Correspondence: Juan J. Díez, Department of Endocrinology, Hospital Ramón y Cajal, Carretera de Colmenar km 9, 28034 Madrid, Spain. Tel.: +34913368065; E-mail: email@example.com
A thyrotropin(TSH)-lowering effect of metformin therapy has been recently reported in patients with type 2 diabetes (T2D) and hypothyroidism. We aimed to evaluate the interplay between metformin therapy and serum TSH concentrations in a group of patients with T2D and normal thyroid function.
Patients and methods
Eight hundred and twenty-eight euthyroid patients with T2D (53% women, mean age 65·9 years, median duration of diabetes 10 years) were retrospectively evaluated. There were 250 patients on metformin treatment (30·2%). Serum concentrations of TSH were measured in all subjects.
Patients on metformin treatment exhibited significantly higher TSH levels [1·63 (1·11–2·24) mU/l] than those found in patients without metformin [1·40 (1·01–2·24) mU/l, P = 0·009]. We found no significant differences in TSH levels in patients who were on therapy with other oral antidiabetics, antihypertensive drugs or hypolipidemic agents in relation to subjects not taking these drugs. Serum TSH was significantly related to gender, body mass index, hyperlipidaemia and the presence of goitre and diabetic macroangiopathy. In multiple regression analysis with TSH as dependent variable, goitre was negatively related to TSH values. Metformin therapy was a nonsignificant variable in this model.
In summary, this is the first survey analysing the relationship between metformin and thyroid function in a large cohort of patients with diabetes. Our data do not support the presence of an independent and significant relationship between TSH values and metformin treatment in euthyroid patients with T2D.
Type 2 diabetes (T2D) and thyroid dysfunction are the most common disorders of the endocrine system occurring in the general population. The association between diabetes and hypothyroidism is known from the sixties. In particular, patients with T2D exhibit a striking high prevalence of thyroid dysfunction, ranging from 5% to 13%, as reported by recent retrospective [2, 3] and prospective [4, 5] studies. Metformin, an oral hypoglycaemic biguanide, has been used for the treatment for T2D for many years, and recently, it is considered the first choice for oral treatment for T2D patients in the absence of contraindications.
Metformin is considered a safe drug with few pharmacological interactions. Several retrospective and prospective studies in a limited number of patients have suggested that therapy with this agent is associated with a significant reduction in serum thyrotropin (TSH) concentrations, without relevant changes in serum thyroxine (T4) and triiodothyronine (T3) levels. This finding has been reported in diabetic patients with primary hypothyroidism both under replacement therapy and untreated.[8-10] The prospective study by Cappelli et al., however, did not report any significant effect of metformin treatment in a group of 54 patients with T2D and intact pituitary–thyroid axis. Furthermore, in a recent study, we have reported that newly diagnosed hypothyroidism in patients with T2D was significantly and directly related not only with thyroid autoimmunity, but also with metformin therapy. These findings suggest an independent association of metformin treatment with newly diagnosed hypothyroidism in patients with diabetes.
To our knowledge, there are no surveys analysing the relationship between metformin and thyroid function in large cohorts of diabetic patients with normal thyroid function. Therefore, the aim of the present study has been to evaluate serum TSH levels in euthyroid patients with T2D, and their relationship with metformin and other antidiabetic treatments in a cross-sectional design.
Patients and methods
We designed a cross-sectional study including all euthyroid patients found in a screening programme for thyroid dysfunction performed during 2004–2010 in our diabetic clinic.[11, 12] All patients were ambulatory and in good health, and had been diagnosed with T2D according to American Diabetes Association criteria. None of them had a recent acute illness or an acute complication of a chronic disease. Patients taking drugs known to affect thyroid function were excluded. The study protocol was approved by the local ethical committee, and all patients gave informed consent before blood sampling.
We registered demographic and anthropometric data, goitre (assessed by physical examination), duration of diabetes, hypertension, hyperlipidaemia, presence or absence of diabetic complications (diabetic retinopathy, nephropathy, ischaemic heart disease, cerebrovascular disease, peripheral artery disease), treatment for diabetes (diet, oral agents and insulin) and analytical data (glucose, cholesterol, triglyceride and haemoglobin A1c). Serum concentrations of TSH were measured in all patients without previous thyroid dysfunction. Thyroid autoimmune status was evaluated by means of the measurement of serum levels of thyroid peroxidase autoantibodies (TPOAb) and thyroglobulin autoantibodies (TGAb). Thyroid autoimmunity was considered negative when both autoantibodies were negative and positive when any of them were positive.
