Introduction

Atrial fibrillation (AF) is the most common sustained cardiac arrhythmia, currently affecting >2 million people in the United States.¹ With an aging population, its prevalence is projected to double over the next 30 years.^2,3 The total health expenditures incurred by patients with AF are almost 5 times those of patients without AF, ranging between $6 and $26 billion.⁴ AF is also associated with considerable morbidity and mortality.⁵ Although the management of AF has received much attention, literature furthering our understanding of AF prevention remains limited.

Editorial see p 1821

Clinical Perspective on p 1834

Several studies have suggested an association between lifetime physical activity and the development of AF.^6,7 However, these studies used self–reported physical activity as a measure of physical function, which is not a direct measure of underlying physiology. The impact of cardiorespiratory fitness (CRF) on risk of AF has not been examined previously in large, multiracial cohort. In addition, it is not known if this relationship is influenced by incident coronary artery disease or incident left ventricular dysfunction, and the interactions between modifiable risk factors of AF (including obesity, hypertension, diabetes mellitus, smoking, and hyperlipidemia) and CRF and incident AF are also not well established.

Thus, we aimed to examine the relationship of CRF with the risk of incident AF among a large, multiethnic cohort of individuals undergoing a graded exercise treadmill test.

Methods

Study Population and Settings

The methods of the Henry Ford Exercise Testing (FIT) Project have been published previously.⁸ In short, the FIT Project is a retrospective cohort study of 69 885 patients who underwent physician-referred treadmill stress testing at affiliated hospitals and clinics across metropolitan Detroit, MI, within the Henry Ford Health System between 1991 and 2009. Data pertaining to exercise parameters, demographics, medications, and medical history were obtained by exercise physiologists and nurses and logged after the exercise test encounter. Patients with known AF (n=1763), heart failure (n=1565), or atrial flutter (n=78) and those with missing data (resting systolic pressure, 521; peak systolic pressure, 190; peak heart rate, 55; metabolic equivalents [METs], 826; total=1592) at the time of the stress test were also excluded from our analysis. The final sample used for the present analysis was 64 561. The FIT Project was approved by the Henry Ford Health System Institutional Review Board.

Graded Exercise Treadmill Testing

All patients underwent clinically indicated, symptom-limited treadmill stress testing following the standard Bruce protocol.⁹ Individuals who underwent the modified Bruce protocol or other non-Bruce protocols were excluded from the project. Resting heart rate and blood pressure were assessed manually immediately before each test. CRF was calculated by the Quinton treadmill controller on the basis of the elevation and speed achieved by the patient during the test and was expressed in METs. The treadmill speed was set initially at 1.7 mph and then increased to 2.5, 3.3, 4.2, 5.0, and 5.5 mph on minutes 3, 6, 9, 12, and 15, respectively. In the first 3 minutes, the grade was set at 10%, followed by a 2% increase every 3 minutes. The patient exercised for 3 minutes in each stage. Termination of testing was based on the American Heart Association/American College of Cardiology guidelines at the discretion of the supervising cardiologist. These included abnormal hemodynamic response, diagnostic ST-segment changes, significant arrhythmias, chest pain, shortness of breath, and unwillingness or inability of the patient to continue with the test. Hypertensive response to exercise was defined as peak systolic blood pressure during stress test of ≥190 mm Hg in women and ≥210 mm Hg in men.¹⁰ Chronotropic incompetence was defined as inability to achieve 85% of the predicted heart rate for a particular age. Predicted heart rate was calculated by 220 minus age. CRF was divided by cutoffs of <6, 6 to 9, 10 to 11, and >11 METs.

Clinical Characteristics

Demographics, body mass index, history of cardiovascular disease, and smoking status were obtained at the time of treadmill testing. Race and sex were self-identified by the patient. Obesity was based on the medical record or self-report. Diabetes mellitus was defined as a prior diagnosis of diabetes mellitus, use of hypoglycemic medications, including insulin, or a database-verified diagnosis of diabetes mellitus. Hypertension was defined as a prior diagnosis of hypertension, use of antihypertensive medications, or a database-verified diagnosis of hypertension. The blood pressure at the time of the test was not used to diagnose hypertension. Dyslipidemia was defined by prior diagnosis of any major lipid abnormality, use of lipid-lowering medications, or a database-verified diagnosis of hypercholesterolemia or dyslipidemia. Known coronary artery disease was defined as prior myocardial infarction, coronary angioplasty, coronary artery bypass surgery, or prior documented obstructive coronary artery disease (CAD) on a previous angiogram.

