When a donor kidney becomes available, doctors often have only hours to decide which of several waiting patients should receive it. A new study from Elmbridge University asked a difficult question: what would an artificial intelligence system choose in the same situation, and would we be comfortable with its answer?
The researchers trained a model on ten years of anonymized data from the national transplant registry and then asked it to allocate kidneys in 2,300 historical cases. In 81 percent of them, it picked the same patient as the transplant committee had.
Where the machine disagreed
The remaining 19 percent are what caught the researchers’ attention. In most of those cases, the model favored younger patients with a better predicted long-term survival of the transplanted organ, even when an older patient had been waiting much longer.
From a purely statistical view, the model maximizes the years a kidney will work. But our allocation rules also value fairness and time spent waiting. A machine doesn’t weigh those things unless we tell it exactly how.
Dr. Leonie Haas, lead author and medical ethicist at Elmbridge University
The current system gives points for waiting time, tissue compatibility, medical urgency and age difference between donor and recipient. Committee members can also consider factors that are hard to put into numbers, such as a patient’s support at home.
Help, not a decision-maker
Transplant surgeons interviewed for the study saw some value in the tool, mainly to flag risks and speed up the analysis when time is short.
“At three in the morning, a second opinion that has read every case in the registry could be useful,” said Dr. Aleksander Ilic, a transplant surgeon in Riverton. “But I would never hand over the final call. Someone has to be accountable to the patient and their family.”
Patient groups warned that any model trained on past decisions could also repeat past biases, for example against patients from rural areas who were referred later. The researchers found some evidence of this and corrected for it, but say it shows why such systems need independent checks.
What comes next
The study’s authors are not proposing that AI should replace transplant committees. Instead, they recommend clear national rules before any hospital uses such tools, including full transparency about how recommendations are made.
The national bioethics council said it will open a public consultation on artificial intelligence in organ allocation early next year.
Comments 3
Useful as a tool, scary as a judge. Good that the researchers say so clearly.
My father waited four years for a kidney. Waiting time has to count for something, whatever the statistics say.
The bias point is the most important part of this article.