Author + information
- Received November 9, 2016
- Revision received February 27, 2017
- Accepted March 14, 2017
- Published online May 15, 2017.
- Morten Fenger-Grøn, MSca,∗ (, )
- Kim Overvad, MD, PhDb,c,
- Anne Tjønneland, MD, PhD, DMScd and
- Lars Frost, MD, PhD, DMSce
- aResearch Unit for General Practice and Section for General Medical Practice, Department of Public Health, Aarhus University, Aarhus, Denmark
- bSection for Epidemiology, Department of Public Health, Aarhus University, Aarhus, Denmark
- cDepartment of Cardiology, Aalborg University Hospital, Aalborg, Denmark
- dDanish Cancer Society Research Center, Copenhagen, Denmark
- eDepartment of Clinical Medicine, Aarhus University, Silkeborg Hospital, Denmark
- ↵∗Address for correspondence:
Mr. Morten Fenger-Grøn, Research Unit for General Practice, Aarhus University, Bartholins Allé 2, Aarhus C, Central Region 8000, Denmark.
Background Obesity is repeatedly emphasized as a risk factor for atrial fibrillation or flutter (AF). However, the underlying evidence may be questioned, as the obvious correlations between various anthropometric measures hamper identification of the characteristics that are biologically driving AF risk, and recent studies suggest that fat carries limited or no independent risk of AF.
Objectives This study sought to assess mutually adjusted associations among AF risk and height, weight, body mass index, hip and waist circumference, waist-to-hip ratio, and bioelectrical impedance-derived measures of fat mass, lean body mass, and fat percentage.
Methods Anthropometric measures and self-reported life-style information were collected from 1993 to 1997 in a population-based cohort including 55,273 persons age 50 to 64 years who were followed in Danish registers until June 2013.
Results During a median of 17 years of follow-up, 3,868 persons developed AF. Adjusted hazard ratios per population SD difference (HRs) showed highly statistically significant, positive associations for all 9 anthropometric measures (HRs ranging from 1.08 [95% confidence interval (CI): 1.05 to 1.12] for waist-to-hip ratio to 1.37 [95% CI: 1.33 to 1.42] for lean body mass). Pairwise mutual adjustment of the 9 measures left the association for lean body mass virtually unchanged (lowest HR: 1.33 [95% CI: 1.28 to 1.39] when adjusting for height), whereas no other association remained substantial when adjusted for lean body mass (highest HR: 1.05 [95% CI: 1.01 to 1.10] for height).
Conclusions Lean body mass was the predominant anthropometric risk factor for AF, whereas no association was observed for either of the obesity-related anthropometric measures after adjustment for lean body mass.
Atrial fibrillation is estimated to affect up to 1 in 4 persons during a lifespan (1,2), and it is associated with a series of adverse effects, such as lowered health-related quality of life (3), dementia (4), heart failure (5), stroke (6), or death (7), and their implied major human and health-economic costs (8,9). Furthermore, atrial fibrillation is one of the few heart diseases with increasing incidence (10). Thus, identification of modifiable risk factors is highly warranted.
Obesity is modifiable and is repeatedly emphasized as a risk factor for atrial fibrillation (11–17). This is supported by documented associations for a series of obesity-related anthropometric measures including weight, body mass index (BMI), waist and hip circumference, waist-to-hip ratio, fat mass, and fat percentage (11–13,15,18,19). However, associations with atrial fibrillation are also documented for height (13,20–22), and lately even for lean body mass in a large cohort with bioelectrical impedance-derived measures (23). This has raised suspicion that the commonly known association between obesity and atrial fibrillation is largely due to the obvious correlations between the various measures. The suspicion was confirmed by a recent study exploiting dual-energy x-ray absorptiometry measures of body fat and lean body mass in post-menopausal women (24) and, to some extent, by an even newer study on a study group of older adults including men (25), who are at markedly higher risk of atrial fibrillation (10).
