Low Serum Thyrotropin Level and Duration of Suppression as a Predictor of Major Osteoporotic Fractures—The OPENTHYRO Register Cohort

Authors


  • Preliminary findings were presented as an abstract at the annual meeting of the American Society for Bone and Mineral Research, Baltimore, MD, October 2013.

ABSTRACT

The relationship between thyrotoxicosis and osteoporotic fractures remains controversial, particularly in men. Register-based cohort study including all patients with a serum thyrotropin (TSH) measurement in the region of Funen 1996–2010. All TSH determinations were done in the same lab, which served all hospitals and General Practice (GP) practices in the region. Persons with raised TSH or a history of thyroid/pituitary disease or use of thyroid medications were excluded. The study population consisted of 222,138 (96%) persons with normal and 9217 (4%) with low TSH (<0.3 mIU/L). A single low TSH at baseline was associated with increased risk of hip fractures (adj HR 1.16, 95% CI 1.07–1.26, p < 0.001) but not major osteoporotic fractures (MOF, adj HR 1.06, 95% CI 0.99–1.12, p = 0.058) over a median follow-up of 7.5 years. When men were analyzed separately, results did not reach statistical significance. We found a significant association between duration of thyrotoxicosis and fracture. For each 6 months in which the mean TSH value was decreased (<0.3 mIU/L), hip fracture risk increased by a factor 1.07 (adj HR, 95% CI 1.04–1.10, p < 0.001) and MOF by 1.05 (adj HR, 95% CI 1.03–1.07, p < 0.001). Overt thyrotoxicosis was associated with an increased risk of hip fractures but not MOF. In euthyroid patients, the risk of fractures increased significantly with each SD unit of TSH decrease: Hip fracture (HR 1.45, 95% CI 1.22–1.71, p < 0.001) and MOF (HR 1.32, 95% CI 1.19–1.46, p < 0.001). In a population-based cohort, a single, first measurement of decreased TSH in patients without known thyroid disease was associated with an increased long-term risk of hip fracture, which remained significant in women but not in men after adjusting for confounders. Moreover, the risk of both hip fracture and MOF increased exponentially by the length of time during which TSH had remained low. © 2014 American Society for Bone and Mineral Research

Introduction

Thyroid dysfunction is associated with a considerable negative impact on quality of life even following treatment.[1] It is preceded and followed by an increased somatic as well as psychiatric morbidity[2, 3] and eventually excess mortality.[4, 5] While the effects on bone turnover and bone mineral density are well established,[6-10] uncertainty still surrounds the magnitude, dose dependency and reversibility[8] of the effect on the long-term risk of osteoporotic fractures in the population as a whole and in particular fracture risk in men and in younger women.[11, 12]

In the 1920s—triggered by case reports published toward the end of the 19th century—Aub et al. demonstrated large increases in calcium and phosphorus excretion and noted thinning of the hand bones on x-ray images in patients with hyperthyroidism.[13] Though thyroid hormones impact on bone metabolism,[11, 14, 15] fractures are relatively rare events, especially in persons without osteoporosis, implying that observational cohorts need to be fairly large to robustly examine the real-world impact of varying degrees and durations of thyroid hormone excess on fracture risk. As discussed below, clinical cohorts drawn from the general population are very useful, but resolving power is limited by the great majority of the population sampled having thyroid status within the normal range. Though providing less detail, especially on anthropometric and lifestyle characteristics that may influence the risk of fractures, register-based cohorts offer the research community the opportunity to identify a fairly large number of individuals with abnormal thyroid function tests and to follow both fracture outcomes and medical intervention.

We examined the association between thyroid status and major osteoporotic fractures in a region of Denmark where all thyrotropin (TSH) measurements were performed in the same clinical biochemistry laboratory, irrespective of whether samples originated from hospitals or from a primary practice. The main objective of the study was to assess whether the level of TSH, as evaluated by a single first measurement, had long term consequences on the risk of major osteoporotic fractures, either (1) classified by standard deviation scores of TSH within the euthyroid range, or (2) TSH below the normal range. In addition, we aimed to examine the dose response relationship among patients with thyrotoxicosis. This was done (1) by classifying thyrotoxicosis as overt or subclinical disease based on thyroxine (T4) and triiodothyronine (T3) levels, and (2) by calculating the cumulative duration of low TSH.

Methods and Study Population

We conducted an open, register-based cohort study which included all adult patients (age 18 +) with a new TSH measurement in the Danish region of Funen (pop 476,580, January 2005) in the period January 1996 to end of December 2010. The OPENTHYRO register cohort was formed by merging information from our clinical biochemistry database with data from the National Hospital Discharge Register (NHDR), filled prescriptions from the National Prescription Database[16] and the National Central Civil Register, the latter holding information about deaths and migration.

