AI tools are being tested to rescue donor hearts that transplant teams now reject

The new tools aim to standardize risk without removing clinicians

The meeting presentation outlined several AI models designed to support this decision-making process. One highlighted tool is TOPHAT, short for Tool Predicting Heart Acceptance for Transplant. Developed by Wayda in collaboration with ISHLT President-Elect Kiran Khush of Stanford Health Care, the web-based model uses 20 donor characteristics to estimate the probability that a transplant center will accept a donor heart based on historical data.

That framing is notable. The tool is not described as directly declaring a heart safe or unsafe. Instead, it estimates acceptance likelihood using patterns from prior decisions. In practice, that could make it useful both as a predictor and as a mirror, showing when a team’s instinct diverges sharply from broader historical behavior.

The source text emphasizes that these systems are intended to synthesize risk rather than replace physicians. That distinction is likely to matter for adoption. In a field as high stakes as heart transplantation, fully automated acceptance decisions would face significant clinical and ethical resistance. Decision support, by contrast, may be easier to integrate because responsibility remains with the transplant team.

The opportunity is in the hearts that may be needlessly lost

The strongest case for AI in this setting comes from the gap between supply and use. If only 30% to 40% of available hearts are transplanted, there is little room for complacency. The source text explicitly notes that research shows not all donor hearts are justifiably discarded. That does not mean every rejected heart should have been used, but it does mean a portion of the current rejection pattern may be avoidable.

For patients waiting on transplant lists, that distinction is not academic. Each avoidable discard can represent a missed chance at survival or recovery. The value of AI here is therefore less about futuristic automation and more about raising the floor on consistency, especially when teams must work quickly across a large and heterogeneous set of donor profiles.

It may also reduce variation across institutions. Some centers are more aggressive than others in taking marginal or complex donors. A strong prediction model could provide a common framework for discussing risk, making decision-making more data-driven across centers that currently rely on local habit, experience, and institutional culture.

What would success look like?

The clearest success metric would be simple: more donor hearts used safely. The source material does not claim that AI has already solved the problem, nor does it provide definitive outcomes data showing system-wide implementation. What it does show is that clinicians and researchers are building tools specifically around a chokepoint that has outsized consequences for transplant access.

If those tools help teams identify hearts that are acceptable but currently overlooked, the impact could be significant without requiring any new breakthrough in organ generation, preservation, or surgery. In that sense, the work is a reminder that some of medicine’s biggest gains can come not only from new therapies, but from better decisions about the resources already available.

The broader message from the transplant meeting is that AI’s most credible role in health care may be narrowly defined and operational. In the donor-heart setting, that means helping humans make a difficult call more quickly, more consistently, and with a better grasp of patterns hidden in historical data. Given the scale of the donor shortage and the narrow decision window, that is a highly practical ambition.

This article is based on reporting by Medical Xpress. Read the original article.