Weight and weight changes throughout life and postmenopausal breast cancer risk: a case-control study in France

The case group included incident cases of in situ or invasive breast cancer diagnosed between April 2005 and March 2007 in women aged 25–75 years who resided in the study areas. Patients were recruited in the main cancer hospital in each area (Centre Eugène Marquis in Rennes and Centre Georges-François Leclerc in Dijon), as well as from smaller public and private hospitals that also recruited breast cancer patients. Among the 1553 eligible cases identified during the study period, 163 refused to participate, 151 could not be contacted, and 7 died before the interview, leaving 1232 cases included in the study sample (participation 79.3 %).

Controls were women without a previous history of breast cancer recruited in the general population of each study area. They were selected by phone and were frequency-matched to the cases by 10-year age group. To avoid selection bias that could arise from differential participation rates across categories of socioeconomic status (SES), we obtained a control group with a distribution by SES category similar to the general population, by using predefined numbers of controls by SES calculated from the census data. To recruit the controls, phone numbers of private homes were selected at random from the telephone directory completed beforehand with unlisted numbers. Phone numbers were dialed up to 15 times at different times of the day and different days of the week until contact could be established with the residents. When a woman was living in the residence reached by phone, she was invited to participate to the study, as long as the predefined number of controls in her age and SES stratum was not complete. When this number was exceeded, the woman was excluded. To obtain the desired number of controls within the limits of age and SES categories, approximately 30,000 phone numbers were dialed for identifying 1,731 eligible controls. Among 1731 controls identified by telephone fulfilling eligibility criteria, 260 declined participation and 154 could not be re-contacted for an in-person interview, leaving 1317 women available for the study (participation 76.1 %).

The study was approved by the French Ethic Committee (CCPPRB Kremlin-Bicêtre, Jan 2005), the National Data Protection Agency (Dec 2004) and the Advisory Committee on the Treatment of Health Research Information (Apr 2004). All participants signed informed consent.

Only postmenopausal women were included in the analysis. Women were considered postmenopausal if they had not menstruated for twelve or more months (natural menopause, n = 936), if they had had bilateral oophorectomy (artificial menopause, n = 93), or if they had used MHT (Menopausal Hormone Therapy) before natural cessation of menstruation (n = 352). Women with unknown menopausal status (n = 199) (hysterectomy before cessation of menstruations or unknown date of last menstruation), were considered postmenopausal if they were aged 50 or more years, the median age at menopause in women with natural menopause (n = 174). Women with unknown menopausal status below 50 years old were excluded from the analysis (n = 25). One woman reporting aberrant low weight was excluded from the analysis. In total, the analysis included 1554 postmenopausal women (739 cases and 815 controls).

Data pertaining to study subjects were obtained from a structured questionnaire during in-person interviews conducted by trained interviewers. We defined a reference date for each study subject, which was the date of diagnosis for the cases and the date of selection for the controls. The age at reference date will be referred to below as current age. Only events that occurred before that date were considered in the analyses. We elicited information on socio-demographic characteristics, history of previous diseases, family history of cancer, history of menstruations, use of oral contraceptives, infertility, reproductive history, residential and occupational history, lifetime consumption of alcohol and tobacco, recreational activities, and dietary habits.

Women were invited to report their height at the age of 20 years, and to report their usual weight one year before reference date (hereinafter referred to as current weight) to avoid reporting weight loss that might be due to cancer development. We also elicited information on weight at the ages of 20, 40 and 50 years.

Information on estrogen (ER) and progesterone receptor (PR) status was obtained from the pathology report. Tumors containing more than 10 % positive cells for hormonal receptors were classified as receptor-positive.

Odds ratios (ORs) of breast cancer were calculated for BMI at age 20, BMI at age 50 and BMI at the current age, and for BMI changes from age 20 to current age, from age 20 to 40, from age 40 to 50, and from age 50 to current age. BMIs were calculated as weight in kilograms divided by height in meters squared (kg/m²). BMI at different ages were categorized according to the WHO classification (<18.5; 18.5–25; 25–30; ≥30 kg/m²), except for BMI at age 20 where we used a single category of BMI ≥ 25 kg/m² due to small number of young obese women ≥30 kg/m². BMI changes during different periods of life were categorized in 3 groups. To enable comparisons between BMI gain during different periods of life, and despite uneven distributions, we sought to use the same cut points for defining BMI gain categories in different periods: BMI gain <1 kg/m²; BMI gain ≥ 1 < 3 kg/m²; and BMI gain ≥ 3 kg/m². For BMI changes from age 20 to current age, the highest category of BMI gain was subdivided in two classes (3–6 kg/m² and ≥ 6 kg/m²). We also fitted models using tertiles or quartiles of BMI gain distribution among controls specific to each exposure period, but the findings were very similar and are not shown. To test dose-response trends, we fitted models where BMI at different ages and BMI changes were introduced as continuous variables, assuming a linear relationship between the variable and breast cancer risk, and reported odds ratios for each increment of 5 kg/m² of BMI or BMI change.

Further analyses were conducted to examine whether low BMI at age 20 (<18.5; ≥ 18.5 kg/m²), age at reference date (<65, ≥ 65 years), and use of menopausal hormone therapy (MHT) (current vs past or never) modified the association of breast cancer with BMI at different ages, and BMI changes. In these analyses, we present only the odds ratios associated with continuous variables for BMI and BMI changes. p-values for interaction between BMI or BMI changes and the stratification variables (i.e. BMI at age 20 years, age at reference date and MHT use) were calculated by comparing models with and without an interaction term using the likelihood ratio test.

