Breast Cancer Screening with Digital Breast Tomosynthesis: Are Initial Benefits Sustained?

Introduction

Breast imaging practices have rapidly adopted digital breast tomosynthesis (DBT) (combined with two-dimensional digital mammography [DM]) since its approval by the U.S. Food and Drug Administration in 2011. Early evidence about the effectiveness of screening with DBT was based on single-site retrospective studies and enriched reader studies (1–3). More recent data come from multisite retrospective studies and prospective trials (4–15). For example, a U.S. multisite retrospective analysis of 281 187 DM examinations and 173 663 DBT examinations demonstrated a 29% increase in the cancer detection rate (CDR) (from 4.2 per 1000 examinations to 5.4 per 1000 examinations) and a simultaneous 15% decrease in the abnormal interpretation rate (AIR) (from 10.7% to 9.1%) with DBT compared with DM (7). In a prospective European trial that included 12 631 mammographic examinations, the CDR increased by 27% (from 6.1 per 1000 examinations to 8.0 per 1000 examinations) and the AIR decreased by 15% (from 6.1% to 5.3%) with DBT (4).

Most published studies on DBT were based on the first or prevalent screening round, with limited data on performance metrics from subsequent or incident screening rounds (16–18). In a single-institution retrospective analysis, screening performance metrics were evaluated for 44 468 examinations during a 4-year period, 1 year before and 3 years after DBT implementation (16). The CDR remained unchanged (16). The AIR increased slightly for years 1–3 after DBT implementation but remained reduced compared with that for the year before DBT implementation (16). A more recent multi-institutional study from the Breast Cancer Surveillance Consortium based on 221 248 DM examinations and 106 126 DBT examinations demonstrated an immediate decrease in the AIR after DBT implementation, which was sustained for 2 years (18).

Because a prevalence effect with regard to cancer detection and false-positive results may occur in studies that evaluate the performance of DBT in screening a population for the first time, studies of DBT performance beyond the initial year of implementation and after the first screening round are warranted. The purpose of this study was to determine whether improved screening performance metrics with DBT are sustained over time at the population level and after the first screening round at the individual level. We hypothesized that certain performance metrics, such as reduced AIR, are sustained.

Materials and Methods

Study Population

The institutional review board exempted this Health Insurance Portability and Accountability Act–compliant retrospective study from the requirement for written informed consent. We conducted a retrospective review of screening mammograms obtained at a single academic medical center from March 1, 2008, to February 28, 2011 (DM group), and from January 1, 2013, to December 31, 2017 (DBT group). Screening mammograms obtained from March 1, 2011, to December 31, 2012, were excluded to avoid selection bias, as screening was performed with both DM alone and with combined DM and DBT during this hybrid period. DBT was fully integrated into our breast imaging clinic by January 2013. Patients with a personal history of breast cancer were also excluded from this study.

The DBT1 group included screening mammograms obtained in 2013, the DBT2 group included screening mammograms obtained in 2014, the DBT3 group included screening mammograms obtained in 2015, the DBT4 group included screening mammograms obtained in 2016, and the DBT5 group included screening mammograms obtained in 2017. Examinations in the DBT1 group could represent a patient’s first DBT examination or, if she had undergone DBT during the hybrid period, could represent a subsequent DBT examination. A woman whose mammogram was included in the DBT2 to DBT5 groups would not have undergone a prior DBT examination (a) if the examination was her baseline or (b) if she had transferred her care from another institution where she underwent DM examinations only.

Of the 304 630 included mammographic examinations, 61 747 were previously included in a study that focused on screening performance metrics in women aged 65 years and older (19) and 141 384 were previously included in a study that focused on the characteristics of screening-detected and interval cancers with DM versus DBT (20). In this study, we compared screening performance metrics with DM versus DBT in all women and evaluated whether improved screening performance metrics with DBT were sustained.

