- Email: [email protected]
Influence of thyroid hormone therapy on the fracture rate — A claims data cohort study
Bone 86 (2016) 86–90
Contents lists available at ScienceDirect
Bone journal homepage: www.elsevier.com/locate/bone
Full Length Article
Influence of thyroid hormone therapy on the fracture rate — A claims data cohort study Annika Viniol a,⁎, Lennart Hickstein b,c, Jochen Walker b,c, Norbert Donner-Banzhoff a, Erika Baum a, Annette Becker a a b c
Department of General Practice/Family Medicine, University of Marburg, Karl-von-Frisch-Str. 4, 35043 Marburg, Germany Health Risk Institute, Spittelmarkt 12, 10117 Berlin, Germany Health Analytics Germany, Elsevier, Jägerstraße 41, 10117 Berlin, Germany
a r t i c l e
i n f o
Article history: Received 13 October 2015 Revised 19 February 2016 Accepted 1 March 2016 Available online 2 March 2016 Keywords: Thyroxine hormones Risk factor Fracture risk
a b s t r a c t Introduction: It has been debated for years whether long-term thyroid hormone intake causes fractures. Not only have previous studies suffered from design limitations, they also reached contradictory conclusions. We investigated thyroid hormones (thyroxine) as a possible risk factor for fractures in a cohort of 6.7 million persons based on administrative data. Methods: The database consists of anonymized settlement data of approximately 70 German statutory health insurances covering a time period of six years. All subjects aged 60 and above were included in the study; subjects with repeated thyroxine prescriptions were assigned to the exposure group; members without thyroxine prescriptions to the control group. Outcome was any incident fracture during a declared time period. In order to calculate fracture risk, we performed multivariate cox regression analyses to adjust for confounders. Results: Of 798 770 subjects fulfilling the inclusion criteria, 11.7% took thyroxine regularly and belong to the exposure group. The final cox regression showed that subjects taking thyroxine have a 6.3% higher risk (HR 1.063; CI 1.046–1.080, p = b.0001) than members of the control group. Discussion: The study supports the assumption that long term thyroxine intake leads to an increase in fracture risk among patients older than 60 years. The findings have implications for long term thyroxine treatment. © 2016 Published by Elsevier Inc.
1. Introduction Thyroid hormones (thyroxine) are among the most frequently prescribed drugs in Germany [1]. According to a national survey, the population prevalence for thyroxine intake is 5.1% [2]. Despite the improvement in iodine supply over the last decade [1,3], annual prescription rates for thyroid hormones have risen [1]. In general, thyroxine is regarded highly by both doctors and patients who perceive the hormone as well tolerated and with few side effects [4]. However, for several years there has been a debate on whether longterm thyroxine therapy might increase the fracture incidence [5]. In their nested case–control study, Turner et al. showed an increased fracture rate among patients (aged 70 or more) under long-term thyroxine therapy in comparison to thyroxine patients whose prescription had ended more than 180 days earlier [6]. Still, the evidence remains contradictory [7–10]. There are no large-scaled cohort studies comparing fracture risk for thyroxine patients versus non thyroxine patients. Regarding to the potential association between thyroxine and fractures, two possible mechanisms have been suggested: The most obvious ⁎ Corresponding author at: Karl-von-Frisch-Str. 4,35043 Marburg, Germany. E-mail address: [email protected] (A. Viniol).
http://dx.doi.org/10.1016/j.bone.2016.03.002 8756-3282/© 2016 Published by Elsevier Inc.
mode of action is that thyroxine leads to a decrease of bone mass [11–13] which in turn might result in osteoporosis and osteoporosis related fractures. It is unclear whether a thyroxine induced hyperthyroidism [14,15] is required for this to happen or whether high-normal thyroxine levels can also lead to lower bone mass [16]. Another possible mode of action is hyperthyroidism induced side effects such as arrhythmia which could lead to falls resulting in fractures [17]. The aim of our study was to investigate whether the intake of thyroxine is an independent risk factor for fractures in women and men aged 60 and above. 2. Methods 2.1. Study design and study population We performed a retrospective cohort study on the Health Risk Institute research database# [18]. The database consists of approximately 6.7 million anonymized insured persons belonging to the German statutory health insurances (SHI) covering a time period of six years (2008–2013). The Health Risk Institute database# consists of anonymized settlement data of approximately 70 SHI. In addition to sociodemographic data, the dataset contains information about
A. Viniol et al. / Bone 86 (2016) 86–90
drugs (prescribed by doctors and dispensed by pharmacies) and diagnostic codes (classified by the International Classification of Diseases (ICD), Version 10) as well as claims for medical procedures. In Germany, universal coverage is required by law; insurance organizations therefore have data on most health care provisions for their members. Data were analyzed in cooperation with the Health Risk Institute#. 2.2. Data assessment Ambulatory care physicians are required to submit their claims four times a year by the end of each quarter. As a result, the data set consists of four time units per year, each representing a 3 month period. A total of 24 units were available for our analysis for a time period of six years (2008–2013). The confounder variables were assessed at the first quarter at index year (1 quarter). The exposure confounder was assessed over the whole study period (index year + 5 years: 24 quarters). The last five years (2009–2013, 20 quarters) served as follow-up time for the outcome of interest. 2.2.1. Inclusion criteria We included all subjects aged 60 and above. Subjects who received a minimum of three prescriptions of the active ingredient levothyroxine (thyroxine) during the first recording year (index year) and a minimum of one thyroxine prescription in each of the following recorded years were assigned to the exposure group. Subject who received thyroxine prescriptions neither during the index year nor during follow-up were assigned to the control group. 2.2.2. Exclusion criteria The following cases were excluded from the study: – Subjects with only one or two thyroxine prescription during the index year. – Subjects who received at least three thyroxine prescriptions during the index year but none during follow-up. – Subjects who did not receive any thyroxine prescription during the index year but at least one during follow-up. – Subjects diagnosed with thyroid cancer (ICD code: C73). – Subjects diagnosed with hyperthyreosis (ICD code: E05.‐).
2.3. Study outcomes The primary outcome was any incident fracture during the follow-up period. It was operationalized by the first coded fracture in the system during follow-up. Fracture diagnoses were identified by the following ICD codes: M80.- (fracture with osteoporosis), S22.- (fracture: rips, sternum, thoracic spine), S12.-(fracture: neck area), S02.-(fracture: head area), S32.-(fracture: lumbar vertebral column, pelvis), S42.-(fracture: shoulder, upper arm), S52.-(fracture: forearm), S62.-(fracture: wrist, hand), S72.-(fracture: femur), S82.-(fracture: lower leg, upper ankle), S92.-(fracture: lower ankle, foot), T02.-(fracture of several regions of the body), and T10.-(fracture: upper extremity). In order to exclude codes for fractures which happened before the time period considered, we evaluated only fracture diagnoses not coded in the three previous quarters (incident fractures). The second outcome was any incident fracture of the pelvis or the femur during the designated time period. This was operationalized by the first coded fracture in the system during follow-up time. Fracture diagnoses were identified by following ICD codes: S32.- + S72.-. We chose this secondary outcome because fractures of the pelvis or the femur are associated with an increased morbidity risk [19–21] and thus carry special clinical relevance.
