Impaired Sensitivity to Thyroid Hormones Is Associated With Diabetes and Metabolic Syndrome

Recent evidence suggests that hyperthyroxinemia may be cross-sectionally and prospectively associated with type 2 diabetes (1). This is at odds with the physiological metabolism-activating action of thyroid hormones, which is considered capable of ameliorating obesity-related morbidities (2). It is also in conflict with the increased incidence of diabetes reported among patients with hypothyroidism (3) and with the increased prevalence of metabolic syndrome described in the upper-normal range of the pituitary thyroid-stimulating hormone (TSH) (alsoknown as thyrotropin) (4), a range associated with lower thyroid activity. Previous evidence is inconsistent (4): some studies find diabetes or metabolic syndrome to be associated with TSH but not with thyroid hormones, whereas others find an association only with thyroid hormones, occasionally linking opposite ends of the thyroid hormone concentration range with diabetes. The effect of thyroid hormones on insulin sensitivity differs by tissue; it enhances glucose uptake in the muscle but reduces it in the liver (5). The overall net effect in hypothyroidism favors insulin resistance, but in hyperthyroidism it may also occasionally favor glucose intolerance (5). These metabolic mechanisms described in clinical thyroid diseases may not be enough to explain the associations found within the normothyroid range.

Physiologically, thyroid hormones and TSH are inversely correlated owing to a negative feedback loop. However, high thyroid hormones coexist with high TSH in individuals with resistance to thyroid hormone syndrome, an inherited autosomal recessive trait (6–8). Yet, a possibly reversible acquired resistance to thyroid hormone, due to homeostatic compensation, has been hypothesized (9). Prolonged fasting produces drops in TSH and increases in pituitary sensitivity to thyroid hormones (10,11). In contrast, morbidly obese individuals tend to have higher levels of both thyroid hormones and TSH (12). Resistance to thyroid hormone manifestations can be systematized into central resistance phenomena, which affect the feedback loop set point in the central nervous system, and peripheral resistance phenomena, which decrease hormones’ metabolic effects. The former are easier to evaluate than the latter because they can be quantified just observing thyroid hormones and TSH concentrations or indices derived from them (13,14), although suppression tests have also been described (15).

Thus, given the thyroid feedback loop, the aforementioned reports showing association of high thyroid hormones or high TSH with the prevalence and incidence of metabolic syndrome and diabetes provide apparently contradicting results. In this work, we hypothesized that these conflicting results might be conciliated if high thyroid hormones and high TSH co-occurrence in diabetes reflected a central resistance to thyroid hormone. If that were the case, this central resistance would probably be the expression of a general reduced sensitivity to thyroid hormones, i.e., not only central but also peripheral. This resistance to thyroid hormone could be one additional characteristic of the cardiometabolic traits predicting diabetes. We analyzed continuous National Health and Nutrition Examination Survey (NHANES) (16) data to gain insight into this issue and propose the Thyroid Feedback Quantile-based Index (TFQI), a new resistance to thyroid hormone index focused on deviations close to normality.

Thyroid hormones and TSH were measured in serum in 5,222 (N₁) adults 20 years of age and older in the continuous NHANES (16) performed in 2007 and 2008, which recruited a representative sample of the noninstitutionalized U.S. population. Data on mortality up to 2011 from the National Death Index are available for all but six individuals. The current study estimates cross-sectional associations between resistance to thyroid hormone indices and diabetes-related conditions as well as longitudinal associations between the indices and the diabetes-related mortality rate. Analyses were performed using the entire NHANES data set (N₁) and several subsets (main complete case sample, N₂ = 5,129; normothyroid sample, N₃ = 4,750; and fasting normothyroid sample, N₄ = 1,997) (see flowchart in Supplementary Data). NHANES participants provided written informed consent to participate in the survey, which was approved by the National Center for Health Statistics (NCHS) Research Ethics Review Board (Continuation of Protocol no. 2005-06).

