Abstract
Background:
In 30 states, women who have had screening mammography are informed of their breast density on the basis of Breast Imaging Reporting and Data System (BI-RADS) density categories estimated subjectively by radiologists. Variation in these clinical categories across and within radiologists has led to discussion about whether automated BI-RADS density should be reported instead.
Objective:
To determine whether breast cancer risk and detection are similar for automated and clinical BI-RADS density measures.
Design:
Case-control.
Setting:
San Francisco Mammography Registry and Mayo Clinic.
Participants:
1609 women with screen-detected cancer, 351 women with interval invasive cancer, and 4409 matched control participants.
Measurements:
Automated and clinical BI-RADS density assessed on digital mammography at 2 time points from September 2006 to October 2014, interval and screen-detected breast cancer risk, and mammography sensitivity.
Results:
Of women whose breast density was categorized by automated BI-RADS more than 6 months to 5 years before diagnosis, those with extremely dense breasts had a 5.65-fold higher interval cancer risk (95% CI, 3.33 to 9.60) and a 1.43-fold higher screen-detected risk (CI, 1.14 to 1.79) than those with scattered fibroglandular densities. Associations of interval and screen-detected cancer with clinical BI-RADS density were similar to those with automated BI-RADS density, regardless of whether density was measured more than 6 months to less than 2 years or 2 to 5 years before diagnosis. Automated and clinical BI-RADS density measures had similar discriminatory accuracy, which was higher for interval than screen-detected cancer (C-statistics: 0.70 vs. 0.62 [P < 0.001] and 0.72 vs. 0.62 [P < 0.001], respectively). Mammography sensitivity was similar for automated and clinical BI-RADS categories: fatty, 93% versus 92%; scattered fibroglandular densities, 90% versus 90%; heterogeneously dense, 82% versus 78%; and extremely dense, 63% versus 64%, respectively.
Limitation:
Neither automated nor clinical BI-RADS density was assessed on tomosynthesis, an emerging breast screening method.
Conclusion:
Automated and clinical BI-RADS density similarly predict interval and screen-detected cancer risk, suggesting that either measure may be used to inform women of their breast density.
Thirty states have laws requiring that women receive some level of notification of breast density (1). The Breast Imaging Reporting and Data System (BI-RADS) breast density categories (2), estimated subjectively by radiologists, is the standard for reporting breast density in the United States. Language regarding notification varies by state, with 10 states providing BI-RADS density information to all women and 20 notifying only those whose breasts are categorized as dense (heterogeneously or extremely dense). About 50% of women who have screening mammography have dense breasts (3–5), which may result in decreased cancer detection and increased cancer risk, leading several states to advise women to talk to their providers about whether supplemental screening is right for them (3).
Concern has been raised about using clinical BIRADS breast density for prevention strategies, calling into question the subjectivity and reproducibility of the measure for individual women. Recent studies of interand intrarater reliability of the BI-RADS categories have reported moderate to substantial agreement (6–9). In clinical practice, 17.2% of women with consecutive mammograms interpreted by different radiologists had discordant BI-RADS density ratings of dense versus nondense, compared with 10.0% who had consecutive mammograms interpreted by the same radiologist (10). The variation in BI-RADS density interpretations within and across radiologists has clinical implications, because breast density assessment may lead to recommendations for supplemental imaging (3), affect risk assessment (11), and guide screening frequency (12).
Automated breast density measures are available with commercial software (Quantra [Hologic], Volpara [Volpara Solutions], PowerLook Density Assessment [iCAD]) to assess automated BI-RADS and volumetric density on digital mammography. Studies have shown that automated and clinical BI-RADS density measures have similar associations with overall cancer risk (13–15). One study conducted in the Netherlands examined whether automated breast density measured with Vol para software predicts cancer detection, defining interval cancer as invasive cancer occurring within 24 months of a negative screening result (15). Wanders and colleagues (15) found that automated dense breast volume, percentage of dense volume, and BI-RADS density were more strongly associated with interval than screen-detected cancer, compared with women who did not develop breast cancer. No study has examined whether automated and clinical BI-RADS density measures similarly predict screen-detected and interval invasive breast cancer risk compared with women who do not develop breast cancer. If automated BI-RADS density measures, which are reportedly more reproducible than clinical measures on repeated examinations (16, 17), can accurately predict cancer detection, automated breast density assessment might be used more widely for breast cancer prevention strategies.
We determined screen-detected and interval invasive breast cancer risk and mammography sensitivity for clinical and automated BI-RADS density measures according to the length of time between density assessment and breast cancer diagnosis.
METHODS
Study Sample
Study participants were from 2 case-control studies nested within large prospective breast imaging cohorts. The San Francisco Mammography Registry (SFMR) participates in the National Cancer Institute-funded Breast Cancer Surveillance Consortium (BCSC) (http://www.bcsc-research.org/index.html) (18). The SFMR obtains annual institutional review board approval and passive permission for data collection and participant enrollment, as well as data linkages for research purposes, and received a federal Certificate of Confidentiality that protects the identities of research participants. For the Mayo Clinic screening cohort, the institutional review board approved a waiver of informed consent and Health Insurance Portability and Accountability Act authorization from the participants. Only persons who had not refused permission to use their medical records for research (according to Minnesota Research Authorization) were included in the Mayo Clinic cohort (19).
The SFMR obtained “for-processing” digital screening examinations from Hologic Selenia machines at 4 facilities since 2006, which served as the underlying imaging cohort. Annual linkage to the California Cancer Registry identified cases of incident invasive breast cancer reported from January 2007 through May 2014. Raw digital screening examinations performed more than 6 months to 5 years before diagnosis (n = 1312) were included for case participants. Two control participants (n = 2603) without previous breast cancer or breast implants were selected from the SFMR imaging cohort and matched to each case participant by age within 5 years, race, date of screening examination within 1 year, mammography machine, and facility. For the Mayo Clinic cohort, for-processing digital images were collected from women in the tristate region of Minnesota, Iowa, and Wisconsin; the images were obtained from Hologic Selenia machines at 1 facility from March 2008 through September 2014. Annual linkage to the Mayo Clinic tumor registry identified cases of incident invasive breast cancer reported through December 2015 (n = 648). Approximately 3 control participants (n = 1806) without previous breast cancer or breast implants were selected from the Mayo imaging cohort and matched to each case participant by age within 5 years, race, state of residence, date of screening examination within 1 year, and mammography machine. We ensured that all control participants had at least 1 normal screening mammogram on or after their corresponding matched case participants’ diagnosis dates.
Interval cancer was defined as invasive breast cancer occurring within 12 months of a negative mammography result (BI-RADS 1 or 2). Screen-detected cancer was defined as invasive cancer occurring within 12 months of a positive mammography result (BI-RADS 0, 4, or 5).
Measurement of Risk Factors
Age, first-degree family history of breast cancer, race/ethnicity, breast biopsy history, height, and weight were obtained from self-report at the time of mammography for the SFMR cohort and from self-report or medical record review (height and weight) for the Mayo cohort. Body mass index was calculated by dividing weight in kilograms by height in square meters (kg/m2). Race/ethnicity was coded by using the expanded definitions currently used in the SEER (Surveillance, Epidemiology, and End Results) program and U.S. vital statistics (non-Hispanic white, non-Hispanic black, Asian/ Pacific Islander, American Indian/Alaska Native, Hispanic, other/mixed race). We calculated the BCSC, version 1.0, 5-year risk score at the time of mammography, which estimates the probability of invasive breast cancer occurring within the next 5 years on the basis of age, race, ethnicity, family history, history of breast biopsy, and clinical BIRADS breast density (20).
