Interpretable SVM Model for Predicting CMV Infection in Seropositive Kidney Transplant Recipients: A Single-Center Retrospective Study.

Background: Cytomegalovirus (CMV) infection is a serious complication after kidney transplantation. Although most recipients are CMV-seropositive (R+), preventive strategies for this group remain controversial, whereas they are relatively well established for CMV-seronegative recipients (R-). Conventional serostatus-based classification alone is insufficient to accurately assess infection risk in R+ individuals. Therefore, we aimed to develop machine learning models that integrate clinical and immune variables to provide a precise risk prediction tool for CMV infection in R+ recipients. Methods: This study included patients from June 2023 to December 2024, and were randomly divided into training and validation cohorts in a 7:3 ratio. Feature selection was performed in the training cohort using the Boruta algorithm. Six machine learning models were applied to identify the best model for predicting CMV infection risk in R+ patients, and model interpretability was assessed using SHAP. Results: Of 162 R+ patients, 51.2% developed CMV DNAemia. Seven key predictors were identified, including T-cell subsets (CD8+, CD4+, CD4+CD27-), recipient age, cold ischemia time, donor type, and prevention strategy. Among these, CD4+ and CD8+ T-cell subset counts were the most influential predictors, with lower counts associated with a higher risk of CMV infection. The support vector machine (SVM) achieved the best discrimination in the validation cohort (AUC, 0.821; 95% CI, 0.692-0.932). Conclusion: The interpretable SVM model showed promising performance for identifying R+ recipients at high risk of CMV infection and potentially individualized prophylactic and monitoring strategies. External validation in prospective cohorts is warranted.