INTRODUCTION

In the United States 25.4 million people are projected to have diabetes by 2011, rising to 37.7 million by 2031.1 Physiologic changes associated with diabetes include increased left atrial size2 and elevated C-reactive protein,3–6 a marker of chronic inflammation. Both findings are associated with heightened risk of atrial fibrillation,3,7–9 a common and serious arrhythmia that confers significant risks for stroke and death.10,11 Many epidemiologic studies have examined atrial fibrillation risk in relation to diabetes or elevated blood glucose, with conflicting results: ten studies observed an association,3,9,12–19 while nine did not.11,20–27 Often, prior studies examined many possible predictive factors and were not designed to evaluate the role of diabetes specifically.3,9,11,14,17,19,22–27 Many studies did not adjust for obesity, which is associated with increased risk of both diabetes and atrial fibrillation. 20,28,29 No study has examined the potential role of diabetes duration or glycemic control. We used data from a large case-control study of newly-recognized atrial fibrillation30 to examine the relationship between diabetes and risk of atrial fibrillation and to explore whether risk differs by diabetes duration and glycemic control. Our hypotheses were that diabetes would be associated with higher risk of atrial fibrillation and that risk would be higher for people with longer duration of diabetes or worse glycemic control.

RESEARCH DESIGN AND METHODS

Setting

This study took place at Group Health (GH), a large, integrated health care delivery system in the United States. Study procedures were approved by the GH Human Subjects Review Committee.

Population; Identification of Cases and Controls

We carried out a population-based case-control study of newly-recognized atrial fibrillation. This study is one of four case-control studies of cardiovascular outcomes with a shared control group.30–33 Because prior studies examined hormone replacement therapy and antihypertensive medications, the study population consists of women who were post- and perimenopausal and men with pharmacologically treated hypertension. These analyses include people age 30-84 with at least four GH visits (to ensure adequate data in the medical record). Women were defined as post-menopausal if their periods had stopped due to natural or surgical menopause. Treated hypertension was defined as having received a diagnosis of hypertension or high blood pressure and also receiving pharmacologic treatment. People were excluded if they had a pacemaker (which could interfere with identification of atrial fibrillation) or were missing data for body mass index (BMI; n = 72). These inclusion and exclusion criteria were ascertained by medical record review.

We used GH electronic data to identify all cases of atrial fibrillation occurring between October 1, 2001, and December 31, 2004 based on International Classification of Diseases, 9th revision (ICD-9) codes for atrial fibrillation or flutter from inpatient and outpatient encounters. Medical records were reviewed to validate the presence of atrial fibrillation and ensure that it was newly recognized. Perioperative atrial fibrillation was included only if it persisted until hospital discharge (10 cases).

A shared group of control subjects was selected from GH enrollment lists, frequency matched to the myocardial infarction cases (the largest case group in the four linked studies) by age, sex, calendar year, and hypertension status. For these analyses, we included only controls who had no history of atrial fibrillation that met study criteria.

Data Collection and Variable Creation

All subjects were assigned an index date, and data on covariates were collected only prior to the index date. For cases, the index date was the date of atrial fibrillation diagnosis, while controls were assigned a random date within that calendar year. Medical record review provided information on clinical characteristics including diabetes, hypertension, hypercholesterolemia, height, weight, valvular heart disease, congestive heart failure, myocardial infarction, angina, history of revascularization, and blood pressure. Telephone interview, completed by 51% of cases and 61% of controls, provided information about smoking, race, and alcohol intake. For people who did not complete the telephone interview, these data were obtained from medical records. Laboratory results and pharmacy data came from GH’s automated databases. Among GH members in this age range, 96% report filling all or almost all prescriptions through GH pharmacies.34

We categorized atrial fibrillation based on its persistence and duration using clinical data from the medical record. Our classification scheme was based on that of the American College of Cardiology/American Heart Association/European Society of Cardiology,35 with minor modifications required because of the data available to us. Transitory atrial fibrillation was defined as an episode lasting 7 days or less with no evidence of recurrence in the next 6 months, persistent/intermittent as lasting longer than 7 days or recurring but with sinus rhythm also present during the next 6 months, and sustained as lasting for at least 6 months with no evidence of sinus rhythm (similar to the ACC/AHA/ESC category of permanent).

