Predicting onset of atrial fibrillation based on ECG P-wave duration: results from the Copenhagen ECG Study

Abstract

Purpose: The PR interval has been shown to be of value in a current risk score for atrial fibrillation (AF). We aimed to determine the discriminative power of P-wave duration in risk prediction of AF. Methods: We included 301,572 individuals, corresponding to one-third of the population in the greater region of Copenhagen. These individuals had a digital electrocardiogram (ECG) recorded in a general practitioner's core facility from 2001-2011. Data on drug use, comorbidity, and outcomes were collected from Danish registries. Individuals were categorized into seven risk groups based on percentiles of P-wave distribution. The association between P-wave duration and the risk of AF was evaluated with the use of Cox regression adjusted for the following covariates that were obtained at the time of study inclusion: age, gender, hypertension, myocardial infraction, heart failure, valvular heart disease, diabetes, hyperthyroidism, heart rate, PR interval, and treatment with beta-blockers or calcium-antagonist. C-statistics was estimated based on Cox regression models. Results: The median follow-up time was 5.5 years. During follow-up, 11,268 individuals developed AF. We found a J-shaped association between P-wave duration and the risk of AF. Having a P-wave below 5th percentile (≤89ms) conferred an increased risk of AF as evidenced by a hazard ratio (HR) of 1.55 (95% CI 1.38-1.75, P<0.001) compared with individuals in the risk group with the lowest risk of AF (reference group; P-wave between 5th to 20th percentile, 90-97ms). The risk of AF increased in a dose response manner from the reference group and upwards reaching a HR of 3.01 (95% CI 2.73-3.32, P<0.001) for having a P-wave above 95th percentile (≥130ms). Adding the P-wave risk categories to a prognostic model containing the above mentioned covariates except for PR interval significantly improved C-statistics from 0.680 (95% CI 0.675-0.685) to 0.711 (95% CI 0.706-0.716, P<0.001). For comparison, adding PR interval to the same model improved C-statistics to from the 0.680 to 0.693 (95% CI 0.688-0.698, P<0.001). Adding P-wave duration to a risk model containing both the mentioned covariates and PR interval further improved C-statistics from the 0.693 to 0.715 (95% CI 0.710-0.719, P<0.001). Conclusions: In this large primary care population we were able to show a J-shaped association between P-wave duration and the risk of incident AF. P-wave duration performed better in risk prediction of AF than PR interval. More importantly, P-wave duration added predictive information on top of the predictive information already contained in PR interval.