Increased Cancer Detection Rate and Variations in the Recall Rate Resulting from Implementation of 3D Digital Breast Tomosynthesis into a Population-based Screening Program

This prospective internally funded investigation demonstrates that the addition of digital breast tomosynthesis to a two-dimensional mammography screening program in a U.S. academic medical center resulted in a significant increase in the cancer detection rate, as well as a decrease in the screening recall rate overall and for patients with heterogeneous or extremely dense breasts and for patients in their 5th or 7th decades.

Abstract

Purpose

To compare the recall and cancer detection rates (CDRs) at screening with digital breast tomosynthesis (DBT) with those at screening with two-dimensional (2D) mammography and to evaluate variations in the recall rate (RR) according to patient age, risk factors, and breast density and among individual radiologists at a single U.S. academic medical center.

Materials and Methods

This institutional review board–approved, HIPAA-compliant prospective study with a retrospective cohort included 85 852 asymptomatic women who presented for breast cancer screening over a 3-year period beginning in 2011. A DBT unit was introduced into the existing 2D mammography screening program, and patients were assigned to the first available machine. Ten breast-subspecialized radiologists interpreted approximately 90% of the examinations. RRs were calculated overall and according to patient age, breast density, and individual radiologist. CDRs were calculated. Single and multiple mixed-effect logistic regression analyses, χ² tests, and Bonferroni correction were utilized, as appropriate.

Results

The study included 5703 (6.6%) DBT examinations and 80 149 (93.4%) 2D mammography examinations. The DBT subgroup contained a higher proportion of patients with risk factors for breast cancer and baseline examinations. DBT was used to detect 54.3% more carcinomas (+1.9 per 1000, P < .0018) than 2D mammography. The RR was 7.51% for 2D mammography and 6.10% for DBT (absolute change, 1.41%; relative change, –18.8%; P < .0001). The DBT subgroup demonstrated a significantly lower RR for patients with extremely or heterogeneously dense breasts and for patients in their 5th and 7th decades.

Conclusion

Implementing DBT into a U.S. breast cancer screening program significantly decreased the screening RR overall and for certain patient subgroups, while significantly increasing the CDR. These findings may encourage more widespread adoption and reimbursement of DBT and facilitate improved patient selection.

^© RSNA, 2015

Introduction

Conventional screening mammography (two-dimensional [2D] mammography) is the only effective population-based early detection strategy that reduces the risk of breast cancer–specific mortality (1–5). For some, potential risks of false-positive results, not detecting breast cancer, or overtreatment of small breast cancers have caused concern about the utility of 2D mammography, as with any imaging examination (6–9).

Digital breast tomosynthesis (DBT) is a new technology designed to overcome some of the limitations of 2D mammography (10–12). Reader and observational studies, simulated clinical settings, retrospective studies, and several industry-funded investigations have suggested that DBT may reduce the screening recall rate (RR), reduce the false-positive rate, and possibly increase the cancer detection rate (CDR) (13–22).

To our knowledge, it remains uncertain, however, whether a prospective study evaluating DBT integration into a U.S. screening mammography program would produce similar results. Furthermore, there are limited data available describing variations in RR reduction with the use of DBT according to patient subgroup and risk factors, as well as among individual interpreting radiologists (21,23).

Our prospective study with a retrospective cohort sought to compare the CDR and screening RR with DBT with those of screening with 2D mammography and to evaluate variations in the RR according to patient age, risk factors, breast density, and examination type and among individual radiologists.

Materials and Methods

Study Design

This study was internally funded. All digital screening mammography examinations performed in a 649-bed academic medical center system (Beth Israel Deaconess Medical Center) from January 3, 2011, to March 15, 2014, were included in this institutional review board–approved, Health Insurance Portability and Accountability Act–compliant prospective investigation with a retrospective cohort. Asymptomatic women who presented for screening were included in our study. Women who had breast symptoms or had received a diagnosis of cancer within the past 5 years were excluded.

Examinations performed prior to December 14, 2012, were performed with 2D mammography machines (Essential [model years 2012 and 2007], DS [model year 2006], and 2000D [model year 2002]; all by GE Healthcare, Fairfield, Conn). One DBT device (Selenia Dimensions [model year 2012]; Hologic, Bedford, Mass) device was implemented on December 15, 2012.

