Survival analysis using machine learning in transplantation: a practical introduction

Area of Science:

• Biostatistics
• Machine Learning in Medicine
• Transplantation Science

Background:

• Survival analysis is crucial in transplantation research.
• Machine learning, specifically Random Survival Forest (RSF), can improve predictive modeling.
• This study introduces RSF for kidney transplant survival analysis.

Purpose of the Study:

• To introduce the application of the RSF model in kidney transplant survival analysis.
• To provide a practical guide for developing and evaluating predictive algorithms.
• To assess the predictive performance of RSF models in this context.

Main Methods:

• A RSF model was applied to a simulated kidney transplant recipient dataset.
• Data were split into training, validation, and test sets using 70%-30% split and 5-fold cross-validation.
• Model performance was evaluated using concordance index (C-index), Integrated Brier Score (IBS), time-dependent AUC, F1 score, accuracy, and precision.

Main Results:

• The RSF model achieved a C-index of 0.774 and an IBS of 0.090.
• High scores were observed for F1 (0.945), accuracy (89.67%), and precision (90.99%).
• Important predictive variables identified were donor age, recipient age, and BMI.

Conclusions:

• The RSF model demonstrates robustness and potential for kidney transplant survival analysis.
• RSF offers advantages in handling complex, censored data in transplantation.
• Further research into hybrid models and clinical integration of RSF is warranted.