Cases

In order to include only incident cases, 1995 was used as a run-in year for clinical biochemistry so that patients with a TSH measurement prior to January 1996 could be excluded. The index date was the date of the first TSH measurement in the study period. Hospital contacts were extracted from the NHDR. We excluded all patients who had previous hospital contacts (inpatient or outpatient visits) for thyroid disorders or who had filled prescriptions for thyroid medications (thyroxine, levothyronine, or thionamides) prior to the index date. Finally, patients with pituitary disorders were excluded from the study because low TSH is not a reliable indicator of thyrotoxicosis in this group of patients. All TSH determinations were done in the same lab, which served all hospitals and GP practices in the region. Fracture outcomes in the form of major osteoporotic fractures[17] were collected by ICD10 diagnosis from the NHDR with data available up to 30 November 2012 (the end of the study). Of 275,467 persons with TSH measurements, 5644 were excluded as they were prevalent cases who were also in the 1995 dataset; 7688 had received thyroid medication; 5 had a record of radio-iodine therapy; 1126 had prior thyroid surgery; 753 had known pituitary disease; 20,260 were under 18 years of age; and, finally, 223 were lost to follow-up due to migration. Of the remaining 239,768 persons, 222,138 had normal TSH (0.3 to 4 mIU/L), while 9217 persons had low TSH (<0.3 mIU/L). Patients with raised TSH (4.0%) were not included.

Assays

All measurements were done in the same laboratory, which served all hospitals and GP practices in the region. Until 2006, TSH concentrations were measured using a time-resolved fluoroimmunoassay based on a direct sandwich technique with three mouse anti-human monoclonal antibodies. For detection, Europium served as a fluorescence enhancer. Analyses were performed with AutoDELFIA equipment (Wallac, Turku, Finland). The analytical limit of detection (LOD) was 0.005 mIU/L, while within-run imprecision was 2.7 CV% (coefficient of variation) at 0.89 mIU/L and 1.3 CV% at 17.6 mIU/L. From 2006, measurements were performed with a solid-phase, two-site chemiluminescent immunometric assay on an Immulite 2000 (Siemens, Erlangen, Germany). The analytical LOD was 0.004 mIU/L. Within-run imprecision was 5.3 CV% at 0.32 mIU/L and 3.9 CV% at 3.3 mIU/L. To assure compatibility between the two analyses, method comparison was performed in 120 patient samples showing comparable means (range 0.008–49 mIU/L) and a regression coefficient of 0.991. Also, an external quality control program (Ringversuche, RfB, Bonn, Germany) assured comparability for the entire period.

Concentrations of total triiodothyronine (TT3) and total thyroxine (TT4) were analyzed using time-resolved fluoroimmunoassays based on competitive binding to T3- or T4-specific antibodies, respectively. Both assays used specific mouse anti-human monoclonal antibodies and Europium served as a fluorescence enhancer. Analyses were performed on an AutoDELFIA equipment (Wallac, Turku, Finland). LOD was 0.3 nmol/L (TT3) and 5 nmol/L (TT4). Within-run imprecision were 3.2 CV% at 1.37 nmol/L (TT3) and 2.8 CV% at 79 nmol/L (TT4).

Outcomes

Incident fractures were identified by ICD-10 codes for inpatient and outpatient contacts with Danish hospitals in the study period. In accordance with current clinical practice, we considered fractures of the hip, humerus, forearm, and spine to be fractures most likely to be associated with osteoporosis (“major osteoporotic fractures”).[18] A distinction was not made between traumatic or nontraumatic fractures as the risk of fractures caused by both high and low impact trauma is increased in osteoporosis.[19] The register has been validated for hip fracture outcomes[20] and has been extensively used in fracture epidemiology and pharmacoepidemiology.[21-23]

Statistical Analysis

Cox proportional hazards models were used with major osteoporotic fracture or hip fracture as the outcome and with censoring at the date of death, date of emigration, or end of study. Using this approach, multiple fractures in the same individual are only counted once. The validity of the proportional hazards assumption was evaluated by inspection of Schoenfeld residuals versus time. Diagnoses prior to the index date were considered baseline comorbid conditions. Prescriptions filled in the past year before the index date were used to identify use of osteoporosis drugs and systemic glucocorticoids so this information could be incorporated into the statistical models. We used SPSS® and SAS® through Statistics Denmark. Differences in baseline demographics were analyzed using parametric statistics or categorical statistical methods depending on the nature of the data. We controlled for individual comorbidities and for Charlson index.[24]

For dose-response analyses by severity of thyroid dysfunction, we divided the low TSH group into “overt thyrotoxicosis” (TSH < 0.3 mIU/L and s-TT3 > 2.2 nmol/L or s-TT4 > 165 nmol/L) and “subclinical thyrotoxicosis” (TSH < 0.3 mIU/L, s-TT3: 1.3–2.2 nmol/L and s-TT4: 60–135 nmol/L). The main analysis was by design dichotomous and compared outcomes between patients with and without low TSH. However, we also undertook an analysis where TSH was used as a continuous variable in patients with TSH within the normal range. We were aware that the biological effects of such subtle TSH variation could only be observed in patients with TSH within the normal range as these patients would not be subject to pharmaceutical intervention.

