Analytical sample

From the total ICMD sample (n = 11,755), for the analyses examining associations between the parity exposure and √MD measures, 10,988 (93.5%) women were eligible for inclusion. The included analytical sample are further described in Fig. 1. In analyses restricted to parous women, 9,773 women were included for the age at first birth analyses. For the breastfeeding analyses, 8,548 were eligible to be included in the ever/never breastfed analyses and among parous women who breastfed, 5,657 were eligible to be included in the lifetime breastfeeding duration analyses.

Fig. 1

Flow chart of the study sample with reasons for exclusion in the grey boxes and the subsequent analytical samples in the dashed boxes for each exposure of interest – parity, age at first birth, ever/never breastfed, and lifetime breastfeeding duration. * This study was excluded from the ever/never breastfed analyses as 100% of women in the US-USC Asian population group reported never breastfeeding. This proportion may be as a result of the random sample selected from the study during the initial consortium development which may not reflect the true proportion of women who breastfed in this specific population group. Given this uncertainty, it was decided to exclude this study from the ever/never breastfed analyses

Study participant characteristics

A summary of the reproductive exposures of the 40 country-ethnicity-specific population groups are described in Table 1 and the distribution of mean parity, age at first birth, and lifetime breastfeeding duration across the population groups are shown in Fig. 2. Among the total included study sample (n = 10,988), the mean age at mammography was 52.7 years (standard deviation (SD) = 8.2) and mean BMI was 27.0 kg/m2 (SD = 5.7). 90.1% (n = 9,895) of women were parous. The mean parity was 2.7 births (SD 1.8), ranging from 1.3 births (SD = 1.1) for Hong Kong Chinese women to 5.5 births (SD = 3.1) for Singapore-Malay women (Fig. 2). Overall, 13% (n = 1,286) of parous women had five births or more. Amongst parous women, the overall mean age at first birth was 24.3 years (SD = 5.1). Among the population groups, women in India had the lowest mean age at first birth (15.4 years, SD = 3.4) and the highest was among Chinese women in Hong Kong (29.0 years, SD = 4.7). Among the parous study sample, 71.8% (n = 7,143) breastfed and the median lifetime breastfeeding duration for women that breastfed was 10.0 months (interquartile range (IQR) = 28.0 months). Within this study sample, across the population groups, the group with the lowest median lifetime breastfeeding duration were white women in the UK (0.0 months, IQR = 6.0) and women in India had the highest median lifetime breastfeeding duration (72.0 months, IQR = 36.0).

Table 1 Characteristics of study participants from 40 country-ethnicity specific population groups from the 27 ICMD contributing individual studies included in this analysis.Fig. 2

Graph showing mean parity (blue), mean age at first birth (red), and mean lifetime breastfeeding duration (green) for the 40 country-ethnicity specific population groups included in this study

Associations between parity and MD measures within ICMD

Population-specific meta-analyses showed inverse associations between increasing parity and √PMD (β =  − 0.05 per birth, 95% CI =  − 0.07, − 0.03) and √DA (β =  − 0.08 per birth, 95% CI: − 0.12, − 0.05) (Figs. 3(a) and (b)). Findings were mostly consistent across population groups for √PMD (I2 = 2.2%, p = 0.430). Moderate heterogeneity was found for √DA (I2 = 35.9%, p = 0.014). Pooled categorical analyses were consistent with findings from the population-specific meta-analyses and demonstrated the linear inverse association between increasing parity and MD. This decreasing pattern was evident with √PMD and √DA among women with up to at least 9 births, though analysis of categories ≥ 8 births were limited by sample size (Fig. 4 (a) and (b)). For both analytic approaches, no clear pattern of association was observed between increasing parity and √NDA (Additional File 1: Fig. S2(a-b)).

Fig. 3

a–b Forest plots showing the associations between parity and √MD measures (a: √PMD and b: √DA) for all women (parous and nulliparous) across the 40 country-ethnicity specific population groups in the ICMD sample. Analyses were adjusted for age at mammogram, BMI, menopausal status, use of hormone replacement therapy, age at menarche, age at first birth, and image parameters

Fig. 4

a–b Pooled analysis plots showing the associations between parity (categorical) and √MD measures (a: √PMD and b: √DA) for all women (parous and nulliparous) in the pooled ICMD sample. Analyses were adjusted for age at mammogram, BMI, menopausal status, use of hormone replacement therapy, age at menarche, age at first birth, and image parameters