Clinical and Automated BI-RADS Density
Practicing radiologists classified breast density as part of routine clinical practice at the time of mammography interpretation by using the BI-RADS density categories (2): (a), almost entirely fatty; (b), scattered fibroglandular densities; (c), heterogeneously dense; and (d), extremely dense.
Volpara, version 1.5.3, the most commonly used 3-dimensional density measure in clinical practice and research settings, is a fully automated method for assessing volumetric breast density. It uses the measured breast thickness and x-ray attenuations in the for-processing image to create estimates of dense and nondense tissue volume for each pixel. Summing the dense pixel volumes provides total dense breast volume. Volpara uses proprietary algorithms to calculate breast thickness and determine dense tissue volume by averaging measures of each breast. For this study, we used the dense breast volume output from the vendor-specific software for each woman, incorporating all 4 views (craniocaudal and mediolateral oblique of both breasts) of raw digital images, as done in the clinical setting. Dividing dense breast volume by total breast volume and multiplying by 100 defines volumetric percentage of density (VPD). Cut points are applied by Volpara to fractionate VPD into 4 categories analogous to BI002DRADS categories. The automated BIRADS categories, a to d, are defined as VPD that is (a), less than 4.5%; (b), 4.5% to 7.49%; (c), 7.5% to 15.49%; and (d), 15.5% or greater (21).
Breast density measures were assessed more than 6 months to 5 years before diagnosis. Mammograms were classified as more than 6 months to less than 2 years or 2 to 5 years before diagnosis. For stratified analysis, we selected mammograms 2 to 5 years before diagnosis for women with images available for both periods. We performed a sensitivity analysis for women who had examinations during both periods and found the results to be consistent with the main findings (Appendix Tables 1 and 2, available at Annals.org).
Statistical Analysis
We calculated frequency distributions of demographic characteristics and risk factors between case participants with screen-detected or interval cancer and control participants.
We used conditional logistic regression to assess the association of clinical and automated BI-RADS density with screen-detected and interval cancer. In these models, the mammogram furthest from the cancer diagnosis was used for each woman. These models were fit overall and stratified by length of time between density measurement and diagnosis (>6 months to <2 years [recent] vs. 2 to 5 years [distant]). Associations were summarized with odds ratios (ORs) and 95% CIs and with areas under the receiver-operating characteristic curve, or c-statistics, which accounted for the matched study design. Bootstrapping was used to test for differences in c-statistics between models. Models were adjusted for age, race/ethnicity, first-degree family history of breast cancer, history of benign results on breast biopsy, and body mass index (continuous). We used the second BI-RADS category as a reference to allow for estimations of risk at the lowest and highest categories and because it is the category with the greatest proportion of averagerisk women (3). Differences in breast cancer associations by study and timing of density measure in relation to breast cancer diagnosis were evaluated by including interaction terms in the models. Differences in risk associations between density measures and interval versus screen-detected cancer were tested by simultaneously estimating the risk for both interval and screen-detected cancer to formally compare the magnitude of associated ORs (“polytomous logistic regression”).
Overall sensitivity was calculated as the number of invasive breast cancer cases within 12 months of a positive mammography result divided by the total number of invasive breast cancer cases and by BI-RADS category. Sensitivity estimates were compared for recent and distant density before diagnosis with a proportion test adjusted for several comparisons.
Analyses were performed by using SAS software, version 9.4 (SAS Institute). Statistical tests were 2-sided, and P values less than 0.050 were considered statistically significant. For more details, see the Supplement (available at Annals.org).
Role of the Funding Source
The National Cancer Institute had no role in the design or conduct of the study or in the reporting of results.
RESULTS
We compared 1609 case participants with screen- detected invasive cancer and 351 with interval invasive cancer with 4409 matched control participants. Of the case participants, 599 had a recent breast density measure (>6 months to <2 years before diagnosis; median, 1.2 years) and 1361 had a distant assessment (2 to 5 years before diagnosis; median, 3.4 years). Women with screen-detected or interval cancer were more likely than control participants to have a family history of breast cancer, dense breasts, high dense breast volume, and high to very high BCSC 5-year risk (Table 1). Compared with the SFMR cohort, women in the Mayo group tended to be older and white and to have a higher body mass index, and fewer had dense breasts (Appendix Table 3, available at Annals.org).
Table 1.
Characteristics of Study Population*
| Characteristic | Women With Screen-Detected Invasive Cancer (n = 1609) |
Women With Interval Invasive Cancer (n = 351) |
Matched Control Participants (n = 4409) |
|---|---|---|---|
| Mean age (SD), y | 60.2 (11.8) | 56.6(12.2) | 59.6 (11.8) |
| Mean body mass index (SD), kg/m2† | 26.9 (5.9) | 24.1 (4.6) | 26.4 (6) |
| Median dense breast volume (IQR), ml‡ | 57.6 (41.4–80.5) | 67.0(45.9–97.8) | 51.5 (37.7–71.4) |
| Family history of breast cancer, n (%)§ | 448 (28.2) | 110(31.3) | 837 (19) |
| History of breast biopsy, n (%)∥ | 358 (22.4) | 99 (28.4) | 788 (17.9) |
| Race/ethnicity, n (%) | |||
| White | 1248 (77.6) | 263 (74.9) | 3526 (80) |
| Asian | 225 (14) | 61 (17.4) | 570 (12.9) |
| Black | 45 (2.8) | 4(1.1) | 89 (2) |
| Hispanic | 36 (2.2) | 8 (2.3) | 110(2.5) |
| Other | 55 (3.4) | 15 (4.3) | 114(2.6) |
| Clinical BI-RADS density, n (%) | |||
| Almost entirely fatty | 207 (12.9) | 19 (5.4) | 801 (18.2) |
| Scattered fibroglandular densities | 650 (40.4) | 74(21.1) | 1764(40) |
| Heterogeneously dense | 578 (35.9) | 160 (45.6) | 1469 (33.3) |
| Extremely dense | 174(10.8) | 98 (27.9) | 375 (8.5) |
| Automated BI-RADS density, n (%)‡ | |||
| Almost entirely fatty | 296 (18.4) | 23 (6.6) | 967 (21.9) |
| Scattered fibroglandular densities | 536 (33.3) | 61 (17.4) | 1395 (31.6) |
| Heterogeneously dense | 522 (32.4) | 116(33) | 1353 (30.7) |
| Extremely dense | 255 (15.8) | 151 (43) | 694(15.7) |
| BCSC 5-y risk, n (%) | |||
| Low (0%−0.99%) | 387 (24.1) | 79 (22.5) | 1373 (31.1) |
| Average (1.00%−1.66%) | 446 (27.7) | 84 (23.9) | 1210 (27.4) |
| Intermediate (1.67%−2.49%) | 399 (24.8) | 93 (26.5) | 1110(25.2) |
| High (2.50%−3.99%) | 282 (17.5) | 68 (19.4) | 528 (12) |
| Very high (≥4.00%) | 95 (5.9) | 27 (7.7) | 188 (4.3) |
BCSC = Breast Cancer Surveillance Consortium; BI-RADS = Breast Imaging Reporting and Data System; IQR = interquartile range.