87
2.4. Statistical analysis We calculated descriptive statistics (mean, standard deviation, frequencies, and percentages) for each of the two groups (exposure and control group). To model the relationship between the dichotomous outcome “fracture” (yes/no) and the exposure variable “thyroxine intake”, we calculated a raw cox regression model with only these two variables. Subsequently, we analyzed the two potential confounders “age” and “sex” which fulfill the criteria for confounding according to clinical considerations. We defined confounder as a risk factor for the outcome “fracture” that is also associated with the exposition “thyroxine intake” [22]. In the context of our analysis of the confounder “age”, we conducted a cox regression separately for every 10-year-age group. The confounder “sex” was assessed by a separate cox regression estimated in the same way. Finally, we performed a cox regression analysis for getting a fracture under thyroxine intake adjusted for both confounders. For cox regression, we used 95% Wald Confidence Limits for the estimated hazard ratio. Censoring occurred for any subject reaching the end of follow-up without experiencing any event. Data extraction and statistical analyses were done using SAS 9.2 (SAS Institute, Cary NC). Comorbidities were assessed by the Charlson index [23] which bases upon a validated ICD codes algorithm [24].
3. Results After the exclusion of 107 235 subjects meeting the exclusion criteria mentioned above, data of 798 770 subjects were eligible. The majority was female (51.8%) and, on average, 71.4 years old (reference: year 2008). While 11.7% (n = 93 252) belonged to the exposure group, 705 518 subjects (88.3%) belonged to the controls. Descriptive comparisons showed exposed subjects to be more frequently female (79.3% vs. 48.2%; SMD 0.63) and slightly younger [age (year: mean): 70.80 vs. 71.48; SDM 0.09] than controls. They suffered from more comorbidities [Charlson Index (mean): 1.87 vs. 1.77, SDM 0.05] and took more medications [No. of different ATC codes without thyroxine (mean): 7.16 vs. 5.81; SDM 0.28]. At descriptive baseline comparison, the exposure group had a higher total fracture rate (7.61% vs. 6.09%; SDM 0.06) and a higher risk of fractures of the pelvis/femur (1.62% vs. 1.54%; SDM 0.01) in comparison to controls. Further details about the baseline characteristics are shown in Table 1. The unadjusted cox regression analysis for fracture risk showed that exposed subjects had a 21.3% higher risk (HR 1.213, CI 1.195–1.231, p = b.0001) than members of the control group. This effect was slightly less pronounced regarding the risk of fractures of the pelvis or the femur (HR 1.096, CI 1.067–1.125, p = b.0001). Adjustment for age (categorized) leads to smaller hazard ratios (all fractures: HR 1.213 (raw model) in comparison to HR 1.300 (age group 60–69), HR 1.289 (age group 70–79), HR 1.129 (age group 80– 89), HR 1.132 (age group ≥ 90)) (Table 2) which is in line with confounding by age. Furthermore, age also seems to be an effect modifier. The observed differences in hazard ratios between age groups (80 and above versus 60 to 79 years) suggest effect modification by age in a way that thyroxine might have a stronger influence on fracture risk in the younger age strata. The influence of thyroxine regarding pelvis/ femur fractures also shows significant but weaker associations with similar trends (Table 2). After adjusting for sex, thyroxine is still a significant risk factor for getting a fracture but hazard ratios decrease in the stratified analysis to HR 1.063 (CI 1.046–1.080, p = b .0001) for women and to HR 1.074 (CI 1.033–1.117, p = 0.0004) for men (Table 3), suggesting confounding by sex, but no effect modification. The final adjusted cox regression for risk of fracture showed that subjects taking thyroxine have a 6.4% higher risk (HR 1.064, CI 1.048– 1.080, p = b.0001) than members of the control group (Table 4). For
88
A. Viniol et al. / Bone 86 (2016) 86–90
Table 1 Raw comparison of exposure and control group at index year [n = 798 770]. Characteristics
Exposure group (thyroxine group) (n = 93 252)
Control group (no thyroxine group) (n = 705 518)
Standardized mean difference
Male subjects [no. (%)] Age [years: mean] Dosage of LT [μg: mean (SD)] Charlson Index [mean (SD)] Medication without thyroxine [no. of different ATC codes, mean (SD)] Any fractures [no. (%)] Fractures of pelvis or femur [no. (%)]
19 352 (20.75) 70.80 90.90 (36.42) 1.87 (2.14) 7.16 (4.98)
365 303 (51.78) 71.48 0 (0) 1.77 (2.18) 5.81 (4.80)
0.63 0.09 1 0.05 0.28
7092 (7.61) 1513 (1.62)
42 940 (6.09) 10 884 (1.54)
0.06 0.01
fractures of pelvis or femur we obtained a HR of 1.030 (CI 1.002–1.058, p = 0.0336). 4. Discussion This study investigated the association between thyroxine intake and the development of bone fractures. An unadjusted cox regression analysis showed that subjects taking thyroxine have a 21.3% higher risk for getting a fracture than subjects without thyroxine (HR 1.213, CI 1.195–1.231, p = b .0001). This effect is less pronounced but still significant after adjustment for the confounder “age” and “sex” (HR 1.064, CI 1.048–1.080, p = b0.001). The results of our study are in accordance with the recent study by Turner et al. [6] who also found an association between thyroxine intake and the fracture risk. While both studies are claim data analyses, they differ not only regarding the selection of the control group and the number of covariates but also regarding the study design: Turner et al. did a nested-case–control study and our study was a cohort study. In contrast to our study's inclusion and exclusion criteria, Turner et al. compared patients with long-term thyroxine therapy with patients whose prescription had ended more than 180 days earlier to gain information on the dosage–agency association. In contrast to the studies on the association between thyroxine and fracture risk by Turner [6] and others [7–10], we decided to include only those covariates in the cox regression model which fulfill the confounder definition [22]. This is to avoid colliding which – if ignored – induces bias by inducing false correlation [25]. Confounding by “age” as shown in our study has previously been described with respect to thyroxine intake and fracture rate [4,26,27]. However, our finding that thyroxine has a stronger influence on fracture risk in younger elderly (60 to 79 years) than in older elderly was unexpected. There is evidence that the thyroxine requirement decreases by age [28–30] which might result in an increased risk for thyroxine induced hyperthyroidism [31] leading to osteoporosis and osteoporosis related fractures. Since the dataset did not contain laboratory values, we could not verify this hypothesis. So far there are no other comparable analyses of this association concerning high age groups. We assume that the detected effect modification within age groups may be linked to the strength of causality. In higher age groups, the frequency of other causes of osteoporosis (independent from thyroxine) and falls increases, so that in turn the attributable risk of thyroxine decreases.
