Electrocardiogram-Based Deep Learning to Prioritize Testing for Transthyretin Amyloid Cardiomyopathy.

Authors

Philip M Croon, Bruno Batinica, Lovedeep S Dhingra, Ryan B Choi, Evangelos K Oikonomou, Sumukh V Shankar, Veer Sangha, Robert M A van der Boon, Michelle Michels, Maarten van Ettinger, Peter-Paul Zwetsloot, Sergio Teruya, Cesia Gallegos Kattan, Edward J Miller, Navid Noory, Oscar M Westin, Sie K Fensman, Steen Hvitfeldt Poulsen, Avneet Singh, Sudarshan Balla, Anouk Achten, Charalambos Vlachopoulos, Alexios S Antonopoulos, Nico Bruining, Julian D Gillmore, Mathew S Maurer, Frederick L Ruberg, Marianna Fontana, Rohan Khera

Published in

JAMA. Aug 28, 2026. Epub Aug 28, 2026.

Abstract

Transthyretin amyloid cardiomyopathy (ATTR-CM) is a treatable cause of heart failure, but diagnosis is often delayed. Accessible tools to prioritize confirmatory evaluation could reduce missed diagnoses when cardiac imaging is limited.
To develop and validate a locally deployable artificial intelligence (AI)-enabled system that identifies patients for further ATTR-CM evaluation from routine electrocardiography (ECG) images.
This diagnostic test study used an AI-ECG model developed within Yale New Haven Health (using ECGs from August 2015-June 2023) and temporally validated (July 2023-July 2025). External validation included 5 multinational cohorts and 3 screening cohorts (older Black and Hispanic adults with heart failure; adults with prior carpal tunnel surgery; 99-3902 individuals per cohort).
Use of AI-ECG for detecting ATTR-CM from routine ECG images or raw 12-lead signals.
Area under the receiver operating characteristic curve (AUROC), sensitivity, specificity, and positive and negative predictive values for ATTR-CM confirmed by cardiac amyloid radionuclide imaging (CARI) or biopsy. An exploratory analysis evaluated sequential screening with AI-ECG followed by AI-enabled echocardiography.
Development used 28 174 ECGs from 11 291 patients (293 with ATTR-CM). In internal validation (44 123 patients; mean age, 68.5 years; 49.7% female), AUROC was 0.84 (95% CI, 0.79-0.89), with sensitivity of 0.72 and specificity of 0.86 at the prespecified threshold, and was maintained in a specificity stress test among patients with features that mimic ATTR-CM (left ventricular hypertrophy or severe aortic stenosis without amyloid; AUROC, 0.81 [95% CI, 0.75-0.86]) and those referred for CARI (AUROC, 0.78 [95% CI, 0.73-0.84]). Across 5 external cohorts, AUROCs ranged from 0.78 to 0.89. In 3 screening cohorts, AUROCs were 0.76 (95% CI, 0.68-0.83) in the SCAN-MP study (n = 645) and 0.79 (95% CI, 0.65-0.93) and 0.91 (95% CI, 0.82-0.97) in 2 CACTUS study cohorts (n = 251, n = 121). Sequential AI-ECG and AI-enabled echocardiography increased the positive predictive value from 0.24 to 0.66 and reduced sensitivity from 0.84 to 0.68.
Locally deployable AI applied to ECG images discriminated ATTR-CM across multinational retrospective and screening cohorts. This approach may provide an accessible first step to prioritize selected patients for echocardiography and confirmatory imaging, although intended use and calibration require prospective evaluation.

PMID:
42663420
Bibliographic data and abstract were imported from PubMed on 28 Aug 2026.

Read full publication at:
Please sign in to see all details.

Stats

1-terrible, 9-excellent. How would you rate this publication? Sign in in to submit your rating.

Post a comment

You need to be signed in to post comments. You can sign in here.

Comments

There are no comments yet.