Subclinical Hypothyroidism and the Risk of Coronary Heart Disease and Mortality FREE

Quiz Ref IDControversy persists on the indications for screening and threshold levels of thyroid-stimulating hormone (TSH) for treatment of subclinical hypothyroidism,^1- 3 defined as elevated serum TSH levels with normal thyroxine (T₄) concentrations. Because subclinical hypothyroidism has been associated with hypercholesterolemia⁴ and atherosclerosis,⁵ screening and treatment have been advocated to prevent cardiovascular disease.³ However, data on the associations with coronary heart disease (CHD) events and mortality are conflicting among several large prospective cohorts.^6- 9 Three recent study-level meta-analyses^10- 12 found modestly increased risks for CHD and mortality, but with heterogeneity among individual studies that used different TSH cutoffs, different confounding factors for adjustment, and varying CHD definitions.¹⁰ Part of the heterogeneity might also be related to differences in participants' age, sex, or severity of subclinical hypothyroidism (as measured by TSH level).⁴ One cohort study suggested particularly high risk in participants with subclinical hypothyroidism and preexisting cardiovascular disease.⁸

Analysis of individual participant data from large cohort studies may reconcile these conflicting data and define the influence of age, TSH levels, and preexisting cardiovascular disease. Individual participant data analysis is considered the best way for synthesizing evidence across several studies because it is not subject to potential bias from study-level meta-analyses (ecological fallacy)¹³ and allows performance of time-to-event analyses.¹⁴

To clarify the cardiovascular risk of subclinical hypothyroidism, we formed the Thyroid Studies Collaboration and conducted an individual participant data analysis of subclinical hypothyroidism and CHD outcomes.

Identification of potential studies was based on protocols developed for our study-level meta-analysis of prospective cohort studies.¹⁰ Briefly, we conducted a systematic literature search of articles in all languages on the association between subclinical thyroid dysfunction and CHD or mortality (cardiovascular and total) published from 1950 to May 31, 2010, in the MEDLINE and EMBASE databases and searched bibliographies of key articles (details are available in the eMethods). To maximize the quality and comparability of the studies, we formulated general inclusion criteria a priori. We included only full-text, published longitudinal cohort studies that (1) measured thyroid function with both serum TSH level and thyroxine (T₄) level at baseline in adults, (2) followed up participants systematically over time, (3) assessed CHD events and/or mortality, and (4) had a comparison group with euthyroidism. We excluded studies that only examined persons taking antithyroid medications, thyroxine replacement or amiodarone, or with overt hypothyroidism (defined as low T₄ and elevated TSH concentrations). Possible studies for inclusion were independently assessed for suitability by 2 of the authors (N.R., J.G.) and any disagreement was resolved by discussion with a third author (D.C.B). The agreement between the 2 reviewers was 99.9% for the first screen (titles and abstracts, κ = 0.98) and 100% for the full-text screen (κ = 1.00).

Investigators from each eligible study were invited to join the Thyroid Studies Collaboration. We collected detailed information about prespecified outcomes and potential confounding variables for each participant. Requested data included individual demographic characteristics, baseline thyroid function (TSH and T₄ levels), baseline cardiovascular risk factors (eg, low- and high-density lipoprotein cholesterol level, diabetes, blood pressure, cigarette smoking), prevalent cardiovascular disease, medication use at baseline (thyroid medication, lipid-lowering and antihypertensive drugs), and outcome data.