Thyroxine might rather be withdrawn in high-age patients in the context of polypharmacy and multimorbidity. Our analyses showed that sex is a confounder but not an effect modificator of the thyroxine–fracture associations as was to be expected against the background of previous literature [32]. Claim data analyses are always susceptible to methodological limitations [33]. While the main variables of our analysis (e.g. socio demographic data, thyroxine prescription rate and fracture diagnoses) can be considered valid indicators, operationalizing incident fractures is rather challenging. Despite extensive efforts there certainly is a risk to overestimate the newly developed fracture diagnosis. Besides, it is possible that fractures which were not referred for imaging were not recorded, e.g. in those patients who did not show thoracic instability. However, both phenomena reflect an undifferentiated information bias which would not influence the observed association. We chose fractures of the pelvis or femur as secondary outcome, not because of their association with osteoporosis but rather for their clinical relevance since they are associated with an increased morbidity risk. Taking into consideration an osteoporosis related subgroup, we should have included vertebral fractures as well. Not having done so might be the reason why the risk for the subgroup of pelvic and femur fractures increased to a lesser extent than that for fractures in general. Osteoporosis induced fractures of the spine might have lead to an increased risk in the unselected group. Another limitation is that we selected ICD code S72 which includes every kind of fracture of the femur even though it is femur fractures that are in fact the most relevant fractures in relation to osteoporosis. The comparatively short follow-up time of six years will lead to an underestimation of the long-term effect of thyroxine and its impact on incident fractures. Given the fact that the average duration of thyroid hormones intake equals 17.5 years [4], we suppose that the majority of our exposed subjects had taken thyroxine for more than six years. Since only 7 years are covered by the data set analyses, it was impossible to demonstrate temporal relationships between thyroxine exposure and fractures. From the patient's safety point of view, thyroxine prescribing habits in Germany are highly problematic. For our research question, however, excessive thyroxine prescribing for indications other than hypothyroidism ensured a stronger contrast between exposed and non-exposed individuals than in previous studies conducted in other countries [34]. We did not adjust for or stratify according to a known diagnosis of
Table 2 Thyroxin and risk of fracture (cox regression analysis — stratified by age). Age groups
Thyroxine intake
60–69 years 70 and 79 years 80 and 89 years ≥90 years
Any fractures
Fractures of pelvis or femur
Hazard ratio
95%-Confidence interval
p
Hazard ratio
95%-Confidence interval
p
1.300 1.289 1.129 1.132
1.270–1.330 1.261–1.317S 1.095–1.164 1.005–1.276
b.0001 b.0001 b.0001 0.0414
1.300 1.289 1.129 1.132
1.170–1.330 1.261–1.317 1.095–1.164 1.005–1.276
b.0001 b.0001 b.0001 0.0414
A. Viniol et al. / Bone 86 (2016) 86–90
89
Table 3 Thyroxin and risk of fracture (cox regression analysis — stratified by sex, adjusted for age). Gender
Female Male
Thyroxine intake Age Thyroxine intake Age
Any fractures
Fractures of pelvis or femur
Hazard ratio
95%-Confidence interval
p
Hazard ratio
95%-Confidence interval
p
1.063 1.052 1.074 1.051
1.046–1.080 1.051–1.052 1.033–1.117 1.050–1.052
b.0001 b.0001 0.0004 b.0001
1.030 1.096 1.010 1.102
1.000–1.060 1.095–1.098 0.938–1.087 1.100–1.104
0.0480 b.0001 0.7977 b.0001
Table 4 Thyroxin and risk of fracture (cox regression analysis — adjusted for sex and age). Any fractures
Thyroxine intake Sex (0 = female; 1 = male) Age (in years)
Fractures of pelvis or femur
Hazard ratio
95%-Confidence interval
p
Hazard ratio
95%-Confidence interval
p
1.064 0.545 1.051
1.048–1.080 0.539–0.551 1.051–1.052
b0.001 b0.001 b0.001
1.030 0.578 1.098
1.002–1.058 0.566–0.589 1.097–1.099
0.0336 b0.001 b0.001
hypothyroidism because a previous study showed a serious lack of validity regarding diagnostic codes for thyroid disease [4].
Ethical approval A claim data study does not require ethical approval.
5. Conclusion Despite the limitations mentioned above, our study supports the assumption that thyroid hormones increase fracture risk among patients older than 60 years. Prospective studies with longer follow-up times are desirable in order to unambiguously answer to our research question. However, since the effect is only small to moderate, sample size and cost of such studies will be prohibitive. In our view, results from this and previous studies justify a cautious approach to prescribing thyroxine. #
The Health Risk Institute (HRI) is founded by SpectrumK, itself a subsidiary of German statutory health insurers, and Elsevier. The HRI pools sickness funds' claims data from 70 different German statutory health insurances to develop patient-level risk predictions and to conduct outcomes research. All insurances are bound to offer the same comprehensive benefit package, as the contents to be reimbursed are set in Social Law (paragraphs 287 SGB V and 75 SGB X of German law). Competing interests Lennart Hickstein is an employee of Elsevier GmbH. Jochen Walker is an employee of Elsevier GmbH and managing director of the Health Risk Institute. Elsevier GmbH is one of the founders of the Health Risk Institute. Annika Viniol, Erika Baum, Norbert Donner-Banzhoff and Annette Becker declare that they have no conflicts of interest in the relation to this article. Author contributorship Annika Viniol planned the study, developed methodological approach, and wrote the manuscript. Lennart Hickstein analyzed data. Jochen Walker developed methodological approach and analyzed data. Norbert Donner-Banzhoff, Erika Baum, and Annette Becker planned the study design. All authors edited the drafted version of the manuscript. Funding This work was supported by human resources of the Department for primary care, University Marburg and the Health Risk Institute in Berlin.
Acknowledgments We also would like to thank Mareike Künkler for providing Englishlanguage editing, improving the precision and fluency of the manuscript.
