Subject patients are above 20 in age who are hospitalized in our hospital or outpatients with arrhythmia symptoms after the clinical research approval. The sinus rhythm electrocardiogram at the time of the patient's participation in the study is put into the artificial intelligence prediction algorithm, and the risk stratification results are blinded and are not informed to both the research director and the subjects. For the low-risk group, after attaching the wearable electrocardiogram to the subject, the electrocardiogram recorded a week later is analyzed to confirm the occurrence of atrial fibrillation.

Subject patients are above 20 in age who are hospitalized in our hospital or outpatients with arrhythmia symptoms after the clinical research approval. The sinus rhythm electrocardiogram at the time of the patient's participation in the study is put into the artificial intelligence prediction algorithm, and the risk stratification results are blinded and are not informed to both the research director and the subjects. For the highrisk group, after attaching the wearable electrocardiogram to the subject, the electrocardiogram recorded a week later is analyzed to confirm the occurrence of atrial fibrillation.

It is a 9.2g wearable electrocardiogram device, mobiCARE, in the form of a patch, and the model name is MC200M.

The AI prediction algorithm classifies patients into high-risk and low-risk categories for predicting paroxysmal atrial fibrillation within a week, based on ECG recordings of those with normal sinus rhythm. The accuracy of the prediction will be assessed through the use of a wearable device that records occurrence of paroxysmal atrial fibrillation over the course of a week.

The artificial intelligence prediction algorithm categorizes patients into high-risk and low-risk groups when predicting paroxysmal atrial fibrillation within one week based on normal sinus rhythm ECG data. The AI prediction algorithm's performance is assessed based on the data obtained from the primary outcome, which involves confirming whether atrial fibrillation recorded through a week-long use of a wearable device. We will gauge the algorithm's effectiveness by evaluating its predictive abilities, encompassing sensitivity, specificity, positive predictive rate, negative predictive rate, and the F1 score.

The predictive capabilities of the artificial intelligence prediction algorithm in risk stratification will be compared to the risk stratification proficiency of the experts. Each expert will be required to answer a questionnaire consisting of 30 ECGs to classify them as high risk or low risk. The questionnaire is composed of three components:

Q1. Atrial fibrillation/flutter risk prediction based on normal sinus rhythm 12-lead ECG and participant's clinical data (Age, gender, comorbidities, laboratory result, EHRA Symptom Score, etc.). The laboratory result could include BUN/Cr, eGFR, liver function test, lipid profile test.

Q2. Further plan required for identification of atrial fibrillation/flutter. Q3. Decisive evidence of atrial fibrillation/flutter risk prediction. The evidence could include normal sinus rhythm 12-lead ECG or participant's clinical data.