Machine Learning Model Cuts Canceled Liver Transplants By 60%

A shortage of donor livers leaves many patients waiting for a transplant, yet nearly half of potential transplants from donors who die after withdrawal of life support must be canceled. For donation after circulatory death, the donor must pass within 30 to 45 minutes, or the organ becomes too damaged for use. Delays often force surgeons to abandon the procedure after preparations have already begun. Now, a new machine-learning model offers a solution by predicting whether a donor is likely to die within that critical window.

Researchers at Stanford Medicine (Stanford, CA), in collaboration with Kyoto University (Kyoto, Japan), trained the artificial intelligence (AI) model using donor demographics, vital signs, neurologic reflexes, ventilator settings, cardiovascular history, and laboratory values. It analyzes this information before preparations start so surgical teams can determine early whether an organ is likely to be viable. The research published in Lancet Digital Health, shows that the model substantially reduced futile procurements while outperforming surgeon judgment.

In the study, the researchers compared multiple machine-learning algorithms using more than 2,000 donor cases from six U.S. transplant centers. The final model predicted whether death would occur in the viable window with 75% accuracy, outperforming surgeons who averaged 65%. It remained accurate even when donor records had missing information. The model also reduced futile procurements by 60%, limiting situations in which the surgical team prepares for an organ retrieval only to discover it can no longer be used.

The tool can be customized to reflect surgeon preferences and institutional protocols, such as calculating donor death from withdrawal of life support or from the start of agonal breathing. A natural-language interface automatically extracts data from donor medical records, simplifying integration into clinical workflows. Although missed opportunities—cases in which death occurs early but no procurement was initiated—remained around 15%, the team has already identified updated algorithms that reduce this to about 10%.

In addition to improving efficiency, the model could increase the number of livers available for transplantation. Donation after circulatory death is expanding, supported by technologies like normothermic machine perfusion that keep organs functioning during transport. By helping clinicians anticipate viability more accurately, the model ensures resources are allocated appropriately and candidates receive transplants sooner. Researchers are now adapting the system for heart and lung transplantation, aiming for a unified AI framework across organ types. Continued refinement may further improve prediction accuracy and reduce missed opportunities.

“By identifying when an organ is likely to be useful before any preparations for surgery have started, this model could make the transplant process more efficient,” said Kazunari Sasaki, MD, senior author on the study. “It also has the potential to allow more candidates who need an organ transplant to receive one.”

Related Links:
Stanford Medicine
Kyoto University