Body Size Across the Life Course, Mammographic Density, and Risk of Breast Cancer

Abstract

Adult body mass index (BMI) is inversely associated with premenopausal breast cancer risk, and childhood and adolescent body size is inversely associated with breast cancer risk in pre- and postmenopausal women. Breast density is inversely related to body size and may play a role in the association of body size with breast cancer risk. The authors conducted a nested case-control study including 1,528 cases and 2,844 controls from the Nurses’ Health Study (1989–2004) and Nurses’ Health Study II (1996–2003). Prior to breast cancer diagnosis, participants reported their body fatness during childhood and adolescence, BMI at age 18 years, and current BMI. Mammographic density was measured by using a computer-assisted thresholding method. The inverse association between adult BMI and premenopausal breast cancer (for BMI ≥30 vs. BMI 20–22.4, odds ratio = 0.64, 95% confidence interval: 0.38, 1.06) (P_trend = 0.36) became positive after adjustment for mammographic density (odds ratio = 1.28, 95% confidence interval: 0.72, 2.30) (P_trend = 0.07). Conversely, the inverse association between childhood and adolescent body size and breast cancer risk remained after adjustment for mammographic density. The inverse association between adult BMI and premenopausal breast cancer risk may be partially due to negative confounding by mammographic density. Conversely, mammographic density does not appear to explain the inverse association between childhood and adolescent body fatness and breast cancer risk.

body mass index, body size, breast neoplasms, mammography

Body mass index (BMI) throughout the lifetime is associated with breast cancer risk, with adult BMI having opposite effects in premenopausal and postmenopausal women. A higher adult BMI is inversely associated with premenopausal breast cancer risk and positively associated with postmenopausal breast cancer risk (1). The increased risk associated with BMI in postmenopausal women is likely explained, in large part, by increases in concentrations of endogenous estrogens and decreased sex hormone-binding globulin after menopause (2, 3). In contrast, the biologic mechanisms of the inverse association between BMI and premenopausal breast cancer remain unresolved. In addition, BMI at age 18 years and body fatness during youth are also inversely associated with breast cancer risk in both pre- and postmenopausal women (4–12). These associations are supported by animal models, epidemiologic studies, and mathematical models of breast cancer etiology that demonstrate that the years before first birth are critical in establishing breast cancer risk (7, 13–18). Although the mechanisms are not understood, it has been hypothesized that rapid adolescent growth may increase breast cancer risk and that girls with more body fat may experience slower adolescent growth (7).

Breast density is one of the strongest and most consistent risk factors for breast cancer. Most studies have reported over a 4-fold increased risk of breast cancer incidence for women with the most dense breasts (19–22), a magnitude that sets it apart from other established breast cancer risk factors, which are usually less than 2-fold (21). Dense breast tissue consists of epithelial cells and connective tissue that appear light on a mammogram, while fat, the other major component of the breast, appears dark. Previous studies have shown that many established breast cancer risk factors, such as parity and age at first birth, are associated with mammographic density and breast cancer risk in the same direction (23, 24). However, the biologic mechanism behind the association between mammographic density and breast cancer is not completely understood, but mammographic density may reflect the amount and proliferation of epithelial and stromal cells in the breast (25, 26) and exposure of the breast to mitogens and mutagens (27). Although breast density is believed to be an independent risk factor for breast cancer (28), it is inversely associated with body size throughout the life course (29, 30). It is therefore important to consider what role (confounder, intermediate, effect modifier), if any, mammographic density may play in the inverse association between body size at different time periods and breast cancer incidence.

Three studies have examined the association between adult body size and breast cancer risk while accounting for mammographic density (28, 31, 32). All observed a stronger positive association between adult body size and postmenopausal breast cancer risk after adjustment for mammographic density, and 2 observed a positive association between body size and premenopausal breast cancer risk (28, 31). However, these studies did not examine what role mammographic density plays in the association between youth body size and breast cancer risk, and only 1 examined the potential interaction between body size and mammographic density (32).

In this study, we investigated the association between body fatness during childhood and adolescence, BMI at age 18 years, and BMI in premenopausal women and breast cancer risk after adjustment for mammographic density, using data from prospective case-control studies nested within the Nurses’ Health Study (NHS) and Nurses’ Health Study II (NHS II). We also examined whether there was an interaction between body size and mammographic density.

MATERIALS AND METHODS

Study population

The NHS was established in 1976 among 121,700 registered nurses aged 30–55 years, and the NHS II was established in 1989 among 116,430 registered nurses aged 25–42 years. All women completed a baseline questionnaire that collected information on demographic and lifestyle factors, anthropometric variables, and disease history. Follow-up questionnaires are sent biennially to participants updating information on newly diagnosed diseases, anthropometric factors, and other risk factors.

