Mammographic density and risk of breast cancer by adiposity: An analysis of four case-control studies

Abstract

The association of mammographic breast density with breast cancer risk may vary by adiposity. To examine effect modification by body mass index (BMI), the authors standardized mammographic density data from four case-control studies (1994–2002) conducted in California, Hawaii, and Minnesota, and Gifu, Japan. The 1,699 cases and 2,422 controls included 45% Caucasians, 40% Asians, and 9% African-Americans. Using ethnic-specific BMI cut points, 34% were classified as overweight and 19% as obese. A single reader assessed density from mammographic images using a computer-assisted method. Logistic regression was used to estimate odds ratios (OR) and 95% confidence intervals (95% CI) while adjusting for potential confounders. Modest heterogeneity in the relation between percent density and breast cancer risk across studies was observed (p_{heterogeneity} = 0.08). Cases had a greater age-adjusted mean percent density than controls: 31.7% versus 28.5%, respectively (p <0.001). Relative to <20 percent density, the ORs for >35 were similar across BMI groups whereas the OR for 20–35 was slightly higher in overweight (OR = 1.69, 95% CI: 1.28, 2.24) and obese (OR = 1.62, 95% CI: 1.12, 2.33) than in normal weight women (OR = 1.49, 95% CI: 1.11, 2.01). Furthermore, limited evidence of effect modification by BMI of the OR per 10% increase in percent density (p_interaction = 0.06) was observed, including subgroup analyses by menopausal status and in analyses that excluded women at the extremes of the BMI scale. Our findings indicate little, if any, modification by BMI of the effects of breast density on breast cancer risk.

Introduction

Breast density assessed from mammography refers to the distribution of fat, connective, and epithelial tissue in the breast. A high percentage of radiologically dense glandular and fibrous stromal tissue on mammographic images is a strong and consistent predictor of breast cancer risk,^1–2 the presence of dense tissue in more than 50% of the breast may account for 30% of breast cancer cases.^3–4 Factors that may modify the risk associated with breast density are of public health importance, particularly when breast density is used as an intermediate endpoint in intervention studies⁵ and as a predictor of individualized breast cancer risk.^6–8

Several possible biologic mechanisms exist by which adiposity may adversely influence breast cancer carcinogenesis including endogenous estrogen levels, insulin and insulin-like growth factors, adipokine production, low-grade inflammatory state, and oxidative stress.^9–10 A stronger association between breast density and breast cancer risk in obese than normal weight women has been suggested in previous studies.^11–15 Body mass index (BMI), a measure of overall adiposity, is positively correlated with the amount of adipose tissue in the breast^16–18 and with compressed breast thickness on mammograms, resulting in lower image contrast.¹⁹ Due to greater compressed breast thickness a given projected area of density on a mammogram in overweight and obese women may represent a higher quantity of dense breast tissue than in normal-weight women.

To examine potential modification of the association between percent breast density and breast cancer risk by adiposity, mammographic data was standardized from four case-control studies representing ethnically diverse populations with a wide variation in BMI. This study contributes to the growing literature on potential modifiers of the effects of percent density on breast cancer risk.^{11, 13}

MATERIALS AND METHODS

Study design

This analysis included participants from four previous case-control studies conducted in different geographic locations: California,¹⁵ Hawaii,¹⁴ and Minnesota²⁰ in the United States, and Gifu²¹ in Japan. All studies included incident breast cancer cases diagnosed between 1994 and 2002 and controls representing the underlying population, and used quantitative methods to assess mammographic density.

The studies in this analysis were selected with the goal of including women with different ethnic backgrounds and a wide range in BMI. The California study included participants from two breast cancer studies that recruited cases from a population-based cancer registry and controls by random-digit dialing in Los Angeles County. The study population represented three different ethnic groups: Caucasians, African Americans, and Asian Americans, including Chinese, Filipino, and Japanese American women.¹⁵ The Hawaii study recruited cases and controls from an ongoing cohort that identified cases through a population-based cancer registry and included three different ethnic groups: Caucasians, Japanese Americans, and Native Hawaiians.¹⁴ Minnesota and Japan recruited women from mammography clinics: primarily Caucasians from the Mayo Clinic in Minnesota and Japanese from a general hospital in Gifu, respectively.^20,21

Eligibility criteria common to the four studies included the availability of a prior screening mammogram without suspicious lesions, no history of cancer except non-melanoma skin cancer, and no history of breast surgery, such as breast reduction and enlargement. Of the 4,248 eligible mammography images, the authors excluded women who were missing BMI or hormone replacement therapy (HRT) data (N = 127). The study population for this analysis included 4,121 subjects: California (N = 1,075), Hawaii (N = 1,274), Minnesota (N = 1,003), and Japan (N = 769). The Institutional Review Boards at all four institutions approved the original projects and this analysis.