A dynamic model was also developed to encompass changes in thyroid dysfunction over time. Here we constructed a time-dependent cumulative covariate to capture the number of 6 month periods during which the mean TSH level had been low. This took into account all episodes of low TSH in both study groups rather than the baseline classification alone. Any 6 month periods in which no further TSH measurement had been performed were considered periods of normal TSH for the purpose of this conservative sub-analysis.

This study was not a clinical trial. The OPENTHYRO project was approved by the local representative of the Danish Data Protection Agency in the Region of Southern Denmark, Ethics Committee approval is not needed. The study was approved for data access by Statistics Denmark (reference 704047).

Results

Population characteristics

Patients with low TSH were significantly more likely to have a history of major osteoporotic fractures and to use osteoporosis medications than patients with normal TSH. In addition, the prevalence of recent prednisolone use and chronic comorbid conditions such as heart failure, neoplasms, diabetes, and renal failure was also significantly higher in patients with low TSH (Table 1A and 1B). Within the first year after baseline TSH measurement, 26.7% of patients with thyrotoxicosis had begun treatment with anti-thyroid drugs (ATD). This was the case for 40.5% of patients with overt thyrotoxicosis and for 5% of patients with subclinical thyrotoxicosis.

Table 1A. Baseline Descriptives for the Study Population (Women) by Initial TSH Level (Funen County, Denmark 01/01-1996 to 31/12-2010 Using 1995 as a Run-in Year)a
WomenLow TSHNormal TSHp
N%N%
  1. aPlease refer to the Methods section for definitions of low TSH, overt thyrotoxicosis, and subclinical thyrotoxicosis.
 66295.1%12240094.9% 
Age (mean, median, range)61.4 y64.1 y (18–106)49.8 y48.4 y (18–107)<0.0001
18–4418722.8%6465397.2%χ2
45–7428646.2%4309393.8%<0.0001
75+189311.4%1465488.6% 
Severity
Overt367355.4%   
Subclinical156423.6%   
Not classified139221.0%   
Fracture history
Major osteoporotic fracture82912.5%92617.6%<0.0001
Hip3365.1%23331.8%<0.0001
Spine430.7%5310.4%0.01
Forearm4196.3%58464.8%<0.0001
Humerus1462.2%17151.4%<0.0001
History of comorbid conditions
Myocardial infarction2053.1%19161.6%<0.0001
Chronic heart failure6249.4%69255.7%<0.0001
Peripheral vascular disease1352.0%14071.2%<0.0001
Dementia661.0%5530.5%<0.0001
Cerebrovascular disease3715.6%36713.0%<0.0001
Hemiplegia410.6%6890.6%0.6
Pulmonary disease6469.8%119009.7%1.0
Rheumatic and collagen diseases6019.1%89857.3%<0.0001
Ulcer disease2593.9%26492.2%<0.0001
Liver disease2253.4%30082.5%<0.0001
Renal disease380.6%4030.3%0.0009
Diabetes2654.0%34012.8%<0.0001
HIV/AIDS10.0%440.0%0.4
Malignancy6109.2%70545.8%<0.0001
Solid metastatic tumor510.8%5870.5%0.001
Charlson index
0398160.1%8477869.3%χ2
1–2204930.9%3127125.5%<0.0001
3+5999.0%63515.2% 
Medications in last year
Prednisolone5768.7%72695.9%<0.0001
Osteoporosis medications1151.7%12621.0%<0.0001
Table 1B. Baseline Descriptives for the Study Population (Men) by Initial TSH Level (Funen County, Denmark 01/01-1996 to 31/12-2010 Using 1995 as a Run-in Year)a
MenLow TSHNormal TSHp
N%N%
  1. aPlease refer to the Methods section for definitions of low TSH, overt thyrotoxicosis, and subclinical thyrotoxicosis.
 25882.5%9973897.5% 
Age (mean, median, range)62.4 y64.4 y (18–96)50.1 y52.2 y (18–105)<0.0001
18–445791.3%4550098.7%χ2
45–7413683.0%4384397.0%<0.0001
75+6415.8%1039594.2% 
Severity
Overt118445.8%   
Subclinical84432.6%   
Not classified56021.6%   
Fracture history
Major osteoporotic fracture1405.4%51045.1%0.5
Hip592.3%10781.08%<0.0001
Spine160.6%6880.7%0.7
Forearm582.2%28602.9%0.06
Humerus261.0%8540.9%0.4
History of comorbid conditions
Myocardial infarction2258.7%50575.1%<0.0001
Chronic heart failure37314.4%77267.8%<0.0001
Peripheral vascular disease1084.2%21132.1%<0.0001
Dementia301.2%4260.4%<0.0001
Cerebrovascular disease2399.2%45464.6%<0.0001
Hemiplegia240.9%6100.6%0.04
Pulmonary disease32712.6%1026810.3%<0.0001
Rheumatic and collagen diseases26410.2%74007.4%<0.0001
Ulcer disease1535.9%33923.4%<0.0001
Liver disease552.1%19051.9%0.4
Renal disease411.6%6570.7%<0.0001
Diabetes1726.7%46024.6%<0.0001
HIV/AIDS10.0%660.1%0.6
Malignancy2288.8%45714.6%<0.0001
Solid metastatic tumor321.2%6030.6%<0.0001
Charlson index
0134251.9%6439764.6%χ2
1–287633.8%2817428.2%<0.001
3+37014.3%71677.2% 
Medications in last year
Prednisolone2258.7%55955.6%<0.0001
Osteoporosis medications140.5%2070.2%0.0003

