futurem2FutureMatch:

Enabling Better Organ Exchange Programs

 Over 123,000 adults and children currently await organ transplants. Unfortunately, only about 30,000 donor organs become available each year. Every day, roughly 20 people die waiting for a “match.”

WHY IT'S IMPORTANT

Donor cycles and chains, in which recipients and live donors who aren’t compatible trade with other recipient/donor pairs, can alleviate donor-organ shortages. But as the number of donor/recipient pairs increases, it’s difficult to calculate the best combination of trades. Tuomas Sandholm and his PhD students at Carnegie Mellon University have been working with the United Network for Organ Sharing (UNOS), the nonprofit organization that manages the national donor organ supply in the U.S. Their software automatically creates UNOS’s kidney-paired donation transplant plans, optimizing organ matches at 142 transplant centers—about 60 percent of such facilities in the U.S.—twice a week

“You have a decision variable for each possible donor/recipient chain, and each possible cycle, a huge number of decision variables … So if you have more memory, as we did with Blacklight, you can just bring in more variables. This enabled us to run a massive number of simulations within our new learning algorithms that learn to match better in a dynamic kidney exchange setting.”

—Tuomas Sandholm, Carnegie Mellon University

futurem

Diagram of an exchange between otherwise incompatible pairs.
 
HOW BLACKLIGHT AND XSEDE  HELPED

Calculating optimum donor exchanges requires both holding massive data in the computer’s memory and many parallel computations. An XSEDE allocation on PSC’s Blacklight supercomputer, supported by XSEDE's Extended Collaborative Support Service and Novel and Innovative Projects Program, provided both of these capabilities. The researchers could expand the number of donor/recipient pairs and broaden the criteria for matches so more hard-to-match patients could find donors. Their simulations on Blacklight suggested that both these goals could be achieved equitably, as defined by stakeholders. The researchers also showed that a combined kidney and liver exchange could find more matches than separate exchanges. UNOS has already enacted many of their improvements.


Machine Learning/Transplantation Medicine