Introduction

Thyroid disease and biochemical thyroid dysfunction are found frequently in diabetes. In the case of hypothyroidism, 6–20% of people with type 2 diabetes have overt hypothyroidism (an increased serum thyrotropin (TSH) and low serum free thyroxine (FT4) concentration) and a further 10% have subclinical hypothyroidism (an increased serum TSH and normal serum FT4) [1]. Thyroid overactivity is much less common, with approximately 4% of people with type 2 diabetes found to have overt thyrotoxicosis (a decreased TSH and raised serum FT4) and the same percentage subclinical thyrotoxicosis (a decreased serum TSH and normal serum FT4) [1].

Overt hypothyroidism affects multiple organ systems. In particular it may cause or contribute to dyslipidaemia [2], hypertension [3], renal dysfunction [4], and cardiac/endothelial dysfunction [5], all of which increase cardiovascular disease (CVD) risk. Although representing relatively mild thyroid dysfunction, subclinical hypothyroidism is also associated with adverse CVD risk factors [4,5,6]. While general population studies of the incidence of CVD in subclinical hypothyroidism have had variable results, an individual participant data meta-analysis found significantly increased risk of CVD and death in those with raised TSH concentrations [7]. In the case of thyrotoxicosis, there is also evidence of an association between overt thyrotoxicosis and CVD events and cardiovascular death [8] and the balance of evidence suggests that this is also the case for subclinical thyrotoxicosis [9].

The CVD risk factors in people with thyroid dysfunction are also frequently found in those with type 2 diabetes, suggesting that mild thyroid disease may be an additional, and potentially treatable, risk factor in this setting. In addition, minor degrees of thyroid dysfunction can act in concert with the insulin resistance that is characteristic of type 2 diabetes to worsen dyslipidaemia [10]. However, the existing studies of the impact of subclinical hypothyroidism on CVD risk in people with type 2 diabetes have been carried out in clinic-based or selected samples of patients [11,12,13] and only rarely in community-based cohorts [14], while there have been no population-based studies of the prognostic implications of overt or subclinical thyrotoxicosis complicating type 2 diabetes.

Given this background, the aim of the present study was to examine whether overt and subclinical thyroid dysfunction detected in a community-based sample of well characterised adults with type 2 diabetes were associated with CVD risk factors, key CVD outcomes and mortality. We hypothesised that thyroid dysfunction would influence CVD risk and directly and/or indirectly increase macrovascular complications and death.

Methods

Participants

The present study included a subset of 1,250 participants with type 2 diabetes and no known history of thyroid disease from Phase II of the Fremantle Diabetes Study (FDS2), a prospective representative community-based study of people living with diabetes in the primary catchment area of Fremantle Hospital. The FDS2 cohort comprises 1,732 participants with diabetes, including 1,483 (85.6%) with type 2 diabetes based on age at diagnosis, nature of first presentation, treatment history, body mass index and evidence of ketonaemia, and, where available, autoantibody positivity, serum insulin/C-peptide concentrations, and genotyping for those at high risk for monogenic forms [15]. Detailed information regarding recruitment and classification of diabetes has been published [15].

Participants with known prevalent thyroid disease (n = 158), based on a history of hypothyroidism, Hashimoto’s thyroiditis, Graves’ disease, nodular goitre, taking thyroxine, liothyronine and/or anti-carbimazole or propylthiouracil, or who had a history of thyroidectomy or 131I therapy by self-report or ascertained from linkage with the Hospital Morbidity Data Collection (HMDC) through the Western Australia Data Linkage System (WADLS) [16], were excluded from analyses. An additional 75 participants were excluded, comprising those taking amiodarone or lithium (n = 5) with the risk of iatrogenic thyroid disease, those with missing medication data (n = 9), or without suitable blood samples (n = 9), and 52 who also participated in the Fremantle Diabetes Study Phase I (FDS1) but who were living outside the catchment area. The FDS2 protocol was approved by the Southern Metropolitan Area Health Service Human Research Ethics Committee. All subjects gave informed consent before participation.

