#2273 A surrogate marker for graft-to-recipient allocation could be identifying patients at risk of postoperative complications using donor data as predictors

Abstract Background and Aims A key challenge in kidney transplantation is optimizing graft allocation to maximize graft survival, recipient survival, and dialysis-free survival time. Variables such as delayed graft failure (DGF) and serum creatinine are widely recognized as reliable prognostic factors, and most allocation algorithms use these values as surrogate markers, relying on donor, recipient, and transplant data. This study aimed to investigate whether other transplant-related outcomes, such as early post-transplant complications, could be predicted with similar accuracy to DGF and serum creatinine. Method We analyzed 831 kidney transplants performed between 1989 and 2020 across three transplant centers using machine learning (ML) models. The dataset included 14 donor variables, 7 transplant variables, and 60 recipient variables. Four problem types were defined based on independent variable combinations: donor-only (D), donor + transplant (DT), donor + recipient (DR), and donor + transplant + recipient (DTR). Of the patients, 355 (42.7%) experienced at least one complication during the first three months post-transplant, while 476 (57.3%) did not. Complications included biopsy proven acute tubular necrosis, vascular and urological complications, lymphocele, urinary tract infection, renal arterial stenosis, or atheromatous disease. Six ML models were used: Support Vector Machine (SVM), Decision Trees (DET), Random Forest (RF), Gradient Boost (GB), Extreme Gradient Boost (XGB), and Multilayer Perceptron (MLP). Performance was evaluated using accuracy and AUC-ROC metrics. Results The best AUC-ROC scores were: • D: 80% (Decision Trees). • DT: 78% (Gradient Boost). • DR: 83% (Random Forest). • DTR: 80% (Random Forest and Gradient Boost). Conclusion Although the highest predictive performance was achieved using donor and recipient data, donor-only data effectively predicted post-transplant complications. This finding underscores its potential use in assessing donor quality and improving graft allocation decisions.