Association Between Serum Free Thyroxine Concentration and Atrial Fibrillation FREE

Thyroid dysfunction is common in the general population, with increasing prevalence of overt thyroid disease with increasing age.¹ Subclinical thyroid dysfunction, indicated biochemically by abnormal serum thyrotropin (TSH) values in association with normal thyroid hormone values, is also common, with a prevalence among adults of up to 12% and increasing with age.^2- 4

The pathophysiological consequences of mild thyroid dysfunction remain unclear. Several studies have shown an effect of subclinical hyperthyroidism on cardiac function or morphology.⁵ Importantly, there is evidence from 2 large prospective cohort studies^6- 7 and a cross-sectional study⁸ of an association between subclinical hyperthyroidism and atrial fibrillation (AF). Atrial fibrillation is an important dysrhythmia representing an independent risk factor for cardiovascular events and stroke.⁹ We have previously reported from a cohort study of 1191 subjects 60 years or older that a single low serum TSH measurement predicts increased all-cause mortality during 10 years of follow-up, an excess mortality specifically reflecting increased deaths from circulatory disorders,¹⁰ although another study has not replicated this.⁶ Subclinical hypothyroidism may also have pathophysiological consequences relevant to the cardiovascular system, with associations described with hyperlipidemia and left ventricular dysfunction^11- 12 and evidence for reversal of these changes with thyroxine (T₄) replacement therapy.¹³ There is conflicting evidence whether these findings translate into cardiovascular morbidity or mortality,^6,14- 19 but a recent meta-analysis suggests that subclinical hypothyroidism is associated with increased coronary heart disease risk.²⁰

Overall, evidence of an association of subclinical thyroid dysfunction with morbidity or mortality has been considered insufficient, at least to direct strategies for screening or therapeutic intervention.²¹ Since AF is one of the most important chronic disorders linked to subclinical thyroid dysfunction that might warrant screening or treatment, we have examined the relationship between thyroid status and the presence of AF on electrocardiogram (ECG) in a large community-based cohort of subjects 65 years or older.

The cohort comprised 5860 subjects 65 years or older and excluded those currently being treated for thyroid dysfunction and those with a history of hyperthyroidism. Subjects were identified and investigated in primary care with tests of thyroid function and resting 12-lead ECG digitally recorded onto a personal digital assistant, transferred to a personal computer, and displayed for analysis using Cardioview Software (MicroMedical Industries Ltd, Sydney, Australia). All ECGs were analyzed for the presence of AF by a single cardiologist (M.D.G.) blinded to thyroid status and patient details.

All current drug therapies were recorded, as were major risk factors for AF previously defined in the Framingham population (smoking, diabetes mellitus, hypertension, heart failure, and ischemic heart disease) and used in similar studies.⁷ Initial recording was based on patient reporting, with validation by inspection of primary care records as previously described.²² The sensitivity and positive predictive value of computerized information regarding major medical diagnoses recorded in family practice in the United Kingdom has been shown to exceed 90%.^23- 24 All subjects gave written informed consent for the study, which was approved by the multicenter research ethics committee.

Serum samples were stored at −80ºC. Serum free T₄, free triiodothyronine (T₃), and TSH were measured by chemiluminescent immunoassay (Advia Centaur; Bayer Diagnostics, Newbury, England). The laboratory reference range for free T₄ was 0.70 to 1.55 ng/dL (9.0-20.0 pmol/L), with an interassay coefficient of variation of 8.2% to 9.8% over the range of 0.64 to 4.27 ng/dL (8.2-54.9 pmol/L). Serum TSH concentration had a reference range of 0.4-5.5 mU/L, with an interassay coefficient of variation of 4.4% to 10.9% over the range of 0.41 to 24.5 mU/L. The lower limit of reporting for the TSH assay was 0.1 mU/L, with a mean functional sensitivity 0.02 mU/L. Free T₄ and TSH concentrations were determined in all; in those with serum TSH concentrations below the reference range, serum free T₃ (reference range, 227.3-422.1 pg/dL [3.5-6.5 pmol/L], interassay coefficient of variation of 4.2%-6.9% over the range of 259.7-1039.0 pg/dL [4.0-16.0 pmol/L]) was additionally measured. The reference ranges used for free T₄, TSH, and free T₃ were those recommended by the manufacturer and used in other studies of this cohort and studies from our unit.^22,25 Subjects were classified according to measurements of serum free thyroid hormone and TSH concentrations as follows:

Data were summarized using medians, interquartile ranges (IQRs), proportions, and associated 95% confidence intervals (CIs). The Kolmogorov-Smirnov test was used to assess normality of free T₄, free T₃, and TSH values. The comparison of median concentrations of free T₄ and TSH used Mann-Whitney and Kruskal-Wallis nonparametric tests. The comparison of proportions was achieved using Pearson χ² and Fisher exact tests, as appropriate. The analysis of factors associated with the prevalence of AF was conducted using a logistic regression model. Backward stepwise elimination was used to produce a parsimonious logistic regression model, and the possibility of nonlinearity and interaction effects was also investigated.