Free thyroxine (fT4) was quantified with the Access Free T4 assay, a two-step enzyme immunoassay, and TSH was quantified with the Access HYPERsensitive hTSH Assay, a two-site immunoenzymatic (“sandwich”) assay. Measurements were performed at the University of Washington (Seattle, WA). The laboratory normality reference ranges were 7.74–20.64 pmol/L for fT4 and 0.34–5.60 mIU/L for TSH.

Thyrotroph T4 Resistance Index (TT4RI) was calculated as fT4 (pmol/L) ⋅ TSH (mIU/L) (13). TSH index (TSHI) was calculated as ln TSH (mIU/L) + 0.1345 ⋅ fT4 (pmol/L) (14).

Given encouraging preliminary results from the description of participants who were simultaneously in the extreme quartiles of fT4 and TSH (see Statistical Analyses and Results sections), we developed a new index to quantify, in a continuous manner, deviations from the median pituitary response (inhibition) to thyroid hormones: the TFQI, which stands for thyroid feedback quantile-based index.

Ranks (order position from minimum to maximum value) of fT4 and TSH were converted to quantiles between 0 and 1, taking into account sampling weights. Fractional quantiles are a generalization that summarizes in a single fraction the ordering number of each part in the sequence and the number of parts, e.g., the upper limit of the second tertile is the value of the quantile 0.66 and the upper limit of the third quartile is the value of the quantile 0.75. That conversion is achieved by applying the population empirical cumulative distribution function (cdf) to hormone concentration. TFQI was calculated as cdf fT4 − (1 − cdf TSH), i.e., the difference between fT4 quantile and the reversed TSH quantile, because they are inversely correlated in the negative feedback loop. This index ranges between −1 and 1. Negative values indicate lower TSH (higher inhibition by fT4) than that expected for the actual fT4 (which means higher sensitivity to fT4). Similarly, positive values indicate higher TSH (lower inhibition by fT4) than that expected for the actual fT4 (lower sensitivity to fT4). Median TSH response to fT4 is represented by the value 0. This index is more stable than TT4RI and TSHI at abnormal ranges of the thyroid gland responsiveness to TSH. That is, TFQI does not reach extreme values in individuals with thyroid gland–caused (primary) clinical hyper- or hypothyroidism (see data points in the top-left and bottom-right areas of TSH versus fT4 scatterplots in Supplementary Fig. 1).

In order to provide an index that can be calculated for any new value or adapted to other populations, an approximation with the same range and interpretation, the Parametric Thyroid Feedback Quantile-based Index (PTFQI) can be obtained from fT4 in pmol/L and TSH in mIU/L using the standard normal cumulative distribution function as follows: Φ((fT4 − μ_fT4)/σ_fT4) − (1 − Φ((ln TSH − μ_{ln TSH})/σ_{ln TSH})), where μ_fT4 = 10.075, σ_fT4 = 2.155, μ_{ln TSH} = 0.4654, and σ_{ln TSH} = 0.7744 for the U.S. population. This can be easily achieved with the Excel or LibreOffice spreadsheet formula =NORM.DIST(fT4_cell,10.075,2.155,TRUE)+NORM.DIST(LN(TSH_cell),0.4654,0.7744,TRUE)−1.

Demographic (including race/ethnicity as nonhispanic white, nonhispanic black, and other), clinical interview, physical examination, and other laboratory variables were obtained following protocols available in NHANES operation manuals (16). Briefly, glucose was measured with the glucose oxidase method and glycohemoglobin (HbA_1c) on the A1c G7 HPLC Glycohemoglobin Analyzer (Tosoh Medics, Inc., San Francisco, CA). Blood was drawn from participants who were visited at any time throughout the day, except for a subsample of participants appointed for a morning visit and requested to fast. In the main sample data set (which includes nonfasting data), high HbA_1c (≥6.5%, i.e., ≥48 mmol/mol) and use of diabetes medication were analyzed, as they can be interpreted as proxies for diabetes. In the fasting subsample, diabetes was defined as fasting (≥8 h) glycemia ≥126 mg/dL or presenting the previously mentioned proxies. Obesity was defined as a BMI ≥30 kg/m². Metabolic syndrome diagnosis was established using the harmonized criteria (17); i.e., diagnosis was issued when meeting three of the following criteria: 1) waist circumference ≥102 cm in men or ≥88 cm in women; 2) triglycerides ≥150 mg/dL; 3) HDL cholesterol <40 mg/dL in men or <50 mg/dL in women; 4) systolic or diastolic blood pressure ≥130 mmHg or ≥85 mmHg, respectively (or receiving medication for hypertension); and 5) glucose ≥100 mg/dL (or receiving medication for diabetes).

