High BMI in childhood was strongly inversely associated with developing mixed/dense breasts and marginally with breast cancer after age 50. MD may explain the inverse association between childhood BMI and breast cancer risk, at least in part. Tallness in childhood was not significantly associated with MD, but it was marginally positively associated with breast cancer risk. Birth weight was not significantly related to MD or breast cancer risk.
Inverse associations between BMI in childhood and MD in this cohort agree with existing evidence based primarily on Caucasian women in European and American populations [17, 18, 22–24], despite different definitions of MD and body size. The only study similar to ours with measured body size (ages 2 to 15 years) in 1,298 British women found an odds of a higher Wolf grade of 0.56 (0.49 to 0.64) per 2.8 kg/m2 in BMI at age 15 ; this is remarkably similar to ours of 0.56 (0.53 to 0.58) per z-score at age 13 years. A study of 628 Scottish women detected an inverse association between having a high-risk mammogram (≥ 25% dense) and BMI at age 18 . Mammographic percentage density (MPD) in 1,893 American women was linked to self-reported weight and adiposity at 7, 12, and 18 years, but a significant inverse association was detected only at age 12 . A significant inverse relation between MPD and self-reported weight before menarche was detected in 3,547 Spanish women . Finally, self-reported BMI at age 8–10 years was significantly inversely associated with percentage and absolute breast density volume in 174 young women aged 25 to 30 years .
Our findings disagree with two studies failing to detect association between MD and self-reported weight at age 10 in 201 US Chinese immigrants  and self-reported body size before and after menarche in 1,531 Mexican women . Overall evidence suggests possible relevance of race/ethnicity in the association of childhood body size and MD, with consistent and robust inverse associations observed in Caucasian women of European and American descent, and none in Asian or Hispanic women.
A weak protective effect of BMI at ages 7 through 13 on breast cancer risk in this cohort is confirmatory [8, 11, 13–15]. However, our finding that the inverse association between childhood BMI and breast cancer diminishes after adjustment for MD conflicts with Harris et al. , where this association was robust to adjustment for MD. Various differences between two studies preclude direct comparisons and possibly explain the conflicting results. Where we benefited from measured childhood anthropometrics and prospective cohort design, Harris et al., in a nested case–control study, retrospectively collected body fatness of women in Nurses’ Health Study at ages 5, 10, and 20 years, by using nine-level figure drawing. Conversely, although we adjusted for birth cohort and age in full model, lacking information on other breast cancer risk factors, Harris et al. matched cases and controls on age, menopausal status, postmenopausal hormone use, and race/ethnicity, and furthermore adjusted for age at menarche, parity/age at first birth, history of breast cancer, and alcohol use, but did not adjust for birth cohort.
Furthermore, breast cancer and MD definitions differ between the studies. Whereas we obtained mammograms and breast cancer information objectively via registries, without possibility for population selection by refusal to participate/release information, Harris et al. relied on self-reports confirmed by medical records in women who gave permission to obtain medical records and mammograms. Additionally, our study was conducted predominantly on postmenopausal women, whereas Harris et al. included a younger population of women at screening with a shorter follow-up (mean time between screening and breast cancer diagnoses of 4.7 years versus 8.6 years in our study), resulting in predominantly premenopausal breast cancer.
Finally, whereas we excluded 98 breast cancer cases diagnosed at screening from main analyses, because of lack of MD, Harris et al. allowed these in analyses . In a sensitivity analysis including 98 women with additional 75 confirmed cases of breast cancer, we found similar associations of BMI with breast cancer risk as seen in main analyses (not shown).
The lack of an association between height at ages 7 to 13 and MD in our study concurs with two studies with measured heights at ages 2 to 15  and at 18 , and a study with self-reported height before menarche . A single study detected significant positive associations between self-reported height at ages 7, 12, and 18 with PMD . The positive and significant associations between childhood height and breast cancer risk in this cohort corroborate current evidence [6, 8, 11, 12], but none of the previous studies tested whether this association could be explained by MD. Likewise, our finding of a borderline significant interaction (P value = 0.08), indicating that the positive association between height and breast cancer may be limited to women with mixed/dense breasts (Table 4), calls for replication.