Odds ratios and 95 % confidence intervals were calculated using unconditional logistic regression models adjusting for the matching variables, i.e. age (5-year age group) and study area, and for breast cancer risk factors in Table 1: age at menarche (≤11, 12, 13, 14, ≥ 15 years), parity (0, 1, 2, 3, ≥ 4 children), age at first full-term pregnancy (<22, 22–24, 25–27, ≥ 28 years), duration of breast-feeding (0, <26, 26–52, >52 weeks), oral contraceptive use (ever, never), family history of breast cancer in first degree relatives (yes, no), MHT use (current, never or past use), recreational physical activity (ever/never), tobacco smoking (never, < 10, ≥ 10 packs-years), and alcohol consumption (≤3, 4–7, 8–14, > 14 glasses per week).

All analyses were also conducted using different categorization for BMI and BMI gain during lifetime, or using weight and weight changes (in kg) instead of BMI. These analyses produced similar findings and are not reported here. We also conducted analyses stratifying the case group according to hormone receptor status of the tumor (ER-positive/ER-negative, PR-positive/PR-negative) using polytomous logistic regression models, but no particular hint emerged from this analysis (not shown).

All analyses were conducted using SAS computer software (version 9.3, Cary, North Carolina).

BMI gain in adulthood has been linked to the risk of postmenopausal breast cancer in previous investigations [5, 8, 10, 13]. Although the timing of weight gain during life may be an important determinant of breast cancer risk, epidemiological evidence is sparse since most studies have assessed weight gain over long periods from early adulthood to the time of cancer diagnosis regardless of specific time periods [5, 8, 9, 14]. Our results suggest that weight gain during the age range period 40–50 years, i.e. in late reproductive period, may be particularly harmful. These findings are consistent with studies that reported increased risk of breast cancer among women who gained weight in middle adulthood [15, 16]. If this is confirmed, it would point to the importance of controlling weight gain in that period of life.

We also observed that the association of breast cancer with current BMI and weight or BMI gain in the age range period 40–50 years was stronger in women above 65 years of age than in younger postmenopausal women, suggesting a relatively long induction period between weight gain and breast cancer occurrence. Alternatively, it is possible that the beneficial effect of adiposity during pre-menopause may compensate the adverse effect of overweight in early post-menopause.

The hypothesis that weight during adolescence or young adulthood may influence breast cancer risk after the menopause is supported by several epidemiological studies reporting an inverse association between weight at a young age and postmenopausal breast cancer [9‐11, 17, 18]. In addition, it was demonstrated that pre-pubertal girls with low weight have higher mammographic density when they become adults [19], and mammographic density is one of the strongest risk factors for breast cancer [20]. It was also postulated that low level of adiposity in the mammary gland may alter breast tissue maturation, making breast tissue more susceptible to carcinogenic stimuli among leaner women [6, 21]. Our data did not confirm the hypothesis of a direct link between low weight at a young age and breast cancer risk after menopause, as no association between BMI at age 20 and postmenopausal breast cancer was detected. However, there was weak indication that weight gain between age 40 and 50 might lead to higher postmenopausal breast cancer risk in women who were leaner at a young age (BMI < 18.5 kg/m²). This finding should be interpreted with care as no statistically significant interaction between BMI at age 20 and BMI gain was seen. Nevertheless, it is consistent with a report from the large US Nurse’s Health Study showing that the association of postmenopausal breast cancer risk with weight gain of 25 kg or more since the age 18 years was stronger in women with BMI below 21 kg/m² at age 18 years than in heavier women (p for interaction 0.05) [13]. This result points to the importance of examining lifelong weight history in order to elucidate the complex relationships between adiposity and breast cancer risk.

The association of postmenopausal breast cancer risk with weight gain between ages 40 and 50 years was apparent only among non-current MHT users, although the interaction between BMI gain and MHT use was not significant. This is consistent with several studies that reported an association between adiposity and postmenopausal breast cancer only among women who did not use MHT [5, 6, 8, 10, 13, 18, 22‐24]. To explain this frequent observation, it has been suggested that the increased levels of circulating estrogens in women treated with hormones are predominant and mask the effects of adiposity on breast carcinogenesis [25].

In our study, incident breast cancers were identified on a population basis in well-defined geographical areas, using inclusion criteria similar to a cancer registry, and using active real-time search in the main cancer hospitals in each area. Controls were carefully selected from the study base controlling for possible differential participation rates across SES categories. In addition, all potentially important confounders were taken into account in the analysis.

The main limitations of the study include the self-reported and recalled history of height and weight. Studies that examined the accuracy of self-reported height and weight compared to measured values consistently reported that height tended to be overestimated and weight to be underestimated by the women [26‐28]. If this applies to our study, then BMI values should be underestimated. Errors due to recalled weight at younger ages are also likely to have occurred, particularly for longer recall [29]. However, we think that misclassification errors due to self-reported or recalled weight were most likely non-differential, and are not probable explanations for the observed associations. Indeed, the cases and the controls were interviewed in the same way using a standardized questionnaire, they were not aware of the specific objectives of the study, and the possible link of weight changes with breast cancer is not a widely known fact among women in France. Moreover the main findings of our study were in line with expectations. As in other studies, chance findings may have occurred especially as we performed a large number of tests. Conversely some associations may have remained undetected due to low statistical power in some analyses. In particular, statistical power was limited to detect interactions between variables. With these limitations in mind, however, we believe that these data are of valuable interest given the scarcity of studies examining the effect of BMI assessed over the lifetime on breast cancer risk, and provide further opportunities for research.