Data collection

Each of the original four case-control studies collected detailed demographic, medical, screening, and reproductive information and self-reported anthropometric measures. In California, cases and controls were interviewed at their homes or offices using structured questionnaires.¹⁵ BMI was recalled five years before the date of diagnosis for cases or the date of the interview for controls. In Gifu, Japan, a self-administered questionnaire that included BMI was completed by the women at the time of the breast cancer screening.²¹ In Hawaii, an extensive questionnaire at cohort entry collected a wide range of information, including BMI,²² and, as part of the nested case-control study, a one-page breast health questionnaire elicited updated information on menopausal status, previous breast surgery, mammograms, and HRT use.¹⁴ In Minnesota, BMI and HRT use at the date of the mammogram were abstracted from medical records and other self-reported information was obtained from a clinical database.²⁰

Mammographic density assessment

Mammograms for the four case-control studies were taken between 1990 and 2003. For women with multiple mammograms available, the mammogram at the time of diagnosis (case) or reference date (control) was selected for each woman in California, Minnesota, and Japan, and the closest prediagnostic mammogram was selected for each woman in Hawaii. To assess the mammograms, all of the original studies except Japan used an interactive computer-assisted assessment method, either the Cumulus²³ or Madena²⁴ software program; Japan used an adaptive threshold technique.²⁵ To assure standardized density measurements across the different studies for this analysis, the original mammographic images from all four studies were read by a single reader (CGW) using Cumulus software.²³ For all studies, the contralateral image was used for cases and an equal proportion of left and right mammograms for controls. Craniocaudal (CC) views were read for all studies except Japan, for which only mediolateral oblique (MLO) views were available. Digital images used in the original studies were obtained for California, Hawaii, and Minnesota (Figure 1). For Japan, the digital images used in the original study were in an incompatible format for the Cumulus software. Therefore, the digital images were printed on films and digitized using a Kodak LS85 Digitizer (Eastman Kodak Company, Rochester, New York). The reader, who was blinded to case status, read 45 batches of 106 images with equal proportions of cases and controls and participants from each study in random order. The reader: 1) manually contoured a line to outline the breast area from the background and pectoralis muscle and 2) set a threshold based on grey level to delineate the radiologically dense from nondense tissue. The software calculated the number of pixels in each of these areas constituting the total and dense areas, respectively. The areas in pixels were converted to a common unit (cm²) based on the digitizer-specific pixel size (Figure 1).

Figure 1.

Summary of mammographic readings for the present study. The images from Japan were re-digitalized to assure consistency in the digital image formats from all four studies. To standardize mammographic density measurements, all mammographic images from the four individual studies were read using the same computer-assisted assessment method to calculate dense and total breast area.

To assess reliability of these measurements, 270 randomly selected mammograms were re-read twice: once within the same batch for a measure of intra-batch reliability and once across batches for a measure of inter-batch reliability. The intraclass correlation coefficient was 0.97 for percent density and 0.96 for the dense area in the breast. Two independent readers re-read images from both the random set of 270 (CB re-read 270; MJY re-read 82) as well as a separate sample of 24 images across all batches. The initial and repeat readings had excellent concordance for dense area (within-person Spearman correlation coefficients of dense area ranged from 0.79–0.96 between the readers and 0.79–0.92 between the readers and the four original studies) and for percent dense area (within-person Spearman correlation coefficients of percent density ranged from 0.75–0.96 between the readers and from 0.75–0.96 between the readers and the four original studies).

