Use of an echocardiographic-based, artificial intelligence system to improve racial disparities in care of patients with valvular heart disease

Abstract Funding Acknowledgements Type of funding sources: None. Introduction Referral for valvular interventions is frequently delayed, and racial disparities in access to care are prominent in the United States. Purpose To assess the effectiveness of an echocardiographic (Echo) – based, artificial intelligence (AI) driven protocol in improving management of patients with severe aortic stenosis (SAS) and severe mitral regurgitation (SMR). Methods From 10/2019 to 11/2022, 126,678 Echo’s (73,453 patients) were screened using a proprietary natural language processing (NLP) system to detect SAS and SMR based on the ACC/AHA guidelines. Aggregated zip code-derived census data and the area deprivation index (ADI) were merged based on each patient’s recorded address at the time of their index Echo. The control group comprised of Echo’s performed 10/2019–2/2020 (prior to implementation of our NLP system) and the intervention group of those between 2/2020–11/2022 (following implementation of NLP). Our AI system uses NLP to detect SAS/SMR from Echo reports and automatically sends a notification to the referring clinician’s inbox with recommendation for referral to appropriate specialist. The endpoint was follow-up defined as the combination of (a) follow-up Echo, (b) follow-up clinic visit with the heart valve (for SAS) or the advanced heart failure clinic (for those with SMR), or (c) follow-up valvular intervention. Results Figure 1 summarizes the sociodemographic and Echo characteristics of the patients, as well as outcomes. Notifications were more likely to be sent out for black patients (OR 2, 1.69–2.34, p