Thyroid-Stimulating Hormone Levels and Incident Depression

Results From the ELSA-Brasil Study

Ana C. Varella; Isabela M. Benseñor; Carolina C.P.S. Janovsky; Alessandra C. Goulart; Marina G. Birck; Itamar S. Santos; Andre R. Brunoni; Paulo A. Lotufo


Clin Endocrinol. 2021;94(5):858-865. 

In This Article


Study Design

The Brazilian Longitudinal Study of Adult Health (ELSA-Brasil) is a cohort of 15,105 civil servants between the ages of 35 and 74 years, from six Brazilian cities (Porto Alegre, Rio de Janeiro, Salvador, São Paulo, Belo Horizonte and Vitória) who were monitored over a period of time and evaluated periodically. They were enrolled between August 2008 and December 2010.[13–15] The second visit took place between 2012 and 2014, four years after the baseline evaluation, with new examinations and collection of biological samples using the same ELSA-Brasil protocols and quality control.

All participants between the ages of 35 and 74 years were eligible to participate, except those who intended to leave the institution at any time in the near future, had severe cognitive or communication impairment, current or recent pregnancy (<4 months prior to the first interview) and, if retired, those living outside the metropolitan area of the study centre. The questionnaire covered a wide range of health-related topics such as, socio-demographic factors, lifestyle, morbidity, cardiovascular risk factors, cognitive function, mental health and others.[16] Participants were also asked about any regular use of medication over the previous two weeks and were instructed to bring their prescriptions and drugs used to the study centre. The study was approved at all six centres by each Institutional Review Board and all participants signed a written informed consent form.

Evaluation of Thyroid Function

Venous blood samples were drawn after an overnight fast and centrifuged at 2500 g for 15 minutes to obtain the serum for biochemistry and determine hormone levels. TSH and free-thyroxine (FT4) levels were determined by a third-generation immunoenzymatic assay (Siemens, Deerfield, IL, USA). Single TSH and FT4 measurements were performed at baseline. Thyroid dysfunction was defined by TSH and FT4 levels, by routine use of thyroid hormones or anti-thyroid medication, such as propylthiouracil or methimazole. Reference ranges for normal TSH levels were 0.4–4.0 mIU/L, and FT4 levels were 10.3–24.45 pmol/L.[17] Participants were classified in the following subgroups based on TSH and FT4 levels and use of thyroid medication. Overt hyperthyroidism (low TSH, high FT4 or use of hyperthyroidism medication), subclinical hyperthyroidism (low serum TSH, normal FT4 and no use of thyroid medication), euthyroidism (normal TSH and no use of thyroid medication), overt hypothyroidism (high TSH, low FT4 or use of levothyroxine) and subclinical hypothyroidism (high TSH, normal FT4 levels and no use of thyroid medication). We excluded participants who were using drugs that could have altered thyroid function or interfered with tests (eg amiodarone, carbamazepine, oxcarbazepine, carbidopa, furosemide, haloperidol, heparin, levodopa, lithium, metoclopramide, phenytoin, propranolol, primidone, rifampicin, valproic acid, sodium valproate or divalproex sodium).

Evaluation of Depression

Depression was assessed by the validated Portuguese version of the Clinical Interview Schedule – Revised (CIS-R). It consists of a structured interview for the measurement and diagnosis of non-psychotic psychiatric morbidity in the community. It was developed by Lewis et al. in 1992.[18] The complete CIS-R version includes 14 sections covering symptoms of depression and anxiety: obsessions, compulsions, panic, phobias, anxiety, worry, worry about physical health, depression, depressive ideas, irritability, fatigue, concentration, sleep and somatic symptoms. The CIS-R also yields International Classification of Diseases (ICD-10) diagnoses. The relevant symptoms reported by the participant are grouped to form ICD-10 diagnoses. It permits several psychiatric diagnoses as follows: depression (MDD, F32.xx), general anxiety disorder (GAD, F41.1), panic disorder (F41.0), obsessive-compulsive disorder (OCD, F42) and mixed anxiety and depressive disorder (MADD, F41.2) in participants with symptoms of anxiety and depression that did not fulfil isolated criteria for MDD or GADs. CIS-R can also be computed as a score with a range of 0 to 57. People are classified as having common mental disorders (CMD) if they present a score ≥12. It can be applied by trained lay-interviewers, making it appropriate for epidemiological studies. At ELSA-Brasil, it was applied by clinical research staff with special training both at baseline and follow-up. The specific section built to diagnose depression contained questions addressing depression, depressive ideas, fatigue, concentration/forgetfulness and sleep disturbance. If the participant fulfils the criteria for depression according to ICD-10, it was classified as depressed and the participants that did not fulfil the ICD-10 criteria as non-depressed in a binary way.