From 1989 to 1990, 32,826 NHS participants and, from 1996 to 1999, 29,611 NHS II participants provided blood samples and completed a brief questionnaire. Details have been described previously (33, 34). We conducted our analyses among cases and controls from breast cancer case-control studies nested within the NHS and NHS II cohorts. For these women, mammograms were also obtained. These nested case-control studies included breast cancer cases diagnosed after blood collection but before June 1, 2004, for the NHS and before June 1, 2003, for the NHS II, as well as controls, who were matched to cases on age, menopausal status, postmenopausal hormone use, race/ethnicity, time of day, month, and fasting status at blood draw. Breast cancer cases were confirmed by medical record review. More detailed information on the identification of breast cancer cases has been described previously (35).

Mammographic collection began in 1995 for the NHS and in 2005 for the NHS II (Figure 1). For the NHS, mammograms were obtained for 1,449 (96.1%) of the cases and 2,427 (95.8%) of the controls from the nested case-control study who were alive and eligible to participate in mammogram collection. Additional details of the collection of mammograms in the NHS have been described previously (36). For the NHS II, mammograms were successfully obtained from 253 (92.3%) NHS II cases and 502 (96.7%) NHS II controls. We further restricted to subjects who had a mammogram date that was before the date of diagnosis for a total of 223 cases and 482 controls from the NHS II and 1,305 cases and 2,362 controls from the NHS. Women for whom mammograms could and could not be obtained were similar with respect to age, body mass index, and circulating hormone levels (36).

Figure 1.

Description of mammogram collection in the Nurses’ Health Study (NHS, 1989–2004) and the Nurses’ Health Study II (NHS II, 1996–2003).

As the association between adult BMI and breast cancer risk differs by menopausal status, the study population was restricted to premenopausal women for the adult BMI and breast cancer analysis. For body fatness during youth and BMI at age 18 years, the inverse association with breast cancer does not differ for premenopausal and postmenopausal breast cancer; therefore, data are combined in these analyses. Women were defined as premenopausal if they reported that their periods had not ceased or reported having a hysterectomy but with at least 1 ovary remaining and were aged 47 years or less for nonsmokers or 45 years or less for smokers. They were defined as postmenopausal if they reported that their natural menstrual periods had stopped permanently, had a bilateral oophorectomy, or had a hysterectomy with at least 1 ovary remaining and were aged 56 years or more for nonsmokers or 54 years or more for smokers. This study was approved by the institutional review boards of the Harvard School of Public Health and Brigham and Women's Hospital. Written authorization was obtained for mammography collection, and implied consent was assumed upon return of the completed questionnaire.

Mammographic density measurements

Mammographic density was assessed by digitizing the craniocaudal views of both breasts with a Lumisys 85 laser film scanner (Lumisys, Sunnyvale, California). For each image, 1 threshold level was set to define the edge of the breast, and a second was set to delineate the dense area of the breast within the original threshold region. Cumulus software (37) calculated the total number of pixels within the entire region of interest and within the area identified as dense (absolute density in cm²). These values were used to calculate the percent of the breast area that was dense and the nondense area. Reproducibility of the percent mammographic density in this study was high, with a within-person intraclass correlation coefficient of 0.93 (38). We used the average of both breasts for the density measures; studies have shown similar results when the density of a random side or the average of the 2 was used (39).

Assessment of body size

NHS and NHS II participants reported their height and weight at baseline; current weight was updated every 2 years. BMI at the time of mammogram was assessed by use of data from the biennial questionnaire before the date of the mammogram. BMI was calculated (weight (kg)/height (m)²). Weight at age 18 years was reported in 1980 (NHS) and 1989 (NHS II). In 1988 (NHS) and 1989 (NHS II), participants were asked to recall their body fatness at ages 5, 10, and 20 years using a 9-level figure drawing (Figure 2) (40).

Figure 2.

The 9-figure body size scale used to assess body fatness at ages 5, 10, and 20 years in the Nurses’ Health Study (1988) and the Nurses’ Health Study II (1989). Reprinted from Genetics of Neurological and Psychiatric Disorders by S. S. Kety et al., eds. (40). Copyright 1983, with permission from Wolters Kluwer.

Covariates

Information on risk factors for breast cancer was collected at baseline and on biennial questionnaires. Participants reported their age, height, and age at menarche at baseline. Personal history of benign breast disease, family history of breast cancer, ages at first birth and menopause, parity, oral contraceptive use, menopausal status, duration of postmenopausal hormone (PMH) use, physical activity, and alcohol consumption were reported at baseline and on follow-up questionnaires. Birth weight was reported in 1991 for NHS II and in 1992 for NHS.

Statistical analyses

Unconditional logistic regression was used to estimate odds ratios and 95% confidence intervals for the associations between adult BMI (in premenopausal women), BMI at age 18 years, and body fatness in childhood and adolescence and breast cancer with and without adjustment for percent density, absolute density, and absolute nondense area. Adult BMI was categorized into <20, 20–<22.5, 22.5–<25, 25–<27.5, 27.5–<30, and ≥30 kg/m². BMI at age 18 years was categorized into <18.5, 18.5–<20, 20–<22.5, 22.5–<25, and ≥25 kg/m². We averaged each participant's values at ages 5 and 10 (childhood) and ages >10 and 20 (adolescent) to obtain estimates of childhood and adolescent body fatness and categorized these into 1, 1.5–2, 2.5–3, 3.5–4, and ≥4.5. Because few participants recalled their body fatness as greater than level 5, the upper categories were combined. Percent density, absolute density, and nondense area were kept as continuous variables.

