Improving deceased donor kidney utilization: predicting risk of nonuse with interpretable models

Area of Science:

• Nephrology
• Transplantation Medicine
• Health Services Research

Background:

• A significant number of deceased donor kidneys are not utilized for transplantation, despite a high demand.
• Early identification of kidneys at risk of nonuse is crucial for implementing effective allocation strategies.
• Existing complex machine learning models for predicting kidney nonuse risk face challenges in practical implementation.

Purpose of the Study:

• To develop simplified and implementable models for predicting the risk of deceased donor kidney nonuse.
• To enhance the practical application of kidney nonuse risk prediction in organ allocation.

Main Methods:

• Proposed simplified models integrate the Kidney Donor Risk Index (KDRI) with a limited set of variables identified through machine learning or expert input.
• Incorporated Organ Procurement Organization (OPO)-level factors influencing kidney disposition into the prediction models.
• Validated the performance of simplified models against more complex, multi-variable approaches.

Main Results:

• The developed simplified models achieved competitive predictive performance compared to complex, data-intensive models.
• The proposed models offer enhanced interpretability and ease of use in clinical practice.
• Identified key factors contributing to kidney nonuse, including variations in Organ Procurement Organization (OPO) practices.

Conclusions:

• Simplified, interpretable models accurately predict deceased donor kidney nonuse risk.
• These models can guide the development of targeted interventions to increase kidney transplantation rates, particularly for "hard-to-place" organs.
• Understanding OPO-specific variations is vital for optimizing organ allocation strategies.