Statistical analyses

A standardized database was created using each study’s primary data. Percent density was modeled as a categorical variable based on tertiles of the control distribution (<20%, 20%–35%, >35%) or as a continuous variable rescaled and expressed per 10% increase in percent density. In a study of 30 premenopausal women the observed correlation coefficients between the CC and MLO image measurements were 0.96 for the breast area and 0.92–0.96 for breast density.²⁶ To correct for the slightly lower percent densities in MLO than the respective CC images, an estimated value for the MLO images was computed based on the formula MLO = 2.6 + 0.86 CC.²⁶ Because Asians experience a higher risk for hypertension and diabetes at lower levels of body mass index (BMI) compared to Caucasians,^27–28 women were classified as normal, overweight, or obese according to ethnic-specific BMI cut points (<23.0, 23.0–27.4, or ≥ 27.5 kg/m² for Asian women, and <25.0, 25.0–29.9, or ≥ 30.0 kg/m² for the other ethnic groups).²⁹ A study-ethnic variable was created for Caucasian, African American, Asian, and Other ethnic groups. Statistical computing was conducted using SAS version 9.2 (SAS Institute Inc., Cary, NC, USA), with a 2-sided p-value of <0.05 considered to be statistically significant.

The study population characteristics by the four studies and case status were compared using Cochran-Mantel-Haenszel χ² for categorical variables and analysis of variance (ANOVA) for continuous variables, applying Tukey’s Test for multiple comparisons and controlling for study effects as appropriate. Multivariable logistic regression models estimated odds ratios (OR) and 95% confidence intervals (95% CI) for breast cancer associated with percent density. All models were adjusted for the following covariates that are known to be associated with breast cancer³⁰ and mammographic density:^4,31 age at mammogram (continuous), BMI (continuous), study-ethnicity (California-Caucasian, California-African American, California-Asian, Hawaii-Caucasian, Hawaii-Asian, Hawaii-Other, Minnesota-Caucasian, Minnesota-Other, Japan-Asian), age at first live birth (no children, <26, 26–30, ≥ 31 years), number of children (0, 1–2, ≥ 3), postmenopausal HRT use (premenopausal, postmenopausal with no HRT use, postmenopausal with any HRT use), and family history of breast cancer (yes, no/unknown). Trend tests were performed by entering the categorical variable as a continuous parameter in the corresponding models.

Heterogeneity across studies and effect modification by BMI were assessed by the inclusion of an interaction term with percent density and study or BMI (continuous) and by separate models for individual studies or BMI groups. P-values were based on the Wald χ² test. Sensitivity analyses were performed including the exclusion of women at the extremes of the BMI scale. Analyses were repeated with area of dense breast tissue (cm²) as the outcome of interest; the variable was modeled as a categorical variable based on tertiles of the control distribution (<21, 21–36, >36 cm²) or as a continuous variable rescaled and expressed per 10% increase in dense area (cm²).

RESULTS

The study population (N = 4,121) included primarily postmenopausal (75%) women of diverse ethnicity: 45% Caucasians residing in Minnesota, California, and Hawaii, 39% Asians residing in Japan, Hawaii, and California, 9% African Americans residing in California, and 7% from other ethnic groups residing in Hawaii, primarily Native Hawaiians, and in Minnesota. Using ethnic-specific BMI cut points, 34% were classified as overweight and 19% as obese. Heterogeneity existed by age at mammogram, BMI, and reproductive factors across the four case-control studies (p_{heterogeneity} < 0.001 for all) (Table 1). Women in Minnesota and Hawaii were older (mean age 68 and 61 years, respectively), more likely to be postmenopausal, and had more children compared to those in California and Japan (mean age 49 and 50 years, respectively). Mean BMI was highest for women in Minnesota and lowest for women in Japan (p < 0.01). The use of HRT among postmenopausal women was uncommon in Japan (3%), whereas over half (60%) in Hawaii reported HRT use. Based on standardized mammographic data across studies, California had lower age-adjusted mean percent density compared to the other studies (p < 0.01). Although, California and Japan had similar dense area values, Japan had lower total breast area and subsequently higher percent density than California.

Table 1.

