Early Life Factors in Relation to Breast Cancer Risk in Postmenopausal Women¹

The hypothesis that breast cancer risk is influenced by in utero exposure to high levels of pregnancy hormones (1) stimulated considerable research in the last decade (2). Although study results are not entirely consistent (2), factors that may reflect high levels of intrauterine estrogens (dizygotic twin births, high birth weight, and left-handedness) have been associated with increased risk (3, 4, 5, 6, 7), whereas factors reflecting low levels of pregnancy estrogens (pre-eclampsia and later birth order) have been associated with decreased risk (8, 9). Other factors of early life, including exposure to infectious agents, have also been implicated as reducing breast cancer risk (10).

This report is based on a large, United States population-based, case-control study of breast cancer that is described in detail elsewhere (11). Briefly, potentially eligible cases were women of ages 50–79 years with a new diagnosis of invasive breast cancer ascertained from January 1992 to December 1994 by statewide registries in Massachusetts, New Hampshire, and Wisconsin. Eligibility required a listed telephone number and a driver’s license determined by self-report (if <65 years of age). Physicians were queried in advance for permission to contact case women. Of 6839 potential case women, 158 (2%) were excluded at their physician’s request, 293 (4%) had died before contact, 83 (1%) could not be located, and 620 (9%) declined to participate, providing an overall participation rate of 83%. Of the 5685 case participants, 26 gave interviews that were considered unreliable by the interviewer, leaving 5659 case women for analysis.

We selected control women in each state from lists of licensed drivers (ages 50–64) and Medicare beneficiaries (ages 65–79). Control women were selected at random to have an age distribution similar to that of case women; as with case women, eligibility required a listed telephone number. Control women who had a history of breast cancer were ineligible. Of 7655 potential controls, 183 (2%) had died, 124 (2%) could not be located, and 1397 (18%) declined to participate, providing an overall participation rate of 78%. Interviews for 23 of the 5951 control participants were considered unreliable by the interviewer, leaving 5928 control women for analysis.

Potential case and control subjects were contacted initially by mail and subsequently interviewed by telephone; participants were enrolled from July 1992 to July 1995. Early life factors of principal interest included the subject’s birth weight, parental smoking while the mother was pregnant with the study subject, the father’s level of education, the mother’s age at the time of the subject’s birth, the subject’s birth rank, the number of older brothers or sisters, and the gender of siblings in the sibship.

Analyses were confined to exposures occurring before a reference date, which was defined for cases as the date of breast cancer diagnosis; control women were assigned a reference date corresponding to the state-specific average time interval between case diagnosis and interview (approximately 1 year). Women were considered postmenopausal if natural (permanent cessation of periods for at least 6 months) or surgical (bilateral oophorectomy) menopause occurred before the reference date. An algorithm (11) was used to classify menopausal status among women who began taking hormones before their periods had stopped or who had undergone a hysterectomy but were uncertain whether their ovaries had been removed. Premenopausal women (267 cases and 292 controls) and women with unknown menopausal status (116 cases and 150 controls) were excluded from the analyses.

During the data collection period, questions were added to the questionnaire and later withdrawn to make room for new areas of inquiry; thus, the number of cases and controls available for comparisons was determined by the length of time a question was in the questionnaire. Data collection was concurrent for all of the early life variables (other than parental smoking) for at least 1 year, allowing assessment of mutual confounding.

ORs³ and 95% CIs from logistic regression models were used to evaluate the association between factors of interest and breast cancer risk (12). Initially, early life factors were evaluated singly in logistic models containing terms for age and state. Multivariate models were also used to evaluate potential confounding by known breast cancer risk factors, including body mass index (kg/m²) at the reference date, religion (Jewish or non-Jewish), family history of breast cancer (breast cancer diagnosed in mother, sister, or daughter), age at first full-term pregnancy, parity (number of pregnancies lasting more than 6 months), age at menopause, and other available early life factors. Parental smoking was assessed in separate models with and without terms for birth weight and/or known risk factors. Our analyses provided little evidence of confounding; thus, the ORs reported here are generally adjusted only for age and state. In general, factors were treated as categorical variables; tests of linear trend were conducted, when possible, by treating factors as continuous variables. Potential interactions between risk factors were assessed using likelihood ratio tests. These tests were also used to assess the appropriateness of adding square terms to the models (e.g., birth weight + birth weight²). All comparisons were based on at least 1300 case and 1600 control women.

There was no clear association between birth weight and breast cancer risk (P for trend = 0.81; Table 1). The relationship between birth weight and risk was roughly consistent with a J-shaped pattern, with relative risks slightly elevated in both the lowest (OR = 1.10 for <2500 g) and the highest (OR = 1.18 for ≥4500 g) birth weight categories when compared with the normative birth weight of 3000–3499 g. However, neither effect was statistically significant, and there was no evidence of a nonlinear trend when birth weight was modeled as a quadratic function (P = 0.44). The weak J-shaped pattern was also observed in analyses using 2500–2999 g as the referent category and ≥4000 g as the highest birth weight category (data not shown). We found no evidence that the effect of birth weight was modified by age at menarche or adult body mass index (data not shown).

Paternal smoking or smoking by the mother (or by both the mother and father) during the mother’s pregnancy was unrelated to breast cancer risk in the adult daughter. We found no evidence of confounding between birth weight and parental smoking (data not shown).

