Atrial fibrillation (AF) is the most common cardiac arrhythmia in the general population (1), and its prevalence is projected to increase substantially over the next decades (2). Individuals with AF have an increased risk of major complications, including death, stroke, and congestive heart failure, even after adjustment for relevant comorbidities ((3),(4),(5),(6),(7),8), such that defining and treating risk factors for AF occurrence is a major public health priority.

Several prospective studies evaluated the relationship between type 2 diabetes (T2D) and incident AF, and discrepant findings were reported ((9),(10),(11),(12),(13),(14),15). A recent meta-analysis found a modest, but statistically significant increased risk of new-onset AF among individuals with T2D (summary relative risk: 1.34 [95% confidence interval (CI): 1.07 to 1.68]) (16). However, significant between-study heterogeneity was observed, part of which might be explained by the fact that studies with more extensive multivariable adjustment were associated with smaller relative risk estimates than studies with less extensive adjustments (p for heterogeneity = 0.053) (16).

Given the close relationship of T2D with obesity and hypertension (17), which are 2 of the strongest risk factors for new-onset AF ((15),(18),(19),20), adequate adjustment in the evaluation of the association between T2D and AF is a relevant issue. The observed association between baseline T2D and AF may be due to differential changes of these risk factors in diabetic compared with nondiabetic subjects, because of the increased risk of weight gain and hypertension among patients with established T2D ((21),(22),23).

Few if any studies on the association between T2D and AF have taken into account changes of risk factors over time after the development of T2D ((9),(10),(11),(12),(13),(14),(15),(),). Because detailed knowledge of these relationships is important to optimize measures for AF disease prevention in the population, the primary aim of this study was to prospectively examine whether changes of major AF risk factors over time and intercurrent cardiovascular events may explain the previously observed association between T2D and AF.

All study subjects participated in the WHS (Women's Health Study), a completed randomized trial among 39,876 women evaluating benefits and risks of low-dose aspirin (100 mg every other day) and vitamin E (600 IU every other day) in the primary prevention of cardiovascular disease and cancer, using a randomized, double-blind, placebo-controlled study design. Details about the conduct of the study have been previously published ((26),(27),28).

Participants were female health professionals in the United States, age ≥45 years and free of cardiovascular disease, cancer, or other major illnesses at study entry. Randomized treatment ended on March 31, 2004, and all women were invited to participate in continued observational follow-up, which for the present analysis was truncated on March 2, 2011. Of the original cohort, we excluded 897 women (2.2%) with a history of AF at baseline and 54 (0.1%) women with a confirmed cardiovascular event (stroke, heart failure, or myocardial infarction) before study entry. In addition, 4,205 (10.5%) women did not participate in the observational follow-up, leaving 34,720 participants for the present analysis. The study was approved by the institutional review board of Brigham and Women's Hospital, Boston, and was monitored by an external data and safety monitoring board.

Mailed questionnaires were used to collect information on baseline characteristics. To obtain information on study outcomes, changes in covariates, and other information, follow-up questionnaires were sent to all participants every 6 months during the first year and every 12 months thereafter. Covariates of interest included age, hypertension, body mass index (BMI) (weight in kilograms divided by height in meters squared), hypercholesterolemia, physical exercise, self-reported race/ethnicity, highest education level, smoking, and alcohol consumption.

Details regarding the ascertainment of incident T2D in the WHS have been previously reported (29). Briefly, participants were asked annually whether and when they had been diagnosed with diabetes since baseline. Confirmation of T2D was conducted using American Diabetes Association diagnostic criteria (30). Self-reported cases were investigated by either telephone interview conducted by a physician or a self-administered supplemental questionnaire that inquired about symptoms, diagnostic testing, and use of diabetes medications. In a validation study (31), the self-administered questionnaire proved highly accurate for confirmation of clinical diabetes compared with medical record review. Thus, since 1999, confirmation of T2D has relied upon supplemental questionnaires. Overall, in 95% of all post-randomization self-reported diabetes, sufficient information for confirmation or disconfirmation of the endpoint was obtained. In addition, among women with available baseline levels of hemoglobin A_1c (HbA_1c) and without a baseline diagnosis of T2D, the prevalence of HbA_1c >6.5% was only 0.5%. Only confirmed cases of incident T2D were included in this study.

