AI ECG Screening Detects Cardiac Amyloidosis After Asthma Misdiagnosis in Routine Care

The case involved a 68-year-old with progressive dyspnea initially attributed to late-onset asthma. Serial ECGs showed low-voltage QRS and pseudo-infarct patterns that the AI flagged as consistent with transthyretin amyloid cardiomyopathy, a diagnosis confirmed by technetium pyrophosphate scan and fat-pad biopsy. Prior physician review had overlooked these subtle voltage discrepancies amid normal spirometry.

Related work from Mayo Clinic and Mount Sinai demonstrates that deep-learning ECG models achieve AUCs of 0.91-0.94 for amyloid detection in retrospective cohorts of 20,000-40,000 patients, outperforming cardiologist visual reads by 15-20 percentage points. These observational studies lack randomization and outcome endpoints, leaving open whether earlier flags translate to fewer hospitalizations.

Deployment of the free tool raises questions about calibration drift across ECG vendors and the rate of false positives that could trigger unnecessary scintigraphy. A forthcoming pragmatic trial at three health systems will track 12-month mortality and quality-of-life metrics against usual care.

Integration timelines depend on liability frameworks and whether payers reimburse confirmatory imaging triggered by AI alerts.