Fracture incidence

During a median follow-up time of 7.5 years (range 0–16.9y), 16,543 patients (13.5% of the low TSH group and 6.9% of the normal TSH group) sustained at least one major osteoporotic fracture. Specifically, incident hip fractures were recorded in 6384 persons, clinical spine fractures in 1809 patients, humerus fractures in 3548 patients and forearm fractures in 7346 patients. Incidence rates for hip fractures were significantly higher in patients with low TSH, irrespective of gender, in the age groups 50–74 and 75+ years. For major osteoporotic fractures, rates in subjects with low TSH were significantly increased in all three age categories and in both genders. Fracture free survival curves for TSH level and age are shown in Figures 1 and 2.

Figure 1.

Fracture free survival for hip fractures, stratified by gender and age range. Population N = 231,355 (9217 with low TSH, 222,138 with normal TSH). Outcomes were followed to 30 November 2012. Unadjusted Kaplan-Meier p-values. Please refer to Table 3 for results from the confounder adjusted analysis of hip fracture outcomes.

Figure 2.

Fracture free survival for major osteoporotic fractures, stratified by gender and age range. Population N = 231,355 (9217 with low TSH, 222,138 with normal TSH). Outcomes were followed to 30 November 2012. Unadjusted Kaplan-Meier p-values. Please refer to Table 3 for results from the confounder adjusted analysis of major osteoporotic fracture outcomes.

Influence of TSH levels in euthyroid subjects

In patients with normal TSH at baseline, N = 222,138, the risk of hip fractures (5746 patients) and major osteoporotic fractures (15,300 patients) increased significantly for each 1 SD that TSH was below the population mean. For hip fractures, the multiply adjusted HR per SDTSH was 1.45 (95% CI 1.22–1.71, p < 0.001) for the total population (women: HR 1.44, 95% CI 1.18–1.76, p < 0.001, men: HR 1.49, 95% CI 1.08–2.06, p = 0.016). For major osteoporotic fractures, the multiply adjusted HR per SDTSH was 1.32 (95% CI 1.19–1.46, p < 0.001). In women, the HR was 1.42 (95% CI 1.26–1.60, p < 0.001) and in men 1.10 (95% CI 0.90–1.35, p = 0.36). Thus, the risk of hip fractures was significantly higher with lower euthyroid TSH levels in both genders while the association with major osteoporotic fractures was observed in women only. Because the association between major osteoporotic fractures and TSH could be driven by hip fractures alone (which also count as major osteoporotic fractures) we also conducted an analysis where hip fractures were excluded from the major osteoporotic fracture outcome, so that only spine, forearm, and humerus fractures were combined as an outcome. The association remained significant after removal of hip fractures (HR 1.31; 95% CI 1.16–1.48, p = 0.001).

Long-term prediction of fracture risk in thyrotoxicosis by initial TSH status

The Cox proportional hazards analysis (Table 3) showed a significant association between low TSH and the subsequent risk of hip fracture (HR 1.28, 95% CI 1.17–1.38, p < 0.001) and major osteoporotic fracture (HR 1.15, 95% CI 1.08–1.22, p < 0.001) after adjustment for age, sex, and Charlson comorbidity index. However, further adjustment for known strong risk factors for fracture, including prior osteoporotic fractures, prednisolone use, and specific comorbid conditions, such as pulmonary and collagen disorders, attenuated the association somewhat. Thus, the adjusted risk estimate remained statistically significant for hip fractures (HR 1.16, 95% CI 1.07–1.26, p < 0.001), while the association with major osteoporotic fractures was no longer statistically significant (HR 1.06, 95% CI 0.99–1.12, p = 0.058). When the analysis was stratified by age categories, the increased risk of hip fractures with low TSH was confined to subjects aged 50 +. If hip fractures were removed from the major osteoporotic fractures outcome, the risk was not statistically significant and the effect estimate was close to unity (HR 1.00, 95% CI 0.93–1.08, p = 0.926).