We included the following a priori potential confounders in the covariate-adjusted models: age at menarche (<12, 12, 13, ≥14 years), parity/age at first birth (nulliparous, age at first birth <25 years/1–4 children, age at first birth 25–29 years/1–4 children, age at first birth ≥30 years/1–4 children, age at first birth <25/≥5 children, age at first birth ≥25 years/≥5 children), family history of breast cancer (yes/no), alcohol consumption (0, <5, 5–<15, ≥15 g/day), and menopausal status (premenopausal, postmenopausal, unknown). We did not include age at menarche and adult BMI in the childhood and adolescent body fatness analyses as they could be intermediate variables on the causal pathway. Birth weight (<5.5, 5.5–6.9, 7–8.4, ≥8.5 pounds; 1 pound = 0.45 kg) was included in the childhood and adolescent body fatness analyses. In analyses that included postmenopausal women, we also adjusted for age at menopause (<46, 46–50, 51–54, ≥55 years) and duration of postmenopausal hormone use (continuous). Physical activity, duration of oral contraceptive use, and height were also assessed but were not included in the covariate-adjusted models as they were not associated with breast cancer in this subcohort. History of benign breast disease was not included in the models because women with dense breasts are more likely to be diagnosed with benign breast disease, which may be a partial surrogate measure for breast density (41, 42). Tests for trend for BMI and BMI at age 18 years were performed by using the midpoint of the interval for each category and for childhood and adolescent body fatness were performed by using a continuous variable. We conducted secondary analyses excluding women diagnosed with breast cancer within 2 years of their mammogram.

Interaction between body size measures and mammographic density was assessed with a likelihood ratio test comparing a model with the cross-product term between the 2 variables with a model having main effects only. Chi-square tests were used to obtain P values for the likelihood ratio test statistics. All tests of statistical significance were 2 sided, and all statistical analyses were performed by using SAS, version 9.1, software (SAS Institute, Inc., Cary, North Carolina).

RESULTS

The combined nested case-control studies in the NHS and NHS II included 1,528 cases (260 premenopausal, 1,265 postmenopausal, 3 unknown) and 2,844 controls (574 premenopausal, 2,267 postmenopausal, 3 unknown). At the time of mammogram, the mean age of NHS participants was 57.9 years (range, 38–84) and for NHS II participants, 45.6 years (range, 34–57). In both premenopausal and postmenopausal women, those whose mammographic density was greater than 50% had a lower BMI, had higher absolute breast density, were more likely to be nulliparous, and were more likely to report a history of benign breast disease (Table 1). Both premenopausal and postmenopausal cases were more likely than controls to have a family history of breast cancer, a history of benign breast disease, and to be nulliparous. Premenopausal cases had a lower BMI than controls, while postmenopausal cases were more likely than controls to have used postmenopausal hormones and to have drunk alcohol. Among the cases, the mean time between mammogram and diagnosis was 4.7 years, the median time was 4.1 years, and the interquartile range was 1.8–6.9 years. Among the premenopausal controls, the Spearman correlation between BMI and percent mammographic density was −0.60 (P < 0.0001), the correlation between BMI and absolute density was −0.21 (P < 0.0001), and the correlation between BMI and nondense area was 0.62 (P < 0.0001). Among all controls, the correlations between percent density and BMI at age 18 years, adolescent body fatness (measured by the averaged body size figures at ages >10 and 20), and childhood body fatness were −0.30, −0.20, and −0.14, respectively (P < 0.0001). A strong positive association was observed between percent density and premenopausal breast cancer risk (for top vs. bottom quartile, odds ratio (OR) = 3.15, 95% confidence interval (CI): 1.92, 5.18) (P_trend < 0.0001), a slightly weaker positive association was observed between absolute density and premenopausal breast cancer risk (for top vs. bottom quartile, OR = 2.14, 95% CI: 1.35, 3.39) (P_trend = 0.001), and a strong inverse association was observed between nondense area and premenopausal breast cancer risk (for top vs. bottom quartile, OR = 0.47, 95% CI: 0.30, 0.73) (P_trend = 0.001).

Table 1.

