(115) Development of Machine Learning Ensemble Model for Predicting DSA Positivity and Immunological Risk Stratification after Kidney Transplantation

Aim: Donor-specific antibody (DSA) positivity significantly impacts graft outcomes after kidney transplantation. This study aimed to develop a machine learning ensemble model to predict DSA positivity and stratify immunological risk using integrated immunological, clinical, and biochemical parameters.

Methods: We retrospectively analyzed clinical and laboratory data from 45 kidney transplant recipients who underwent mycophenolic acid (MPA) monitoring between May 2023 and March 2025. A total of 382 MPA tests were performed, with 120 parameters recorded per test, resulting in 45,840 data points. Parameters included 23 medication types, 86 laboratory tests, and 11 clinical variables, covering HLA typing, PRA results, and primary renal disease (PRD). Five patients were identified as DSA-positive. Machine learning models (CatBoost, XGBoost, and multilayer perception [MLP]) were trained independently. An ensemble model using soft voting was constructed, assigning higher weights to XGBoost, followed by MLP and CatBoost. Model performance was evaluated using AUROC, accuracy, recall, precision, and F1-score. Feature importance analysis was conducted for CatBoost and XGBoost.

Results: The ensemble model achieved an AUROC of 0.728 and an accuracy of 0.843. Recall, precision, and F1-score were 0.692, 0.681, and 0.686, respectively. Individual model AUROCs were 0.669 for CatBoost, 0.716 for XGBoost, and 0.694 for MLP. CatBoost identified PRD, monocytes, total calcium, RBC count, age, MCV, TIT(min), PT (INR), and total cholesterol as key features. XGBoost highlighted PRA (%), hemoglobin, PRD, PT (%), total cholesterol, amylase, HLA MN, and eGFR-MDRD (IDMS). Immune-related parameters such as PRA (%), HLA MN, PT (INR), aPTT, and eGFR consistently ranked among the top predictors, emphasizing the relevance of immune status, coagulation, and renal function in DSA development.

Conclusion: This study developed a machine learning-based ensemble model that effectively predicted DSA positivity and stratified immunological risk in kidney transplant recipients. Integration of immunological, coagulation, and renal function parameters may facilitate early identification of high-risk patients and inform optimization of post-transplant immunosuppression strategies. Further validation in larger, multi-center cohorts is warranted.