A family history of CAD was defined as a compatible history in a first-degree relative. A history of smoking was defined as current smoking obtained by self-report at the time of the stress test. Evidence of a sedentary lifestyle was established through a question survey, which asked participants whether they exercised regularly or not. History of active medication use at the time of stress test was obtained by the interviewing registered nurse or exercise physiologist. These medications were categorized into common indications such as antihypertensive medications and lipid-lowering medications. Use of inhaled medications for lung disease was considered a marker of chronic lung disease. By the same token, the use of thyroid medications was presumed to be a marker of underlying thyroid disease. A history of AF was defined as a previous clinical diagnosis of at least paroxysmal AF. Data on risk factors and medication use were supplemented and cross-checked by means of a retrospective search of the electronic medical record, as well as pharmacy claims filed by the patients. Left ventricular ejection fraction reported on echocardiography was also obtained but was available in only a subset of individuals (n=11 089). Left ventricular dysfunction was defined as a left ventricular ejection fraction <50% as measured by 2-dimensional echocardiography in the follow-up period. Risk factors were considered absent when they were reported as such at the time of stress testing or did not meet criteria for a database-verified diagnosis.

Patients with incident CAD were defined as those who were subsequently diagnosed with CAD, that is, a composite of nonfatal myocardial infarction, percutaneous coronary intervention, and coronary artery bypass grafting. Myocardial infarction and revascularization were ascertained through linkage with administrative claims files from services delivered by the system-affiliated group practice or reimbursed by the system’s health plan. Linkage was performed with the use of appropriate International Classification of Diseases, Ninth Revision (410.xx) codes and Current Procedural Terminology codes for percutaneous coronary intervention (92920–92944) and coronary artery bypass surgery. Data collection and subsequent chart documentation were performed by qualified nursing staff.

Outcome Ascertainment

The primary measured outcome of this study was new-onset AF. This diagnosis was ascertained by linkage with claims files from services delivered by the system-affiliated group practice or reimbursed by the system’s health plan. Linkage was performed by use of the appropriate International Classification of Diseases, Ninth Revision codes (427.31 for AF). A new diagnosis was considered present when the appropriate International Classification of Diseases, Ninth Revision code was identified in ≥3 separate follow-up encounters. Patients who developed isolated atrial flutter were not included in this outcome. International Classification of Diseases, Ninth Revision code 427.31 has been shown to be valid representation of AF (positive predictive value, 70%–95%; median, 79%).^8,11

Statistical Analysis

The characteristics of the study population were described at baseline across METs categories. We presented categorical variables as percentages and continuous variables as mean±SD. The trends across categories of METs were evaluated with linear (for continuous variables) or logistic (for categorical variables) regression.

To describe the observed incidence rate of AF across METs categories, we plotted cumulative incidence estimates using the Kaplan-Meier method. We compared differences between the curves using a log-rank test. The number of patients at risk was calculated at different time points. We also plotted 95% confidence intervals (CIs) of the curves. Because competing risk of mortality is a concern, we plotted competing risks of AF and mortality across time.¹² We tested differences between curves using the Fine and Gray method.¹³ Age-, sex-, and race-adjusted probabilities for risk of AF across METs were plotted for the high fitness group (>11 METs) because there is a known concern about a higher risk of AF at higher levels of METs.¹⁴

Hazard proportionality was confirmed for key variables via log-log plots and Martingale residuals. Nested Cox proportional hazards models were used to examine the association between peak METs and incident AF.

We adjusted analyses for 3 different models. Model 1 was adjusted for age, sex, and race (white, black, other). Model 2 was adjusted for model 1 covariates and history of hypertension, diabetes mellitus, sedentary habits, obesity, family history of CAD, smoking, hyperlipidemia, known CAD, hypertensive response, and chronotropic incompetence. Model 3 was adjusted for the covariates of model 2 and thyroid medications, digoxin, treated lung disease, β-blocker use, angiotensin-converting enzyme inhibitor use, angiotensin receptor blocker use, calcium channel blocker use, and use of all lipid-lowering medications, including statins. METs were examined as continuous and categorical groups. In addition, we examined the dose-response relationship between METs (as an independent variable) and the risk of AF (as dependent variable) using restricted cubic spline regression adjusted for model 3. Knots were placed at the 5th, 50th, and 95th percentiles of METs (4, 10, and 13 METs, respectively).^15,16

We also performed a sensitivity analysis examining the relationship between CRF and AF after adjustment for model 3 covariates and incident CAD as a time-dependent covariate. We also performed 3 subgroup analyses examining the relationship of CRF with AF in individuals who had subsequent left ventricular dysfunction (ejection fraction <50%), in individuals with CRF >10, and in individuals with established CAD (n=8495).