Details about AF endpoint confirmation have been previously published ((18),32). Briefly, participating women were asked about incident AF diagnoses at baseline, after 48 months, and annually thereafter. Women who participated in the continued observational follow-up who indicated the occurrence of an incident AF event on ≥1 yearly questionnaires were sent a supplemental questionnaire to collect additional information and to obtain written informed consent for medical record review. For all deceased participants who reported AF during the trial and extended follow-up period, family members were contacted to obtain consent and additional relevant information. An endpoint committee of cardiovascular physicians reviewed medical records for reported events according to predefined criteria. An AF event was confirmed if there was electrocardiographic evidence of AF or if a medical report clearly indicated a personal history of AF. Only confirmed AF events were included in the present study.

Collection of cardiovascular endpoints occurred through questionnaires, letters, and phone calls, as described previously ((8),26). A blinded endpoint committee of physicians adjudicated all events according to predefined criteria (26). Information on stroke and myocardial infarction was collected from the beginning of the study. Incident diagnoses of congestive heart failure were reported for the first time at the 48-month questionnaire. Heart failure was confirmed if participants met either the Framingham Heart Study (33) or the Cardiovascular Health Study (34) criteria for the presence of heart failure, and both definite and probable cases were included in the analysis.

Blood samples at study entry were obtained from 28,345 women. Levels of HbA_1c were estimated using the Tina-Quant turbidimetric inhibition immunoassay on a Hitachi 911 (Roche Diagnostics, Indianapolis, Indiana) autoanalyzer using packed red blood cells (35). The assay is specific for HbA_1c, standardized against the approved International Federation of Clinical Chemists reference method, and traceable to the Diabetes Control and Complications Trial by use of a conversion factor. Values of HbA_1c presented in this study are Diabetes Control and Complications Trial aligned. The coefficient of variation for HbA_1c computed from blinded simultaneously analyzed quality controls was 7.2%.

Wilcoxon rank-sum tests for continuous variables and chi-square tests for categorical variables were used to compare baseline characteristics of women with and without T2D at study entry. Person-years of follow-up were calculated from the date of return of the baseline questionnaire to the first occurrence of new-onset AF, death, loss to follow-up, or March 2, 2011.

Cox proportional hazards models were constructed to calculate hazard ratios (HRs) and 95% CIs for incident AF and to adjust for potential confounders. Age-adjusted models were further adjusted for adult height, smoking, exercise, alcohol consumption, education, race/ethnicity, and hypercholesterolemia. Because of the close relationship among BMI, hypertension, and T2D (17), these variables were added in separate steps to the multivariable models.

To evaluate the influence of covariate changes over time, we then constructed multivariable Cox models where T2D and all other covariates were updated during follow-up whenever follow-up values were available. We adjusted these multivariable models according to the same previously described pre-specified order, but added a final step where we additionally adjusted for intercurrent cardiovascular events, defined as stroke, myocardial infarction, or congestive heart failure. Finally, to assess the potential bidirectional nature of the relationship between T2D, hypertension, and BMI, we constructed a similar series of multivariable Cox models for BMI and hypertension, where T2D was added as a covariate in a separate step.

To gain further insights on the role of glucose homeostasis in AF development, we assessed the relationship between approximate quartiles of HbA_1c levels and incident AF in age- and multivariable-adjusted Cox models among women with available HbA_1c levels at baseline. Tests for linear trend were performed by assigning all women the quartile-specific median HbA_1c value. To evaluate the effect of elevated HbA_1c values, we repeated the same analyses after dichotomizing HbA_1c levels at the 95th percentile. Finally, we also specifically assessed the association between HbA_1c and incident AF among women with baseline T2D.

Categorical variables were entered into the Cox models using binary indicator variables. Multiplicative interaction terms between T2D and several baseline characteristics were evaluated in the fully adjusted baseline models using likelihood ratio tests. All analyses were performed using SAS version 9.2 (SAS Institute Inc, Cary, North Carolina). A 2-tailed p value <0.05 was pre-specified to indicate statistical significance.

Baseline characteristics stratified by the presence or absence of T2D are presented in (Table 1). At study entry, 937 women (2.7 %) had T2D. Women with T2D were significantly older, had a higher BMI, a higher prevalence of hypertension, a lower educational level, and exercised less frequently compared with women without T2D (all p < 0.0001).