References [1] U. Schwabe, D. Paffrath, Arzneiverordnungs-Report 2010 (Report of the Prescription of Pharmaceuticals 2010), Springer Medizin Verlag, Heidelberg, 2010. [2] H.U. Melchert, B. Görsch, Beiträge zur Gesundheitsberichterstattung des Bundes: Schilddrüsenhormone und Schilddrüsenmedikamente bei Probanden in den Nationalen Gesundheitssurveys (Articles about health commentatorship of the federation: thyroid hormones at participants of the National health survey), RobertKoch-Institut, Berlin, 2002. [3] M. Thamm, U. Ellert, Symposium zur Studie zur Gesundheit von Kindern und Jugendlichen in Deutschland — Schilddrüsenvergrößerung und Jodverorgung in Deutschland (Symposium of the study about the health of German children and adolescent — thyroid diseases and iodine supply), Bundesgesundheitsbl. (Fed. Health Newspaper) 5 (6) (2006). [4] A. Viniol, S. Bösner, E. Baum, N. Donner-Banzhoff, Forgotten drugs: long-term prescriptions of thyroid hormones — a cross-sectional study, Int. J. Gen. Med. 6 (2013) 329–334. [5] G.P. Leese, R.V. Flynn, Levothyroxine dose and fractures in older adults, Br. Med. J. (Clin. Res. Ed.) 342 (2011) d2250. [6] M.R. Turner, X. Camacho, H.D. Fischer, P.C. Austin, G.M. Anderson, P.A. Rochon, L.L. Lipscombe, Levothyroxine dose and risk of fractures in older adults: nested case– control study, Br. Med. J. (Clin. Res. Ed.) 342 (2011) d2238. [7] P. Vestergaard, L. Rejnmark, L. Mosekilde, Influence of hyper — and hypothyroidism, and the effects of treatment with anti thyroid drugs and levothyroxine on fracture risk, Calcif. Tissue Int. 77 (2005) 139–144. [8] M.C. Sheppard, R. Holder, J.A. Franklyn, Levothyroxine treatment and occurrence of fracture of the hip, Arch. Intern. Med. 162 (2002) 338–343. [9] D.C. Bauer, Risk for fracture in women with low serum levels of thyroid-stimulating hormone, Ann. Intern. Med. 134 (2001) 561. [10] B. Wejda, G. Hintze, B. Katschinski, T. Olbricht, G. Benker, Hip fractures and the thyroid: a case–control study, J. Intern. Med. 237 (1995) 241–247. [11] B. Uzzan, J. Campos, M. Cucherat, P. Nony, J.P. Boissel, G.Y. Perret, Effects on bone mass of long term treatment with thyroid hormones: a meta-analysis, J. Clin. Endocrinol. Metab. 81 (1996) 4278–4289. [12] W.E. Duncan, A. Chang, B. Solomon, L. Wartofsky, Influence of clinical characteristics and parameters associated with thyroid hormone therapy on the bone mineral density of women treated with thyroid hormone, Thyroid 4 (1994) 183–190. [13] G.M. Stall, S. Harris, L.J. Sokoll, B. Dawson-Hughes, Accelerated bone loss in hypothyroid patients overtreated with L-thyroxine, Ann. Intern. Med. 113 (1990) 265–269. [14] M.R. Blum, D.C. Bauer, T.-H. Collet, H.A. Fink, A.R. Cappola, B.R. da Costa, C.D. Wirth, R.P. Peeters, B.O. Asvold, W.P.J. den Elzen, R.N. Luben, M. Imaizumi, A.P. Bremner, A. Gogakos, R. Eastell, P.M. Kearney, E.S. Strotmeyer, E.R. Wallace, M. Hoff, G. Ceresini, F. Rivadeneira, A.G. Uitterlinden, D.J. Stott, R.G.J. Westendorp, K.-T. Khaw, A. Langhammer, L. Ferrucci, J. Gussekloo, G.R. Williams, J.P. Walsh, P. Juni, D. Aujesky, N. Rodondi, Subclinical thyroid dysfunction and fracture risk: a meta-analysis, JAMA 313 (2015) 2055–2065.
90
A. Viniol et al. / Bone 86 (2016) 86–90
[15] B. Mohammadi, V. Haghpanah, S.M. Tavangar, B. Larijani, Modeling the effect of levothyroxine therapy on bone mass density in postmenopausal women: a different approach leads to new inference, Theor. Biol. Med. Model. 4 (2007) 23. [16] P. Hadji, O. Hars, G. Sturm, T. Bauer, G. Emons, K.D. Schulz, The effect of long-term, non-suppressive levothyroxine treatment on quantitative ultrasonometry of bone in women, Eur. J. Endocrinol. 142 (2000) 445–450. [17] R.W. Flynn, S.R. Bonellie, R.T. Jung, T.M. MacDonald, A.D. Morris, G.P. Leese, Serum thyroid-stimulating hormone concentration and morbidity from cardiovascular disease and fractures in patients on long-term thyroxine therapy, J. Clin. Endocrinol. Metab. 95 (2010) 186–193. [18] F. Andersohn, J. Walker, Characteristics and external validity of the German Health Risk Institute (HRI) database, Pharmacoepidemiol. Drug Saf. 25 (2016) 106–109. [19] S.R. Cummings, L.J. Melton, Epidemiology and outcomes of osteoporotic fractures, Lancet 359 (2002) 1761–1767. [20] F. Bleibler, A. Konnopka, P. Benzinger, K. Rapp, H.-H. König, The health burden and costs of incident fractures attributable to osteoporosis from 2010 to 2050 in Germany—a demographic simulation model, Osteoporos. Int. 24 (2013) 835–847. [21] O. Johnell, J.A. Kanis, A. Oden, I. Sernbo, I. Redlund-Johnell, C. Petterson, C. de Laet, B. Jonsson, Mortality after osteoporotic fractures, Osteoporos. Int. 15 (2004) 38–42. [22] M.S. Porta, A Dictionary of Epidemiology, fifth ed. Oxford University Press, Oxford, New York, 2008. [23] M.E. Charlson, P. Pompei, K.L. Ales, C.R. MacKenzie, A new method of classifying prognostic comorbidity in longitudinal studies: development and validation, J. Chronic Dis. 40 (1987) 373–383. [24] H. Quan, V. Sundararajan, P. Halfon, A. Fong, B. Burnand, J.-C. Luthi, L.D. Saunders, C.A. Beck, T.E. Feasby, W.A. Ghali, Coding algorithms for defining comorbidities in ICD-9-CM and ICD-10 administrative data, Med. Care 43 (2005) 1130–1139. [25] I. Shrier, R.W. Platt, Reducing bias through directed acyclic graphs, BMC Med. Res. Methodol. 8 (2008) 70.
[26] S.K.G.M.e.a., T. Spreng, Stürze älterer Menschen: Ergebnisse einer epidemiologischen Erhebung, ZFA, 2000. [27] R.D. Clark, S.R. Lord, I.W. Webster, Clinical parameters associated with falls in an elderly population, Gerontology 39 (1993) 117–123. [28] L.L. Somwaru, A.M. Arnold, N. Joshi, L.P. Fried, A.R. Cappola, High frequency of and factors associated with thyroid hormone over-replacement and under-replacement in men and women aged 65 and over, J. Clin. Endocrinol. Metab. 94 (2009) 1342–1345. [29] A.D. Mooradian, Normal age-related changes in thyroid hormone economy, Clin. Geriatr. Med. 11 (1995) 159–169. [30] C.T. Sawin, A. Geller, J.M. Hershman, W. Castelli, P. Bacharach, The aging thyroid. the use of thyroid hormone in older persons, JAMA 261 (1989) 2653–2655. [31] S.R. Cummings, M.C. Nevitt, W.S. Browner, K. Stone, K.M. Fox, K.E. Ensrud, J. Cauley, D. Black, T.M. Vogt, Risk factors for hip fracture in white women, N. Engl. J. Med. 332 (1995) 767–774. [32] K.L. Holloway, S.L. Brennan, M.A. Kotowicz, G. Bucki-Smith, A.G. Dobbins, E.N. Timney, L.J. Williams, J.A. Pasco, Age- and sex-related patterns of first fracture and fracture prevalence, Calcif. Tissue Int. 96 (2015) 38–44. [33] M.A. Laux G, Grenzen und Chancen der Verwendung von Routinedaten für die Versorgungsforschung (Constraints and chances of claim data in health service research), in: Versorgungsforschung in der Hausarztpraxis (Health service reserach at general practice). [34] W.P.J. den Elzen, A.A. van de Lefèbre –Fliert, M.S., V. Virgini, P.M. Kearney, N. Kerse, C.D. Mallen, M.C., J.C. MyCarthy, R.A., T. Rosemann, H. Schers, d.W.M., D.J. Stott, A. Warner, R.G.J. Westendorp, G.J., N. Rodondi, Internationale variatie in behandelstrategieën van Internationale variatie in behandelstrategieën van huisartsen bij ouderen met subklinische hypothyreoïdie, NHG-Wetenschapsdag, Abstractboek 2013, pp. 29–30.