In the present study, we used updated data from the Danish Diet, Cancer, and Health cohort (23) to depict the correlations between the 9 anthropometric measures, and to assess their associations with risk of atrial fibrillation or flutter (AF) under mutual adjustment. Particularly focusing on the complementary pair comprising fat mass and lean body mass, we further challenge the dogma of fat as the important independent driver of AF risk.
We performed a prospective cohort study in which persons recruited for the Danish Diet, Cancer, and Health study (26) were followed in Danish nationwide registers. Information on death or emigration was collected from the Danish Civil Registration System, which was established in 1968, and holds historical and updated electronic records for all persons in Denmark (27).
Information on hospital diagnoses was obtained from the Danish National Registry of Patients, which was founded in 1976. The registry includes dates of all admissions and discharges at nonpsychiatric hospitals in Denmark, as well as a primary discharge diagnosis and possibly 1 or more secondary diagnoses, as coded by the discharging physician (28). Since January 1, 1995, information from outpatient and emergency room visits has also been included. Until 1993, diagnoses were classified according to the Danish version of the International Classification of Diseases-8th Revision (ICD-8) and thereafter according to the national version of the ICD-10.
The Danish Diet, Cancer, and Health study cohort was established between December 1993 and May 1997 through invitation of 80,996 men and 79,729 women who were 50 to 64 years of age (26,29). Eligible cohort members were born in Denmark, were living in the Copenhagen or Aarhus areas, and had no previous cancer diagnosis according to the Danish Cancer Registry (30). For the present study, we excluded participants with an AF event before the date of their recruitment, according to the definition provided later in the text, as well as persons for whom lifestyle–related questionnaire data were missing or insufficient.
The primary exposures of interest were the anthropometric measures: height, weight, lean body mass (fat-free mass), and waist and hip circumference; as well as the derived measures: fat mass, fat percentage, BMI, and waist-to-hip ratio. Trained laboratory technicians at 2 study clinics in Aarhus and Copenhagen, Denmark, collected all anthropometric measures at the time of enrollment into the study. Fat and lean body mass were estimated using previously developed and validated sex-specific equations on bioelectrical impedance (31), as measured using a BIA 101-F device (Akern/RJL, Florence, Italy). Sensing electrodes were placed over the wrist and ankle, and current electrodes placed over the metacarpals or metatarsals on nonfasting participants lying relaxed, with legs approximately 45° apart and arms 30° from the torso. Further description of the measurements can be found elsewhere (23).
Comorbidity was assessed using date of first hospital contact that included a primary or secondary diagnosis for hypertension (ICD-8: 400 to 404, 410.09, 411.09, 412.09, 413.09, 414.09, 435.09, 437.00, 437.01, 437.08, 437.09, and 438.09; ICD-10: I10 to I15), diabetes (ICD-8: 249, 250; ICD-10: E10 to E14), ischemic heart disease (ICD-8: 410 to 414; ICD-10: I20 to I25), congestive heart failure (ICD-8: 425.99, 427.09, 427.10, 427.11, 427.19, 427.99, 428.99; ICD-10: I50), or mitral and/or aortic valve disease (ICD-8: 394 to 396; ICD-10: I05, I06, I08, I34, I35). Additional information on diabetes was drawn from the Danish National Diabetes Register (32). These data were supplemented with participants’ self-reports of hypertension, hypercholesterolemia, diabetes, and related medical treatment, as well as of menopausal status and hormone replacement therapy for women.
Likewise, we used participants’ questionnaire self-reports at recruitment to obtain data on educational level; physical activity during working hours and during leisure time (33); smoking habits; and intake of alcohol, specific foods, and nutrients, as calculated from a detailed semiquantitative food-frequency questionnaire (34,35).
The outcome of interest was an inpatient or outpatient contact registered with a first-time primary diagnosis of atrial fibrillation or flutter (AF), between which the present diagnostic classification system does not discriminate (ICD-8 codes 427.93 and 427.94; ICD-10 code I48). In a sensitivity analysis, we included secondary diagnoses as well. Prior validation has shown a positive predictive value above 90% for these diagnoses in the studied cohort (36). AF diagnoses reported only from emergency rooms have proven considerably less reliable (36), and were not included.