Study population characteristics¹ of the four case control studies N =4,121

	California		Hawaii		Minnesota		Japan		All
Characteristic	Cases	Controls	Cases	Controls	Cases	Controls	Cases	Controls	Cases	Controls
N	625	450	607	667	330	673	137	632	1,699	2,422
Ethnicity, %
Caucasian	44	51	32	28	96	96	0	0	46	44
Asian	24	15	54	47	2	1	100	100	36	42
African American	32	34	0	0	0	0	0	0	12	6
Other	0	0	14	25	2	3	0	0	5	8
Age at mammogram, years	48.9	49.2	61.4	59.8	67.7	67.5	53.1	49.3	57.4	57.2
Postmenopausal at mammogram, %	54	59	90	85	98	99	53	43	75	73
Time between mammogram and diagnosis, years	0.4	-- --	1.4	-- --	0.7	-- --	0.1	-- --	0.8	-- --
BMI, kg/m2, %2
Normal	56	49	49	44	32	33	57	60	49	46
Overweight	27	33	33	37	35	40	35	34	31	36
Obese	17	18	19	19	33	27	8	6	20	18
Body mass index, kg/m2	25.1	25.9	24.8	25.5	28.2	27.7	23.1	22.5	25.4	25.4
Family history of breast cancer, %	15	9	17	12	16	13	9	4	15	10
Parous, %	78	84	85	89	88	87	87	97	83	89
Age at first live birth, years	24.5	23.6	24.8	24.6	23.8	23.9	25.7	25.1	24.6	24.4
Children, N	2.0	2.2	2.4	2.6	2.7	2.8	1.9	2.3	2.3	2.5
Any postmenopausal HRT use3	40	56	60	60	16	25	3	3	40	37
Percent density4	27.6	23.4	35.3	30.3	31.9	29.1	31.9	31.1	31.7	28.5
Dense area (cm2)4,5	27.7	23.8	32.2	25.9	33.3	26.7	26.2	23.8	29.7	25.0
Total breast area (cm2)4,5	129.2	134.1	107.6	109.7	124.8	121.1	104.9	95.5	116.1	114.2

¹Mean values are provided unless indicated otherwise. Percentages may not add to 100 due to rounding.

²Overweight cut point ≥ 23 kg/m² for Asian women and ≥ 25 kg/m² for other ethnic groups. Obese cut point ≥ 27.5 kg/m² for Asian women and ≥ 30 kg/m² for other ethnic groups.

³Includes postmenopausal women only.

⁴All mammographic images were read using the same computer-assisted assessment method to standardize across studies. Presented with adjustment for age.

⁵Geometric mean.

Abbreviations: BMI: body mass index; HRT: hormone replacement therapy

Overall, cases were more likely to have a family history of breast cancer, to be nulliparous, and to have fewer children than controls. Age-adjusted mean percent density was significantly greater for cases than controls: 31.7% versus 28.5%, respectively (p < 0.001). BMI was inversely associated with breast density as expected on the basis of previous studies; the estimated age-adjusted mean percent density was 36.5%, 26.8%, and 19.3% in normal, overweight, and obese women, respectively (p_trend < 0.001).

Percent density was positively associated with breast cancer risk in multivariable analyses (Table 2). In separate models by study, the estimated OR per 10% increase in percent density was higher for Hawaii and Minnesota than California and Japan (1.20 and 1.23 compared to 1.13 and 1.04, respectively). Modest heterogeneity in the relation between percent density and breast cancer risk was observed across the four studies (p_{heterogeneity} = 0.08) and was greatly reduced when considering only the California, Hawaii, and Minnesota studies (p_{heterogeneity} = 0.99). In the total population, a 10% increase in percent density corresponded with an estimated OR of 1.16 for breast cancer (95% CI: 1.10, 1.21) that only slightly changed when the total population was modeled without Japan (OR: 1.18; 95% CI: 1.12, 1.24). Evidence of a dose response was observed given higher tertiles of percent density (p_trend < 0.001),

Table 2.

Breast cancer risk by percent density based on standardized mammographic data¹ across four case-control studies, N = 4,121