During a median follow-up of 16.4 years (interquartile range: 15.6 to 16.8 years), 1,079 (3.1%) women developed incident AF. The age-adjusted incidence rate for new-onset AF was 3.97 and 1.99 per 1,000 person-years of follow-up among women with and without baseline T2D, respectively ((Table 2), upper section). Accordingly, in Cox proportional-hazards models, the age-adjusted HR for women with baseline T2D was 1.95 (95% CI: 1.49 to 2.56; p < 0.0001). Multivariable adjustment for established cardiovascular risk factors, but not BMI and hypertension, had little effect on this risk estimate (HR: 1.87: 95% CI: 1.41 to 2.47; p <0.0001). Although additional adjustment for hypertension and BMI substantially attenuated this relationship, a significant effect of baseline T2D persisted in the fully adjusted model (HR: 1.37; 95% CI; 1.03 to 1.83; p = 0.03). BMI and hypertension were strong and significant risk factors of incident AF in all models, and the effect of adding T2D to these models was small (6).

Among women without T2D at baseline, 2,730 (7.9%) developed new-onset T2D during follow-up. Sixty-eight (2.5 %) of these women had a subsequent diagnosis of new-onset AF, which corresponded to an age-adjusted incidence rate of 3.62 events per 1,000 person-years among women with new-onset T2D, as shown in (Table 2). In multivariable models that did not adjust for hypertension and BMI, T2D was significantly associated with incident AF (HR: 1.58; 95% CI: 1.30 to 1.93; p < 0.0001) (Table 2, lower section). When BMI and hypertension were added to this model, the association between T2D and new-onset AF was weakened and no longer statistically significant (HR: 1.19; 95% CI: 0.97 to 1.45; p = 0.10).

Among women with T2D, significantly more AF events were preceded by a cardiovascular event compared with women without T2D (16 of 124 [12.9%] vs. 39 of 955 [4.1%]; p < 0.001). After additional adjustment for these intercurrent events, the multivariable adjusted relative risk was further attenuated (HR: 1.14; 95% CI: 0.93 to 1.40; p = 0.20), as shown in (Table 2). Again, BMI and hypertension were strongly associated with incident AF in all models and adding T2D only minimally altered these results (6).

Subgroup analyses are displayed in (Table 3). The relationship between baseline T2D and incident AF was stronger among women age <65 years (HR: 1.54; 95% CI: 1.13 to 2.11) than among women age ≥65 years (HR: 0.85; 95% CI: 0.42 to 1.74) (p for interaction = 0.01). Again, in women age <65 years, the effect of T2D on incident AF was attenuated in fully adjusted time-updated Cox models (HR: 1.24; 95% CI 0.99 to 1.55). Consistent findings with nonsignificant interaction tests were obtained for all other subgroups assessed.

Among 28,345 women with available blood samples at baseline, 25,290 were eligible for the present analysis, of whom 24,890 (98%) had a valid HbA_1c measurement. In this subcohort, 835 confirmed incident AF events occurred. HbA_1c at study entry was not significantly associated with incident AF. The age- and multivariable-adjusted models for quartiles of HbA_1c are shown in (Table 4), and the p values for linear trend across quartiles were nonsignificant. When extreme levels above the 95th percentile (i.e., >5.6%) were examined, the age-adjusted association was significant (HR: 1.53; 95% CI: 1.20 to 1.96), but was highly attenuated and became nonsignificant after multivariable adjustment (HR: 1.13; 95% CI: 0.87 to 1.48). Finally, stratifying women by the presence or absence of T2D at baseline provided similar results. Among the 647 women with T2D at baseline, the multivariable HR for each 1% increase in HbA_1c was 1.09 (95% CI: 0.93 to 1.27). The corresponding HR among the nondiabetic women was 0.87 (95% CI: 0.71 to 1.07).

In this large, prospective cohort of initially healthy middle-aged women, baseline T2D was a modest, but statistically significant risk factor of incident AF after multivariable adjustment. Similar to previous studies on this issue, we found that the risk of incident AF among women with T2D was increased approximately 2-fold after adjustment for age, and that this risk was attenuated to about 1.4 after more extensive multivariable adjustment for baseline risk factors ((9),(12),(14),16). The present study added to this literature by showing that much of the remaining association between T2D and AF in these previous studies could be accounted for by the development over time of hypertension, obesity, and cardiovascular disease. We also found that the effect of T2D on incident AF was stronger in younger women (p for interaction = 0.01). The reason for this finding was unclear, but it might suggest that environmental factors leading to hypertension and obesity are less important in younger individuals with T2D.