Characteristics of the Study Population at Time of Mammogram by Menopausal Status According to Categories of Mammographic Density, Nurses’ Health Study (1989–2004) and Nurses’ Health Study II (1996–2003)

	Mammographic Density, %
	<10			10–24			25–49			50–74			≥75
	Mean	No.	%	Mean	No.	%	Mean	No.	%	Mean	No.	%	Mean	No.	%
Premenopausal
No. of participants		71			99			326			276			62
Absolute density, cm2	15.1			46.2			81.1			102.6			112.7
Nondense area, cm2	237.7			209.4			132.0			67.3			27.1
Age at mammogram, years	47.1			46.5			45.6			45.1			45.2
Height, inchesa	64.6			64.6			65.0			65.0			65.0
BMI at age 18 years, kg/m2	24.1			22.4			20.9			20.2			20.1
BMI at mammogram, kg/m2	32.6			28.6			25.2			22.7			21.3
Childhood body fatnessb	3.6			2.8			2.6			2.4			2.3
Adolescent body fatnessc	3.9			3.1			2.9			2.6			2.6
Age at menarche, years	12.0			12.2			12.4			12.6			12.5
Age at first birth, years	24.8			25.7			26.5			26.0			26.6
No. of children (among parous women)	2.6			2.7			2.4			2.4			2.3
Nulliparous			8.5			7.1			11.0			19.2			16.1
Physical activity, METs/week	12.8			16.9			16.7			20.9			14.8
Alcohol consumption, g/day	2.9			3.8			4.6			4.7			2.6
Family history of breast cancer			5.6			5.1			10.1			7.6			14.5
History of benign breast disease			42.3			37.4			47.6			60.1			67.7
Postmenopausal
No. of participants		648			1,119			1,247			468			50
Absolute density, cm2	12.1			36.9			62.3			87.7			84.5
Nondense area, cm2	217.8			178.6			111.7			59.9			20.8
Age at mammogram, years	60.6			60.0			57.3			54.8			52.7
Height, inchesa	64.6			64.5			64.7			64.8			64.5
BMI at age 18 years, kg/m2	22.7			21.3			20.7			20.6			19.8
BMI at mammogram, kg/m2	29.6			26.5			24.6			23.3			21.4
Childhood body fatnessb	2.9			2.4			2.3			2.2			2.5
Adolescent body fatnessc	3.2			2.6			2.5			2.5			2.5
Age at menarche, years	12.5			12.5			12.6			12.6			12.5
Age at first birth, years	25.1			25.1			25.2			25.6			25.3
No. of children (among parous women)	3.5			3.4			3.0			2.7			2.5
Nulliparous			5.1			5.6			7.4			10.9			12.0
Physical activity, METs/week	15.3			17.5			17.9			18.2			14.9
Alcohol consumption, g/day	4.9			5.1			6.1			6.6			5.4
Family history of breast cancer			12.0			13.7			12.5			13.0			14.0
History of benign breast disease			34.9			44.8			54.9			66.9			70.0
Age at menopause, years	49.7			49.8			49.6			49.9			49.1
Current postmenopausal hormone use			26.7			32.2			45.2			38.5			40.0
Duration of PMH use, months	35.2			42.1			44.6			36.1			36.6

Abbreviations: BMI, body mass index; MET, metabolic task; PMH, postmenopausal hormone.

a

One inch = 2.54 cm.

b

Average of childhood fatness at ages 5 and 10 years by 9-level figure drawing (Figure 2, reprinted with permission).

c

Average of adolescent fatness at ages >10 and 20 years by 9-level figure drawing (Figure 2, reprinted with permission).

A borderline significant inverse association was observed between BMI and premenopausal breast cancer, with a covariate-adjusted odds ratio of 0.64 (95% CI: 0.38, 1.06) (P_trend = 0.36) comparing women having a BMI ≥30 with women having a BMI of 20–22.4, but there was no linear trend. After adjustment for percent density, there was a positive linear trend of borderline significance, and a BMI ≥30 was no longer associated with breast cancer risk (OR = 1.28, 95% CI: 0.72, 2.30) (P_trend = 0.07), but women with a BMI of 27.5–29.9 were at an increased risk (OR = 2.86, 95% CI: 1.44, 5.68). Adjustment for absolute density did not materially alter the results, and adjustment for the nondense area affected the effect estimates similarly to adjustment for percent density (Table 2).

Table 2.

Odds Ratios and 95% Confidence Intervals of Premenopausal Breast Cancer by Body Mass Index at Mammogram Among Women in the Nurses’ Health Study (1989–2004) and the Nurses’ Health Study II (1996–2003)

	Body Mass Index												Ptrenda
	<20b		20–22.4b		22.5–24.9b		25–27.4b		27.5–29.9b		≥30b
	OR	95% CI	OR	95% CI	OR	95% CI	OR	95% CI	OR	95% CI	OR	95% CI
Age-adjusted modelc	0.80	0.44, 1.47	1.00	Referent	0.81	0.54, 1.22	1.06	0.68, 1.66	1.77	0.95, 3.32	0.63	0.39, 1.04	0.42
Covariate-adjusted modeld	0.86	0.47, 1.58	1.00	Referent	0.85	0.56, 1.28	1.08	0.68, 1.71	1.75	0.92, 3.33	0.64	0.38, 1.06	0.36
Covariate-adjusted model + percent mammographic densitye	0.75	0.40, 1.38	1.00	Referent	1.04	0.68, 1.60	1.59	0.97, 2.59	2.86	1.44, 5.68	1.28	0.72, 2.30	0.07
Covariate-adjusted model + absolute mammographic densityf	0.86	0.47, 1.59	1.00	Referent	0.82	0.54, 1.25	1.06	0.67, 1.68	1.71	0.89, 3.27	0.68	0.41, 1.13	0.52
Covariate-adjusted model + nondense areag	0.81	0.44, 1.49	1.00	Referent	0.99	0.65, 1.51	1.51	0.92, 2.47	2.73	1.36, 5.46	1.12	0.62, 2.01	0.23