Survey data were linked with the National Death Index by the NCHS. Besides the leading cause of death, records also include contributing causes of death in a 20 entity axis set of codes for multiple causes of death. The public use linked mortality data set provided by the NCHS includes a flag variable for diabetes when ICD-10 codes E10–E14 appear as one of the 20 contributing or multiple causes of death. We used this flag, i.e., diabetes-related death, as our end point.

Mean and SD of thyroid parameters were calculated for population sex, age, and race strata. TSH and TT4RI were processed in the logarithmic scale given their skewed distributions. We created scatterplots of TSH versus fT4 and plotted population-based percentiles of the indices.

The interesting results observed when we compared participants who were either at the lowest or at the highest quartiles of fT4 and TSH simultaneously (see Supplementary Data and Results) led us to develop TFQI and PTFQI (described above). These indices enabled us to study pituitary feedback response with a continuous variable.

We created population-based quartiles for each resistance to thyroid hormone index (TFQI, PTFQI, TT4RI, and TSHI). Differences, odds ratios (ORs), and rate ratios (RRs) across quartiles, using the lowest as reference, were estimated with generalized linear, logistic, and Poisson regression models, respectively. Model 1 adjusted for sex, age, and race. Model 2 further adjusted for BMI. Model 3 further adjusted for initially presenting HbA_1c ≥6.5% and diabetes medication use. For each trend, significance was tested with each index quartile ordinal as a continuous variable. Given that there were no diabetes-related deaths within the first quartile of three indices, this outcome was compared between values above and below the median, and trend was tested with each index as a continuous variable.

In the main sample and the normothyroid sample, interquartile comparisons were performed for BMI, obesity, HbA_1c ≥6.5%, diabetes medication use, and diabetes-related mortality. BMI was modeled with linear regression; obesity, HbA_1c ≥6.5%, and diabetes medication were modeled with logistic regression, and death was modeled with Poisson regression.

In the fasting subsample, ORs for diabetes, metabolic syndrome, and metabolic syndrome criteria were calculated for 1-SD increase of resistance to thyroid hormone indices except for TT4RI. This index was analyzed using the logarithmic scale, and its OR was calculated for a TT4RI multiplicative increase by an SD equivalent (× 2^{SD log2 TT4RI}). Diabetes, metabolic syndrome, and metabolic syndrome criteria were modeled with logistic regression.

As part of sensitivity analyses, we further adjusted models for education level, physical activity performed at work or during leisure time, and daily sedentary time. All analyses were performed with statistical computing software R (18), version 3.4.4, with the package “survey” (19) to account for the NHANES complex survey design.

The 5,222 participants in the sample represented the U.S. population 20 years of age and older, and they had a mean (SD) age of 46.9 (16.6) years (Table 1). Indices of resistance to thyroid hormone were higher among older people and lower among black individuals (Table 1).

In a preliminary analysis, obesity prevalence and, especially, HbA_1c ≥6.5% and use of diabetes medication showed a tendency to be higher among participants whose fT4 and TSH were simultaneously above their 75th percentiles (Supplementary Data).

A scatterplot of TSH versus fT4 shows the negative correlation of these variables, while the contour lines show selected percentiles for each one of the resistance to thyroid hormone indices (Fig. 1 and Supplementary Fig. 1). TT4RI and TSHI (Supplementary Fig. 1) take extreme values (>95th and <5th percentile) for almost all patients diagnosed with clinical hypo- and hyperthyroidism. This seemingly inappropriate result together with our observation that a quantile-based approach (Supplementary Data) uncovered the above-mentioned associations and provided more moderate values among patients with clinical thyroid gland–caused (primary) disease is what led us to develop TFQI.

