The potential association between thyroid hormones and breast cancer has been investigated in a large number of studies without conclusive results. This study investigated triiodothyronine (T3) levels in relation to breast cancer mortality in a population with no breast cancer patients at baseline. An additional aim was to study T3 levels in relation to mortality from other cancers and all-cause mortality.
Design and methods
This was a population-based prospective cohort study including 2185 women in whom T3 levels were measured as part of a preventive health project, i.e. before diagnosis in women who later developed breast cancer. Mean follow-up was 24.1 years and record-linkage to The Swedish Cause-of-Death registry identified 471 women who died: 26 out of breast cancer and 182 from other cancers. Mortality was assessed using a Cox's analysis, yielding hazard ratios (HRs), with 95% confidence intervals. Analyses of T3 as a continuous variable were repeated for pre- and peri/postmenopausal women separately.
T3 levels were positively associated with the risk of breast cancer-specific death in the age-adjusted analysis: HR for T3 as a continuous variable was 2.80 (1.26–6.25). However, the crude analysis did not reach statistical significance. Breast cancer mortality was even higher in postmenopausal women: 3.73 (1.69–8.22), but stratified analyses included few events. There were no statistically significant associations between T3 levels and deaths from other cancers, age-adjusted HR: 1.09 (0.72–1.65) or all-cause mortality (1.25:0.97–1.60).
This study, the first of its kind on prospectively measured T3 levels, indicates that T3 levels are positively associated with breast cancer-specific mortality and that this is not related to a general effect on all-cause mortality.
Thyroid disorders are common in women and an association with breast cancer would have a large impact. A large number of cross-sectional studies have investigated the association between breast cancer and thyroid disease, or levels of thyroid hormones, but these studies have been inconclusive (1, 2, 3). Two recent prospective studies have, however, found a positive association between pre-diagnostic thyroid hormone levels and breast cancer risk (4, 5). These studies did not find any corresponding association between TSH and risk (ibid).
In spite of the potentially large clinical relevance, there are only two small studies, including 84 and 47 patients respectively, on thyroid hormones and clinical outcome, i.e. survival following breast cancer, but they found no clear associations (6, 7). A problem in studies measuring thyroid hormones in breast cancer patients is, however, that the disease per se, surgery, or adjuvant treatment may affect hormonal levels.
In order to investigate whether there is a clinically important association between thyroid hormone levels and breast cancer, the most relevant measurement is probably the breast cancer-specific mortality in a population with no breast cancer at baseline; this would reflect the net result of incidence and survival in a population. The only previous study on thyroid conditions and breast cancer mortality to our knowledge is Goldman et al. (8), but they found no association. They did, however, not include information on thyroid hormonal levels.
This study is a population-based, prospective cohort study including 2185 women with no breast cancer at baseline and in whom triiodothyronine (T3) levels were measured as part of a preventive health project, i.e. before diagnosis in women who developed breast cancer. In a previous study based on this cohort, we have found a positive association between T3 levels and the risk of breast cancer (4), but it is not known whether this leads to a corresponding increase in breast cancer mortality.
The aim of this study was to investigate serum levels of T3 in relation to breast cancer mortality in a population with no breast cancer patients at baseline. An additional aim was to study T3 levels in relation to all-cause mortality.
Materials and methods
The Malmö Preventive Project
Originally, 10 902 women participated in the Malmö Preventive Project. The project was established in 1974 when residents in Malmö, a city in southern Sweden, were invited to participate in a health survey. Entire birth cohorts, men and women, were examined until 1992 when the department closed. Approximately 70% of invited subjects participated (9).
All women answered a questionnaire concerning sociodemographic information, lifestyle habits, and medical history. Questions on reproductive factors, e.g. menopausal status, were only included in women screened from April 1983 and onward. BMI (kg/m2) was assessed at baseline examination (9). A subject was considered to have a history of goiter if the question ‘have you been treated for goiter’ was answered with ‘yes’. There was no available information on specific type of previous thyroid disorders or type of treatment. The study was approved by the ethics committee at Lund University: Dnr 652/2005 and Dnr 501/2006.
