The few studies on hormone variations by maternal and pregnancy characteristics have focused on maternal measures in mid- or late pregnancy, primarily estrogens. Because the timing of carcinogenic events is unknown, characterising hormones over the entire pregnancy may provide additional insights into the biological mechanisms for maternal and perinatal breast cancer risk factors.

The study sample, described previously (Potischman et al, 2005), includes 109 Hispanic, 56 African American and 255 Caucasian women 18–42 years of age (median 29.6), who delivered a live, singleton birth in an uncomplicated pregnancy. A total of 38% of the women were nulliparous.

In analyses mutually adjusted for the pregnancy factors and including gestational age, androgen and oestrogen concentrations were lower in older mothers (Table 2); the greatest incremental difference was between those <30 and 30+ years (e.g., P<0.001 for androstenedione; other data not shown). In contrast, maternal age was positively associated with progesterone. The androgens, estrogens and prolactin were higher in the 160 nulliparous than in the 260 parous women. Maternal BMI (and maternal weight; data not shown) was positively associated with the androgens and inversely associated with progesterone and SHBG, and maternal height was inversely associated with the estrogens and SHBG though not with the other hormones. Current smokers had lower progesterone and prolactin concentrations, whereas alcohol users had lower oestradiol only.

The positive association of maternal BMI and androstenedione was limited to mothers of female fetuses (P=0.01); there were no other interactions by offspring sex. Differences in hormones by race/ethnicity were reported in an earlier paper from this study (Potischman et al, 2005). The associations between the pregnancy factors and hormones were similar with adjustment for race/ethnicity and there were no meaningful interactions demonstrated (data not shown).

Birth weight, an established prenatal breast cancer risk factor (Michels and Xue, 2006), is hypothesised to be mediated by foetal oestrogen exposure (Michels et al, 1996). Third trimester maternal oestriol is elevated in high birth weight pregnancies (Kaijser et al, 2000; Mucci et al, 2003; Peck et al, 2003; Troisi et al, 2003; Nagata et al, 2006), but data for second trimester estrogens are conflicting (Kaijser et al, 2000; Wuu et al, 2002). Our data show no association of birth weight with first trimester estrogens seeming to suggest that if oestrogen explains the association of birth weight and breast cancer risk, the critical exposure window may be later in pregnancy. Yet associations of birth weight with cord estrogens are unclear (Simmons et al, 1994; Shibata et al, 2002; Troisi et al, 2003; Nagata et al, 2006), and breast cancer risk in women prenatally exposed to diethylstilbestrol does not differ by trimester of first exposure (Palmer et al, 2006). Other prenatal factors could act differently. For example, a more pronounced breast cancer risk is suggested for women who were in utero during the Dutch famine in early compared with later gestation (Painter et al, 2006).

Long-term maternal breast cancer risk is reduced with a full-term pregnancy, particularly when it occurs at a young age (National Cancer Institute, 2003). Epidemiologic data indicate protection increases with gestational length (Vatten et al, 2002), suggesting that the important hormonal events must occur late in pregnancy. However, protection from a full-term pregnancy would also be consistent with cumulative dose effects of specific hormones throughout the pregnancy, or early hormonal events which require the subsequent action of others to induce the relevant molecular changes. Descriptive data characterising the hormonal profile of the entire pregnancy are lacking with regard to parity. The elevated first trimester oestradiol we observed in first compared with subsequent pregnancies has also been shown for the second (Wuu et al, 2002; Arslan et al, 2006) and early third trimesters (Wuu et al, 2002), but perhaps not at delivery (Troisi et al, 2003). The elevated prolactin concentrations in our data for first pregnancies have also been noted in the second trimester (Xu et al, 2003; Arslan et al, 2006), but not later in the pregnancy (Xu et al, 2003), and the first trimester androgen elevations have not been shown at delivery (Troisi et al, 2003). Our data also indicate that oestrogen and androgen concentrations are higher in pregnancies that occur in younger than in older women. There is limited and conflicting information for maternal estrogens by age at any point during the pregnancy (Panagiotopoulou et al, 1990; Kaijser et al, 2002; Troisi et al, 2003), whereas the androgen findings are consistent with studies using measurements made later in the pregnancy (Carlsen et al, 2003; Troisi et al, 2003).

The explanation for the protective effect of a first pregnancy and the inverse relationship of this protection with the age at which the pregnancy occurs remains speculative. The hypothesis most commonly invoked is that irreversible molecular changes (driven by hormonal exposures) to the breast that prevent tumour initiation (Medina, 2005) must occur early in life to precede the initial stage of breast carcinogenesis, and that most of these changes occur with the first pregnancy. An alternative explanation, however, is that the operative hormonal influences may be different in first pregnancies, particularly those occurring at young ages. Our finding of higher circulating estrogens, androgens and prolactin in both a first pregnancy and, independently, in those occurring at younger ages could be consistent with this latter explanation. Subsequently reduced concentrations of these hormones in the pregnant and non-pregnant state may also be involved.

The study subjects were participants in a health care plan that provides prenatal care. The MGH study's participation rate is high and thus unlikely to represent an unusual population of women who seek medical care early in pregnancy. Whereas the reproducibility data were high for oestradiol and oestrone, there was substantial laboratory imprecision for other analytes that based on the quality control data suggested sporadic, random measurement errors. Finally, it is unknown how well a single first trimester hormone measurement represents early pregnancy exposure.