Birth weight was not associated with MD in this cohort, consistent with three [16–18] and in contrast to three studies showing significant positive associations [19–21]. Cerhan et al. found significant positive associations of birth weight with MPD only in the postmenopausal group of the 1,893 US women , whereas Tamimi et al.  presented data on 893 Swedish postmenopausal women only. Pearce et al.  detected significant and positive associations in a mixed group of 199 pre- and postmenopausal British women. The differences between our and these studies [19–21] may be due to adjustment for additional covariates, such as BMI/weight at mammography, HRT use, menopausal status, parity, age at first pregnancy, alcohol consumption, and so on. Pearce et al. [17–20] showed that adjustment for the most complete set of confounders of all mentioned studies resulted in higher and statistically significant effect estimates of an association between birth weight and MD, as compared with a crude estimate . However, studies failing to detect an association between birth weight and MD did not observe differences between crude and adjusted models [17, 18]. Additional adjustment for BMI at age 13 in our analyses, next to birth cohort and age at screening, changed our OR from 0.98 (0.90 to 1.07) to 1.11 (1.02 to 1.22), in agreement with Pearce et al. . However, we chose not to adjust for BMI at age 13 in main analyses, as childhood BMI may be an intermediate variable on the causal pathway between birth weight and MD. In any case, a critical assessment of relevant covariate adjustment is necessary when comparing estimates of association between birth weight and MD.
We found no association between birth weight and breast cancer, in contrast to the vast literature [4, 6–10]. Also in contrast to our findings, an earlier Danish study by Ahlgren et al. based on the same data source that included 106,504 women, but without information on MD, detected a significant, positive association between birth weight and breast cancer . However, this study  included younger women and had a 9-year shorter follow-up (until 2000) than ours, resulting in different age distribution of breast cancer cases. In our study, based on screened women older than 50 years, 69.9% and 92.2% of the breast cancer cases were older than age 60 and 55, respectively, whereas Ahlgren et al. had 4.1% and 18.0% of cases older than age 60 and 55, respectively. Nonetheless, the age-specific associations agree rather well: for breast cancers older than age 60, Ahlgren et al.  found a relative risk (RR per kg birth weight) of 0.77 (0.56 to 1.07), whereas we detected HRs of 0.82 (0.66 to 1.02); for breast cancer cases aged 50 to 54 and 55 to 59 years, Ahlgren et al. reports RR of 1.08 (0.92 to 1.25) and 1.09 (0.91 to 1.32), respectively, where we, for breast cancer diagnosed between 50 and 59 years, found OR of 1.01 (0.75 to 1.36).
Thus, these two sets of partially overlapping Danish data point at a lack of an association between birth weight and breast cancer from ages 50 to 59 years and a slight indication of a negative association at ages 60 and older. A careful look at the existing literature also supports this notion. Reviews by Ruder et al.  and Michels and Xue  suggest that evidence of a birth-weight effect is mixed and strongest for premenopausal breast cancer . Xu et al.  showed that the OR from meta-analyses based on studies of premenopausal breast cancer is 1.37 (0.98 to 1.92), whereas postmenopausal is 1.13 (0.85 to 1.51). Indeed, studies with data on both pre- and postmenopausal breast cancer generally find associations with birth weight only for the first group. Oberg et al.  reports significant and positive associations between birth weight and breast cancer diagnosed before age 50, but inverse nonsignificant associations for cancers after age 50 . Similarly, three studies [36–38] found positive associations between birth weight and premenopausal, and inverse [36, 37] or neutral  associations with postmenopausal breast cancer.
The mechanisms behind our finding that MD may be a mediator explaining the inverse association between childhood body fatness and breast cancer risk are not well understood. A pathway suggesting direct influence of childhood body fatness on the development of mammary tissue during adolescence  is likely, and supported by strong inverse associations with MD observed in current and previous studies [17, 18, 22–24]. One hypothesis suggests the relevance of sex hormones, higher in girls with more body fat, which are associated with earlier differentiation of breast tissue, resulting in cells less susceptible to malignant transformations . Another theory involves adolescent growth, as childhood body fatness is associated with lower levels of insulin-like growth factor 1 [41, 42] and slower adolescent growth, a possible pathway to reduced breast cancer risk [5, 6]. In any case, our findings concur with the increasing evidence that the early life exposures and years before first pregnancy, when the mammary glands differentiate and the terminal structure of mammary tissue is determined, are critical in establishing breast cancer risk [5, 6].
The current study benefited from a large cohort of women with prospectively collected data on anthropometric childhood factors, MD, and breast cancer, with minimal possibility of recall, information, or selection bias. We detected a strong effect of birth cohort on MD, finding that younger cohorts of women (born in 1945 to 49) had significantly higher MD than women from the oldest cohorts (1930 to 1934), in agreement with Hellman et al. . Our study expands on evidence provided by Harris et al.  about the influence of MD on the association between body size in childhood and breast cancer risk, adding novel results on birth weight and height.