Baseline description of the cohort was done by sex and presented as percentages for discrete covariates, medians and 10th to 90th percentile intervals for continuous or pseudo-continuous covariates, and means and SDs for the 9 anthropometric exposures of interest. Correlations between these measures were assessed in terms of Pearson’s correlation coefficient and depicted by scatterplots, truncated and thinned out to prevent identification of individuals.
Associations between exposures and risk of AF were estimated in Cox proportional hazards regression models with age as the underlying time scale and delayed entry on the day of recruitment into the study (37). The observation time was ended by death, emigration, end of follow-up (June 30, 2013), or a hospital diagnosis of AF, whichever came first.
Associations and corresponding 95% confidence intervals (CIs) for the 9 studied anthropometric measures were reported as adjusted hazard ratios per sex-specific SD difference (HRs). In additional analyses of selected exposures, we explored possible nonlinear associations by 4-knotted restricted cubic spline regression.
All exposures were studied individually and under mutual pairwise adjustment in multiply-adjusted models that also allowed for: sex, baseline-registered smoking status, educational level, and physical activity (categorical variables) (Table 1); fruit and vegetable intake, alcohol consumption, and total energy intake (variables modeled applying 4-knotted restricted cubic splines on sex-specific deviation from the mean); and comorbidities (time-dependent). Among women, we also adjusted for hormone replacement therapy and menopausal status, as reported at baseline.
Focusing on the exposure pair comprising lean body mass and fat mass, we performed a series of sensitivity analyses by omitting selections of adjustment covariates. In additional analyses, we estimated independent associations for men and women.
The proportionality assumption of the Cox model was evaluated using tests on the basis of scaled Schoenfeld residuals, and by enriching the analysis models with an interaction term between the exposures and time. Clear violation of proportionality was seen for sex, for which adjustment was consequently performed by stratification. All analyses were performed using Stata version 12.1 (StataCorp LLC, College Station, Texas).
The Diet, Cancer, and Health study was approved by the Regional Committee on Health Research Ethics in Copenhagen and Aarhus, Denmark, and by the Danish Data Protection Agency. Written informed consent was obtained from all participants.
Of 57,053 persons accepting the invitation for the Diet, Cancer, and Health Study, 564 received a cancer diagnosis before baseline, 378 had registered AF, and 838 gave incomplete information on lifestyle factors. This left 55,273 participants for the present study, of whom 47.6% were men.
The detailed baseline characteristics of the cohort are presented in Table 1. The cohort was studied for up to 19.6 years (median 16.9 years), during which 3,868 participants developed AF (1,473 women and 2,395 men) with a median time to event of 11.5 years.
With only a few exceptions, the 9 studied anthropometric measures were all positively correlated with each other, generally with Pearson’s R >0.6 in both sexes (Online Figure 1).
As expected, highly significant, positive associations with risk of AF were seen for all 9 measures. Adjusted HRs ranged from 1.08 (95% CI: 1.05 to 1.12) for waist-to-hip ratio to 1.37 (95% CI: 1.33 to 1.42) for lean body mass (Central Illustration, top).
However, when also adjusting for lean body mass, no appreciable association was observed for any of the other anthropometric measures (Central Illustration, middle). Focusing on the upper limits of the CIs, we found that the strongest adjusted association compatible with the data was the HR of 1.10 observed for height (estimate: 1.05; 95% CI: 1.01 to 1.10).
The association for lean body mass remained virtually unchanged when adjusted for any of the 8 other anthropometric measures, despite strong mutual correlations. Even the widths of the CIs were only marginally affected. The lowest lower HR CI limit for the lean body mass association was 1.26, which was observed when adjusting for weight (Central Illustration, bottom).
On the chosen scales for the associations described in the preceding text, no differences between sexes were substantive, nor were they statistically significant (Online Figure 2). Table 2, which includes estimated associations for all possible pairs of exposures, revealed no similar patterns for any other anthropometric measure.
The observed incidence rates of AF for the 2 sexes developed slightly differently with age, but the estimated association for lean body mass accounted well for the difference in risk levels between men and women below 70 years of age (Figure 1).
