Patient blood management for cardiac surgery - Validation of a blood transfusion prediction model for a low transfusion population

<b>INTRODUCTION / PURPOSE</b><br/>Red blood cell (RBC) transfusions after cardiac surgery are associated with increased mortality, morbidity, and costs. Timely identification of cardiac surgery patients at risk for transfusion is key. Patient blood management (PBM) strategies include<br/>managing preoperative anemia, minimizing blood loss, and tolerating perioperative anemia as part of an `Enhanced Recovery after Cardiac Surgery´ program. Some prediction models were developed, e.g. the Transfusion Risk and Clinical Knowledge Score (TRACK), yet have rarely been validated in modern cardiac surgery populations. Therefore, we aimed to validate and update the TRACK model for a low-transfusion-rate cardiac surgery population.<br/><b>MATERIALS AND METHOD</b><br/>External validation of TRACK was performed in 4072 consecutive adult patients receiving on-pump cardiac surgery from 2015 – 2022, and in the same population coefficients were updated. Preoperative antiplatelet therapy (APT) was added to the<br/>updated TRACK to reflect current clinical practice. TRACK prediction performance was assessed with ROC curves (Figure 1). Net reclassification improvement determined accuracy of each model. A nomogram was developed for daily clinical use with updated TRACK coefficients (Figure 2).<br/><b>FINDINGS</b><br/>External validation of TRACK was performed in 4072 consecutive adult patients receiving on-pump cardiac surgery from 2015 – 2022, and in the same population coefficients were updated. Preoperative antiplatelet therapy (APT) was added to the updated TRACK to reflect current clinical practice. TRACK prediction performance was assessed with ROC curves (Figure 1). Net reclassification improvement determined accuracy of each model. A nomogram was developed for daily clinical use with updated TRACK coefficients (Figure 2).<br/><b>DISCUSSION / CONCLUSION</b><br/>Validation of an RBC transfusion prediction model in a low-transfusion-rate population showed good discriminative value but highlights the importance of selecting the correct model for institutional demographics. Although transfusion<br/>prediction models may be part of PBM strategies, their effectiveness should be evaluated on reducing RBC transfusions.