Contents lists available at ScienceDirect
Bone journal homepage: www.elsevier.com/locate/bone
Full Length Article
Influence of thyroid hormone therapy on the fracture rate — A claims data cohort study Annika Viniol a,⁎, Lennart Hickstein b,c, Jochen Walker b,c, Norbert Donner-Banzhoff a, Erika Baum a, Annette Becker a a b c
Department of General Practice/Family Medicine, University of Marburg, Karl-von-Frisch-Str. 4, 35043 Marburg, Germany Health Risk Institute, Spittelmarkt 12, 10117 Berlin, Germany Health Analytics Germany, Elsevier, Jägerstraße 41, 10117 Berlin, Germany
a r t i c l e
i n f o
Article history: Received 13 October 2015 Revised 19 February 2016 Accepted 1 March 2016 Available online 2 March 2016 Keywords: Thyroxine hormones Risk factor Fracture risk
a b s t r a c t Introduction: It has been debated for years whether long-term thyroid hormone intake causes fractures. Not only have previous studies suffered from design limitations, they also reached contradictory conclusions. We investigated thyroid hormones (thyroxine) as a possible risk factor for fractures in a cohort of 6.7 million persons based on administrative data. Methods: The database consists of anonymized settlement data of approximately 70 German statutory health insurances covering a time period of six years. All subjects aged 60 and above were included in the study; subjects with repeated thyroxine prescriptions were assigned to the exposure group; members without thyroxine prescriptions to the control group. Outcome was any incident fracture during a declared time period. In order to calculate fracture risk, we performed multivariate cox regression analyses to adjust for confounders. Results: Of 798 770 subjects fulfilling the inclusion criteria, 11.7% took thyroxine regularly and belong to the exposure group. The final cox regression showed that subjects taking thyroxine have a 6.3% higher risk (HR 1.063; CI 1.046–1.080, p = b.0001) than members of the control group. Discussion: The study supports the assumption that long term thyroxine intake leads to an increase in fracture risk among patients older than 60 years. The findings have implications for long term thyroxine treatment. © 2016 Published by Elsevier Inc.
1. Introduction Thyroid hormones (thyroxine) are among the most frequently prescribed drugs in Germany [1]. According to a national survey, the population prevalence for thyroxine intake is 5.1% [2]. Despite the improvement in iodine supply over the last decade [1,3], annual prescription rates for thyroid hormones have risen [1]. In general, thyroxine is regarded highly by both doctors and patients who perceive the hormone as well tolerated and with few side effects [4]. However, for several years there has been a debate on whether longterm thyroxine therapy might increase the fracture incidence [5]. In their nested case–control study, Turner et al. showed an increased fracture rate among patients (aged 70 or more) under long-term thyroxine therapy in comparison to thyroxine patients whose prescription had ended more than 180 days earlier [6]. Still, the evidence remains contradictory [7–10]. There are no large-scaled cohort studies comparing fracture risk for thyroxine patients versus non thyroxine patients. Regarding to the potential association between thyroxine and fractures, two possible mechanisms have been suggested: The most obvious ⁎ Corresponding author at: Karl-von-Frisch-Str. 4,35043 Marburg, Germany. E-mail address: [email protected] (A. Viniol).
http://dx.doi.org/10.1016/j.bone.2016.03.002 8756-3282/© 2016 Published by Elsevier Inc.
mode of action is that thyroxine leads to a decrease of bone mass [11–13] which in turn might result in osteoporosis and osteoporosis related fractures. It is unclear whether a thyroxine induced hyperthyroidism [14,15] is required for this to happen or whether high-normal thyroxine levels can also lead to lower bone mass [16]. Another possible mode of action is hyperthyroidism induced side effects such as arrhythmia which could lead to falls resulting in fractures [17]. The aim of our study was to investigate whether the intake of thyroxine is an independent risk factor for fractures in women and men aged 60 and above. 2. Methods 2.1. Study design and study population We performed a retrospective cohort study on the Health Risk Institute research database# [18]. The database consists of approximately 6.7 million anonymized insured persons belonging to the German statutory health insurances (SHI) covering a time period of six years (2008–2013). The Health Risk Institute database# consists of anonymized settlement data of approximately 70 SHI. In addition to sociodemographic data, the dataset contains information about
A. Viniol et al. / Bone 86 (2016) 86–90
drugs (prescribed by doctors and dispensed by pharmacies) and diagnostic codes (classified by the International Classification of Diseases (ICD), Version 10) as well as claims for medical procedures. In Germany, universal coverage is required by law; insurance organizations therefore have data on most health care provisions for their members. Data were analyzed in cooperation with the Health Risk Institute#. 2.2. Data assessment Ambulatory care physicians are required to submit their claims four times a year by the end of each quarter. As a result, the data set consists of four time units per year, each representing a 3 month period. A total of 24 units were available for our analysis for a time period of six years (2008–2013). The confounder variables were assessed at the first quarter at index year (1 quarter). The exposure confounder was assessed over the whole study period (index year + 5 years: 24 quarters). The last five years (2009–2013, 20 quarters) served as follow-up time for the outcome of interest. 2.2.1. Inclusion criteria We included all subjects aged 60 and above. Subjects who received a minimum of three prescriptions of the active ingredient levothyroxine (thyroxine) during the first recording year (index year) and a minimum of one thyroxine prescription in each of the following recorded years were assigned to the exposure group. Subject who received thyroxine prescriptions neither during the index year nor during follow-up were assigned to the control group. 2.2.2. Exclusion criteria The following cases were excluded from the study: – Subjects with only one or two thyroxine prescription during the index year. – Subjects who received at least three thyroxine prescriptions during the index year but none during follow-up. – Subjects who did not receive any thyroxine prescription during the index year but at least one during follow-up. – Subjects diagnosed with thyroid cancer (ICD code: C73). – Subjects diagnosed with hyperthyreosis (ICD code: E05.‐).
2.3. Study outcomes The primary outcome was any incident fracture during the follow-up period. It was operationalized by the first coded fracture in the system during follow-up. Fracture diagnoses were identified by the following ICD codes: M80.- (fracture with osteoporosis), S22.- (fracture: rips, sternum, thoracic spine), S12.-(fracture: neck area), S02.-(fracture: head area), S32.-(fracture: lumbar vertebral column, pelvis), S42.-(fracture: shoulder, upper arm), S52.-(fracture: forearm), S62.-(fracture: wrist, hand), S72.-(fracture: femur), S82.-(fracture: lower leg, upper ankle), S92.-(fracture: lower ankle, foot), T02.-(fracture of several regions of the body), and T10.-(fracture: upper extremity). In order to exclude codes for fractures which happened before the time period considered, we evaluated only fracture diagnoses not coded in the three previous quarters (incident fractures). The second outcome was any incident fracture of the pelvis or the femur during the designated time period. This was operationalized by the first coded fracture in the system during follow-up time. Fracture diagnoses were identified by following ICD codes: S32.- + S72.-. We chose this secondary outcome because fractures of the pelvis or the femur are associated with an increased morbidity risk [19–21] and thus carry special clinical relevance.
