Abstract
BACKGROUND: Survival analysis is a critical tool in transplantation studies. The integration of machine learning techniques, particularly the Random Survival Forest (RSF) model, offers potential enhancements to predictive modeling and decision-making. This study aims to provide an introduction to the application of the RSF model in survival analysis in kidney transplantation alongside a practical guide to develop and evaluate predictive algorithms. METHODS: We employed a RSF model to analyze a simulated dataset of kidney transplant recipients. The data were split into training, validation, and test sets using split sample (70%-30%) and cross-validation (5-folds) techniques to evaluate model performance. Hyperparameter tuning strategies were employed to select the best model. The concordance index (C-index) and Integrated Brier Score (IBS) were used for internal validation. Additionally, time-dependent AUC, F1 score, accuracy, and precision were evaluated to provide a comprehensive assessment of the model's predictive performance. Finally, a Cox Proportional Hazards model was fitted to compare the results of the main metrics between both models. All analyses were supported by step-by-step code to ensure reproducibility. FINDINGS: The RSF model obtained a C-index of 0.774, an IBS of 0.090. The F1 score was of 0.945, accuracy was 89.67 and precision was 90.99%. The time-dependent ROC analysis produced an AUC of 0.709, indicating a moderate predictive performance. Lastly, the analysis shows that the three most important variables are donor age, BMI, and recipient age. CONCLUSIONS: This study demonstrates the robustness and potential of the RSF model in kidney transplant analysis, achieving strong validation metrics and highlighting its advantages in managing complex, censored data, while emphasizing the need for further exploration of hybrid models and clinical integration.