To maximize the comparability of the studies, we used a common definition of subclinical hypothyroidism. Based on expert reviews^1- 2 and definitions used in the Cardiovascular Health Study,^6,15 we defined subclinical hypothyroidism as a serum TSH level of 4.5 mIU/L or greater to less than 20 mIU/L, with a normal T₄ concentration; and euthyroidism was defined as a serum TSH level of 0.5 mIU/L or greater and less than 4.5 mIU/L. Because the Whickham Survey used a first-generation TSH radioimmunoassay, which gives higher measured TSH values than current assays,¹⁶ a TSH range of 6.0 mIU/L or greater to less than 21.5 mIU/L was used for this individual participant data analyses, as in the original and recent analysis of this study.^17- 18 In that study, a serum TSH level of 6.0 mIU/L corresponded to the 97.5th percentile of the group with negative thyroid antibodies,¹⁸ which is close to the modern level of 4.5 mIU/L for the current generation of assays. For T₄ level, we used site- and method-dependent specific cutoffs (eTable) because T₄ measurements show greater intermethod variation than do sensitive TSH assays. The Whickham Survey measured total T₄ level.¹⁸ Participants with abnormal T₄ values, results suggestive of nonthyroidal illness (low TSH and FT₄ levels) or low TSH level (<0.5 mIU/L) were excluded (n = 3023). Some studies had participants with missing T₄ values (eTable); we considered participants with a TSH level of 4.5 mIU/L to 19.9 mIU/L and a missing T₄ level as having subclinical hypothyroidism because most adults with this degree of TSH elevation have subclinical and not overt hypothyroidism.¹⁹ We performed a sensitivity analysis excluding those with a missing T₄ level.

Outcome measures were CHD events, CHD mortality, and total mortality. To limit outcome heterogeneity observed with previous study-level meta-analyses,^10- 12 we used more homogeneous outcome definitions. Similar to the current Framingham risk score,²⁰ we limited cardiovascular mortality to CHD mortality or sudden death (eTable). A CHD event was defined as nonfatal myocardial infarction or CHD death (equivalent to hard events in the Framingham risk score²⁰) and hospitalization for angina or coronary revascularization (coronary artery bypass grafting or angioplasty).⁶ We performed a sensitivity analysis with hard events only.

Using previously described criteria¹⁰ and new information from study authors, we systematically evaluated the following key indicators of study quality¹³: methods of outcome adjudication and ascertainment, accounting for confounders, and completeness of follow-up ascertainment. Two reviewers (N.R., J.G.) rated all studies for quality.

We used separate Cox proportional hazard models to assess the associations of subclinical hypothyroidism with CHD events and mortality for each cohort (SAS version 9.2, SAS Institute Inc, Cary, North Carolina). Pooled estimates for each outcome were calculated using random-effects models, based on the variance model according to DerSimonian and Laird,²¹ as recommended^14,22 and published in recent 2-stage individual participant data analyses.²³ Results were summarized using forest plots (Review Manager version 5.0.24, Nordic Cochrane Centre, Copenhagen, Denmark). The research authors of 1 study with 14 CHD outcomes^5,10 declined to participate; as recommended,²⁴ we included the published summary estimate from that study in the random-effects models in a sensitivity analysis. To assess heterogeneity across studies, we used the I² statistic, which describes the total variation across studies attributable to heterogeneity rather than chance (I² >50% indicating at least moderate statistical heterogeneity).²⁵

Primary analyses were adjusted for age and sex, and then for traditional cardiovascular risk factors (systolic blood pressure, smoking, total cholesterol, diabetes) that were available in all cohorts (except for the Birmingham Study,²⁶ which was excluded from this analysis). We considered the age- and sex-adjusted model as the primary analysis because some traditional risk factors are potential mediators of the relationship between subclinical hypothyroidism and CHD.⁴

To explore sources of heterogeneity, we performed several predefined subgroup and sensitivity analyses. We conducted stratified analyses by age, sex, race, TSH concentrations, and preexisting cardiovascular disease. Based on expert reviews^1- 2 and previous studies,^7,15 subclinical hypothyroidism was stratified according to the following TSH concentration categories: 4.5-6.9 mIU/L (mild elevation), 7.0-9.9 mIU/L (moderate elevation), and 10.0-19.9 mIU/L (marked elevation). In the study-specific analyses stratified by age or TSH level, some strata had participants without either CHD deaths or CHD events (for 1 study²⁷). For these study-specific analyses, we used penalized likelihood methods²⁸ to obtain hazard ratios (HRs) and confidence intervals (CIs). As done in previous studies,^7,27,29 after including all participants in the primary analyses, we performed sensitivity analyses excluding participants who had thyroid hormone use at baseline and during follow-up. To calculate age- and sex-adjusted rates per 1000 person-years, we first fit Poisson models³⁰ to the pooled data, then standardized the fitted rate in the euthyroidism group to the overall age and sex distribution of the pooled sample. Finally, to obtain rates in the TSH groups consistent with the meta-analytic results, we multiplied the standardized rates in the euthyroidism group by the summary meta-analytic HRs. We checked the proportional hazard assumption using graphical methods and Schoenfeld tests (all P > .05). We used the Egger test³¹ and age- and sex-adjusted funnel plots to assess for publication bias.