Abbreviations: CI, confidence interval; OR, odds ratio.

a

Determined by using category medians.

b

Number of cases/number of controls by body mass index category (kg/m²): <20, 19/46; 20–22.4, 77/151; 22.5–24.9, 60/146; 25–27.4, 48/94; 27.5–29.9, 23/26; ≥30, 31/100.

c

Unconditional logistic regression, age adjusted.

d

Unconditional logistic regression adjusted for age at menarche (<12, 12, 13, ≥14), parity/age at first birth (nulliparous, age at first birth <25 years/1–4 children, age at first birth 25–29 years/1–4 children, age at first birth ≥30 years/1–4 children, age at first birth <25 years/≥5 children, age at first birth ≥25 years/≥5 children), family history of breast cancer (yes/no), and alcohol (nondrinker, 0–<5, 5–<15, ≥15 g/day).

e

Unconditional logistic regression adjusted for the variables above plus average percent mammographic density (continuous).

f

Unconditional logistic regression adjusted for the variables above plus average absolute mammographic density (continuous).

g

Unconditional logistic regression adjusted for the variables above plus average nondense area (continuous).

An inverse association was observed between BMI at age 18 years and breast cancer risk. Compared with women having a BMI of 20–22.4, women with a BMI ≥25 had a covariate-adjusted odds ratio of 0.68 (95% CI: 0.52, 0.87) (P_trend = 0.02). Adjustment for percent density attenuated the association while separate adjustment for absolute density and nondense area caused a smaller attenuation of the association (Table 3).

Table 3.

Odds Ratios and 95% Confidence Intervals of Breast Cancer by Body Mass Index at Age 18 Years Among Women in the Nurses’ Health Study (1989–2004) and the Nurses’ Health Study II (1996–2003)

	Body Mass Index at Age 18 Years										Ptrenda
	<18.5b		18.5–19.9b		20–22.4b		22.5–24.9b		≥25b
	OR	95% CI	OR	95% CI	OR	95% CI	OR	95% CI	OR	95% CI
Age-adjusted modelc	0.99	0.81, 1.22	1.03	0.87, 1.22	1.00	Referent	0.97	0.80, 1.16	0.69	0.54, 0.89	0.02
Covariate-adjusted modeld	0.97	0.79, 1.20	1.03	0.87, 1.23	1.00	Referent	0.96	0.80, 1.16	0.68	0.52, 0.87	0.02
Covariate-adjusted model + percent mammographic densitye	0.92	0.75, 1.14	0.96	0.81, 1.14	1.00	Referent	1.07	0.89, 1.30	0.84	0.65, 1.09	0.78
Covariate-adjusted model + absolute mammographic densityf	1.00	0.81, 1.23	1.03	0.87, 1.22	1.00	Referent	1.00	0.83, 1.21	0.73	0.57, 0.95	0.06
Covariate-adjusted model + nondense areag	0.94	0.76, 1.15	1.01	0.85, 1.20	1.00	Referent	1.01	0.84, 1.22	0.76	0.59, 0.99	0.29

Abbreviations: CI, confidence interval; OR, odds ratio.

a

Determined by using category medians.

b

Number of cases/number of controls by body mass index category (kg/m²): <18.5, 185/339; 18.5–19.9, 314/559; 20–22.4, 616/1,124; 22.5–24.9, 238/449; ≥25, 99/260.

c

Unconditional logistic regression, age adjusted.

d

Unconditional logistic regression adjusted for age at menarche (<12, 12, 13, ≥14), parity/age at first birth (nulliparous, age at first birth <25 years/1–4 children, age at first birth 25–29 years/1–4 children, age at first birth ≥30 years/1–4 children, age at first birth <25 years/≥5 children, age at first birth ≥25 years/≥5 children), family history of breast cancer (yes/no), and alcohol (nondrinker, 0–<5, 5–<15, ≥15 g/day).

e

Unconditional logistic regression adjusted for the variables above plus average percent mammographic density (continuous).

f

Unconditional logistic regression adjusted for the variables above plus average absolute mammographic density (continuous).

g

Unconditional logistic regression adjusted for the variables above plus average nondense area (continuous).

Average body fatness during both childhood and adolescence was significantly inversely associated with breast cancer risk. The greatest decrease in risk was observed for adolescent body fatness with a covariate-adjusted odds ratio of 0.58 (95% CI: 0.44, 0.78) (P_trend < 0.0001) comparing those with level 4.5 or higher with level 1. The corresponding odds ratio for childhood body fatness was 0.67 (95% CI: 0.52, 0.86) (P_trend = 0.0002). Adjustment for percent density slightly altered the associations; however, they remained marginally significant. Adjustment for absolute density and nondense area altered the associations only marginally (Table 4).

Table 4.

