Article Impact Level: HIGH Data Quality: STRONG Summary of JHLT Open https://doi.org/10.1016/j.jhlto.2026.100698 Dr. K. Chen et al.
Points
- NYU Langone Health researchers developed a multi-modal artificial intelligence model combining electrocardiogram readings and blood biomarkers to noninvasively detect heart transplant rejection.
- Machine learning algorithms were trained on five thousand three hundred electrocardiogram readings paired with biopsy records from over two thousand three hundred adult heart transplant recipients.
- Testing demonstrated that the combined model correctly identified ninety four percent of patients not experiencing rejection while reducing false positive results generated by blood tests alone.
- Integration of electrical signals with cell free donor DNA and immune transcriptional biomarkers helped discriminate true allograft rejection from diagnostic biomarker noise in transplant recipients.
- Study authors concluded that prospective multi-center clinical trials are required to validate the multi-modal artificial intelligence framework before widespread clinical adoption in transplant centers.
Summary
This study evaluated a multi-modal artificial intelligence model integrating electrocardiogram (EKG) recordings with molecular blood biomarkers to detect cardiac allograft rejection noninvasively. Published in JHLT Open by researchers at NYU Langone Health, the investigation addressed the limitations of invasive endomyocardial biopsies—the current diagnostic gold standard—and the high false-positive rates of standalone blood tests. The research sought to determine whether cross-compartment concordance across electrical, immune-transcriptional, and donor-derived cell-free DNA signals could improve diagnostic specificity and prevent unnecessary biopsies in adult heart transplant recipients.
The investigators trained AI models on 5,300 EKG recordings matched with endomyocardial biopsy records from 2,357 adult heart transplant recipients treated between 2018 and 2024. Model performance was evaluated in a test cohort of 38 heart transplant recipients to discriminate between no/mild rejection and moderate/severe rejection requiring therapeutic intervention. The multi-modal AI model demonstrated improved specificity over single-modality approaches, correctly identifying 94% of patients without rejection. While blood biomarker testing alone generated 19 false-positive predictions that would have triggered invasive biopsies, the combined multi-modal model successfully filtered out biomarker noise, though AUC differences across modalities did not achieve statistical significance in this preliminary cohort.
The authors conclude that integrating EKG electrophysiological signals with immune-transcriptional and donor-DNA biomarkers provides a promising noninvasive approach to discriminate true cardiac allograft rejection from diagnostic noise. By demonstrating higher specificity, the cross-compartment AI model has the potential to spare transplant recipients from unnecessary invasive biopsies and enable earlier therapeutic management. However, given the small test set, the authors emphasize that a pre-specified, prospective multi-center trial powered for at least 50 rejection events is necessary to validate these findings for clinical implementation.
Link to the article: https://www.jhltopen.org/article/S2950-1334(26)00219-3/fulltext
References
Chen, K., Koesmahargyo, V., Ronan, R., Barbhaiya, C. R., Bernstein, S. A., Kushnir, A., Garber, L., Yang, F., Aizer, A., Reyentovich, A., Chinitz, L. A., Goldberg, R. I., & Jankelson, L. (2026). Multi-modal AI: Integrating electrocardiography with molecular biomarkers for noninvasive detection of cardiac allograft rejection. JHLT Open, 100698. https://doi.org/10.1016/j.jhlto.2026.100698