Patients were scheduled and screened on the basis of the availability of each unit. An exceedingly small number of patients (estimated to be fewer than 0.05% of all patients), either at the time of scheduling or on arrival in the department, specifically requested 2D or DBT screening. The technologist and staff radiologist resolved each of these requests on a case-by-case basis, and these were not individually tracked.

Screening Protocol and Examination Interpretation

Before the implementation of DBT screening, all technologists and radiologists were trained specifically in how to acquire and interpret DBT images. All radiologists received more than 8 hours of U.S. Food and Drug Administration–approved tomosynthesis training before interpreting DBT examinations.

Our Food and Drug Administration–approved DBT protocol is for patients undergoing DBT to receive a combination examination, comprising a three-dimensional (3D) and 2D mammography acquisition using the two conventional views (craniocaudal and mediolateral oblique) of each breast.

Ten board-certified radiologists in a subspecialized breast imaging practice (including R.E.S., S.V., J.P., V.D., V.J.F., S.P., P.J.S., and T.S.M.) interpreted 76 620 (89.2%) of all included examinations. Each of these radiologists spends more than 80% of his or her time in breast imaging. All except three were fellowship trained in breast imaging, with an average of 15.6 years of breast imaging experience. The three radiologists without a fellowship had 27, 35, and 41 years of breast imaging experience. Each of these radiologists interpreted more than 150 2D mammography and DBT examinations during the study period.

The remainder of the examinations (9232 [10.8%]) were interpreted by 12 other lower-volume general radiologists. These examinations were excluded from analysis because of the low sample size and heterogeneity of this cohort.

Data Extraction

For each examination, the following information was extracted: age, personal history of breast cancer, history of breast biopsies with benign results, BRCA gene mutation positivity, family history of breast cancer (with “family” defined as first-degree relatives), examination type (2D mammography or DBT), whether the examination was a baseline examination, Breast Imaging Reporting and Data System assessment category, and anonymized interpreting radiologist identifier. Breast density was reported as fatty, scattered, heterogeneous, or extremely dense. Examinations for which a combination of density terms or nonstandard terms were used were excluded from the RR variations analysis.

Statistical Analysis

RRs were calculated for the two types of examinations overall and were stratified according to patient age at screening, breast density, family history and personal history of breast cancer, history of BRCA mutation, personal history of a breast biopsy with a benign result, whether this was the patient’s first mammographic examination, and radiologist. The Fisher exact test (24) was performed to determine the statistical significance of the RRs between the two screening methods overall and stratified according to patient risk factors (the nine predictors). P < .05 was considered to indicate a significant difference. Percentage reduction in RR and the 95% confidence interval (CI) were calculated (25,26). The number of patients that needed to be screened with DBT to prevent one recall was calculated by comparing RRs between DBT and 2D mammography [1/(RR of 2D mammography − RR of DBT)]. Simple logistic regression analysis was performed for each predictor, and unadjusted odds ratios and associated 95% CIs were obtained.

A mixed-effects logistic regression analysis (27) was performed to evaluate for a relationship between RR and each evaluated parameter. Note that the interaction effect of radiologist and examination type was also considered in the model to investigate if the effects of individual radiologists on the risk of recall varied according to the two types of examination. The F test was used to determine if the effect of a predictor was significant. P < .05 again was considered to indicate a significant difference. Odds ratios and the associated 95% CIs were obtained. For variables with more than two levels, such as patient age at screening, breast density, and radiologist, the Bonferroni correction method was applied to the calculation of the 95% CIs of the odds ratios. All analyses were performed by using software (SAS 2012; SAS Institute, Cary, NC).

Results

Overall Results

Of the 85 852 examinations included in our study, 5703 (6.6%) were performed with DBT and 80 149 (93.4%) were performed with 2D mammography.

Cancer Detection

Two-dimensional mammography was used to detect 280 carcinomas, or 3.5 per 1000 (95% CI: 3.1, 3.9). DBT was used to detect 31 carcinomas, or 5.4 per 1000 (95% CI: 3.7, 7.8). DBT helped detect 54.3% more carcinomas per 1000 (+1.9 per 1000, P < .0018) than 2D mammography.

One hundred ninety-seven (70.4%) of the carcinomas detected with 2D mammography and 16 (51.6%) of those detected with DBT were invasive carcinomas, corresponding to an invasive CDR of 2.46 per 1000 for 2D mammography and 2.81 per 1000 for DBT (+0.35 per 1000; +14.2%; P = .61).