BMI, obesity prevalence, HbA_1c ≥ 6.5%, and use of diabetes medication were progressively higher across TFQI quartiles after adjustment for sex, age, and race (all P < 0.01). ORs of the fourth versus the first TFQI quartile for HbA_1c ≥6.5% and use of diabetes medication were 1.73 (95% CI 1.32, 2.27) and 1.66 (1.31, 2.10), respectively, and the association was independent of BMI, given that after adjustment, OR estimations were only partially reduced to 1.43 (1.07, 1.92) and 1.39 (1.10, 1.75), respectively (P_trend <0.05) (Table 2). Participants had a median follow-up of 3 years and 11 months, during which there were 27 deaths with diabetes listed within the entity axis codes for multiple causes of death. In nine of these cases, diabetes was the leading cause of death. For participants with TFQI above the median (versus below it), sex-, age-, and race-adjusted RR for diabetes-related death was 4.81 (95% CI 1.01, 22.94). TFQI was linearly associated with the rate for diabetes-related death (P = 0.015), and the association remained statistically significant even after adjustment for BMI (P = 0.035). One-fourth of deaths occurred among participants without initial presence of proxies for diabetes. However, in spite of a similar magnitude of association for being above the median, RR 3.51 (0.78, 15.85), such a small number of events made it impossible to detect a statistically significant association after adjustment for the initial condition. All results were similar in examination of PTFQI instead of TFQI (Supplementary Data). Adjustment for education, physical activity, and sedentary time did not substantially change these results (Supplementary Data).

TT4RI and TSHI were cross-sectionally associated with obesity, although not with diabetes proxies. Nonetheless, both were prospectively associated with diabetes-related mortality independently of initial presence of diabetes proxies (Supplementary Data). All of these results persisted in the normothyroid subsample (Supplementary Data) with the exception of RR for mortality. Although the magnitudes for its indicators were very similar, they failed to reach statistical significance, probably due to the statistical loss of power caused by excluding two deceased participants who had marginally subclinical hypothyroidism (Fig. 1 and Supplementary Data).

We tested cross-sectional associations of the indices with diabetes and metabolic syndrome in the fasting subsample of normothyroid participants. Only TFQI (and its approximation, PTFQI) was associated with diabetes based on glycemia, HbA_1c, and medication. All resistance to thyroid hormone indices were associated with metabolic syndrome in general and with HDL and blood pressure metabolic syndrome criteria. Lastly, TT4RI and TSHI were also associated with triglyceride criterion (Table 3).

Based on a sample representative of the U.S. population, we provide evidence of the association between indices measuring resistance to thyroid hormone and prevalence of obesity, diabetes, and metabolic syndrome as well as between these indices and the incidence of diabetes-related deaths. These associations were also identified within the normal fT4 and TSH ranges. These results not only reconcile previous reports linking diabetes to hypothyroidism (3) and to hyperthyroxinemia (1) but also call for a new interpretation of thyroid hormone changes associated with diabetes.

As far as we are aware, this is the first time that indices of resistance to thyroid hormone have been evaluated in the general population and associated with cardiometabolic health characteristics. Our analyses showed that TT4RI and TSHI indices seemed to systematically classify thyroid gland–caused (primary) hypothyroidism and hyperthyroidism (clinical and subclinical) as resistant or very sensitive to thyroid hormones, respectively. Thus, we proposed a new index, TFQI. TFQI is based on the empirical joint distribution of fT4 and TSH with the advantage of not yielding extreme values in cases of thyroid gland dysfunction. Its parametric version (PTFQI) can readily be used as a formula to calculate the index for any particular individual with reference to the U.S. population, or it can be adapted to other reference populations.

