Deceased-Donor Kidney Transplant Outcome Prediction Using Artificial Intelligence to Aid Decision-Making in Kidney...

Area of Science:

• Nephrology
• Transplantation immunology
• Artificial intelligence in medicine

Background:

• Optimizing kidney transplant recipient-donor pairing for graft survival is challenging.
• Existing risk prediction models have limited accuracy and calibration.
• Modern decision-support tools have not been adequately compared to traditional models.

Purpose of the Study:

• To develop a highly accurate risk-stratification index using artificial intelligence (AI).
• To compare AI-driven risk prediction with existing models like the kidney donor profile index (KDPI).
• To enhance deceased donor selection and kidney allocation schemes.

Main Methods:

• Utilized UNOS database (156,749 deceased kidney transplants, 2007-2021) for training and validation.
• Developed and assessed four machine learning models, including Deep Cox mixture.
• Evaluated models on death-censored graft survival using discrimination (CTD) and calibration (IBS) metrics.
• Conducted decision curve analysis and external validation using UK Transplant data.

Main Results:

• The AI-based Deep Cox mixture model demonstrated superior discriminative performance (AUC 0.66-0.68) and CTD (0.66).
• AI-based D-TOP significantly outperformed the KDPI model (CTD 0.59, AUC 0.60).
• Adequate calibration was achieved with an IBS of 0.12 for the AI model.

Conclusions:

• AI-based D-TOP is a more effective tool than KDPI for evaluating kidney transplant pairs.
• This AI approach can potentially improve deceased donor selection and optimize graft survival.
• Advanced computing, including AI, is expected to significantly impact future kidney allocation strategies.