Artificial intelligence (AI) has become a driving force in modern healthcare, and its implications for organ donation and transplantation are growing rapidly. Drawing on decades of research and clinical experience, the presentation offered a comprehensive look at how AI is poised to reshape key dimensions of transplant care, from diagnostics to decision support.
Challenges
- The risk of bias and misinformation in AI applications due to unvetted or incomplete data
- Increasing complexity in clinical decision-making
- Limited expert availability in urgent situations
- Persistent inefficiencies in organ offer evaluation and patient surveillance
Innovations Introduced
- Unsupervised Machine Learning: Meaningful patterns can be detected without predefined assumptions or a priori knowledge. An example is identifying mycophenolate use as a leading predictor of poor vaccine response.
- Neural Networks and Decision Algorithms: Predictive algorithms are helping to forecast post-transplant outcome and risk, prioritize candidates, and enhance organ allocation decisions.
- Generative AI and Large Language Models (LLMs): These models draft patient summaries, prioritize transplant candidates, and communicate clinical information with clarity and compassion.
Results and Applications
1. Early results highlight AI’s growing potential in clinical settings. In one study using complex nephrology case scenarios, ChatGPT’s accuracy in solving the quizzes increased significantly within a 6-month time span. ChatGPT 3.5 began with a score of 42% accuracy, then GPT 4.0 reached 58%, and finally GPT 4V achieved 83%. ChatGPT’s scores beat transplant fellows (75%) and nearly matched the accuracy of program directors. Notably, these models were not trained on transplant-specific data, demonstrating the power of generative AI tools.¹
2. In another application, LLMs gave accurate and reliable answers to real transplant questions from online forums, especially when the questions were based on widely agreed-upon clinical guidelines. The best-performing questions had an accuracy outcome of 92%, a complete score of 90%, and a safe score of 96%. The lowest-scoring questions were linked to smaller amounts of data in the network or disagreements in the field.²
3. AI tools are also actively supporting:
- Patient Surveillance: Identifies patients who are at risk of “getting into trouble,” who are then flagged for medical providers to investigate
- Evaluating Organ offers: Biopsy interpretations, particularly during overnight hours
- HLA matching and prediction of waiting times
- Identification of ideal recipients for high-KDPI and marginal organs
- Automated donor referral processes, which have already led to increased authorization and transplant rates in pilot studies
Conclusions
AI is particularly useful in scenarios that are too computationally intensive or complicated for a human, when an expert is unavailable, and when certain tasks are not possible without the proper bandwidth or staff. AI’s role was underscored as a support tool rather than a substitute for clinical expertise, and it is crucial for the transplant community to embrace AI’s role in improving precision, efficiency, and equity in transplant care.