Among 4440 reports identified, 12 prospective studies met eligibility criteria (eFigure) and 11 prospective cohorts in the United States, Europe, Australia, Brazil, and Japan agreed to provide individual participant data (Table 1). The final sample included 55 287 adults comprising 3450 with subclinical hypothyroidism (6.2%) and 51 837 with euthyroidism. Zero to 8.3% of participants reported thyroid hormone use at baseline (all excluded in 5 studies) and 0% to 12.6% reported thyroid hormone use during follow-up. The median follow-up ranged from 2.5 to 20 years, with total follow-up of 542 494 person-years.

All 11 cohort studies reported total and CHD deaths, and 7 studies also reported CHD events among 25 977 participants. For the quality of individual studies, all studies reported outcome adjudication without knowledge of thyroid status; 4 of 7 studies reporting CHD events used formal adjudication procedures^6- 8,27; and 4 of 11 studies reporting CHD deaths mainly used death certificates.^26,33- 35 All studies had 5% or less loss to follow-up.

During follow-up, 9664 participants died (2168 of CHD) and 4470 participants had CHD events (among 7 studies). In age- and sex-adjusted analyses, the overall HR for participants with subclinical hypothyroidism compared with euthyroidism was 1.18 (95% CI, 0.99-1.42) for CHD events (24.0 vs 20.3/1000 person-years for participants with euthyroidism), 1.14 (95% CI, 0.99-1.32) for CHD mortality (5.5 vs 4.9/1000 person-years), and 1.09 (95% CI, 0.96-1.24) for total mortality (23.1 vs 21.1/1000 person-years; Figure 1). We found heterogeneity across studies for CHD events (I² = 59%) and total mortality (I² = 66%), but not for CHD mortality (I² = 0%). We subsequently examined whether heterogeneity was related to differences in risks by degree of subclinical hypothyroidism and age. The risk of CHD events (P<.001 for trend) and CHD death (P = .005 for trend) increased with higher TSH level, but not for total mortality (Figure 2). In stratified analyses, participants with TSH levels of 10 mIU/L or greater had significantly increased risk of CHD events (HR, 1.89 [95% CI, 1.28-2.80]; n = 70 events/235; 38.4 vs 20.3/1000 person-years for participants with euthyroidism) and CHD mortality (HR, 1.58 [95% CI, 1.10-2.27]; n = 28 deaths/333; 7.7 vs 4.9/1000 person-years) compared with participants with euthyroidism. The risk for CHD associated with subclinical hypothyroidism appeared to be somewhat higher in younger participants, but the number of outcomes in the younger age group was small, and there was no significant trend in CHD risk across age groups. Otherwise, the risk estimates for CHD events, CHD mortality, and total mortality did not differ significantly according to age, sex, race, or preexisting cardiovascular disease, except an increase in CHD events and CHD mortality among white but not among nonwhite participants with subclinical hypothyroidism (Table 2). All results were similar after further adjustment for traditional cardiovascular risk factors.