Clinical and biochemical data.

All participants underwent a detailed assessment at study entry which provided a comprehensive range of clinical, biochemical and questionnaire data. Each participant was invited to return for biennial face to face assessments interspersed with biennial postal questionnaires. Serum TSH, FT4 and antibodies to thyroperoxidase (anti-TPO) were measured in baseline samples at the nationally accredited PathWest laboratories at Fremantle Hospital, Western Australia, by automated electrochemiluminescence immunoassays using reagents from Roche Diagnostics Australia (Castle Hill, NSW, Australia) and a Cobas E601 analyser. Between-run imprecision, expressed as coefficients of variation, were ≤ 4.7% for FT4, ≤ 3.5% for TSH and ≤ 8.1% for anti-TPO antibodies. Reference ranges were 0.34–5.1 mIU/L for serum TSH, 12–22 pmol/L for serum FT4 (derived from a study of blood donors in the Fremantle area) and < 35 U/L for anti-TPO (as recommended by the manufacturer). Standard care testing of serum glucose, HbA1c, serum urea and electrolytes, serum lipids, serum uric acid, liver function, urine albumin and creatinine, and full blood count were performed using standard automated methods at PathWest Fremantle Hospital. Chronic complications were ascertained using standard criteria [17].

The FDS2 cohort was followed from baseline (2008–2011) until end-December 2016 for mortality outcomes and to end-June 2016 for incident myocardial infarction (MI) and stroke using linkage through the WADLS. The WADLS includes all public and private hospitalisations and death registrations within WA. Relevant International Classification of Disease (ICD)-9-CM and ICD-10-AM diagnosis codes were used to identify prevalent and incident MI and stroke in the HMDC (MI: ICD-9-CM codes 410 and ICD-10-AM codes I21, I22; stroke: ICD-9-CM codes 430–432, 433.01, 433.11, 433.21, 433.31, 433.81, 433.91, 434.01, 434.11, 434.91 and, ICD-10-AM codes I60-I64). Causes of death specified on the death certificate or in a coroner’s report were reviewed independently by two study physicians (DGB and TMED) and classified as cardiovascular (cardiac, cerebrovascular or sudden death) or otherwise.

Undiagnosed prevalent hypo- and hyperthyroidism at baseline were defined by an abnormal TSH. An elevated TSH (> 5.1 mIU/L) defined overt hypothyroidism in participants with a subnormal FT4 (< 12 pmol/L), and subclinical hypothyroid in those with normal FT4. Those with a low TSH (< 0.34 mIU/L) and elevated FT4 (> 22 pmol/L) were considered overtly hyperthyroid, while a low TSH and normal FT4 was categorised as subclinical hyperthyroidism. Since thyroid function tests are not part of usual-care biochemical screening in type 2 diabetes, they were performed retrospectively in the present study and were thus not used as the basis for clinical management decisions.

Statistical analysis

The statistical analyses were performed using SPSS for Windows (version 25; SPSS Inc., Chicago, IL, USA). Data are presented as proportions, mean ± standard deviation (SD), geometric mean (SD range) or median [interquartile range (IQR)] as appropriate. For independent samples, two-way comparisons for proportions were by Fisher’s exact test, for normally distributed variables by Student’s t-test and for non-normally distributed variables by Mann–Whitney U-test. Serum TSH and FT4 were analysed using categories based on the assay reference ranges (TSH < 0.34, 0.34–2.9, 3.0–5.1 and > 5.1 mIU/L; FT4 < 12, 12–22 and > 22 pmol/L). A conservative Bonferroni approach was used to adjust for multiple comparisons. All clinically plausible variables with P ≤ 0.20 in bivariable analyses were considered for entry into multivariable models. Multiple logistic regression (backward conditional, variable entry and exit set at P < 0.05 and P ≥ 0.05, respectively) was used to identify independent associates of prevalent hypothyroid disease. Determinants of all-cause and cardiovascular mortality, incident MI, and incident stroke were determined using Cox proportional hazards regression modelling (backward conditional, variable entry and exit set at P < 0.05 or P < 0.01 and P ≥ 0.05 or P ≥ 0.01, respectively, depending on the number of outcomes). In the survival analyses, the lowest two TSH categories were combined due to small cell numbers.