Percentages may not sum to 100 due to rounding.
Data missing for 73 participants (1.1%).
Measured with Volpara software (Volpara Solutions).
Defined as having a mother, sister, or daughter with breast cancer. Data missing for 35 participants (0.6%).
Data missing for 17 participants (0.3%).
Screen-Detected and Interval Cancer Risk for Automated and Clinical Breast Density Measured More Than 6 Months to 5 Years Before Diagnosis
Of women whose breast density was assessed by automated BI-RADS, those with extremely dense breasts had a 5-fold greater risk for interval cancer (OR, 5.65 [95% CI, 3.33 to 9.60]) and a 1.4-fold greater risk for screen-detected cancer (OR, 1.43 [CI, 1.14 to 1.79]) than those with scattered fibroglandular densities (Table 2). This difference in ORs for density between detection modes was statistically significant (P for heterogeneity < 0.001). Similar statistically significant differences in the association between density and detection mode were found for clinical BI-RADS density (Table 2). Automated and clinical BI-RADS density measures had similar discriminatory accuracy, which was higher for interval than screen-detected cancer (c- statistics: 0.70 vs. 0.62, P < 0.001, and 0.72 vs. 0.62, P < 0.001, respectively). Associations between clinical and automated BI-RADS density and interval and screen-detected cancer were similar in both study cohorts (Appendix Table 4, available at Annals.org).
Table 2.
Risk for Screen-Detected and Interval Cancer for BI-RADS Breast Density Measured More Than 6 Months to 5 Years Before Cancer Diagnosis
| Variable | Screen-Detected Cancer |
Interval Cancer |
P Value for Interval vs. Screen- Detected Cancer |
||||
|---|---|---|---|---|---|---|---|
| Case Participants/ Control Participants, n/N |
OR (95% CI)* | C-Statistic (95% CI) |
Case Participants/ Control Participants, n/N |
OR (95% CI)* | C-Statistic (95% CI) |
||
| Clinical BI-RADS density | - | - | 0.62 (0.61–0.64) | - | - | 0.72 (0.69–0.75) | <0.001 |
| Almost entirely fatty | 207/691 | 0.62 (0.51–0.75) | - | 19/110 | 0.74(0.41–1.36) | - | |
| Scattered fibroglandular densities | 650/1481 | 1.00 (reference) | - | 74/283 | 1.00 (reference) | - | |
| Heterogeneously dense | 578/1195 | 1.26 (1.09–1.46) | - | 160/274 | 2.51 (1.74–3.61) | - | |
| Extremely dense | 174/283 | 1.83(1.44–2.32) | - | 98/92 | 5.09(3.11–8.35) | - | |
| Automated BI-RADS densityt† | - | - | 0.62 (0.60–0.63) | - | - | 0.70 (0.66–0.73) | <0.001 |
| Almost entirely fatty | 296/828 | 0.73 (0.61–0.87) | - | 23/139 | 0.73(0.42–1.29) | - | |
| Scattered fibroglandular densities | 536/1182 | 1.00 (reference) | - | 61/213 | 1.00 (reference) | - | |
| Heterogeneously dense | 522/1112 | 1.20 (1.02–1.41) | - | 116/241 | 2.22(1.44–3.43) | - | |
| Extremely dense | 255/528 | 1.43 (1.14–1.79) | - | 151/166 | 5.65 (3.33–9.60) | - | |
BI-RADS = Breast Imaging Reporting and Data System; OR = odds ratio.
Adjusted for age, body mass index, family history of breast cancer, history of breast biopsy, and race/ethnicity.
Measured with Volpara software (Volpara Solutions).
screen-Detected and Interval Cancer Risk by Recent and Distant Breast Density Measures Before Cancer Diagnosis
Among women who had recent automated BIRADS density measures before cancer diagnosis, those with extremely dense breasts compared with those with scattered fibroglandular densities had a 5-fold greater risk for interval cancer and a 1.4-fold greater risk for screen-detected cancer than control participants (Table 3). Likewise, among women with distant density measures before cancer diagnosis, those with extremelydense breasts had a 6-fold greater risk for interval cancer and a 1.4-fold greater risk for screen-detected cancer than those with scattered fibroglandular densities (Table 3). The differences in density effects between detection modes were statistically significant for density measured at both time points before diagnosis: recent (P < 0.001) and distant (P < 0.001).
Table 3.
Risk for Screen-Detected and Interval Cancer for Recent and Distant BI-RADS Breast Density Measures Before Cancer Diagnosis
| Variable | Screen-Detected Cancer |
Interval Cancer |
P Value for Interval vs. Screen- Detected Cancer |
||||
|---|---|---|---|---|---|---|---|
| Case Participants/ Control Participants, n/N |
OR (95% CI)* | C-Statistic (95% CI) |
Case Participants/ Control Participants, n/N |
OR (95% CI)* | C-Statistic (95% CI) |
||
| Recent breast density measure† | |||||||
| Clinical BI-RADS density | - | - | 0.62 (0.59–0.65) | - | - | 0.72 (0.68–0.77) | 0.023 |
| Almost entirely fatty | 53/167 | 0.64 (0.44–0.94) | - | 7/47 | 0.61 (0.23–1.62) | - | |
| Scattered fibroglandular densities | 171/403 | 1.00 (reference) | - | 34/122 | 1.00 (reference) | - | |
| Heterogeneously dense | 164/320 | 1.36(1.02–1.81) | - | 61/104 | 2.67 (1.50–4.77) | - | |
| Extremely dense | 62/87 | 2.06(1.35–3.15) | - | 47/36 | 5.98(2.81–12.7) | - | |
| Automated BI-RADS density‡ | - | - | 0.62 (0.59–0.65) | - | - | 0.70 (0.65–0.75) | <0.001 |
| Almost entirely fatty | 64/209 | 0.56(0.38–0.81) | - | 6/63 | 0.37(0.14–1.01) | - | |
| Scattered fibroglandular densities | 144/298 | 1.00 (reference) | - | 26/77 | 1.00 (reference) | - | |
| Heterogeneously dense | 162/306 | 1.31 (0.96–1.78) | - | 51/99 | 2.44(1.20–4.97) | - | |
| Extremely dense | 80/164 | 1.38 (0.92–2.07) | - | 66/70 | 5.39(2.30–12.6) | - | |
| Distant breast density measure§ | |||||||
| Clinical BI-RADS density | - | - | 0.62 (0.60–0.64) | - | - | 0.71 (0.67–0.76) | <0.001 |
| Almost entirely fatty | 154/524 | 0.61 (0.49–0.77) | - | 12/63 | 0.84(0.38–1.84) | - | |
| Scattered fibroglandular densities | 479/1078 | 1.00 (reference) | - | 40/161 | 1.00 (reference) | - | |
| Heterogeneously dense | 414/875 | 1.22(1.02–1.45) | - | 99/170 | 2.57 (1.57–4.19) | - | |
| Extremely dense | 112/196 | 1.72 (1.29–2.31) | - | 51/56 | 4.62 (2.38–8.98) | - | |
| Automated BI-RADS density‡ | - | - | 0.62 (0.60–0.64) | - | - | 0.69 (0.65–0.74) | <0.001 |
| Almost entirely fatty | 232/619 | 0.79 (0.64–0.97) | - | 17/76 | 1.07 (0.53–2.17) | - | |
| Scattered fibroglandular densities | 392/884 | 1.00 (reference) | - | 35/136 | 1.00 (reference) | - | |
| Heterogeneously dense | 360/806 | 1.15(0.95–1.40) | - | 65/142 | 2.11 (1.20–3.69) | - | |
| Extremely dense | 175/364 | 1.44(1.10–1.89) | - | 85/96 | 6.11 (3.07–12.2) | - | |
BI-RADS = Breast Imaging Reporting and Data System; OR = odds ratio.