Sensitivity analyses yielded similar results, with increased risks of CHD events and mortality in those with TSH levels of 10 mIU/L or greater (Table 3). Risk estimates were slightly higher for those with TSH levels of 10 mIU/L or greater after excluding those who took thyroid medication during follow-up. Estimates were lower for subclinical hypothyroidism overall after limiting the analyses to 4 studies with formal adjudication procedures, but slightly higher for those with TSH levels of 10 mIU/L or greater. The effect of increasing TSH level on CHD events did not significantly differ according to age (P = .87 for interaction). We found no evidence of publication bias, either with visual assessment of age- and sex-adjusted funnel plots or with the Egger test for mortality data (P = .39 for CHD mortality and P = .97 for total mortality) and little evidence of publication bias for CHD events (P = .13 for CHD events).

Quiz Ref IDIn this analysis of 55 287 individual participants from 11 prospective cohort studies, subclinical hypothyroidism was associated with an increased risk of CHD events and CHD mortality in those with higher TSH levels. There was a significant trend of increased risk at higher serum TSH concentrations, and the risk of both CHD mortality and CHD events was significantly increased in participants with TSH levels of 10 mIU/L or greater.Quiz Ref IDThese associations persisted after adjustment for traditional cardiovascular risk factors, and did not significantly differ by age, sex, race, or preexisting cardiovascular disease. Compared with participants with euthyroidism, the overall HR for CHD events with subclinical hypothyroidism was 1.18 (95% CI, 0.99-1.42) and the overall HR for CHD mortality was 1.14 (95% CI, 0.99-1.32). Minimal TSH elevations were not associated with an increased risk of CHD events and CHD mortality. Our results clarify the CHD risk of subgroups of adults with subclinical hypothyroidism, which could not be adequately addressed in previous study-level meta-analyses^10- 12 or in single cohort studies performed in more limited age groups or without TSH stratification.^{6- 7,26- 27}

These results are generally consistent with previous study-level meta-analyses showing modest increased risks of CHD events and cardiovascular mortality associated with subclinical hypothyroidism.^10- 11 However, these meta-analyses could not accurately explore potential differences related to participant characteristics (eg, age, TSH concentrations) because of potential bias without individual participant data analysis (ecological fallacy),¹³ and they also were limited by clinical heterogeneity,^10,36 with individual studies using varying TSH cutoffs, confounding factors for adjustment, and CHD definitions. Among 11 cohorts, only 2 studies previously reported results stratified by TSH level. One study⁹ reported an increased risk of CHD events in participants with a TSH level of 10.0 mIU/L or greater (HR, 2.2; 95% CI, 1.2-4.2) and the other study⁷ reported an increased risk of cardiovascular mortality (HR, 2.26; 95% CI, 0.54-9.45) but not CHD events (HR, 0.96; 95% CI, 0.35-2.61) over 4 years among adults aged 70 to 79 years with TSH levels of 10 mIU/L or greater. However, the HR for CHD events increased to 1.28 (95% CI, 0.68-2.39) with extended follow-up to 8 years in the present data. In overall pooled data, we found statistical heterogeneity among individual study findings for CHD events (I² = 59%), but not for CHD death. Part of the heterogeneity might be related to different CHD risks across age, race, and TSH subgroups.

Our individual participant data analysis found that the CHD outcomes in adults with subclinical hypothyroidism did not differ significantly across age groups. For the specific age group of 80 years or older, there was no significant increased risk of total mortality, CHD mortality, or CHD events in contrast to a single previous study that found reduced mortality associated with increasing TSH concentrations.^27,37 Previous study-level meta-analyses have found increased risks of CHD events and cardiovascular mortality associated with subclinical hypothyroidism, particularly in studies with a mean age of younger than 65 years,^10- 11 but this was not confirmed by our individual participant data analysis. We found some evidence for increased risks of CHD events and mortality in younger adults with subclinical hypothyroidism, but there also were large 95% CIs without significant trend across age groups (Figure 2). Moreover, the effect of increasing TSH level on CHD events did not significantly differ according to age. In contrast to a previous study suggesting that adults with subclinical hypothyroidism and preexisting cardiovascular disease might be at particularly high cardiovascular risk,⁸ we found no significant effect of baseline preexisting cardiovascular disease on outcomes.

