ER-VISION-AI Study (ER-VISION-AI)

Multimodal Visual Language Model-Assisted Diagnostic Strategy in the Emergency Department: A Prospective Multicenter Randomized Controlled Trial (ER-VISION-AI Study)

Prospective, multicenter, randomized, open-label, blinded-endpoint (PROBE-like) clinical trial evaluating whether physician-supervised Generative Pre-trained Transformer (GPT)-assisted multimodal diagnostic support improves diagnostic concordance in emergency department patients presenting with acute cardiopulmonary symptoms.

Study Overview

• Status: Not yet recruiting
• Conditions: Chest Pain; Dyspnea; Emergency Department Patients; Acute Cardiopulmonary Disease
• Intervention / Treatment: Diagnostic test: Generative Pre-trained Transformer (GPT)-assisted multimodal visual language model (VLM) diagnostic support; Diagnostic test: Conventional emergency department diagnostic evaluation

• Study Type: Interventional
• Enrollment (Estimated): 1000
• Phase: Not Applicable

Contacts and Locations

• Study Contact: Name: Yeji Kim, PhD; Phone Number: +82-10-2724-7740; Email: lexie6169@gmail.com

Study Locations

• South Korea
  • Seoul, South Korea, 07804
    • Ewha Womans University Mokdong Hospital
    • Contact: Yeji Kim, PhD; Phone Number: +82-10-2724-7740; Email: lexie6169@gmail.com

Participation Criteria

Eligibility Criteria

• Ages Eligible for Study: Adult; Older Adult
• Accepts Healthy Volunteers: No

Description

Inclusion Criteria:

• Age ≥18 years
• Presentation to a participating emergency department with acute cardiopulmonary symptoms, including chest pain, dyspnea, palpitations, syncope, dizziness, or fever accompanied by cardiopulmonary symptoms
• Performance of both a standard 12-lead electrocardiogram and chest radiography during the initial emergency department evaluation
• Availability of initial clinical assessment, vital signs, laboratory findings, and all mandatory clinical information required for the multimodal AI workflow
• Expected emergency department observation or hospital admission for at least 24 hours
• Ability and willingness to provide written informed consent

Exclusion Criteria:

• Inability or refusal to provide written informed consent
• Requirement for immediate life-saving intervention that precludes completion of the study workflow
• Death before completion of the initial emergency department diagnostic assessment
• Electrocardiographic quality insufficient for reliable physician or Artificial intelligence (AI) interpretation
• Chest radiographic quality insufficient for reliable physician or Artificial intelligence (AI) interpretation
• Cardiac pacing rhythm
• Missing mandatory clinical information required for the multimodal Artificial intelligence (AI) workflow
• Previous enrollment in the ER-VISION-AI trial
• Inability to establish a blinded adjudicated reference diagnosis

Study Plan

How is the study designed?

Design Details

• Primary Purpose: Diagnostic
• Allocation: Randomized
• Interventional Model: Parallel Assignment
• Masking: None (Open Label)

Number of Arms

2

Arms and Interventions

Participant Group / Arm	Intervention / Treatment
Experimental: GPT-assisted multimodal visual language model (VLM) diagnostic strategy; Participants receive physician-supervised Generative Pre-trained Transformer (GPT)-assisted multimodal diagnostic support integrating electrocardiography, chest radiography, structured clinical information, laboratory findings, vital signs, and relevant medical history. Treating physicians remain responsible for all diagnostic and therapeutic decisions.	Diagnostic test: Generative Pre-trained Transformer (GPT)-assisted multimodal visual language model (VLM) diagnostic support; A Generative Pre-trained Transformer (GPT)-based multimodal visual language model integrates electrocardiograms, chest radiographs, structured clinical information, laboratory findings, vital signs, and relevant clinical history to generate diagnostic suggestions and differential diagnoses for physician-supervised clinical decision support.
Active Comparator: Conventional physician-guided diagnostic strategy; Participants undergo standard emergency department diagnostic evaluation according to routine clinical practice without Generative Pre-trained Transformer (GPT)-assisted diagnostic support.	Diagnostic test: Conventional emergency department diagnostic evaluation; Routine emergency department diagnostic evaluation performed according to standard clinical practice without AI-assisted diagnostic support.

What is the study measuring?

Primary Outcome Measures

Outcome Measure	Measure Description	Time Frame
Diagnostic concordance between the final emergency department diagnosis and the blinded adjudicated reference diagnosis established at hospital discharge.	Diagnostic concordance between the treating physician's final emergency department diagnosis and the blinded adjudicated reference diagnosis based on the prespecified principal diagnostic category.	During the index hospitalization, up to hospital discharge (average 3 days)