Adjusted for age, body mass index, family history of breast cancer, history of breast biopsy, and race/ethnicity.
Defined as >6 mo to <2 y before cancer diagnosis.
Measured with Volpara software (Volpara Solutions).
Defined as 2–5 y before cancer diagnosis.
Similar statistically significant differences in the association between density measures and detection mode were found for recent and distant clinical BIRADS density measures before diagnosis (Table 3).
No statistically significant interactions were observed between the time point of density measure and the associations of automated and clinical BI-RADS density with interval breast cancer (P = 0.27 and P = 0.84, respectively) or screen-detected cancer (P = 0.22 and P = 0.83, respectively).
Recent and distant clinical and automated BI-RADS density measures before cancer diagnosis had greater discriminatory accuracy for interval than screen-detected cancer, but discrimination was similar across the 2 measures (Table 3 and Appendix Table 1).
Mammography Sensitivity for Clinical and Automated BI-RADS Density
Mammography sensitivity was similar between automated and clinical BI-RADS density categories: fatty, 93% versus 92%; scattered fibroglandular densities, 90% versus 90%; heterogeneously dense, 82% versus 78%; and extremely dense, 63% vs. 64%, respectively (Table 4). Sensitivity was greater for scattered fibroglandular densities, heterogeneously dense, and extremely dense categories for distant automated and clinical BI-RADS density measures than for recent measures before diagnosis (Table 4). Sensitivity was similar for women who had examinations available for both periods (Appendix Table 2).
Table 4.
Mammography Sensitivity by Clinical and Automated BI-RADS Breast Density, by Timing of Density Measure Before Cancer Diagnosis
| Variable | Almost Entirely Fatty |
Scattered Fibroglandular Densities |
Heterogeneously Dense |
Extremely Dense |
Overall |
|---|---|---|---|---|---|
| Density measure >6 mo to 5 y before cancer diagnosis | |||||
| Clinical BI-RADS density | |||||
| Interval cancer cases | 19 | 74 | 160 | 98 | 351 |
| Screen-detected cancer cases | 207 | 650 | 578 | 174 | 1609 |
| Sensitivity (95% CI), % | 92(88–95) | 90 (88–92) | 78(75–81) | 64 (58–70) | 82 (80–84) |
| Automated BI-RADS density* | |||||
| Interval cancer cases | 23 | 61 | 116 | 151 | 351 |
| Screen-detected cancer cases | 296 | 536 | 522 | 255 | 1609 |
| Sensitivity (95% CI), % | 93 (90–96) | 90 (87–92) | 82 (79–85) | 63 (58–68) | 82 (80–84) |
| Recent breast density measure† | |||||
| Clinical BI-RADS density | |||||
| Interval cancer cases | 7 | 34 | 61 | 47 | 149 |
| Screen-detected cancer cases | 53 | 171 | 164 | 62 | 450 |
| Sensitivity (95% CI), % | 89(81–97) | 84(79–89)‡ | 73 (67–79)§ | 57 (48–66)∥ | 75 (72–79) |
| Automated BI-RADS density* | |||||
| Interval cancer cases | 23 | 61 | 116 | 151 | 149 |
| Screen-detected cancer cases | 296 | 536 | 522 | 255 | 450 |
| Sensitivity (95% CI), % | 92 (85–98) | 85(79–90)¶ | 76(70–82)¶ | 55 (47–63)¶ | 75 (72–79) |
| Distant breast density measure** | |||||
| Clinical BI-RADS density | |||||
| Interval cancer cases | 12 | 40 | 99 | 51 | 202 |
| Screen-detected cancer cases | 154 | 479 | 414 | 112 | 1159 |
| Sensitivity (95% CI), % | 93(89–97) | 92 (90–95)‡ | 81 (77–84)§ | 69 (62–76)∥ | 85 (83–87) |
| Automated BI-RADS density* | |||||
| Interval cancer cases | 17 | 35 | 65 | 85 | 202 |
| Screen-detected cancer cases | 232 | 392 | 360 | 175 | 1159 |
| Sensitivity (95% CI), % | 93 (90–96) | 92 (89–94)¶ | 85(81–88)¶ | 67 (62–73)¶ | 85 (83–87) |
BI-RADS = Breast Imaging Reporting and Data System.
Measured with Volpara software (Volpara Solutions).
Defined as >6 mo to <2 y before cancer diagnosis.
P = 0.002.
P = 0.039.
P = 0.046.
P = 0.039.
Defined as 2–5 y before cancer diagnosis.
DISCUSSION
We found automated and clinical BI-RADS breast density measures to have similar ability to predict interval and screen-detected invasive cancer, regardless of timing of density measure, recent or distant from cancer diagnosis. We also found that automated and clinical BI-RADS density more strongly predicted interval than screen-detected cancer. This finding suggests that either automated or clinical BI-RADS measures could be used to inform women of their breast density and associated interval and screen-detected cancer risk. Automated BI-RADS density is more reproducible than clinical BI-RADS density on repeated measures (16, 17) between screening assessments at different facilities, whereas clinical BI-RADS has modest interrater reproducibility if different radiologists at the same facility or different facilities assess a woman’s breast density on consecutive examinations (6–8).
Breast density may affect breast cancer detection by increasing the growth rate of tumors or by masking them. Masking is the phenomenon in which both tumors and dense breast tissue appear white on mammograms, limiting the discrimination of breast cancer from normal tissue. Dense tissue also increases tumor aggressiveness, resulting in a greater proportion of advanced-stage cases of breast cancer, especially advanced-stage interval cancer (3), being diagnosed in women with dense breasts than in those with nondense breasts (22). Given these 2 mechanisms, it is not surprising that BI-RADS density has greater discriminatory accuracy in predicting interval than screen-detected cancer. Finally, on average, breast density declines about 2% per year (23), such that breast density measured several years apart shows similar associations with breast cancer risk.
The first study to report that automated BI-RADS density measured with Volpara on digital mammography is more strongly associated with interval than screened-detected cancer defined interval cancer as invasive cancer occurring within 24 months of a negative screening result (15). We extend the literature by reporting, in what we believe is the largest study to date, that automated BI-RADS density is more strongly associated with interval than screen-detected cancer when interval cancer is defined as invasive cancer occurring within 12 months of a negative screening result, which is the standard definition in the United States (2, 3). In addition, we compared automated with clinical BIRADS density, the standard for reporting breast density in the United States, and show that the 2 measures have similar predictive ability. Consistent with our results, area measures of breast density assessed on film-screen mammography in research settings have been found to be more strongly related to interval than screen-detected breast cancer risk (24, 25).
