Thyroid hormones and breast cancer association according to menopausal status and body mass index

The present study is derived from the population-based case-control study “Risk factors for BC in Mexico: mammographic patterns, C-peptide, and growth factors, a multicenter study” (CAMA), which was conducted in three cities of Mexico (Monterrey, Veracruz, and Mexico City) from January 2004 to December 2007 [42]. In summary, the CAMA study included consecutive women with incident BC (cases (n = 1000)), aged 35–69 years, who were required to have had a minimum of 5 years of residency in the study cities and who were recruited from 12 public hospitals (5 from the Mexican Social Security Institute (Instituto Mexicano de Seguro Social – IMSS), 2 from the Institute of Security and Social Services of State Workers (Instituto de Seguridad y Servicios Sociales de los Trabajadores de Estado – ISSSTE), and 5 from the Ministry of Health (Secretaría de Salud – SS)). Nurses from the field staff were based at each hospital Monday to Friday from January 2004 to December 2007. The inclusion criteria for the cases were (a) histopathological confirmation of BC (median of 3 days between diagnosis and inclusion in the study); (b) no previous treatment (radiotherapy, chemotherapy, or antiestrogens) in the last 6 months; and (c) absence of pregnancy. The response rate of patients with BC was 94%. Controls (n = 1074) were selected based on a probabilistic multistage design and were randomly selected considering the catchment area of each of the participating hospitals. Mammography was performed and women with Breast Imaging Reporting and Data System (BI-RADS) categories I and II were included in the study. The response rate of controls was 87%. The control group was matched to the patients according to quinquennial age, health institution affiliation, and state of residence. In-person interviews were conducted at the hospital with the patients and at the homes of the controls to obtain information on their sociodemographics, reproductive health, breast pathology, lifestyle, and co-morbidities. Information on the perception of body image in different stages of life (six stages) was also included. The perception was measured with the use of six pictograms that represent silhouettes from very thin to very obese. In both the patients and the controls, blood samples and anthropometric measurements were obtained at participating hospitals by personnel following standardized procedures and  who were blinded to the study hypotheses. For the analyses of the present study, out of 1000 cases and 1070 controls we excluded 50 and 45 women, respectively, because they answered “yes” to the question “Has a doctor diagnosed you with thyroid disease?” In the remaining 950 cases and 1029 controls, we determined serum thyroid hormone concentrations in a random subsample of 645 cases, among whom 3 % had in situ BC, and 697 controls (Fig. 1). Characteristics (e.g., age, residence, breastfeeding, first-degree BC family history, BMI, and parity) of the subsample and the total population from the CAMA study were not statistically different (data not shown).

Blood samples were obtained from the participants after they had fasted for at least 8 h. The samples were centrifuged at 3200 rpm at room temperature for 15 min, and the serum was separated and maintained at − 20 °C for 3 weeks and then at − 80 °C until use. Total concentrations of total T3 (TT3), total T4 (TT4), thyroid stimulating hormone (TSH), thyroglobulin (Tg), and thyroid peroxidase (TPO) and thyroglobulin (Tg) antibodies (TgAb) were determined in triplicate using a chemiluminescence assay in a Beckman Coulter UniCel DxI 800 system (Brea, CA, USA) according to the manufacturer’s instructions. The coefficients of intra-assay and inter-assay variation were less than 6% for the hormones TT3, TT4, TSH, and Tg. In addition, at the Children’s Hospital Oakland Research Institute, 106 ancestry information markers were genotyped using multiplex polymerase chain reaction (PCR) and Sequenom’s unique baseline extension methodology (Sequenom Inc., San Diego, CA, USA). The details of these markers have been published elsewhere [43, 44]. The laboratory personnel who performed the measurements were blinded to the condition (case or control) of each participant.

In order to obtain high-quality body measurements, trained nurses were assessed for intra-observer and inter-observer reliability until consistent and accurate anthropometric measurements were obtained. We used validated and standardized protocols and calibrated instruments according to Lohaman’s recommendations [45]. Weight was measured using a digital scale (Tanita Corporation of America, Inc., Arlington Heights, IL, USA) and recorded to the nearest 0.1 kg. Height was measured using a stadiometer (SECA, Hamburg, Germany) to the nearest millimeter. Waist circumference was measured at the level of the navel with the patient in standing position, and the hip circumference was measured at the most prominent level of the buttocks with the woman in a standing position. The BMI was calculated as the weight (in kilograms) divided by the height (in meters) squared. The waist-to-hip ratio (WHR) was calculated by dividing the waist circumference (in centimeters) by the hip circumference (in centimeters). The cutoff points used to estimate the association for BMI, waist circumference, hip circumference, and WHR were established according to the distribution of control patients into tertiles, while the cases were assigned according to the controls’ cutoff points. Tertiles were used for the anthropometric variables because Mexican women are mostly in the overweight and obesity categories. In addition, the participating women were asked to select the silhouette that best represented their body shape (using six pictograms that represent body shape from very thin to extremely obese) at different stages of their lives (before and immediately after menarche, between 18 and 20 years of age, before their first pregnancy, between 25 and 35 years, and their current body shape at the time of the interview). The correlation between BMI and silhouette has been reported as 0.67 in adult women [46], and in the study population, the correlation was p = 0.69 (p ≤ 0.001). A trajectory analysis based on a group approach was performed as proposed by Nagin [47]; this approach is based on the identification of groups with different individual trajectories in the study population over time and looks for the most homogeneous clusters. To identify the optimal model (number of groups and trajectories), we used the Bayesian information criterion (BIC) [47, 48]. To place enough subjects in each category for the statistical analysis, the body shape silhouettes were combined into categories. At childhood, at adolescence, at age 18–20 years, and at age before first pregnancy, the thin category included the women who selected silhouette 1. Silhouette 2 was included for the median category, and for obesity, the women who selected silhouettes 3, 4, 5, and 6 were included. For body silhouettes of women aged 25–35 years and for their current body silhouette, the thin category included the women who selected silhouette 1 and 2. Silhouette 3 was included for the median category, and for obesity, the women who selected silhouettes 4, 5, and 6 were included [48].

