Breast Cancer Population Attributable Risk Proportions Associated with Body Mass Index and Breast Density by Race/Ethnicity and Menopausal Status

Body mass index (BMI) and breast density are well-established risk factors for breast cancer (1–4). Higher breast density is associated with increased risk in both premenopausal and postmenopausal women (5, 6). Overweight and obesity are associated with increased risk of postmenopausal breast cancer (3, 4). Most large studies and meta-analyses found an inverse association between BMI and a premenopausal diagnosis of breast cancer, including a 2018 study by the Premenopausal Breast Cancer Collaborative Group (7). However, a 2017 meta-analysis by Chen and colleagues showed no association between increased BMI and a premenopausal breast cancer diagnosis (3), and a 2019 systematic literature review and meta-analysis by the World Cancer Research Fund International found an overall inverse association between BMI and a premenopausal breast cancer diagnosis but noted variability across regions; although European studies showed inverse associations, North American studies showed nonsignificant or inverse associations, and Asian studies showed positive associations (8). Studies evaluating the association between premenopausal BMI and future breast cancer (including breast cancers that may occur after the menopausal transition) show mixed results (8–11).

Distributions of BMI and breast density vary by race/ethnicity and menopausal status. BMI is generally higher among Black and Hispanic women (12), which may be associated with lower breast density (13, 14). Breast density is well studied among Asian women, who have a high prevalence of heterogeneously/extremely dense breasts (15). However, most studies report results only for specific races/ethnicities or subgroups and for BMI or breast density separately, but not both (16–18).

Identifying risk factors that are both strongly associated with increased risk and are highly prevalent would help researchers understand potential causes of a large proportion of breast cancers and help develop personalized screening strategies and other interventions to improve early detection and prevention. Studies on breast cancer population attributable risk proportions (PARP) associated with BMI or breast density by menopausal status found that they accounted for large proportions of breast cancers (19–22), but these studies did not report results by race/ethnicity. We used U.S.-based Breast Cancer Surveillance Consortium (BCSC) data to estimate associations and PARPs of BMI and breast density with invasive breast cancer by race/ethnicity and menopausal status.

In this cohort study, we selected all screening and diagnostic mammography examinations between January 1, 1994, and December 31, 2016 among women ages 35–84 years at seven U.S.-based BCSC registries (www.bcsc-research.org): Carolina Mammography Registry, New Hampshire Mammography Network, Vermont Breast Cancer Surveillance System, San Francisco Mammography Registry, Kaiser Permanente Washington Registry, Metro Chicago Breast Cancer Registry, and New Mexico Mammography Project. We excluded observations on women with a personal history of invasive breast cancer, ductal carcinoma in situ (DCIS), mastectomy, or breast implants, with invasive breast cancer or DCIS diagnosed within 3 months of the mammogram to remove prevalent cases, or with unknown American College of Radiology Breast Imaging Reporting and Data System (BI-RADS; ref. 23) breast density, menopausal status, or BMI. Observations with multiple self-reported races or race/ethnicity other than the four under study (non-Hispanic White, non-Hispanic Black, Asian American, or Hispanic/Latina) were excluded because of small sample sizes. The final study cohort included 3,786,802 observations from 1,071,653 women with a median 5.2 years of follow-up, among whom 66,419 observations were from 21,253 women who were diagnosed with invasive breast cancer. Of these, 1,369,826 observations were from 518,852 premenopausal women who developed 7,337 breast cancers, and 2,416,976 observations were from 679,966 postmenopausal women who developed 15,256 breast cancers. Some women (N = 127,165) became postmenopausal during the study, contributing to both cohorts.

BCSC registries and the Statistical Coordinating Center received Institutional Review Board approval for active or passive consenting processes or a waiver of consent to enroll participants, link and pool data, and perform analysis. All procedures were Health Insurance Portability and Accountability Act compliant, and registries and the Coordinating Center received a Federal Certificate of Confidentiality and other protections for the identities of women, physicians, and facilities.

Our main outcome was a primary, invasive breast carcinoma diagnosis. Diagnoses of invasive breast carcinoma and DCIS were obtained by linkage with pathology databases and regional Surveillance, Epidemiology, and End Results or state and regional tumor registries.

