Thyroid-Stimulating Hormone and Thyroid Hormones and Incidence of Diabetes: Prospective Results of the Brazilian Longitudinal Study of Adult Health (ELSA-BRASIL)

Background

Thyroid dysfunction and type 2 diabetes are prevalent endocrine disorders. The high frequency of thyroid dysfunction in people with diabetes and of diabetes in those with thyroid dysfunction suggests that both diseases share underlying mechanisms (1,2). Besides, diabetes is characterized by hyperglycemia from impaired insulin secretion and/or insulin resistance, and thyroid hormones (TH) influence insulin and glucose homeostasis (1). Subclinical hyperthyroidism has been associated with increased hepatic gluconeogenesis (3), increased glucose response after an oral glucose tolerance test (4), insulin degradation (5), and resistance (4,6). On the other hand, subclinical hypothyroidism is associated with decreased use of peripheral glucose (7), higher glucose levels (8), decreased glucose oxidation/glycogen synthesis (9), and insulin resistance as well (10,11).

Few studies explored prospectively the relationship between thyrotropin (TSH) and TH with incident diabetes. The Rotterdam study showed an association between higher TSH and lower free thyroxine (fT4) levels with incident diabetes (12), but these findings were not confirmed in another study in the country (13). Two large cohort studies in Korea did not find an increased risk of diabetes (14,15), but they reported that higher increases of TSH and TH levels during the follow-up were associated with increased risk of diabetes (14,15). Finally, a large Chinese study found a significant association between the highest TSH quartile and metabolic syndrome, but not with diabetes (16).

In clinical practice, thyroid function is determined by TSH and fT4 levels. Free triiodothyronine (fT3) is less used, mostly to identify hyperthyroidism with fT4 in the normal range. fT3:fT4 ratio is not used in clinical practice, but has been used in studies to estimate deiodinase activity. There is scarce information about fT3:fT4 ratio and the risk of diabetes (17). As obesity leads to an increased conversion rate, the hypothesis of a positive association between fT3:fT4 ratio and incident diabetes seems reasonable.

This study investigated the prospective association of TSH, TH, and fT3:fT4 ratio with the risk of developing diabetes using data from the Brazilian Longitudinal Study of Adult Health (ELSA-Brasil), a highly admixed sample of European, African-American, and native Brazilian descendants, who were not evaluated in previous studies. We hypothesized that the relationship of TSH and TH with incident diabetes is a U-shaped curve with both low and high levels associated with diabetes.

Methods

Thyroid function and diabetes

Venous blood samples were drawn after an overnight fast from 6.30 am to 9.00 am. TSH (normal range: 0.4–4.0 mIU/L), fT4 (0.93–1.7 ng/dL), and fT3 (0.20–0.44 ng/dL) were determined by a third-generation immunoenzymatic assay (Roche Diagnostic, Manheim, Germany). The analysis included euthyroid participants (TSH levels 0.4–4.0 mIU/L and no use of levothyroxine/anti-thyroid drugs) and participants with subclinical hyperthyroidism (TSH levels <0.4 mIU/L and fT4 levels 0.93–1.7 ng/dL) and subclinical hypothyroidism (TSH >4.0 mIU/L with fT4 0.93–1.7 ng/dL). Participants with overt thyroid diseases were excluded (22). We excluded participants using medication that alters thyroid function (amiodarone, carbamazepine, carbidopa, furosemide, haloperidol, heparin, levodopa, lithium, metoclopramide, phenytoin, propranolol, primidone, rifampicin, steroids, and valproic acid) and those with diabetes at baseline from the analysis (Fig. 1).

OPEN IN VIEWER

FIG. 1.

Study sample flowchart. HyperSC, subclinical hyperthyroidism; HypoSC, subclinical hypothyroidism.

The definition of diabetes included a self-reported medical diagnosis of diabetes, use of drugs to treat diabetes, fasting plasma glucose levels ≥7.0 mmol/L, 2-h glucose levels ≥11.1 mmol/L, or HbA1c ≥6.5%. Prediabetes was defined according to the ADA (23).

Results

The baseline characteristics of 7948 participants (7177 [90.3%] euthyroid; 726 [9.1%] with subclinical hypothyroidism; and 45 [0.6%] subclinical hyperthyroidism) are shown in Table 1. The mean age of participants was 50.2 (SD 8.6) years, 54.4% female, being mostly white (53.6%). The most frequent cardiovascular risk factors were dyslipidemia (43.1%) and hypertension (28%). Additional description of participants' baseline characteristics by thyroid status (Supplementary Table S1) and according to quintiles of TH and fT3:fT4 ratio is shown in Supplementary Tables S2–S4.

After a mean follow-up of 8.2 (SD 0.53) years, the incidence of diabetes was 14.8%. Participants who developed diabetes were slightly older, less educated, or physically active at baseline compared to those who did not develop diabetes (Table 2).