Blood samples were taken after an overnight fast with the patient in the supine position. The serum samples were analyzed for T3 and TSH in women born in 1928 and 1941 and examined in 1983 and 1984. In women born in 1935 (examined from 1990 to 1992), T3 was only measured in a subset of all women, i.e. those with pathological TSH values, a history of thyroid disease, or those with an enlarged thyroid gland at examination. In addition to this, the attending physician could also decide to analyze T3 (4). No analysis of thyroxin (T4) had been performed at baseline in these women. T3 was measured by a double antibody RIA (reference interval 0.9–3.2 mmol/l) (10). Only six women had a value above the upper reference limit.
Study cohort and follow-up on mortality
Among 10 902 women, reproductive data including menopausal status had been assessed in 8051 subjects. T3 had been measured in 2383 women (1161 born in 1928, 907 born in 1941, and 315 born in 1935). Women with prevalent breast cancer (n=35), goiter (n=167), or both (n=4) were identified and excluded from the study. Finally, the study population consisted of 2185 women with information on T3 and without breast cancer at baseline or a record of goiter. Information on vital status and cause of death was retrieved from the Swedish Cause-of-Death Registry up until 31 December 2010.
Quartile cut-points for T3 were based on the distribution among all women in the study population. T3 levels were investigated in relation to factors known to be associated with subsequent mortality, i.e. age at baseline, educational levels, alcohol consumption, and BMI. A Kendall's τ-b test was used giving correlation coefficients (τb) and corresponding P values. Smoking was compared in relation to T3 quartiles using a χ2 test. A Bonferroni's correction was performed for P values in pairwise χ2 comparisons multiplying the P value with the number of comparisons. A two-sided P value <0.05 was regarded as statistically significant. Missing categories were not included in these tests.
The primary endpoint was breast cancer as underlying cause of death. In addition to this, death from other cancers, death from other causes, and all causes were included in the main analysis. The distribution of the above mortality-related factors was examined in different groups defined by vital status and cause of death. These distributions were tested using the χ2 test as described earlier for smoking.
Each woman was followed from baseline until the end of follow-up, 31 December 2010, or until she died. Mean follow-up was 24.1 years (s.d. 5.3) and total follow-up included 52 579 person-years. A Cox's proportional hazards analysis was used to estimate hazard ratios (HRs) with a confidence interval (CI) of 95% in relation to T3 levels. T3 levels were introduced as a continuous variable in the main analyses. T3 was not normally distributed (P<0.001 using a one sample Kolmogorov–Smirnov test), but to give estimates more readily interpretable, the original T3 values were used to obtain HRs. However, T3 values were also transformed using the natural logarithm in order to confirm whether the associations were statistically significant or not. Analyses were adjusted first for age at baseline and in a final model for all potential confounders. The limited number of breast cancer deaths did not allow inclusion of all covariates in the same model, but in relation to this endpoint, age and one additional factor at a time were included in the final model. In this way, a maximum of three variables were included in each model. In tables, only the model with the largest change from the crude HR for T3 was reported. The primary endpoint, breast cancer as cause of death, was also analyzed with T3 as a continuous variable stratified for menopausal status as we, in our previous study on T3 levels and breast cancer incidence, found that the association was considerably stronger in postmenopausal women compared with premenopausal. The stratified analyses did only allow the inclusion of one additional covariate, and we considered age at baseline to be most important. The stratified analyses were further evaluated by including a term for interaction in the model between T3 levels and menopausal status. A P value <0.05 was considered as a statistically significant interaction. Finally, breast cancer death was analyzed in relation to different T3 quartiles where breast cancer-specific mortality was calculated per 10 000 person-years. Corresponding HRs were calculated as earlier.
It is possible that subclinical disease may affect T3 levels, and in a sensitivity analysis, all analyses were repeated excluding all women who died before 2 years following baseline examination. SPSS version 18.0 (Lund, Sweden) was used for all analyses.
