The usefulness of a classification and regression tree algorithm for detecting perioperative transfusion-related pulmonary complications

Abstract

BACKGROUNDTransfusion-related acute lung injury (TRALI) and transfusion-associated circulatory overload (TACO) are leading causes of transfusion-related mortality. An electronic medical record-based screening classification and regression tree (CART) algorithm was previously developed for predicting transfusion-related pulmonary complications. In the Republic of Korea, TRALI is not sufficiently recognized and an accurate TRALI incidence has not been reported. Therefore, we carried out this study to assess the incidence of TRALI and to determine whether the CART algorithm can be applied to our hospital data. STUDY DESIGN AND METHODSA retrospective analysis of all patients who received any type of transfusion during anesthesia was performed. After the patients were diagnosed by the relevant diagnostic criteria, they were reclassified by the CART algorithm. The validity of the algorithm was evaluated with sensitivity, specificity, likelihood ratios, and misclassification rate. RESULTSAmong 1948 patients who had received 11,269 units of transfusion, 26 TRALI and 20 TACO cases were identified. The incidence of TRALI among the transfused patients was 1.33% and per unit of transfused blood product was 0.23%. The sensitivity and specificity of the TRALI algorithm were estimated to be 73.1% (95% confidence interval [CI], 53.9%-86.3%) and 57.0% (95% CI, 52.5%-61.4%). For TACO, the sensitivity and specificity were 90.0% (95% CI, 69.9%-97.2%) and 56.0% (95% CI, 51.6%-60.4%), respectively. CONCLUSIONSLow specificity of the CART algorithm adopted previously indicated its limited diagnostic value in the Republic of Korea. A new algorithm is needed to facilitate the detection of transfusion-related complications.