The cohort comprised 5860 subjects. The median age of the cohort was 72 years (range, 65-98 years; 50.9% female and 49.1% male). Overall distributions of age and free T₄, free T₃, and TSH values were significantly different from normal (P<.01); nonparametric statistics were therefore used to summarize and analyze these parameters. Fourteen subjects (0.2%) were found to have previously undiagnosed overt hyperthyroidism (median free T₄, 1.93 ng/dL [24.8 pmol/L], IQR, 1.40-2.28 ng/dL [18.0-29.3 pmol/L]; free T₃, 506.5 pg/dL [7.8 pmol/L], IQR, 474.0-571.4 pg/dL [7.3-8.8 pmol/L]; and TSH, <0.1 mU/L in all). Of the 5860 subjects, 126 (2.2%) had subclinical hyperthyroidism (median free T₄, 1.20 ng/dL [15.5 pmol/L], IQR, 1.06-1.31 ng/dL [13.6-16.9 pmol/L]; and free T₃, 311.7 pg/dL [4.8 pmol/L], IQR, 285.7-344.2 pg/dL [4.4-5.3 pmol/L]). Of these 126 subjects, serum TSH was undetectable (<0.1 mU/L) in 27; TSH concentration was reported as 0.1 mU/L in 11 and was low but detectable (0.2-0.3 mU/L) in the remaining 88. Of the 5860 subjects, 5519 (94.4%) were euthyroid (median free T₄, 1.11 ng/dL [14.3 pmol/L], IQR, 1.01-1.22 ng/dL [13.0-15.7 pmol/L]; and TSH, 1.6 mU/L, IQR, 1.1-2.3 mU/L). There were 167 subjects (2.9%) who had subclinical hypothyroidism (median free T₄, 0.98 ng/dL [12.6 pmol/L], IQR, 0.89-1.06 ng/dL [11.4-13.6 pmol/L]; and TSH, 6.8 mU/L, IQR, 6.0-8.8). Twenty-three subjects (0.4%) had overt hypothyroidism (median free T₄, 0.58 ng/dL [7.5 pmol/L], IQR, 0.45-0.62 ng/dL [5.8-8.0 pmol/L]; and TSH, 40.6 mU/L, IQR, 16.7-52.2). There was a preponderance of women over men in the hyperthyroid (58%) and subclinical hypothyroid (65%) categories (P<.001). Ten subjects fell outside the prespecified diagnostic categories defined in the “Methods” section. Of these 10, 9 had a minor elevation of free T₄ (range, 1.56-1.86 ng/dL [20.1-23.9 pmol/L]), with normal free T₃ and TSH values, which were low but detectable in 7 (range, 0.2-0.3 mU/L) or undetectable in 2. The remaining subject had low free T₄ (0.65 ng/dL [8.4 pmol/L]) and free T₃ (201.3 pg/dL [3.1 pmol/L]), with serum TSH lower than 0.1 mU/L. One additional subject had an erroneous free T₄ result reflecting familial dysalbuminemic hyperthyroxinemia. These 11 subjects were excluded from analyses based on thyroid status categories but were included in other analyses.

The prevalence of AF evident on resting ECG in the whole cohort was 4.8%, with a prevalence of 3.1% (91/2982) in women and a higher prevalence in men at 6.6% (189/2878) (odds ratio [OR], 2.23; P<.001). Of the cases of AF evident on ECG, 150 (53%) were newly found; a history of AF had been recorded in the remainder. The prevalence of AF according to category of thyroid status is given in Table 1; this indicates little variation in prevalence compared with the numerically dominant euthyroid group, with the exception of the subclinical hyperthyroid group. A comparison of the prevalence of AF between the euthyroid and subclinical hyperthyroid subjects allowing for differences between the sexes was achieved with a logistic regression of AF prevalence against thyroid status and sex. This confirmed the higher prevalence of AF in men (P<.001) and in subclinical hyperthyroid subjects when compared with euthyroid subjects (P = .01). A similar analysis showed no significant difference in AF prevalence between the euthyroid subjects and those in other categories of thyroid status (P = .57). The prevalence of AF in the 126 subjects with subclinical hyperthyroidism was 3.7% (1/27) in those with undetectable TSH and 11.1% (11/99) in those with low but detectable TSH (P = .46).