Women in higher T3 quartiles were relatively old (τb=0.29; P<0.001), they had a lower educational level (τb=−0.10; P<0.001), a lower alcohol consumption (τb=−0.07; P<0.001), and they were more often overweight than women in lower T3 quartiles (τb=0.17; P<0.001; Table 1). There was no statistically significant association between T3 levels and smoking (overall P=0.82).
Distribution of potential determinants for mortality according to serum T3 level. Column percent P values are given in italics.
|T3 quartile (T3 range (mIU/l))|
|Factor||1 (n=494 (<1.50))||2 (n=704 (1.60–1.80))||3 (n=456 (1.90–2.00))||4 (n=531 (2.10–4.30))||All (n=2185 (0.60–4.30))||Kendall's τ-b coefficient (τb; P values)|
|Less than every week||55.3||60.7||62.5||60.1||59.7||(<0.001)|
Overall P value from the χ2 test. Not including missing values.
Pairwise χ2 tests, with corrected P values (multiplied with 3). Not including missing values.
Women who died from cancers other than breast cancer, or from causes other than cancer, were considerably older at baseline compared with women alive at the end of follow-up (P<0.001; Table 2). Women who died from causes other than cancer during follow-up had a lower educational level (P=0.006), reported a lower alcohol consumption (P=0.03), and had a higher BMI (P<0.001) than women alive at the end of follow-up. Current smoking was more common among women who died from cancers other than breast cancer, and all causes other than cancer, compared with women alive at the end of follow-up (P<0.001).
Distribution of potential determinants for mortality according to vital status. Column percent P values are given in italics.
|Factor||Alive (n=1714)||Dead from breast cancer (n=26)||Dead from other cancers (n=182)||Dead from other causes (n=263)||Dead from all causes (n=471)||χ2, P valuea|
|Less than every week||59.9||50.0||63.2||57.0||59.0|
Overall P value from the χ2 test tested for alive/dead from breast cancer/dead from other cancers/dead from other causes. Not including missing values.
Pairwise χ2 tests, with corrected P values (multiplied with 3). Not including missing values.
In this study, we observed a positive association between T3 levels and the risk of death from breast cancer (Table 3). The crude analysis did not reach statistical significance, but the risk increased following adjustment for age and other potential confounders. This association with mortality from breast cancer was only apparent among postmenopausal women compared with premenopausal. However, the terms for interaction between T3 levels and menopausal status did not reach statistical significance. All statistically significant results in the continuous analyses were confirmed using the logarithmic T3 value (data not shown). In the quartile analysis, the risk of breast cancer death was higher in the fourth T3 quartile compared with the first quartile (Table 4). The limited number of events did not allow a quartile analysis stratified for menopausal status.
Mortality in relation to serum T3 levels. The model with the largest influence on the HR, compared with the crude model, included alcohol consumption in addition to age at baseline; HR and corresponding P values presented in table. All these multivariate models showed statistically significant HRs for T3.
|Cause of death||Subjects in analysis (n)||Dead (n)||HR (95% CI), crude T3||P value, T3||HR (95% CI), age-adjusted T3||P value, T3||HR (95% CI), adjusteda T3||P, value T3|
|All||2185||26||2.22 (0.92–5.37)||0.077||2.80 (1.26–6.25)||0.012||2.82b (1.25–6.37)||0.01b|
|Premenopausalc||863||14||0.81 (0.17–3.76)||0.784||0.93 (0.20–4.37)||0.930||–||–|
|Postmenopausalc||1322||12||4.30d (1.88–9.86)||0.001||3.73e (1.69–8.22)||0.001||–||–|
|Other cancers||2185||182||1.45 (0.99–2.11)||0.055||1.09 (0.72–1.65)||0.685||0.98 (0.64–1.51)||0.937|
|Other causes||2185||445||1.60 (1.26–2.02)||<0.001||1.17 (0.90–1.52)||0.237||1.07 (0.82–1.41)||0.610|
|All causes||2185||471||1.63 (1.30–2.05)||<0.001||1.25 (0.97–1.60)||0.087||1.16 (0.89–1.50)||0.280|
Adjusted for age at baseline, educational level, smoking status, alcohol consumption, and BMI (continuous).
