Heart-transplant rejection is treatable when detected, but the tests used to screen for it can generate false alarms that lead to invasive biopsies. A new retrospective study combined an electrocardiogram with two molecular blood markers in an attempt to separate genuine moderate or severe rejection from biological noise.[1][2]
The model was trained on about 5,300 ECGs from 2,357 adult transplant recipients treated between 2018 and 2024. Biopsy records served as the reference standard, with ECGs matched to biopsies performed within the preceding month.[2][3]
Three views of the same transplanted heart
What each signal contributes
| Input | What it measures | Why it can help |
|---|---|---|
| Electrocardiogram | Electrical activity across the heart | May capture physiological changes associated with injury or rejectionSource 1Source 2 |
| Gene-expression blood marker | Immune-related gene activity | Can indicate a cellular rejection responseSource 2 |
| Donor-derived cell-free DNA | Fragments of donor DNA in the recipient’s blood | Can rise when transplanted tissue is injuredSource 2 |
| Biopsy | Tissue examined directly | Used as the study’s reference for rejection severitySource 1Source 2 |
The value of combining modalities is not that they all say the same thing. It is that they observe different compartments. A blood marker may be sensitive to an immune signal without showing how the heart is functioning; an ECG may add physiological information that helps the model decide whether the molecular alert represents clinically important rejection.
The test result is promising—and small
The held-out test group contained 38 male and female transplant recipients. The combined model correctly identified 94% of patients who were not experiencing moderate or severe rejection. By contrast, the model using the blood biomarkers alone produced 19 false-positive predictions that would have triggered biopsies. In the retrospective analysis, the combined model would have filtered out those false alarms.[2][3]
That is a specificity result: the ability to recognise people who do not have the target condition. It does not by itself establish sensitivity, clinical outcomes or how the system behaves when deployed prospectively across different hospitals.
Why false positives matter
An endomyocardial biopsy involves passing a catheter into the heart to remove small tissue samples. It is a familiar part of transplant follow-up, but it consumes specialist resources and carries discomfort and procedural risk. A non-invasive screen that safely rules out more false alarms could reduce unnecessary procedures while preserving biopsies for cases that still need tissue confirmation.
The next test is clinical workflow, not another retrospective curve
The study grouped cases as no or mild rejection versus moderate or severe rejection, because the more serious category is the one more likely to change treatment. A useful prospective trial would need to predefine when the model changes a biopsy decision, monitor missed rejection, and test performance across centres, devices and patient populations.
The important contribution is the cross-compartment design. Instead of asking one blood test or one ECG to solve the entire problem, the model combines molecular warning signs with the heart’s electrical behaviour. The evidence is early, but it shows a plausible route to making post-transplant screening more selective rather than simply more sensitive.
Sources & further reading
- Multi-modal AI: Integrating electrocardiography with molecular biomarkers for noninvasive detection of cardiac allograft rejection JHLT Open · 25 Sep 2026 · Accessed 15 Oct 2026
- AI may help catch heart transplant rejection without biopsies NYU Langone Health · 6 Oct 2026 · Accessed 15 Oct 2026
- AI may help catch heart transplant rejection without biopsies Medical Xpress · 6 Oct 2026 · Accessed 15 Oct 2026