In 1954, the world’s initial successful organ transplantation occurred at Brigham and Women’s Hospital, involving a kidney given from one twin to another. At that moment, a cohort of physicians and researchers had accurately hypothesized that the recipient’s antibodies would be less likely to reject an organ from a genetically identical twin. One Nobel Prize and several decades later, progress in immune-suppressing medications boosted the feasibility of and need for organ transplants. Currently, over 1 million organ transplants have been conducted in the United States, more than any other nation globally.
The remarkable extent of this accomplishment was made feasible due to improvements in organ matching systems: The premier computer-based organ matching system was introduced in 1977. Despite ongoing advancements in computing, medicine, and matching technologies over the years, more than 100,000 individuals in the U.S. are presently listed on the national transplant waiting list, with 13 individuals succumbing each day while awaiting an organ transplant.
Most computational investigations in organ allocation concentrate on the preliminary phases, when patients on the waiting list are being prioritized for organ transplants. In a recent paper presented at the ACM Conference on Fairness, Accountability, and Transparency (FAccT) in Athens, Greece, researchers from MIT and Massachusetts General Hospital emphasized the concluding, less-examined stage: when an offer is extended and the physician at the transplant center determines on behalf of the patient whether to accept or decline the proposed organ.
“I don’t think we were tremendously shocked, but we were certainly disappointed,” co-first author and recent MIT PhD graduate Hammaad Adam remarks. Utilizing computational models to scrutinize transplantation data from over 160,000 transplant candidates in the Scientific Registry of Transplant Recipients (SRTR) from 2010 to 2020, the researchers discovered that physicians were generally less inclined to accept liver and lung offers for Black candidates, leading to further obstacles for Black patients in the organ allocation framework.
For livers, Black patients encountered a 7 percent lower probability of offer acceptance compared to white patients. In the case of lungs, the gap widened even further, with a 20 percent reduced chance of offer acceptance than white patients with comparable characteristics.
The findings do not inherently suggest clinician bias as the primary factor. “The more significant implication is that even if there are reasons that validate clinical decision-making, there could be clinical conditions we did not account for, that are more prevalent among Black patients,” Adam clarifies. If the wait-list fails to recognize certain patterns in decision-making, they may generate barriers in the process even if the system itself is “impartial.”
The researchers further highlight that considerable variability in offer acceptance and risk tolerances among transplant centers may complicate the decision-making process. Their FAccT paper references a 2020 article published in JAMA Cardiology, which determined that candidates on the wait-list listed at transplant centers with lower acceptance rates have a higher likelihood of mortality.
Another major observation was that an offer is more likely to be accepted if the donor and candidate share the same race. The paper characterizes this trend as “concerning,” considering the historical inequities in organ procurement that have restricted donations from racial and ethnic minority groups.
Prior research conducted by Adam and his associates has sought to address this disparity. Last year, they assembled and published Organ Retrieval and Collection of Health Information for Donation (ORCHID), the first multi-center dataset detailing the performance of organ procurement organizations (OPOs). ORCHID comprises 10 years’ worth of OPO data and aims to facilitate research addressing bias in organ procurement.
“Engaging in meaningful work in this field requires time,” says Adam, noting that the entire organ allocation initiative took years to finalize. To his knowledge, only one paper thus far investigates the relationship between offer acceptance and race.
While the bureaucratic and highly interdisciplinary nature of clinical AI projects may deter computer science graduate students from pursuing them, Adam dedicated himself to the project throughout his PhD under the guidance of associate professor of electrical engineering Marzyeh Ghassemi, an affiliate of the MIT Jameel Clinic and the Institute of Medical Engineering and Sciences.
To graduate students keen on pursuing clinical AI research endeavors, Adam advises them to “liberate [themselves] from the cycle of publishing every four months.”
“I found it liberating, to be honest — it’s acceptable if these collaborations take time,” he states. “It’s difficult to sidestep that. I made a deliberate choice a few years back and was content undertaking that work.”
This research received backing from the MIT Jameel Clinic. This investigation was partially funded by Takeda Development Center Americas Inc. (successor in interest to Millennium Pharmaceuticals Inc.), a NIH Ruth L. Kirschstein National Research Service Award, a CIFAR AI Chair at the Vector Institute, and by the National Institutes of Health.