All these resistance to thyroid hormone indices measure central sensitivity/resistance, i.e., the grade of pituitary gland inhibition by fT4 levels. Thus, they evaluate the set point of the central regulation of thyroid hormone concentration. Among participants with high values, peripheral resistance could also be present because, despite higher fT4, we find in this group higher prevalence of obesity, diabetes, and metabolic syndrome, phenotypes usually associated with hypothyroidism.

Thyroid hormones increase energy expenditure and thermogenesis (20), increase glucose and fatty acid oxidation in muscle (21) and the liver (22), increase adipose tissue lipolysis (23), and promote lower body weight (24). In sum, they have an overall effect on metabolism that would prevent diabetes development. Interestingly, thyroid hormones also sensitize to catecholamines, which increase liver hepatic glycogenolysis, neoglucogenesis, and glucose output (25); reduce insulin anabolic actions in the liver; and increase glucose intestinal absortion (5), though the latter effect plays a minor role in glucose level derangement. In overt hyperthyroidism, these changes could be responsible for decompensation of preexisting diabetes and even trigger ketoacidosis (25), but they may be less important within the normal range of thyroid hormones and unlikely to cause diabetes onset in a previously metabolically normal individual.

Besides secretion regulation by the hypothalamic-pituitary-thyroid axis, thyroid hormone action is modulated (6,26) at cell membrane transport, enzymatic modification into active and inactive species by deiodinases, and nuclear receptors. Nutritional signals can modulate the thyroid system during some of the regulation, secretion, and action steps. Hypothalamic neurons inhibit secretion when leptin levels fall, providing a system to protect against starvation (27). Meanwhile, an increase in leptin derived from adipose tissue accumulation may indicate availability of energetic substrates, and, thus, the subsequent activation of the thyroid axis could also be interpreted as a homeostatic regulation. Accordingly, high TSH—within normal levels—is associated with increased adiposity (28) and consistent with near-hypothyroidism hormonal end effects. This can be interpreted as a deficient thyroid hormone production in relative terms, unable to meet the metabolic increase required to compensate and avoid more fat accumulation. Interestingly, in morbidly obese individuals thyroid hormones are also elevated (12), hinting at a downstream signal problem. Activity of deiodinases (29), epigenetic modification of histones, and interaction with signaling of other transcription factors (30) also regulate, at the cellular level, thyroid hormone actions on metabolism. In addition, in obesity a steady consumption of high-calorie foods may even overpower the thyroid stimulus that increases resting metabolic rate (31), leading to a biochemical phenotype compatible with an apparent resistance to thyroid hormone. We previously proposed a similar interpretation for the origin of insulin resistance in obesity (32).

From a clinical point of view, there seems to be a gradient of increasing thyroid hormones levels—within the normal range—across early stages of diabetes (33). Furthermore, prospectively, higher levels of thyroid hormones are associated with incident diabetes (1) in spite of the fact that hypothyroidism is the disorder that clearly increases diabetes risk (3). As resistance to thyroid hormone is one of the differential diagnoses when both fT4 and TSH are elevated (34), our results offer an explanation for thyroid profiles commonly found in patients with diabetes. That is, at the population level, measurements of resistance to thyroid hormone are cross-sectionally associated with diabetes, independently of the degree of obesity as measured with BMI, suggesting that there might be other underlying mechanisms linking diabetes and resistance to thyroid hormone. Furthermore, prospectively, these measurements were also associated with diabetes-related deaths, even independently of the initial diabetes status.

Measurements of resistance to thyroid hormone were also cross-sectionally associated with metabolic syndrome on a consistent basis. Previous attempts to relate metabolic syndrome with the thyroid axis either showed an association with TSH but not with fT4 (35) or yielded inconsistent results. We hypothesized that whereas a metabolic status with excess energy may stimulate the thyroid axis, the thyroid boost to energy expenditure fails to compensate. This results in an expression compatible with resistance to thyroid hormone. Thyroid stimuli still increase, although insufficiently, the uncoupled thermogenic oxidative processes, which can be observed as an increase of oxidative stress in metabolic syndrome (36). A sustained compensatory TSH stimulus could also explain, in part, the recently reported association of metabolic syndrome, insulin resistance, and diabetes with thyroid enlargement, thyroid nodules, and thyroid cancer (37). We cannot exclude that other conditions, like immunity disorders, could also explain a coincidence of elevated thyroid hormones, due to autoimmune thyroid disease, and latent autoimmune diabetes of adults. However, from a population perspective, energy imbalance is much more prevalent.