We assessed for effect modification by strata of age (<40, 40–49, 50–59, ≥60 years), sex, race (black, white, other), history of obesity, history of hypertension, history of smoking, history of diabetes mellitus, and history of hyperlipidemia. These findings were presented as a forest plot of hazard ratios (HRs) and 95% CIs. Because the interaction for obesity was significant (P=0.02), we examined the dose-response relationship between METs (as an independent variable) and the risk of AF (as a dependent variable) using restricted cubic spline regression adjusted for model 3 with knots placed at the 5th, 50th, and 95th percentiles of METs in obese and nonobese individuals. All analyses were performed with SAS version 9.3 (SAS Inc, Cary, NC) and the R package for competing risk analysis (R Foundation, Vienna, Austria; download available at http://www.R-project.org).¹⁷ Statistical significance was defined as P≤0.05 for main effects and interaction.

Results

The mean age of study participants without AF at baseline was 54±13 years. Compared with adults with excellent fitness (METs >11), adults with poor fitness (METs <6) were more often female (61% versus 20%) and black (40% versus 18%) and more likely to have hypertension (86% versus 44%), diabetes mellitus (33% versus 8%), hyperlipidemia (45 versus 39%), obesity (21% versus 9%), sedentary lifestyle (25% versus 17%), and known CAD (29% versus 5%; P for all <0.001; Table 1). The poor fitness group was less likely to have non-Hispanic whites (56% versus 73%).

Over a median follow-up period of 5.4 years (interquartile range, 2.8–8.6 years), there were 4616 new cases of AF. The unadjusted 5-year cumulative incidence rates of AF across categories of CRF (<6, 6–9, 10–11, and >11 METs) were 18.8%, 9.5%, 5.0%, and 3.7%, respectively. Time-to-event analysis was performed for categories of CRF and risk of AF with the Kaplan-Meier method (log-rank test, P<0.001; Figure 1). To observe competing risk of mortality, adjusted risks of AF and mortality were plotted against time (Figure I in the online-only Data Supplement). The risk curves were significantly different (P<0.001).

• Download figure
• Download PowerPoint

In a multivariable Cox regression model adjusted for potential confounders, we found a 7% reduction in the risk of incident AF for every 1 greater MET achieved (HR, 0.93; 95% CI, 0.92–0.94; P<0.001; model 3). The relationships of various categories of CRF with risk of AF are shown in Table 2. The association of model 3 covariates with incident AF is shown in Table I in the online-only Data Supplement.

In a sensitivity analysis, the relationship of CRF with AF after adjustment for incident CAD as a time-varying covariate in addition to model 3 covariates remained statistically significant (HR, 0.92; 95% CI, 0.91–0.93; P<0.001).

In a subgroup analysis of 2363 patients with left ventricular dysfunction (left ventricular ejection fraction <50%), CRF remained significantly associated with lower risk of AF (HR, 0.96; 95% CI, 0.94–0.99; P=0.01). When the analysis was restricted to only individuals with >10 METs, CRF remained significantly associated with lower risk of AF (HR, 0.94; 95% CI, 0.90–0.98; P=0.001). Findings remained consistent when analyses were restricted to individuals with known cardiovascular disease (Table 3).

In addition, we assessed for effect modification of the relationship between METs achieved and incident AF by strata of AF risk factors (Figure 2). We found that the magnitude of the inverse association between METs achieved and incident AF was greater for obese individuals. Age, sex, race, history of hypertension, smoking, or diabetes mellitus did not significantly alter the relationship between physical fitness and incident AF.

• Download figure
• Download PowerPoint

When cubic spline regression was used to explore the adjusted association between higher levels of CRF and risk of AF, we observed an approximately linear relationship, with HRs significantly >1 for CRF <10 METs (Figure 3). Because there was a significant interaction for obesity, we also examined risk of AF across METs for obese and nonobese individuals in a similar restricted cubic spline model adjusted for model 3 covariates of the Cox model. The plot showed a higher risk of AF for obese individuals at lower METs that rapidly approached and overlapped the risk of AF in nonobese participants at higher levels of METs (Figure II in the online-only Data Supplement).

• Download figure
• Download PowerPoint

We also explored age-, sex-, and race-adjusted predicted probabilities of AF in individuals with METs >10, which demonstrated an almost linear inverse relationship of CRF with AF (Figure III in the online-only Data Supplement).

Discussion

To the best of our knowledge, this is the first large report to examine the association between CRF as measured by exercise treadmill testing and risk of incident AF. In this analysis, we observed that higher levels of CRF were inversely associated with the lower risk of incident AF. We found a substantial dose-response relationship with a 7% lower risk for every 1-MET increase in CRF. This relationship was demonstrated over a median follow-up period of 5.4 years and appeared to be more pronounced in obese and older individuals. In addition, the relationship did not attenuate at higher fitness levels.

