – 0.029; p = .40).

Conclusion

The original TRACK model overestimates blood transfusion risk in a low-transfusion-rate population. Risk classification significantly improved by updating the original TRACK coefficients. Using the uTRACK model provides more accurate identification of patients at risk of receiving red blood cell transfusions in a low transfusion rate population.

Trial Registry

ClinicalTrials.gov (https://clinicaltrials.gov), registration number: NCT05581238.