From a total population of 1112 patients included in our screening programme, we excluded 179 patients who exhibited previously known thyroid dysfunction. In the remaining group of 933 patients, there were 828 subjects with normal serum TSH concentration and 105 with newly diagnosed thyroid dysfunction (68 of them with elevated TSH and 37 with low TSH levels). For the purpose of this study, we selected and retrospectively evaluated 828 patients with normal serum TSH concentrations.
Additionally, we also studied the relationship between TSH and metformin in the group of 933 patients with T2D and no previously known thyroid dysfunction, that is, including the group of 105 patients found to have elevated or low TSH levels in our screening programme. We performed this additional analysis to evaluate the effects of metformin without the restriction to patients with normal TSH.
Fasting samples of venous blood were obtained from an antecubital vein between 08:00 and 09:00 for the estimation of hormonal and general analytical data. Serum TSH concentrations were determined by using commercially available immunochemiluminescent assay (Immulite; Diagnostic Products Corporation, Los Angeles, CA, USA). The sensitivity of the assay was 0·004 mU/l. The mean intra- and interassay coefficients of variation were less than 10%. Normal values for TSH were 0·4–5·0 mU/l. TPOAb and TGAb were measured using a chemiluminescent immunoassay system (Immulite Thyroid Autoantibody; Siemens Medical Solutions Diagnostic Ltd., Llanberis, Gwynedd, UK). Positivity for TPOAb and TGAb was considered when the titre of these autoantibodies was at least 100 U/ml and at least 340 U/ml, respectively.
For quantitative variables, results are expressed as mean ± SD for normally distributed data and as median (interquartile range) for nonparametric data. Adjustment to normal distribution was tested by the Kolmogorov test. Categorical variables are described as percentages (%). For comparisons of means between two groups of subjects, the Student's t-test was used for normally distributed data and the Mann–Whitney U-test was employed for nonparametric data. For comparison of means among more than two groups, we used repeated measures analysis of variance or the Kruskal–Wallis analysis of variance by ranks, as necessary. Multiple regression analysis and multivariate logistic regression analysis were used to study the relationship between quantitative and qualitative variables. For ratio comparisons, the chi-square test or Fisher's exact test was used. Differences were considered significant when P < 0·05.
Main characteristics of the studied patients
In the study population of euthyroid patients with diabetes, there were 440 women (53·1%) and 388 men (46·9%) aged (mean ± SD) 65·9 ± 12·3 years. 297 patients (35·9%) were obese [body mass index (BMI) ≥ 30 kg/m2]. Median duration of diabetes was 10 (5–16) years and the percentages of patients with diabetic microangiopathy, macroangiopathy, hypertension and hyperlipidaemia were, respectively, 38·2%, 26·4%, 67·8% and 59·7%. Diet was used as the only therapy for diabetes in 127 (15·3%) patients. 246 patients (29·7%) were treated with oral agents and 455 (55·0%) with insulin (with or without oral antidiabetics). There were 250 patients on metformin therapy, 50 of them in monotherapy, 109 in combination with other oral agents and 91 in combination with insulin. Mean fasting blood levels of glucose and haemoglobin A1c were 9·3 ± 3·4 mm and 7·8 ± 1·6%, respectively.
Comparison of data from patients with (n = 250) and without (n = 578) metformin therapy is shown in Table 1. Patients on metformin therapy were younger, had a shorter duration of diabetes, higher values of BMI and serum triglyceride, and had higher percentage of females and hyperlipidaemia. On the contrary, patients taking metformin were less prone to have microangiopathy, macroangiopathy and hypertension. No differences were found between both groups in the percentages of goitre or thyroid autoimmunity, or in the levels of glucose, haemoglobin A1c and cholesterol.
Table 1. Clinical and analytical features of studied diabetic patients with and without metformin therapy
Patients without metformin therapy
Patients on metformin therapy
*P < 0·05
**P < 0·01
***P < 0·001
Data are the number of patients (percentage), mean ± SD for normally distributed data and median (interquartile range) for nonparametric data. The total number of patients used to calculate the values of means, medians and percentages in each group is indicated by n.