At each mammography examination, self-reported information on age, race/ethnicity, height and weight, and menopausal status was obtained from questionnaires. BMI was calculated as weight in kilograms divided by height in meters squared (kg/m²) and categorized using modified World Health Organization cutoffs (24): underweight (<18.5 kg/m²), normal weight (18.5–24.9 kg/m²), overweight (25.0–29.9 kg/m²), obesity class I (30.0–34.4 kg/m²), and obesity class II/III (≥35.0 kg/m²). Observations were considered premenopausal at the time of examination if the woman reported continued menstrual periods, current use of hormonal birth control, or not knowing if periods had stopped permanently (N = 1,286,595), or were <45 years old in the absence of other information (N = 83,231; ref. 25). Observations were considered postmenopausal at the time of examination if the woman reported periods that stopped naturally, no period for more than 365 days, hysterectomy with bilateral oophorectomy, or current hormone therapy use (N = 2,406,567), or were 55 years or older in the absence of other information (N = 10,409; ref. 25). Menopausal status was considered missing otherwise (N = 325,220) and these observations were removed from the main analysis (25). We performed two sensitivity analyses categorizing women with missing menopausal status as premenopausal or postmenopausal based on two different age cutoffs of 50 and 52 as proxies for age at menopause (Supplementary Tables S1A–S1F; ref. 25). Radiologists recorded BI-RADS breast density as almost entirely fat, scattered fibroglandular densities, heterogeneously dense, or extremely dense (23).

Separate analyses were performed for observations on the premenopausal and postmenopausal cohorts. Because women may contribute multiple mammograms and the number contributed may be related to overall breast cancer risk, we calculated frequencies, proportions, and HRs by inversely weighting each woman's observations by the total number of mammograms she contributed to each of the premenopausal and postmenopausal cohorts (26).

We fit separate Cox proportional hazards regression models by race/ethnicity and menopausal status to estimate associations between BMI and breast density with invasive breast cancer. SEs were calculated using the robust sandwich variance estimator to account for correlation among multiple observations from the same woman (26). Follow-up began 3 months after each mammogram to remove prevalent cases. In sensitivity analysis, we found that HRs computed from starting follow-up at 6 months did not differ from HRs computed from starting follow-up at 3 months (Supplementary Table S2). Models adjusted for age as a categorical variable to be consistent with the PARP analysis described below, with age categorized into the following groups: 35–49, 50–59, 60–69, and 70–85 years. HR and confidence intervals (CI) were similar when adjusting for age as a continuous versus categorical variable. Models were stratified by BCSC registry to allow for separate baseline hazards. The proportional hazards assumption was tested using Schoenfeld residuals in goodness-of-fit tests (27) with all P > 0.10. In sensitivity analyses, we repeated analyses for Asians using previously suggested lower cutoff points for overweight and obesity (Supplementary Table S3; ref. 28). In sensitivity analyses, we also fit separate Cox models adjusting individually and simultaneously for first-degree family history of breast cancer, history of prior breast biopsy, age at first live birth, or postmenopausal hormone therapy use (Supplementary Tables S4A–S4E).

We calculated partial PARPs using a publicly available SAS macro (29) that did not allow use of continuous variables, so age was categorized into the same groups used for estimating the HRs. PARPs for breast density shifted the proportion of women in the heterogeneously or extremely dense categories to scattered fibroglandular densities, while holding the proportion of women in the almost entirely fat category unchanged. PARPs for BMI shifted the proportion of overweight and obese women to normal BMI, while holding the proportion of underweight women unchanged. Observed proportions used to calculate PARPs were computed as total person-time in each combination of age, BMI, and breast density categories to match the analysis used to compute HRs (29). We calculated Wald-based CIs for pairwise differences in PARPs using estimates of the SEs from the macro and assuming unequal variances.

Statistical analyses were performed using SAS/STAT version 14.2. Tests of statistical significance used a two-sided α = 0.05.

Age distributions were similar across racial/ethnic groups among premenopausal and postmenopausal women, although premenopausal Hispanic women tended to be younger than other races/ethnicities (Table 1). Black women had the highest proportion of premenopausal (82.8%) and postmenopausal (82.7%) overweight or obese women, followed by Hispanic (63.0% premenopausal, 67.6% postmenopausal), White (49.7% premenopausal, 58.3% postmenopausal), and Asian women (29.3% premenopausal, 34.9% postmenopausal). Asian women had the highest proportion of dense breasts among premenopausal (81.2%) and postmenopausal (54.7%) women, followed by White women (61.3% premenopausal, 38.2% postmenopausal), Hispanic women (56.6% premenopausal, 30.9% postmenopausal), and Black women (56.0% premenopausal, 32.1% postmenopausal).

