Unpredictable Blood Product Usage in Orthotropic Liver Transplantation: A Foregone Conclusion or A Problem Solved by Artificial Intelligence?

Abstract

Abstract Introduction/Objective Orthotropic liver transplantation (OLT) is a complex procedure requiring coordination between operative and transfusion medicine staff. Unanticipated blood product demands during these surgeries strain blood bank inventory, especially when units are needed for patients with significant alloantibodies. Currently, transfusion medicine services issue set numbers of crossmatched red blood cell (RBC) units for these procedures. At our institution, 10 units are issued for each OLT. This project categorized OLT patients into three groups by RBC unit requirements: <5 units, 5-9 units and >10 units. Furthermore, we identified pre-operative clinical and lab factors predicting placement into each category. From this data, we developed an artificial intelligence platform (CCBBAI-1). Methods/Case Report We collected data from 474 adults (>18 years old) that underwent OLT from Jan. 1, 2021, to Dec. 31, 2022. Demographic, anticoagulant, and lab testing data generated within 24 hours of admission were obtained. Data analysis was performed in R-studio. CCBBAI-1 was built using machine intelligence learning optimizer platform (MILO). Results (if a Case Study enter NA) Based on univariate regression analysis (p-value<0.05, 95% confidence interval (CI)), the following lab variables were predictive of >10 RBC units issued: high INR (OR 3.48 [CI 2.35-5.26]), admission heparin dosage (2.61 [1.56-4.31]), total bilirubin (1.06 [1.04-1.08]), conjugated bilirubin (1.08 [1.04-1.11]), PTT (1.04 [1.03-1.06]), BUN (1.03 [1.02-1.04]), and creatinine (1.55 [1.33-1.83]). Low albumin (1.50 [1.13-2.22]) and MCV (1.04 [1.02-1.07]) also projected a 10 RBC unit procedural requirement. The univariate regression analysis variables predictive in categorizing patients requiring 5-9 units vs < 5 units were as follows: decreased albumin (2.17 [1.42-3.33]), decreased serum calcium level (1.72 [1.13-2.63]) and elevated INR (2.14 [1.36-3.40]). The mean number of units issued for each patient group including those requiring >10, 5-9, and <5 units were 22, 7, and 2 units, respectively. CCBBAI-1 achieved a sensitivity of 87.0% and specificity of 70.0% in classifying patients requiring >10 RBC units. Patients needing <10 units were analyzed in a second module to determine if they needed 5-9 units, achieving a sensitivity of 64.7% and a specificity of 62.4%. Conclusion Our study divided OLT patients into three categories and identified predictive laboratory values characterizing each group, a step towards improving transfusion medicine efficacy in this context.