Cardiovascular
Longitudinal assessment of the time-dependent association between cigarette smoking and incident atrial fibrillation in the Atherosclerosis Risk in Communities (ARIC) study Oluseye Ogunmoroti* Oluseye Ogunmoroti Ashely I. Naimi David Benkeser Elsayed Z. Soliman Lin Yee Chen Faye L. Norby Alanna M. Chamberlain Alvaro Alonso
Background: Prior research investigating the link between cigarette smoking and atrial fibrillation (AF) does not account for the complex history of cigarette smoking overtime and the influence of time-dependent confounding. This study estimates the longitudinal time-dependent association between cigarette smoking and AF.
Methods: The analytical study sample comprised 13,201 adults aged 45-64 years enrolled between 1987 and 1989 and followed up until 2019. Cigarette smoking status was measured at 7 visits by self-report questionnaire and categorized as current, former and never. Incident AF was identified by ECG, hospital discharge records and death certificates. We compared estimates from 2 cox regression models. The standard model estimated the association of baseline cigarette smoking with AF, adjusting for time-fixed confounders (Figure). The time-dependent model estimated the association of cigarette smoking at multiple visits with AF, adjusting for both time-fixed and time-dependent confounders.
Results: At baseline, 55% of participants were women with mean age (SD) of 54.1 (5.7). After a median (IQR) follow-up of 28 (20, 31) years, 3,094 AF cases (23.4%) were identified. Compared with never smokers, estimates from the standard model showed that current and former smokers had an increased AF risk (HR: 1.77 [95% CI: 1.60, 1.95] and 1.12 [1.03, 1.23], respectively). While the time-dependent model showed an increased AF risk for current and former smokers, the magnitude of risk was lower (1.18 [1.12, 1.25] and 1.06 [1.02, 1.11], respectively).
Conclusion: The time-dependent cox model, accounting for the complex history of cigarette smoking and time-dependent confounding, showed an attenuated risk of AF among current and former cigarette smokers. These findings suggest that the estimates from the standard cox model may be biased. Further analysis using more advanced methods such as the parametric g-formula may be warranted to confirm and quantify the bias.