Including secondary diagnoses for AF in the definition of the outcome slightly attenuated the association estimated for lean body mass (weakest association HR: 1.29; 95% CI: 1.22 to 1.37) observed under adjustment for weight), but the general pattern was unaltered (Online Table 1).
Applying flexible modeling for the complementary pair of measures comprising lean body mass and fat mass revealed limited deviations from linearity in both associations. Only small changes were seen when successively omitting adjustment for covariates to control for unintended overadjustment (Figure 2). For fat mass, the association presented a weak U-shape that was raised to significance when omitting adjustment for comorbidity, but it still showed a very modest slope compared with the slope for lean body mass.
Splitting the follow-up into periods before and after the mean time since inclusion (8.3 years) showed an anticipated tendency toward a slight attenuation of associations for the modifiable measures over the 20 years of follow-up. Still, the overall pattern was remarkably consistent. In the latter period, the associations for lean body mass ranged from HR: 1.29 (95% CI: 1.23 to 1.36), when adjusting for height, to HR: 1.38 (95% CI: 1.32 to 1.45), when adjusting for waist circumference, whereas height was the only measure for which we found any long-term association when adjusting for lean body mass (HR: 1.07; 95% CI: 1.02 to 1.13).
In this large, population-based prospective cohort study including 3,868 AF outcomes, we confirmed expected positive associations with AF risk for all 9 of the investigated anthropometric measures: height, weight, BMI, hip and waist circumference, waist-to-hip ratio, and bioimpedance-derived fat mass, lean body mass, and fat percentage. However, correlations between the measures were also confirmed.
Therefore, the key feature of this study was the systematic mutual adjustment of the anthropometric characteristics, although the yielded estimates for certain exposure pairs were, indeed, more suited for elucidation of methodological mechanisms than for clear biological interpretation.
This approach evoked a spectacularly simple pattern, which suggests that lean body mass was the predominant anthropometric driver of AF risk, whereas none of the traditional obesity-related measures proved to have any independent influence. The pattern was consistent between the sexes and over time. Flexible modeling of fat and lean body mass did not reveal any decisive underlying nonlinearity.
These results thus embrace the numerous observations of associations between obesity-related measures and AF risk, as well as the ambiguous earlier findings reported from the same cohort (23), and the recent findings in post-menopausal women of no independent association between obesity and AF (24).
Below 70 years of age, the estimated association for lean body mass accounted well for the observed and well-known, but not fully explained, AF incidence differences between men and women. Above 70 years of age, the sex difference predicted from lean body mass exceeded the observed difference, possibly because predisposing comorbidities have a more decisive influence. This fits well with the latest findings of a slightly less-marked association between lean body mass and AF among older people (25).
Study strengths and limitations
The major strength of the study was the power provided by the large number of cases and virtually complete follow-up over a long period of time. Still, several limitations must be carefully considered in the interpretation of the findings. Residual confounding is always a concern in observational studies. However, in this case, we regarded this as a minor issue as quite decent information for adjustments was available, and the adjustments still seemed to have very limited influence.
Change of baseline-assessed exposures during long follow-ups is another concern. In general, this entails a bias toward the null that is greatest for exposures with the largest temporal variability, and therefore, a larger downward bias on the estimated associations could be expected for the fat-related measures than for height and perhaps lean body mass. However, although we did find slight attenuation of associations over time, no changes were substantial. Besides, if the indication of a stronger influence from lean body mass than (directly) from fat was a product of variability bias, we would expect this pattern to be particularly marked in the last part of the study period; reassuringly, this was not observed. Still, it would have been interesting to have follow-up measurements of anthropometry, and perhaps even assessments of fat and lean body mass by methods other than impedance measurement, which may not be considered the most precise approach (24).
