Predictive Modeling of Massive Transfusion Requirements During Liver Transplantation and Its Potential to Reduce Utilization of Blood Bank Resources. (May 2017)

• Record Type: Journal Article
• Authors: Pustavoitau, Aliaksei; Lesley, Maggie; Ariyo, Promise; Latif, Asad; Villamayor, April J.; Frank, Steven M.; Rizkalla, Nicole; Merritt, William; Cameron, Andrew; Dagher, Nabil; Philosophe, Benjamin; Gurakar, Ahmet; Gottschalk, Allan
• Is Part Of: Anesthesia & analgesia. Volume 124:Number 5(2017)
• Publication Date: 2017-05
• Subjects: Anesthesiology -- Periodicals; Anesthesia; Anesthesiology; Analgesia; Analgesics; Anesthesiology -- Periodicals; 617.9605
• DOI: 10.1213/ANE.0000000000001994
• Languages: English
• ISSNs: 0003-2999

Abstract

Abstract : BACKGROUND: Patients undergoing liver transplantation frequently but inconsistently require massive blood transfusion. The ability to predict massive transfusion (MT) could reduce the impact on blood bank resources through customization of the blood order schedule. Current predictive models of MT for blood product utilization during liver transplantation are not generally applicable to individual institutions owing to variability in patient population, intraoperative management, and definitions of MT. Moreover, existing models may be limited by not incorporating cirrhosis stage or thromboelastography (TEG) parameters. METHODS: This retrospective cohort study included all patients who underwent deceased-donor liver transplantation at the Johns Hopkins Hospital between 2010 and 2014. We defined MT as intraoperative transfusion of > 10 units of packed red blood cells (pRBCs) and developed a multivariable predictive model of MT that incorporated cirrhosis stage and TEG parameters. The accuracy of the model was assessed with the goodness-of-fit test, receiver operating characteristic analysis, and bootstrap resampling. The distribution of correct patient classification was then determined as we varied the model threshold for classifying MT. Finally, the potential impact of these predictions on blood bank resources was examined. RESULTS: Two hundred three patients were included in the study. Sixty (29.6%) patients met the definition for MT and received a median (interquartile range) of 19.0 (14.0–27.0) pRBC units intraoperatively compared with 4.0 units (1.0–6.0) for those who did not satisfy the criterion for MT. The multivariable model for predicting MT included Model for End-stage Liver Disease score, whether simultaneous liver and kidney transplant was performed, cirrhosis stage, hemoglobin concentration, platelet concentration, and TEG R interval and angle. This model demonstrated good calibration (Hosmer-Lemeshow goodness-of-fit test P = .45) and good discrimination ( c statistic: 0.835; 95% confidence interval, 0.781–0.888). A probability cutoff threshold of 0.25 was found to misclassify only 4 of 100 patients as unlikely to experience MT, with the majority such misclassifications within 4 units of the working definition for MT. For this threshold, a preoperative blood ordering schedule that allocated 6 units of pRBCs for those unlikely to experience MT and 15 for those who were likely to experience MT would prevent unnecessary crossmatching of 338 units/100 transplants. CONCLUSIONS: When clinical and laboratory parameters are included, a model predicting intraoperative MT in patients undergoing liver transplantation is sufficiently accurate that its predictions could guide the blood order schedule for individual patients based on institutional data, thereby reducing the impact on blood bank resources. Ongoing evaluation of model accuracy and transfusion practices is required to ensure continuing performance of the predictive model.