Secondary Outcome Measures

Outcome Measure	Measure Description	Time Frame
Diagnostic concordance after Generative Pre-trained Transformer (GPT)-assisted diagnostic support	Diagnostic concordance between the physician's final emergency department diagnosis after Generative Pre-trained Transformer (GPT)-assisted multimodal diagnostic support and the blinded adjudicated reference diagnosis in participants assigned to the intervention group.	During the index emergency department visit (average 6 hours)
Time from emergency department presentation to final diagnosis	Time required from emergency department presentation until establishment of the physician's final emergency department diagnosis.	During the index emergency department visit (average 6 hours)
Diagnostic reclassification after Generative Pre-trained Transformer (GPT)-assisted evaluation	Frequency of changes between the physician's initial working diagnosis and the final emergency department diagnosis after review of Generative Pre-trained Transformer (GPT)-generated diagnostic recommendations.	During the index emergency department visit (average 6 hours)
Physician diagnostic confidence	Physician-reported diagnostic confidence recorded before and after Generative Pre-trained Transformer (GPT)-assisted diagnostic support using the prespecified study assessment scale.	During the index emergency department visit (average 6 hours)
Physician acceptance of Generative Pre-trained Transformer (GPT)-generated diagnostic recommendations	Frequency of physician acceptance, modification, or rejection of Generative Pre-trained Transformer (GPT)-generated diagnostic recommendations in the intervention group.	During the index emergency department visit (average 6 hours)
Emergency department disposition accuracy	Accuracy of emergency department disposition decisions, including discharge, hospital admission, or intensive care unit admission, compared with the adjudicated reference diagnosis.	Up to hospital discharge (average 3 days)
Emergency department length of stay	Length of stay in the emergency department measured from patient presentation until emergency department discharge or hospital admission.	Up to hospital discharge (average 3 days)
Hospital length of stay	Total duration of hospitalization from admission until hospital discharge.	Up to hospital discharge (average 3 days)
In-hospital mortality	All-cause mortality occurring during the index hospitalization.	Up to hospital discharge (average 3 days)
30-day all-cause mortality	All-cause mortality occurring within 30 days after the index emergency department visit.	30 days
30-day emergency department revisit	Revisit to any emergency department for any cause within 30 days after the index emergency department visit.	30 days
30-day hospital readmission	Hospital readmission for any cause within 30 days after discharge from the index hospitalization.	30 days

Collaborators and Investigators

• Sponsor: Ewha Womans University Mokdong Hospital
• Collaborators: Ewha Womans University Seoul Hospital

Publications and helpful links

General Publications

• Topol EJ. High-performance medicine: the convergence of human and artificial intelligence. Nat Med. 2019 Jan;25(1):44-56. doi: 10.1038/s41591-018-0300-7. Epub 2019 Jan 7.
• Lopez-Puerta JM, Fernandez-Marin MR, Martin Benlloch JA, Lorente R. Spinal osteoid osteoma recurring as an aggressive osteoblastoma. Neurocirugia (Engl Ed). 2020 May-Jun;31(3):146-150. doi: 10.1016/j.neucir.2019.06.002. Epub 2019 Sep 2. English, Spanish.
• ANCA-associated vasculitis. Nat Rev Dis Primers. 2020 Aug 27;6(1):72. doi: 10.1038/s41572-020-0212-y. No abstract available.
• Kim TH, Kim CH, Choi SG. Radiation-induced angiosarcoma (RIAS) of the maxilla: a case report. J Korean Assoc Oral Maxillofac Surg. 2020 Aug 31;46(4):288-291. doi: 10.5125/jkaoms.2020.46.4.288.
• Li R, Chen X, Wang Y. Adverse events analysis of Relugolix (Orgovyx(R)) for prostate cancer based on the FDA Adverse Event Reporting System (FAERS). PLoS One. 2024 Oct 22;19(10):e0312481. doi: 10.1371/journal.pone.0312481. eCollection 2024.
• Asravor RK. Uncovering the forgotten story of the impact of Human Immunodeficiency Virus/Acquired Immunodeficiency Syndrome on economic growth in Ghana: A gender analysis. Int J Health Plann Manage. 2023 Sep;38(5):1495-1509. doi: 10.1002/hpm.3675. Epub 2023 Jun 23.
• Shakiba M, Nazemipour M, Mansournia N, Mansournia MA. Protective effect of intensive glucose lowering therapy on all-cause mortality, adjusted for treatment switching using G-estimation method, the ACCORD trial. Sci Rep. 2023 Apr 10;13(1):5833. doi: 10.1038/s41598-023-32855-3.
• Hsu HW, Chiu MC, Shoemaker D, Yang CS. Viral infections in fire ants lead to reduced foraging activity and dietary changes. Sci Rep. 2018 Sep 10;8(1):13498. doi: 10.1038/s41598-018-31969-3.

Study record dates

Study Major Dates

• Study Start (Estimated): January 1, 2027
• Primary Completion (Estimated): December 31, 2028
• Study Completion (Estimated): December 31, 2029

Study Registration Dates

• First Submitted: July 18, 2026
• First Submitted That Met QC Criteria: July 21, 2026
• First Posted (Actual): July 27, 2026

Study Record Updates

• Last Update Posted (Actual): July 27, 2026
• Last Update Submitted That Met QC Criteria: July 21, 2026
• Last Verified: July 1, 2026

More Information

Terms related to this study

• Keywords: Randomized controlled trial; Artificial intelligence; Emergency department; Electrocardiography; Chest radiography; Clinical decision support; Large language model; Multimodal AI
• Additional Relevant MeSH Terms: Pain; Neurologic Manifestations; Pathologic Processes; Disease Attributes; Respiratory Tract Diseases; Respiration Disorders; Signs and Symptoms, Respiratory; Pathological Conditions, Signs and Symptoms; Signs and Symptoms; Emergencies; Dyspnea; Chest Pain
• Other Study ID Numbers: ER-VISION-AI study

Plan for Individual participant data (IPD)

• Plan to Share Individual Participant Data (IPD)?: UNDECIDED
• IPD Plan Description: The investigators have not yet determined whether de-identified individual participant data (IPD), including the analyzable dataset and supporting documentation, will be shared with researchers outside the study team. A final decision will be made after completion of the study, taking into consideration institutional policies, participant privacy, ethical requirements, and applicable regulations.

Drug and device information, study documents

• Studies a U.S. FDA-regulated drug product: No
• Studies a U.S. FDA-regulated device product: No