CRF is better referred to as a phenotype of physical function and is influenced by factors other than exercise training such as genetics.¹⁸ A closely related entity, self-reported physical activity, has been evaluated in many observational studies. In contrast, physical activity represents behavior and may not reflect one’s health such as CRF. Even though there might be fundamental differences between CRF and self-reported physical activity, the present study supports the thought that being fit is inversely associated with incident AF. In the Atherosclerosis Risk in Communities (ARIC) study, 14 219 participants with poor levels of physical activity (<1–149 min/wk of moderate physical activity or 1–74 min/wk of vigorous physical activity) had 1.56 times the risk of AF.¹⁹ Similarly, women enrolled in the Women’s Health Initiative (WHI) Study who achieved at least 7.5 MET·h/wk were found to be at a modestly lower risk of AF (HR, 0.86; 95% CI, 0.75–0.98; P=0.03).²⁰ However, this association did not remain significant after adjustment for BMI (a measure of obesity). The authors commented that BMI plays an important role in determining the relationship between physical activity and AF. In our study, the relationship between METS achieved and AF remained significant even after adjustment for obesity.

We found that the inverse association of CRF with AF was stronger in obese patients than nonobese patients. A similar interaction was observed in an analysis from the WHI, in which obese women had a lower risk of AF from increased levels of physical activity.²¹ Our study has important implications for obese individuals and suggests that obese individuals may potentially benefit by strategies that can improve CRF, provided that CRF is causally associated with AF. There are many plausible mechanisms for this interaction. Obesity is thought to promote AF by increasing left atrial size, which allows the development and propagation of re-entrant electric impulses.^22,23 Chronic stretch of pulmonary vein ostia owing to high intracardiac pressures in the obese is known to occur. These ostia act as a nidus for AF and are frequently ablated to cure AF.²⁴ Sleep apnea, which is closely related to obesity, is also an important risk factor for AF. The prevalence of AF in patients with sleep apnea is 6 times higher than that of the general population.²⁵ In a recent meta-analysis of 5 prospective studies, patients with obstructive sleep apnea were at more than twice the risk for AF (odds ratio, 2.38; 95% CI, 1.57–3.62; P<0.001) compared with patients without sleep apnea.²⁶ The mechanism underlying AF in obese patients and patients with sleep apnea is thought to be attributable to increased NADPH oxidase activity and dysfunctional nitric oxide synthase, resulting in increased oxidative stress in the atrial cardiomyocytes and hence AF.²⁷ Increased inflammation that occurs at a result of oxidative stress was also found to be associated with an increased risk of AF.^28–31 In this study, we observed a greater risk of AF among obese individuals with lower levels of CRF. With increasing levels of CRF, however, this risk gradually declined so that their risk of AF closely approached that of nonobese individuals. The relatively steep decline in AF risk suggests that CRF may be especially important for obese patients, although it must be restated that this is merely a hypothesis-generating analysis that requires further study.

Another important finding in our study was the inverse linear relationship between CRF and AF, which notably did not show evidence of plateauing or reversal at higher levels of CRF. This is contrast to previous studies in which extreme levels of fitness were found to be associated with an increased risk of AF. Indeed, a number of studies have suggested that extreme and sustained exercise likely promotes AF, particularly among elite endurance athletes.^32,33 Potential “cardiotoxicity” resulting from excessive participation in endurance exercise has been described by O’Keefe et al.¹⁴ Most epidemiological studies, however, have failed to show a higher risk for AF with exercise training in general populations, likely owing to an underrepresentation of highly trained athletes and resultant selection bias.³⁴ The mechanisms explaining the association between exercise and AF have not been fully elucidated, but inflammation, atrial remodeling, and autonomic imbalance are among those most commonly cited.^7,35–37 These mechanisms do not seem to be important in nonathletes, perhaps because of the less severe hemodynamic changes one would expect with moderate levels of exercise.¹⁸ Other potential reasons why we did not observe increased levels of AF among subjects with high levels of CRF may relate to selection bias (because elite athletes are rarely referred for exercise stress testing) and the fact that exercise stress testing is not a form of sustained endurance. These findings are merely exploratory and need further investigation in future studies.

Limitations

The present study has several important limitations. First, incident AF was identified by administrative claims files, which did not include direct evaluation of electrocardiograms. Therefore, a number of persons with undiagnosed AF and those with paroxysmal AF may have been missed as incident cases, attenuating our results. Second, our assessment of baseline physical fitness was based on a single measurement, preventing us from assessing changes in fitness over time or issues related to individual variability. Third, our study population comprised persons referred for stress testing, which undoubtedly carries a greater burden of cardiovascular disease at baseline than the general population and may have led to referral bias. Finally, residual confounding is always a concern with observational studies, especially with covariates assessed via self-report or the medical record rather than direct measurement.

Conclusions

Increased CRF is associated with a lower risk of AF independently of age, sex, or race and other traditional risk factors. This association was strongest in obese patients. This study provides further evidence in support of the benefits of higher overall fitness. Future studies are required to evaluate these findings and to assess whether an increase in fitness is associated with incident AF.

Acknowledgments

We acknowledge Zeina Dardari, MS, for statistical support.

Footnotes