Identifying patients for blood conservation strategies.

Background: Generally, only the type of operation is used to estimate the need for perioperative homologous blood transfusion. This study quantified the extent to which the estimation could be improved if, in addition, simple patient characteristics were taken into account.

Methods: Retrospective data on 24 509 consecutive adult surgical patients were used to derive and validate three models to predict perioperative homologous transfusion. The first model was a univariable model with type of operation as the only predictor. The second and third models were a full and a simplified multivariable logistic regression model. The performance of the multivariable models was tested in two validation sets: in similar patients who had operations in the same general hospital (internal validation) and in patients who had operations in a university hospital (external validation). The areas under the receiver-operator characteristic (ROC) curve were compared with that found in the derivation set.

Results: There were no important differences in characteristics between the derivation and validation sets. The ROC area of the model including surgery only was 0.92 (99 per cent confidence interval (c.i.) 0.91 to 0.94) and that of the full and simplified multivariable models 0.95 (99 per cent c.i. 0.94 to 0.96) and 0.94 (99 per cent c.i. 0.93 to 0.95) respectively. The latter two were significantly different from the first one. In the external validation set the ROC area of the simplified model was 0.84 (95 per cent c.i. 0.83 to 0.86). Patients who had a preoperative haemoglobin level lower than 13 g/dl and underwent major invasive surgery had the highest risk (43 per cent) of transfusion.

Conclusion: A simple algorithm using type of operation and haemoglobin concentration was effective in identifying patients likely to need perioperative homologous blood transfusion.