Table 2 shows associations of BMI and breast density with invasive breast cancer stratified by race/ethnicity and menopausal status. Comparing obese II/III with normal BMI among premenopausal women, White women had an increased risk of breast cancer (HR = 1.18, 95% CI = 1.09–1.28) with a positive dose response (P_trend < 0.001). HRs associated with premenopausal BMI among Black, Asian, and Hispanic women were not statistically significant.

Postmenopausal overweight/obese BMI showed increased HR estimates for breast cancer across all races/ethnicities studied (P_trend for increasing BMI <0.001). Comparing obese II/III with normal BMI, associations were strongest for Asian women (HR = 2.21, 95% CI = 1.60–3.05) and moderate for Black women (HR = 1.76, 95% CI = 1.47–2.11). Associations were weakest, but still elevated, for White (HR = 1.43, 95% CI = 1.36–1.50) and Hispanic women (HR = 1.37, 95% CI = 1.13–1.66).

Premenopausal and postmenopausal breast density were moderate-to-strong breast cancer risk factors across all races/ethnicities (P_trend for increasing breast density <0.001). Comparing extremely dense breasts with scattered fibroglandular densities, Hispanic women had the strongest associations with a premenopausal HR of 2.43 (95% CI = 1.94–3.06) and a postmenopausal HR of 2.06 (95% CI = 1.67–2.54). Moderate-to-strong risk associated with breast density was observed in White women (premenopausal HR = 2.05, 95% CI = 1.91–2.19; postmenopausal HR = 1.62, 95% CI = 1.53–1.72) and Black women (premenopausal HR = 1.96, 95% CI = 1.39–2.76; postmenopausal HR = 1.69, 95% CI = 1.23–2.31). Asian women had the weakest associations with a premenopausal HR of 1.80 (95% CI = 1.43–2.26) and a postmenopausal HR of 1.49 (95% CI = 1.24–1.78).

Table 3 shows breast cancer PARPs associated with BMI and breast density by race/ethnicity and menopausal status. PARPs for BMI were larger for postmenopausal versus premenopausal women (12.0%–28.3% vs. 1.0%–9.9%). CIs for PARPs associated with premenopausal BMI among Black, Asian, and Hispanic women included zero and did not rule out a protective effect. The PARP associated with premenopausal BMI among White women was statistically significantly greater than zero but small (PARP = 3.7%, 95% CI = 0.6–6.7). Postmenopausal BMI accounted for a large proportion of breast cancers among Black women (PARP = 28.3%, 95% CI = 17.4–38.5), nearly double the PARPs of other races/ethnicities (12.0%–15.4%).

PARPs for breast density were larger for premenopausal versus postmenopausal women (23.9%–35.0% vs. 13.0%–16.7%). Premenopausal breast density accounted for roughly one third of breast cancers among Hispanic (PARP = 35.0%), Asian (PARP = 33.2%), and White (PARP = 30.4%) women, and 23.9% of breast cancers among Black women (PARP = 23.9%, 95% CI = 10.6–36.4). Postmenopausal breast density accounted for a similar, substantial proportion of breast cancers across all races/ethnicities (13.0%–16.7%).

Among premenopausal women, PARPs for reducing both high breast density and high BMI simultaneously were similar across races/ethnicities (31.7%–35.8%). PARPs for reducing both high breast density and high BMI among postmenopausal women were lower than premenopausal women and similar among White, Asian, and Hispanic women (23.8%–27.8%), but were larger among Black women (PARP = 40.9%, 95% CI = 28.2–52.2) driven by the large PARP associated with postmenopausal BMI among Black women.

Pairwise differences in PARPs across races/ethnicities did not show statistically significant differences except the PARP associated with postmenopausal BMI among Black women was significantly higher than for other races/ethnicities (difference in PARPs for Black versus White = 12.9%, 95% CI = 1.9–24.0; Black vs. Asian = 13.6%, 95% CI = 0.8–26.3; Black vs. Hispanic = 16.3%, 95% CI = 2.1–30.6).

Sensitivity analyses showed very little change in in the distributions of BMI and BI-RADS breast density by race/ethnicity and menopausal status when women with missing menopausal status were categorized as premenopausal or postmenopausal based on two different age cutoffs of 50 years and 52 years as proxies for age at menopause (Supplementary Tables S1A–S1F). HRs also showed very little change except for moderate changes in the lowest BMI category of underweight and only among Black and Hispanic women where we note that sample sizes are smallest, and the CIs are very wide. PARPs also showed very little change with most changing less than 1% and none more than 3%.