Odds Ratios and 95% Confidence Intervals of Breast Cancer Self-reported Body Fatness at Young Ages Among Women in the Nurses’ Health Study (1989–2004) and Nurses’ Health Study II (1996–2003)

	Average Body Fatness										Ptrenda
	1		1.5–2		2.5–3		3.5–4		≥4.5
	OR	95% CI	OR	95% CI	OR	95% CI	OR	95% CI	OR	95% CI
	Average during childhood (ages 5–10 years)b
Age-adjusted modelc	1.00	Referent	0.91	0.77, 1.08	0.84	0.70, 1.01	0.71	0.57, 0.87	0.66	0.52, 0.84	<0.0001
Covariate-adjusted modeld	1.00	Referent	0.92	0.77, 1.09	0.85	0.70, 1.02	0.73	0.59, 0.90	0.67	0.52, 0.86	0.0002
Covariate-adjusted model + percent mammographic densitye	1.00	Referent	0.91	0.77, 1.08	0.89	0.74, 1.08	0.83	0.66, 1.03	0.77	0.60, 0.99	0.02
Covariate-adjusted model + absolute mammographic densityf	1.00	Referent	0.89	0.75, 1.06	0.84	0.69, 1.01	0.74	0.60, 0.92	0.69	0.54, 0.89	0.0005
Covariate-adjusted model + nondense areag	1.00	Referent	0.93	0.78, 1.10	0.88	0.73, 1.07	0.78	0.63, 0.97	0.72	0.56, 0.92	0.003
	Average during adolescence (ages >10–20 years)h
Age-adjusted modelc	1.00	Referent	0.85	0.68, 1.07	0.77	0.61, 0.97	0.71	0.55, 0.90	0.57	0.43, 0.76	<0.0001
Covariate-adjusted modeld	1.00	Referent	0.85	0.67, 1.07	0.79	0.62, 0.99	0.73	0.57, 0.94	0.58	0.44, 0.78	<0.0001
Covariate-adjusted model + percent mammographic densitye	1.00	Referent	0.84	0.66, 1.06	0.82	0.64, 1.03	0.82	0.64, 1.06	0.71	0.53, 0.95	0.05
Covariate-adjusted model + absolute mammographic densityf	1.00	Referent	0.81	0.64, 1.02	0.75	0.59, 0.95	0.72	0.56, 0.93	0.59	0.44, 0.79	0.0004
Covariate-adjusted model + nondense areag	1.00	Referent	0.87	0.69, 1.09	0.83	0.65, 1.04	0.79	0.61, 1.02	0.65	0.49, 0.87	0.005

Abbreviations: CI, confidence interval; OR, odds ratio.

a

Determined by using category medians.

b

Number of cases/number of controls by average body fatness category according to 9-level figure drawing with upper levels combined (Figure 2, reprinted with permission): 1, 456/725; 1.5–2, 433/758; 2.5–3, 294/564; 3.5–4, 183/416; ≥4.5, 119/290.

c

Unconditional logistic regression, age adjusted.

d

Unconditional logistic regression adjusted for age at menarche (<12, 12, 13, ≥14), parity/age at first birth (nulliparous, age at first birth <25 years/1–4 children, age at first birth 25–29 years/1–4 children, age at first birth ≥30 years/1–4 children, age at first birth <25 years/≥5 children, age at first birth ≥25 years/≥5 children), family history of breast cancer (yes/no), and alcohol (nondrinker, 0–<5, 5–<15, ≥15 g/day).

e

Unconditional logistic regression adjusted for the variables above plus average percent mammographic density (continuous).

f

Unconditional logistic regression adjusted for the variables above plus average absolute mammographic density (continuous).

g

Unconditional logistic regression adjusted for the variables above plus average nondense area (continuous).

h

Number of cases/number of controls by average body fatness category according to 9-level figure drawing with upper levels combined (Figure 2, reprinted with permission): 1, 167/238; 1.5–2, 478/806; 2.5–3, 445/834; 3.5–4, 266/545; ≥4.5, 136/343.

Among premenopausal women, no significant interaction between adult BMI and percent density was observed (P_interaction = 0.15). Premenopausal women with a low mammographic density of <25% and BMI ≥25 had a covariate-adjusted odds ratio of 0.30 (95% CI: 0.14, 0.66) compared with women who had a high mammographic density of ≥50% and BMI <25, while women with low mammographic density and a BMI <25 had a covariate-adjusted odds ratio of 0.29 (95% CI: 0.11, 0.77) compared with the same reference group (Table 5). The association between percent density and breast cancer risk was similar among women with childhood body fatness levels above and below 3.5 (P_interaction = 0.25) (Table 6). The association between percent density and breast cancer risk was also similar among women with adolescent body fatness levels above and below 3.5 (P_interaction = 0.11) (data not shown). We conducted sensitivity analyses excluding women diagnosed with breast cancer within 2 years of their mammogram, and these results were similar to those of the main analyses (data not shown).

Table 5.