The increased risk of CHD events associated with higher TSH levels in our study might be related to the known effects of thyroid hormone on the heart and metabolism, consistent with the concept that subclinical hypothyroidism is a milder form of overt hypothyroidism.^38- 39Quiz Ref IDIncreased systemic vascular resistance, arterial stiffness, altered endothelial function, increased atherosclerosis, and altered coagulability have been reported to be associated with subclinical hypothyroidism and may accelerate development of CHD.^4,39- 40 The fact that adjustment for traditional cardiovascular risk factors did not alter risks could favor this hypothesis. Other potential mechanisms include elevated cholesterol level,^4,39 although adjustment for cholesterol level did not remove the associations in our data. Adults with higher TSH concentrations also are more likely to develop overt hypothyroidism,⁴¹ and it is possible that this progression explains the association with subclinical hypothyroidism. Alternative explanations for the observed results are bias in the selection of included studies, bias and quality problems in the original studies, publication bias, and unmeasured confounders.⁴² Sensitivity analyses pooling higher-quality studies yielded similar results. Whereas one randomized controlled trial has shown benefits with thyroxine treatment of subclinical hypothyroidism on intima-media thickness⁴⁰ and another has shown benefits with thyroxine treatment of subclinical hypothyroidism on brachial artery endothelial function,⁴³ the potential causal relationship can only be proven by randomized controlled trials of thyroxine replacement and clinical outcomes.³⁶

Among the strengths of our study, an individual participant data analysis is the preferred way to perform time-to-event analyses to avoid biases associated with the use of aggregate data in meta-regression for subgroup analysis and to allow standardization of definitions of predictors, outcomes, and adjustment for potential confounders.^14,22 We included all available international and published data on these associations. Among the limitations of our study, the individual participant data analysis included predominantly white populations, except for 2 studies conducted in Japan³⁴ and Brazil.³⁵ Results for subgroups at risk of CHD mortality generally had wider 95% CIs than those for CHD events, reflecting less statistical power. However, post hoc calculations showed 80% power to detect meaningful differences between overall subclinical hypothyroidism and euthyroidism groups for each outcome. Specifically, our study had adequate power to detect an HR of 1.18 or higher for CHD events, an HR of 1.30 or higher for CHD mortality, and an HR of 1.13 or higher for total mortality. Even with this very large amount of individual participant data, our power for subgroup analyses was limited among those with TSH levels of 10 mIU/L or greater or adults younger than 50 years because of the limited number of CHD events and deaths. Thyroid function testing was performed only at baseline, and we have no data on how many participants progressed from euthyroidism to subclinical hypothyroidism, from subclinical to overt hypothyroidism, or who normalized their TSH level over time, which is a limitation of all published large cohorts.^6- 7,33 In addition, free triiodothyronine (T₃) was not available in most cohorts, and thus could not be included in thyroid status classification. Commencement of thyroid medication during follow-up by up to 12.6% of participants might have attenuated any true effects of subclinical hypothyroidism, as illustrated by the sensitivity analysis excluding such participants.

In summary, combining all available data from large prospective cohorts among 55 287 individual participants suggests that subclinical hypothyroidism is associated with an increased risk of CHD in those with higher TSH levels. The risk of both CHD mortality and CHD events, but not of total mortality, increases with higher concentrations of TSH and is significantly elevated in adults with TSH levels of 10 mIU/L or greater. Quiz Ref IDConversely, minimal TSH elevations are not associated with an increased risk of CHD events and CHD mortality. Our finding of no increased risk of CHD among the high proportions of adults with minimal TSH elevations is also important because many patients with minimal TSH elevations are currently treated in clinical practice.⁴⁴ Our results might help refine a TSH threshold at which larger clinical benefits of thyroxine replacement would be expected.^4,45 Our study cannot address whether these risks are attenuated or abolished by thyroxine replacement. Given the high prevalence of subclinical hypothyroidism,^2,19 this question needs to be addressed in an appropriately powered randomized controlled trial.