Could fewer biopsies be needed after a heart transplant? NYU combines ECG AI with blood markers to assess rejection

After a heart transplant, promptly detecting when the immune system is attacking the transplanted organ remains an ongoing challenge for patients and their medical teams. Heart muscle biopsies provide direct tissue evidence, but they also involve an invasive procedure. A research team at NYU Langone Health is exploring the use of AI to analyze ECGs and blood tests together, hoping to assess rejection risk more accurately and reduce unnecessary biopsies.

The study was published online in the Journal of Heart and Lung Transplantation on September 25, and the university announced the results on October 6. According to the university’s press release and the same study description distributed through PR Newswire, the team trained the model on 5,300 ECG recordings from 2,357 adult heart transplant recipients, then used biopsy records to assess its predictive performance. Both sources came from the research institution; they do not represent two independent studies.

ECGs record the heart’s electrical activity, while blood tests provide another layer of clues: gene activity associated with cellular rejection and fragments of donor DNA in recipients’ blood. The research team noted that although these blood markers can help detect rejection, they may produce false positives, flagging patients without rejection for further biopsy. The purpose of the integrated model is to test whether electrical signals from the heart can complement the interpretation of blood tests.

The study description states that the integrated analysis used data from patients who received care between 2018 and 2024, pairing ECGs with biopsies performed within the preceding month. Biopsy results were divided into two groups: “no rejection or mild rejection” and “moderate or severe rejection.” This reflects the clinical situation in which more severe cases generally require treatment adjustments. However, this grouping also means that the model does not treat all degrees of rejection as the same warning sign.

In a test involving another 38 recipients, the integrated model correctly identified 94% of patients who did not have rejection. By comparison, interpretation based on blood tests alone incorrectly flagged 19 people as potentially needing a biopsy. The researchers estimated that the integrated model could allow these people to avoid the procedure. This is a potential benefit inferred from predictions; it cannot yet be equated with safely eliminating 19 biopsies in clinical practice.

The 94% figure describes performance in identifying patients without rejection. It cannot be interpreted directly as sensitivity for detecting rejection, nor is it sufficient to establish that the model can replace biopsies. The test included only 38 people, and the press release did not fully report the rate of missed diagnoses or statistical uncertainty. The gap of up to one month between ECGs and biopsies also means that the data do not all reflect patients’ condition at the same point in time. If the model is to inform decisions to omit biopsies, the key question remains whether it can reduce false alarms while avoiding missed cases of rejection that require treatment.

The study was funded by NYU Langone, and the team plans to validate the model in more patients and at multiple transplant centers. The next phase needs to establish whether this approach to interpretation applies to patients and testing workflows at different centers, and whether it is safe when actually used to guide biopsy scheduling. The current results offer clues toward less invasive monitoring, but more complete clinical evidence is still needed before routine care can change.