Subclinical Hypothyroidism and Survival: The Effects of Heart Failure and Race

Source cohort

We conducted a retrospective cohort study using data from the NHANES III, a survey executed by the National Center for Health Statistics (NCHS) from 1998 to 1994 to provide national estimates of the health and nutritional status of U.S. children and adults. The survey employed a complex, multistage, stratified, clustered sampling design in which children, elderly, Black, and Mexican-American participants were oversampled to enable precise estimation within these subgroups (29). At the time of enrollment, participants' self-reported sociodemographics, comorbid conditions, and medications were assessed; blood and urine samples were collected for measurement of serum TSH, total T₄ (TT₄), creatinine, and urine albumin and creatinine (details below). The NHANES III study protocol was approved by the NCHS Institutional Review Board, and all participants gave their informed written consent before participation. The current study was deemed exempt by the Partners Healthcare Institutional Review Board.

We included NHANES III participants ages ≥18 years old at enrollment, who had baseline serum TSH, TT₄, and creatinine measurements and had available follow-up data (n = 15 467). From these, we excluded participants with secondary hypothyroidism (TSH < lower limit of normal [LLN] and TT₄ < LLN) (30) due to sparse representation (n = 2). Given that the comparison of interest was between primary hypothyroidism (subclinical and overall) vs euthyroidism with mortality within strata of CHF and race, we excluded participants with primary hyperthyroidism (TSH < LLN; n = 560), secondary hyperthyroidism (TSH > upper limit of normal [ULN] and TT₄ > ULN; n = 8) (30), and missing data on CHF status (n = 18); race was not missing for any patients. The final analytical cohort contained 14 879 participants. Within the NHANES III, race and comorbidity status (including CHF) were self-identified by participants.

Laboratory measures

TSH was measured with a chemiluminescence immunometric assay (Nichols Institute Diagnostics, San Juan Capistrano, California), and TT₄ was measured with an RIA (Roche Molecular Biochemicals, Indianapolis, Indiana). The normal reference ranges for TSH and TT₄ were 0.39 to 4.6 mIU/L and 4.5 to 13.2 μg/dL, respectively (1, 31). Spot urine albumin was measured by solid-phase fluorescent immunoassay, and spot urine and serum creatinine were measured by the Jaffe rate reaction (31). Microalbuminuria and macroalbuminuria were defined as an albumin to creatinine ratio (ACR) of 30 to 299 mg/g and ≥300 mg/g, respectively (32). Serum creatinine measurements were calibrated to the Cleveland Clinic laboratory by subtracting 0.23 mg/dL as previously described (33). Estimated glomerular filtration rate (eGFR) was calculated using the Chronic Kidney Disease Epidemiology Collaboration equation (eGFR = 141 × min[Scr/κ, 1]^α × max[Scr/κ, 1]^−1.209 × 0.993^Age × 1.018 [if female] × 1.159 [if Black]; Scr is serum creatinine, κ = 0.7 for females and 0.9 for males, α = −0.329 for females and −0.411 for males, min indicates the minimum of Scr/κ or 1, and max indicates the maximum of Scr/κ or 1) (34).

Exposure ascertainment

In co-primary analyses, we sought to examine how mortality was associated with hypothyroidism overall and with subclinical hypothyroidism. In the former analyses, participants with TSH higher than ULN were considered hypothyroid, and participants with TSH within the reference range were considered euthyroid. In the latter analyses, hypothyroid participants were parsed into subclinical and overt disease on the basis of their TT₄ levels (subclinical is within reference range; overt is below LLN) (30). Given that TT₄ levels may not be concordant with unbound, biologically active free T₄ levels (35), secondary analyses were also conducted in which subclinical and overt hypothyroidism were defined by severity of TSH derangement (subclinical is ≤10 mIU/L; overt is >10 mIU/L); this cutoff was selected based on the TSH threshold at which treatment of subclinical hypothyroidism is often recommended based on published guidelines (36).

Outcome ascertainment

The primary outcome of interest was all-cause mortality. Mortality data were available through December 31, 2006, and were ascertained through linkage of NHANES III to death certificate data from the National Death Index (NDI). The NDI is a central computerized index of annually updated death record information compiled from data submitted by state vital statistics offices to the NCHS. The NDI is considered the gold standard for death ascertainment for epidemiologic research, because it has been shown to provide superior coverage of deaths compared with other sources (37). Participants were considered at risk the day after baseline TSH and TT₄ measurements until death or end of follow-up. Secondary analyses considered death from cardiovascular causes; this was ascribed on the basis of International Statistical Classification of Diseases, Injuries, and Causes of Death codes, 10th revision codes (Supplemental Table 1, published on The Endocrine Society's Journals Online web site at http://jcem.endojournals.org) (38, 39). Cause-of-death information was available for 98.8% of the source cohort.

