AI-guided population screening for cardiac ATTR amyloidosis in heart failure patients within ambulatory health service's database
I Davarashvili, O F R A Krieger, Y A E L Konstatinovsku, O F R A Appelman, I Z H A R Laufer, A D V A Tzuk Onn, S H A H A R Dekel, S H L O M O Vinker, A V I S H A GrupperAbstract
Background
Transthyretin amyloidosis (ATTR) is a progressive and potentially fatal multisystem disease caused by amyloid fibril deposition, frequently involving the heart. Cardiac ATTR amyloidosis (ATTR-CM) typically presents as restrictive cardiomyopathy and heart failure, often with preserved or mildly reduced ejection fraction. Due to heterogeneous and staggered cardiac and extra-cardiac manifestations, ATTR-CM remains significantly underdiagnosed, resulting in delayed treatment and poor outcomes. Artificial intelligence (AI)–based tools may facilitate earlier identification of patients at risk.
Methods
An AI-based machine-learning algorithm was design to identify patients with suspected ATTR-CM within a large healthcare system database. Echocardiographic data from 2019 onward were screened, yielding 60,800 patient records. Inclusion criteria were age ≥65 years, left ventricular ejection fraction >45%, interventricular septal or posterior wall thickness ≥12 mm, diastolic dysfunction, and estimated pulmonary artery systolic pressure >35 mmHg. The algorithm further incorporated weighted ATTR "red flags," including atrial fibrillation, prior pacemaker implantation, atrioventricular conduction disease, carpal tunnel syndrome, spinal stenosis, biceps tendon rupture, peripheral neuropathy, orthostatic hypotension, and autonomic dysfunction. Patients flagged by the AI model were referred for confirmatory imaging using technetium-99m pyrophosphate (PYP) SPECT or cardiac magnetic resonance (CMR).
Results
The AI-based screening identified 166 patients with high suspicion for ATTR-CM. Due to logistical and technical limitations, 63 patients underwent confirmatory imaging (61 PYP SPECT, 2 CMR). ATTR-CM was diagnosed in 6 patients (9.5% of those imaged). All diagnosed patients demonstrated concordant echocardiographic structural changes and multiple extra-cardiac manifestations consistent with ATTR amyloidosis.
Conclusion
An AI-based algorithm integrating echocardiographic parameters with extra-cardiac clinical features successfully identified patients with previously unrecognized cardiac ATTR amyloidosis within a large heart failure population. This approach may improve early detection, optimize referral for advanced imaging, and reduce diagnostic delay in ATTR-CM.