We evaluated the combined influence of multiple sex and growth hormones on the risk of postmenopausal breast cancer on the basis of several scores. The scores ranked women by the number of hormones above the age- and batch-adjusted geometric mean and weighted the hormone values by their individual associations with breast cancer risk or proliferation of MCF-7 breast cancer cells. In each instance, there was a positive linear relationship between the score and risk of breast cancer overall and ER-positive disease. The risks in the highest categories were about double those in which only one hormone at a time was considered. When considering separate scores for estrogens and androgens, the estrogens had a slightly stronger association.
Very few studies have evaluated the relationship between multiple hormones simultaneously and breast cancer risk, and those that have considered interactions between two hormones generally have not revealed significant multiplicative interactions [2
]. However, additive models were not considered. Trichopoulos et al.
] measured estrone, estradiol, androstenedione, DHEAS, testosterone and IGF-1 in 29 prospective cases and 58 controls. The authors constructed a score summing the number of hormones with levels above the age-adjusted mean and observed that women with four to six versus one to three hormones above the mean did not have a different risk of breast cancer (RR = 1.13, 95% CI = 0.43 to 3.00, P
trend = 0.53). Women with no hormones above the mean had a substantially lower risk of breast cancer compared to all other women (RR = 0.11, 95% CI = 0.01 to 0.90), suggesting that sex hormones may act as permissive factors for breast cancer growth. In contrast, our results do not support this hypothesis, as we observed a strong linear trend as the number of hormones above the geometric mean increased, which was more pronounced for ER-positive disease, although we also noted that women with no hormones above the mean versus any had a lower risk of breast cancer (RR = 0.55). Our results are similar regardless of how the hormone score was defined, in part because of the high correlation between the scores. The difference in results between the two studies likely stems from the greater statistical power in our study, since when we created a score using the same hormones as Trichopoulos et al.
], we observed a linear association (RR, one-unit increase in number of hormones above the mean = 1.18, 95% CI = 1.07 to 1.31, P
trend = 0.001).
Estrogens and prolactin likely influence the risk of breast cancer by inducing cell proliferation and tumor growth through the ER and prolactin receptor, respectively [32
]. On the other hand, androgens have been hypothesized to increase breast cancer risk either directly by increasing cellular growth and proliferation or indirectly via conversion to estrogens [34
]. In in vitro
studies, androgens have been shown to either increase or decrease cell proliferation, depending upon the model system used [34
]. Biological data have examined the combined effect of hormones. For example, in an Nb rat model, the simultaneous administration of testosterone and estradiol led to the development of invasive mammary carcinoma [35
], but neither hormone alone induced carcinogenesis. Furthermore, in MCF-7 cells, both DHEA [24
] and DHEAS [20
] can increase proliferation independently of aromatase activity or binding of the ER. Similarly, in a breast cancer cell line induced to endogenously produce high levels of prolactin, proliferation was magnified by the addition of estradiol [37
]. These biological studies suggest that hormones could act synergistically or through different cellular mechanisms (not only through the ER) to increase proliferation and ultimately breast carcinogenesis. Our results generally support this hypothesis in that the risk of breast cancer increased as the hormone scores increased.
We also observed that estrogens may have a slightly stronger influence on breast cancer risk than androgens. Our results are consistent with those of most prior studies [2
], which have shown that when estradiol and testosterone (correlation approximately 0.30 to 0.35) are placed in the same model, the association with estradiol was essentially unchanged, but the association with testosterone was attenuated. However, investigators who conducted a combined analysis of nine prospective studies comprising 663 cases (including the cases in the present analysis) reported that estradiol and testosterone have independent associations from the other [1
]. A complexity in the pooled analysis is that some studies used indirect assays to measure estradiol and testosterone, whereas others used direct assays, potentially leading to different correlations between these hormones across studies. The differences between studies may be due to sample size, different correlations between hormones examined or, in our study, consideration of multiple estrogen and androgens simultaneously. Additional research is needed to resolve the independent role of androgens and estrogens, as well as to determine the mechanism of action for androgens.
When considering the hormone score counting the number of hormones above the mean, the addition of IGF-1 and c-peptide increased the relative risk for the top category; however, this result was not replicated in the score weighted by individual hormone associations, possibly due to different reference groups. Researchers in two prior studies [7
] observed that the risk of breast cancer was highest among women with high levels of both IGF-1 and at least one sex hormone (for example, testosterone or estradiol). Also, proliferation of MCF-7 cells was two times higher among cells treated with both IGF-1 and estradiol than when treated with IGF-1 or estradiol alone [38
]. Although more studies are needed, it is possible that that IGF-1 and/or c-peptide, though not strongly associated with breast cancer risk individually, may be associated with a higher risk when considered in combination with other hormones.
This study has several strengths and limitations. First, case-control sets were assayed together, which reduces variability in hormone measures, and the assays had excellent coefficients of variation. Second, we had only one measure of each hormone to reflect long-term exposure; however, the intraclass correlations (ICCs) over two to three years were greater than 0.60 (except for prolactin, with an ICC of approximately 0.45) [39
], suggesting good reproducibility over time. Importantly, in the same reproducibility study, the ICCs of the three hormone scores in our study ranged from 0.72 to 0.85 over two to three years, and the ICCs of the individual hormones generally were similar for women with high versus low hormone scores at the baseline blood draw. Third, our study was relatively large. Although this increased the study's power to detect statistically significant associations, our sample size for subanalyses (for example, interactions based on PMH use) was limited and as such we were not able to precisely assess whether a smaller subset of hormones could capture most of the variation across all hormones or directly compare associations for ER-positive versus ER-negative tumors. Not all the assays were measured for some women, although this was a random loss. When women with one missing hormone were included in the analysis (using the median level in the population for the missing hormone), the results were similar (data not shown). Fourth, although the score using the PIm
was based on independent biological data, the score has some limitations. Since there was not a single study comparing all hormones, we drew data from multiple studies that used different metrics of proliferation and may have used different strains of MCF-7 cells. However, we standardized measures for each individual hormone by the comparable assay results for estradiol and averaged results across multiple studies. Also, other metrics, such as apoptosis or binding affinity to ER, would have been of interest but were not available for all hormones. Finally, some potential hormones of interest, such as melatonin and SHBG, either were not measured or had many missing values.