We defined diabetes as present if there was a physician diagnosis in the medical record. Diabetes was classified as pharmacologically treated if the record indicated use of antihyperglycemic medications at the index date and as untreated if the person was not receiving medications. A similar approach was used to categorize hypertension and hypercholesterolemia. Diabetes duration was defined as time from the date of the first antihyperglycemic medication prescription to the index date. No other information on diabetes duration was available to us. Clinical coronary artery disease was defined as history of coronary artery bypass grafting, angioplasty, definite or probable angina, or hospitalized myocardial infarction. We calculated BMI as the most recent weight (kilograms) divided by the square of height (meters) and defined obesity as BMI ≥30 kg/m2.36

The GH computerized laboratory database, in existence since 1988, provided data on glycemic control. Prior to 1996, the GH laboratory measured total glycosylated hemoglobin, which we converted to equivalent hemoglobin A1c values using the formula 1.7 + 0.6*(total glycosylated hemoglobin) (equation developed by William C. Butts, PhD, and provided by Dr. Kim Riddell, GH laboratories). To examine long-term glycemic control, we calculated a simple linear time-weighted average of all values prior to index date.

Statistical Analysis

We performed descriptive analyses comparing characteristics of atrial fibrillation cases and controls. We also examined characteristics associated with diabetes among control subjects. Multivariable logistic regression was used to estimate the risk of newly-recognized atrial fibrillation associated with diabetes. All ORs and 95% CIs were calculated with adjustment for the matching variables of age, sex, calendar year, and treated hypertension. We selected additional covariates for final models using the change-in-estimate approach, including only those which altered the primary risk estimate by 10% or more. Variables considered were race, BMI, smoking, alcohol intake, number of visits in the prior year, valvular heart disease, claudication, chronic lung disease, hypercholesterolemia, blood pressure, total and HDL cholesterol, and use of specific antihypertensive agents. We did not include coronary artery disease and congestive heart failure because they may be in the causal pathway between diabetes and atrial fibrillation. Only BMI altered the main risk estimate by 10% or more, so final models were adjusted only for matching variables and BMI. We carried out sensitivity analyses adjusting for a large number of cardiovascular risk factors and diseases to explore the potential for residual confounding.

Because there were few people with untreated diabetes and our primary analyses indicated no association between untreated diabetes and atrial fibrillation, we excluded people with untreated diabetes from further analyses. We repeated our main analysis for subgroups defined by age, sex, and the presence of hypertension, obesity, hypercholesterolemia, coronary artery disease, or congestive heart failure. We used multiplicative interaction terms and the likelihood ratio test to determine whether the relationship between treated diabetes and atrial fibrillation differed significantly between subgroups. We also examined whether the association differed according to the duration and persistence of atrial fibrillation, using polytomous logistic regression to compare the three types of cases (transitory, persistent/intermittent, and sustained) to controls and assessing statistical significance with the Wald test.

Multivariable logistic regression was used to examine the association of two aspects of the exposure, duration of diabetes and glycemic control, with atrial fibrillation risk. We did this to investigate whether there was a dose-response relationship, hypothesizing that people with longer or more intense exposure to diabetes would have a higher risk of atrial fibrillation. The referent group was people without diabetes. Finally, in analyses limited to people with treated diabetes, we modeled duration and glycemic control simultaneously as continuous variables, adjusting each variable for the other.

Analyses were carried out using Stata/MP versions 9.2 and 10.1 for Windows (StataCorp LP, College Station, TX).

RESULTS

We identified 1,410 cases of newly-recognized atrial fibrillation during the study period. Five hundred and forty eight (39%) had transitory atrial fibrillation, 631 (45%) persistent/intermittent, 207 (15%) sustained, and 24 (1.7%) could not be categorized due to missing data. 2203 controls were selected from GH enrollment lists, stratified by age, sex, calendar year, and hypertension status.

Cases were slightly older than controls and more likely to have valvular heart disease, coronary artery disease, or chronic congestive heart failure (Table 1). Among controls, people with treated diabetes were younger and more often male compared to those without diabetes (Table 2). Controls with treated diabetes were also more likely to be obese and to have hypercholesterolemia, coronary artery disease, and congestive heart failure.