Limitations of this study include the lack of adjustment for other relevant breast cancer covariates at the time of screening, including menopausal status, age at menarche, age at first giving birth, parity, hormone replacement therapy (HRT) use, socioeconomic status, education, physical activity, alcohol use, and others, and possible bias in our estimates due to confounding. However, studies with covariate information available suggest that full adjustment for these enhanced crude associations between birth weight and MD, making them statistically significant [21, 23], whereas it did not affect associations of BMI to MD [23, 24, 27] or of height to MD . Furthermore, we did not have information on BMI at the time of screening, which Lope et al.  adjusted for, showing no change in estimates of an association between prepubertal weight and MD, as compared with crude estimates. However, adult BMI could be an intermediate variable on the causal pathway between childhood BMI and MD, as pointed by Harris et al., who therefore did not to adjust for it in their analyses . Finally, Dorgan et al. showed that association of BMI at age 8 to 10 with two different measures of MD was robust to adjustment for adult BMI: effect estimates for percentage dense breast volume were attenuated by a half, but remained inverse and statistically significant, whereas estimates for absolute dense breast volume remained virtually unchanged . Thus, evidence from literature suggests limited possibility of bias in current study due to confounding. Furthermore, adjustment for adult body size for variables, which are on the causal pathway between birth weight/childhood body size and MD/breast cancer risk, is arguably inappropriate and may lead to an artifactual statistical effect .
Another limitation is the possibility of BMI tracking, implying that the findings of inverse associations between childhood BMI and MD in adult life would be expected if BMI were tracked through life. As correlations between child and adult BMI strengthen with age, if the observed associations were due to BMI tracking, we would expect the associations of BMI and MD to be much stronger at 13 years of age versus 7 years of age; however, this is not the case here (Table 3). Furthermore, if BMI tracking were to account for the observed associations, then adult BMI should have a stronger association with MD than childhood BMI. Again, this seems not to be the case here, as a related Danish cohort study on adult anthropometry (without data on childhood anthropometry) and MD in 5,937 women reported an inverse association between BMI at ages 50 to 65 and MD. Per SD increase in adult BMI, the odds of having mixed/dense breasts were 0.51 (0.48 to 0.54) (unpublished data); this estimate is similar to the estimates of associations of childhood BMI and MD observed in our study (Table 3). Although unlikely, even if the observed associations between BMI and MD are largely due to tracking, it still leaves open the possibility that the causal processes creating the association between BMI and MD could be operating early in life. If, on the contrary, we assume we had found no association between childhood BMI and MD/breast cancer, whereas one had adult BMI, this would also be very important, because it would indicate that the adult association is based on weight gain in adulthood, and hence the scenario for both exploring the mechanisms and opportunities for prevention would be very different. Finding the associations in childhood that account for the adult association exclude this possibility. Still, the robustness of associations observed in current study to adjustment for other breast cancer risk factors, including BMI to address the possibility of tracking in more detail, will be examined in a subsequent study, in a subset of women from the current data set who participated in Danish Diet, Cancer and Health cohort .
The participation rate in the Copenhagen mammography screening program in the period from 1991 to 2001 was between 67% and 70% , and women who refused to participate were more likely to be unmarried, older, of non-Danish origin, and have less contact with health care (primary physician or dentist), than did screened women . However, a U-shaped curve was found for an association between education and screening nonparticipation, reflecting high rates of nonparticipation among both, women with highest and lowest education .
We used a dichotomized outcome of high (mixed/dense breasts) and low (fatty breasts) MD, as no other measure of MD was available. In contrast, a wide variety of measures of MD were used in related studies, including the Boyd semiquantitative scale with six levels (A to F) [18, 23], Wolfe score of four qualitative categories [17, 21], BI-RADS , PMD [19, 20, 22, 26, 27], or absolute and percentage dense breast volume . However, the dichotomous outcome has been used successfully earlier in a study of MD and breast cancer mortality , and showed an expected doubling of the breast cancer risk in women with mixed/dense compared with women with fatty breasts, with HR or 2.34 (1.97 to 2.78) (Table 3), in agreement with Boyd et al. . Furthermore, we successfully validated dichotomous MD measure in a subset of 118 women from this study who had their negative screening mammograms reevaluated and assigned BI-RADS, for a related study . Specifically, in these 118 women, we compared dichotomous MD outcome (fatty, which should be equivalent to BI-RADS code 1 and part of 2; and mixed/dense, which should be equivalent to part of BI-RADS code 2, 3, or 4) with BI-RADS code, and found rather good agreement: among the 31 women coded as having fatty breasts, 32% were estimated as having BI-RADS code 1, 61% BI-RADS code 2, and 7% BI-RADS code 3, whereas among 87 women with mixed/fatty breasts, 1% had BI-RADS code 1, 31% BI-RADS code 2, 62% BI-RADS code 3, and 6% BI-RADS code 4 at reevaluation.