Maternal Pelvic Size Not Predictive of Daughter's Breast Cancer or Ovarian Cancer in a Large Swedish Cohort

Barker et al. recently reported that maternal pelvic size predicted breast and ovarian cancer in 4,102 Finnish women born between 1934 and 1944, and followed from 1971 to 2003 (1, 2). They investigated four standard measures of maternal pelvic size: intercristal distance (the maximal distance between the iliac crests), interspinous distance (distance between the anterior-superior iliac spines), the difference between the intercristal and interspinous distances, and external conjugate distance (distance from the front of the pubic bone to the fifth lumbar vertebrae). Two hundred and six of their cohort members developed primary breast cancer, with higher incidence among women whose mothers had a larger intercristal distance and a larger interspinous-intercristal difference. Thirty-nine developed primary ovarian cancer, with higher incidence among women whose mothers had a larger interspinous distance. Barker et al. hypothesize that pelvic size is a marker for the mother's hormone profile, and that this in utero exposure increases the daughter's risk.

The Uppsala Birth Cohort Multigenerational Study (UBCoS Multigen) has been described previously (3-5). Briefly, the cohort comprises all live births between 1915 and 1929 in the Uppsala Academic Hospital, Sweden. Archived obstetric records provided information about cohort members and their mothers at birth, and record linkage provided data from routine registers up to 2002. This included the Swedish Cancer Registry, which was established on January 1, 1960.

Of the 6,781 females live births, 751 died before 1960, 63 emigrated before 1960, and 202 were never traced. Of the remainder, 3,845 of 5,765 (66.7%) had maternal pelvic measurements, and these women comprise the starting population for our analyses. Pelvic measurements were more complete after 1924 (31.7-50.1% with data pre-1924 versus ≥93.4% after) and for primiparous women (72.1% with data versus 63.2% for multiparous). There was no evidence (P > 0.05) that missing pelvic measurements predicted breast or ovarian cancer in daughters.

We fitted Cox proportional hazards models, running separate models for breast and ovarian cancer. Follow-up started on January 1, 1960 and continued until the date of death, emigration, diagnosis with any primary cancer, or until December 31, 2002. The woman's age defined the time scale. To adjust for possible cohort or period effects, we divided birth years into three bands (1915-1919, 1920-1924, and 1925-1929) and included this as a categorical variable in all models. We used the same cutoffs as Barker et al. for categorizing pelvic measurements and we also present analyses using the continuous measurements.

We then assessed the effect of adjusting for each potential confounder listed in Table 1 individually, and of adjusting for all of them, simultaneously. We banded the continuous confounders into between five and eight categories of approximately equal size, and modeled these as categorical variables.

Barker et al. report that the effect of intercristal distance on breast cancer was greatest in multiparous mothers and infants born at ≥40 weeks, and that the effect of interspinous distance on ovarian cancer was greatest in mothers who had menarche before age 14 and were <160 cm tall. We conducted sensitivity analyses restricting our analyses to these subgroups, except for mother's height, which was not recorded for our cohort. To assess whether the effect of maternal pelvic size was modified by menopausal status, we also conducted separate analyses for ages <50 and ≥50 years, using these as proxies for premenopausal and postmenopausal ages.

The characteristics of the study population and their mothers are summarized in Table 1. Among our 3,845 study members, 273 developed primary breast cancer and 52 developed primary ovarian cancer by December 31, 2002 (total person-years at risk, 142,826.3).

Table 2 presents the hazard ratios for breast and ovarian cancer for each of the pelvic measures. In no case was there evidence of an association (P > 0.1); this remained true when entering the pelvic measurements as categorical variables or with quadratic terms. There was likewise no evidence at the 5% level of an association after adjusting for any potential confounder listed in Table 1, after stratifying our analyses between women ages <50 and those ages ≥50 years, or after restricting our analyses to the subgroups in which Barker et al. report the greatest effect.

The Uppsala Birth Cohort Multigenerational Study provides a unique opportunity to test the hypotheses proposed by Barker et al. (1, 2). Unlike their findings from Finland, our Swedish cohort provided no evidence that maternal pelvic size predicts daughters' breast or ovarian cancer. This is despite a close similarity in our methods, and a close similarity between our study populations in terms of pelvic sizes and cancer incidence. Our null findings were robust to adjustment for confounders and sensitivity analyses, including analyses restricted to the subgroups in which Barker et al. report the largest effect. The one subgroup analysis we did not have the data to replicate was restricting the analysis of interspinous distance and ovarian cancer to shorter mothers. However, we found no evidence (P = 0.17) for this effect in the whole population, whereas in the Finnish cohort the whole-population P value was 0.008 (1).

This nonreplication cannot be attributed to insufficient power. Our cohort is of a similar size to the Finnish cohort (3,854 versus 4,201 females) and, because of the longer follow-up, contains somewhat more cancer cases (273 versus 206 breast cancers, 52 versus 39 ovarian cancers). For example, Barker et al. report a hazard ratio for breast cancer of 1.23 per 1 cm increase in the intercristal-interspinous difference. With 273 cancers among 3,845 women, and a SD of 1.3, our cohort would have 99.4% power to detect this at the 5% significance level. This nonreplication likewise cannot be attributed to poor measurement of exposure or outcome, as both have previously shown positive findings in other studies. For example, larger maternal pelvic size does protect against stroke in our cohort (6), in a way which replicates findings from the Finnish cohort (7). Similarly, breast cancer in our cohort is predicted by birth size (3), in a way consistent with the existing literature (8).

In summary, this cohort provides no evidence that maternal pelvic size predicts daughters' incidence of breast or ovarian cancers. This therefore casts doubt on a relationship between these factors.

No potential conflicts of interest were disclosed.

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.

We thank Rawya Mohsen and Bitte Modin for their contribution to data preparation, to Richard Silverwood for help with statistical power calculations, and to Amal Khanolkar and Flaminia Chiesa for comments during the preparation of this manuscript.