The use of follow-up data from Danish nationwide registries, which cover an entire nation’s free hospital system, ensured very complete registration of AF diagnoses for which a high positive predictive value is documented (36). However, underdiagnosis of AF is a generally recognized problem that carries a risk of detection bias. Most likely, persons with characteristics that are known to confer risk of cardiovascular disease will experience the most intense monitoring, which could result in overestimation of the AF risk associated with these characteristics. In itself, this can hardly explain the identification of lean body mass as the predominant anthropometric risk factor. Yet, in the main analyses, the outcome was defined by primary diagnoses for AF only, thus disregarding hospital contacts registered as primarily caused by other conditions. If several conditions were handled during the same hospital contact, this analytical approach could go beyond the reduction of upward bias, and cause downward bias for traditional cardiac risk factors. Therefore, the observation that the sensitivity analyses based on both primary and secondary AF diagnoses gave results very similar to the main analyses is important. We also chose to adjust for cardiac comorbidity as time-dependent covariates. This has potentially important implications for the interpretation of the estimated associations, particularly for the fat-related measures. Thus, the resulting estimates describe the possible direct influence of fat on AF risk, and disregard indirect influence of obesity on the risk of myocardial infarction and heart failure, which in turn may increase the risk of AF. Few would question the existence of such indirect causal pathways between fat and AF risk, which is also confirmed in the lower part of Figure 2, where no correction for comorbidity was included. Still, omitting this correction did not induce any major changes in the overall picture. An additional weakness in the AF data is that we were not allowed to distinguish between initial, paroxysmal, persistent, and permanent AF, and these conditions may not share the relevant etiology.
Finally, the studied cohort comprises one-third of a random sample collected in the area around the 2 largest cities in Denmark who responded. This sample may constitute a rather selected, healthy, and socioeconomically-privileged group (26). Obviously, the observed AF incidence and the distributions of the anthropometric characteristics are specific for the studied cohort and period; hence, the specific values estimated for the studied associations cannot be directly generalized to other settings. However, as the selection for the cohort took place before the beginning of the study period, it is unlikely to cause bias on the estimated risk associations within the cohort. If the estimated associations reflect biological mechanisms, it appears likely that they would be present in larger segments of the human population. Still, it must be considered that lean body mass could be a marker of other personal characteristics in urban, industrialized settings than in more rural or deprived areas.
Although we consider our results consistent and unambiguous, we recommend they be interpreted with caution and used to inspire future research, rather than to guide interventions.
First, firm conclusions on causality cannot generally be justified from observational designs. However, it appears plausible that a causal relation between lean body mass and AF is induced by a close relation between high lean body mass and left ventricular hypertrophy or left atrium enlargement, which are commonly acknowledged to provide favorable substrates for AF. Thus, similar to our findings for lean body mass, left atrium size has previously been reported to be closely correlated to obesity-related measures and yet to stand out as more important for AF risk than any of these measures (11). Unfortunately, our study did not include measurements of cardiac dimensions, and to our knowledge, the association between lean body mass and these more specific measures is not well studied in similar, healthy cohorts. Further links have been hypothesized between skeletal muscle activity and pulmonary vein ectopic AF trigger activity, but solid evidence is sparse (38). Overall, we believe that a causal mechanism behind the presented associations must be suspected, but is far from clear.
Second, if we assume that the estimated associations represent causal relations, manipulating anthropometric characteristics may still not induce risk changes similar to the differences observed for the natural variations. Even if it does, interventions aiming to minimize lean body mass, while allowing fat mass to “run wild,” would be against all reason in a broader health perspective, and they could not be justified even from a very narrow AF risk perspective. Additionally, such interventions appear to be most hypothetical. Thus, from a common-sense perspective, it seems plausible that high fat mass causes high lean body mass, and this is partly confirmed by the observed covariation between fat and lean body mass. Intriguingly, our results might, in fact, support traditional antiobesity interventions for targeting AF risk, because aggressive weight loss interventions are known to entail loss of lean body mass, which is unwarranted in the prevention of most conditions.
A further, intriguing consideration is whether high lean body mass represents the common denominator that may solve the apparent paradox that both obesity and physical exercise confer risk of AF (39). However, in this regard, the presented results can only serve as inspiration for more detailed pathophysiological studies.
This study identifies lean body mass as the predominant anthropometric risk factor for AF and questions the role played by fat in the AF etiology. Although the study provides no reason to advise against limiting fat mass, and possibly not even against maintaining a certain lean body mass, our findings may have substantial implications for the future understanding of this frequent and serious condition.