Table 2 gives AF prevalence divided according to age and sex. Logistic regression of AF prevalence against age and sex revealed a significant rise with increasing category of age (P<.001) and higher prevalence in men than in women in all age groups (P<.001). The median ages of subjects with and without AF were 77 and 72 years, respectively (P<.001).

There was a significant difference in serum free T₄ concentration between patients with AF and those without in the whole cohort (median free T₄, 1.14 ng/dL [14.7 pmol/L], IQR, 0.12-1.27 ng/dL [13.5-16.4 pmol/L] vs 1.10 ng/dL [14.2 pmol/L], IQR, 1.00-1.22 ng/dL [12.9-15.7 pmol/L]; P<.001) but no significant difference in TSH concentration (median TSH, 1.6 mU/L in both groups; P = .82). Figure 1 illustrates the increasing prevalence of AF with increasing serum free T₄ concentration for the whole cohort, the smooth curve being derived from logistic regression of the presence of AF vs free T₄ level. Figure 2 shows a smoothed plot of the prevalence of AF vs free T₄ for different ages and sexes derived from an additive effects logistic regression of the presence of AF vs free T₄, sex, and age.

Logistic regression analysis, taking into account risk factors for AF defined previously in the Framingham cohort (smoking, diabetes mellitus, hypertension, heart failure, and ischemic heart disease) (Table 3), showed a persistent association of AF with increasing serum free T₄ concentration, subclinical hyperthyroidism, age, and male sex, as well as the expected risk associated with diabetes mellitus, hypertension, and heart failure. Serum TSH, smoking, and ischemic heart disease were not significant independent predictors of AF (P = .94, .33, and .12, respectively). A difference in serum free T₄ concentration of 1 pmol/L (0.08 ng/dL) was associated with an OR for the finding of AF of 1.08 (95% CI, 1.03-1.14; P<.001), while larger differences in free T₄ concentrations of 2.5 pmol/L (0.19 ng/dL), 5.0 pmol/L (0.39 ng/dL), and 10.0 pmol/L (0.78 ng/dL) were associated with larger ORs of 1.21 (95% CI, 1.06-1.38), 1.47 (95% CI, 1.13-1.91), and 2.17 (95% CI, 1.29-3.63), respectively.

In view of the well-recognized influence of overt thyroid dysfunction on AF risk,²⁶ further analysis was performed excluding those with previously undiagnosed overt hyperthyroidism and hypothyroidism. The remaining group comprised 5812 subjects, 279 (4.8%) of whom had AF on ECG. Serum free T₄ concentration remained higher in those with AF than in those without (median free T₄, 1.14 ng/dL [14.7 pmol/L], IQR, 1.05-1.27 ng/dL [13.5-16.4 pmol/L] vs 1.10 ng/dL [14.2 pmol/L], IQR, 1.00-1.22 ng/dL [12.9-15.7 pmol/L]; P<.001). Atrial fibrillation was found more commonly in men than in women (189/2862 [6.6%] vs 90/2950 [3.1%]; adjusted OR, 2.40; P<.001). The median age of those with AF remained higher than in those without (age, 77 vs 72 years; P<.001). Logistic regression analysis of the significant risk factors for this subgroup that excluded those with overt thyroid dysfunction produced similar results to those seen for the whole cohort (Table 4).

Further subgroup analysis confined to those 5519 euthyroid subjects in the cohort with normal serum TSH concentrations (0.4-5.5 mU/L) again revealed higher free T₄ concentrations in those with AF compared with those without (median free T₄, 1.13 ng/dL [14.6 pmol/L], IQR, 1.05-1.26 ng/dL [13.5-16.2 pmol/L] vs 1.10 ng/dL [14.2 pmol/L], IQR, 1.01-1.21 ng/dL [13.0-15.6 pmol/L]; P = .001). Logistic regression (Table 4) again revealed increasing category of age and male sex to be independent predictors of AF (prevalence of AF, 6.5% [men] vs 2.9% [women]; adjusted OR, 2.44; P<.001). Serum free T₄ was again an independent predictor of AF. The final column in Table 4 used a more stringent definition of euthyroidism, namely TSH concentration in the range of 0.4-4.0 mU/L, inclusive. Serum free T₄ remained an independent predictor of AF (P = .009), as did other factors significant when the euthyroid range of TSH concentration was defined as 0.4 to 5.5 mU/L.

