Simple risk model and score for predicting of incident atrial fibrillation in Japanese

Investigating regarding a predicted risk score of incident atrial fibrillation (AF) for an Asian general population has not been enough. Whether addition of electrocardiogram (ECG) variables to risk factors improves prediction of incident AF is unclear in a context that ECGs are extensively used at medical check-ups and outpatient clinics in Japan.

Participants undergoing periodic health check-ups during 2008-2014 followed-up by December 2015 including 96,841 (65.1% male) aged 40-79 years were pooled to derive prediction models and risk scores for incident AF. Multivariable Cox regression identified clinical risk factors associated with incident AF in 7 years among 65,984 eligible participants including 349 AF cases.

A 7-year prediction model ("Simple-model") including the variables of age, waist circumference, diastolic blood pressure, alcohol consumption, heart rate, and cardiac murmur, had good discrimination (C-statistic, 0.77), requiring no blood sampling. Addition model of the ECGs variables ("Added-model") including left ventricular hypertrophy, atrial enlargement, atrial premature contraction, and ventricular premature contraction, improved significantly the overall model discrimination (C-statistic, 0.78; categorical net reclassification improvement, 0.063; 95%CI, 0.031-0.099). The risk scores derived from the two models respectively showed an approximation of the observed and predicted probability for each score. Participants with score ≤4 or ≥9 points had, respectively, ≤1% and ≥5% predicted probability of incident AF in 7 years. The receiver-operating characteristics curve for the risk score of the added-model was significantly higher than the simple-model (0.769 vs 0.753, p<0.001). Atrial enlargement on ECG and the highest age group were the highest risk points of the significant predictors.

We developed 7-year risk scores for incident AF using usually available clinical factors including ECGs in primary care. These risk scores could identify individuals with high risk of incident AF at health check-up and outpatient clinics.