Independent Contribution of Diabetes to Increased Prevalence and Incidence of Atrial Fibrillation Open Access

The study site was Kaiser Permanente Northwest (KPNW), a 480,000-member group-model HMO that uses clinical practice guidelines to assist clinicians with patient management. We used an observational longitudinal study design that capitalizes on the comprehensive medical utilization data maintained by KPNW, including an electronic medical record of all patient encounters, laboratory results that are analyzed by a single regional laboratory using standardized methods, and dispenses from pharmacies located in all clinics. The institutional review board of the Kaiser Permanente Center for Health Research reviewed and approved the study.

Since 1989, KPNW has maintained a registry of patients with known diabetes. The criteria for registry entry include an inpatient or outpatient diagnosis of diabetes (ICD-9-CM codes 250.xx), use of an antihyperglycemic drug, or at least two blood glucose test results above diagnostic levels for diabetes. For the current study, we selected 10,213 patients who were members of the diabetes registry as of 1 January 1999 and 7,159 patients who entered the registry by 31 December 2004 (total = 17,372), defining the study index date as either 1 January 1999 (for pre-1999 registrants) or date of registry entry. We then matched these patients to KPNW members without diabetes on year of birth and sex. For these nondiabetic subjects, we defined their index date as that of their matched diabetic subject. The prevalence of atrial fibrillation was calculated as of the index date. To conduct incidence analyses, we excluded patients with prevalent atrial fibrillation and 1,143 patients who had a history of stroke. The remaining patients were then followed until they died, left the health plan, or until 31 December 2008, whichever came first.

Patients were identified with prevalent atrial fibrillation if they had a diagnosis (ICD-9-CM 427.31) recorded in the medical record on or before their index date. Incident diagnoses were similarly identified any time after index until the end of follow-up among those patients who did not have prevalent atrial fibrillation. We also identified comorbidities present at baseline (ICD-9-CM code), including ischemic heart disease (410.x–414.x), history of stroke or cerebrovascular disease (430.x, 431.x, 432.x, 434.x, 435.x, 436.x, and 437.1), valvular disease (424.0–424.3), heart failure (428.x), hypertension (401.x–405.x), and history of depression (296.2x, 296.3x, 300.4, 309.1, and 311). Height, weight, and blood pressure were captured in the electronic medical record: for this study, baseline values of weight and blood pressure were the last measures before the index date. Baseline laboratory values, including LDL cholesterol, HDL cholesterol, triglycerides, serum creatinine (for estimation of glomerular filtration rate), and A1C (among patients with diabetes) were also those recorded latest on or before index date.

We conducted bivariate comparisons of baseline covariates among patients with and without diabetes using t tests for continuous variables and χ² tests for categorical variables. We calculated and compared age- and sex-adjusted incidence rates of atrial fibrillation per 1,000 person-years using regression for incidence densities. We then used Cox proportional hazards regression analysis to determine risk of incident atrial fibrillation for diabetes, controlling for other risk factors. All analyses were conducted with SAS (version 8.2; SAS Institute, Cary, NC).

Mean age of the study sample was 58.4 ± 11.5 years, and 51.2% were men (Table 1). Patients with diabetes were more likely to be nonwhite (11.9 vs. 6.4%, P < 0.0001), to have a higher BMI (33.6 ± 7.3 vs. 28.9 ± 5.7 kg/m², P < 0.0001), and to have higher blood pressure (136/80 vs. 132/79 mmHg, P < 0.0001). Comorbidities were more common among patients with diabetes, including ischemic heart disease (10.6 vs. 9.8%, P = 0.014), history of stroke (4.4 vs. 2.9%, P < 0.0001), hypertension (47.1 vs. 26.9%, P < 0.0001), heart failure (4.4 vs. 1.8%, P < 0.0001), and history of depression (15.8 vs. 11.7%, P < 0.0001).

The prevalence of atrial fibrillation was significantly greater among patients with diabetes (3.6 vs. 2.5%, P < 0.0001) (Fig. 1,A). Prevalence increased with age in both groups but was significantly exaggerated among diabetic patients. These same relationships were observed among men (Fig. 1,B) and women (Fig. 1 C). Although men had a higher prevalence of atrial fibrillation in all age-groups regardless of diabetes, the difference in prevalence between those with and without diabetes was greater in women than in men. For example, among men aged 65–74 years, atrial fibrillation was 30% more prevalent among patients with diabetes (7.9 vs. 6.1%, P = 0.037). Among women, atrial fibrillation prevalence was 61% higher in diabetic patients in the 65–74 years age-group (6.1 vs. 3.8%, P = 0.002).

Over a mean follow-up of 7.2 ± 2.8 years, the 16,057 diabetic patients without prevalent atrial fibrillation or a history of stroke at baseline developed atrial fibrillation at an age- and sex-adjusted incidence rate of 9.1 per 1,000 person-years (95% CI 8.6–9.7) compared with a rate of 6.6 (6.2–7.1) among the 16,471 nondiabetic patients (Fig. 2,A). Thus, the age- and sex-adjusted relative risk of atrial fibrillation was 38% (28–49%) greater among individuals with diabetes. Among men, the age-adjusted incidence of atrial fibrillation per 1,000 person-years was 10.8 (9.9–11.7) and 8.3 (7.6–9.1), respectively, for patients with and without diabetes (Fig. 1,B), producing a 30% (17–43%) greater relative risk among men with diabetes. Among women, age-adjusted atrial fibrillation incidence was 7.6 (6.9–8.4) and 5.0 (4.5–5.6), respectively (Fig. 1 C), with a higher relative risk attributable to diabetes of 52% (34–70%).