Statistical analyses

Baseline characteristics between exposure groups were compared using 2-sample t tests, Wilcoxon rank sum tests, and χ² tests as dictated by data type. Analogous methods were used to estimate associations between exposure and mortality in all analyses. Unadjusted associations between exposure and mortality were estimated using Kaplan-Meier plots, log-rank testing, and unadjusted Cox proportional hazards models. Multivariable Cox proportional hazards models were adjusted for potential confounders of the exposure-mortality association based on published data that included age, sex, categories of race/ethnicity (non-Hispanic Black and non-Hispanic White referred to as Black and White, respectively, in this report; Mexican-American; and other race/ethnicity), diabetes, active smoking status, hypertension, hypercholesterolemia, previous stroke, previous myocardial infarction (MI), pre-existing CHF, body mass index (BMI), ACR (categorized as <30, 30–299, or ≥300 mg/g or missing), and eGFR (categorized as <60 or ≥60 mL/min/1.73 m²). Sensitivity analyses were also conducted in which 1) multivariable Cox models were additionally adjusted for cardiovascular medications (amiodarone, angiotensin-converting enzyme [ACE] inhibitors, β-blockers, dihydropyridine and nondihydropyridine calcium channel blockers, digoxin, thiazide and loop diuretics, statins, niacin, gemfibrozil, and long-acting nitrates) and 2) participants on baseline thyroid supplementation (n = 260) and thyroid-suppressive therapy (n = 3) were excluded. The proportional hazards assumption was confirmed graphically and through Schoenfeld residual testing.

Stratum-specific estimates were estimated within subgroups defined by race, pre-existing CHF, and pre-existing atherosclerotic disease (previous MI or stroke). Statistical significance of effect modification was assessed by likelihood ratio testing comparing models with and without the corresponding 2-way factor-by-exposure cross-product terms (eg, CHF × subclinical hypothyroidism). An interaction P value of <.05 was considered statistically significant. Analyses were performed using Stata MP version 10.1 (StataCorp, College Station, Texas).

Cohort description

The overall study cohort consisted of 14 879 participants, 14 130 (95.0%) of whom were euthyroid and 749 (5.0%) of whom had hypothyroidism overall; among the latter, 691 had subclinical hypothyroidism (representing 4.6% of the total cohort and 92.3% of participants with hypothyroidism overall). Compared with euthyroidism, participants with subclinical hypothyroidism tended to be older, female, and White; had higher prevalence of low eGFR, diabetes, micro- and macroalbuminuria, CHF, hypertension, hypercholesterolemia, previous stroke, previous MI, higher BMI, and exogenous thyroid hormone use; and had a lower prevalence of smoking (Table 1). Similar patterns of association were observed when comparing patients with hypothyroidism overall with euthyroidism (Table 1) and when comparing subclinical hypothyroidism as defined by TSH levels with euthyroidism (Supplemental Table 2).

Subclinical hypothyroidism and mortality according to pre-existing CHF

Within the overall cohort, participants contributed a total of 199 299 patient-years of at-risk time during which 3337 all-cause deaths (1477 cardiovascular deaths) were observed. Median time at risk was 14.3 years. In the overall cohort, a significant association between subclinical hypothyroidism and hypothyroidism overall with all-cause mortality was not observed: adjusted hazard ratios (aHRs) = 1.04 (with 95% confidence interval [CI] of 0.91–1.18) and 1.02 (95% CI, 0.90–1.15), respectively. In the primary analyses, pre-existing CHF modified the association between subclinical hypothyroidism (P interaction = .03) and hypothyroidism overall (P interaction = .01) with all-cause mortality. Subclinical hypothyroidism was significantly associated with greater all-cause mortality compared with euthyroidism in those with CHF (aHR = 1.44 [95% CI, 1.01–2.06]) but not in those without CHF (aHR = 0.97 [95% CI, 0.85–1.11]) (Figure 1A). Similarly, hypothyroidism overall was significantly associated with greater all-cause mortality compared with euthyroidism in those with CHF (aHR = 1.47 [95% CI, 1.05–2.05]) but not in those without CHF (aHR = 0.95 [95% CI, 0.83–1.09]) (Figure 1B). In sensitivity analyses adjusted for cardiovascular medications, results were qualitatively similar (Table 2). After exclusion of participants using baseline thyroid medications, associations between subclinical hypothyroidism and hypothyroidism overall with mortality in participants with pre-existing CHF were even more potent (Table 2).