Among women who undergo mammography in the United States, 83% are screened every 12 to 35 months and 8% every 36 months or more (26). Thus, the opportunity to assess breast density on mammography for use in risk prediction models is variable. Boyd and colleagues (27) assessed percentage of mammo-graphic density on digitized film-screen mammography examinations in 3 screening programs in Canada using a continuous computer-assisted measure. Consistent with our results, the authors reported a higher percentage of breast density in women receiving a diagnosis of screen-detected or non-screen-detected cancer compared with those who did not develop breast cancer, up to 8 years after study entry. However, in contrast to our study, in which we found that associations with interval and screen-detected cancer risk were similar for recent and distant breast density measures before cancer diagnosis, Boyd and colleagues (27) reported a 17-fold higher risk for non-screen-detected cancer in the 1 to 2 years after a screening examination. Of note, the risk was 3.9-fold greater 2 to 4 years after a screening examination and 8.9-fold greater when risk was measured 4 to 8 years after screening (27). These results suggest that the ability to identify women at increased risk for interval cancer several years before diagnosis would allow improved screening strategies to be implemented to detect cancer earlier and reduce the risk for interval cancers. For example, the need for supplemental imaging could be predicted several years before cancer detection to optimize the chance to decrease interval cancer risk.
We examined mammography sensitivity to determine the absolute effect of breast density on the risk for interval and screen-detected invasive cancer. We found that automated and clinical BI-RADS density measures had similar sensitivity for each of the 4 BI-RADS categories, with slightly higher values for density measured further from the cancer diagnosis. Destounis and colleagues (28) reported that sensitivity decreased from the lowest to highest automated BI-RADS density categories (95% to 65%) but less so for clinical BI-RADS (82% to 66%). Wanders and colleagues (29) reported lower mammography sensitivity values from the lowest to highest automated BI-RADS density categories (86% to 61%) when the median time from measurement to diagnosis was longer than 2 years. The mammography sensitivity we report for density measured 2 to 5 years before diagnosis is slightly greater than that reported by Wanders and colleagues, probably because we defined interval cancer as invasive cancer diagnosed within 12 months, as opposed to 24 months, of a negative screening result. Longer screening intervals allow more time for missed cancer to grow and become symptomatic, such that interval cancer rates are higher in women who have biennial versus annual screening (30). Also, mammography sensitivity was slightly greater for the 2- to 5-year group, because the longer the period before breast cancer diagnosis, the higher the risk for both screen-detected and interval cancer, with a disproportionately higher risk for slow-growing screen-detected cancer.
Linkage to state tumor registries to enhance the completeness of identifying interval cancer cases was a strength of our study. We examined Volpara automated density measures that are available in clinical practice. Other commercially available automated volumetric breast density software (Quantra and PowerLook Density Assessment) might be tested to verify our results. We used clinical BI-RADS density assessments when the definitions from the fourth BI-RADS edition were available in clinical practice. Breast density distributions during the available periods of the fourth and fifth BIRADS editions in the BCSC are similar, suggesting that our results are clinically applicable (Miglioretti DL. Personal communication.). California and Minnesota density laws were enacted after clinical BI-RADS measures were collected for this study. We used a case-control design for economical assessment of automated density measures from several examinations for each study participant. Our study’s design did not allow us to assess the positive predictive value of mammography by breast density. Our matched control participants had a distribution of 5-year breast cancer risk similar to that of the population-based BCSC cohort (3), suggesting that our results are generalizable to women undergoing screening mammography. Our population was predominantly white and Asian. Although cancer detection has not been shown to vary by race/ethnicity (31) despite differences in breast density across racial/ethnic groups (20, 32), studies should be repeated in black and Hispanic women to ensure generalizability of results across all racial/ethnic groups. Finally, breast tomosynthesis is an emerging breast screening technique, with 30% of mammography machines in the United States producing tomosynthesis images as of 1 February 2018 (33). Volpara density measures are similar on digital and tomosynthesis C-View (Hologic) images (34), and no evidence has been published that the interval cancer rate or mammography sensitivity is different for digital mammography versus tomosynthesis (35). However, the contribution of volumetric density measures to breast cancer risk for tomosynthesis needs to be established.
This study looked at the timing of automated and clinical BI-RADS density measures and found that measures close to breast cancer diagnosis and those up to 5 years before were similar in predicting interval and screen-detected cancer risk. These findings suggest that automated or clinical BI-RADS measures may be used to inform women of their breast density and predict their risk for interval and screen-detected cancer, even as long as 5 years before cancer diagnosis. Because automated BI-RADS breast density is more reproducible than clinical density (16, 17) and is being used increasingly in the clinical setting, our results suggest that automated density measures may be used to predict risk and help identify women most in need of supplemental screening. Future research should focus on developing prediction models comparing automated with clinical BI-RADS density to determine whether repeated automated or clinical measures more accurately predict the 5-year cumulative risk for interval cancer.
Acknowledgment:
The authors thank the study participants, mammography facilities, and radiologists for the data they provided for this study.
Primary Funding Source: National Cancer Institute.
Grant Support: In part by grants R01 CA177150, P01 2CA154292, and R01 CA207085 from the National Institutes of Health, National Cancer Institute.
Appendix Table 1.
Risk for Screen-Detected and Interval Cancer for Recent and Distant BI-RADS Breast Density Measures Before Cancer Diagnosis, Among Women With Mammograms in Both Time Periods
| Variable | Screen-Detected Cancer |
Interval Cancer |
P Value for Interval vs. Screen- Detected Cancer |
||||
|---|---|---|---|---|---|---|---|
| Case Participants/ Control Participants, n/N |
OR (95% CI)* | C-Statistic (95% CI) |
Case Participants/ Control Participants, n/N |
OR (95% CI)* | C-Statistic (95% CI) |
||
| Recent breast density measure† | |||||||
| Clinical BI-RADS density | 0.63(0.60–0.66) | 0.75 (0.69–0.80) | 0.003 | ||||
| Almost entirely fatty | 54/218 | 0.44 (0.30–0.63) | 5/43 | 0.40(0.14–1.17) | |||
| Scattered fibroglandular densities | 244/509 | 1.00 (reference) | 27/86 | 1.00 (reference) | |||
| Heterogeneously dense | 190/359 | 1.26(0.98–1.64) | 60/99 | 1.83(0.97–3.47) | |||
| Extremely dense | 40/70 | 1.54(0.94–2.52) | 26/20 | 4.31 (1.64–11.3) | |||
| Automated BI-RADS density‡ | 0.63 (0.60–0.66) | 0.75 (0.69–0.80) | <0.001 | ||||
| Almost entirely fatty | 131/354 | 0.71 (0.53–0.95) | 8/58 | 0.87 (0.31–2.42) | |||
| Scattered fibroglandular densities | 191/394 | 1.00 (reference) | 23/91 | 1.00 (reference) | |||
| Heterogeneously dense | 153/289 | 1.23(0.92–1.66) | 48/60 | 5.85 (2.56–13.3) | |||
| Extremely dense | 53/119 | 1.10(0.70–1.72) | 39/39 | 9.79(3.51–27.3) | |||
| Distant breast density measure§ | |||||||
| Clinical BI-RADS density | 0.63(0.60–0.66) | 0.75 (0.69–0.80) | <0.001 | ||||
| Almost entirely fatty | 65/238 | 0.49 (0.34–0.70) | 4/44 | 0.49(0.15–1.59) | |||
| Scattered fibroglandular densities | 225/482 | 1.00 (reference) | 23/89 | 1.00 (reference) | |||
| Heterogeneously dense | 188/365 | 1.26(0.96–1.65) | 62/87 | 2.83(1.45–5.51) | |||
| Extremely dense | 50/71 | 2.09(1.30–3.34) | 29/28 | 4.40(1.75–11.1) | |||
| Automated BI-RADS density‡ | 0.63 (0.60–0.66) | 0.75 (0.69–0.80) | <0.001 | ||||
| Almost entirely fatty | 118/302 | 0.80(0.59–1.07) | 6/50 | 0.52 (0.18–1.53) | |||
| Scattered fibroglandular densities | 192/410 | 1.00 (reference) | 26/86 | 1.00 (reference) | |||
| Heterogeneously dense | 151/317 | 1.19(0.88–1.60) | 39/66 | 2.72 (1.28–5.77) | |||
| Extremely dense | 67/127 | 1.41 (0.92–2.16) | 47/46 | 6.20(2.39–16.1) | |||
BI-RADS = Breast Imaging Reporting and Data System; OR = odds ratio.