COMPETENCY IN MEDICAL KNOWLEDGE: When adjusted for other anthropometric measures, lean body mass is the predominant risk factor for AF, and when adjusted for lean body mass, obesity-related measures show no association with the risk of developing AF.
TRANSLATIONAL OUTLOOK: Further research into the etiology of AF should focus on factors related to lean body mass, rather than parameters related to obesity, and investigate whether loss of muscle mass along with fat mass reduces the risk of developing AF.
For a supplemental table and figures, please see the online version of this article.
The study was supported by the Danish Council for Strategic Research (grant 09-066965). The Diet, Cancer and Health cohort study was funded by the Danish Cancer Society. Mr. Fenger-Grøn is supported by a grant from the Faculty of Health, Aarhus University, Denmark, and from the Lundbeck Foundation, Denmark. Funders had no role in the design and conduct of the study; management, analysis, and interpretation of the data; preparation, review, or approval of the manuscript; or decision to submit the manuscript for publication. The authors have reported that they have no relationships relevant to the contents of this paper to disclose.
- Abbreviations and Acronyms
- atrial fibrillation or flutter
- body mass index
- confidence interval
- hazard ratio (per sex-specific standard deviation in the population)
- International Classification of Diseases
- Received November 9, 2016.
- Revision received February 27, 2017.
- Accepted March 14, 2017.
- 2017 American College of Cardiology Foundation
- Lloyd-Jones D.M.,
- Wang T.J.,
- Leip E.P.,
- et al.
- Dorian P.,
- Jung W.,
- Newman D.,
- et al.
- Shinbane J.S.,
- Wood M.A.,
- Jensen D.N.,
- Ellenbogen K.A.,
- Fitzpatrick A.P.,
- Scheinman M.M.
- Wolf P.A.,
- Abbott R.D.,
- Kannel W.B.
- Benjamin E.J.,
- Wolf P.A.,
- D'Agostino R.B.,
- Silbershatz H.,
- Kannel W.B.,
- Levy D.
- Murphy N.F.,
- MacIntyre K.,
- Stewart S.,
- Hart C.L.,
- Hole D.,
- McMurray J.J.
- Tedrow U.B.,
- Conen D.,
- Ridker P.M.,
- et al.
- Magnani J.W.,
- Hylek E.M.,
- Apovian C.M.
- Nalliah C.J.,
- Sanders P.,
- Kottkamp H.,
- Kalman J.M.
- Lee J.J.,
- Yin X.,
- Hoffmann U.,
- Fox C.S.,
- Benjamin E.J.
- Verdecchia P.,
- Dagenais G.,
- Healey J.,
- et al.,
- for the Ongoing Telmisartan Alone and in Combination With Ramipril Global Endpoint Trial/Telmisartan Randomized AssessmeNt Study in ACE iNtolerant subjects with cardiovascular Disease Investigators
- Psaty B.M.,
- Manolio T.A.,
- Kuller L.H.,
- et al.
- Rosengren A.,
- Hauptman P.J.,
- Lappas G.,
- Olsson L.,
- Wilhelmsen L.,
- Swedberg K.
- Rosenberg M.A.,
- Patton K.K.,
- Sotoodehnia N.,
- et al.
- Azarbal F.,
- Stefanick M.L.,
- Assimes T.L.,
- et al.
- Karas M.G.,
- Yee L.M.,
- Biggs M.L.,
- et al.
- Overvad K.,
- Tjønneland A.,
- Haraldsdóttir J.,
- Ewertz M.,
- Jensen O.M.
- Tjønneland A.,
- Overvad K.,
- Haraldsdóttir J.,
- Bang S.,
- Ewertz M.,
- Jensen O.M.
- Korn E.L.,
- Graubard B.I.,
- Midthune D.
- Mont L.,
- Elosua R.,
- Brugada J.
- Andersen K.,
- Rasmussen F.,
- Held C.,
- Neovius M.,
- Tynelius P.,
- Sundström J.