Covariate-adjusteda Odds Ratios and 95% Confidence Intervals of Premenopausal Breast Cancer Among Women in the Nurses’ Health Study (1989–2004) and the Nurses’ Health Study II (1996–2003) According to Body Mass Index and Percent Mammographic Densityb

	Percent Mammographic Density
	<25		25–<50		≥50
	OR	95% CI	OR	95% CI	OR	95% CI
BMI <25c	0.29	0.11, 0.77	0.73	0.48, 1.10	1.00	Referent
BMI ≥25d	0.30	0.14, 0.66	0.96	0.55, 1.70	1.31	0.67, 2.57

Abbreviations: BMI, body mass index; CI, confidence interval; OR, odds ratio.

a

Unconditional logistic regression adjusted for age at menarche (<12, 12, 13, ≥14), parity/age at first birth (nulliparous, age at first birth <25 years/1–4 children, age at first birth 25–29 years/1–4 children, age at first birth ≥30 years/1–4 children, age at first birth <25/≥5 children, age at first birth ≥25/≥5 children), family history of breast cancer (yes/no), alcohol (nondrinker, 0–<5, 5–<15, ≥15 g/day), and body mass index (continuous).

b

P_{heterogeneity} = 0.15 (based on a likelihood ratio test comparing a model having the cross-product term between body mass index and percent mammographic density with a model having main effects only).

c

Number of cases/number of controls by percent mammographic density category: <25, 5/34; 25–<50, 53/130; ≥50, 98/179.

d

Number of cases/number of controls by percent mammographic density category: <25, 23/107; 25–<50, 54/84; ≥50, 25/29.

Table 6.

Covariate-adjusteda Odds Ratios and 95% Confidence Intervals of Breast Cancer Among Women in the Nurses’ Health Study (1989–2004) and Nurses’ Health Study II (1996–2003) According to Average Body Fatness During Childhood and Percent Mammographic Densityb

	Percent Mammographic Density
	<10		10–<25		25–<50		≥50
	OR	95% CI	OR	95% CI	OR	95% CI	OR	95% CI
Average body fatness <3.5c,d	0.43	0.33, 0.57	0.53	0.42, 0.66	0.83	0.68, 1.01	1.00	Referent
Average body fatness ≥3.5e	0.36	0.24, 0.55	0.58	0.39, 0.85	0.75	0.53, 1.08	1.18	0.76, 1.83

Abbreviations: CI, confidence interval; OR, odds ratio.

a

Unconditional logistic regression adjusted for parity/age at first birth (nulliparous, age at first birth <25 years/1–4 children, age at first birth 25–29 years/1–4 children, age at first birth ≥30 years/1–4 children, age at first birth <25/≥5 children, age at first birth ≥25/≥5 children), family history of breast cancer (yes/no), birth weight (<5.5, 5.5–6.9, 7–8.4, ≥8.5 pounds), alcohol (nondrinker, 0–<5, 5–<15, ≥15 g/day), menopausal status (premenopausal, postmenopausal, unknown), age at menopause (<46, 46–<50, 50–<55, ≥55 years), total duration of postmenopausal hormone use (continuous), and childhood body size (continuous). (One pound = 0.45 kg.)

b

P_{heterogeneity} = 0.34 (based on a likelihood ratio test comparing a model having the cross-product term between childhood body fatness and percent mammographic density with a model having main effects only).

c

Number of cases/number of controls by percent mammographic density category: <10, 113/309; 10–<25, 283/617; 25–<50, 486/723; ≥50, 301/398.

d

Body fatness categories according to 9-level figure drawing with some levels combined (Figure 2, reprinted with permission).

e

Number of cases/number of controls by percent mammographic density category: <10, 54/214; 10–<25, 78/196; 25–<50, 108/214; ≥50, 62/82.

DISCUSSION

In this study, adult BMI was inversely associated with premenopausal breast cancer risk, an association that became positive with adjustment for percent mammographic density. Among all women, a greater BMI at age 18 years and greater body fatness in childhood and adolescence were associated with decreased risk of breast cancer. Adjustment for percent density partly attenuated these associations, but a significant inverse association still remained between childhood and adolescent body fatness and breast cancer risk.

Our results for adult BMI in premenopausal women are largely consistent with those of 2 studies in which adjustment for mammographic density altered the association between body size and premenopausal breast cancer risk. Brisson et al. (31) reported a change in the odds ratio of 1.2–2.7 for the effect of body weight on premenopausal breast cancer risk after adjustment for nodular densities and homogeneous density. In contrast to our results, another case-control study reported no significant associations for premenopausal breast cancer risk and BMI or breast density even with mutual adjustment (32). Most recently, Boyd et al. (28) reported a change in the odds ratio of 0.76–1.47 among premenopausal women for the association between BMI and breast cancer after adjusting for percent density. This suggests that the inverse association observed between adult BMI and premenopausal breast cancer risk may partially be due to negative confounding by mammographic density (28).