Because it is well recognized that amiodarone therapy is associated with a rise in serum free T₄ concentration (typically within the normal range)²⁷ and this therapy is prescribed in AF, an analysis was repeated excluding results from the 25 subjects in the cohort prescribed this drug, to avoid any potential bias of association. After this exclusion, serum free T₄ remained an independent predictor of AF, with a minimal difference in OR (per 1-pmol/L change [0.08 ng/mL] of free T₄, 1.07; 95% CI, 1.02-1.13; P = .01), and the association with subclinical hyperthyroidism remained significant (OR, 1.92; 95% CI, 1.03-3.70; P = .04).

The present study has revealed an association between the finding of AF on resting ECG and serum free T₄ concentration in a large community-based elderly cohort, even in biochemically euthyroid subjects with normal serum TSH concentrations. As expected, there was an association of AF with increasing age and male sex, as well as other recognized risk factors including hypertension, heart failure, and diabetes mellitus,²⁸ the overall prevalence of AF being similar to that described in other cohorts of similar age.^29- 32 Taking previously recognized risk factors into account, serum free T₄ was an independent predictor of the presence of AF in the cohort as a whole, and this association was sustained after exclusion of those with overt thyroid dysfunction. Furthermore, when the analysis was further restricted to those classified as euthyroid (with normal serum TSH concentration), this independent relationship between serum free T₄ and AF was still evident.

It is well recognized that overt hyperthyroidism is associated with AF, the reported prevalence being 5% to 15%.^26,33 This association of AF and overt hyperthyroidism is more common with increasing age and with underlying cardiovascular disease, especially structural heart disease^26,34- 35 and is known to be associated with complications, especially stroke.^34,36 The present study revealed only a low prevalence of previously undiagnosed overt hyperthyroidism detected by screening; none of these subjects had AF.

A topic of considerable debate and importance is the relevance of subclinical hyperthyroidism to AF risk, especially in older age groups. Several studies have indicated an association between subclinical hyperthyroidism and AF prevalence or incidence. Ten years of follow-up of more than 2000 subjects 60 years or older, who were part of the Framingham cohort, revealed an adjusted relative risk of 3.1 (95% CI, 1.7-5.5) for development of AF in the undetectable TSH group and 1.6 (95% CI, 1.0-2.5) in those with low but detectable TSH.⁷ This cohort was different from the present one in that it included subjects with overt hyperthyroidism and taking thyroid hormones, which may account for the higher prevalence of subclinical thyroid dysfunction than described here. Similar findings have recently been reported by Cappola et al⁶ investigating incident AF in 3233 US community-dwelling subjects 65 years or older. After exclusion of those with prevalent AF, those with subclinical hyperthyroidism (1.6% of the cohort) had a greater incidence of AF compared with those with normal thyroid function over a period of 12 years (adjusted hazards ratio, 1.98; 95% CI, 1.29-3.03).

A further large but retrospective cross-sectional study compared the prevalence of AF in 1338 subjects with overt or subclinical hyperthyroidism due to autonomous thyroid nodules or Graves disease with AF prevalence in a control group of 22 300 subjects admitted to a hospital.⁸ The AF prevalence was 13.8% in patients with overt hyperthyroidism, 12.7% in those with subclinical hyperthyroidism, and 2.3% in euthyroid controls. The relative risk of AF in those with subclinical hyperthyroidism was 5.2 (95% CI, 2.1-8.7) compared with controls. The results of the present study, in terms of the finding of increased AF prevalence in subclinical hyperthyroidism, are similar, but Auer et al⁸ did not explore any relationship of AF with thyroid hormone concentrations. A further difference is that in the present study, classification of subjects with regard to their thyroid status was restricted to biochemical investigation, and those classified subclinical hyperthyroid were not limited to those with proven underlying thyroid disease.