87
2.4. Statistical analysis We calculated descriptive statistics (mean, standard deviation, frequencies, and percentages) for each of the two groups (exposure and control group). To model the relationship between the dichotomous outcome “fracture” (yes/no) and the exposure variable “thyroxine intake”, we calculated a raw cox regression model with only these two variables. Subsequently, we analyzed the two potential confounders “age” and “sex” which fulfill the criteria for confounding according to clinical considerations. We defined confounder as a risk factor for the outcome “fracture” that is also associated with the exposition “thyroxine intake” [22]. In the context of our analysis of the confounder “age”, we conducted a cox regression separately for every 10-year-age group. The confounder “sex” was assessed by a separate cox regression estimated in the same way. Finally, we performed a cox regression analysis for getting a fracture under thyroxine intake adjusted for both confounders. For cox regression, we used 95% Wald Confidence Limits for the estimated hazard ratio. Censoring occurred for any subject reaching the end of follow-up without experiencing any event. Data extraction and statistical analyses were done using SAS 9.2 (SAS Institute, Cary NC). Comorbidities were assessed by the Charlson index [23] which bases upon a validated ICD codes algorithm [24].
3. Results After the exclusion of 107 235 subjects meeting the exclusion criteria mentioned above, data of 798 770 subjects were eligible. The majority was female (51.8%) and, on average, 71.4 years old (reference: year 2008). While 11.7% (n = 93 252) belonged to the exposure group, 705 518 subjects (88.3%) belonged to the controls. Descriptive comparisons showed exposed subjects to be more frequently female (79.3% vs. 48.2%; SMD 0.63) and slightly younger [age (year: mean): 70.80 vs. 71.48; SDM 0.09] than controls. They suffered from more comorbidities [Charlson Index (mean): 1.87 vs. 1.77, SDM 0.05] and took more medications [No. of different ATC codes without thyroxine (mean): 7.16 vs. 5.81; SDM 0.28]. At descriptive baseline comparison, the exposure group had a higher total fracture rate (7.61% vs. 6.09%; SDM 0.06) and a higher risk of fractures of the pelvis/femur (1.62% vs. 1.54%; SDM 0.01) in comparison to controls. Further details about the baseline characteristics are shown in Table 1. The unadjusted cox regression analysis for fracture risk showed that exposed subjects had a 21.3% higher risk (HR 1.213, CI 1.195–1.231, p = b.0001) than members of the control group. This effect was slightly less pronounced regarding the risk of fractures of the pelvis or the femur (HR 1.096, CI 1.067–1.125, p = b.0001). Adjustment for age (categorized) leads to smaller hazard ratios (all fractures: HR 1.213 (raw model) in comparison to HR 1.300 (age group 60–69), HR 1.289 (age group 70–79), HR 1.129 (age group 80– 89), HR 1.132 (age group ≥ 90)) (Table 2) which is in line with confounding by age. Furthermore, age also seems to be an effect modifier. The observed differences in hazard ratios between age groups (80 and above versus 60 to 79 years) suggest effect modification by age in a way that thyroxine might have a stronger influence on fracture risk in the younger age strata. The influence of thyroxine regarding pelvis/ femur fractures also shows significant but weaker associations with similar trends (Table 2). After adjusting for sex, thyroxine is still a significant risk factor for getting a fracture but hazard ratios decrease in the stratified analysis to HR 1.063 (CI 1.046–1.080, p = b .0001) for women and to HR 1.074 (CI 1.033–1.117, p = 0.0004) for men (Table 3), suggesting confounding by sex, but no effect modification. The final adjusted cox regression for risk of fracture showed that subjects taking thyroxine have a 6.4% higher risk (HR 1.064, CI 1.048– 1.080, p = b.0001) than members of the control group (Table 4). For
88
A. Viniol et al. / Bone 86 (2016) 86–90
Table 1 Raw comparison of exposure and control group at index year [n = 798 770]. Characteristics
Exposure group (thyroxine group) (n = 93 252)
Control group (no thyroxine group) (n = 705 518)
Standardized mean difference
Male subjects [no. (%)] Age [years: mean] Dosage of LT [μg: mean (SD)] Charlson Index [mean (SD)] Medication without thyroxine [no. of different ATC codes, mean (SD)] Any fractures [no. (%)] Fractures of pelvis or femur [no. (%)]
19 352 (20.75) 70.80 90.90 (36.42) 1.87 (2.14) 7.16 (4.98)
365 303 (51.78) 71.48 0 (0) 1.77 (2.18) 5.81 (4.80)
0.63 0.09 1 0.05 0.28
7092 (7.61) 1513 (1.62)
42 940 (6.09) 10 884 (1.54)
0.06 0.01
fractures of pelvis or femur we obtained a HR of 1.030 (CI 1.002–1.058, p = 0.0336). 4. Discussion This study investigated the association between thyroxine intake and the development of bone fractures. An unadjusted cox regression analysis showed that subjects taking thyroxine have a 21.3% higher risk for getting a fracture than subjects without thyroxine (HR 1.213, CI 1.195–1.231, p = b .0001). This effect is less pronounced but still significant after adjustment for the confounder “age” and “sex” (HR 1.064, CI 1.048–1.080, p = b0.001). The results of our study are in accordance with the recent study by Turner et al. [6] who also found an association between thyroxine intake and the fracture risk. While both studies are claim data analyses, they differ not only regarding the selection of the control group and the number of covariates but also regarding the study design: Turner et al. did a nested-case–control study and our study was a cohort study. In contrast to our study's inclusion and exclusion criteria, Turner et al. compared patients with long-term thyroxine therapy with patients whose prescription had ended more than 180 days earlier to gain information on the dosage–agency association. In contrast to the studies on the association between thyroxine and fracture risk by Turner [6] and others [7–10], we decided to include only those covariates in the cox regression model which fulfill the confounder definition [22]. This is to avoid colliding which – if ignored – induces bias by inducing false correlation [25]. Confounding by “age” as shown in our study has previously been described with respect to thyroxine intake and fracture rate [4,26,27]. However, our finding that thyroxine has a stronger influence on fracture risk in younger elderly (60 to 79 years) than in older elderly was unexpected. There is evidence that the thyroxine requirement decreases by age [28–30] which might result in an increased risk for thyroxine induced hyperthyroidism [31] leading to osteoporosis and osteoporosis related fractures. Since the dataset did not contain laboratory values, we could not verify this hypothesis. So far there are no other comparable analyses of this association concerning high age groups. We assume that the detected effect modification within age groups may be linked to the strength of causality. In higher age groups, the frequency of other causes of osteoporosis (independent from thyroxine) and falls increases, so that in turn the attributable risk of thyroxine decreases.