Table 2 displays the Cox regression models of time to development of atrial fibrillation. After controls were included for other known risk factors, patients with diabetes had a 16% greater risk of developing atrial fibrillation (hazard ratio [HR] 1.16 [95% CI 1.05–1.28]). However, atrial fibrillation risk associated with diabetes was substantially higher among women (1.26 [1.08–1.46]) and was not statistically significant among men (1.09 [0.96–1.24]). In addition, age ≥65 years tripled the risk of atrial fibrillation overall but had a greater effect among women, whereas BMI ≥30 kg/m² increased atrial fibrillation risk among men but was not a significant predictor in the women only model. White race, systolic blood pressure ≥140 mmHg, ischemic heart disease, valvular disease, hypertension, and heart failure were all significantly associated with atrial fibrillation risk in both sexes.

In this observational, age- and sex-matched cohort, longitudinal study of 34,744 patients with and without diabetes, we found that atrial fibrillation was 44% more prevalent and 38% more likely to develop when diabetes was present. Controlling for other known risk factors such as hypertension and heart failure approximately halved the risk of incident atrial fibrillation. However, this finding differed between men and women. Diabetes was not a significant predictor of atrial fibrillation among men after controlling for other risk factors but was highly significant among women.

Recent studies have provided evidence of how comorbidities can have important modulatory effects on the progression and complications of atrial fibrillation (15). Thus far, studies evaluating the specific role of diabetes have been lacking. In the recent development of a risk score for atrial fibrillation, the Framingham Heart Study did not find diabetes to be a significant predictor of atrial fibrillation risk (16). This finding is contrary to a previous Framingham report that found diabetes to be a strong independent risk factor (10). Consistent with the latter study, our findings suggest that in the present population, diabetes made a significant contribution to the prevalence and incidence of atrial fibrillation, independent of other established risk factors such as hypertension and congestive heart failure. Although this result needs to be confirmed in other populations, there may be important public health implications. The steadily rising prevalence of diabetes may further increase the prevalence of atrial fibrillation, already the most common arrhythmia diagnosis around the globe. Conversely, any efforts made to prevent and treat diabetes may reduce the burden of atrial fibrillation.

These findings also underscore the importance of understanding the mechanistic links responsible for this association. Published studies point to abnormalities of the autonomic nervous system as being an important mechanism in initiation and maintenance of atrial fibrillation. In observational clinical studies, initiation of atrial fibrillation episodes have been ascribed to variations in autonomic tone (17). Similar associations with altered autonomic tone have been made in conditions such as sleep apnea (18). This work has been extended in a canine model of atrial fibrillation in which researchers have observed that discharges from the stellate ganglia and vagal nerves often precede the onset of paroxysmal atrial fibrillation (19). In fact, targeted ablation of these latter structures appears to prevent the onset of paroxysmal atrial fibrillation in dogs, and such therapies are being evaluated in humans (20,–22). Analogous to the well-described peripheral autonomic neuropathy observed in diabetes, cardiac autonomic neuropathy is a serious but overlooked complication of diabetes (23,24). Taken together, these findings lend credence to the hypothesis that cardiac autonomic dysfunction may be an important mechanism for the increased propensity for atrial fibrillation in diabetes and warrants further careful study.

The increased predilection for atrial fibrillation among female diabetic subjects compared with male diabetic subjects is a novel finding that needs to be explored further. A prolonged intrinsic QT interval is a well-recognized sex-related difference in cardiac electrophysiological properties. Women have a longer intrinsic QT interval, a phenomenon that has been attributed to higher levels of estrogens affecting the cardiac repolarizing potassium ion channels (25). Although this finding is unlikely to affect the propensity for atrial fibrillation, there could be other, yet undiscovered, sex differences in atrial ion channel properties or the remodeling of the autonomic tone in female diabetic subjects that could account for the current observations.

An observational study has some inherent limitations. In a subgroup of patients, atrial fibrillation is an asymptomatic and paroxysmal phenomenon, such that it may escape detection unless it is observed incidentally. It is possible that patients with diabetes may be more likely to have electrocardiograms or ambulatory monitoring in the absence of symptomatic atrial fibrillation compared with nondiabetic patients. Conversely, diabetes is undiagnosed in ∼25% of individuals (1), suggesting that the prevalence of diabetes may have been underestimated in this population. If so, we may be underestimating the effect of diabetes on increased prevalence of atrial fibrillation in this population. Finally, unmeasured variables such as socioeconomic status, education, and alcohol intake could account for some of the differences we report.

In conclusion, in this population, diabetes was an independent determinant of atrial fibrillation prevalence and incidence. Prevalence was higher at baseline and there was an exaggerated increase of atrial fibrillation incidence over time in diabetic versus nondiabetic patients. Women with diabetes had an unexplained higher predilection for atrial fibrillation compared with men. These findings have potential public health implications, especially in women, and emphasize the need for further investigation of the mechanistic links between diabetes and atrial fibrillation.

S.S.C. is the Pauline and Harold Price Chair in Cardiac Electrophysiology Research at the Heart Institute, Cedars-Sinai Health System, Los Angeles, California.

No potential conflicts of interest relevant to this article were reported.

Parts of this study were presented in abstract form at the 30th annual meeting of the Heart Rhythm Society, Boston, Massachusetts, 13–16 May 2009, and at the 45th annual meeting of the European Association for the Study of Diabetes, 29 September–2 October 2009.