Using the third quintile as reference, no association was found for TSH, TH, and fT3:fT4 ratio with the risk of developing diabetes after multivariable adjustment. However, the increase in 1-unit (1-U) of fT4 decreased the risk of diabetes (RR 0.94 [CI 0.91–0.99]), while the increase of 1-U of the fT3:fT4 ratio increased risk (RR 1.37 [CI 1.15–1.63] of diabetes (Table 3). The sensitivity analysis considering participants with diabetes at the 4-year follow-up confirmed in the 8.2-year follow-up was nonsignificant (Supplementary Table S5).

Participants with overweight or obesity showed a significantly higher risk of diabetes in the lowest quintile of fT4 (RR 1.04 [CI 1.01–1.07]) using the third quintile as reference. In addition, in the same subset of participants, low fT3:fT4 ratio presented a modest inverse association (RR 0.96 [CI 0.93–0.98]) compared to those within the normal-weight range using the same reference. In the analyses stratified by BMI, participants with overweight or obese presented both an inverse association for diabetes for a 1-U increase of fT4 and an increased risk for diabetes at each 1-U increase of fT3:fT4 ratio (Table 4).

Incidence of prediabetes after the 8.2-year follow-up was 1657 (53.6%) cases. No association was found between TSH, TH, and fT3:fT4 ratio with the incidence of prediabetes (Supplementary Table S6). Analyses stratified by sex (Supplementary Table S7) showed no association.

Discussion

In this prospective study, 14.8% of participants euthyroid or with subclinical thyroid dysfunction, and without diabetes at baseline, developed diabetes over an 8.2-year follow-up. We did not confirm our hypothesis that the relationship of TSH and TH with incident diabetes presents a U-shaped curve (reference in the third quintile), but using fT4 and fT3:fT4 ratio as continuous variables, we found a slight inverse association for diabetes and an increased risk of diabetes with the increase in each unit of fT4 and the conversion ratio, respectively.

A sensitivity analysis performed among participants with overweight or obese showed an association of low fT4 levels (first quintile) with a modest incidence of diabetes; and low levels of fT3:fT4 ratio were associated with a modest inverse association for diabetes. Our results using continuous variables are in line with the findings in quintiles: each 1-U increase in fT4 was associated with a modest inverse association with the risk of diabetes, and a 1-U increase in the conversion ratio increases the risk of diabetes. Both results using quintiles and continuous variables highlight the effect of BMI on the association of TH with incident diabetes.

Our most important finding is the influence of BMI in the relationship between TH and incident diabetes observed when considering TSH, TH, and the conversion ratio examined as quintiles and as continuous variables. Obesity is a risk factor for diabetes associated with insulin resistance and impaired insulin secretion (29 –33). Thyroid functioning plays a key role in energy metabolism and weight balance. Thyroid dysfunction is associated with excess body fat, although this relationship is not fully understood (34 –36)—hypothyroidism is associated with increased body weight (29,36 –38), and weight loss seems to reduce TSH and fT3 (36,39,40). Due to the relationship of obesity with endocrine abnormalities, a stronger relationship between TH and incident diabetes among overweight/obese people looks reasonable (35,36,41,42), corroborating our findings.

Moreover, the weight gain may also change TH levels trying to increase energy expenditure, preventing more weight gain (36,41,43). Other mechanisms may be involved in the association between fT4 and diabetes beyond obesity. A recent study of Gu et al. (44) reported that low free TH levels within the reference range were significantly associated with high glucose levels and insulin resistance. Particularly, low-normal fT4 levels were associated with metabolic syndrome and insulin resistance (42,45). Moreover, low levels of TH were associated with a decrease in peripheral insulin sensitivity and glucose intolerance. These findings suggest that other mechanisms beyond overweight/obesity may explain the possible association of low fT4 and diabetes (46). We could assume that these mechanisms may overlap with the insulin resistance associated with overweight/obesity, which would help explaining our findings of low fT4 levels among those with weight excess.

The fT3:fT4 ratio reflects deiodinase activity. fT4 must be converted to fT3, which has a higher affinity to TH receptors and is considered the biologically active TH. Moreover, this peripheral process of conversion produces around 80% of T3 in the body. Thus, the fT3:fT4 ratio is very important in maintaining circulating T3 and offering it to tissues (47). Higher deiodinase activity with higher peripheral conversion of fT4 into fT3 was reported as a strategy to mitigate changes caused by weight gain (36,41,43).

Our analysis considering fT4 and the fT3:fT4 ratio as continuous variables presented complementary results compared to the analysis in quintiles. The use of TH as continuous variables increased our power to detect small, but significant linear associations among TH and risk of diabetes. In addition, the analysis presented in quintiles was used to explore whether the associations followed a nonlinear pattern (U-shape or J-shape), although quintile analyses alone cannot formally detect a statistically significant departure from nonlinearity.