Compared to the main population-specific meta-analysis results, stratified meta-analysis findings did not differ by menopausal status (Additional File 1: Fig. S3(a-b), data not shown for √DA or √NDA). However, in the stratified meta-analyses among pre-menopausal women, moderate heterogeneity was observed across population groups for √PMD (I2 = 48.3%, p = 0.001), √DA (I2 = 38.8%, p = 0.010) and √NDA (I2 = 42.5%, p = 0.004) though the sample of pre-menopausal women was smaller (Additional File 1: Fig. S3a, data not shown for √DA or √NDA). Compared to the primary pooled analysis among all women, the patterns observed in the stratified pooled analysis among pre-menopausal women appeared slightly weaker for √PMD and were consistent for √DA and √NDA, respectively (Additional File 1: Fig. S4(a) and data not shown for √DA or √NDA). Findings from the stratified pooled analyses for post-menopausal women were consistent for √PMD and √DA to the findings observed in the main pooled analysis (Additional File 1: Fig. S4(b) and data not shown for √DA). Analysis of the higher parity categories were limited for the stratified pooled analyses due to the smaller sample sizes of pre- or post-menopausal women in those categories. Compared to the main population-specific meta-analysis findings, sensitivity analyses did not show any differences between increasing parity and the √MD measures when nulliparous women were excluded from the analysis (see Additional File 1: Fig. S5(a-c)).

Associations between age at first birth and MD measures within ICMD

Among parous women, positive associations were observed between √PMD (β = 0.06, 95% CI = 0.03, 0.10) and √DA (β = 0.06, 95% CI = 0.02, 0.10) respectively per five-year increase in age at first birth (Fig. 5(a) and (b)). An inverse association was observed for √NDA per five-year increase in age at first birth (β =  − 0.06, 95% CI =  − 0.11, − 0.01) (Additional File 1 Fig. S6(a)). Heterogeneity was moderate across population groups for √PMD (I2 = 19.8%, p = 0.142) and √NDA (I2 = 22.2%, p = 0.112), and low for √DA (I2 = 3.9%, p = 0.401). In pooled categorical analyses, compared to the reference category (22–23 years), a consistently increasing pattern beginning at the youngest age categories (12–21 years) and slowing down at older age categories (≥ 24 years) was observed for √PMD (Fig. 6 (a)). No consistent pattern was observed with √DA (Fig. 6 (b)). For √NDA, a curvature pattern was observed, increasing initially among women who first gave birth between age 12–15 years and decreasing among those who gave birth for the first time aged 16–21 years, compared to the reference category (22–23 years) (Additional File 1: Fig. S6(b).

Fig. 5

a–b Forest plot showing the associations between age at first birth and √MD measures (a: √PMD and b: √DA) among parous women across 39 country-ethnicity specific population groups in the ICMD sample. Analyses were adjusted for age at mammogram, BMI, menopausal status, use of hormone replacement therapy, age at menarche, parity, and image parameters

Fig. 6

a–b Pooled analysis plots showing the associations between age at first birth (categorical) and √MD measures (a: √PMD and b: √DA) among parous women in the pooled ICMD sample. Analyses were adjusted for age at mammogram, BMI, menopausal status, use of hormone replacement therapy, age at menarche, parity, and image parameters

Population-specific meta-analyses stratified by menopausal status found consistent associations only amongst those who were post-menopausal women at MD assessment at the time of mammography compared to the primary meta-analysis findings (Additional File 1: Fig. S7(a-b)). For stratified pooled analyses, no consistent trends were observed for √PMD or √DA among pre-menopausal women, whilst similar patterns were seen for both √MD measures among post-menopausal women to those observed in the main pooled analysis among all parous women in the ICMD sample (Additional File 1: Fig. S8(a-b) and data not shown for √DA). Further, the patterns observed among pre- and post-menopausal women for √NDA in the stratified pooled analyses were consistent with those observed in the primary pooled analysis (Data not shown).

Associations between breastfeeding and MD measures within ICMD

Among parous women, no strong evidence of an association was found between ever/never breastfed and √PMD (β =  − 0.04, 95% CI =  − 0.11, 0.04), √DA (β =  − 0.04, 95% CI =  − 0.14, 0.06), or √NDA (β =  − 0.00, 95% CI =  − 0.12, 0.12) (Fig. 7 (a) and (b) and Additional File 1: Fig. S9). No heterogeneity was observed across population groups for √PMD (I2 = 0.0%, p = 0.576) or √DA (I2 = 1.2%, p = 0.447). Whilst heterogeneity between population groups was low for √NDA (I2 = 25.5%, p = 0.103). Findings from the meta-analyses stratified by menopausal status were similar to the findings from the primary population-specific meta-analysis (Additional File 1: Fig. S10(a-b)).

Fig. 7

a–b Forest plot showing the associations between ever/never breastfed and √MD measures (a: √PMD and b: √DA) among parous women across 30 country-ethnicity specific population groups in the ICMD sample. Analyses were adjusted for age at mammogram, BMI, menopausal status, use of hormone replacement therapy, age at menarche, parity, age at first birth, and image parameters

For lifetime breastfeeding duration, associations between each six-month increase in lifetime breastfeeding duration and √PMD were (β = 0.01, 95% CI =  − 0.00, 0.02) and √DA (β = 0.01, 95% CI = 0.00, 0.03) (Fig. 8 (a) and (b)). No association was observed with √NDA per six-month increase in lifetime breastfeeding duration (Additional File 1: Fig. S11(a)). Low heterogeneity was found across population groups for both √MD measures, √PMD (I2 = 16.2%, p = 0.224) and √DA (I2 = 18.7%, p = 0.191). Results from the pooled categorical analyses showed no consistent pattern between increasing lifetime breastfeeding duration in categories and the √MD measures, compared to the reference category (≤ 1 month) (Figs. 9 (a) and (b) and Additional File 1: Fig. S11(b)).