A similar pattern of results was observed in secondary analyses of subclinical hypothyroidism defined by TSH levels only. Subclinical hypothyroidism was significantly associated with greater all-cause mortality compared with euthyroidism in those with CHF (aHR = 1.77 [95% CI, 1.19–2.64]) but not in those without CHF (aHR = 0.97 [95% CI, 0.84–1.12]) (P interaction = .01; Supplemental Figure 1). In sensitivity analyses adjusted for cardiovascular medications, results were qualitatively similar (Table 2). After exclusion of participants using baseline thyroid medications, the association between subclinical hypothyroidism with mortality in participants with pre-existing CHF was even more potent (Table 2).

Hypothyroidism overall and mortality according to race

Race modified the association between hypothyroidism overall with all-cause mortality (P interaction = .03). Hypothyroidism overall was significantly associated with greater all-cause mortality compared with euthyroidism in Black participants (aHR = 1.44 [95% CI, 1.03–2.03) but not in those who were non-Black (aHR = 0.95 [95% CI, 0.83–1.08) (Figure 2B). In sensitivity analyses adjusted for cardiovascular medications, results were qualitatively similar (Table 2). After exclusion of participants using baseline thyroid medications, the association between hypothyroidism overall with mortality in Black participants was even more potent (Table 2).

In stratified analyses, subclinical hypothyroidism was associated with greater all-cause mortality compared with euthyroidism in Black participants (HR 1.51 [95% CI, 1.05–2.17) but not in other races (HR = 0.96 [95% CI, 0.84–1.10]); however, tests of interaction (P interaction = .07) did not reach statistical significance (Figure 2A). In sensitivity analyses adjusted for cardiovascular medications, results were qualitatively similar (Table 2). After exclusion of participants using baseline thyroid medications, tests of interaction for race were statistically significant (P interaction = .005) (Table 2).

Similarly, in stratified analyses, subclinical hypothyroidism defined by TSH levels trended toward greater all-cause mortality only among Black participants (aHR = 1.48 [95% CI, 0.96–2.28]), but not in those of other races (aHR = 0.98 [95% CI, 0.85–1.14]); however, estimates and tests for interaction did not reach statistical significance (P interaction = .08; Supplemental Figure 2). In sensitivity analyses adjusted for cardiovascular medications, results were qualitatively similar. After exclusion of participants using baseline thyroid medications, tests of interaction for race were statistically significant (P interaction = .03) (Table 2).

Subclinical hypothyroidism and mortality considering both pre-existing CHF and race

The concurrent effects of race and CHF on the association between subclinical hypothyroidism and hypothyroidism overall with all-cause mortality were considered (Table 3). Among participants without CHF, race modified the association between subclinical hypothyroidism and mortality: aHR (95% CI) was 1.43 (0.96–2.14) among Blacks and 0.89 (0.77–1.03) among non-Blacks (P interaction = .002). Similarly, among participants with CHF, race modified the association between subclinical hypothyroidism and mortality; aHR (95% CI) was 2.68 (0.84–8.61) among Blacks and 1.36 (0.92–2.02) among non-Blacks (P interaction = .03).

Among participants without CHF, race modified the association between hypothyroidism overall and all-cause mortality; aHR (95% CI) was 1.55 (1.06–2.27) among Blacks and 0.87 (0.75–1.00) among non-Blacks (P interaction = .005). In contrast, among participants with CHF, hypothyroidism overall was associated with elevated risk that did not differ according to race (P interaction = .7).

When cardiovascular mortality was considered (Table 4), race modified the association between subclinical hypothyroidism and mortality among participants without CHF (P interaction = .01) but not among participants with CHF (P interaction > 0.9). Similarly, race modified the association between hypothyroidism overall and mortality among participants without CHF (P interaction = .004) but not among participants with CHF (P interaction = .4).

Subclinical hypothyroidism and mortality according to pre-existing atherosclerotic disease

In stratified analyses, subclinical hypothyroidism was associated greater all-cause mortality compared with euthyroidism in participants with atherosclerotic disease (aHR = 1.34 [95% CI, 1.04–1.74]) but not in those without (aHR = 0.91 [95% CI, 0.73–1.14]); however, tests of interaction did not reach statistical significance (P interaction = .2). Subclinical hypothyroidism defined by TSH levels was also associated with greater all-cause mortality compared with euthyroidism in participants with atherosclerotic disease but not in those without (aHR = 1.32 [95% CI, 1.00–1.75] and 0.96 [95% CI, 0.82–1.13], respectively); however, tests of interaction did not reach statistical significance (P interaction = .2). Similarly, hypothyroidism overall was associated with greater all-cause mortality compared with euthyroidism in participants with atherosclerotic disease but not in those without (aHR = 1.30 [95% CI, 1.02–1.66] and 0.93 [95% CI, 0.81–1.08]); however, tests of interaction did not reach statistical significance (P interaction = .08).