Thyroxine might rather be withdrawn in high-age patients in the context of polypharmacy and multimorbidity. Our analyses showed that sex is a confounder but not an effect modificator of the thyroxine–fracture associations as was to be expected against the background of previous literature [32]. Claim data analyses are always susceptible to methodological limitations [33]. While the main variables of our analysis (e.g. socio demographic data, thyroxine prescription rate and fracture diagnoses) can be considered valid indicators, operationalizing incident fractures is rather challenging. Despite extensive efforts there certainly is a risk to overestimate the newly developed fracture diagnosis. Besides, it is possible that fractures which were not referred for imaging were not recorded, e.g. in those patients who did not show thoracic instability. However, both phenomena reflect an undifferentiated information bias which would not influence the observed association. We chose fractures of the pelvis or femur as secondary outcome, not because of their association with osteoporosis but rather for their clinical relevance since they are associated with an increased morbidity risk. Taking into consideration an osteoporosis related subgroup, we should have included vertebral fractures as well. Not having done so might be the reason why the risk for the subgroup of pelvic and femur fractures increased to a lesser extent than that for fractures in general. Osteoporosis induced fractures of the spine might have lead to an increased risk in the unselected group. Another limitation is that we selected ICD code S72 which includes every kind of fracture of the femur even though it is femur fractures that are in fact the most relevant fractures in relation to osteoporosis. The comparatively short follow-up time of six years will lead to an underestimation of the long-term effect of thyroxine and its impact on incident fractures. Given the fact that the average duration of thyroid hormones intake equals 17.5 years [4], we suppose that the majority of our exposed subjects had taken thyroxine for more than six years. Since only 7 years are covered by the data set analyses, it was impossible to demonstrate temporal relationships between thyroxine exposure and fractures. From the patient's safety point of view, thyroxine prescribing habits in Germany are highly problematic. For our research question, however, excessive thyroxine prescribing for indications other than hypothyroidism ensured a stronger contrast between exposed and non-exposed individuals than in previous studies conducted in other countries [34]. We did not adjust for or stratify according to a known diagnosis of
Table 2 Thyroxin and risk of fracture (cox regression analysis — stratified by age). Age groups
Thyroxine intake
60–69 years 70 and 79 years 80 and 89 years ≥90 years
Any fractures
Fractures of pelvis or femur
Hazard ratio
95%-Confidence interval
p
Hazard ratio
95%-Confidence interval
p
1.300 1.289 1.129 1.132
1.270–1.330 1.261–1.317S 1.095–1.164 1.005–1.276
b.0001 b.0001 b.0001 0.0414
1.300 1.289 1.129 1.132
1.170–1.330 1.261–1.317 1.095–1.164 1.005–1.276
b.0001 b.0001 b.0001 0.0414
A. Viniol et al. / Bone 86 (2016) 86–90
89
Table 3 Thyroxin and risk of fracture (cox regression analysis — stratified by sex, adjusted for age). Gender
Female Male
Thyroxine intake Age Thyroxine intake Age
Any fractures
Fractures of pelvis or femur
Hazard ratio
95%-Confidence interval
p
Hazard ratio
95%-Confidence interval
p
1.063 1.052 1.074 1.051
1.046–1.080 1.051–1.052 1.033–1.117 1.050–1.052
b.0001 b.0001 0.0004 b.0001
1.030 1.096 1.010 1.102
1.000–1.060 1.095–1.098 0.938–1.087 1.100–1.104
0.0480 b.0001 0.7977 b.0001
Table 4 Thyroxin and risk of fracture (cox regression analysis — adjusted for sex and age). Any fractures
Thyroxine intake Sex (0 = female; 1 = male) Age (in years)
Fractures of pelvis or femur
Hazard ratio
95%-Confidence interval
p
Hazard ratio
95%-Confidence interval
p
1.064 0.545 1.051
1.048–1.080 0.539–0.551 1.051–1.052
b0.001 b0.001 b0.001
1.030 0.578 1.098
1.002–1.058 0.566–0.589 1.097–1.099
0.0336 b0.001 b0.001
hypothyroidism because a previous study showed a serious lack of validity regarding diagnostic codes for thyroid disease [4].
Ethical approval A claim data study does not require ethical approval.
5. Conclusion Despite the limitations mentioned above, our study supports the assumption that thyroid hormones increase fracture risk among patients older than 60 years. Prospective studies with longer follow-up times are desirable in order to unambiguously answer to our research question. However, since the effect is only small to moderate, sample size and cost of such studies will be prohibitive. In our view, results from this and previous studies justify a cautious approach to prescribing thyroxine. #
The Health Risk Institute (HRI) is founded by SpectrumK, itself a subsidiary of German statutory health insurers, and Elsevier. The HRI pools sickness funds' claims data from 70 different German statutory health insurances to develop patient-level risk predictions and to conduct outcomes research. All insurances are bound to offer the same comprehensive benefit package, as the contents to be reimbursed are set in Social Law (paragraphs 287 SGB V and 75 SGB X of German law). Competing interests Lennart Hickstein is an employee of Elsevier GmbH. Jochen Walker is an employee of Elsevier GmbH and managing director of the Health Risk Institute. Elsevier GmbH is one of the founders of the Health Risk Institute. Annika Viniol, Erika Baum, Norbert Donner-Banzhoff and Annette Becker declare that they have no conflicts of interest in the relation to this article. Author contributorship Annika Viniol planned the study, developed methodological approach, and wrote the manuscript. Lennart Hickstein analyzed data. Jochen Walker developed methodological approach and analyzed data. Norbert Donner-Banzhoff, Erika Baum, and Annette Becker planned the study design. All authors edited the drafted version of the manuscript. Funding This work was supported by human resources of the Department for primary care, University Marburg and the Health Risk Institute in Berlin.
Acknowledgments We also would like to thank Mareike Künkler for providing Englishlanguage editing, improving the precision and fluency of the manuscript.