Adjusted for age at baseline and one factor at a time out of educational level, smoking status, alcohol consumption, and BMI (continuous).
Menopausal status at baseline.
Interaction between T3 and menopausal status: P value=0.07.
Interaction between T3 and menopausal status: P value=0.12.
Breast cancer-specific mortality in relation to serum T3 levels. The model with the largest influence on the HR for the fourth quartile, compared with the crude model, included smoking status in addition to age at baseline; HR and corresponding P value presented in table. All these multivariate models showed statistically significant HRs for the fourth T3 quartile, except for the model including age at baseline and BMI where this association was only borderline significant (P=0.05002).
|T3 (quartile)||Subjects (n)||Dead from breast cancer (n)||Mortality/10 000 person-years||HR (95% CI), crude T3||HR (95% CI), age-adjusted T3||HRa (95% CI), adjusted T3|
|2||704||10||5.8||1.82 (0.57–5.81)||2.07 (0.65–6.63)||2.07 (0.65–6.61)|
|3||456||3||2.8||0.91 (0.20–4.05)||1.27 (0.28–5.77)||1.27 (0.28–5.76)|
|4||531||9||7.4||2.48 (0.76–8.09)||3.61 (1.08–12.1)||3.65 (1.09–12.2)|
|Overall P value||–||–||–||0.30||0.15||0.14|
Adjusted for age at baseline and one factor at a time out of educational level, smoking status, alcohol consumption, and BMI (continuous).
The risk of deaths from cancers other than breast cancer, causes other than breast cancer, and all causes were positively associated with T3 levels in the unadjusted analyses (Table 3). However, following adjustment for age, and subsequently for all potential confounders, these estimates were close to unity. When women who died within 2 years following baseline were excluded from the analyses, all results were similar (data not shown).
This is the first prospective study on T3 levels in relation to breast cancer mortality. It shows that pre-diagnostic T3 levels are positively associated with the risk of breast cancer-specific death and that this is not related to a general effect on cancer death or all-cause mortality.
It may be asked whether breast cancer cases in this cohort are representative of the whole breast cancer population. This cohort mainly comprised middle-aged women and 70% of the women invited to the health examination attended (9). As we have no information about exposure to the studied risk factors in women outside this cohort, absolute risks may not be applicable to all age groups or to the general population. However, as there was a wide distribution of T3 levels, it was possible to make internal comparisons between subjects with relatively low and high values respectively. Hence, we consider that our estimates of relative risks were not considerably affected by selection bias.
Incomplete follow-up may affect the results. However, the Swedish Cause-of-Death Registry has been validated and found to have a completeness of about 97.3% in 2008 (11). Moreover, correctness of cause of death due to malignancies has been shown to be higher than 90% in the Swedish Cause-of-Death Registry (12).
Subjects with elevated T3 levels may already be within the health care system, possibly because of related symptoms, and due to this, the diagnosis of breast cancer might be established earlier than for euthyroid subjects. If true, this would lead to early diagnosis and an expected decrease in breast cancer mortality in this group. However, this is not likely as the contrary was found in this study. Furthermore, general mammography screening was introduced in Sweden in 1991 and stands for the majority of detection of breast cancer cases (13). It is also unlikely that thyroid status does affect participation in mammography screening. It should also be noted that most women with T3 levels in the fourth quartile in this study still had levels within the normal range, and these women probably did not have any symptoms of thyroid disease. It is not likely that the results in this study were due to a detection bias.
The method of analysis for T3 remained the same throughout the study period and all analyses were performed using a standardized collection of blood samples. T3 levels were analyzed following blood collection in 1983, 1984 and between 1990 and 1992. There are no recorded values for the coefficient of variance (CV) from the laboratory covering this period. However, reported within-laboratory CV at the time for T3 were in the order of 9–11% according to a large European analysis including 150 laboratories (14). Following this, we consider that T3 levels had a good reliability and that misclassification was probably low.
