Conference Coverage

AI-ECG and AI-Echo Flag Unrecognized ATTR-CM Risk

Edited by:
Brian McCurdy

Key Highlights

  • AI-ECG achieved an AUROC of 0.78, while AI-Echo achieved 0.88 in their respective diagnostic evaluation groups.
  • AI-ECG flagged 4.8% of the broader cohort without confirmed disease as having probable ATTR-CM.
  • Reported PYP positive predictive value reached 76.8% among patients with positive AI-ECG and AI-Echo screens.

Artificial intelligence applied to electrocardiograms (ECGs) and echocardiograms (AI-Echo) identified probable, clinically unrecognized transthyretin amyloid cardiomyopathy (ATTR-CM) across a US healthcare network. Combining the two modalities improved case detection, according to a TRACE-AI analysis being presented at the 2026 American Heart Association meeting.

The retrospective study included adults aged 50–95 with an ECG and echocardiogram recorded within the preceding 5 years and within 1 year of each other. Eight of the TRACE-AI network’s 15 US health systems had contributed data. The researchers distributed software to participating systems to run model predictions locally.

Investigators ascertained ATTR-CM status using technetium pyrophosphate (PYP) scintigraphy or ATTR-directed therapy within 1 year after the index study. They evaluated screening results at predefined operating thresholds targeting 90% specificity and examined associations between screen positivity and mortality in patients without confirmatory testing.

Study Findings

AI-ECG predictions were available for 289,848 adults, including 144,281 women (49.8%). Among 2,307 patients with PYP results or ATTR therapy, 484 had confirmed ATTR-CM. In the subgroup with disease ascertainment within one year after the index ECG (n=1458), AI-ECG achieved an AUROC of 0.78 (95% CI, 0.75–0.80). In the broader cohort without confirmed disease, it flagged 12,517 patients (4.8%) as having probable ATTR-CM. Positive screening was associated with mortality (adjusted hazard ratio [HR], 1.51; 95% CI, 1.45–1.56).

AI-Echo predictions were available for 13,031 participants. The model achieved an AUROC of 0.88 among 525 with valid ATTR information and flagged 382 patients without confirmatory testing (2.9%) as probable cases. Among patients evaluated with both AI modalities and PYP, reported PYP positive predictive value increased from 6.6% with dual-negative screens to 47% with ECG positivity, 61.3% with echo positivity, and 76.8% with dual positivity. Among patients without confirmatory testing, dual positivity was associated with mortality (adjusted HR, 1.84; 95% CI, 1.28–2.63).

Clinical Implications

According to the authors, the findings support a tiered, AI-guided approach to identifying patients who may benefit from confirmatory ATTR-CM testing. They concluded that combining AI-ECG with AI-Echo, when available, improved case detection. The reported probable cases were screening classifications, not confirmed diagnoses.

Expert Commentary

“A tiered, AI-guided pathway holds promise for the timely identification of patients who would benefit from confirmatory ATTR-CM testing,” the researchers concluded.

Reference

Batinica B, Oikonomou EK, Croon PM, et al. Artificial intelligence for the screening of ATTR cardiomyopathy in the US: a multicenter analysis from the TRACE-AI network. Abstract DP2293. Presented at: American Heart Association Scientific Sessions 2026; November 9, 2026.