Fig. 8

a–b Forest plot showing the associations between lifetime breastfeeding duration and √MD measures (a: √PMD and b: √DA) among parous women that breastfed across 28 country-ethnicity specific population groups in the ICMD sample. Analyses were adjusted for age at mammogram, BMI, menopausal status, use of hormone replacement therapy, age at menarche, parity, age at first birth, and image parameters

Fig. 9

a–b Pooled analysis plots showing the associations between lifetime breastfeeding duration (categorical) and √MD measures (a: √PMD and b: √DA) among parous women that breastfed in the pooled ICMD sample. Analyses were adjusted for age at mammogram, BMI, menopausal status, use of hormone replacement therapy, age at menarche, parity, age at first birth, and image parameters

Compared to findings from the main meta-analysis, stratified meta-analyses demonstrated similar associations with the √MD measures per six-month increase in lifetime breastfeeding duration among pre- and post-menopausal women (Additional File 1: Fig. S12(a-b) and data not shown for √DA and √NDA). For pooled categorical analyses stratified by menopausal status, the patterns observed were consistent with those observed in the main pooled analysis for lifetime breastfeeding duration and the √MD measures (Additional File 1: Fig. S13(a-b) and data not shown for √DA and √NDA).

Compared to the findings from the primary meta-analysis examining the association between a six-month increase in lifetime breastfeeding duration and the √MD measures, similar associations were observed in the sensitivity analysis when shorter (< 1 month) and longer (> 200 months) lifetime breastfeeding durations were excluded (Additional File 1: Fig. S14(a-c)). Further, pooled analysis findings did not differ between increasing lifetime breastfeeding duration categories and the √MD measures when parous women that did not breastfeed were included as the reference category, compared to the main pooled analyses which only included parous women who breastfed in the ICMD sample (data not shown). To further investigate the breastfeeding associations, we conducted additional analyses to investigate associations between an approximate per birth breastfeeding duration and the √MD measures. Meta-analysis findings examining the associations between a six-month increase in breastfeeding duration per birth (derived from dividing lifetime breastfeeding duration by parity) and the √MD measures were: √PMD (β = 0.03, 95% CI =  − 0.01, 0.06), √DA (β = 0.05, 95% CI = 0.01, 0.09), and √NDA (β = 0.02, 95% CI =  − 0.04, 0.07) (Additional File 1: Fig. S15(a-c)). These findings showed similar associations for √PMD and √NDA, and stronger for √DA compared to the primary meta-analysis examining the associations between the lifetime breastfeeding duration exposure and the √MD measures. While the patterns observed in the pooled categorical analyses for breastfeeding duration per birth and the √MD measures were consistent with patterns observed in the pooled analyses examining lifetime breastfeeding duration (Additional File 1: Fig. S16(a-c)).

Sensitivity analysis

Findings from the sensitivity analyses excluding the five studies that collected covariate data at least two years before or after time of mammography did not differ to the main population-specific meta-analyses examining the associations between each of the reproductive factors (parity, age at first birth, ever/never breastfed, or lifetime breastfeeding duration) and the √MD measures. In these sensitivity analyses, associations between increasing parity and √PMD were (β =  − 0.05 per birth, 95% CI =  − 0.07, − 0.02), √DA (β =  − 0.08 per birth, 95% CI: − 0.12, − 0.04) and √NDA (β = 0.00 per birth, 95% CI: − 0.04, 0.04) (data not shown). Findings from the sensitivity analysis examining the associations between a five year increase in age at first birth and the √MD measures were: √PMD (β = 0.06, 95% CI = 0.02, 0.10), √DA (β = 0.07, 95% CI = 0.01, 0.12) and √NDA (β =  − 0.05, 95% CI =  − 0.11, 0.01) (data not shown). For the breastfeeding exposures, the associations between ever/never breastfed and √PMD were (β =  − 0.07, 95% CI =  − 0.15, 0.04), √DA (β =  − 0.07, 95% CI =  − 0.18, 0.04), and √NDA (β = 0.03, 95% CI =  − 0.08, 0.15). Whilst findings from the sensitivity analysis examining the association between a six-month increase in lifetime breastfeeding duration and the √MD measures were: √PMD (β = 0.01, 95% CI =  − 0.00, 0.02), √DA (β = 0.01, 95% CI = 0.00, 0.03), and √NDA (β = 0.01, 95% CI =  − 0.00, 0.02) (data not shown).