Patient and Clinician Perceptions of a Deep Learning Model for Automated Screening for Patients At-Risk for Undiagnosed Structural Heart Disease in the Emergency Department: Cross-sectional Survey Study (Preprint)
Year: 2026
Type: preprint
Study design: Observational Study
Abstract: BACKGROUND EchoNext is a prospectively validated deep learning model that identifies patients at risk for underdiagnosed structural heart disease (SHD) from 12-lead electrocardiograms (ECGs) routinely acquired in the emergency department (ED). Successful translation of Artificial Intelligence (AI)-based clinical decision support tools from technical validation to clinical impact depends on stakeholder acceptance, yet few studies have assessed both patient and clinician perspectives on a specific AI tool embedded in a defined ED workflow. OBJECTIVE This study aimed to characterize ED patient and clinician perspectives on AI-enabled SHD screening using EchoNext, and to identify perceived facilitators and barriers to adoption prior to health system-wide deployment. METHODS Two structured cross-sectional surveys were administered to patients and clinicians at a single urban academic medical center. Patients aged 18-65 presenting with non-urgent ED complaints completed an in-person survey assessing AI familiarity, comfort with AI-assisted clinical tasks, and factors influencing acceptance. ED clinicians completed an electronic survey that included a standardized clinical vignette describing AI-ECG-generated best practice alert, and assessed trust, perceived utility, implementation concerns, and adoption priorities. Descriptive statistics were used to summarize responses. RESULTS Between July and October 2024, 102 patient surveys were completed (median age 34 years; 47% female; 49% white; 20% Asian; 20% Hispanic; 15% Black or African). While 56% reported knowing little to nothing about AI in healthcare, 72% were comfortable with AI identifying potential SHD, increasing to 94% when AI recommendations were further reviewed by a physician. Further, 84% expressed comfort with being contacted after their ED visit to schedule an ultrasound recommended by AI. Factors most strongly associated with acceptance included physician review (97%), sharing test results (95%), and model accuracy (93%) whereas disclosure of the use of AI was less influential (72%). Notably, 79% believed AI would make healthcare somewhat better or much better, though acceptance varied by age and race/ethnicity. Among 52 clinicians surveyed, 75% reported they would trust an AI-ECG tool’s recommendations. While 85% believed an AI-ECG tool would reduce missed SHD diagnoses, only 63% were confident it would improve outcomes. Accuracy and workflow integration ranked as the highest priorities for adoption; however, concerns regarding patient follow-up (95%) and medicolegal liability (86%) were common. CONCLUSIONS Acceptance of AI-enabled SHD screening was broadly favorable among both ED patients and clinicians but consistently conditional on physician oversight and implementation infrastructure. Patient concerns centered on trust and transparency, while clinician concerns centered on follow-up capacity and medicolegal accountability. This suggests that barriers to deployment are organizational frameworks rather than technical acceptance, and that sustainable integration of an AI-ECG tool requires structured follow-up pathways, liability frameworks, and targeted patient communication strategies, particularly for populations with lower baseline comfort and limited access to longitudinal care.
Language: English
Subfield: Health Informatics
Field: Medicine
Domain: Health Sciences
Open Access status: green
OpenAlex ID: W7170901816