Immunosuppressive regimens and long-term kidney transplant outcomes: A dual survival modeling framework

Optimizing immunosuppressive therapy remains central to improving long-term outcomes after kidney transplantation. Both induction and maintenance therapies are widely used, yet their comparative effectiveness across diverse recipient populations requires further evaluation. This national retrospective cohort study analyzed 228,855 deceased-donor kidney transplant recipients using data from 2000-2024. Multivariable Cox proportional hazards (PH) models were used for clinical inference, and four machine learning (ML) survival models: random survival forest (RSF), support vector machine (SVM), penalized Cox regression (CoxNet), and extreme gradient boosting optimized with the Cox partial likelihood (XGBoost-Cox), were developed to assess predictive performance for death-censored graft failure and all-cause patient mortality. Model performance was evaluated using the concordance index (C-index) and time-dependent area under the curve (tdAUC). Maintenance regimens incorporating calcineurin inhibitors (CNI) and mycophenolate mofetil (MMF) were associated with lower hazards for both graft failure (CNI + MMF: hazard ratio [HR] 0.72, 95% confidence interval [CI] 0.70-0.74; CNI + MMF+steroids: HR 0.84, 95% CI 0.82-0.87) and patient mortality (CNI + MMF: HR 0.78, 95% CI 0.76-0.81; CNI + MMF+steroids: HR 0.90, 95% CI 0.88-0.93). Among induction therapies, antithymocyte globulin (ATG) was associated with lower hazards for both outcomes, whereas interleukin-2 receptor (IL-2R) antagonists and Alemtuzumab demonstrated neutral associations. Combined ATG + IL-2R therapy was associated with higher hazard of graft failure (HR 1.09). Recipient diabetes, dialysis dependence, older age, and higher Kidney Donor Profile Index were strong adverse predictors. Traditional Cox regression achieved robust discrimination (graft failure C-index: 0.685; patient mortality C-index: 0.704), comparable to ML survival models. These findings support the continued association of CNI and MMF maintenance regimens with favorable long-term transplant outcomes while demonstrating variation across induction strategies. The dual analytical framework integrating classical Cox modeling with ML survival methods, suggests that Cox models remain highly competitive for clinical inference, whereas ML approaches provide complementary predictive value to support individualized post-transplant risk stratification.