Table 1 Characteristics of People with Atrial Fibrillation (Cases) and Controls
Full size table
Table 2 Characteristics Associated with Diabetes Among Control Subjects
Full size table

Treated diabetes was present in 17.9% of cases and 14.1% of controls (OR 1.40, 95% CI 1.15-1.71, adjusted for age, sex, calendar year, treated hypertension, and BMI; Table 3). In sensitivity analyses, the OR was 1.45 (95% CI 1.16-1.81) after additional adjustment for race, smoking, hypercholesterolemia, prior visits, systolic blood pressure, use of ACE inhibitors or beta-blockers, and total and HDL cholesterol. Untreated diabetes was not associated with risk of atrial fibrillation (adjusted OR 1.04, 95% CI 0.75-1.45). The association between treated diabetes and atrial fibrillation was stronger in people who were obese (OR 1.64, 95% CI 1.27-2.12) than in those who were not (OR 1.10, 95% CI 0.80-1.52; p = 0.02 for the interaction) but did not differ substantially by sex, age, or the presence of treated hypertension, hypercholesterolemia, coronary artery disease, or congestive heart failure (all p values >0.10). The OR for transitory atrial fibrillation in relation to treated diabetes was 1.35 (95% CI 1.03-1.78); for persistent/intermittent atrial fibrillation, 1.36 (1.06-1.76); and for sustained atrial fibrillation, 1.71 (1.17-2.49; p = 0.51 for differences among the three case groups).

Table 3 Risk of Newly Recognized Atrial Fibrillation in Relation to Diabetes
Full size table

Duration of pharmacologic treatment for diabetes, which we examined as a surrogate for diabetes duration, was available for 91% of people with treated diabetes. The median duration was 8.2 years and was higher for cases than controls (9.6 versus 7.1 years). Figure 1 shows the risk of atrial fibrillation in relation to duration. Compared to people without diabetes, the adjusted OR for duration ≤5 years was 1.07 (95% CI 0.75-1.51); for >5 but ≤10 years, 1.51 (95% CI 1.05-2.16); and for >10 years, 1.64 (95% CI 1.22-2.20). Among people with treated diabetes, the adjusted OR per additional 1 year of duration was 1.03 (95% CI 1.01-1.06).

Figure 1
figure 1

Risk of newly-recognized atrial fibrillation by duration of treated diabetes. The diamond represents the adjusted OR and the vertical line, the 95% CI. ORs are adjusted for age, sex, calendar year, treated hypertension, and body mass index. People with no history of diabetes are the referent group.

Full size image

At least one measure of hemoglobin A1c was available for 99% of people with treated diabetes, and the median number of measures was 16 (IQR 10-25) over a median of 9.4 years. Compared to people without diabetes, increased atrial fibrillation risk was seen for average hemoglobin A1c above 7 (Fig. 2). Adjusted ORs, compared to people without diabetes, were 1.06 (95% CI 0.74-1.51) for average hemoglobin A1c ≤ 7; 1.48 (95% CI 1.09-2.01) for A1c above 7 but ≤8; 1.46 (95% CI 1.02-2.08) for A1c above 8 but ≤9; and 1.96 (95% CI 1.22-3.14) for A1c > 9. Among people with treated diabetes, the adjusted OR per 1 unit increase in hemoglobin A1c was 1.14 (95% CI 0.96-1.35).

Figure 2
figure 2

Risk of newly-recognized atrial fibrillation by average level of hemoglobin A1c. The diamond represents the adjusted OR and the vertical line, the 95% CI. ORs are adjusted for age, sex, calendar year, treated hypertension, and body mass index. People with no history of diabetes are the referent group.

Full size image

CONCLUSIONS

In this population-based case-control study, people receiving pharmacologic treatment for diabetes had 40% higher risk of developing atrial fibrillation than people without diabetes. Risk was higher with longer duration of treated diabetes, and there was a suggestion of higher risk with worse glycemic control.

Prior studies examining risk of atrial fibrillation in relation to diabetes or impaired glucose metabolism have yielded conflicting results.3,9,11–27 Our results are similar to those reported by several prospective studies,12,14,18 including the Framingham Heart Study,14 all of which reported ORs in the range of 1.4 to 1.6. Other studies found an association between higher blood glucose and increased risk of atrial fibrillation.9,17 However, numerous studies did not observe an association between diabetes and atrial fibrillation,11,20–27 and one study observed an association in women but not men (adjusted OR 1.26 [1.08-1.46] for women and 1.09 [0.96-1.24] for men).13 The discrepant findings may reflect methodologic differences including sample size, leading to low power in some studies, as well as the definition and method of ascertaining diabetes and atrial fibrillation. Unlike many prior studies, our primary focus was on the diabetes-atrial fibrillation relationship, and so we designed our statistical analyses accordingly. Unlike most prior studies, we adjusted for BMI, and so our findings support an independent association of diabetes with atrial fibrillation beyond the known association with obesity.20,28,29

A large proportion (39%) of our cases had “transitory” atrial fibrillation, defined as a single episode lasting 7 days or less without recurring in the ensuing 6 months. Many prior studies did not report information about duration or persistence of atrial fibrillation11,13,14,16,17,19,20,25–27 or used a classification scheme not directly comparable to ours.9,12,22 Based on their methods, some prior studies would have been more likely to detect permanent than paroxysmal atrial fibrillation.12,19,25,27 If diabetes conveys increased risk of paroxysmal atrial fibrillation and these cases were classified as not having atrial fibrillation in other studies, then our risk estimates may be higher than those from studies with such misclassification. We found that people with diabetes were at increased risk for all subtypes of atrial fibrillation, regardless of its duration or persistence.

