Enhancing post-kidney transplant prognostication: an interpretable machine learning approach for longitudinal outcome...

Area of Science:

• Nephrology
• Transplantation Medicine
• Biomedical Data Science

Background:

• Kidney transplantation is a life-saving treatment for end-stage renal disease.
• Long-term graft survival and patient mortality remain significant challenges post-transplant.
• Existing prediction models often lack dynamic updating capabilities, limiting their long-term utility.

Purpose of the Study:

• To develop and validate a two-stage machine learning (ML) framework for dynamic, annual risk prediction of graft loss and death in kidney transplant recipients.
• To assess the performance improvement gained by incorporating longitudinal data compared to baseline-only models.
• To identify the most effective ML model for accurate and interpretable risk stratification.

Main Methods:

• Utilized a multi-center cohort from the Swiss Transplant Cohort Study (STCS) with 13 years of follow-up data.
• Developed a two-stage ML framework for dynamic, annual risk prediction.
• Trained and evaluated five ML models, including LightGBM, comparing longitudinal versus baseline data inputs.

Main Results:

• Incorporating longitudinal data significantly enhanced predictive performance over baseline-only models.
• The LightGBM model demonstrated superior performance, achieving an AUROC of 0.896 for graft loss and 0.797 for death.
• Dynamic ML models provide improved accuracy in predicting patient and graft outcomes.

Conclusions:

• Dynamic, interpretable ML models offer a practical and scalable approach to personalized risk stratification in kidney transplantation.
• These models can aid in guiding tailored follow-up strategies and timely interventions for transplant recipients.
• Improved risk prediction can lead to better long-term outcomes for kidney transplant patients.