Table 3. Proportional Hazards Model
 Low TSH, allpSubclinicalpOvertp
HRHRHR
Nhyper = 9,217Nhyper = 2,408Nhyper = 4,857
Comparator N = 222,138Comparator N = 222,138Comparator N = 222,138
  1. Bold typeface indicates main analyses while non-bold typeface indicates sub-analyses.
  2. Confounder adjusted model: Age, sex, Charlson index, prednisolone in last year, osteoporosis treatment in last year, diabetes, dementia, heart failure, malignancy, liver disease, rheumatic diseases, pulmonary disease, prior major osteoporotic fracture, year, site (GP practice vs hospital).
  3. Interaction terms for the hip fracture outcome: Age p = 0.17, sex p = 0.84, site p = 0.005.
  4. Interaction terms for the major osteoporotic fracture outcome: Age p = 0.18, sex p = 0.91, site p = 0.001.
Hip fracture (N = 6,384)1.16 (1.07–1.26)<0.0011.13 (0.97–1.31)0.121.14 (1.01–1.29)0.04
Women1.17 (1.06–1.28)<0.0011.11 (0.93–1.32)0.251.17 (1.03–1.34)0.02
Men1.17 (0.95–1.42)0.141.26 (0.91–1.74)0.160.95 (0.67–1.34)0.76
18–490.93 (0.44–1.99)0.860.60 (0.08–4.28)0.610.49 (0.12–1.97)0.31
50–741.33 (1.15–1.53)<0.0011.23 (0.92–1.65)0.161.10 (0.89–1.35)0.38
75+1.13 (1.02–1.26)0.041.23 (0.92–1.65)0.161.12 (0.96–1.31)0.16
Major osteoporotic fracture (N = 16,543)1.06 (0.99–1.12)0.0581.13 (1.014–1.26)<0.051.01 (0.93–1.10)0.86
Women1.05 (0.98–1.12)0.151.11 (0.99–1.25)0.081.01 (0.93–1.11)0.80
Men1.09 (0.94–1.27)0.231.18 (0.93–1.51)0.180.92 (0.73–1.17)0.52
18–491.02 (0.82–1.27)0.861.23 (0.81–1.87)0.340.98 (0.74–1.31)0.91
50–741.07 (0.97–1.16)0.171.19 (1.01–1.42)0.030.97 (0.86–1.10)0.65
75+1.04 (0.96–1.13)0.361.07 (0.93–1.24)0.361.03 (0.91–1.16)0.68

We noted a significant statistical interaction with the site of initial assessment (GP practice versus hospital clinic) and performed a post hoc analysis to examine this (shown at the end of the results section).

Dose response: Long-term prediction of fracture risk in thyrotoxicosis by initial TSH, s-TT3, and s-TT4 status

Overt thyrotoxicosis (Table 3), defined from TSH, TT3, and TT4 levels at presentation, was modestly but significantly associated with an increased risk of hip fracture in the multiply adjusted analysis (HR 1.14, 95% CI 1.01–1.29, p = 0.04). For subclinical thyrotoxicosis, the association was short of statistical significance (HR 1.13, 95% CI 0.97–1.31, p = 0.12), although the mean effect size was similar. By contrast, the risk of major osteoporotic fractures was increased significantly in subclinical thyrotoxicosis but not in overt thyrotoxicosis.

Dose response: Long-term prediction of fracture risk in thyrotoxicosis by cumulative exposure to low TSH level

A proportional hazards model in which a time-dependent covariate had been defined was used to capture the cumulative duration of low TSH (Table 4). For each 6 months in which the mean TSH value was below reference, hip fracture risk increased by a factor of 1.07 (HR, 95% CI 1.04–1.10, p < 0.001), and major osteoporotic fracture risk increased by a factor of 1.05 (HR, 95% CI 1.03–1.07, p < 0.001). This estimate was unaffected by removal of hip fractures from the major osteoporotic fractures outcome. Taken together, these results indicate a doubling of the risk of hip fractures and a 63% increase in risk of a major osteoporotic fracture (MOF) with 5 years of low TSH.

