Real-time Quality Assessment of Donor Kidneys During Normothermic Machine Perfusion Using Perfusate and Urinary...

Objective, real-time criteria for kidney graft quality during normothermic machine perfusion (NMP) are lacking, contributing to subjective acceptance decisions and organ discard. We evaluated real-time perfusate/urinary biomarkers and perfusion parameters during NMP and developed an exploratory predictive model for delayed graft function (DGF).

This predefined secondary analysis used the NMP arm of a randomized trial (n = 41). Primary nonfunction cases (n = 3) were excluded a priori. The remaining immediate graft function/DGF cohort (n = 38) was split 70/30 into training and test sets; all screening and feature selection were restricted to the training set. Variables without missing data were grouped according to their correlation structure and physiological interpretability, and the feature with the largest absolute effect size within each group was selected. An L2-penalized logistic regression model was evaluated using cross-validation and a held-out test set (n = 12).

DGF occurred in 20 of 41 (49%) and primary nonfunction in 3 of 41 (7%) cases. In the training set (immediate graft function n = 12; DGF n = 14), DGF was associated with higher venous Po₂ at 60 min and lower total urine production >0-120 min. Urinary phosphate at 120 min met the prespecified threshold for practical relevance and was retained as a complementary model feature. The resulting 3-feature regression model achieved a test set area under the curve of 0.81 and an accuracy of 0.83.

Lower urine output and higher venous Po₂ during NMP were associated with DGF. The regression model showed preliminary internal discrimination and warrants validation in larger independent cohorts.