In sensitivity analysis using more conservative BMI cutoff points suggested for Asian women (28), we did not find meaningful differences in HRs (Supplementary Table S3). However, lower cutoff points shifted 21.3% of premenopausal Asian women from normal BMI to overweight and 6.8% from overweight to obese I, and shifted 22.5% of postmenopausal Asian women from normal to overweight and 8.3% from overweight to obese I. Using lower cutoff points for Asian women increased their premenopausal BMI PARP from 3.3% (95% CI = −5.1 to 11.7) to 4.7% (95% CI = −6.7 to 16.0) and postmenopausal BMI PARP from 14.7% (95% CI = 8.1–21.2) to 18.6% (95% CI = 10.0–27.0).

Sensitivity analysis showed very limited evidence of confounding (i.e., HRs changed by <10.0%) when adjusting individually for age at first live birth, history of prior breast biopsy, use of hormone replacement therapy, and first-degree family history of breast cancer (Supplementary Tables S4a–S4d). When adjusting for all confounders simultaneously, some HRs in the highest BI-RADS breast density categories changed by approximately 10%, but we note that these groups have relatively small sample sizes, particularly among Black and Hispanic women, and changes were smaller for White women where sample sizes were largest (Supplementary Table S4E).

Our study is the first to directly investigate and report important racial/ethnic differences in PARPs associated with BMI and breast density among premenopausal and postmenopausal White, Black, Hispanic, and Asian women. Across all racial/ethnic groups examined, breast density was a moderate-to-strong risk factor with a clear dose response for increasing breast density that accounted for statistically and clinically significant proportions of invasive breast cancer among both premenopausal (23.0%–35.0%) and postmenopausal (13.0%–16.7%) women. Premenopausal BMI was not significantly associated with future breast cancer risk except for a small effect in White women; however, postmenopausal BMI was a strong risk factor with a clear dose response for increasing BMI for all races examined. If all overweight/obese postmenopausal women achieved a normal BMI, breast cancer incidence could be reduced by 12%–15% in White, Asian, and Hispanic women and 28% in Black women.

Our findings are consistent with other studies in identifying breast density as a strong and prevalent risk factor for breast cancer and postmenopausal BMI as accounting for a large proportion of breast cancers (19–22, 30, 31). Our findings that HRs for premenopausal BMI showed small, but statistically significant, increased risk of future breast cancer among White women, but no significant differences among Black, Asian, and Hispanic women, likely differ from most prior studies showing strong, inverse associations between BMI and premenopausal breast cancer (7, 8) because our study evaluated premenopausal BMI and future breast cancer, including cancers that may occur after the menopausal transition.

Both premenopausal and postmenopausal PARPs associated with breast density were generally similar across races/ethnicities. However, the corresponding risk-factor prevalences and HRs show important differences across races/ethnicities. Although Hispanic women had lower proportions of dense breasts, Hispanic women had the strongest associations between premenopausal breast density and breast cancer risk. In contrast, Asian women had the largest proportions of dense breasts but modest associations between breast density and breast cancer risk. Screening strategies that consider both breast density and risk may be particularly important for Hispanic and Asian women and risk prediction models should take into account the different prevalences and magnitudes of association with breast density across races/ethnicities when calculating absolute risk.

Postmenopausal BMI was a strong risk factor for all race/ethnicities examined but was a particularly important risk factor among Black women. Black women had the highest prevalences of overweight and obesity, and strong associations between postmenopausal BMI and breast cancer risk, leading to almost twice the PARP as other races/ethnicities. Postmenopausal Asian women had the strongest associations between BMI and breast cancer risk, but the lowest prevalences of overweight/obesity resulting in similar PARPs as White and Hispanic women. In contrast, postmenopausal Hispanic women had the second highest prevalences of overweight/obesity but the weakest associations between BMI and breast cancer, resulting in PARPs similar to Asian and White women. By 2030, overall obesity (categories I/II/III) and severe obesity (categories II/III) are projected to rise to 49% and 24%, respectively, with severe obesity becoming the most prevalent BMI category among women and the highest prevalence of obesity projected among Black adults, followed by Hispanic and White adults (32). While prevention efforts to avoid and reduce overweight and obesity should be a focus in all women, reducing weight in postmenopausal Black women could result in the largest reductions in breast cancer risk.