Results

Associates of baseline thyroid dysfunction

At baseline, the 1,250 participants with type 2 diabetes had a mean ± SD age of 65.3 ± 11.5 (range 16.4–97.1) years, 56.5% were males, and their median [IQR] diabetes duration was 8.0 [2.5–15.6] years. The baseline characteristics of participants by biochemically defined thyroid disease are shown in Table 1. Sixty-one participants had subclinical hypothyroidism had four had a TSH > 10.0 mU/L, the highest being 12.0 mU/L. In bivariable analyses, compared to those with normal serum TSH, those with subclinical hypothyroidism were older at baseline, more likely to be anti-TPO positive, had lower eGFR, and higher systolic blood pressure. Independent associates of subclinical hypothyroidism included anti-TPO positivity, systolic blood pressure, and eGFR (negatively) (Table 2). In bivariable analyses, the 15 participants with overt hypothyroidism had significantly higher prevalence of anaemia and anti-TPO positivity than the euthyroid participants. There were too few participants with overt hypothyroidism for multivariable analysis. There were no significant differences between the hyperthyroid and euthyroid participants, but the numbers of hyperthyroid participants were too low for any meaningful comparisons.

Table 1 Baseline characteristics of 1,250 type 2 diabetes participants by biochemically defined thyroid disease
Full size table
Table 2 Multiple logistic regression model of independent associates of undiagnosed prevalent subclinical hypothyroidism in 1,232* participants with type 2 diabetes
Full size table

Thyroid dysfunction and incident stroke and myocardial infarction

Among the 1,219 participants with no known thyroid disease and no stroke prior to enrolment, 42 had an incident stroke during 7,603 person-years (mean ± SD 6.2 ± 1.6 years) of follow-up. Among 1,149 participants with no known thyroid disease and no MI prior to enrolment, 103 had an incident MI during 7,129 person-years (mean ± SD 6.2 ± 1.7 years) of follow-up. There were no bivariable associations between incident stroke or MI and either TSH or FT4 categories (P ≥ 0.14; Supplementary Table 2). The most parsimonious Cox models of incident stroke and incident MI with TSH categories added are shown in Table 3. TSH categories did not add significantly to the models for either outcome (P ≥ 0.22). In particular, the incidence of stroke for those with a TSH < 3.0 mU/L was not statistically different to that for participants with a TSH 3.0–5.1 mU/L. No incident strokes or MIs occurred in those with a high FT4, and only three incident strokes and four incident MIs were recorded in those with low FT4, so FT4 categories were not added to the respective most parsimonious Cox models.

Table 3 TSH category added to most parsimonious Cox proportional hazards models of independent predictors of incident stroke and myocardial infarction, using time as the timeline, in participants with type 2 diabetes and no known history of thyroid disease, and of stroke or myocardial infarction, respectively
Full size table

All-cause and cardiovascular mortality

Of 1,250 participants with type 2 diabetes and no known thyroid disease at baseline, 215 (17.2%) died, including 37 (17.2%) from cardiovascular causes during 9,520 person-years (mean ± SD: 6.7 ± 1.7 years) of follow-up. Baseline characteristics by vital status are summarised in Supplementary Table 1. There was no significant relationship between baseline serum TSH categories and subsequent all-cause and cardiovascular mortality (P = 0.12 and P = 0.18, respectively). Kaplan–Meier survival analysis (data not shown) showed borderline statistical significance for the effect of serum TSH categories on all-cause and cardiovascular mortality (log rank test, P = 0.057 and 0.044, respectively), but TSH categories did not add to the most parsimonious Cox model in each case (Table 4).