		Percent Density, % (tertiles)2
		< 20		20–35		>35		10% Increase2
Study		Cases/ Controls	OR	Cases/ Controls	OR (95% CI)	Cases/ Controls	OR (95% CI)	Cases/ Controls	OR (95% CI)
California		169/175	1.00	195/119	1.58 (1.14, 2.20)	261/156	1.56 (1.11, 2.19)	625/450	1.13 (1.03, 1.22)
Hawaii		150/239	1.00	203/198	1.85 (1.36, 2.51)	254/230	2.08 (1.50, 2.86)	607/667	1.20 (1.11, 1.29)
Minnesota		126/314	1.00	111/218	1.50 (1.08, 2.08)	93/141	2.13 (1.46, 3.12)	330/673	1.23 (1.11, 1.37)
Japan		34/145	1.00	32/205	0.92 (0.52, 1.64)	71/282	1.99 (1.08, 3.66)	137/632	1.04 (0.91, 1.19)
Four studies		479/873	1.00	541/740	1.55 (1.30, 1.84)	679/809	1.91 (1.58, 2.30)	1,699/2,422	1.16 (1.10, 1.21)
Three studies without Japan		445/728	1.00	509/535	1.65 (1.38, 1.98)	608/527	1.90 (1.56, 2.31)	1,562/1,790	1.18 (1.12, 1.24)
p value3									0.08
p value4									0.99

¹Mammographic images were read using the same computer-assisted assessment method to standardize density measurements across studies.

²Estimated odds ratio (95% confidence interval) adjusted for body mass index, age at mammogram, age at first live birth, number of children, menopausal status, hormone replacement therapy use, family history of breast cancer, and study-ethnicity as a combined variable.

³p value for heterogeneity across the four studies.

⁴p value for heterogeneity across the California, Hawaii, and Minnesota studies.

Relative to <20 percent density, the ORs for >35 were similar across BMI level whereas the OR for 20–35 was slightly higher in overweight (OR = 1.69) and obese (OR = 1.62) than in normal weight women (OR = 1.49) (Table 3). Similarly, slightly higher estimated ORs per 10% increase in percent density were observed for overweight and obese than normal weight participants (1.18 and 1.25 compared to 1.14, respectively) (p_interaction = 0.06). Estimated ORs were similar in analyses that used the same BMI cut points (<25, 25–29.9, ≥ 30 kg/m²) across ethnic groups (data not shown). However, the modification by BMI of the effect of percent density was weaker in subgroup analyses by menopausal status at the mammographic assessment (p_interaction = 0.32 for premenopausal and p_interaction = 0.19 for postmenopausal women) and by ethnicity (p_interaction = 0.54 for Asian and p_interaction = 0.50 Caucasian). Furthermore, the modification by BMI of the effect of percent density was weaker in analyses without Japan (p_interaction = 0.23), in analyses that excluded HRT users (p_interaction = 0.10), and in analyses that excluded 248 (6%) women at the extremes of the BMI scale (<18.5 BMI or >38 BMI) (p_interaction = 0.49) (data not shown). Dense breast area was associated with breast cancer risk to a similar degree; the estimated overall OR was 1.11 (95% CI: 1.07, 1.14) per 10 cm² increase in density in the total population with similar estimates across studies (p_{heterogeneity} = 0.29) and across BMI levels (p_interaction = 0.35) (Table 4). The joint effects of percent density or dense breast area and BMI level on the estimated ORs provide limited evidence of multiplicative interaction (Table 5).

Table 3.

Breast cancer risk by percent density and BMI level based on standardized mammographic data¹, N = 4,121

