Validation of an Artificial Intelligence–Derived ECG Algorithm for Detecting Cardiac Amyloidosis in Patients With Heart Failure With Preserved Ejection Fraction: Clinical Application and Prognostic Implications

Background Cardiac transthyretin amyloidosis (ATTR‐CA) is frequently underdiagnosed and commonly presents as heart failure with preserved ejection fraction (HFpEF). Early identification enables disease‐modifying therapy but remains challenging in routine practice, so simple, widely available screening tools are needed. Methods In this multicenter validation study, 885 patients with HF from 2 European centers were included. The internal validation cohort comprised 560 patients with preserved ejection fraction and analyzable ECGs: 149 with ATTR‐CA, 318 with HFpEF, and 93 with hypertrophic cardiomyopathy. External validation used an independent cohort of 107 patients (72 ATTR‐CA, 31 HFpEF, 4 hypertrophic cardiomyopathy). Standard 12‐lead ECGs were analyzed blindly by 3 independent observers using a previously developed, 2‐step, artificial intelligence–derived, visually interpretable ECG algorithm. Results The ECG pattern was present in 82.6% of patients with ATTR‐CA, versus 10.2% with HFpEF and 6.5% with hypertrophic cardiomyopathy ( P <0.001). Internal‐cohort accuracy was high: area under the curve 0.87 (95% CI, 0.84–0.90), sensitivity 83% (95% CI, 76%–88%), specificity 91% (95% CI, 88%–93%), and negative predictive value 93% (95% CI, 91%–96%). In the external cohort, the area under the curve was 0.84 (95% CI, 0.76–0.92), with sensitivity 89% (95% CI, 78%–94%) and specificity 79% (95% CI, 63%–90%). The pattern was strongly associated with ATTR‐CA (odds ratio, 46 [95% CI, 27–80]; P <0.001) and with reduced 3‐year survival (log‐rank P =0.007). Conclusions A visually interpretable, artificial intelligence–derived ECG algorithm enables effective screening for ATTR‐CA among patients with HFpEF. Its simplicity and compatibility with standard ECG systems support broad clinical implementation.