The Association of age at Menarche and Adult Height With Mammographic Density in the International Consortium of Mammographic Density

Sarah V. Ward; Anya Burton; Rulla M. Tamimi; Ana Pereira; Maria Luisa Garmendia; Marina Pollan; Norman Boyd; Isabel dos-Santos-Silva; Gertraud Maskarinec; Beatriz Perez-Gomez; Celine Vachon; Hui Miao; Martín Lajous; Ruy López-Ridaura; Kimberly Bertrand; Ava Kwong; Giske Ursin; Eunjung Lee; Huiyan Ma; Sarah Vinnicombe; Sue Moss; Steve Allen; Rose Ndumia; Sudhir Vinayak; Soo-Hwang Teo; Shivaani Mariapun; Beata Peplonska; Agnieszka Bukowska-Damska; Chisato Nagata; John Hopper; Graham Giles; Vahit Ozmen; Mustafa Erkin Aribal; Joachim Schüz; Carla H. Van Gils; Johanna O. P. Wanders; Reza Sirous; Mehri Sirous; John Hipwell; Jisun Kim; Jong Won Lee; Caroline Dickens; Mikael Hartman; Kee-Seng Chia; Christopher Scott; Anna M. Chiarelli; Linda Linton; Anath Arzee Flugelman; Dorria Salem; Rasha Kamal; Valerie McCormack; Jennifer Stone

Disclosures

Breast Cancer Res. 2022;24(49) 

In This Article

Methods

Study Design and Participants

We examined two markers of developmental growth, age at menarche and adult height, in relation to measures of MD in the ICMD. The ICMD methodology and contributing studies are discussed in detail elsewhere.[19] Briefly, the consortium pooled individual-level data from studies investigating MD and its putative determinants in breast cancer-free women, purposefully including studies from diverse countries and ethnic groups with different underlying breast cancer incidence rates. In total, 11,755 women were included from 27 studies in 22 countries, forming 40 location and ethnicity-specific 'population groups' (Figures 1, 2, 3, 4). Population groups included the broad ethnic groups of Black, East Asian, South Asian, Hawaiian, Mestizo, Middle Eastern and White women (see Table 1 for breakdown by country). In each population group, there were approximately 200 pre- and 200 post-menopausal women aged 35 years or older at the time of mammography. Mammograms were originally taken as part of organized screening (n = 13 studies), opportunistic or community-based screening (n = 8), mammography trials (n = 3) or for research (n = 3).

Figure 1.

Association of age at menarche (per year) with per cent density. Forest plot depicting results from a meta-analysis of the association of age at menarche (per year) with square-root per cent density of the breast, in studies from the International Consortium on Mammographic Density. Effect estimates for each separate population group are shown, as well as the combined effect estimate, from random effects model

Figure 2.

Association of age at menarche (per year) with dense area. Forest plot depicting results from a meta-analysis of the association of age at menarche (per year) with square-root dense area of the breast, in studies from the International Consortium on Mammographic Density. Effect estimates for each separate population group are shown, as well as the combined effect estimate, from random effects model

Figure 3.

Association of adult height (per 10 cm increment) with per cent density. Forest plot depicting results from a meta-analysis of the association of adult height (per 10 cm increment) with square-root per cent density of the breast, in studies from the International Consortium on Mammographic Density. Effect estimates for each separate population group are shown, as well as the combined effect estimate, from random effects model

Figure 4.

Association of adult height (per 10 cm increment) with dense area. Forest plot depicting results from a meta-analysis of the association of adult height (per 10 cm increment) with square-root dense area of the breast, in studies from the International Consortium on Mammographic Density. Effect estimates for each separate population group are shown, as well as the combined effect estimate, from random effects model

In the current study, further exclusions were made from the total 11,755 women. First, women from very small ethnic groups within each study were excluded (n = 129 across four countries), then women with no MD information due to poor image quality (n = 529), inconsistent age at first birth compared to age at menarche (n = 5) and 'nulliparous' women who had children (n = 6), implausible/missing BMI (n = 2), missing parity (n = 93) and missing age at menarche (n = 310). This resulted in a total of 10,681 women for analyses.