As a potentially modifiable risk factor, BMI is an attractive target for intervention and risk reduction. Sustained weight loss in women ages 50 and older has been shown to reduce breast cancer risk (33). In addition, changes in lifestyle factors such as increased vigorous physical activity and exercise, maintaining health body weight and body composition, breast feeding, reduced alcohol consumption, and increased consumption of fruits and vegetables are all associated with BMI (24, 34) and all are associated with reduced breast cancer risk (35). As an extreme example, dramatic weight loss associated with bariatric surgery among severely obese women has been shown to reduce overall and estrogen receptor–positive breast cancer risk (36, 37).

Breast density reduction is also associated with reduced risk of breast cancer (38). Reduced breast cancer risk following weight loss or bariatric surgery may be due, in part, to reductions in total volume of dense breast tissue, although results for other breast density measures are conflicting (39–41). Alcohol consumption increases breast cancer risk (35) but the effects on breast density remains unclear (42). Preventive tamoxifen reduces breast density with the largest reductions in women under 45 years, in whom density is highest (43), and reduces breast cancer risk in clinical trials by approximately one third (44) among women who experienced at least a 10% density reduction (45). Studies on aromatase inhibitors (AI) and breast density reduction have mixed results (46–48). One recent study found that AIs were associated with larger volumetric percent density reductions among postmenopausal women than tamoxifen, whereas tamoxifen was associated with larger density reductions among premenopausal women than AIs (49). Studies of gonadotropin-releasing hormone agonist (GnRHA) show promise in chemoprevention and treatment through ovarian hormone suppression and may be associated with reduced breast density (50–52). However, preventative tamoxifen, AIs, and GnRHA use are generally not prescribed for women of average risk and use is low among high-risk women due to side effects (51, 52). BMI is inversely associated with qualitative measures of breast density, such as BI-RADS breast density, but is not cross-sectionally associated with some quantitative measures such as dense tissue volume (15). Weight loss impacts breast density by decreasing overall breast volume and volume of fibroglandular tissue, which leads to percent fibroglandular tissue remaining unchanged or in some cases increasing (39). As a result, BI-RADS or other qualitative measures of density may not show the impact of weight loss on decreased volume of dense tissue (39, 53). Given limited strategies to reduce breast density in otherwise average-risk women, secondary prevention efforts for women with dense breasts should focus on identifying those at high risk of advanced breast cancer who might benefit from supplemental screening (54).

Some studies show that Asians are at higher risk of weight-related diseases than Whites at similar BMI or weight-gain levels (55), perhaps due to higher percent body fat at similar BMI levels (56). Using more conservative BMI cutoff points suggested for Asians (28) did not meaningfully change the HRs for breast cancer, but increased the proportion of overweight/obese women which slightly increased the PARPS.

Study limitations include inability to evaluate differential effects of BMI and breast density by country of ancestry, place of birth, or acculturation, which may modify breast cancer risk in subgroups of races/ethnicities (57–61). Even with very large study cohorts and multiple observations per woman, some estimated CIs were wide due to small samples, for example in the highest obesity categories among Asian women and the highest breast density categories among Black and Hispanic women. We were unable to evaluate quantitative measures of breast density; however, BI-RADS breast density is the most collected density measure in clinical practice in the United States and used in breast cancer risk prediction models (62–64). We did not evaluate potential interactions between BMI and BI-RADS breast density due to the added complexity given results are subdivided by menopausal status and race/ethnicity (8 separate subgroups) and concerns about lack of power for detecting interactions due to relatively small sample sizes for all but White women. However, a prior study found no significant interactions between BMI and BI-RADS breast density among premenopausal nor postmenopausal women, suggesting any interactions may be small (65). Although BMI is not a perfect measure of adiposity, it was readily available in the medical record. We were not able to analyze other measures such central adiposity, waist-to-hip ratios, or visceral versus subcutaneous fat distribution because these measures were not available in the medical record.

Study strengths include the prospective BCSC cohort, which is broadly representative of the U.S. population, has larger sample sizes than other studies for the four largest U.S. races/ethnicities, links women to state and/or regional tumor registries for near complete capture of breast cancer diagnoses, and covers the spectrum of breast imaging facilities from mobile vans to university hospitals. We were able to include women receiving either screening or diagnostic mammograms to improve the generalizability of our population. The large BCSC cohort allowed evaluation of both breast density and BMI by race/ethnicity and menopausal status.

