In this prospective study we observed a significant inverse association between body fatness during childhood and adolescence and incidence of breast cancer in premenopausal women, with approximately 50% lower risk for the most overweight compared with the most lean in childhood. The magnitude of the decrease in risk was greater for childhood body fatness than for body fatness at older ages. The inverse association was independent of later BMI and menstrual cycle characteristics, suggesting that body fatness at young ages may influence breast cancer risk through other biologic pathways.
Our findings are consistent with the results of some other studies that have examined this relationship. Le Marchand and colleagues [12
] linked prospectively recorded information on height and weight from census data for over 38,000 women to the Hawaii Tumor Registry, from which 607 cases of breast cancer were identified. In that study BMI at ages 5–9 years, 10–14 years, and 20–24 years were each inversely associated with breast cancer incidence, but the strongest association was for BMI from ages 10–14 years, with an odds ratio of 0.51 for the highest versus the lowest tertile. A second prospective study of 3447 women born at the University Hospital of Helsinki [13
] obtained anthropometric measurements from birth and school health records and linked this information to the National Hospital Discharge Registry and the Cause of Death registry, identifying 177 incident cases of breast cancer. BMI at ages 7–15 years were inversely associated with breast cancer risk, with a RR of 0.83 for each 1 kg/m2
increase in BMI at age 7 years. Both studies, however, lacked information on other important breast cancer risk factors that might confound these associations. In a prospective cohort study conducted in Norway and Sweden [15
], the investigators observed inverse associations of perceived body shape at age 7 years and BMI at age 18 years with incidence of premenopausal breast cancer, with approximately 30% decreased risk for those who were fat or very fat at age 7 years compared with those who were average. In contrast to our findings, the association was strongest for adult BMI at enrollment, and the associations for perceived body shape at age 7 years and BMI at age 18 years were no longer significant after this adjustment. However, only 733 cases were included in that study, which could explain the lack of significance. A recent record linkage study conducted in Denmark that included over 3000 breast cancer cases [16
] also observed a modest inverse association for BMI at age 14 years, based on information from school health records, but data on adult BMI were not available.
Other epidemiologic studies have assessed body fatness at young ages through recall. Most of these utilized a case-control design in which women with breast cancer and cancer-free controls were asked to categorize their relative weight compared with other girls at specific ages. The majority found inverse associations for body fatness during the childhood and teenage years, observing a 30–50% decrease in risk for those who reported being heavier or much heavier compared with those who recalled being thin or average size [4
]. Several others [8
], however, have not found such inverse associations.
Two previous studies have assessed the relationship between body fatness at young ages and breast cancer risk using the same 9-level figure drawing as in our study. In a large Swedish population-based case-control study of 3345 cases of invasive breast cancer and 3454 controls between ages 50 and 74 years [6
], body fatness at ages 7 and 18 years were both inversely associated with postmenopausal breast cancer risk, with a RR of 0.38 for figure level 7 or greater versus level 4 at age 7 years. In a largely retrospective analysis within the earlier Nurses' Health Study [14
], body fatness at ages 5, 10, and 20 years were each inversely associated with risk of premenopausal breast cancer, with the strongest association for age 10 years. When body fatness at these ages were mutually adjusted for one another, the RR for figure level 5 or greater compared with level 1 was 0.60 for body fatness at age 10 years. A similar pattern was also observed for postmenopausal breast cancer.
Ours is one of the first prospective studies to examine the relation between childhood body fatness and breast cancer incidence, and we were able to control for a broad range of factors, both in early life and adulthood, that were not available in earlier studies. In addition, unlike most previous studies, we adjusted for later BMI using several different variables, including the cumulative average of BMI at age 18 years and all subsequent BMI reports to obtain the best long-term measure. Even with this adjustment, the inverse associations for body fatness during childhood and adolescence remained strong and statistically significant, suggesting that greater body fatness at early stages of life, perhaps even before puberty, may lower breast cancer risk. In addition, in the analyses stratified by current BMI, greater childhood body fatness was associated with reduced risk of breast cancer among both lean and heavy women, which indicates that greater childhood body fatness may confer a lasting protective effect. The biologic mechanisms that would explain this, however, are not well understood. One theory postulates that more overweight girls may experience slower pubertal growth and sexual maturation, despite their earlier menarche [6
]. In the Harvard Longitudinal Study of Child Health and Development [34
], leaner body mass at age 10 years was predictive of more rapid adolescent growth, and in the Nurses' Health Study [14
] adolescents in the highest two quintiles of estimated growth rate had nearly 50% increased risk of premenopausal breast cancer. Rapid adolescent growth may increase breast cancer risk by increasing levels of growth hormones and epithelial proliferation in the breast or by decreasing the amount of time for repair of DNA damage [19
The effect of body fatness at young ages may also be mediated through hormonal pathways. Obesity in pre-adolescent and adolescent girls is associated with higher basal insulin levels [35
], which can impair oocyte maturation and stimulate androgen production in the ovary [31
]. Hyperinsulinemia is also associated with decreased plasma levels of sex hormone binding globulin, leading to increases in free (unbound) testosterone and estradiol, and the aromatization of excess androgen to estrogen in adipose tissue may also increase estrogen levels [38
]. Greater waist:hip ratio has been associated with higher serum concentrations of testosterone and estradiol in prepubertal and pubertal girls in some studies [31
] but not all studies [40
]. High levels of androgens in adolescent girls are associated with metabolic features of polycystic ovary syndrome [42
], greater frequency of anovulatory cycles [37
], and reduced fertility later in life [43
]. In a previous study conducted among participants in this cohort [44
], higher BMI at age 18 years was associated with increased risk of irregular and long menstrual cycles between ages 18 and 22 years as well as increased risk of ovulatory infertility in adulthood [44
], and greater body fatness at age 10 years was also associated with moderately increased risk of menstrual cycle irregularities and nulliparity in the present analysis after adjustment for other factors. In addition, menstrual cycle regularity and length were related to breast cancer risk among NHS II participants during the first few years of follow up [45
]. However, the observed associations for body fatness at young ages in the present study were nearly identical among participants with no history of irregular menstrual cycles or infertility, suggesting that these are not intermediate factors and that other mechanisms may be involved.
