The Influence of AI Support on Surgeons’ Decision-Making in Kidney Transplant Cases

Improving kidney utilization requires timely, informed decisions across multiple stages of the transplant process. For 4 years, this research project has developed and evaluated explainable artificial intelligence (XAI) systems to enhance decision quality, efficiency, and trust in this high-stakes context. A stakeholder engagement process with transplant surgeons, Organ Procurement Organization (OPO) coordinators, and patient representatives identified key decision challenges and AI information needs, emphasizing interpretable, optional AI features and user control over information depth. Building on these insights, our team developed two deep learning models, one predicting kidney transplant likelihood for candidates and another identifying hard-to-place kidneys. Pilot human-subject studies examined how users interact with AI recommendations and uncertainty information in risky and non-risky contexts.

In the final year of this effort, we are conducting a randomized controlled trial with transplant stakeholders to evaluate the effectiveness of our AI tools. Using SimUNet, a web-based simulator replicating the United Network for Organ Sharing (UNOS) DonorNet interface, we will assess how XAI influences surgeons’ decisions and how timing, before versus after an initial decision, affects their use. Surgeons review multiple de-identified donor cases across Pre-decision AI (AI score shown before making a decision), Post-decision AI (AI score shown after an initial independent decision as a second opinion), and No AI conditions. We will analyze surgeons’ alignment with real-world transplant decisions and outcomes, as well as their decision time. Preliminary work suggests that the post-decision AI condition will maximize performance, while pre-decision AI will maximize efficiency, presenting a trade-off for system outcomes.