Table 4. Estimating the Adjusted HR for Fracture as a Function of Number of 6 Month Time Intervals in Which the Mean TSH Level Was Below Reference
 HR (95% CI) per 6 months of mean TSH < 0.3 mIU/Lap
  1. This analysis also serves the purpose of accounting for any subsequent recovery of normal thyroid function due to treatment or occurring spontaneously. Population, N = 231,355.
  2. aConfounder adjusted model: Age, sex, Charlson index, prednisolone in last year, osteoporosis treatment in last year, diabetes, dementia, heart failure, malignancy, liver disease, rheumatic diseases, pulmonary disease, prior major osteoporotic fracture, year, site (GP practice versus hospital).
Hip fracture (N = 6,384)1.07 (1.04–1.10)<0.001
Women1.07 (1.04–1.10)<0.001
Men1.07 (0.99–1.14)0.05
Subclinical1.08 (1.04–1.12)<0.001
Overt1.06 (1.03–1.10)0.0005
Major osteoporotic fracture (N = 16,543)1.05 (1.03–1.07)<0.001
Women1.05 (1.03–1.07)<0.001
Men1.05 (0.99–1.10)0.09
Subclinical1.09 (1.06–1.11)<0.0001
Overt1.04 (1.02–1.07)0.0004

General practice or hospital clinic

There was a statistically significant interaction between the type of health provider who requested the initial TSH assessment and the impact of low TSH on fracture risk, with the risk association being considerably stronger for tests requested in a primary practice setting. Thus, a low TSH in a primary practice setting was associated with an adjusted HR for hip fracture of 1.30 (95% CI 1.10–1.52, p < 0.001, Table 5) while the same observation made in a hospital setting was associated with an adjusted HR of only 1.12 (95% CI 1.01–1.23, p = 0.03). This was true also for major osteoporotic fractures, where a statistically significant association was present for the primary practice setting but not for the hospital setting.

Table 5. Low TSH (Based on Biochemistry at First Contact): Post Hoc Analysis, Stratified by Health Provider Requesting the First TSH Assessment. Cox Proportional Hazards Model
 GP practicepHospital clinicp
(N = 103,290)(N = 128,065)
HR (adjusted model)aHR (adjusted model)a
  1. Bold typeface indicates main analyses while non-bold typeface indicates sub-analyses.
  2. aConfounder adjusted model: Age, sex, Charlson index, prednisolone in last year, osteoporosis treatment in last year, diabetes, dementia, heart failure, malignancy, liver disease, rheumatic diseases, pulmonary disease, prior major osteoporotic fracture, and year. Analysis performed due to a significant interaction term (p = 0.01).
Hip fracture (N = 6,384)1.30 (1.10–1.52)0.0011.12 (1.01–1.23)0.03
Women1.30 (1.09–1.55)0.0011.12 (1.01–1.25)0.04
Men1.32 (0.90–1.96)0.171.12 (0.89–1.42)0.33
Major osteoporotic fracture (N = 16,543)1.19 (1.07–1.32)0.0011.01 (0.94–1.09)0.74
Women1.17 (1.05–1.31)0.0011.00 (0.93–1.08)0.99
Men1.24 (0.95–1.63)0.121.05 (0.88–1.25)0.62

Discussion

This study demonstrates that a single, first measurement of low TSH in women without known thyroid disease is associated with an increased long-term risk of hip fractures. The risk increase in men was of the same magnitude but not statistically significant. There was no significant effect modification by age or gender so effects may be homogenous across genders and age strata. Fracture rates were of course lower in younger patients and in men. Though the risk of both fracture outcomes showed an exponential increase with increasing low TSH duration, the association was strongest for hip fractures, which is the fracture type associated with the greatest societal burden[25-27] and risk of mortality.[28] The present study was population-based in the sense that we were able to collect all TSH measurements performed in a well-defined geographical area over a long period of time, but this is not a random sample of the population, as TSH measurements were presumably done on a clinical indication. However, in addition to identifying several thousand patients with low TSH for evaluation of outcomes in a real world setting—providing enough study power to examine cumulative severity and time relationships—the strengths of the study include the use of a single central laboratory and a uniform collection of outcomes and co-morbidity data through national health registers with capture of all hospital contacts and filled prescriptions.

Pathophysiology

In the late 1970 s, thyrotoxicosis was shown to act to increase activation frequency with expansion of the osteoclastic resorption surfaces in the human skeleton, creating a reversible bone loss which is more pronounced at the cortical bone surface.[29] Indeed, within bone biology, thyrotoxicosis is viewed as the archetypal example of a reversible bone loss. This is true only to the extent that the increased bone resorption does not lead to substantial trabecular perforations, which are unable to subsequently heal.[30] Thus, the amount of damage to bone strength would be expected to be a function of the initial trabecular thickness, the extent to which balance by the remodeling cycle is negative, and the duration and severity of thyrotoxicosis. The first two factors are strongly related to age and gender with elderly women being, on average, more susceptible to suffer an irreversible bone loss. Accordingly, a return of normal or near normal BMD has been observed in several clinical studies upon successful treatment of hyperthyroidism.[31-33] As discussed below, observational studies have also demonstrated a certain reversibility of the excess fracture risk in patients with thyrotoxicosis, though fracture risk at some sites may remain increased even a decade after the onset of thyrotoxicosis.