Thyrotropin values according to diabetes-related characteristics
Median (IQR) serum TSH concentration in our euthyroid diabetic sample was 1·44 (1·03–2·12) mU/l. These values were slightly higher in female [1·54 (1·05–2·18) mU/l] than in male patients [1·40 (1·01–2·08) mU/l, N.S.]. We found significantly elevated TSH levels in patients with obesity in relation to normal weight subjects, as well as in diabetic patients with hyperlipidaemia in relation to patients without this diagnosis. TSH levels were significantly lower in patients with palpable goitre and with macroangiopathy (Fig. 1). We found no relationship between TSH values and the presence of microangiopathy and hypertension.
Although we found a significant relationship between TSH and obesity and between TSH and goitre (Fig. 1), goitre and obesity were nonrelated variables in our patients. In fact, the prevalence of goitre among patients with obesity (18 out of 289, 6·2%) was very similar to that found in nonobese patients (29 out of 467, 6·2%). BMI was 30·3 ± 5·7 kg/m2 in subjects with goitre and 29·6 ± 5·3 kg/m2 in subjects without goitre.
Serum TSH concentration was not different in patients classified according to their antidiabetic treatment modality, that is, diet [1·48 (1·09–2·10) mU/l], oral antidiabetics [1·49 (1·03–2·14) mU/l] and insulin therapy [1·41 (1·01–2·12) mU/l]. Patients taking metformin exhibited significantly higher TSH levels [1·63 (1·11–2·24) mU/l] than those found in patients without metformin treatment [1·40 (1·01–2·24) mU/l, P = 0·009]. Nevertheless, we found no significant differences in TSH levels in patients who were on therapy with sulphonylureas, meglitinides or thiazolidinediones in relation to subjects not taking these drugs. No relationship was found between TSH and treatment with antihypertensive drugs (calcium antagonist, renin–angiotensin system inhibitors, beta blockers and diuretics) or hypolipidaemic agents (statins and fibrates) (data not shown).
When including in our analysis the group of 105 patients with thyroid dysfunction found in the screening programme (69 with elevated TSH and 37 with low TSH), we again found that patients taking metformin showed TSH concentration [1·71 (1·13–2·49) mU/l, n = 283] higher than that found in patients not taking this drug [1·40 (0·98–2·19) mU/l, n = 650, P < 0·001]. TSH was not related with other drugs in the analysis performed without restriction to the patients with euthyroid.
Metformin and other variables according to TSH values
Patients with TSH levels in the upper tertile (1·90–5·00 mU/l) exhibited a BMI value (30·3 ± 6·0 kg/m2) significantly higher than that found in subjects with TSH levels in the middle or in the lower tertile. Patients in the upper TSH tertile were also characterized by a lower percentage of goitre and macroangiopathy, and also by higher serum triglyceride levels (Table 2).
Table 2. Main clinical and analytical features of patients with diabetes classified according to tertiles of serum TSH concentration
BMI, body mass index; RAS, renin–angiotensin system.
Data are the mean ± SD for normally distributed data, the median (interquartile range) for nonparametric data and the number (percentage) for qualitative variables. Significant values (P < 0·05) are highlighted in bold.
60·5 ± 11·5
65·9 ± 12·7
65·2 ± 12·7
28·8 ± 4·8
29·6 ± 4·9
30·3 ± 6·0
Duration of diabetes
9·4 ± 3·4
9·3 ± 3·3
9·4 ± 3·4
Haemoglobin A1c (%)
7·84 ± 1·50
7·75 ± 1·59
7·85 ± 1·70
5·2 ± 1·0
5·2 ± 1·3
5·0 ± 1·1
The percentage of patients on metformin therapy was also significantly different in patients classified according to tertiles of TSH levels, being higher in the upper tertile (34·5%) in relation to the middle (31·0) or the lower tertile (25·0%). We did not find significant differences in the proportion of other medical therapies in patients classified according to TSH tertiles. In the analysis without restriction to euthyroid patients, the proportions of patients with metformin in the upper (35·0%) and middle (32·2%) tertiles of TSH levels were significantly higher than that found in the lower tertile (23·8%, P = 0·007).