Exposures of Interest and Confounders/Modifiers

Individual-level data on sociodemographic and lifestyles factors were harmonized across all ICMD studies. Age at menarche data was self-reported in adulthood and collected as integers or categories, with the median values of each category assumed for continuous analyses. Categorical analyses were performed using previously used cut-points (< 12, 12 to < 13, 13 to < 14, 14 to < 15, and 15 years and older). Height was recorded in centimetres (cm) in the majority of studies and converted to centimetres for those recorded in feet and inches. Height was examined both as a continuous variable (per 10 cm increase) and as a categorical variable (< 155, 155 to < 160, 160 to < 165, ≥ 165 cm).

Weight was recorded in kilograms for most studies and converted from stones and pounds for all other studies. A measure of BMI was calculated for all participants as kg/m2. Information on the method of height and weight ascertainment was also collected, either as self-reported (n = 3370, 32%) or measured (n = 7061, 66%) and was not known in one study (n = 249, 2%).

Other variables included in the present analyses included age at mammogram, parity, age at first birth, and use of hormone therapy at the time of mammography. Study-specific definitions were used and then harmonized for menopausal status, as has been described in detail previously.[20]

Mammographic Density Measurement

In the ICMD, MD was measured centrally from digitized film mammograms and raw or processed digital images by one of three experienced assessors (authors VM, IdSS, NB) using the software program Cumulus.[21] For each woman, one mammographic image (cranio-caudal or medio-lateral oblique view) was measured. To assess the intra- and inter-assessor reliability, approximately 20% of the images were re-measured, providing repeated measures for a subset of women, as detailed previously.[19] This resulted in a total of 12,586 MD measurements for the 10,681 women in present analyses. The MD measures used in these analyses were DA (cm2) and PD (PD = 100 × DA/breast area), as they were considered the most aetiologically relevant. The measures from processed images were corrected to a raw image equivalent using published equations.[22]

Statistical Methods

Descriptive analyses were conducted in the broad ethnic groups of Black, East Asian, South Asian, Hawaiian, Mestizo, Middle Eastern and White women to summarise the data. Two analytical approaches were then taken to examine the associations between age at menarche and adult height with the outcomes PD and DA. Both of these approaches used the more specific population groups to take into account ethnic differences between participants. Both MD outcomes were first square-root-transformed to normalize residuals. PD and DA are both area measures, so if they are considered as squares, this transformation implies that regression beta-coefficients can be interpreted as the effect on the length of the side of a square, e.g. if DA = 25 cm, √DA = 5 cm (a square of 5 × 5 cm), and beta = + 0.1 cm, then the length of the DA square increases from 5.0 to 5.1 cm, and the corresponding DA increases from 25 to 5.12 = 26.0 cm.[20]

In the first analytical approach, population-specific associations were examined and their effect estimates combined using meta-analytic approaches with a random effects model. Forest plots were used to display population-specific effect estimates. Second, individual-level pooled analyses were performed using multi-level models with density measures clustered for an individual, who was clustered within their population group. Individual-level clustering was used to account for women with repeated measures, and population group clustering to account for differences in ethnicity between groups.

In both approaches, all models were adjusted for age at mammogram (cubed due to best fit), menopausal status, use of hormone therapy, mammogram view, calibration method, mammogram reader, parity and BMI (quadratic or cubed terms depending on best fit). To evaluate the possible independent effects of each exposure of interest, models for age at menarche were additionally adjusted for height and models for height were additionally adjusted for age at menarche.

Subgroup analyses by menopausal status, parity, BMI category and anthropometric ascertainment method were also performed using the multi-level pooled models. An additional adjustment for age at first birth was included for parous subsets.

Sensitivity Analyses

We also performed sensitivity analyses adjusting for a population-specific weight-for-height index, instead of BMI, as BMI and height are inversely correlated. Whilst BMI aims to be a measure of weight independent of height, it is a simplified index and is not completely independent of height.[23] The weight–height relationship has been shown to vary according to age and gender, such that height is inversely associated with BMI in (white) adults, but the magnitude of the association is larger for women and increases with age.[24] The weight–height relationship may differ in an international study where body sizes and shapes differ, and it is therefore important to examine MD associations with height independent of BMI or body fatness. Additional analyses were conducted to take these potential issues into account, using a weight-for-height index that was defined as the ratio of an individual's weight to their population-specific expected weight. The expected weight was generated using a population group-specific relationship of weight as a function of k 1 × height k 2 where both k 1 and k 2 were optimized separately for each population group. The median value of k 2 was 1.30 (inter-quartile range [IQR] 1.19–1.35; Additional file 1: Table S1), which is lower than the power of 2 which is used in BMI (i.e. weight/height2).

processing....