Using Machine Learning to Examine Pre-Transplant Factors Influencing De novo HLA-Specific Antibody Development Post-Kidney Transplant

The development of de novo donor-specific antibodies (DSAs) against HLA is associated with premature graft failure in kidney transplantation. However, rates and factors influencing de novo DSA formation vary widely across the literature. We aimed to identify pre-transplant factors influencing the development of de novo HLA-specific antibodies following kidney transplantation using machine learning. Data from 460 kidney transplant recipients at a single centre between 2009-2014 were analysed. Pre-transplant variables were collected, and post-transplant sera were screened for HLA antibodies. Positive samples were investigated using Single Antigen Bead (SAB) testing. Machine learning models (Classification and Regression Trees, Random Forest, XGBoost, CatBoost) were trained on a training set of pre-transplant data to predict de novo DSA formation, with and without SMOTE oversampling. Model performance was evaluated on an independent testing set using F1 scores, and feature importance was assessed using SHAP. In the full cohort analysis, XGBoost models performed the best, with F1 scores of 0.54-0.59 without SMOTE and 0.72-0.79 with SMOTE. The strongest predictors were pre-transplant HLA antibodies, number of kidney transplants, cold ischemia time (CIT), recipient age and female gender. SHAP dependence plots showed that pre-existing HLA antibodies and past transplants increased the risk of de novo DSA development. In the unsensitised subgroup analysis, model performance was poor. Machine learning models can be used to identify pre-transplant risk factors for de novo HLA-specific antibody development in kidney transplantation. Monitoring and risk-stratifying patients based on these factors may help guide preventive immunological strategies and recipient selection to improve long-term allograft outcomes. Translational statementThis study identified pre-transplant risk factors for the development of de novo HLA-specific antibody in kidney transplantation. Monitoring and risk-stratifying patients based on these factors may help guide preventive immunological strategies and recipient selection to improve long-term allograft outcomes.