Values in patients negative for thyroid autoantibodies
Thyroid autoantibodies levels were available in 564 patients. Results were positive in 20 and negative in 544 patients. In the group of 544 patients negative for thyroid autoantibodies, TSH values were significantly higher in those taking metformin [1·64 (1·11–2·19) mU/l] than in those not taking this drug [1·37 (1·01–2·19) mU/l, P = 0·029].
A significant correlation was found between TSH and BMI (ρ = 0·087, P = 0·014) and between TSH and serum triglyceride levels (ρ = 0·133, P < 0·001) in univariate regression analysis. These significant correlations persisted when analysing only patients with negative autoimmunity (n = 544; ρ = 0·116, P = 0·008 for BMI and ρ = 0·161, P < 0·001, for triglyceride levels). We performed a multiple regression analysis with TSH as dependent variable including all the variables found to be related with TSH levels in Fig. 1, as possible confounders. This analysis showed that goitre was negatively related to TSH values (Table 3). Metformin, BMI, macroangiopathy and hyperlipidaemia were nonsignificant variables in this model. When studying the subgroup of patients with negativity for thyroid autoantibodies (n = 544), we found that goitre persisted as the only significant variable in the model. This multiple regression analysis including the group of 105 patients with elevated or low TSH showed that metformin was a nonsignificant variable in the model and goitre was near the statistical significance (P = 0·059).
Table 3. Multiple regression analysis for TSH as dependent variable in the whole group of euthyroid patients with diabetes and in the subgroup of patients with negative results for thyroid autoantibodies
Patients with negative thyroid autoantibodies
Significant values (P < 0·05) are highlighted in bold.
In a logistic regression analysis model with metformin therapy as dependent variable, we found that TSH levels and metformin therapy were significantly related (OR for the highest vs the lowest tertile of TSH levels, 1·583, 95% CI, 1·095–2·289, P = 0·015). However, when including BMI, macroangiopathy and hyperlipidaemia as independent variables in the model, serum TSH concentration was a nonsignificant variable. Similar results were found when analysing our patients without restriction to euthyroid subjects.
Results of the present study show that serum TSH concentration in a sample of euthyroid patients with T2D is related not only with metformin therapy, but also with some other clinical variables (Fig. 1), including goitre. Several studies have reported that TSH levels are increased in obese subjects,[14, 15] and our metformin group had significantly more obese subjects than the group of patients not taking this drug. Contrary to the findings of Cappelli et al.  who recently found a significant relationship between thyroid nodules and morbid obesity, we could not find any significant relationship between the presence of goitre and obesity, although our results are limited by the absence of ultrasonographic information. Interestingly, there were no relationship between TSH values and any of the analysed pharmacological therapies, that is, insulin, oral antidiabetics, antihypertensive and lipid-lowering drugs.
Obesity, goitre and other clinical variables might interact with the interplay between TSH and metformin therapy as confounding factors. In fact, our multiple regression analysis showed that the relationship between metformin and TSH values lost its significance, being goitre the only significant variable in the model. In our multivariate logistic regression analysis, the significant relationship between TSH and metformin was lost when introducing in the model some confounding variables, such as BMI, macroangiopathy and hyperlipidaemia. Similar results were found when analysing patients without restriction to euthyroid subjects, that is, including patients found to have elevated or low TSH concentrations.
Although our patients with and without metformin therapy were not matched for confounding variables, these findings suggest that TSH and metformin are not independently related. Our results also suggest that thyroid autoimmunity do not modify this lack of significant relationship between metformin and TSH.
We could not find any significant relationship between metformin therapy and thyroid autoantibodies. This lack of relationship might be accounted for by the rather low prevalence of thyroid autoimmunity in our population (3·5%). In agreement with this finding, our previous study  showed an overall prevalence of thyroid autoimmunity in diabetic patients without thyroid dysfunction slightly higher (5·6%), and no differences between patients with and without metformin. Another possible influencing factor is the fact that we could not measure thyroid autoantibodies in all our patients.