Breast cancer risk increased with the number of years of the father’s education (P = 0.01); risk was 22% greater for women whose fathers had at least 12 years of education, relative to those whose fathers had 8–11 years of education. The results were comparable when additionally adjusted for the subject’s level of education.

The subject’s birth rank, adjusted for the mother’s age at the time of the subject’s birth, was inversely associated with breast cancer risk (P = 0.03); risk was lower only for women who were fifth or later in birth rank. Further analyses showed that risk was lower only for women whose older siblings were female (P for trend = 0.03). For women with at least three older sisters, relative to those without older sisters, risk was 26% lower. Risk was unrelated to the number of older brothers (data not shown). Risk was also unaffected by the gender ratio of the sibship or by membership in an all-female sibship (relative to having at least one brother). We found no association with the overall size of the sibship or the number of younger siblings (data not shown).

Risk appeared to increase with increasing mother’s age at the time of the subject’s birth, and the association was significant after adjustment for the mother’s number of previous pregnancies (P for trend = 0.04). Relative to women whose mothers were ages 25–29 years at the time of their birth, those whose mothers were at least 40 years of age had a 27% higher risk. We also evaluated whether maternal age modified the influence of key breast cancer risk factors (Table 2). In two groups, women whose first birth was at age 30 years or later and women who were first born, risk appeared to be elevated for those whose mothers were at least 35 years of age, relative to those whose mothers were less than 20 years of age at the time of the subject’s birth. However, the interactions between maternal age and the subject’s birth order (first or later), age at first birth, parity, and family history of breast cancer were not statistically significant (Table 2).

Birth weight may be associated with intrauterine estrogen levels; thus, an influence on adulthood risk of breast cancer is plausible (13). The results of some (6, 14) but not all (15) previous epidemiological studies indicated an increased risk associated with higher birth weight. In two previous studies, strong J-shaped patterns of risk were observed in younger women (5, 16). In one of these studies, the highest level of birth weight was associated with a >3-fold risk among women of age 30 years or less and a 70% increase for women of age 45 years or less, but no increase was observed among women ages 50–64 (5). The second study found a pronounced J-shaped relationship among women of age 37 years or younger (16). The weak J-shaped pattern noted in our study may reflect a weaker relationship between birth weight and breast cancer risk among older women (ages 50–79 years in our study). It is also possible, however, that our findings were attenuated by errors of self-report. Reasonably good validity of self-reported birth weight has been shown, even in studies including older women, although the effects were attenuated in self-reported data (6). In our study, slightly more than half of participants were unable to report their birth weight, which is consistent with the possibility of exposure misclassification.

The association between maternal smoking during pregnancy and lower birth weight may be mediated by reduced estrogen levels associated with smoking (17); however, one study showed only slightly lower estrogen levels in pregnant smokers (18). Our study and previous studies (4) found no evidence that parental smoking decreases the risk of breast cancer in adult daughters.

In our data, breast cancer risk increased with the father’s level of education, which may be a marker of a multigenerational lifestyle factor, such as diet. Some prior studies, however, found no association between father’s occupational level (15) or mother’s socioeconomic status (9) and breast cancer risk.

Maternal blood levels of total estrogens (19) and free or total estradiol (19, 20) are somewhat lower during second pregnancies than during first pregnancies. Similarly, cord blood levels of estradiol, estrone, and progesterones are lower for later-born children than for first-born children (21). Consistent with these findings, a combined analysis of three case-control studies noted lower risk for premenopausal women who were second born, compared with those who were first born (22). In our study, the lower risk associated with later birth rank was due to the protective effect of having older sisters. Study participants had been asked to provide the gender and birth dates of siblings and half-siblings, but were not asked to distinguish half-siblings who were born to the same mother. Consequently, our results may have been attenuated by the inclusion of half-sisters who shared the same father, although this was probably an uncommon event. Older sisters often serve as care-givers to their younger siblings. Although speculative, it is conceivable that the protective effect of older sisters reflects early exposure to infectious agents acquired by older sisters. In support of this possibility, limited evidence suggests that breast cancer risk may be increased by delayed primary exposure to EBV (10). We found no evidence that risk was influenced by membership in sibships that were mostly or entirely female; thus, our findings do not support a role for maternal hormones that might influence offspring gender ratio.

Our results are similar to those of a previous study noting a higher risk for daughters born to older mothers (23), although most found no association (4, 5, 24, 25, 26). We found no evidence of the J-shaped pattern of risk that has been observed in studies of very young women (15, 16). Serum levels of pregnancy estrogens (total estrogens, estradiol, total estriol, and human placental lactogen) do not appear to increase with maternal age (19); thus, the mechanism of possible influence is unclear.

Comparable with two previous studies (25, 26), our data indicate that the influence of maternal age does not differ for nulliparous and parous women. Our findings are also consistent with a previous study showing an increased risk for women whose first birth was delayed and who were born of older mothers (26), although this is not always observed (25). Consistent with previous studies (25, 26), we found that maternal age was not modified by a family history of breast cancer. Our results and those of a previous report (24) suggested higher risk among first-born children of older mothers, but the finding was not statistically significant in either study.

We acknowledge the helpful comments of Dr. Meir J. Stampfer, the dedication of personnel at each study center, and the women in our three states whose generosity made this research possible.