Device Trial Radar · NCT07173673
A Prospective, Multi-center, Randomized Clinical Trial to Evaluate the Detection of Atrial Fibrillation Using Artificial Intelligence-Enhanced Electrocardiography (SmartECG-AFrisk) Compared With Usual Care in Patients With Suspected Atrial Fibrillation: DEEP-AF
Last checked: Nov 19, 2026Latest source record: Sep 15, 2025 (this record)Source: ClinicalTrials.govMethods & limitations
Status
Not yet recruiting
Enrollment
1,230
Planned (sponsor's estimate)
Start
Nov 2025
Primary completion
Nov 30, 2027
Study completion Nov 30, 2027
Sponsor
- Lead sponsor
- Yonsei UniversityAcademic, hospital or other
- Sponsor type
- Academic, hospital or other
- First posted
- Sep 15, 2025
Design
- Study type
- Interventional
- Phase
- Not applicable (typical for device studies)
- Allocation
- Randomized
- Model
- Parallel
- Masking
- None
- Primary purpose
- Diagnostic
- FDA-regulated device
- No
- Who can join
- All sexes, from 30 years
- Healthy volunteers
- No
Summary
As written by the sponsor on the registry.
"The DEEP-AF study is a prospective, multi-center, randomized clinical trial evaluating the effectiveness of an artificial intelligence-enhanced electrocardiography algorithm (SmartECG-AFrisk) for early detection of atrial fibrillation (AF) in adults with suspected AF but no prior diagnosis. A total of 1,230 participants will be enrolled across 13 centers in Korea and randomized 1:1 into standard care or AI-guided care arms. In the standard care arm, diagnostic evaluation follows clinical guidelines with symptom-based use of 12-lead ECG, Holter, or patch ECG. In the AI-guided arm, baseline 12-lead ECGs are analyzed using SmartECG-AFrisk to calculate an AF risk score. Participants are classified as high-risk (score ≥50) or low-risk (\<50), and monitoring strategies are determined accordingly, enabling targeted ECG monitoring for high-risk individuals. The primary objective is to compare the 6-month incidence of newly diagnosed AF between the two arms. Secondary endpoints include AF detection differences between risk groups, healthcare resource utilization per AF diagnosis, anticoagulation initiation rates, major clinical events (stroke, embolism, bleeding, mortality), and patient satisfaction. This study aims to demonstrate whether integrating AI-driven ECG risk stratification into routine care improves AF detection and optimizes healthcare resource use in real-world clinical practice.
Conditions
- Atrial Fibrillation
Interventions
Usual Care (General Practice)Other
Participants receive routine care based on current clinical guidelines. Symptom-driven evaluation is performed by physicians, including at least one diagnostic test within 6 months such as a standard 12-lead ECG, Holter monitoring, or patch ECG. The choice and frequency of monitoring are determined by physician discretion, reflecting real-world practice patterns.
AI-ECG Guided Care (SmartECG-AFrisk)Device
Participants undergo SmartECG-AFrisk analysis of baseline 12-lead ECGs recorded in sinus rhythm. The algorithm calculates an atrial fibrillation risk score, classifying participants as high-risk (score ≥50) or low-risk (\<50). Monitoring strategies are adapted accordingly: high-risk participants undergo targeted and potentially repeated ECG monitoring using 12-lead ECG, Holter, or patch ECG, while low-risk participants follow standard guideline-based care.
Primary outcomes
What the sponsor says the trial is designed to measure.
- New diagnosis of atrial fibrillation within 6 months. (confirmed by ECG, Holter, or patch ECG)Time frame: Baseline (randomization) to 6 months after enrollment. Event date is defined as the first ECG/Holter/patch ECG documenting AF during the 6-month follow-up.
Locations
1 site in 1 country. Site status is shown where the registry gives one.
Yonsei University College of Medicine
Seoul, South Korea
Contact details for sites and study staff are not shown here. To ask about taking part, use the contact information on the registry record.
For information only. Nothing on this page is investment, legal, tax or medical advice, and nothing here is an offer to sell or a solicitation of an offer to buy any security. The data comes from public government sources, which can be incomplete, delayed or wrong. Check the original record before relying on it.