Clinical Validation of SMD-AFECG for Predicting New-Onset Atrial Fibrillation Within 2 Hours Using Single-Lead ECG Data

This clinical study is designed as a retrospective, single-center, single-arm, superiority confirmatory clinical performance study.

Pre-screening Electronic medical records and the VitalDB database at Seoul National University Hospital will be reviewed to identify potentially eligible data. The study will use continuous single-lead electrocardiogram data collected through VitalDB between September 1, 2022, and before September 30, 2025, from patients who were 19 years of age or older at the time of data collection.

Screening and Data Collection A screening number will be assigned sequentially to each potentially eligible case. Eligibility will be determined by reviewing the predefined inclusion and exclusion criteria. For eligible cases, continuous single-lead electrocardiogram data, signal loss, heart rate, the occurrence and timing of atrial fibrillation episodes, the electrocardiogram collection period, demographic information, previous history of atrial fibrillation, and pregnancy status will be collected. Directly identifying personal information will not be collected.

Test Dataset Creation Eligible data will be classified according to the presence or absence of a previous history of atrial fibrillation and the occurrence of an atrial fibrillation episode. The datasets will be categorized as new-onset atrial fibrillation positive or negative and repeated atrial fibrillation positive or negative. A subject number will then be assigned to each eligible dataset.

The planned sample size is 797 cases, including 462 cases for the evaluation of new-onset atrial fibrillation and 335 cases for the evaluation of repeated atrial fibrillation.

Reference Standard Establishment Two physicians with at least 5 years of relevant clinical experience will independently review the electrocardiogram data and relevant medical records. They will assess the suitability of the electrocardiogram data, confirm the presence or absence and timing of atrial fibrillation episodes, and verify the previous history of atrial fibrillation. Any disagreement between the physicians will be resolved according to the procedures specified in the study protocol. The final agreed classification will be used as the reference standard.

Application of the Investigational Device The medical device operator will receive the test electrocardiogram datasets labeled only with screening numbers and will not be provided with the reference-standard results or positive or negative group classifications. The blinded datasets will be analyzed using SMD-AFECG to generate the predicted risk of atrial fibrillation occurring within 2 hours. The software-generated results will be recorded in the electronic data capture system.

Statistical Analysis The software-generated predictions will be compared with the reference-standard classifications. The primary performance measures are the areas under the receiver operating characteristic curves for the prediction of new-onset atrial fibrillation and repeated atrial fibrillation. Two-sided 95% confidence intervals will be calculated using DeLong's method with a Wald-type confidence interval.

The performance criterion for new-onset atrial fibrillation will be met if the lower bound of the 95% confidence interval for the area under the curve is greater than 0.7753. The performance criterion for repeated atrial fibrillation will be met if the corresponding lower bound is greater than 0.8299. The study will be considered successful only if both primary performance criteria are met.

Secondary performance measures include the area under the precision-recall curve, precision, recall, F1 score, sensitivity, specificity, positive predictive value, negative predictive value, confusion-matrix results at different thresholds, and prediction time horizon.

Because this is a retrospective study using previously collected data, there will be no direct participant contact, additional examination, treatment, or change in clinical care.

Inclusion criteria

• Existing continuous single-lead electrocardiogram data collected through VitalDB at Seoul National University Hospital from September 1, 2022, to before September 30, 2025
• Participant aged 19 years or older at the time of electrocardiogram data collection
• Availability of an electrocardiogram dataset meeting one of the following criteria:

• No previous history of atrial fibrillation and a first atrial fibrillation episode, with continuous atrial fibrillation-free electrocardiogram data for the 2 hours before onset and electrocardiogram data for 1 hour after onset (NOAF-positive group)
• No previous history of atrial fibrillation and no atrial fibrillation episode, with continuous atrial fibrillation-free electrocardiogram data for the 2 hours before a selected index time and electrocardiogram data for 1 hour after the index time (NOAF-negative group)
• Previous history of atrial fibrillation and an atrial fibrillation episode, with continuous atrial fibrillation-free electrocardiogram data for the 2 hours before onset and electrocardiogram data for 1 hour after onset (RAF-positive group)
• Previous history of atrial fibrillation and no atrial fibrillation episode, with continuous atrial fibrillation-free electrocardiogram data for the 2 hours before a selected index time and electrocardiogram data for 1 hour after the index time (RAF-negative group)
• If atrial fibrillation occurred within 2 hours after the start of electrocardiogram recording, availability of electrocardiogram data from the start of recording through up to 1 hour after atrial fibrillation onset, with a minimum recorded electrocardiogram duration of 10 minutes

Exclusion criteria

• Electrocardiogram data considered to be of inadequate quality because of noise, poor equipment handling, unclear signals, artifacts, or similar problems
• Electrocardiogram data with signal loss of 5 percent or greater
• Electrocardiogram data with a heart rate below 30 beats per minute or above 300 beats per minute
• Data collected from a participant who was pregnant or breastfeeding at the time of data collection
• Data previously used for training or validation of the artificial intelligence model
• Data considered unsuitable for this clinical study by the principal investigator