To investigate de association between TSH levels at baseline and incident depression, we used data from the CIS-R applied at baseline and over a 4-year follow-up. The results of CIS-R at baseline were used to rule out prevalent cases of depression. Therefore, in the longitudinal analysis only incident cases of depression were included.

Study Population

Of the 14,014 participants examined at follow-up, we excluded those with depression diagnosis at baseline (n = 592) and those with no information on depression diagnosis at any visit (n = 88). We also excluded those using medication known to alter thyroid disease (n = 334), those missing thyroid function data (n = 41) as well as participants with overt thyroid disease (n = 1057). Finally, we excluded TSH outliers (n = 6), resulting in a sample of 11,896 participants for analysis.

It is important to note that ELSA-Brasil is a cohort of healthy workers. In addition, in the event of participants coming to the research centre complaining of fever, pain or any acute symptom, they would have been rescheduled. Also, frequency of cardiovascular diseases and cancer at baseline were low. Participants undergoing chemotherapy, radiation protocol or any other treatment for cancer were not included in the sample. Therefore, results of thyroid tests are not biased by other systemic diseases or acute infections.

Other Variables

The covariates used in this study were sex, age (years), race (White, Mixed, Black and Asian and Indigenous), education (lower than High School degree, High School degree and College degree), family income (less than or over 20,000 dollars per year, at an exchange rate of US$ 1.00 = R$ 2.00), smokers and alcohol consumption (defined as never, former and current use), the use of antidepressants and benzodiazepines, body mass index (BMI) (calculated by dividing the weight in kilograms by the height in metres squared, measured according to standard protocols),[19] also, kidney function and comorbidities. The glomerular filtration rate (GFR, mL/min/1.73 m2) was used to assess kidney function and was estimated by the Chronic Kidney Disease Epidemiology Collaboration (CKD-EPI) method.[20] Diabetes (classified as self-reported previous diagnosis, medication use, fasting glucose ≥7.0 mmol/L, an oral glucose tolerance test ≥11.1 mmol/L and/or glycated haemoglobin ≥6,5%; ≥47.5 mmol/mol),[21] cardiovascular diseases (cardiac surgery and/or myocardial infarction), stroke and any type of cancer.

Statistical Analysis

TSH levels were categorized in quintiles: Q1 (0.0–0.96), Q2 (0.97–1.34), Q3 (1.35–1.78), Q4 (1.79–2.58), Q5 (2.59–35.5). Sample characteristics were presented according to TSH levels. Categorical variables were compared using chi-square test and presented as counts and proportions. Continuous variables were compared using ANOVA and presented as means and standard deviations (± SD).

Our main analysis was to evaluate the association between TSH levels at baseline and incident depression among participants with subclinical thyroid disease or normal thyroid function together. A secondary analysis was conducted to evaluate TSH levels and depression in euthyroid participants only. All analyses were stratified by sex since prevalence of thyroid diseases and depression is known to be higher in women compared with men.[12]

In our main analyses, Poisson regression models with robust variance estimator were built using quintiles of TSH as the main independent variable with the 3rd quintile as reference and incident depression as the dependent variable. To account for possible confounders, two adjusted models were built. Model 1 was adjusted for age and sex, and Model 2 was further adjusted for race, education, BMI, smoking, alcohol consumption, use of antidepressants/benzodiazepines, kidney function and comorbidities (diabetes, stroke, cardiovascular disease[22] and cancer). Results were presented as relative risks (RR) with their respective 95% confidence intervals (95% CI).

We also performed some supplementary analyses shifting the reference category from the 3rd to 1st TSH quintile to verify the presence of any possible associations between higher TSH levels and incidence of depression using 1st TSH quintile as the reference group. We conducted this analysis among participants with subclinical and normal thyroid function together and restricted it to euthyroid individuals.

Furthermore, a sensitivity analysis of those with TSH levels ≥ 10 mIU/L (among all participants and also women only) was performed. It is worth noting that due to low statistical power, we were unable to run analysis of TSH levels above 10 mIU/L in men due to the small number of individuals with high TSH levels in this category.

Finally, we conducted supplementary analysis categorizing participants as having subclinical hyperthyroidism, subclinical hypothyroidism or euthyroidism. Poisson regression models with robust variance estimator were applied using subclinical thyroid dysfunction as the main independent variable and incident depression as the dependent variable. Adjustments were the same as those used in the main analyses. These analyses were also stratified by sex. Importantly, we analysed subclinical hyperthyroidism in the women only, because no men had both subclinical hyperthyroidism and incident depression. All analyses were carried out using the Statistical Package for Social Sciences (SPSS) version 22, with level of significance set at 0.05.