Imaging Technique and Interpretation

Screening mammography was performed with full-field DM or DBT (Selenia Dimensions; Hologic, Bedford, Mass). All DBT examinations consisted of tomosynthesis and conventional DM images. One of 40 breast imaging radiologists (17 in the DM group and 39 in the DBT group, 16 of whom were included in both groups) interpreted the examinations. Their overall breast imaging experience ranged from 1 year to 34 years, and DBT experience ranged from 1 year to 6 years. Mammograms were interpreted by using terminology from the American College of Radiology Breast Imaging Reporting and Data System (BI-RADS) Atlas (21).

Standard Definitions

Standard BI-RADS definitions were used to classify mammograms as true-positive, true-negative, false-positive, or false-negative (21). These definitions were used in our previous study (19). Performance metrics—including CDR, AIR, positive predictive value of recall (PPV1), positive predictive value of biopsies recommended (PPV2), positive predictive value of biopsies performed (PPV3), sensitivity, and specificity—were calculated by using standard formulas from BI-RADS (21). These definitions were also used in our previous study (19). The proportion of invasive cancers was calculated by dividing the number of invasive cancers by the total number of invasive and in situ cancers.

Data Collection

The following information was retrieved from the mammography information system (MagView Mammography Information System; MagView, Burtonsville, Md): age, race, breast density, history of breast cancer, imaging modality (DM or DBT), presence of a prior screening mammogram, number of previous DBT screening mammograms, reader, results of pathologic examinations within 1 year after mammography, and histologic type of breast cancer. In addition, estrogen receptor, progesterone receptor, and human epidermal growth factor receptor 2 statuses were collected for invasive cancers in the DBT groups. To capture all cancer diagnoses, data were also obtained from tumor registries from five hospitals within our health care system (Brigham and Women’s Faulkner Hospital, Boston, Mass; Dana-Farber Cancer Institute, Boston, Mass; Massachusetts General Hospital, Boston, Mass; Newton-Wellesley Hospital, Newton, Mass; and North Shore Medical Center, Salem, Mass). At the time of writing, data from these registries were available through December 31, 2017.

Statistical Analysis

All data were analyzed with statistical software (R, version 3.5.1; R Foundation for Statistical Computing, Vienna, Austria). The Wilcoxon signed rank test (for continuous variables) and the Pearson χ² test (for categorical variables) were used to compare age, race, breast density, and presence of a prior screening mammogram among the women in the DM and DBT groups. For each performance metric and for the proportion of invasive cancers, multivariable logistic regression models were estimated by using generalized estimating equations with an independent correlation structure and robust sandwich standard errors to account for multiple examinations from one patient. Generalized estimating equation models were estimated for each of the performance metrics within (a) the DM group versus the DBT1, DBT2, DBT3, DBT4, and DBT5 groups and (b) the group with no prior DBT examinations versus the groups with one, two, three, and four or more prior DBT examinations. The models were adjusted for age, race, breast density, and reader. The first model was also adjusted for presence of a prior screening mammogram. In addition, we conducted a sensitivity analysis on the subset of women with no prior DM or DBT screening examinations. Adjusted odds ratios (AORs), 95% confidence intervals, and the corresponding Wald P values were calculated for each model. P < .05 was considered to indicate a statistically significant difference.

Results

Study Population for DM versus DBT Analyses

The DM group was composed of 99 582 DM examinations in 55 086 women (mean age, 57.3 years ± 11.6 [standard deviation]), and the DBT group was composed of 205 048 examinations in 76 276 women (mean age, 58.2 years ± 11.2) (P < .001) (Fig 1). Small but statistically significant differences were observed in mean age, race, breast density, and presence of a prior screening mammogram among the DM and DBT groups (P < .001 for all) (Table 1).

Figure 1:

Download as PowerPointOpen in Image Viewer

Screening Performance Metrics with DM versus DBT

In the DM group, the CDR was 4.6 per 1000 examinations (458 of 99 582 examinations), AIR was 7.3% (7305 of 99 582 examinations), PPV1 was 6.3% (458 of 7305 examinations), PPV2 was 35.7% (458 of 1283 examinations), PPV3 was 37.2% (458 of 1230 examinations), sensitivity was 84.2% (458 of 544 examinations), and specificity was 93.1% (92 191 of 99 038 examinations). In the DBT groups, the overall CDR was 5.3 per 1000 examinations (1084 of 205 048 examinations), AIR was 7.1% (14 615 of 205 048 examinations), PPV1 was 7.4% (1084 of 14 615 examinations), PPV2 was 39.2% (1084 of 2762 examinations), PPV3 was 41.4% (1084 of 2617 examinations), sensitivity was 87.3% (1084 of 1242 examinations), and specificity was 93.4% (190 275 of 203 806 examinations). Table 2 presents the performance metrics for the DM and DBT groups.

