PO88 - Artificial intelligence - enhanced EKG predicts presence of cardiac sarcoidosis with a high sensitivity
Emma Johns (United States)1; Selene Rubino (United States)1; Jenna Davison (United States)1; Andrew Rosenbaum (United States)1; Konstantinos Siontis (United States)1; Courtney Arment (United States)1;
1 - Mayo Clinic;
Keywords: Artificial intelligence; screening; epidemiology;
Select the theme: Cardiac Sarcoidosis
Type: Original Papers
Presentation: Poster Presentation

Introduction: Early detection of cardiac sarcoidosis (CS) is a diagnostic challenge, particularly given that 25% of patients may have isolated cardiac involvement. There is a critical need for effective screening strategies to prevent the development of life-threatening arrhythmias in CS. Recently, our institution developed an AI algorithm for the detection of CS on EKG (3). We appraised its sensitivity in a separate retrospective cohort of CS patients followed by Rheumatology.

Objectives: We measured the sensitivity of our AI-EKG algorithm for detection of CS. We measured whether clinical characteristics affected sensitivity of the algorithm.

Materials and Methods: Patients were identified using Mayo Data Explorer, a self-service web application that enables SQL queries of the medical record.

We included adult Rheumatology patients with ICD codes consistent with sarcoidosis who were seen in any of the 3 Mayo clinic sites. The diagnosis of CS was based on the treating rheumatologist assessment.

Patients seen prior to 5/1/2018 or for whom cardiac sarcoidosis was later attributed to chronic infections or malignancy were excluded. Patients used in the training of the original algorithm were excluded.

Results: 153 cases were included out of 177 reviewed. The mean age at time of cardiac sarcoidosis diagnosis was 58.9 years (SD 12). 68 (44%) patients were female. Patients were 79.9% Caucasian, 16.6% Black, and 2.2% Asian. “Probable” CS was found in 60%, whereas “presumed, isolated”, “presumed, systemic”, and “definite” comprised 15%, 14.4%, and 10% of patients respectively. 17.6% had clinically isolated cardiac sarcoidosis.

Sensitivity at the time of earliest EKG was 0.948. Segregating by female sex, non-White race, or isolated cardiac involvement did not significantly change the sensitivity of the test. Sensitivity at the time of the last EKG rose to 0.974, despite initiation of steroids and/or DMARDs in most patients.

Conclusions: We found the AI-EKG was 94.8% sensitive for detection for cardiac sarcoidosis in rheumatology patients.