Preoperative Predictive Modeling of Recurrent Graft Failure: Development and Validation of a 12-Month Prognostic Tool in Kidney Transplant Recipients.

BACKGROUND: Kidney transplantation is among the most effective treatments for end-stage renal disease. However, kidney transplant (KT) recipients remain at a relatively high risk of mortality, with underlying causes yet to be fully elucidated. This study aim to develop a machine learning model to predict 12-month graft survival in KT recipients and to analyze the relationship between key postoperative indicators and graft survival. METHODS: We retrospectively collected clinical data from 368 KT recipients who underwent allogeneic transplantation between April 2016 and September 2022. A novel feature selection method was employed to exclude variables with weak correlations to outcomes. Eight machine learning algorithms were compared to develop an optimal predictive model, with a primary focus on efficiency. RESULTS: Among the evaluated models, the light gradient boosting machine model demonstrated superior predictive performance, achieving an area under the receiver operating characteristic curve of 0.80 and 0.71 for internal and external validations, respectively. Notably, we redefined the threshold for assessing renal dysfunction in KT recipients, proposing an estimated glomerular filtration rate of 45 mL/min/1.73 m² as a more precise standard. CONCLUSION: The proposed machine learning model and the newly established estimated glomerular filtration rate threshold provide accurate assessments of graft dysfunction severity, offering robust support for clinical decision-making in kidney transplantation management.