Finding undiagnosed ATTR-CM patients using AI technology in clinical settings

Project Summary: Transthyretin Amyloidosis with cardiac myopathy, ATTR-CM, represent a serious healthcare issue. ATTR-CM is involved in 13% of heart failure, 16% of transcatheter aortic-valve replacement, and 5% of individuals with presumed hypertrophic cardiomyopathy. The primary challenge is that most

patients are undiagnosed or their diagnosis is delayed for multiple years. Since the damage ATTR-CM causes to the heart is progressive, diagnosis delays strongly impact prognosis and increase mortality. Diagnosis is problematic for two reasons: ATTR-CM has a variable presentation and the prevalence is not high. Thus,

ATTR-CM is often not considered during diagnosis and a more common diagnosis with similar symptoms is given erroneously. Up to 98% of patients are not diagnosed due to the low prevalence and variable presentation. One study found that 32% of ATTR-CM patients had previously been misdiagnosed as having

more common cardiovascular diseases. A readily-available genetic test can be used to detect hATTR and 99mTc-DPD-scintigraphy can be used to diagnose ATTR-CM (both hereditary and wild type). Fortunately, once diagnosed, ATTR is treatable. Thus, the main challenge for ATTR-CM is diagnosis, not therapy. An

effective and economical precision screening system is needed to find the individuals most at risk of ATTR- CM. Those identified via precision screening could be tested and, treated with effective therapy resulting in saved lives and reduced healthcare costs. Atomo’s goal in this SBIR Fast-Track proposal is to create, optimize

and implement an AI-based Clinical Decisions Support System (CDSS) to identify probable yet undiagnosed ATTR patients before they develop CM. For this work, we are partnering with Dr. Dan Rader and PENN Medicine. Dr. Rader is the Seymour Gray Professor of Molecular Medicine and Chair of the Department of

Genetics at the Perelman School of Medicine of the University of Pennsylvania. Dr. Rader also directs the Penn Medicine Biobank. We would utilize the BioBank to identify True Positive patients to train and evaluate an AI model to find probable yet undiagnosed ATTR individuals. The model would be used in a pilot, most

likely as a quality improvement initiative. To complete this work, Atomo will leverage its proven ML technologies that have been used and verified clinically, with published in peer-reviewed journals. The ATTR AI model would be commercialized as an Insights As A Service (IaaS) CDSS.