Predicting Early Allograft Function After Normothermic Machine Perfusion

Background. Normothermic ex situ liver perfusion is increasingly used to assess donor livers, but there remains a paucity of evidence regarding criteria upon which to base a viability assessment or criteria predicting early allograft function. Methods. Perfusate variables from livers undergoing normothermic ex situ liver perfusion were analyzed to see which best predicted the Model for Early Allograft Function score. Results. One hundred fifty-four of 203 perfused livers were transplanted following our previously defined criteria. These comprised 84/123 donation after circulatory death livers and 70/80 donation after brain death livers. Multivariable analysis suggested that 2-h alanine transaminase, 2-h lactate, 11 to 29 mmol supplementary bicarbonate in the first 4 h, and peak bile pH were associated with early allograft function as defined by the Model for Early Allograft Function score. Nonanastomotic biliary strictures occurred in 11% of transplants, predominantly affected first- and second-order ducts, despite selection based on bile glucose and pH. Conclusions. This work confirms the importance of perfusate alanine transaminase and lactate at 2-h, as well as the amount of supplementary bicarbonate required to keep the perfusate pH > 7.2, in the assessment of livers undergoing perfusion. It cautions against the use of lactate as a sole indicator of viability and also suggests a role for cholangiocyte function markers in predicting early allograft function.

Linear regression models with MEAF scores as outcomes, regressed on the perfusion variables, were fitted after first normalising the outcome variables of interest. MEAF differed significantly from a normal distribution. The best fitting transformation for the MEAF score was a square root transformation. Univariable linear regression models were fitted with each of the perfusion variables in turn as the exposure variable. For perfusion variables that had skewed distributions and all positive values, the natural log-transform of the variable was also considered as a predictor and if this was more strongly associated with the viability score than the variable on the original scale then this association is reported instead. The perfusion variables that showed the strongest associations with the MEAF score in univariable models were then considered as candidate variables to build a multivariable model for the prediction of the MEAF score.
For the model selection process for the multivariable model, analysis was restricted to the patients who had complete data on all of the candidate variables to be considered for inclusion in the model (the complete case dataset). The distribution of the MEAF score was assessed for normality in the complete case dataset and it was found that a square-root transformation of the MEAF best fitted a Normal Distribution. Therefore, the square-root of the MEAF score was used as the outcome variable in the linear regression model, as it was in the exploratory analysis.
Univariable linear regression models with each of the candidate variables as covariates and the square-root of the MEAF score as the outcome variable were fitted in turn in the complete case dataset and the candidate variables were ranked in order of the strength of their association with the MEAF score. Each of the candidate variables were then added one at a time to a multivariable linear regression model, with the most strongly associated variables from univariable models included first, and each time a variable was added it was retained in the model if it significantly improved the model fit (at the p<0.1 level). Any variables with effects becoming non-significant on adding new variables to the multivariable model were removed. The final model was arrived at once there was no further improvement in model fit from addition of any of the other candidate variables.
For the 42 cases which were excluded from the multivariable analysis because of missing values from the dataset, those missing variables were as follows: ALT at 60 minutes n=9 ALT at 120 minutes n=15 Glucose rate of fall n=1 Lactate at 3 hours n=2 Lactate at 4 ours n=4 Bile volume at 2 hours n=3 Bile volume at 3 hours n=6 Bile volume at 4 hours n=1 Bile perfusate glucose peak difference n=1 There were no livers where there were missing data for bicarbonate administration, lactate at 1 and 2 hours, lowest bile glucose or peak bile pH. Biopsies were not done in any of these livers to aid in decision making.
✘: criterion not met; ✔: criterion met; o: no value recorded, either because the liver had not been weighed (for the lactate) or no bile was produced (for the bile parameters). Livers in the first row met all our previously published criteria; those in the second row met them all, even though only one of the two bile glucose conditions was met. Note weightadjusted lactate was not met if the liver was not weighed. The other livers not meeting the initial criteria, but where a large proportion of livers were transplanted, were perfusions requiring >30mls supplementary bicarbonate.    Table S7. Distributions of candidate machine perfusion variables and the MEAF score in the full dataset (n= 150) and for livers with complete data on all candidate variables as well as the MEAF (n = 108)