Strengths and limitations of the study

The Danish health service is particularly suited to this type of study because of the availability of long-term data—the National Hospital Discharge Register was established in 1977 and a complete register of all filled prescriptions followed in 1995—and because the registers capture all hospital contacts whether in public or private hospitals.[34] Another strength of the study is that it provides important information about the widely different prognostic implications of a low TSH level diagnosed in general practice and the same observation made in a hospital setting. As would be expected, fracture risk was much lower in patients attending a GP practice than patients examined in hospitals, but this was true to some extent across the full spectrum of thyroid function. More important to the public health message, the risk of hip fracture was increased by about 30% in patients with low TSH in general practice while the increase was a mere 12% in hospital patients. First, the base risk of fractures is much higher in the hospital setting, resulting in higher fracture rates in the hospital comparator group than in the general practice comparator group and a much more modest relative contribution from TSH levels. Second, low TSH levels in general practice are less likely to be the result of non-thyroidal illness and hence are more predictive of long-term thyroid status. Overt rather than subclinical disease was the most frequent finding both in patients diagnosed in the hospital setting and in the general practice setting. This is accounted for by the design of the study, which was based on new cases rather than prevalent cases. Prevalence, which is a function of disease duration and incidence rates, is higher for subclinical disease.

It is a limitation of the study that we could not address prescriptions prior to 1995 or diagnosis codes from primary practice, though we had access to hospital diagnoses as far back as 1977. Thus, while we would not have missed prior hospital treatment for thyroid diseases, milder cases treated exclusively in GP practice prior to the establishment of the National Prescription Database in 1995 would not come to our knowledge. This is unlikely to be a significant cause of bias, however. Our outcome data were captured through national health databases, which have very good accuracy for hip fractures [20, 35] but which have substantial undercapture of spine fractures as recently demonstrated in patients with COPD.[36] In general, we found that the association with hip fractures was more pronounced and robust than the association with major osteoporotic fractures. This may be influenced by the undercapture of some major osteoporotic fractures, especially spine fractures, and an additional reason could be that hip fractures are fall-related to a greater extent than other osteoporotic fractures. We did, however, not observe an effect modification by age—and hip fractures are very rare in young or middle-aged subjects—and we found that the relationship between duration of low TSH and the risk of major osteoporotic fractures persisted after the removal of hip fractures from the outcome.

The absence of information on height, weight, and body mass index as well as lifestyle habits means that we cannot determine to what extent the association with fracture risk is explained by low bone mass or by nonskeletal factors such as muscle weakness or falls. However, the absence of such information does not really detract from the strong association with fracture risk and the clinical message remains that—irrespective of mechanism—low TSH adds to the risk of osteoporotic fractures and the risk rises exponentially with the duration of the disease if untreated. In the context of societal impact, an important strength is the size of the study, which allows for the calculation of highly precise risk estimates even for relatively rare outcomes such as hip fractures.

We deliberately excluded patients who had known thyroid disorders and patients who received treatment with antithyroid drugs or thyroid hormones. Thus, the large group of patients whose TSH levels were due to the thyroxine treatment regimen following thyroid surgery were not included in the study, which aimed to address the long-term consequences of newly diagnosed low TSH. We were also able to adjust for the use of medications such as prednisolone that would influence both fracture risk and TSH levels and for osteoporosis treatment in general.