Type 2 diabetes requires a broad clinical management beyond glucose control, focused on energy balance. Monitoring resistance to thyroid hormone could be used to evaluate the intended reversal of the abnormal energy balance. Interestingly, where traditional interventions of food intake reduction and physical exercise increase seem to fail, new treatments trying to modify the activity of uncoupling proteins have been proposed (2,38,39).

This study is based on a representative sample of the U.S. population with strict data collection protocols. Consistency in results between cross-sectional associations and longitudinal mortality analyses support our conclusions. However, some limitations exist. Indices of resistance to thyroid hormone increase markedly with aging. Whereas all analyses are adjusted for age and other potential confounders, some residual confounding may remain. Given that many factors may be associated with diabetes and metabolic syndrome, it is not possible to adjust for unavailable variables and unknown factors, and the potential residual confounding must be considered when interpreting the study results. One antidiabetes oral agent, metformin, lowers TSH (leaving fT4 unchanged) among patients with hypothyroidism (37), and there is some evidence hinting that it sensitizes the pituitary gland and peripheral tissues to thyroid hormone action (40). Thus, our results may underestimate the true association between resistance to thyroid hormone and diabetes; i.e., the association may be even stronger than the statistically significant association reported here. Because our analysis focused on resistance to thyroid hormone indices, which are all fT4 based, fT3 was intentionally excluded from the models, since it was considered to be downstream in the action pathway. Actually, deiodinases balance may play a role in the modulation of sensitivity to thyroid hormones (29), and future projects should be devoted to specifically study their role in resistance to thyroid hormone in obesity, metabolic syndrome, and diabetes. Diabetes and related outcomes were based on one single measurement of HbA_1c or fasting plasma glucose, which is a common limitation in population-based studies. Finally, the small number of observed diabetes-related deaths limits the statistical power needed to find subtle effects as well as to perform subgroup analyses. Notwithstanding, the association was strong enough to bear statistically significant results.

Higher values in indices of resistance to thyroid hormone are associated with obesity, metabolic syndrome, and diabetes for the entire population in general and also for normothyroid individuals in particular. Higher values are also prospectively associated with diabetes-related mortality. These associations with diabetes morbidity and mortality were independent of obesity. We propose the new TFQI to identify mild levels of acquired resistance in the general population. Type 2 diabetes is in part driven by an energy balance problem, and the associated central resistance measured by indices of resistance to thyroid hormone may be the result of the physiologic contraregulation against it. These indices may facilitate the monitoring and evaluation of new therapies that focus on energy expenditure.

Funding. The research activity of M.L. is funded by Agencia Aragonesa para la Investigación y el Desarrollo (ARAID). This work was supported by grants PI14/00009 and PI17/01709 from Instituto de Salud Carlos III (Ministry of Science, Innovation and Universities), and cofunded by European Regional Development Fund/European Social Fund “Investing in your future.”

None of the funding sources had any role in the writing of the manuscript or the decision to submit it for publication.

Duality of Interest. No potential conflicts of interest relevant to this article were reported.

Author Contributions. M.L. conceptualized the research question and designed the study. M.L. and B.M.-F. performed the statistical data analysis. M.L., B.M.-F., J.M.L.-B., and F.C. interpreted the results. M.L. and B.M.-F. wrote the manuscript. R.M.-G., J.A.C., P.G.-C., A.C., and F.C. revised the manuscript for important intellectual content. M.L., B.M.-F., J.M.L.-B., R.M.-G., J.A.C., P.G.-C., A.C., and F.C. approved the final version of the manuscript for publication. M.L. is the guarantor of this work and, as such, had full access to all the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.