References [1] U. Schwabe, D. Paffrath, Arzneiverordnungs-Report 2010 (Report of the Prescription of Pharmaceuticals 2010), Springer Medizin Verlag, Heidelberg, 2010. [2] H.U. Melchert, B. Görsch, Beiträge zur Gesundheitsberichterstattung des Bundes: Schilddrüsenhormone und Schilddrüsenmedikamente bei Probanden in den Nationalen Gesundheitssurveys (Articles about health commentatorship of the federation: thyroid hormones at participants of the National health survey), RobertKoch-Institut, Berlin, 2002. [3] M. Thamm, U. Ellert, Symposium zur Studie zur Gesundheit von Kindern und Jugendlichen in Deutschland — Schilddrüsenvergrößerung und Jodverorgung in Deutschland (Symposium of the study about the health of German children and adolescent — thyroid diseases and iodine supply), Bundesgesundheitsbl. (Fed. Health Newspaper) 5 (6) (2006). [4] A. Viniol, S. Bösner, E. Baum, N. Donner-Banzhoff, Forgotten drugs: long-term prescriptions of thyroid hormones — a cross-sectional study, Int. J. Gen. Med. 6 (2013) 329–334. [5] G.P. Leese, R.V. Flynn, Levothyroxine dose and fractures in older adults, Br. Med. J. (Clin. Res. Ed.) 342 (2011) d2250. [6] M.R. Turner, X. Camacho, H.D. Fischer, P.C. Austin, G.M. Anderson, P.A. Rochon, L.L. Lipscombe, Levothyroxine dose and risk of fractures in older adults: nested case– control study, Br. Med. J. (Clin. Res. Ed.) 342 (2011) d2238. [7] P. Vestergaard, L. Rejnmark, L. Mosekilde, Influence of hyper — and hypothyroidism, and the effects of treatment with anti thyroid drugs and levothyroxine on fracture risk, Calcif. Tissue Int. 77 (2005) 139–144. [8] M.C. Sheppard, R. Holder, J.A. Franklyn, Levothyroxine treatment and occurrence of fracture of the hip, Arch. Intern. Med. 162 (2002) 338–343. [9] D.C. Bauer, Risk for fracture in women with low serum levels of thyroid-stimulating hormone, Ann. Intern. Med. 134 (2001) 561. [10] B. Wejda, G. Hintze, B. Katschinski, T. Olbricht, G. Benker, Hip fractures and the thyroid: a case–control study, J. Intern. Med. 237 (1995) 241–247. [11] B. Uzzan, J. Campos, M. Cucherat, P. Nony, J.P. Boissel, G.Y. Perret, Effects on bone mass of long term treatment with thyroid hormones: a meta-analysis, J. Clin. Endocrinol. Metab. 81 (1996) 4278–4289. [12] W.E. Duncan, A. Chang, B. Solomon, L. Wartofsky, Influence of clinical characteristics and parameters associated with thyroid hormone therapy on the bone mineral density of women treated with thyroid hormone, Thyroid 4 (1994) 183–190. [13] G.M. Stall, S. Harris, L.J. Sokoll, B. Dawson-Hughes, Accelerated bone loss in hypothyroid patients overtreated with L-thyroxine, Ann. Intern. Med. 113 (1990) 265–269. [14] M.R. Blum, D.C. Bauer, T.-H. Collet, H.A. Fink, A.R. Cappola, B.R. da Costa, C.D. Wirth, R.P. Peeters, B.O. Asvold, W.P.J. den Elzen, R.N. Luben, M. Imaizumi, A.P. Bremner, A. Gogakos, R. Eastell, P.M. Kearney, E.S. Strotmeyer, E.R. Wallace, M. Hoff, G. Ceresini, F. Rivadeneira, A.G. Uitterlinden, D.J. Stott, R.G.J. Westendorp, K.-T. Khaw, A. Langhammer, L. Ferrucci, J. Gussekloo, G.R. Williams, J.P. Walsh, P. Juni, D. Aujesky, N. Rodondi, Subclinical thyroid dysfunction and fracture risk: a meta-analysis, JAMA 313 (2015) 2055–2065.
90
A. Viniol et al. / Bone 86 (2016) 86–90
[15] B. Mohammadi, V. Haghpanah, S.M. Tavangar, B. Larijani, Modeling the effect of levothyroxine therapy on bone mass density in postmenopausal women: a different approach leads to new inference, Theor. Biol. Med. Model. 4 (2007) 23. [16] P. Hadji, O. Hars, G. Sturm, T. Bauer, G. Emons, K.D. Schulz, The effect of long-term, non-suppressive levothyroxine treatment on quantitative ultrasonometry of bone in women, Eur. J. Endocrinol. 142 (2000) 445–450. [17] R.W. Flynn, S.R. Bonellie, R.T. Jung, T.M. MacDonald, A.D. Morris, G.P. Leese, Serum thyroid-stimulating hormone concentration and morbidity from cardiovascular disease and fractures in patients on long-term thyroxine therapy, J. Clin. Endocrinol. Metab. 95 (2010) 186–193. [18] F. Andersohn, J. Walker, Characteristics and external validity of the German Health Risk Institute (HRI) database, Pharmacoepidemiol. Drug Saf. 25 (2016) 106–109. [19] S.R. Cummings, L.J. Melton, Epidemiology and outcomes of osteoporotic fractures, Lancet 359 (2002) 1761–1767. [20] F. Bleibler, A. Konnopka, P. Benzinger, K. Rapp, H.-H. König, The health burden and costs of incident fractures attributable to osteoporosis from 2010 to 2050 in Germany—a demographic simulation model, Osteoporos. Int. 24 (2013) 835–847. [21] O. Johnell, J.A. Kanis, A. Oden, I. Sernbo, I. Redlund-Johnell, C. Petterson, C. de Laet, B. Jonsson, Mortality after osteoporotic fractures, Osteoporos. Int. 15 (2004) 38–42. [22] M.S. Porta, A Dictionary of Epidemiology, fifth ed. Oxford University Press, Oxford, New York, 2008. [23] M.E. Charlson, P. Pompei, K.L. Ales, C.R. MacKenzie, A new method of classifying prognostic comorbidity in longitudinal studies: development and validation, J. Chronic Dis. 40 (1987) 373–383. [24] H. Quan, V. Sundararajan, P. Halfon, A. Fong, B. Burnand, J.-C. Luthi, L.D. Saunders, C.A. Beck, T.E. Feasby, W.A. Ghali, Coding algorithms for defining comorbidities in ICD-9-CM and ICD-10 administrative data, Med. Care 43 (2005) 1130–1139. [25] I. Shrier, R.W. Platt, Reducing bias through directed acyclic graphs, BMC Med. Res. Methodol. 8 (2008) 70.
[26] S.K.G.M.e.a., T. Spreng, Stürze älterer Menschen: Ergebnisse einer epidemiologischen Erhebung, ZFA, 2000. [27] R.D. Clark, S.R. Lord, I.W. Webster, Clinical parameters associated with falls in an elderly population, Gerontology 39 (1993) 117–123. [28] L.L. Somwaru, A.M. Arnold, N. Joshi, L.P. Fried, A.R. Cappola, High frequency of and factors associated with thyroid hormone over-replacement and under-replacement in men and women aged 65 and over, J. Clin. Endocrinol. Metab. 94 (2009) 1342–1345. [29] A.D. Mooradian, Normal age-related changes in thyroid hormone economy, Clin. Geriatr. Med. 11 (1995) 159–169. [30] C.T. Sawin, A. Geller, J.M. Hershman, W. Castelli, P. Bacharach, The aging thyroid. the use of thyroid hormone in older persons, JAMA 261 (1989) 2653–2655. [31] S.R. Cummings, M.C. Nevitt, W.S. Browner, K. Stone, K.M. Fox, K.E. Ensrud, J. Cauley, D. Black, T.M. Vogt, Risk factors for hip fracture in white women, N. Engl. J. Med. 332 (1995) 767–774. [32] K.L. Holloway, S.L. Brennan, M.A. Kotowicz, G. Bucki-Smith, A.G. Dobbins, E.N. Timney, L.J. Williams, J.A. Pasco, Age- and sex-related patterns of first fracture and fracture prevalence, Calcif. Tissue Int. 96 (2015) 38–44. [33] M.A. Laux G, Grenzen und Chancen der Verwendung von Routinedaten für die Versorgungsforschung (Constraints and chances of claim data in health service research), in: Versorgungsforschung in der Hausarztpraxis (Health service reserach at general practice). [34] W.P.J. den Elzen, A.A. van de Lefèbre –Fliert, M.S., V. Virgini, P.M. Kearney, N. Kerse, C.D. Mallen, M.C., J.C. MyCarthy, R.A., T. Rosemann, H. Schers, d.W.M., D.J. Stott, A. Warner, R.G.J. Westendorp, G.J., N. Rodondi, Internationale variatie in behandelstrategieën van Internationale variatie in behandelstrategieën van huisartsen bij ouderen met subklinische hypothyreoïdie, NHG-Wetenschapsdag, Abstractboek 2013, pp. 29–30.