Compared with that in the DM group, the adjusted odds of AIR were lower in the DBT1, DBT2, DBT3, and DBT5 groups (AOR, 0.85–0.93; P < .001 to P = .02) (Table 3). The odds of PPV1 were higher in the DBT3 group than in the DM group (AOR, 1.28; P = .01); however, the odds of PPV1 in the other DBT groups were similar to that in the DM group (AOR, 1.11–1.21; P = .11 to P = .35) (Table 3). The odds of specificity were higher in the DBT1, DBT2, DBT3, and DBT5 groups (AOR, 1.10–1.20; P < .001 to P = .004) compared with the DM group (Table 3). There were no differences in CDR (P = .08 to P = .95), PPV2 (P = .08 to P = .95), PPV3 (P = .12 to P = .84), or sensitivity (P = .10 to P = .87) (Table 3).

We also performed a sensitivity analysis in the subset of women with no prior DM or DBT screening examinations and found similar results. Compared with the DM group, the odds of AIR were lower in all DBT groups (AOR, 0.71–0.78; P < .001 to P = .008). Compared with the DM group, the odds of specificity were higher in the DBT1, DBT2, DBT3, and DBT5 groups (AOR, 1.30–1.41; P < .001 to P = .007). There were no differences in CDR (P = .12 to P = .95), PPV1 (P = .40 to P = .83), PPV2 (P = .13 to P = .94), or PPV3 (P = .20 to P = .86).

Screening-detected Cancers with DM versus DBT

A total of 1542 cancers were diagnosed in the DM and DBT groups, 1116 (72.4%) of which were invasive and 426 (27.6%) of which were in situ (Fig 2). The proportions of invasive cancers in the DM, DBT1, DBT2, DBT3, DBT4, and DBT5 groups were 65.9% (302 of 458), 69.6% (135 of 194), 80.3% (159 of 198), 80.4% (181 of 225), 70.2% (151 of 215), and 74.6% (188 of 252), respectively. Compared with the DM group, the DBT2, DBT3, and DBT5 groups had higher proportions of invasive relative to in situ cancers (AOR, 1.71–1.87; P = .005 to P = .051). The DBT1 and DBT4 groups had similar proportions of invasive cancers compared with the DM group (AOR, 1.08–1.45; P = .12 to P = .70). Estrogen receptor, progesterone receptor, and human epidermal growth factor receptor 2 statuses were available for 96.9% (789 of 814) of invasive cancers in the DBT groups; 85.6% (675 of 789) were estrogen receptor–positive and/or progesterone receptor–positive and human epidermal growth factor receptor 2–negative; only 5.1% (40 of 789) were triple-negative.

Figure 2a:

Download as PowerPointOpen in Image Viewer

Figure 2b:

Download as PowerPointOpen in Image Viewer

Figure 2c:

Download as PowerPointOpen in Image Viewer

Figure 2d:

Download as PowerPointOpen in Image Viewer

Study Population for DBT Analyses Based on Number of Previous DBT Screening Examinations

Of the 205 048 DBT examinations included in the previous analyses, 10 334 (5.0%) were excluded because no prior screening examinations were available at the time of interpretation, thus restricting our analysis of the group with no prior DBT examinations to those who had undergone prior DM examinations (Fig 1). An additional 19 200 examinations (9.4%) were excluded because the prior screening examinations had not been performed at our institution, and therefore the number of prior DBT examinations could not be determined by using our institution’s mammography information system (MagView) (Fig 1). The study population was thus composed of 175 514 DBT examinations in 61 240 women (mean age, 59.1 years ± 10.9 [standard deviation]) (Fig 1). Small but statistically significant differences were observed in mean age, race, and breast density among the DBT groups (P < .001 for all) (Table 4).