Comparison with other studies

The relative impact of low TSH on hip fracture is somewhat less pronounced in the present study than previously found in the pivotal Study of Osteoporotic Fractures (SOF). Briefly, Bauer et al.[15] found a threefold increased risk of hip fractures in women with low TSH while the risk in the present study was increased by a mere 17%, even in women with overt thyrotoxicosis after adjustment for numerous risk factors (Table 3). The difference in the apparent importance of thyrotoxicosis on fracture outcomes may lie not only in the TSH level threshold but also in the difference between a nested design with undisclosed thyroid status and a clinical real-world population with the disclosure of thyroid status. Thus, the number of women with low TSH within the Study of Osteoporotic Fractures was low, but resolving power was enhanced by performing a nested case control study within the cohort, which also reduced the number of TSH measurements done to 148 women with hip fracture, 149 women with vertebral fractures, and 398 random control women.[15] Because thyroid hyperfunction was only diagnosed postfracture, albeit using baseline blood samples, patients may have been exposed to sustained low TSH for several years at the time of fracture. By contrast, the impact on fracture risk in the present study reflects the situation where TSH levels are disclosed to the requesting physician, and most patients with abnormally low TSH would be offered treatment unless the condition had proved self-limiting. The present study provides time- and dose-resolved data to address this in several ways. First, overt thyrotoxicosis at debut was not a stronger risk factor for fracture than was subclinical thyrotoxicosis. This suggests that overt thyrotoxicosis was indeed unlikely to be left untreated for a sustained period of time. Second, in patients whose TSH levels at baseline were within the normal range, intervention would not have been clinically indicated. Accordingly, we observed a strong association with fracture risk amounting to a 45% increase in hip fracture risk in both genders for each SD reduction—relative to population mean—in TSH level in euthyroid subjects. This finding very closely matches the observation of a 43% increase in nonvertebral fractures for each 1 SD difference in TSH levels in 2374 euthyroid, postmenopausal women at five European study sites[14]—one of a small number of studies that have addressed the relationship between TSH levels and fractures in euthyroid individuals by specifically excluding persons with abnormal thyroid function. In the MrOs study, Waring et al.[11] found an increase in fracture risk of HR 1.21(95% CI, 1.00–1.47) for each SD decrease in TSH over the full spectrum of TSH values including abnormal ones. Within the Scottish GP-based TEARS study,[37] 1020 persons with low TSH (<0.3 mU/l), were compared with 11,012 persons with normal TSH for fracture outcomes. A strongly increased risk of admission with osteoporotic fractures was seen in the low TSH population, HR 2.02 (95% CI 1.55–2.62). This estimate was adjusted for age, sex, history of hyperthyroidism, history of osteoporotic fracture, and diabetic status. It is not clear if the effect size would have been reduced if the authors had had the opportunity to account for other chronic disorders or use of glucocorticoids and osteoporosis drugs, as was the case in the present study. In addition to the fracture studies a number of studies have linked low TSH to low BMD in this setting.[38-40] In accordance with previous studies, we observed no increased fracture risk in younger men and women[12] though interaction terms with age were not statistically significant. Third, we observed an exponential increase in fracture risk with time, in a time-resolved analysis, in patients who did not quickly become euthyroid and were exposed to low TSH levels for an extended period of time. Hip fracture risk increased by 50% with 3 years of low TSH. Taken together, these results highlight the importance of timely intervention.

The single largest study to examine the association between thyrotoxicosis and the risk of osteoporotic fractures was a study by Vestergaard et al.,[41] which was conducted in 2002 and used Danish national health data for the period 1977–2000 in a case-control design of 124,655 fracture cases and 373,962 control subjects. The study included 5270 patients with thyrotoxicosis and found an increased risk of fractures in general in the first 5 years after diagnosis while hip fracture risk remained significantly increased beyond 10 years after the first diagnosis of thyrotoxicosis. Unlike the present study, researchers did not have access to serum biochemistry and hence had to rely on hospital diagnoses and on prescriptions filled for antithyroid drugs. The latter showed no statistical association with fracture risk.

In the present study we found an association between low TSH levels and the risk of major osteoporotic fractures in the total population, though statistically significant in women only. In our study, the types of osteoporotic fractures that were more strongly associated with TSH values were fractures of the forearm—which is a rare osteoporotic fracture in men[42]—and fractures of the hip. In a recent report from the MrOS study in the US, TSH levels were significantly associated with the risk of hip fractures but not with the risk of other appendicular fractures.[11] This again is most likely explained by the relative rarity of non-hip appendicular fractures. However, significant gender by drug interaction has been reported in a recent analysis of levothyroxine dose and fracture risk in men and women aged 70 +. This supports the observation that postmenopausal women could be more susceptible to the skeletal consequences of excess thyroid hormones than elderly men.[43]

Conclusions

In conclusion, in a population-based observational register cohort, a single, first measurement of low TSH in a patient without known thyroid disease was associated with an increased long-term risk of hip fracture, which remained significant after adjusting for confounders. Moreover, the risk increased exponentially by the length of time during which TSH remained low. Timely intervention is recommended even in mild/subclinical thyrotoxicosis diagnosed in primary practice due to the increased risk of fractures—and also pronounced adverse health outcomes from other organ systems (1–5)—in prolonged thyrotoxicosis.

Disclosures

No authors received specific funding for this project. Bo Abrahamsen: Grants from or conducted trials for Novartis, Nycomed/Takeda, and Amgen. Advisory board member Nycomed/Takeda, Merck, and Amgen. Speaker's fees from Nycomed/Takeda, Amgen, Merck, and Eli Lilly. Laszlo Hegedüs is supported by an unrestricted grant from the Novo Nordisk Foundation.

Acknowledgments

Authors' roles: All authors contributed to the design and interpretation of the study and to writing the manuscript. HLJ, ASL, and BA collated the data. BA performed the statistical analysis and takes responsibility for the accuracy of this.

Ancillary