We found that overweight/obesity and dense breasts accounted for a large proportion of breast cancers in White, Black, Hispanic, and Asian women despite large differences in risk-factor distributions and variation in associations with breast cancer. Breast density was a strong and important risk factor for breast cancer in both premenopausal and postmenopausal women of all races/ethnicities examined, with larger effects in premenopausal women. BMI was a strong risk factor among postmenopausal women, especially in Black women, but not among premenopausal women. Primary and secondary prevention efforts and risk prediction models should consider racial/ethnic differences in risk associations and PARPs for breast density and BMI.

M.C.S. Bissell reports grants from NCI and PCORI during the conduct of the study. K. Kerlikowske reports grants from NCI and PCORI during the conduct of the study, as well as nonpaid consulting work for the STRIVE study conducted by GRAIL. B.L. Sprague reports grants from NIH and PCORI during the conduct of the study. J.A. Tice reports grants from NCI during the conduct of the study. C.C. Gard reports personal fees from Kaiser Permanente Washington Health Research Institute (biostatistical consulting) during the conduct of the study. L.M. Henderson reports grants from NIH/NCI during the conduct of the study. T. Onega reports grants from NCI and PCORI during the conduct of the study. D.L. Miglioretti reports grants from NCI and PCORI during the conduct of the study. No potential conflicts of interest were disclosed by the other authors.

The funding agencies had no role in the study's design; the collection, analysis, or interpretation of the data; the writing of the manuscript; or the decision to submit the manuscript for publication. The content is solely the responsibility of the authors and does not necessarily represent the official views of the NIH Office of Research on Women's Health, the NCI, or the Patient-Centered Outcomes Research Institute (PCORI), its Board of Governors, or Methodology Committee.

M.C.S. Bissell: Conceptualization, data curation, software, formal analysis, investigation, visualization, methodology, writing–original draft, project administration, writing–review and editing. K. Kerlikowske: Conceptualization, supervision, funding acquisition, investigation, methodology, writing–review and editing. B.L. Sprague: Investigation, methodology, writing–review and editing. J.A. Tice: Investigation, writing–review and editing. C.C. Gard: Investigation, writing–review and editing. K.Y. Tossas: Investigation, writing–review and editing. G.H. Rauscher: Investigation, writing–review and editing. A. Trentham-Dietz: Investigation, writing–review and editing. L.M. Henderson: Investigation, writing–review and editing. T. Onega: Investigation, writing–review and editing. T.H.M Keegan: Supervision, investigation, methodology, writing–review and editing. D.L. Miglioretti: Conceptualization, formal analysis, supervision, funding acquisition, visualization, methodology, writing–original draft, project administration, writing–review and editing.

This research was funded by the NCI and the NIH Office of Research on Women's Health (3P01CA154292-07S1, to M.C.S. Bissell, K. Kerlikowske, and D.L. Miglioretti). Data collection for this work was supported by funding from the NCI (P01CA154292, to K. Kerlikowske, B.L. Sprague, G.H. Rauscher, A. Trentham-Dietz, L.M. Henderson, T. Onega, and D.L. Miglioretti; U54CA163303, to B.L. Sprague); the Patient-Centered Outcomes Research Institute (PCS-1504-30370, to K. Kerlikowske, B.L. Sprague, G.H. Rauscher, A. Trentham-Dietz, L.M. Henderson, T. Onega, D.L. Miglioretti); and the Agency for Healthcare Research and Quality (R01 HS018366-01A1, to G.H. Rauscher). Additional resources were funded by the UC Davis Comprehensive Cancer Center Support Grant awarded by the NCI (P30CA093373). The collection of cancer and vital status data used in this study was supported in part by several state public health departments and cancer registries throughout the United States. For a full description of these sources, please see: http://www.bcsc-research.org/work/acknowledgement.html.

The authors thank Chris Tachibana, PhD, who provided scientific editing, and Deborah Seger, who provided data management, as part of their positions at Kaiser Permanente Washington Health Research Institute and did not receive additional compensation besides their salary. The authors also thank the participating women, mammography facilities, and radiologists for the data they have provided the Breast Cancer Surveillance Consortium (BCSC) for this study. You can learn more about the BCSC at http://www.bcsc-research.org/.

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.