Adjusted for age, body mass index, family history of breast cancer, history of breast biopsy, and race/ethnicity.
Defined as >6 mo to <2 y before cancer diagnosis.
Measured with Volpara software (Volpara Solutions).
Defined as 2–5 y before cancer diagnosis.
Appendix Table 2.
Mammography Sensitivity by Clinical and Automated BI-RADS Breast Density, by Timing of Density Measure Before Cancer Diagnosis, Among Women With Mammograms in Both Time Periods
| Variable | Almost Entirely Fatty |
Scattered Fibroglandular Densities |
Heterogeneously Dense |
Extremely Dense |
Overall |
|---|---|---|---|---|---|
| Recent breast density measure* | |||||
| Clinical BI-RADS density | |||||
| Interval cancer cases | 5 | 27 | 60 | 26 | 118 |
| Screen-detected cancer cases | 54 | 244 | 190 | 40 | 528 |
| Sensitivity (95% CI), % | 92 (84–99) | 90(86–94) | 76(71–81) | 61 (49–72) | 82 (79–85) |
| Automated BI-RADS density† | |||||
| Interval cancer cases | 8 | 23 | 48 | 39 | 118 |
| Screen-detected cancer cases | 131 | 191 | 153 | 53 | 528 |
| Sensitivity (95% CI), % | 94 (90–98) | 89 (85–93) | 76 (70–82) | 58 (48–68) | 82 (79–85) |
| Distant breast density measure‡ | |||||
| Clinical BI-RADS density | |||||
| Interval cancer cases | 4 | 23 | 62 | 29 | 118 |
| Screen-detected cancer cases | 65 | 225 | 188 | 50 | 528 |
| Sensitivity (95% CI), % | 94(89–100) | 91 (87–94) | 75(70–81) | 63 (53–74) | 82 (79–85) |
| Automated BI-RADS density† | |||||
| Interval cancer cases | 6 | 26 | 39 | 47 | 118 |
| Screen-detected cancer cases | 118 | 192 | 151 | 67 | 528 |
| Sensitivity (95% CI), % | 95 (91–99) | 88 (84–92) | 79 (74–85) | 59 (50–68) | 82 (79–85) |
BI-RADS = Breast Imaging Reporting and Data System.
Defined as >6 mo to <2 y before cancer diagnosis.
Measured with Volpara software.
Defined as 2–5 y before cancer diagnosis.
Appendix Table 3.
Characteristics of Study Population, by Study Cohort
| Variable | San Francisco Mammography Registry Control Participants (n = 2603) |
Mayo Clinic Control Participants (n = 1806) |
|---|---|---|
| Mean age (SD), y | 58.3(11.9) | 61.3(11.4) |
| Mean body mass index (SD), kg/m2 | 24.9 (5.1) | 28.6 (6.5) |
| Median dense breast volume (IQR), mL* | 50.7 (36.3–71.9) | 52.5 (39.7–70.6) |
| Family history of breast cancer, n (%)† | 453 (17.4) | 384(21.4) |
| History of breast biopsy, n (%) | 315 (12.1) | 473 (26.2) |
| Race/ethnicity, n (%) | ||
| White | 1768 (67.9) | 1758 (97.3) |
| Asian | 549 (21.1) | 21 (1.2) |
| Black | 87 (3.3) | 2 (0.1) |
| Hispanic | 101 (3.9) | 9 (0.5) |
| Other | 98 (3.8) | 16 (0.9) |
| Clinical BI-RADS density, n (%) | ||
| Almost entirely fatty | 369 (14.2) | 432 (23.9) |
| Scattered fibroglandular densities | 1018 (39.1) | 746 (41.3) |
| Heterogeneously dense | 939 (36.1) | 530 (29.3) |
| Extremely dense | 277 (10.6) | 98 (5.4) |
| Automated BI-RADS density, n (%)* | ||
| Almost entirely fatty | 448 (17.2) | 519 (28.7) |
| Scattered fibroglandular densities | 760 (29.2) | 635 (35.2) |
| Heterogeneously dense | 876 (33.7) | 477 (26.4) |
| Extremely dense | 519 (19.9) | 175 (9.7) |
| BCSC 5–y risk, n (%) | ||
| Low (G%-G.99%) | 952 (36.6) | 421 (23.3) |
| Average (1.GG%–1.66%) | 756 (29) | 454(25.1) |
| Intermediate (1.67%–2.49%) | 583 (22.4) | 527 (29.2) |
| High (2.5G%–3.99%) | 239 (9.2) | 289 (16) |
| Very high (≥4.GG%) | 73 (2.8) | 115(6.4) |
BCSC = Breast Cancer Surveillance Consortium; BI-RADS = Breast Imaging Reporting and Data System; IQR = interquartile range.
Measured with Volpara software (Volpara Solutions).
Mother, sister, or daughter with breast cancer.
Appendix Table 4.
