Methods

Our study drew participants from the Offspring Cohort seventh examination (1998 to 2001) and Third-Generation Cohort first examination (2002 to 2005) who additionally participated in the first examination of the multidetector computed tomography (MDCT) substudy from 2002 to 2005. The inclusion criteria for the MDCT substudy were: (1) subjects residing in New England at the time of the examination; (2) women ≥40 years and not pregnant and men ≥35 years; and (3) <160 kg body weight due to the weight restriction of MDCT scanner. We identified 3,453 eligible subjects who underwent MDCT imaging for the quantification of pericardial fat, intrathoracic fat, and abdominal visceral fat. We excluded participants based on the following exclusion criteria: (1) prevalent AF (n = 86); (2) missing follow-up information for incident AF (n = 2); (3) missing covariates (n = 48); (4) ≤45 years of age (n = 1,145) as having AF under age 46 years is uncommon and is characteristically different from having AF after the age of 45 years; and/or (5) previous coronary artery bypass graft surgery (n = 37). Compared with participants excluded from the analysis (n = 1,318), participants included in the analysis (n = 2,135) had higher mean age, body mass index (BMI), systolic blood pressure, and pericardial, intrathoracic, and abdominal visceral fat volumes; more likely to have diabetes and use antihypertensive medication; and were less likely to currently smoke, and have a history of previous heart failure and myocardial infarction ( Supplementary Table 1 ). The study protocol was approved by the Institutional Review Board of the Boston University Medical Center and Massachusetts General Hospital. All participants provided written informed consent.

The cases of AF, which included AF and atrial flutter, were identified based on the 12-lead electrocardiograms at rest recorded during the Framingham Heart Study examination or electrocardiograms and Holter monitors from the external hospital records and primary medical doctor records. All incident cases of AF were adjudicated by 2 Framingham Heart Study cardiologists.

Participants underwent MDCT imaging (LightSpeed Ultra, General Electric, Milwaukee, Wisconsin) in the chest and abdomen in the supine position for the assessment of pericardial fat, intrathoracic fat, and abdominal visceral fat using a protocol described elsewhere. Briefly, 48 contiguous 2.5-mm thick MDCT scans in the chest (120 kVp, 400 mA, temporal resolution 330) were acquired for the assessment of pericardial and intrathoracic fat. For abdominal VAT assessment, a total of 25 contiguous 5-mm thick slices of the MDCT scans in the abdomen (120 kVp, 400 mA, gantry rotation time 500 ms, table feed 3:1) were obtained. The state-of-the-art workstation tool (Aquarius 3D Workstation, TeraRecon Inc., San Mateo, California) was subsequently used to quantity the adipose tissue volumes from the MDCT images in a semiautomatic manner. The trained reader outlined the pericardium to differentiate the adipose tissue accumulated in the pericardial sac and thorax. The adipose tissue detected within the pericardial sac was defined as pericardial fat. Due to the technical limitations, precisely identifying different layers of pericardium through CT imaging was challenging. Accordingly, pericardial fat included epicardial fat that was contiguous to the myocardium and the fat accumulated outside the parietal layer of the serous pericardium. The adipose tissue quantified in the pericardium and in the thorax from the level of the right pulmonary artery to the diaphragm and the chest wall to the descending aorta was labeled as total thoracic fat. The differences in adipose tissue volume between the total thoracic fat and pericardial fat were defined as intrathoracic fat. Similarly, the abdominal muscular wall was manually outlined to differentiate and quantify abdominal subcutaneous and visceral adipose tissue. We previously reported the high reproducibility of pericardial, intrathoracic, and abdominal visceral fat measurements with interobserver and intraobserver reproducibility of ≥0.95.

Covariates were selected from the AF risk prediction model previously established by the Cohorts for Heart and Aging Research in Genomic Epidemiology-AF Consortium. The covariates were assessed at the Offspring cohort seventh examination (1998 to 2001) and Third-Generation cohort first examination (2002 to 2005). BMI was calculated using participants’ weight in kilograms and height in meters squared (kg/m ² ). Seated systolic and diastolic blood pressures at rest were measured. Diabetes mellitus was diagnosed as fasting plasma glucose ≥126 mg/dl or use of insulin or oral hypoglycemic medications. Clinic physicians asked a series of questions to ascertain participants’ smoking status and use of antihypertensive medications. Current smoking was dichotomized on the basis of smoking ≥1 cigarette per day within the year preceding the Framingham Heart Study examination. Events of heart failure and myocardial infarction were adjudicated by 3 physician investigators at the Framingham Heart Study based on available medical records.

Age- and gender-adjusted Pearson correlation coefficients among the MDCT-measured fat depots (pericardial, intrathoracic, and abdominal visceral fat) and BMI were computed. Age- and gender-adjusted AF incidence rates were acquired according to tertiles of MDCT-assessed fat depots. The MDCT-measured fat depots were standardized to a mean of 0 and standard deviation (SD) of 1 to enable the comparison among the effect sizes of the analysis. We constructed multivariable-adjusted Cox proportional hazard regression models to determine the association between pericardial fat, intrathoracic fat, and abdominal visceral fat with incident AF with a separate model performed for each association tested. For all Cox models, proportionality assumptions were validated. We derived hazard ratio (HR) and 95% confidence interval (CI) per 1-SD increment in the MDCT fat measurements. Two different covariate adjustments were considered: (1) age- and gender-adjusted model and (2) multivariable-adjusted model that accounted for gender and risk factors for AF, including age, systolic blood pressure, diastolic blood pressure, current smoking, the use of antihypertensive medication, diabetes mellitus, a history of heart failure, and a history of myocardial infarction.

As secondary analyses, additional covariate-adjusted models were constructed. We further adjusted the multivariable-adjusted model for (1) BMI or (2) abdominal visceral fat for pericardial and intrathoracic fat models. Moreover, all the ectopic fat depots were additionally entered in the same multivariable-adjusted model to explore the association among the different fat depots with incident AF. Tests for age and gender interactions were separately conducted based on the multivariable model.

We created age- and gender-adjusted cumulative AF incidence curves by the tertiles of MDCT-measured fat depots using adjusted method. Tertile 3 corresponded to higher ectopic fat volume, compared with tertile 1. A 2-sided p value less than 0.05 were considered statistically significant in all analyses. All the statistical analyses were performed using SAS, version 9.4 (SAS Institute, Cary, North Carolina). Given that we had 162 cases of incident AF, we had 80% power to discover an adjusted HR of 1.26 or larger at 0.05 significance level.