High levels of sex hormones in overweight prepubertal and pubertal girls may also have a more direct protective effect on breast tissue. Several experiments have shown that neonatal, prepubertal, or pubertal administration of estrogen, prolactin, progesterone, or testosterone in rats leads to differentiation of cells of the mammary gland as well as a substantial reduction in the incidence of mammary tumors following exposure to chemical carcinogens [46
]. Hence, some have recently hypothesized that the timing of exposure to estrogens and other hormones may determine their effects on breast tissue [51
]. Hilakivi-Clarke [51
] has suggested that early estrogen exposure may reduce breast cancer risk by increasing the expression of tumor suppressor genes such as BRCA1
, inducing differentiation of immature breast cells into more mature ductal structures in addition to stimulating epithelial growth. Higher levels of estrogens may be protective in the breasts of young girls, which are less likely to contain malignant cells, but harmful in older women, whose breasts are more likely to have acquired transformed cells.
Of course, alternative explanations for our findings cannot be ruled out entirely. Although two well designed validation studies [25
] demonstrated that long-term recall of body fatness using this figure drawing has high correlations with BMI at the same ages, no participants in either of those studies recalled their figure as greater than level 7 at young ages; hence, the accuracy of recall at the highest levels of body fatness could not be assessed. Furthermore, the validation studies showed that women who were obese had a greater tendency to underestimate their body fatness at young ages than those who were lean, which could exaggerate the observed association for less extreme levels of body fatness. However, this would not explain the overall association, and the decreasing trend that we observed in age at menarche across all levels of body fatness at age 10 years is strong evidence of the validity of our assessment. We also repeated the analyses using the middle category of body fatness at each age and during childhood and adolescence as the referent group, to evaluate whether the observed inverse association could possibly be explained by higher risk among participants who were extremely lean at young ages. When we did this, we still observed significantly lower risk for the most overweight compared with the middle category. For example, for average childhood body fatness, the multivariate RR for the most overweight compared with the middle category (level 2.5–3) was 0.53 (95% CI 0.39–0.72), whereas the multivariate RR for the most lean was 1.11 (95% CI 0.95–1.31); this argues against elevated risk among the most lean as the major explanation for our findings. Other unmeasured factors, especially those during early life and childhood, could also confound the associations for body fatness at young ages, although a confounder would have to be very strong to account for an association of this magnitude.
Detection bias is another possibility because women who are obese as adults may be less likely to get regular screening mammograms [53
], which could delay or reduce the chance of detection. In this population, however, the probability of having a screening mammography was not appreciably related to childhood body fatness or current BMI among women in several age groups. For example, among women ages 50–54 years in 1999, 69.2% of those who were figure level 1 at age 10 years reported having had a screening mammogram within the preceding 2 years, as compared with 71.8% of those who were figure level 5 or greater at age 10 years. These percentages were similar when examined according to current BMI in 1999 among women ages 50–54 years, although a slightly greater proportion of those in the intermediate category of BMI (23–24.9 kg/m2
) reported having had a recent screening mammogram compared with those in either of the two extreme categories (74.4% compared with 69.6% for those with BMI <21 kg/m2
and 69.0% for those with BMI ≥ 30 kg/m2
). We still observed a strong inverse association between early body fatness and breast cancer risk among women who reported having at least one screening mammography. In addition, if easier detection in lean women were the main explanation for our findings, we would have expected the association to become weaker when in situ
cases were excluded, which is not what occurred. The results stratified by tumor size showed an inverse association for both large and small tumors, also arguing against a detection bias.