		Percent Density, % (tertiles)2
		< 20		20–35		>35		10% Increase2
BMI level		Cases/ Controls	OR	Cases/ Controls	OR (95% CI)	Cases/ Controls	OR (95% CI)	Cases/ Controls	OR (95% CI)
All
Normal		126/228	1.00	241/332	1.49 (1.11, 2.01)	464/551	1.96 (1.46, 2.64)	831/1,111	1.14 (1.06, 1.21)
Overweight		164/361	1.00	203/302	1.69 (1.28, 2.24)	164/215	1.87 (1.37, 2.56)	531/878	1.18 (1.08, 1.28)
Obese		189/284	1.00	97/106	1.62 (1.12, 2.33)	51/43	1.94 (1.19, 3.15)	337/433	1.25 (1.10, 1.41)
p value3									0.06
Postmenopausal
Normal		106/194	1.00	187/257	1.45 (1.04, 2.01)	286/271	2.25 (1.61, 3.14)	579/722	1.20 (1.11, 1.30)
Overweight		142/330	1.00	164/228	1.84 (1.36, 2.49)	116/125	2.32 (1.63, 332)	422/683	1.31 (1.18, 1.45)
Obese		167/245	1.00	77/88	1.50 (1.01, 2.24)	37/33	1.86 (1.06, 3.25)	281/366	1.24 (1.08, 1.43)
p value3									0.19
Premenopausal4
Normal		20/34	1.00	54/74	1.72 (0.80, 3.70)	177/276	1.47 (0.73, 3.00)	251/384	1.01 (0.89, 1.13)
Overweight		22/31	1.00	37/73	1.14 (0.50, 2.58)	48/90	0.97 (0.44, 2.11)	107/194	0.94 (0.79, 1.12)
Obese		20/39	1.00	20/18	3.96 (1.10, 14.24)	13/10	2.91 (0.74, 11.43)	53/67	1.26 (0.89, 1.79)
p value3									0.32
Asian									0.32
Normal		35/72	1.00	81/160	0.91 (0.52, 1.59)	195/318	1.42 (0.82, 2.45)	311/550	1.08 (0.98, 1.20)
Overweight		50/127	1.00	87/135	1.94 (1.20, 3.14)	78/109	2.35 (1.39, 3.99)	215/371	1.20 (1.03, 1.38)
Obese		44/51	1.00	33/27	1.54 (0.70, 3.37)	9/13	0.71 (0.24, 2.16)	86/91	1.09 (0.83, 1.43)
p value3									0.54
Caucasian
Normal		72/138	1.00	127/126	2.21 (1.48, 3.30)	209/174	2.62 (1.75, 3.90)	408/438	1.23 (1.12, 1.35)
Overweight		69/180	1.00	79/132	1.74 (1.14, 2.64)	61/71	2.37 (1.46, 3.83)	209/383	1.29 (1.13, 1.47)
Obese		110/164	1.00	38/56	1.22 (0.73, 2.04)	23/20	1.87 (0.94, 3.75)	171/240	1.24 (1.04, 1.49)
p value3									0.50

¹Mammographic images were read using the same computer-assisted assessment method to standardize density measurements across studies.

²Estimated odds ratio (95% confidence interval) adjusted for body mass index, age at mammogram, age at first live birth, number of children, menopausal status, hormone replacement therapy use, family history of breast cancer, and study-ethnicity as a combined variable.

³p value for interaction with body mass index (continuous).

⁴Minnesota was not included because of small sample of premenopausal women (N = 12).

Table 4.

Breast cancer risk by dense breast area and BMI level based on standardized mammographic data¹, N = 4,121

		Dense Area, cm2 (tertiles)2
		< 21		21–36		>36		10% Increase2
Obesity level		Cases/ Controls	OR	Cases/ Controls	OR (95% CI)	Cases/ Controls	OR (95% CI)	Cases/ Controls	OR (95% CI)
Overall		441/819	1.00	535/785	1.36 (1.15, 1.62)	723/818	1.77 (1.49, 2.09)	1,699/2,422	1.11 (1.07, 1.14)
BMI level
Normal		181/301	1.00	273/425	1.11 (0.86, 1.44)	377/385	1.77 (1.36, 2.30)	831/1,111	1.14 (1.08, 1.20)
Overweight		137/326	1.00	176/254	1.76 (1.30, 2.37)	218/298	1.88 (1.40, 2.53)	531/878	1.09 (1.04, 1.15)
Obese		123/192	1.00	86/106	1.51 (1.02, 2.24)	128/135	1.61 (1.13, 2.32)	337/433	1.12 (1.05, 1.18)
p value3									0.35

¹Mammographic images were read using the same computer-assisted assessment method to standardize density measurements across studies.

²Estimated odds ratio (95% confidence interval) adjusted for body mass index, age at mammogram, age at first live birth, number of children, menopausal status, hormone replacement therapy use, family history of breast cancer, and study-ethnicity as a combined variable.

³p value for interaction with body mass index (continuous).

Table 5.

