New AI Model Could Spare Heart Transplant Patients Unnecessary Heart Biopsies
Researchers at NYU Langone have trained a model on 5,300 ECGs combined with two existing blood biomarkers. In a small test group, it identified 94% of patients without rejection — and would have spared more patients unnecessary biopsies than blood tests alone.
After a heart transplant, patients live with a permanent monitoring task: detecting whether the body is rejecting the new organ before the damage becomes serious. Today this is done mainly in two ways — blood tests that provide early warning but also many false alarms, and heart biopsies, which are the definitive answer but require threading a catheter into the heart. A new study from NYU Langone Health suggests that a routine, inexpensive ECG could help reduce the number of times a biopsy becomes necessary.
The study, published online on September 25, 2026 in Journal of Heart and Lung Transplantation (JHLT Open), describes an AI model that combines ECG readings with two already established blood biomarkers. In a test group of 38 transplant recipients, the combined model correctly identified 94% of patients who did not experience rejection, according to the researchers. By comparison, a model based on blood tests alone misflagged 19 patients as possibly needing a biopsy — patients the combined model would initially have spared the procedure.
The background: blood tests that trigger unnecessary biopsies
The two blood biomarkers the model builds on are already in use. One measures gene activity linked to cellular rejection; the other measures fragments of donor DNA circulating in the recipient's blood. Previous research has shown these tests are effective at detecting rejection, but they frequently produce false positives that lead to unnecessary biopsies, the researchers have said (News Medical).
That is where the ECG data comes in. An electrocardiogram records the heart's electrical activity through sensors on the skin, is routine, inexpensive and non-invasive — and, according to the researchers, full of physiological information that current rejection monitoring does not exploit. The idea is not to replace the biopsy as the reference standard, but to filter better: to distinguish the patients who actually need a biopsy from those the blood test flagged incorrectly.
How the model is built
The researchers trained AI models on patterns in 5,300 ECG readings from 2,357 adult heart transplant recipients treated between 2018 and 2024. Each ECG was matched against a biopsy performed up to one month earlier — a time window supported by previous research — so that the biopsy records served as the benchmark for the models' predictive ability.
They built two variants: one model trained exclusively on ECG data, and one that combined ECG readings with the results of the two blood tests commonly used to predict rejection risk. Rejection was divided into two categories: no or mild rejection versus moderate or severe rejection.
In the test group of a further 38 male and female recipients, it was the combined model that performed best, correctly identifying 94% of patients without rejection.
A cautious reading of the numbers
Several caveats are worth highlighting. First, the 94% figure refers to correct identification of non-rejection in a small test group — not an overall diagnostic accuracy for the model. How well the model detects patients who are actually rejecting is not stated in the available source material. Second, all the figures are researcher-reported: they come from News Medical's coverage (published October 6, 2026) of NYU Langone's material, and the underlying journal article could not be retrieved directly for independent verification. Third, the test group of 38 patients is small, and small groups carry wide margins of uncertainty.
The researchers themselves describe the study as the first to combine these blood biomarkers with ECG readings in a single AI model and to compare the analysis directly against biopsy results. The claim is attributed to the researchers and has not been independently verified.
What the researchers themselves say
The study's senior author Lior Jankelson, MD, PhD — associate professor in the Leon H. Charney Division of Cardiology at NYU Grossman School of Medicine and associate professor of biomedical engineering at NYU Tandon School of Engineering — points to the ECG's untapped content:
"Our findings highlight that electrocardiograms contain an abundance of physiological information that can be used to substantially improve detection accuracy and enable earlier diagnosis and treatment for patients with heart transplant rejection," Jankelson said.
What remains before the clinic
The model is not in clinical use, and crucial validation remains. As a next step, the researchers plan to test the model in more patients at multiple transplant centers — a necessary step to determine whether the results from a single institution's data generalize to other patient populations and clinical practices. The study was funded by NYU Langone.
Until that validation is available, the clinical impact remains unclear: whether the model will actually reduce the number of biopsies in practice, or improve outcomes for patients. What the research shows so far is a proof of principle — that the electrical activity of a transplanted heart, read alongside blood tests patients are already taking, carries enough information to sort better than the blood tests alone.