Only one cross-sectional study has assessed fT3:fT4 ratio as a risk factor for incident diabetes (17), in 15,296 Chinese participants reporting a lower risk of diabetes at higher conversion rates, contrasting with our findings. Compensatory hormonal feedbacks in different moments related to the degree of obesity as well as differences in the population sample and study design may also contribute to these different findings.

Considering the association of high TSH levels with increased risk of diabetes (12,14,15) and the plausible pathophysiologic basis for the association of overactive (7 –11) and underactive thyroid function (3 –6) with diabetes, we hypothesized that our results would show a U-shaped curve for this hormone; however, we did not confirm it and found no association of TSH and diabetes incidence. A recent meta-analysis of 12 prospective studies found a J-shaped curve with high, but not low TSH levels being associated with an increased risk of diabetes (48), different from our results.

Taking into consideration all results together, an association between TSH levels and incident diabetes is very unlikely. Although TSH is very sensitive to TH variation, it has been recently proposed that serum TSH alone may not be the most reliable marker of thyroid function for other outcomes like cardiovascular disease. TH has shown a stronger association compared to TSH (49,50). This could be more outstanding in obese patients as higher and lower TH levels and increased peripheral conversion were observed in many studies (35,41,43,50), including ours.

Considering the entire study population, our results for TSH levels and their relationship with diabetes risk are different from other previous observational studies that showed positive (12,14 –16) or inverse associations (17). However, differing from them, in this analysis, we included not only euthyroid participants but also participants with subclinical thyroid dysfunction. Three studies reported an association of high-normal TSH levels and the risk of diabetes using the first quintile as the reference in euthyroid participants. A prospective study of 8452 participants found an association between high TSH levels and diabetes incidence (12).

A retrospective cohort in Korea (15) of 6235 individuals reported that the highest tertile of TSH was associated with an increased risk of incident diabetes compared to the first tertile. Another retrospective cohort of 17,061 subjects in Korea (14) reported that each increase of 1 μIU/mL in TSH levels was associated with a hazard ratio of 1.13 [CI 1.07–1.20] for increased risk of diabetes. On the other hand, our findings are in line with two other large prospective studies (13,15) that did not find any association of TSH levels with incident diabetes (13).

Our findings on TH are in line with other studies, but with the advantage of including subclinical thyroid dysfunction as well. The Korean study showed that the highest tertile of TH change was protective against diabetes during 25,692 person-years (15). Data from the Rotterdam study (12) for fT4 were similar to the Korean study and our findings.

Differences regarding study population and design across studies such as ethnicity in the samples of Asian (14 –17) and European (12,13), cross-sectional (16,17) or prospective design (12 –15), age strata (>60 years or not), and sex distributions of the samples (13 –17) may partially explain the discrepancies and the high degree of heterogeneity among results.

Our findings have some strengths. The large sample size, prospective design, and the availability of robust data with many variables that could be considered in the multivariable adjustment corroborated to study the intricate relationship between thyroid function and diabetes. We also included two different approaches considering TSH, TH, and the conversion rate both in quintiles (U-shape association) and as continuous variables obtaining complementary results. Besides, a highly admixed sample, including a large proportion of those characterizing themselves as being non-white, is another strength that differs our study from the previous ones performed with Europeans (12,13) and Asians (14 –17).

In addition, ELSA-Brasil data were collected under strict quality control and standardized protocols providing reliable estimates (51). Although having a slightly higher number of women, the sample includes a large number of men. The diagnosis of incident diabetes was very robust and based on different combinations of previous medical history, use of medication, oral glucose tolerance test, and HbA1c, although a second test to confirm diabetes was not done in each one of the visits. We have information on fT3 and fT3:fT4 ratio, which allowed us to have a more comprehensive evaluation of thyroid function. Although our results showed modest associations, the health impact may be higher than expected since the number of people with diabetes and obesity is rising worldwide, especially in low- and middle-income countries like Brazil (52 –54).

The analysis has also some limitations. TSH and TH were determined only once in each visit. These results did not include participants with overt thyroid diseases because of the high frequency of treatment, which limits the potential for detection of stronger associations. Another limitation is that some participants have the diagnosis of incident diabetes based only on metformin use, which may also be prescribed for prediabetes. It is also important to highlight that the main focus of the analysis is to explore TSH, TH, and fT3:fT4 ratio as risk factors for incident diabetes; hence, no conclusion about the use of fT3 or the fT3:fT4 ratio in the diagnosis or screening of diabetes in euthyroid or individuals with subclinical thyroid diseases can be inferred based solely on these findings.

Conclusion

Our findings showed a modest linear association with higher fT4 levels being protective for diabetes, while a high conversion rate is associated with increased risk of diabetes considering the entire sample. However, our most important result is a modest association of fT4 with incident diabetes and the low conversion rate being protective for diabetes in overweight and obese participants. These findings were also confirmed by an analysis considering TH and the conversion ratio as continuous variables. Thus, TH levels and the conversion rate might represent additional risk factors for incident diabetes, especially among the population with overweight and obese. The findings from this secondary analysis need to be confirmed in future studies.