Premium
A data‐driven patient blood management strategy in liver transplantation
Author(s) -
Metcalf R. A.,
Pagano M. B.,
Hess J. R.,
Reyes J.,
Perkins J. D.,
Montenovo M. I.
Publication year - 2018
Publication title -
vox sanguinis
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 0.68
H-Index - 83
eISSN - 1423-0410
pISSN - 0042-9007
DOI - 10.1111/vox.12650
Subject(s) - medicine , liver transplantation , packed red blood cells , blood transfusion , blood loss , retrospective cohort study , surgery , blood volume , transplantation
Background and Objectives Blood utilization during liver transplant has decreased, but remains highly variable due to many complex surgical and physiologic factors. Previous models attempted to predict utilization using preoperative variables to stratify cases into two usage groups, usually using entire blood units for measurement. We sought to develop a practical predictive model using specific transfusion volumes (in ml) to develop a data‐driven patient blood management strategy. Materials and Methods This is a retrospective evaluation of primary liver transplants at a single institution from 2013 to 2015. Multivariable analysis of preoperative recipient and donor factors was used to develop a model predictive of intraoperative red‐blood‐cell ( pRBC ) use. Results Of 256 adult liver transplants, 207 patients had complete transfusion volume data for analysis. The median intraoperative allogeneic pRBC transfusion volume was 1250 ml, and the average was 1563 ± 1543 ml. Preoperative haemoglobin, spontaneous bacterial peritonitis, preoperative haemodialysis and preoperative international normalized ratio together yielded the strongest model predicting pRBC usage. When it predicted <1250 ml of pRBC s, all cases with 0 ml transfused were captured and only 8·6% of the time >1250 ml were used. This prediction had a sensitivity of 0·91 and a specificity of 0·89. If predicted usage was >2000 ml, 75% of the time blood loss exceeded 2000 ml. Conclusion Patients likely to require low or high pRBC transfusion volumes were identified with excellent accuracy using this predictive model at our institution. This model may help predict bleeding risk for each patient and facilitate optimized blood ordering.