Risk for Screen-Detected and Interval Cancer for BI-RADS Breast Density Measured More Than 6 Months to 5 Years Before Cancer Diagnosis, by Study Cohort
| Variable | Screen-Detected Cancer |
Interval Cancer |
P Value for Interval vs. Screen- Detected Cancer |
||||
|---|---|---|---|---|---|---|---|
| Case Participants/ Control Participants, n/N |
OR (95% CI)* | C-Statistic (95% CI) |
Case Participants/ Control Participants, n/N |
OR (95% CI)* | C-Statistic (95% CI) |
||
| San Francisco Mammography Registry | |||||||
| Clinical BI-RADS density | 0.63(0.61–0.65) | 0.72 (0.68–0.75) | <0.001 | ||||
| Almost entirely fatty | 114/303 | 0.66(0.50–0.86) | 13/66 | 0.63(0.29–1.40) | |||
| Scattered fibroglandular densities | 407/834 | 1.00 (reference) | 54/184 | 1.00 (reference) | |||
| Heterogeneously dense | 394/745 | 1.26(1.05–1.52) | 107/194 | 2.25(1.44–3.51) | |||
| Extremely dense | 139/208 | 1.80 (1.36–2.38) | 84/69 | 5.23(2.97–9.21) | |||
| Automated BI-RADS densityt | 0.62 (0.60–0.64) | 0.71 (0.67–0.75) | <0.001 | ||||
| Almost entirely fatty | 158/371 | 0.75 (0.59–0.96) | 16/77 | 0.96(0.47–1.96) | |||
| Scattered fibroglandular densities | 314/628 | 1.00 (reference) | 39/132 | 1.00 (reference) | |||
| Heterogeneously dense | 385/704 | 1.36(1.11–1.67) | 85/172 | 2.36(1.36–4.10) | |||
| Extremely dense | 197/387 | 1.48 (1.13–1.94) | 118/132 | 5.58(2.93–10.6) | |||
| Mayo Clinic | |||||||
| Clinical BI-RADS density | 0.64(0.62–0.67) | 0.73 (0.68–0.79) | 0.001 | ||||
| Almost entirely fatty | 93/388 | 0.57 (0.42–0.76) | 6/44 | 0.78(0.26–2.31) | |||
| Scattered fibroglandular densities | 243/647 | 1.00 (reference) | 20/99 | 1.00 (reference) | |||
| Heterogeneously dense | 184/450 | 1.27 (0.99–1.64) | 53/80 | 3.70(1.83–7.48) | |||
| Extremely dense | 35/75 | 1.91 (1.17–3.10) | 14/23 | 4.10(1.34–12.6) | |||
| Automated BI-RADS density† | 0.66 (0.63–0.68) | 0.71 (0.65–0.76) | <0.001 | ||||
| Almost entirely fatty | 138/457 | 0.69 (0.53–0.90) | 7/62 | 0.41 (0.15–1.10) | |||
| Scattered fibroglandular densities | 222/554 | 1.00 (reference) | 22/81 | 1.00 (reference) | |||
| Heterogeneously dense | 137/408 | 0.97 (0.73–1.28) | 31/69 | 2.06 (0.98–4.30) | |||
| Extremely dense | 58/141 | 1.53(1.01–2.33) | 33/34 | 7.45 (2.65–20.9) | |||
BI-RADS = Breast Imaging Reporting and Data System; OR = odds ratio.
Adjusted for age, body mass index, family history of breast cancer, history of breast biopsy, and race/ethnicity.
Measured with Volpara software (Volpara Solutions).
Footnotes
Disclosures: Dr. Miglioretti reports grants from the National Institutes of Health during the conduct of the study, and personal and travel fees from Hologic outside the submitted work. Dr. Vachon reports grants from the National Cancer Institute during the conduct of the study and funding from Grail outside the submitted work. Authors not named here have disclosed no conflicts of interest. Disclosures can also be viewed at www.acponline.org/authors/icmje/ConflictOfInterestForms.do?msNum=M17–3008.
Reproducible Research Statement: Study protocol and statistical code: Available from Mr. Scott (e-mail, scott.christopher @mayo.edu). Data set: Available after study aims of funded grants are addressed and with appropriate contracts.
References
- 1.DenseBreast-info. Dense breast tissue, dense breasts. 2018. Accessed at www.densebreast-info.org on 28 February 2018.
- 2.American College of Radiology. American College of Radiology Breast Imaging Reporting and Data System Atlas (BI-RADS Atlas). Vol. 5 Reston, VA: American College of Radiology; 2013. [Google Scholar]
- 3.Kerlikowske K, Zhu W, Tosteson AN, Sprague BL, Tice JA, Lehman CD, et al. ; Breast Cancer Surveillance Consortium. Identifying women with dense breasts at high risk for interval cancer: a cohort study. Ann Intern Med. 2015;162:673–81. [PMID: ] doi: 10.7326/M14-1465 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 4.Sprague BL, Gangnon RE, Burt V, Trentham-Dietz A, Hampton JM, Wellman RD, et al. Prevalence of mammographically dense breasts in the United States. J Natl Cancer Inst. 2014;106 [PMID: ] doi: 10.1093/jnci/dju255 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 5.Engmann NJ, Golmakani MK, Miglioretti DL, Sprague BL Kerlikowske K; Breast Cancer Surveillance Consortium. Population-attributable risk proportion of clinical risk factors for breast cancer. JAMA Oncol. 2017;3:1228–36. [PMID: ] doi: 10.1001/jamaoncol.2016.6326 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 6.Ekpo EU, Ujong UP, Mello-Thoms C, McEntee MF. Assessment of interradiologist agreement regarding mammographic breast density classification using the fifth edition of the BI-RADS Atlas. AJR Am J Roentgenol. 2016;206:1119–23. [PMID: ] doi: 10.2214/AJR.15.15049 [DOI] [PubMed] [Google Scholar]
- 7.Gard CC, Aiello Bowles EJ, Miglioretti DL, Taplin SH, Rutter CM. Misclassification of Breast Imaging Reporting and Data System (BI-RADS) mammographic density and implications for breast density reporting legislation. Breast J. 2015;21:481–9. [PMID: ] doi: 10.1111/tbj.12443 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 8.Spayne MC, Gard CC, Skelly J, Miglioretti DL, Vacek PM, Geller BM. Reproducibility of BI-RADS breast density measures among community radiologists: a prospective cohort study. Breast J. 2012; 18:326–33. [PMID: ] doi: 10.1111/j.1524-4741.2012.01250.x [DOI] [PMC free article] [PubMed] [Google Scholar]
- 9.Melnikow J, Fenton JJ, Whitlock EP, Miglioretti DL, Weyrich MS, Thompson JH, et al. Supplemental screening for breast cancer in women with dense breasts: a systematic review for the U.S. Preventive Services Task Force. Ann Intern Med. 2016;164:268–78. [PMID: ] doi: 10.7326/M15-1789 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 10.Sprague BL, Conant EF, Onega T, Garcia MP, Beaber EF, Herschorn SD, et al. ; PROSPR Consortium. Variation in mammo- graphic breast density assessments among radiologists in clinical practice: a multicenter observational study. Ann Intern Med. 2016; 165:457–64. [PMID: ] doi: 10.7326/M15-2934 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 11.Tice JA, Miglioretti DL, Li CS, Vachon CM, Gard CC, Kerlikowske K. Breast density and benign breast disease: risk assessment to identify women at high risk of breast cancer. J Clin Oncol. 