To our knowledge, this is the first study to examine the influence of mammographic density on the association between body fatness in childhood and adolescence and breast cancer risk. Mammary tissue develops during adolescence, but the terminal structures of the mammary gland do not differentiate until the first pregnancy (43). This suggests that the years before first birth are critical in establishing breast cancer risk. Animal models, epidemiologic studies, and mathematical models of breast cancer etiology support this theory (7, 13–18). However, the mechanisms through which youth body fatness decreases breast cancer risk are not clear. Childhood and adolescent adiposity are associated with mammographic density independent of adult body size (30), and our results suggest that mammographic density may be a weak negative confounder of the childhood and adolescent body size and breast cancer association. However, unlike the adult BMI and premenopausal breast cancer association, negative confounding by mammographic density does not appear to explain the inverse association between childhood and adolescent body fatness and breast cancer risk. Girls with more body fat may have higher levels of sex hormones that could lead to earlier differentiation of breast tissue, resulting in cells less susceptible to malignant transformation (44). However, Baer et al. (45) reported no association between BMI in girls aged 8–10 years and estrogen and progesterone levels. Alternatively, body fatness may influence breast cancer risk through its relation with adolescent growth. Childhood body fatness is associated with slower adolescent growth; peak height growth velocity, a measure of adolescent growth, has been associated with breast cancer risk (7). Furthermore, childhood and adolescent body size are inversely associated with adult insulin-like growth factor 1 levels in the NHS II (46). Childhood growth velocity and height are correlated with insulin-like growth factor 1 levels measured at ages 5–8 years (47). These findings suggest a possible role of the insulin-like growth factor axis in the inverse association between childhood and adolescent body fatness and breast cancer risk.

Adjustment for percent mammographic density consistently moved the odds ratio for the association between BMI and premenopasual breast cancer in a positive direction. Conversely, adjustment for absolute density did not materially alter the association between BMI and breast cancer. The association between BMI and percent density was due primarily to the contribution of the nondense area of the breast, consisting mainly of fatty tissue. Adjustment for the nondense area resulted in changes in the effect estimates that were similar to, but slightly weaker than, adjustment for percent density. The dense area of the breast, because it harbors the epithelial and stromal cells, is hypothesized to influence breast cancer risk. Nevertheless, percent density has consistently been most strongly associated with breast cancer risk.

Body size in childhood and adolescence was recalled by participants, which may have resulted in misclassification of these exposures. However, the long-term recall of body fatness levels by using the 9-level figure diagram has been shown to be highly correlated with BMI at the same age (48). BMI at age 18 years and BMI at mammogram may also have been affected by measurement error. Recalled weight at age 18 years and self-reported weight in adults have been validated, with correlations of 0.84 and higher between self-reported weight and directly measured weight (49, 50). Because the information on all body size measures was collected prior to mammograms and breast cancer diagnosis, any misclassification of these variables will be nondifferential.

To our knowledge, this is the first study to examine what role mammographic density plays in the association among childhood and adolescent body size, BMI at age 18 years, and breast cancer risk. We also have high follow-up rates, cancer cases that have been medically confirmed, and data on many covariates, including known risk factors for breast cancer that have been collected and updated at 2-year intervals.

In conclusion, our findings suggest that the inverse association between adult BMI and premenopausal breast cancer risk may be partially due to negative confounding by mammographic density. However, we cannot exclude the possibility that adult BMI and mammographic density operate through the same pathway to influence risk of premenopausal breast cancer. In addition, mammographic density does not appear to explain the inverse association between childhood and adolescent body fatness and breast cancer risk. Future studies that examine the role of the insulin-like growth factor axis in childhood and adolescent growth may help to elucidate the mechanisms underlying this relation.

Abbreviations

Abbreviations
 
• BMI

  body mass index

 
• CI

  confidence interval

 
• NHS

  Nurses’ Health Study

 
• NHS II

  Nurses’ Health Study II

 
• OR

  odds ratio

Author affiliations: Obstetrics and Gynecology Epidemiology Center, Department of Obstetrics, Gynecology, and Reproductive Biology, Brigham and Women's Hospital, Harvard Medical School, Boston, Massachusetts (Holly R. Harris, Karin B. Michels); Department of Epidemiology, Harvard School of Public Health, Boston, Massachusetts (Holly R. Harris, Rulla M. Tamimi, Walter C. Willett, Susan E. Hankinson, Karin B. Michels); Department of Medicine, Channing Laboratory, Brigham and Women's Hospital, Harvard Medical School, Boston, Massachusetts (Rulla M. Tamimi, Walter C. Willett, Susan E. Hankinson, Karin B. Michels); and Department of Nutrition, Harvard School of Public Health, Boston, Massachusetts (Walter C. Willett).

This work was supported by the National Cancer Institute at the National Institutes of Health (grant R01 CA114326 to K. B. M., grant P01 CA87969 to S. E. H., grant CA50385 to W. C. W., grants CA124865 and CA131332 to R. M. T.); the National Institutes of Health (training grant T32 ES007069 to H. R. H.); the Maternal Child and Health Bureau at the US Department of Health and Human Services (grant 5T76MC00001 (formerly MCJ201) to H. R. H.); and the Breast Cancer Research Foundation.

Conflict of interest: none declared.