Breast cancer risk by the joint effects of breast density and BMI level based on standardized mammographic data¹, N = 4,121

	Density Measurement (tertiles)2,3
	T1		T2		T3
BMI level	Cases/ Controls	OR (95% CI)	Cases/ Controls	OR (95% CI)	Cases/ Controls	OR (95% CI)
	Percent Density,%
Normal	126/228	1.00	241/332	1.45 (1.08, 1.94)	464/551	1.85 (1.41, 2.44)
Overweight	164/361	0.84 (0.61, 1.16)	203/302	1.40 (1.02, 1.93)	164/215	1.59 (1.13, 2.23)
Obese	189/284	1.17 (0.75, 1.83)	97/106	1.88 (1.16, 3.02)	51/43	2.12 (1.20, 3.74)
	Dense Area, cm2
Normal	181/301	1.00	273/425	1.11 (0.86, 1.44)	377/385	1.79 (1.39, 2.31)
Overweight	137/326	0.72 (0.53, 0.99)	176/254	1.27 (0.93, 1.74)	218/298	1.35 (1.00, 1.82)
Obese	123/192	1.11 (0.71, 1.76)	86/106	1.66 (1.03, 2.70)	128/135	1.70 (1.07, 2.67)

¹Mammographic images were read using the same computer-assisted assessment method to standardize density measurements across studies.

²Estimated odds ratio (95% confidence interval) adjusted for body mass index, age at mammogram, age at first live birth, number of children, menopausal status, hormone replacement therapy use, family history of breast cancer, and study-ethnicity as a combined variable.

³Tertiles for percent density <20, 20–35, >35%, respectively and for dense breast area <21, 21–36, >36 cm², respectively.

DISCUSSION

In this analysis comprised of ethnically diverse pre- and postmenopausal women from four case-control studies, women with more than 35 percent density had a 91% higher breast cancer risk than women with less than 20 percent density. Risk estimates in relation to percent density were heterogeneous across studies, with stronger estimates in Hawaii and Minnesota than California and Japan. The estimated breast cancer risk per 10% increase in percent density was slightly higher for overweight and obese women than those with normal BMI. Although the formal test for interaction was marginally significant, any evidence of modification of risk by BMI was weakened in sensitivity analyses that included only the California, Hawaii, and Minnesota studies; in analyses that excluded women at the extremes of the BMI scale; and in separate analyses by menopausal status and ethnicity. Furthermore, BMI did not appear to modify the effect of dense breast area on breast cancer risk. Our findings indicate little, if any, modification by BMI of the effects of breast density on breast cancer risk.

Modification by BMI of breast density in relation to breast cancer risk has been suggested by other studies with smaller sample sizes^11–15 and only a few have reached statistical significance.^11,15 Direct comparison with some of these studies^12–13 is difficult because breast density was classified by different mammographic assessment methods, in particular, parenchymal patterns¹³ and a computer program that uses the pectoralis major muscle to establish the threshold¹² compared to the interactive-threshold technique used in this study. A study in primarily postmenopausal Chinese women (491 cases and 982 controls) observed statistically significant joint effects by BMI and percent density;¹¹ for example women with BMI ≥ 26.7 kg/m² and >75% percent density compared to BMI <26.7 kg/m² and <10% percent density had a 9-fold higher risk (OR = 9.53) consistent with multiplicative interaction. The distribution of percent density in our study was much lower and only 48 women (2%) had percent density >75%. Furthermore, in this study the suggestive modifying effects by BMI on percent density and breast cancer risk were weaker in Asian-specific models. Unfortunately, the present study is unable to eliminate the possibility of residual confounding by adiposity, especially since the suggestive effect modification by BMI was only observed with percent density and not with dense breast area, in which is weakly associated with BMI.^16,32 In a recent analysis, models with dense breast area were more predictive than models with percent density adjusted for non-dense area as a measure of adiposity.³³ This suggests that a measure of adiposity gives little or no extra information on risk other than that which is contained in the dense area. The authors did not report effect modification by this measure of adiposity.

Differences in compressed breast thickness during mammography may explain the effect modification by BMI. Breast tissue in different individuals may have similar projected dense areas on screen-film and digital mammography but differ substantially in thickness. In particular, the compressed breast thickness of heavier women is greater and results in decreased image contrast and less uniformity of the displayed breast tissue than that of thinner women.^19,34 This may explain the finding of Stuedal et al. of a weak association between breast density and breast cancer risk among women with large breasts while adjusting for BMI.³⁵ Specifically, the difference in this finding from that of the present study may be due to inaccurate measurements of dense tissue in fatty breasts that have a small proportion of dense tissue. Other possible explanations include differences in BMI ranges, ethnic compositions, and steroid hormone and adipokine patterns between the study populations.