Our study is the first to examine risk of atrial fibrillation in relation to diabetes duration and glycemic control. We observed higher risk with longer duration of pharmacologic treatment for diabetes (used as a surrogate for diabetes duration) and worse glycemic control. This pattern supports a dose-response effect, with greater exposure to hyperglycemia conferring increased risk. A dose-response effect is considered to support causality. Our risk estimates suggest higher atrial fibrillation risk for people with treated diabetes for more than 5 years or average hemoglobin A1c above 7.0. These findings could indicate a threshold effect, which is further supported by our finding that risk was not elevated among people with diabetes who were not receiving pharmacologic treatment (OR 1.04, 95% CI 0.75-1.45). Within GH, people with diabetes who are not receiving pharmacologic treatment probably are earlier in the course of diabetes or have milder disease than those receiving treatment. GH laboratory data confirm that the average hemoglobin A1c in the untreated group was lower than in the treated group. Still, our risk estimates have wide confidence limits, and we cannot rule out a clinically meaningful elevated risk for atrial fibrillation in people with diabetes who are not receiving pharmacologic treatment or have short duration of diabetes or excellent glycemic control.

In subgroup analyses, we observed an association between diabetes and risk of atrial fibrillation for people who were obese but no association in those who were not obese. This result should be interpreted cautiously because it was unexpected and arose from analyses examining many subgroups. It is possible that the physiology of diabetes differs by body mass index, with obese individuals having higher levels of insulin resistance and perhaps altered levels of some hormones compared to non-obese individuals. Further investigation of this potential interaction is warranted because it could shed light on the mechanism by which diabetes may lead to atrial fibrillation.

Several physiologic mechanisms could underlie a causal relationship between diabetes and atrial fibrillation.37 People with diabetes have higher levels of C-reactive protein,3–6 a marker of systemic inflammation, which may promote myocardial fibrosis and diastolic dysfunction. Diabetes is associated with left atrial enlargement2 which is thought to allow the development and propagation of reentrant electrical circuits. Diabetes also causes neural remodeling in the atria, including parasympathetic denervation and heterogeneous sympathetic denervation.38 People with diabetes are at higher risk of coronary artery disease and congestive heart failure, which may contribute to the development of atrial fibrillation. Finally, after accounting for obesity, people with diabetes have higher prevalence of obstructive sleep apnea,39 which may lead to atrial fibrillation.40

Our study has limitations. Our population consists of post- and perimenopausal women and hypertensive men and is predominantly white, which may limit generalizability. Because diabetes was identified in the course of routine clinical care, some people in the study probably had unrecognized diabetes. We identified only atrial fibrillation that came to clinical attention, so some transitory or asymptomatic cases may have been missed. It is possible that people with diabetes were more likely to have their atrial fibrillation recognized than people without diabetes, which could have resulted in bias. We lacked information about actual duration of diabetes and so we examined as a surrogate measure the duration of pharmacologic treatment for diabetes. Many people have diabetes for years before it is clinically detected. These factors may have led to misclassification of diabetes duration. We lacked information about some characteristics (e.g. physical activity and thyroid disease) and so there may be residual confounding.

In conclusion, we observed that pharmacologically treated diabetes was associated with a 40% increased risk of developing atrial fibrillation, and that risk was higher in people with longer duration of diabetes and worse glycemic control, providing support for a causal association. These results may increase understanding of the spectrum of cardiovascular disease associated with diabetes. Our findings suggest clinicians should have heightened suspicion for atrial fibrillation in people with diabetes, particularly those presenting with relevant symptoms. It may also be useful to screen for diabetes in people newly diagnosed with atrial fibrillation. In addition, given the high and rising prevalence of diabetes, our results have important implications for the future health burden of atrial fibrillation. Future research should investigate treatment approaches to reduce the risk of atrial fibrillation among people with diabetes.