Background – Diagnosis of cardiac transthyretin amyloidosis (cardiac ATTR) can be challenging due to the difficulty of collating nonspecific clinical findings across the electronic health record (EHR). Natural language processing (NLP) can scan the EHR to identify potential cardiac ATTR diagnoses and assess treatment appropriateness, closing care gaps and enhancing patient outcomes.

Methods – A set of queries using regular expressions searched for diagnosis codes, evidence from the patient’s present illness narrative, positive PYP, Congo red on heart biopsy, or positive ATTR genetics with cardiac involvement. We defined cardiac ATTR as a patient meeting any of the criteria. Those patients with clinical suspicion were deemed “suspected” for cardiac ATTR.

Results – Out of 3 million clinic notes and reports since June 2022, 1,850 patients had at least one document or procedure referencing amyloid, and 19 had an ICD-10 code documenting the disease. Cardiac ATTR was identified in 144 patients (precision 87% in a randomly selected cohort of 31 patients). Of those, 67 (47%) had no evidence of either being on medication or being offered. Among the 437 patients suspected for cardiac ATTR, 91% did not have a pending diagnostic order.

Conclusion – Significant care gaps exist in medication management for diagnosed cardiac ATTR patients and in confirmatory scanning for those with suspected ATTR. Integrating NLP into clinical workflows can address these gaps, improving diagnostic accuracy and patient outcomes.

VIEW THE PUBLICATION

VIEW THE POSTER