A semi-quantitative Food Frequency Questionnaire (FFQ) adapted from Willett [49] to the Mexican population and validated in Mexico City [49, 50] was used for the present study. To measure caloric consumption, participants were asked to report frequency of consumption of a typical serving of 104 items in the past year, and responses were converted to average daily consumption. To calculate intakes, we used the nutrient database developed by the National Institute of Nutrition in Mexico [51] and, when necessary, the US Department of Agriculture food composition tables [52].

To measure physical activity, a semi-structured interview to estimate an individual’s time spent performing different physical activities (sleep and light-, moderate-, and vigorous-intensity physical activity) was applied. The interview was based on the 7-day recall questionnaire proposed by Sallis et al. (1985) [53]. For the present study, weekly hours of moderate-intensity physical activity (activities that are tiresome but that do not result in breathlessness) were used. The patients were asked to report physical activity in a typical week 1 year before the appearance of signs and symptoms, to reduce the possibility of reverse causation bias, whereas the controls were asked to report physical activity for the year prior to the survey [42].

Descriptive statistics (medians, interquartile ranges, means, SD and proportions) were calculated for both the premenopausal and postmenopausal women who were categorized as either cases or controls. We described sociodemographic, reproductive health, anthropometry, breast pathology, lifestyle, comorbidities, thyroid function parameters and clinical characteristics of the patients (cases) in terms of clinical stage (early ≤ IIA; advanced ≥ IIB) and histological grade (1, 2, or 3).

To assess the association between BC and TT3 or TT4 serum concentrations, two logistic regression models were used, one for premenopausal and one for postmenopausal women. To build each model, bivariate models were constructed for each variable of interest and potential confounders, then variables with a p value ≤0.20 in the bivariate models were included in each final model. In order to build the most parsimonious models that still explain the data, we left the variables with a p value <0.05 [54, 55]. The dependent variable was BC (yes/no), and the independent variables of interest were TT3 and TT4, which were incorporated into the models as standardized continuous variables (Z = (x-μ)/σ). For each model, odds ratios (OR) and 95% confidence intervals (95% CI) were obtained. For continuous variables such as thyroid function parameters and calorie consumption, we estimated the odds of BC for each increase in SD.

The following are the variables that were considered as potential confounders: (a) sociodemographic variables: age (years), entitlement to a health institution (IMSS, ISSSTE, and Ministry of Health), city of residence (Mexico City, Monterrey, or Veracruz), economic index (low, medium, or high), and educational level (last complete school grade); (b) reproductive health: age at menarche (years), age at menopause (years), time of exposure to endogenous hormones (age at menopause in years to age of menarche in years), parity (number of children born alive), ever use of hormonal contraception (yes/no), age at first full-term pregnancy (years), use of hormones for menopause for more than 1 month (yes/no), and breastfeeding (months); (c) anthropometric measurements: height (cm); (d) breast pathology: personal history of benign breast disease (yes/no) and family history of BC (mother, grandmother, or sisters) (yes/no); (e) lifestyles: hours of moderate-intensity physical activity per week [42, 53], alcohol consumption (consumed on average one or more alcoholic drinks a month for a year (yes/no)), tobacco consumption (smoked at least 100 cigarettes in her lifetime (yes/no), and daily calorie intake (Kcal) [49‐52, 56]; (f) percentage of indigenous ancestry informative markers [44, 57]; (g) comorbidities: diabetes mellitus diagnosed by a physician (yes/no); and (h) the other thyroid function parameters: TSH, Tg, and Tg and thyroperoxidase TPO antibodies (TPO Ab) (Z = (x-μ)/σ). The percentage of ancestry informative markers was considered as a potential confounder due to its potential association with the thyroid hormone profile [57] and because it has been associated with BC [44].

The effect modification of obesity (BMI, waist circumference, hip circumference, WHR, and trajectories according to the silhouettes) was assessed for the association between TT3 or TT4 and BC in premenopausal and postmenopausal women. Multiplicative interactions were evaluated, which considered the thyroid hormones as continuous variables, the anthropometric variables in tertiles (waist circumference (tertiles), hip circumference (tertiles), BMI (tertiles), WHR (tertiles)), and the different trajectories of weight change (constantly low, constantly mid-range, moderate increase, strong increase, or constantly high). We focus our results on the potential effect modification by BMI since this variable has been associated more consistently as a protector against BC in premenopausal women and as a risk factor in postmenopausal women [23, 58‐79]. The results for the rest of the anthropometric variables are presented in additional tables. Given that missing values were lower than 5%, we did not include them in the analysis of multiple models. All models were then analyzed with goodness of fit, model specification, collinearity, and influential values tests according to the procedure proposed by Hosmer and Lemeshow [55]. The analysis was performed using the STATA v13 software (StataCorp, College Station, TX, USA).