Abstract 17180: Developing Risk Scores to Facilitate Pre-operative Prediction of Transfusion and Massive Blood Transfusion Associated With Adult Cardiac Surgery
Transfusion of red blood cells (RBC), while necessary when a patient is anaemic, has been shown to be associated with adverse outcomes. A risk score enabling better pre-operative prediction of transfusion, and in particular massive blood transfusion (MBT), would permit better targeting of blood conservation interventions, or enrich patient populations for clinical trials evaluating blood management interventions. The aim of this study was to produce risk scores for blood transfusion and for MBT that had good transferability between cardiac centres. Using cardiac and transfusion data collected from two UK cardiac centres, data from 24 hospitals contributing to the UK Cardiac Surgery National Transfusion Audit and one Italian cardiac dataset, we developed two multivariate logistic regression models, one for predicting blood transfusion, the other for MBT (4 or more units of RBC). The AnyTransfusion model was derived using data from all four datasets (N=33,960), while the MBT model used only UK data (N=21,343) as RBC units were not recorded in the Italian dataset. For each outcome, leave-one-centre-out cross validation was used to select the model with smallest deviance from candidate models. The final logistic models were transformed into a score using the log odds ratios. In the pooled dataset the AnyTransfusion model had an area under the ROC curve (AUC) of 0.776 [0.771-0.780], with AUCs in Bristol, Wolverhampton, Milan and the Audit datasets of 0.846 [0.838-0.853], 0.710 [0.701-0.719], 0.764 [0.750-0.779] and 0.795 [0.783-0.807] respectively. The AnyTransfusion Score achieved greater AUCs than existing scores: TRACK, TRUST, and the Papworth Bleeding Score. The MBT model had an AUC of 0.807 [0.799-0.815] across the UK data as whole, and 0.826 [0.815-0.838], 0.783 [0.762-0.804], and 0.777 [0.761-0.793] for the Bristol, Wolverhampton and Audit datasets respectively. In conclusion, the AnyTransfusion and MBT scores both performed well in predicting RBC transfusion and MBT respectively. These scores may be used to direct clinical management in cardiac surgery patients as well as to refine clinical trial design.
- © 2013 by American Heart Association, Inc.