Our data suggested that the adiposity and hypertension associated with T2D were more important for AF development than glucose homeostasis, a concept that was supported by at least 2 lines of evidence. First, we found no relationship between HbA_1c levels and incident AF in women with and without established T2D who had available blood samples. However, a recent study found that elevated HbA_1c levels were associated with an increased risk of AF occurrence in the subgroup of participants with established T2D (25). The reasons for these differential findings were unclear, but might be related to the fact that a more severe endpoint (AF hospitalizations) was assessed in the earlier study or that a different sample population was evaluated (community-based cohort vs. initially healthy female health professionals). Alternatively, the relatively small subgroup of women with T2D and available HbA_1c levels in our study might have limited our power to detect such an association. The upper limit of the 95% CI in our analysis suggested that our study could not exclude an increased risk of incident AF up to 27% for each 1% increase in HbA_1c among women with T2D, an order of magnitude well in line with the risk estimate of the previous analysis (25). In contrast, the tight 95% CIs (HR: 0.87; 95% CI: 0.71 to 1.07) for HbA_1c in the much larger group of nondiabetic women did confidently rule out even a modest increase in risk. Second, recent data from the Framingham Heart Study investigators showed that insulin resistance did not independently increase the risk of incident AF, with a multivariable adjusted HR for new-onset AF associated with insulin resistance of 1.03 (95% CI: 0.72 to 1.46) (36).

In contrast to these negative findings, the strong and independent relationships of obesity and elevated blood pressure, 2 important T2D-related comorbidities (17), with AF development were demonstrated in several independent studies ((18),(19),(20),). These relationships persisted in models that took into account changes of covariates over time ((18),19), and were only minimally affected in the present study by adding T2D to the models. Population attributable risks around 20% for overweight and obesity ((19),37), and up to 25% for elevated blood pressure (37), further highlighted the importance of overweight and elevated blood pressure as AF risk factors. Because patients with T2D are more likely to have additional weight gain and increases in blood pressure, taking into account changes in these risk factors over time was important to isolate any residual association attributable to diabetes independent of these factors. Interestingly, obesity, hypertension, T2D, and AF were all strongly related to blood levels of inflammatory biomarkers in earlier studies, such that inflammation might be an important pathophysiological link between these variables ((19),(38),39). With regard to clinical implications, our data suggest that prevention of AF among individuals with T2D should focus on the strict control of body weight and blood pressure (40).

The occurrence of cardiovascular events is a strong risk factor for incident AF ((14),37) and an important determinant of an adverse outcome in AF patients (8). Because patients with T2D have an increased risk of developing cardiovascular events (41), we hypothesized that intercurrent cardiovascular events could also mediate AF risk in these patients. However, although significantly more AF events were preceded by a cardiovascular event in women with T2D, the majority of AF events occurred independently of an overt cardiovascular event also among individuals with T2D. Accordingly, our multivariable models suggested that once changes in blood pressure and body weight were taken into account, only a small part of the relationship between baseline T2D and new-onset AF might be mediated by nonfatal cardiovascular events.

Strengths of the present study included the prospective design, the large sample size, and complete long-term follow-up with a large number of events confirmed by medical record review and regular covariate update. Potential study limitations should also be considered. First, most participants of the present study were Caucasian women, and generalization of our findings to different populations should be done cautiously. Second, some AF cases might have been undetected because of asymptomatic AF episodes. The only way to obtain a true picture of the AF prevalence would be by long-term continuous Holter monitoring, which unfortunately is not feasible in large long-term cohort studies of this type. Third, defining the onset of the initial AF episode might be challenging. Fourth, underdiagnosis of T2D might have occurred, given the lack of systematic T2D screening. Fifth, subclinical hyperthyroidism was repeatedly associated with an increased risk of AF and might also be related to T2D ((42),43). Because information on thyroid function was not available in this study, we were unable to assess its role in the relationship between T2D and incident AF.

In this large cohort of initially healthy women, T2D at study entry was associated with a modest increase in risk of incident AF. Most of this increased risk seemed to be mediated through excess weight gain and a higher prevalence of hypertension among individuals with T2D, such that the impact of diabetes beyond these risk factors was less important. Therefore, strategies for AF prevention in diabetic patients should focus on the control of T2D-associated comorbidities, in particular weight maintenance and blood pressure control.