It is also important to consider true variation in T3 values as there is a well-known circadian and seasonal variation (15). Tracking of individuals, that is ranking of individuals over time, is, however, quite stable (16), and a true variation over time would most likely have led to an undifferential misclassification of T3 and, hence, attenuated risks.
The current study is the largest prospective study to date on thyroid hormone levels and breast cancer mortality, but the number of deaths due to breast cancer was still very limited. CIs were wide and the statistical power was relatively low. This led to an imprecision in the estimates related to T3 and breast cancer mortality. This was specifically apparent in the quartile analysis where CI was very wide, and the overall P value did not reach statistical significance. Moreover, statistical power was an even more serious problem in analyses stratified for menopausal status, and these results, although statistically significant, should be regarded with caution.
An additional problem related to the limited number of breast cancer deaths was that not all confounders could be included at the same time in the final model for this endpoint. However, the models including age at baseline and one additional factor yielded statistically significant positive associations between T3 levels and breast cancer death; this indicates that these factors did not seriously confound the observed association.
Concerning death from cancers other than breast cancer, death from causes other than breast cancer, and all-cause mortality, there were a substantially larger number of events, and we consider that the lack of any strong associations was not merely the result of poor statistical power, i.e. a type II error.
In order to simplify the interpretation of HR obtained in the continuous analysis, we used original values even if they were not normally distributed. However, all statistically significant results in the continuous analyses were confirmed using the logarithmic T3 value. This is also what would be expected given the high correlation between raw and log-transformed T3 values.
The potential association between thyroid hormones and outcome following breast cancer has been discussed for more than a century. Several early studies used thyroid hormones/extracts as breast cancer treatment, but the effect was not apparent (2). We have found that no studies were published during the last 50 years on this topic. Goldman et al. (8), the only study on thyroid conditions and breast cancer mortality, found no excess risk of breast cancer death in women with thyroid disorders. Smyth et al. (17) followed 195 breast cancer patients and they found that positive TPO-Ab status was associated with a favorable survival, i.e. a lower risk of recurrent disease and death. Fiore et al. (6) examined 47 patients, among whom 16 died, and also found that positive thyroid antibody status (TPO-Ab and/or Tg-Ab) was associated with a low mortality, i.e. a good survival. These findings on recurrence-free survival and overall survival in relation to TPO-Ab status were not confirmed by Jiskra et al. (7), including 84 patients where 11 died. Fiore et al. (6) and Jiskra et al. (7) also investigated T4 levels and TSH in relation to overall survival, but they found no clear associations. A problem in the interpretation of the above studies is that thyroid hormones, TSH and TPO-Ab, may indeed be affected by prevalent breast cancer, stress, or treatment. Generally, previous studies have been hampered by a low number of cases and short follow-up, and we have found no study on pre-diagnostic thyroid function and survival following breast cancer. Concerning mortality, no study has investigated thyroid function or T3 levels in relation to breast cancer-specific mortality in a ‘breast cancer healthy’ population.
All-cause mortality was associated with relatively high T3 levels in the crude analysis, but adjusted for age and other potential confounders, all estimates were close to unity. There are several recent meta-analyses and reviews on the potential association between thyroid conditions and all-cause mortality. Overt hyperthyroidism was associated with a slightly increased risk of all-cause mortality (RR=1.21:1.05–1.38) in a meta-analysis by Brandt et al. (18), including more than 30 000 patients. All the included studies were adjusted for age, but none for education, BMI, or alcohol consumption, and only some for smoking. Considering subclinical conditions, subclinical hyperthyroidism was positively associated with all-cause mortality in a meta-analysis of seven cohort studies (HR=1.41:1.12–1.79) using a general population as reference (19). Interestingly, another meta-analysis including ten prospective studies found no statistically significant association between subclinical hyperthyroidism and all-cause mortality (20). This study excluded subjects with goiter, and even in the highest quartile, most subjects were within the reference limits for T3. All in all, this makes it difficult to compare our results to studies based on overt thyroid conditions. Some of the previous studies also included patients treated with T4; that is, these studies may reflect an effect caused by exposure to exogenous hormones.