Vigersky et al.  fortuitously observed a reversible metformin-induced suppression of TSH, without changes in free T4 or free T3, in a patient with hypothyroidism on levothyroxine treatment. This finding was also substantiated in another three patients with chronic hypothyroidism that started metformin therapy for diabetes. In a prospective study in 8 obese, diabetic women with primary hypothyroidism on levothyroxine replacement therapy, the administration of metformin, 1700 mg daily for 3 months, was accompanied by a significant reduction in TSH levels from a mean of 3·11 mU/l to 1·18 mU/l, which reverted after withdrawal of the drug. No significant changes in free T4 or free T3 levels were observed in this study. Interestingly, in this study, patients showed a significant decrease in body weight that might account for, at least in part, the decline in TSH levels. Another prospective long-term study in patients with T2D showed that metformin therapy was accompanied by a significant reduction in TSH levels in 29 patients with treated hypothyroidism and in 18 patients with untreated hypothyroidism. These authors did not find significant changes in body mass index or in free T4 levels in their patients. Furthermore, in this prospective study, the authors did not find any effect of metformin therapy in a group of 54 euthyroid patients with diabetes treated during 1 year.
The association of metformin therapy with the presence of newly diagnosed thyroid dysfunction has been specifically assessed in previous studies. We reported that the presence of not previously known hypothyroidism was significantly related not only with thyroid autoimmunity, but also with the presence of metformin treatment (OR 2·51, 95% confidence interval 1·28–4·92). On the contrary, the presence of thyroid hyperfunction in patients with T2D was significantly related with age and goitre, but not with metformin therapy.
It is interesting that the TSH-reducing effect of metformin has been reported only in diabetic subjects with hypothyroidism, but not in patients with euthyroidism. In fact, our findings in euthyroid patients with diabetes are in line with those of Oleandri et al.  In a prospective, randomized study, these authors assessed the effects of 3-month treatment with metformin on anthropometric, metabolic and hormonal variables, including TSH and thyroid hormones, in a group of 28 patients with abdominal obesity. Their results showed that metformin did not modify the effect of diet on serum TSH concentrations in patients with obesity. The effects of metformin treatment on the thyroid hormone profile have been evaluated in a recent prospective study including 24 euthyroid women with polycystic ovary syndrome. Serum levels of TSH before starting metformin therapy did not significantly change after a 4-month course of metformin. Similarly, there were no significant changes in serum-free T4 in this group of women with euthyroidism. Furthermore, no TSH-lowering effect could be demonstrated in the group of patients with T2D and integrity of the pituitary–thyroid axis studied by Cappelli et al., as mentioned above.
Mechanism of action of metformin is complex and multifactorial.[20, 21] This drug may change the affinity and/or number of thyroid hormone receptors, may increase the central dopaminergic tone or may act directly on TSH regulation, thus enhancing the effect of thyroid hormones in the pituitary.[8, 22] Metformin may also enhance gastrointestinal absorption of levothyroxine or may produce subtle changes in thyroid hormone protein-binding or affect thyroid hormone metabolism. All these mechanisms would act in diabetic hypothyroid subjects; however, our results and those of other authors in euthyroid subjects with diabetes  or with polycystic ovary syndrome  do not support that similar mechanisms are operative in subjects with integrity of the pituitary–thyroidal axis. It is also possible that the TSH-lowering effect of metformin is developed slowly as suggested the data from Cappelli et al.  who found that metformin did not exert any acute change in TSH in 11 patients with treated hypothyroidism.
We are unaware of studies analysing the relationship between serum TSH and metformin therapy in large series of diabetic patients with euthyroidism. The relative small number of patients in our sample is the main limitation of this study. The cross-sectional design of the study prevents drawing firm conclusions as to the effect played by metformin on the serum levels of TSH. Our studied sample is skewed between the metformin and nonmetformin group, particularly with respect to obesity, a factor with known relationship with TSH. We could not measure serum T4 levels in our euthyroid subjects and thyroid autoantibodies were not available in the whole sample. Neither were we able to assess the effects of the duration of metformin treatment on TSH values in our patients. The lack of information of switching from a therapeutic regimen to another could represent another factor of confusion.
In summary, we found significant relationships between TSH levels and metformin therapy, obesity, macroangiopathy and hyperlipidaemia. However, in multivariate analysis, only the presence of goitre remained as a significant variable. These results agree with the findings of some studies showing that metformin effect on TSH is not observed in patients with normal thyroid function. Further prospective studies in large samples of euthyroid patients with diabetes are needed to clarify the effect of different doses and duration of treatment with metformin on thyroid economy.
Declaration of Interest
The authors have no conflict of interest in relation to this article.