Screening Performance Metrics Based on Number of Prior DBT Screening Examinations

In the DBT groups, the overall CDR was 5.1 per 1000 examinations (896 of 175 514 examinations), AIR was 6.4% (11 153 of 175 514 examinations), PPV1 was 8.0% (896 of 11 153 examinations), PPV2 was 43.1% (896 of 2080 examinations), PPV3 was 44.9% (896 of 1996 examinations), sensitivity was 86.2% (896 of 1039 examinations), and specificity was 94.1% (164 218 of 174 475 examinations). Table 5 presents the performance metrics for the groups with no prior DBT examinations, one prior DBT examination, two prior DBT examinations, three prior DBT examinations, and four or more prior DBT examinations.

Compared with that in the group with no prior DBT examinations, the odds of CDR were lower in the groups with one prior DBT examination, two prior DBT examinations, three prior DBT examinations, and four or more prior DBT examinations after adjustment for age, race, breast density, and reader (AOR, 0.64–0.78; P = .001 to P = .054) (Table 6). The odds of AIR were also lower in the groups with one prior DBT examination, two prior DBT examinations, three prior DBT examinations, and four or more prior DBT examinations (AOR, 0.79–0.91; P < .001 to P = .002) (Fig 3). Compared with the group with no prior DBT examinations, the odds of PPV1 were lower in the groups with one prior DBT examination and three prior DBT examinations (AOR, 0.81 and 0.76, respectively; P = .04) (Table 6). The odds of sensitivity were lower (AOR, 0.31–0.52; P = .003 to P = .049) and the odds of specificity were higher (AOR, 1.08–1.26; P < .001 to P = .01) in the groups with one prior DBT examination, two prior DBT examinations, three prior DBT examinations, and four or more prior DBT examinations compared with the group with no prior DBT examinations (Table 6). There were no differences in PPV2 (P = .15 to P = .72) or PPV3 (P = .25 to P = .48) among the DBT groups (Table 6).

Figure 3a:

Download as PowerPointOpen in Image Viewer

Figure 3b:

Download as PowerPointOpen in Image Viewer

Screening-detected Cancers Based on Number of Prior DBT Screening Examinations

A total of 896 cancers were diagnosed in the DM and DBT groups, 668 (74.6%) of which were invasive and 228 (25.4%) of which were in situ. The proportions of invasive cancers in the groups with no prior DBT examinations, one prior DBT examination, two prior DBT examinations, three prior DBT examinations, and four or more prior DBT examinations were 75.7% (156 of 206), 75.0% (192 of 256), 73.8% (138 of 187), 73.3% (85 of 116), and 74.0% (97 of 131), respectively. Compared with the group with no prior DBT examinations, all DBT groups had similar proportions of invasive relative to in situ cancers (AOR, 0.88–1.14; P = .61 to P = .91).

Discussion

We found that the benefits of reduced abnormal interpretation rate and improved specificity with digital breast tomosynthesis (DBT) were sustained beyond the first DBT screening round. We did not find increased cancer detection rates (CDRs) after DBT implementation at a population level but did observe a preferential ratio of invasive relative to in situ cancers in the 2nd, 3rd, and 5th years after implementation. The highest CDR was observed with a woman’s first DBT examination.

Our results are consistent with those of other studies that have demonstrated a reduction in AIR with DBT, but the reduction (0.4%–0.7%) was smaller than previously reported (7,22–24). A larger reduction could be expected in practices that have higher AIRs with DM. Our study contributed to the existing literature by demonstrating that the reduction in AIR was sustained after the first screening round. The decrease in unnecessary diagnostic imaging may result in savings in health care expenses, time, and resources; compensate for the additional time required to interpret DBT examinations; and reduce patients’ anxiety (25,26). We also found that DBT leads to a small (0.5%–0.8%) but significant increase in specificity, which is consistent with results from a recent study (15). Of note, the performance metrics in year 4 (DBT4) did not follow the same trends as in other years. Although adjusted for age, race, breast density, presence of a prior screening mammogram, and reader, the analyses may have been affected by year-to-year changes in patient demographics, patients’ risk profiles, radiologists’ experience level with DBT and their examination volume, and recall patterns.