The biological mechanism explaining the association of T3 levels and breast cancer-specific mortality is not clear. However, the presence of thyroid hormone receptors in human breast and breast cancer cell lines has been established (21, 22, 23, 24), and the proliferative effect of T3 has been confirmed by various experimental studies on breast cancer. Our findings are in line with these data (25, 26, 27, 28).
It has been shown that T3 binds and stimulates the estrogen receptor, acting in synergy with estrogen on breast cancer cell lines, potentiating the estrogenic effect, and enhancing cell proliferation (29). The role of estrogen in carcinogenesis of the breast is well known and the possibility that this effect may be even stronger in conjunction with relatively high levels of T3 could in part explain the results in this study.
The positive association between T3 and breast cancer was specifically strong in postmenopausal women. Therefore, an imbalance between estrogen and T3 with an increased T3/estradiol ratio may be more important than a pure synergistic effect between these two hormones for the risk of developing breast cancer. It has been suggested that this imbalance might enhance breast cancer development (25). This theory is in accordance with our findings that T3 levels in postmenopausal women are positively associated with breast cancer mortality. An important issue for future studies will be to evaluate the widespread, long-term use of T4 treatment in TSH-suppressive doses for benign thyroid disease (30, 31), and the management of women with subclinical hyperthyroidism (32, 33, 34). Such studies will contribute with important evidence whether or not to treat mild and/or subclinical thyroid disorders.
In conclusion, the present prospective cohort study, the first of its kind on prospectively measured T3 levels, indicates that T3 levels are positively associated with breast cancer-specific mortality and that this is not related to a general effect on all-cause mortality.
Declaration of interest
The authors declare that there is no conflict of interest that could be perceived as prejudicing the impartiality of the research reported.
This work was supported by The Swedish Cancer Society, The Ernhold Lundström Foundation, The Einar and Inga Nilsson Foundation, The Malmö University Hospital Cancer Research Fund, The Malmö University Hospital Funds and Donations, The Breast Cancer network at Lund University (BCLU), and The Region Skåne (ALF).
JiskraJBarkmanovaJLimanovaZLánskáVSmutekDPotlukovaEAntosovaM. Thyroid autoimmunity occurs more frequently in women with breast cancer compared to women with colorectal cancer and controls but it has no impact on relapse-free and overall survival. Oncology Reports2007181603–1611.
Thorell JL & Larson SM. Eds. Radioimmunoassay and related techniques. In Methodology and clinical applications pp. 202–203. St luois: C M Mosby 1978
The National Board of Health and Welfare. Cause-of-Death – history methods and validity (in Swedish) www.socialstyrelsen.se/Publikationer2010/2010-4-33 (accessed 19 April 2012)
JohanssonLABjorkenstamCWesterlingR. Unexplained differences between hospital and mortality data indicated mistakes in death certification: an investigation of 1,094 deaths in Sweden during 1995. Journal of Clinical Epidemiology2009621202–1209. (doi:10.1016/j.jclinepi.2009.01.010).
The National Board of Health and Welfare. Mammography – questions and answers (in Swedish) http://www.sos.se/FULLTEXT/114/2002-114-5/2002-114-5.htm (accessed 17 December 2008)
AndersenSPedersenKMBruunNHLaurbergP. Narrow individual variations in serum T(4) and T(3) in normal subjects: a clue to the understanding of subclinical thyroid disease. Journal of Clinical Endocrinology and Metabolism2002871068–1072. (doi:10.1210/jc.87.3.1068).
HaentjensPVan MeerhaegheAPoppeKVelkeniersB. Subclinical thyroid dysfunction and mortality: an estimate of relative and absolute excess all-cause mortality based on time-to-event data from cohort studies. European Journal of Endocrinology2008159329–341. (doi:10.1530/EJE-08-0110).
Lopez-BarahonaMFialkaIGonzalez-SanchoJMAsuncionMGonzalezMIglesiasTBernalJBeugHMunozA. Thyroid hormone regulates stromelysin expression, protease secretion and the morphogenetic potential of normal polarized mammary epithelial cells. EMBO Journal1995141145–1155.