2015;33: 3137–43. [PMID: ] doi: 10.1200/JTO.2015.60.8869 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 12.Trentham-Dietz A, Kerlikowske K, Stout NK, Miglioretti DL, Schechter CB, Ergun MA, et al. ; Breast Cancer Surveillance Consor¬tium and the Cancer Intervention and Surveillance Modeling Network. Tailoring breast cancer screening intervals by breast density and risk for women aged 50 years or older: collaborative modeling of screening outcomes. Ann Intern Med. 2016;165:700–12. [PMID: ] doi: 10.7326/M16-0476 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 13.Brandt KR, Scott CG, Ma L, Mahmoudzadeh AP, Jensen MR, Whaley DH, et al. Comparison of clinical and automated breast density measurements: implications for risk prediction and supplemental screening. Radiology. 2016;279:710–9. [PMID: ] doi: 10.1148/radiol.2015151261 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 14.Jeffers AM, Sieh W, Lipson JA, Rothstein JH, McGuire V, Whittemore AS, et al. Breast cancer risk and mammographic density assessed with semiautomated and fully automated methods and BIRADS. Radiology. 2017;282:348–55. [PMID: ] doi: 10.1148/radiol.2016152062 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 15.Wanders JOP, Holland K, Karssemeijer N, Peeters PHM, Veldhuis WB, Mann RM, et al. The effect of volumetric breast density on the risk of screen-detected and interval breast cancers: a cohort study. Breast Cancer Res. 2017;19:67 [PMID: ] doi: 10.1186/s13058-017-0859-9 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 16.Alonzo-Proulx O, Mawdsley GE, Patrie JT, Yaffe MJ, Harvey JA. Reliability of automated breast density measurements. Radiology. 2015;275:366–76. [PMID: ] doi: 10.1148/radiol.15141686 [DOI] [PubMed] [Google Scholar]
- 17.García E, Diaz O, Martí R, Diez Y, Gubern-Mérida A, Sentís M, et al. Local breast density assessment using reacquired mammo-graphic images. Eur J Radiol. 2017;93:121–7. [PMID: ] doi: 10.1016/j.ejrad.2017.05.033 [DOI] [PubMed] [Google Scholar]
- 18.Ballard-Barbash R, Taplin SH, Yankaskas BC, Ernster VL, Rosenberg RD, Carney PA, et al. Breast Cancer Surveillance Consortium: a national mammography screening and outcomes database. AJRAm J Roentgenol. 1997;169:1001–8. [PMID: ] [DOI] [PubMed] [Google Scholar]
- 19.St Sauver JL, Grossardt BR, Yawn BP, Melton LJ 3rd, Pankratz JJ, Brue SM, et al. Data resource profile: the Rochester Epidemiology Project (REP) medical records-linkage system. Int J Epidemiol. 2012; 41:1614–24. [PMID: ] doi: 10.1093/ije/dys195 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 20.Tice JA, Cummings SR, Smith-Bindman R, Ichikawa L, Barlow WE, Kerlikowske K. Using clinical factors and mammographic breast density to estimate breast cancer risk: development and validation of a new predictive model. Ann Intern Med. 2008;148:337–47. [PMID: ] [DOI] [PMC free article] [PubMed] [Google Scholar]
- 21.Volpara Solutions. Volpara Density User Manual Version 1.5.3. Rochester, NY: Volpara Solutions; 2017. [Google Scholar]
- 22.Kerlikowske K, Cook AJ, Buist DS, Cummings SR, Vachon C, Vacek P, et al. Breast cancer risk by breast density, menopause, and postmenopausal hormone therapy use. J Clin Oncol. 2010;28: 3830–7. [PMID: ] doi: 10.1200/JCO.2009.26.4770 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 23.Cuzick J, Warwick J, Pinney E, Warren RM, Duffy SW. Tamoxifen and breast density in women at increased risk of breast cancer. J Natl Cancer Inst. 2004;96:621–8. [PMID: ] [DOI] [PubMed] [Google Scholar]
- 24.Krishnan K, Baglietto L, Apicella C, Stone J, Southey MC, English DR, et al. Mammographic density and risk of breast cancer by mode of detection and tumor size: a case-control study. Breast Cancer Res. 2016;18:63 [PMID: ] doi: 10.1186/s13058-016-0722-4 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 25.Boyd NF, Huszti E, Melnichouk O, Martin LJ, Hislop G, Chiarelli A, et al. Mammographic features associated with interval breast cancers in screening programs. Breast Cancer Res. 2014;16:417 [PMID: ] doi: 10.1186/s13058-014-0417-7 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 26.Lehman CD, Arao RF, Sprague BL, Lee JM, Buist DS, Kerlikowske K, et al. National performance benchmarks for modern screening digital mammography: update from the breast cancer surveillance consortium. Radiology. 2017;283:49–58. [PMID: ] doi: 10.1148/radiol.2016161174 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 27.Boyd NF, Guo H, Martin LJ, Sun L, Stone J, Fishell E, et al. Mammographic density and the risk and detection of breast cancer. N Engl J Med. 2007;356:227–36. [PMID: ] [DOI] [PubMed] [Google Scholar]
- 28.Destounis S, Johnston L, Highnam R, Arieno A, Morgan R, Chan A. Using volumetric breast density to quantify the potential masking risk of mammographic density. AJR Am J Roentgenol. 2017;208: 222–7. [PMID: ] doi: 10.2214/AJR.16.16489 [DOI] [PubMed] [Google Scholar]
- 29.Wanders JO, Holland K, Veldhuis WB, Mann RM, Pijnappel RM, Peeters PH, et al. Volumetric breast density affects performance of digital screening mammography. Breast Cancer Res Treat. 2017; 162:95–103. [PMID: ] doi: 10.1007/s10549-016-4090-7 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 30.Houssami N, Hunter K. The epidemiology, radiology and biological characteristics of interval breast cancers in population mammography screening. NPJ Breast Cancer. 2017;3:12 [PMID: ] doi: 10.1038/s41523-017-0014-x [DOI] [PMC free article] [PubMed] [Google Scholar]
- 31.Henderson LM, Benefield T, Nyante SJ, Marsh MW, Greenwood-Hickman MA, Schroeder BF. Performance of digital screening mammography in a population-based cohort of black and white women. Cancer Causes Control. 2015;26:1495–9. [PMID: ] doi: 10.1007/s10552-015-0631-3 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 32.McCarthy AM, Keller BM, Pantalone LM, Hsieh MK, Synnestvedt M, Conant EF, et al. Racial differences in quantitative measures of area and volumetric breast density. J Natl Cancer Inst. 2016;108 [PMID: ] doi: 10.1093/jnci/djw104 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 33.U.S. Food and Drug Administration. Mammography Quality Standards Act and Program. 2018. Accessed at www.fda.gov/Radiation-EmittingProducts/MammographyQualityStandardsActandProgram/FacilityScorecard/ucm113858.htm on 28 February 2018.
- 34.Machida Y, Saita A, Namba H, Fukuma E. Automated volumetric breast density estimation out of digital breast tomosynthesis data: feasibility study of a new software version. Springerplus. 2016;5:780 [PMID: ] doi: 10.1186/s40064-016-2519-4 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 35.Conant EF, Beaber EF, Sprague BL, Herschorn SD, Weaver DL, Onega T, et al. Breast cancer screening using tomosynthesis in combination with digital mammography compared to digital mammography alone: a cohort study within the PROSPR Consortium. Breast Cancer Res Treat. 2016;156:109–16. [PMID: ] doi: 10.1007/s10549-016-3695-1 [DOI] [PMC free article] [PubMed] [Google Scholar]