Problem
When a kidney becomes available for transplant, information about the donor kidney is uploaded into the United Network for Organ Sharing (UNOS) UNet database. UNOS-UNET gives a tender offer to transplant surgeons through an allocation program. Problems with the current system include surgeon-to-surgeon variability, individual surgeons’ inconsistent decisions over time, an effort-intensive donor kidney evaluation process, lost opportunity to transplant usable kidneys, and use of poor-quality kidneys. Problems lead to patient harm.
Solution
Our advanced AI solution would support Transplant Centers’ go/no-go decision for matching kidney transplant donors to recipients using Generative AI, Large Language Models, and predictions. It would compare deceased donor data in UNET, waitlisted recipient data in the DIHI Data Pipeline and MaestroCare, and two sources of unstructured narrative data: (1) case narratives from monthly M&M meetings and (2) retrospective review of organ offers that Duke declined but were transplanted elsewhere.
Impact
Large language modeling blended with generative AI and machine learning would generate informative summaries about the patient match and establish thresholds for a “go” versus a “no go” decision regarding a kidney offer. An AI solution would introduce far greater standardization of the matching process than current practice, leveraging LLM and Generative AI to use unstructured narrative data to match donors to recipients. We would leap toward streamlined operations and expand our reach.