Table 4 Most parsimonious Cox proportional hazards models of independent predictors of all-cause and cardiovascular mortality, using time as the timeline, in 1,250 participants* with type 2 diabetes and no known history of thyroid disease, with TSH categories and FT4 categories added separately
Full size table

FT4 categories were bivariably associated with all-cause mortality (P = 0.024; Supplementary Table 1). After adjustment for the most parsimonious Cox model of all-cause mortality (Table 4), FT4 categories added significantly (trend P = 0.001). Compared with FT4 of 12–22 pmol/L (reference category; 197 deaths in 1188 participants (16.6%)), a low FT4 was associated with a 76% increased risk of death (15 deaths in 55 participants (27.3%)) and high FT4 with a greater than sixfold increased risk. However, there were only seven participants with high FT4, three (42.9%) of whom died, and so the data need to be interpreted with caution. As there were no cardiovascular deaths in those with a high FT4 and only one in those with low FT4, FT4 categories were not added to the most parsimonious Cox model for cardiovascular mortality.

Discussion

In this study of well characterised, community-based people with type 2 diabetes and no known thyroid disease, we found that previously undiagnosed subclinical hypothyroidism was significantly and independently associated with a higher systolic blood pressure and impaired renal function. Despite these relationships, there were no independent statistically significant associations between serum TSH concentrations and incident stroke, MI, or cardiovascular or all-cause mortality over a more than six-year average follow-up period. Undetected thyrotoxicosis, whether overt or subclinical, was rare in our cohort. These findings have potential implications for type 2 diabetes management in that they do not support attempts to identify and/or treat mild thyroid dysfunction outside usual care.

The association of diastolic hypertension with overt hypothyroidism in the general population is well recognised [3]. Reports of a similar association in subclinical hypothyroidism have been inconsistent, with evidence of a modest association driven by studies in which the subclinically hypothyroid participants were significantly older than those who were euthyroid [18]. In the present study, participants with subclinical hypothyroidism were also older than those who were euthyroid, but the association of systolic blood pressure with subclinical hypothyroidism was robust after adjustment for age, suggesting that age is not the driver of this association in people with type 2 diabetes. Impaired renal function was also independently associated with subclinical hypothyroidism in the present study, as has been seen in other studies in diabetes [12, 19, 20]. In people with chronic kidney disease, there is evidence that those with persistent subclinical hypothyroidism have a greater rate of eGFR decline and a greater incidence of end-stage renal disease compared to those with transient subclinical hypothyroidism or who remain euthyroid [21]. If chronic subclinical hypothyroidism has a similar adverse impact on renal function in type 2 diabetes, which is itself a strong risk factor for chronic kidney disease, its detection and possible treatment may be more important than in the general population.

We found no evidence that an increased TSH was independently associated with subsequently increased cardiovascular or all-cause mortality. Both low and high FT4 were significantly associated with all-cause mortality in multivariable analyses but the numbers in these groups were very low (only seven participants had high FT4) which questions the clinical significance of this observation. Subclinical hypothyroidism was associated with neither all-cause nor cardiovascular mortality after adjustment for other variables in female participants with type 2 diabetes enrolled in the earlier FDS Phase I (FDS1) [14], in a clinic-based sample of people with type 2 diabetes [19], and in a small-scale study of patients with type 2 diabetes undergoing renal dialysis [11]. Taken together, these findings questions whether subclinical hypothyroidism has prognostic implications in type 2 diabetes.

There have been a range of studies of the association between subclinical hypothyroidism and mortality in the general population, the results of which have been inconsistent. One individual participant data meta-analysis found that subclinical hypothyroidism, although not associated with all-cause mortality, was associated with incident coronary heart disease (CHD) and with CHD mortality at TSH concentrations > 10 mU/L, especially if those starting thyroid replacement were omitted from the analysis [7]. More recent study-level meta-analyses found significant associations between overt or subclinical hypothyroidism and both all-cause mortality and incident CVD [22, 23]. These reports revealed a modest increased mortality risk associated with persistent subclinical hypothyroidism in the general population. Our observation that this is not the case in type 2 diabetes may reflect that people with diabetes managed in a well-resourced health system such as in Australia may be both screened and treated relatively intensively for comorbidities such as thyroid disease and more rigorously managed for CVD risk factors than people without diabetes.