An intriguing and novel finding in the present study is the association between the presence of AF on ECG and serum free T₄ concentration, a relationship still evident when those with thyroid dysfunction were excluded and when those prescribed amiodarone were excluded. An expected similar (but negative) association with serum TSH concentration was not evident, despite TSH being considered a more sensitive index of thyroid status than free T₄. Circulating TSH reflects the negative feedback effects of T₄ and T₃ on the pituitary gland, as well as a variety of other influences such as nonthyroidal illnesses and drug therapies.³⁷ It is possible that serum free T₄ is a more sensitive index of cardiac “thyroid status” than TSH and that the present findings reflect particular sensitivity of the heart to T₄. Several potential mechanisms could be invoked for the effect of T₄ on AF risk, including elevation of left atrial pressure secondary to increased left ventricular mass and impaired ventricular relaxation,³⁸ enhanced automaticity and triggered activity in pulmonary vein cardiomyocytes,³⁹ ischemia resulting from raised resting heart rate, and increased atrial ectopic activity.⁴⁰

Free T₄ concentrations may themselves be modulated by “nonthyroidal” factors such as illnesses and drug therapies,³⁷ the typical influence being reduction in free T₄ concentration, and these same factors are likely to increase risk of AF. However, we found that low free T₄ concentration was associated with the lowest prevalence of AF, arguing against any nonspecific influence of illness state or drug therapies.

The present finding of an association in euthyroid subjects between serum free T₄ concentration and AF is consistent with a recent study of 403 ambulatory men aged 73 to 94 years in whom higher free T₄ concentration within the reference range was associated with lower physical function, independent of age and illness,⁴¹ lending support to the view that variation of free T₄ concentration, even within the normal range, may have deleterious physiological consequences.

Major strengths of our study were the recruitment of a large population-based cohort of older subjects that excluded those with current thyroid disease or previous hyperthyroidism and the documentation of major risk factors for AF. The prevalence of AF was determined by resting 12-lead ECG, with interpretation by a single blinded observer. Subgroup analysis was performed after exclusion of those with overt and subclinical thyroid dysfunction, and we performed analyses taking into account age, sex, and known risk factors, as well as category of thyroid status and serum free T₄ concentration. A weakness is that incident data for AF and other vascular end points, including mortality, have yet to be recorded for this cohort.

The present study examined a community-based cohort of elderly subjects and specifically excluded those with current thyroid disease and previous hyperthyroidism. The finding of increased likelihood of AF in those with subclinical hyperthyroidism lends support to the small number of studies linking this biochemistry with AF risk.^6- 8 We found no association of subclinical hypothyroidism with AF, confirming the results in previous studies.^6- 8 The finding that subclinical hyperthyroidism detected as a result of screening is associated with AF contributes to the debate²¹ about the value of screening at-risk populations, such as elderly subjects, known to have a higher prevalence of undetected (and usually mild) thyroid dysfunction. This debate needs to be informed by evidence from trials investigating whether treatment prevents or reverses AF. In those receiving T₄ replacement therapy, the findings add support to recommendations²¹ to adjust T₄ doses until TSH and free T₄ concentrations are within the reference range. In euthyroid subjects with normal TSH values, the finding of higher risk of AF in those with high but normal serum free T₄ concentrations raises intriguing new questions about reducing risk of this important and common dysrhythmia in the general population.

Correspondence: M. D. Gammage, MD, FRCP, Department of Cardiovascular Medicine, Division of Medical Sciences, University of Birmingham, Queen Elizabeth Hospital, Edgbaston, Birmingham B15 2TH, England (m.d.gammage@bham.ac.uk).

Accepted for Publication: December 24, 2006.

Author Contributions: Drs Gammage, Holder, and Franklyn had full access to the data and take responsibility for its integrity and accuracy of analysis. Study concept and design: Gammage, Parle, Holder, Roberts, Hobbs, Wilson, Sheppard, and Franklyn. Acquisition of data: Gammage, Parle, Holder, Roberts, Hobbs, Wilson, Sheppard, and Franklyn. Analysis and interpretation: Gammage, Parle, Holder, Roberts, Hobbs, Wilson, Sheppard, and Franklyn. Drafting and revision of manuscript: Gammage, Parle, Holder, Roberts, Hobbs, Wilson, Sheppard, and Franklyn. Statistical analysis: Holder. Obtained funding: Gammage, Parle, Holder, Roberts, Hobbs, Wilson, Sheppard, and Franklyn. Study supervision: Gammage, Parle, Holder, Roberts, Hobbs, Wilson, Sheppard, and Franklyn.

Financial Disclosure: None reported.

Funding/Support: This study was funded by a project grant from the Health Foundation, with matched support from the Primary Care Research Trust.

Role of the Sponsor: The funding agencies played no part in study design or execution.

Acknowledgment: We are grateful to the participating research nurses and 20 primary care practices and their staff and patients who took part in this study, as well as to all other members of the Birmingham Elderly Thyroid Study Team.