Artificial Intelligence-enabled Large-scale Electrocardiogram Feature Extraction and Exploring Association Between the Extracted Features and Mortality, Stroke or Various Health Outcome of Interest

• In this study, large-scale ECG data (Electrocardiogram data of all patients stored in the MUSE system by measuring standard 12-guided ECG at Severance Health Checkup at Severance Hospital from November 1, 2005 to October 31, 2022) are combined with electronic medical records, National Health Insurance Corporation data, and National Statistical Office death cause data, and the artificial intelligence algorithm is used to extract ECG features to analyze the association between death, stroke, and various health conditions, and to conduct external verification or transfer learning using public databases (e.g., UK Biobank data).
• Intended to use a web-based artificial intelligence platform to distribute computational loads generated during large-scale data processing and improve analysis accuracy and efficiency.

Study Overview

• Status: Not yet recruiting
• Conditions: Health Status(Death, Stroke Etc)

Detailed Description

• All patient IDs obtained from the main office are replaced by research IDs (de-identified IDs), so the actual ID is not exposed and other personal identification information (name, resident registration number) is not collected.

• Research Methods:

  1. Electrocardiogram extraction based on the criteria of subjects.
  2. Combined with extracted ECG data and National Insurance Corporation data (+ National Statistical Office cause of death data).
  3. Health out of interest (HOI) definition. Includes death, stroke, etc.
  4. The defined HOI can be extracted from Yonsei Medical Center data or from National Insurance Service data or Statistics Korea's cause of death data.
  5. Artificial intelligence model training with electrocardiogram (and clinical information diagram if necessary) as input, utilizing supervised deep learning algorithms if there is a label and unsupervised learning algorithms if there is no label.
  6. Performance evaluation for supervised learning artificial intelligence models.
  7. In the case of unsupervised learning artificial intelligence models, the association/correlation between extracted features and HOI or predictability/detectability analysis.
  8. Transfer learning can be performed by adding external verification or dielectric data to the learned model using public databases.
  9. External verification can be performed using external additional data by mounting the learned model on a web-based artificial intelligence platform.
  10. Considering large-scale data, computing workloads can be distributed using web-based artificial intelligence platforms.
  11. The analysis results can be anonymized and the analysis results can be provided to researchers through a web-based artificial intelligence platform.

• Study Type: Observational
• Enrollment (Estimated): 3000000

Contacts and Locations

• Study Contact: Name: Hui-Nam Pak; Phone Number: 82-2-2228-8459; Email: hnpak@yuhs.ac

Study Locations

• Korea, Republic of
  • Seoul, Korea, Republic of
    • Yonsei University Health System
    • Contact: Hui-Nam Pak; Phone Number: 82-2-2228-8459; Email: hnpak@yuhs.ac

Participation Criteria

Eligibility Criteria

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

• Sampling Method: Non-Probability Sample

Study Population

All patients stored in the MUSE system after measuring a standard 12-guided electrocardiogram at Severance Health Checkup at Severance Hospital from November 1, 2005 to October 31, 2022

Description

Inclusion Criteria:

All patients stored in the MUSE system after measuring a standard 12-guided electrocardiogram at Severance Health Checkup at Severance Hospital from November 1, 2005 to October 31, 2022

Study Plan

How is the study designed?

What is the study measuring?

Primary Outcome Measures

Outcome Measure	Measure Description	Time Frame
Number of patients with Mortality	Investigating the reproducibility of mortality prediction (number of patients who died regardless of any cause) from ECG data measured within 1 year before death using artificial intelligence algorithms	2 years
Number of patients with Stroke, Atrial Fibrillation, Dementia	Investigating the correlation of certain patterns of ECG with the possibility of stroke/Atrial Fibrillation, Dementia using artificial intelligence algorithms.	2 years

Collaborators and Investigators

• Sponsor: Yonsei University
• Investigators: Principal Investigator: Hui-Nam Pak, Yonsei University

Study record dates

Study Major Dates

• Study Start (Estimated): December 1, 2023
• Primary Completion (Estimated): December 1, 2025
• Study Completion (Estimated): December 1, 2025

Study Registration Dates

• First Submitted: November 22, 2023
• First Submitted That Met QC Criteria: December 21, 2023
• First Posted (Actual): December 22, 2023

Study Record Updates

• Last Update Posted (Estimated): January 1, 2024
• Last Update Submitted That Met QC Criteria: December 22, 2023
• Last Verified: December 1, 2023

More Information

Terms related to this study

• Additional Relevant MeSH Terms: Cardiovascular Diseases; Vascular Diseases; Cerebrovascular Disorders; Brain Diseases; Central Nervous System Diseases; Nervous System Diseases; Stroke
• Other Study ID Numbers: 4-2022-1506

Plan for Individual participant data (IPD)

• Plan to Share Individual Participant Data (IPD)?: NO

Drug and device information, study documents

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