In our practice, which had a CDR of 4.6 per 1000 examinations with DM, the CDR did not change after implementation of DBT. A recent study also found no change in cancer detection with DBT in the setting of a high CDR with DM, which suggests that the benefits of cancer detection with DBT may be reduced in facilities that achieve a high CDR with DM alone (24). In addition, our study cohort preferentially reflected patients’ subsequent rather than first DBT examinations (because some women underwent their first DBT examination during the excluded hybrid time period). Had we included all patients’ first DBT examinations, the CDR may have been higher. Our analyses based on number of previous DBT examinations did, in fact, demonstrate that the highest CDR was observed with a woman’s first examination. Although the CDR did not change at the population level, we observed a preferential ratio of invasive relative to in situ cancers in the 2nd, 3rd, and 5th years after implementation. This preferential ratio has also been reported in previous studies and may contribute to improving patient outcomes from screening mammography (4,7,27,28). In addition, 85.6% of invasive cancers detected after integration of DBT were estrogen receptor–positive and/or progesterone receptor–positive and human epidermal growth factor receptor 2–negative, which is consistent with other studies demonstrating that DBT is associated with favorable tumor characteristics (15,29).

We found no difference in sensitivity after DBT integration. This finding differs from the results of a recent European study, in which a significant increase in sensitivity (from 54.1% to 70.5%) occurred with DBT (15). In our study, the sensitivities of both DM and DBT were higher (84.2%–89.6%) and false-negative examination rates were lower (less than one per 1000 examinations) than those observed in the European study, which may in part reflect differences in screening intervals and utilization between the U.S. and European screening settings. Similarly, other studies of screening performance in the United States have not found differences in sensitivity (9,23).

Our study has several limitations. The comparisons among the DM and DBT groups may have been affected by performance over time, but this temporal comparison was necessary because our institution transitioned from DM to DBT in 2013. Because this was a retrospective study, we cannot infer that the performance differences were solely due to the change in technology, as there were other differences between the two time periods of our study, including patient demographics and interpreting radiologists. The P values we report from this study were not adjusted for multiple comparisons; however, we prespecified all models and hypothesis tests and report individual P values and confidence intervals where appropriate. In addition, some of the significant differences in performance metrics between the DM and DBT groups were quite small and thus may not translate to improved clinical outcomes.

In conclusion, our results suggest that the benefits of digital breast tomosynthesis (DBT) extend beyond the first screening round. In particular, the benefits of reduced false-positive examinations and higher specificity with DBT are sustained after the first screening round. Although there was no change in cancer detection rate after implementation of DBT at the population level, we observed a preferential ratio of invasive relative to in situ cancers in the 2nd, 3rd, and 5th years after implementation of DBT, which may contribute to improving patient outcomes from screening mammography. This information may inform policies for integrating this relatively new technology into population-level screening programs.

Disclosures of Conflicts of Interest: M.B. disclosed no relevant relationships. S.M. disclosed no relevant relationships. P.A.D. disclosed no relevant relationships. A.M.M. disclosed no relevant relationships. K.P.L. Activities related to the present article: disclosed no relevant relationships. Activities not related to the present article: has grants/grants pending from GE Healthcare. Other relationships: disclosed no relevant relationships. C.D.L. Activities related to the present article: disclosed no relevant relationships. Activities not related to the present article: has grants/grants pending from GE Healthcare. Other relationships: disclosed no relevant relationships.

Author Contributions

Author contributions: Guarantor of integrity of entire study, M.B.; study concepts/study design or data acquisition or data analysis/interpretation, all authors; manuscript drafting or manuscript revision for important intellectual content, all authors; approval of final version of submitted manuscript, all authors; agrees to ensure any questions related to the work are appropriately resolved, all authors; literature research, M.B., P.A.D., K.P.L., C.D.L.; clinical studies, M.B., C.D.L.; statistical analysis, M.B., S.M.; and manuscript editing, all authors