Because mammographically dense tissue is hypothesized to reflect the proliferation of breast stroma and epithelium and subsequent susceptibility to genetic damage, breast cancer risk is likely related to quanity of dense tissue and more acurately measured by volume than projected area. Body size and body fat explain the largest amount of variation in percent density across individuals^36–37 due primarily to a positive association with nondense tissue.¹⁸ On the other hand, although volume measures have not been shown to improve risk predictions based on area measures,^38–40 volumetric methods may provide insight into the complex interrelationship of percent density, BMI, and breast cancer risk as higher dense volume has been observed in heavier women.⁴⁰

Several possible biologic mechanisms exist by which women with high BMI and high percent density may have additional risk for breast cancer. BMI is thought to elevate postmenopausal breast cancer risk by increasing the aromatization of androgens into estrogens,⁴¹ however, only limited evidence supports an association between circulating estrogens and breast density.^42–43 Alternative biological mechanisms include the obesity-related dysregulation of metabolic conditions (e.g., hyperinsulemia) and adipokines associated with the immune response (e.g., leptin, adiponectin) and the inflammatory response (e.g., tumor necrosis factor-α, interleukin-6) that may contribute to the upregulation of cellular proliferation pathways and an aggressive tumor phenotype.^9–10

The major strength of this study is the use of standardized mammographic data from four well-designed studies of ethnically diverse women with a wide range of BMI. Across studies, the mean BMI ranged from 23 kg/m² in Japan to 28 kg/m² in Minnesota, and Hawaii had the greatest within-study range in BMI (13–62 kg/m²). The design of the original studies allows the results to be widely generalizable. For example, the California, Japan, and Minnesota studies were closer to community-based than referral or high-risk populations, and the Hawaii study was nested within a cohort generalizable to different ethnic and social groups within the U.S. population.²² Further, assessing density with the same method by a single reader provided a high level of standardization despite differences in mammography techniques across studies. Other strengths include the availability of information on important confounders, such as parity and HRT status, generally at the time of the mammogram.

A notable limitation in the present study is the comparability of data across studies. For example, different clinical mammography techniques, including positioning and compression of the breast, could explain the heterogeneity in risk estimates across studies despite the standardization of density assessment. In particular, the Japanese mammograms differ from the other studies in many respects, including the use of MLO instead of CC views and the digital versus film mammography. Also, the heterogeneity among the study decreased in sensitivity analyses that included only the California, Hawaii, and Minnesota studies. A potential source of bias includes the non-responders within the four previous case-control studies, since the prevalence of dense breast area and obesity may have differed among respondents and non-responders. Another potential source of bias is the self-report of data; however, validity studies have concluded that the use of self-reported data does not necessarily bias analyses in epidemiological studies.⁴⁴ A potential source of confounding is the unreliability of information abstracted from medical records. It is possible that we were unable to observe a stronger modification of BMI because the original studies did not evaluate the most relevant etiologic time period. This limitation is unlikely because the combined study included women of diverse ages (range 25–93 years). The cross-section design, specifically the assessment of density from the mammogram at the time of diagnosis or the closest prediagnostic mammogram, may affect the etiological relation. However, this is unlikely, because in longitudinal studies the estimated risk associated with percent density were equally strong regardless of the timing of the mammographic assessment.^{14, 20} Other potential sources of confounding include the possibility that BMI ascertained years earlier could have changed substantially, especially at perimenopausal ages, and the confounding effect of extreme values of BMI.

Data from the present study add to the accumulating evidence of the modification by BMI of the effect of breast density on breast cancer risk. Further studies of the complex interrelation of breast density, BMI, and breast cancer risk are needed to understand the implications of the suggestive increased risk among overweight and obese women with high breast density, particularly when considering breast density as a predictor of individualized breast cancer risk.

Novelty and impact.

This paper offers little evidence in a sample of primarily postmenopausal women of diverse ethnicity that BMI modifies the association between breast density and breast cancer risk. The findings contribute to the available data on potential modifiers of the effects of percent density on breast cancer risk that are of public health importance, particularly when breast density is used as a predictor of individualized breast cancer risk.

Abbreviations

BMI

body mass index

CC

craniocaudal

CI

confidence interval

HRT

hormone replacement therapy

MLO

mediolateral oblique

OR

odds ratio