We similarly found that an increased TSH was not a determinant of incident stroke or MI. A systematic review of general population studies found no association between stroke and either subclinical hypothyroidism or subclinical hyperthyroidism [24]. A subsequent individual participant data meta-analysis used the same studies as in the systematic review but included unpublished data from other studies which substantially increased the number of participants and thus study power. There was still no association overall, but younger participants with subclinical hypothyroidism were at significantly increased risk of stroke [25]. More recent, large population-based studies from Scandinavia have also found no independent association between subclinical hypothyroidism and stroke [26, 27]. Our observations are therefore in accord with larger studies involving predominantly participants without diabetes. Interestingly, reports of reduced stroke risk at higher TSH concentrations within the reference interval have been published [13], but we found no indication of a differential incidence of stroke in those with TSH < 3.0 mU/L or 3.0–5.1 mU/L.

Our observation that an increased TSH was not a determinant of incident MI is consistent with our previous observation from FDS1 of CHD in women with type 2 diabetes [28], as well as other studies of people with type 2 diabetes [11, 19] and the results from participants with diabetes and a TSH within the conventional reference interval in the Second Manifestations of ARTerial disease study [13]. General population studies have produced inconsistent results, but there may be an association between subclinical hypothyroidism and CHD at TSH levels > 10 mU/L [7, 27, 29, 30]. We interpret our results and those of other authors as indicating that the excess risk of CHD due to subclinical hypothyroidism in type 2 diabetes is minimal, if present at all.

The present study had limitations. The low numbers with subclinical hypothyroidism in our cohort may have limited our ability to draw conclusions regarding its relationship with endpoints, especially for the less common outcomes. We also had limited power to examine independent associates of overt hypothyroidism and the associated risk of death. Thyroid function was performed only at baseline and so we were unable to assess the mortality risk for those with transient versus persistent subclinical hypothyroidism. We did not measure serum free T3, a variable which may have clarified the status of the few participants with a high FT4. We were unable to adjust our analyses for incident thyroid replacement therapy or other interventions with implications for thyroid function as this information was not available to us for all participants. Such additional data may have strengthened associations with mortality, as shown in one meta-analysis [7]. There were too few participants with positive TPO antibodies among those with raised TSH to allow a meaningful analysis of the effect of this variable on mortality and cardiovascular outcomes, and too few Aboriginal people to assess the possible consequences of concomitant thyroid disease and type 2 diabetes in this vulnerable sub-group. The uptake of the glucagon-like peptide 1 receptor agonists and sodium-glucose co-transporter-2 inhibitors, classes of blood glucose-lowering therapies with glucose-independent benefits for cardiovascular disease, was relatively low in the FDS2 cohort (< 5% in each case at the end of the follow-up period) given that they were first introduced into Australia when the study was in progress. The strengths of the study are its representative, community-based sample and the comprehensive prospective data collection.

Although thyroid dysfunction affects a third or more of people with type 2 diabetes [1], routine biochemical screening for thyroid disease is not mentioned in many major national and international guidelines, and a recent thyroid-specific position paper recommends against screening in patients with type 2 diabetes [31]. This may be appropriate given that most of our patients in whom thyroid disease was newly detected (77.2%) had subclinical hypothyroidism, a condition without a clear relationship to CVD events and all-cause mortality in type 2 diabetes and for which there are no trial data indicating benefit of replacement therapy in preventing these outcomes [1]. Nevertheless, even in a relatively resource rich healthcare environment such as Australia, the present data suggest that in people with type 2 diabetes and coincident hypertension and/or renal impairment, a possible contribution of thyroid disease should be considered as part of clinical assessment.