Eighty-three (29.6%) of the carcinomas detected with 2D mammography and 15 (48.5%) of those detected with DBT were in situ carcinomas, corresponding to an in situ carcinoma CDR of 1.04 per 1000 for 2D mammography and 2.63 per 1000 for DBT (P < .0006).

RR Variations according to Patient Subgroup

Examinations with nonstandard or unspecified breast density categories and those reported by radiologists other than the 10 breast imaging–specialized radiologists were excluded from the RR variations analysis. Among the 75 760 patients included, 70 173 (92.63%) underwent 2D mammography and 5587 (7.37%) underwent DBT.

The average patient age at screening was 57.62 years ± 10.89 (standard deviation) for 2D mammography and 55.68 years ± 9.74 for DBT. Baseline characteristics of the patients who underwent 2D mammography and DBT are shown in Table 1.

Table 1.

Characteristics of Patients according to Type of Examination

Note.—Data in parentheses are percentages. The average patient age at screening was 55.68 years ± 9.74 (standard deviation) for patients who underwent 3D examinations and 57.62 years ± 10.89 for patients who underwent 2D examinations. Breast densities and ages of the patients in the 2D mammography and DBT groups were similar. Patients who underwent DBT were more likely to have a personal history of breast cancer, to have a family history of breast cancer, to have a personal history of a breast biopsy with a benign result, and to be undergoing a baseline examination. Radiologist 1 appeared to be the most experienced radiologist for the 2D examination, followed by radiologist 2; radiologist 2 dominated all the other radiologists in the use of the 3D examination.

Table 2 shows the RRs for the two types of examinations overall and stratified according to patient age at screening, breast density, family history of breast cancer, personal history of breast cancer, personal history of a breast biopsy with a benign result, first mammogram, and radiologist. Percentage reductions in RR and 95% CIs are also presented.

Table 2.

RRs and Number of DBT Examinations Needed to Prevent One Recall for Patients Who Underwent 2D Mammography and DBT

Note.—Data in percentages are 95% CIs. RR was not calculated for patients with a BRCA mutation because none of the four patients with a BRCA mutation who underwent DBT were recalled. NA = not available.

The overall RR was 6.10% for DBT and 7.51% for 2D mammography (P < .0001). RRs stratified according to breast density were lower with DBT than with 2D mammography for all density groups, with significance found for heterogeneously dense (7.33% vs 9.31%, P = .0048) and extremely dense (4.74% vs 6.54%, P = .0429) breasts. The number of patients needed to screen to prevent one recall was 50.51 for heterogeneously dense breasts and 55.56 for extremely dense breasts.

When RRs were stratified according to patient age, reduced RRs were seen with DBT relative to 2D mammography for all age groups, with significant differences observed in patients in the 40–49-year age group (8.66% vs 10.93%, P = .0075) and the 60–69-year age group (3.66% vs 5.86%, P = .0006). The number of patients needed to screen to prevent one recall was 44.05 for patients between 40 and 49 years of age and 45.45 for patients between 60 and 69 years of age.

For patients who were undergoing their first (or baseline) mammographic examination, the RR was 9.88% for DBT and 19.60% for 2D mammography (P < .0001). The differences in RRs between 2D mammography and DBT for patients with a family history of breast cancer (P = .1776), a personal history of breast cancer (P = .5259), and a personal history of a breast biopsy with a benign result (P = .5998) were not statistically significant. There were also no statistically significant differences in RRs between 2D mammography and DBT when RRs were stratified by interpreting radiologist (P = .0871–.9135).

The results of the simple logistic regression analysis (Table 3) indicated that patients who underwent DBT, patients with a personal history of breast cancer, and patients with a personal history of a breast biopsy with a benign result were less likely to be recalled. However, patients with greater breast density, younger patients, patients with a family history of breast cancer, and patients who were undergoing a baseline examination were more likely to be recalled overall. Presence of a BRCA mutation did not have a statistically significant effect on the risk of recall. Furthermore, it appeared that radiologists 2, 3, 4, and 6 were less likely to recall patients compared with radiologist 1, who interpreted more examinations.

Table 3.

Associations of Parameters Related to Examination Type, Patient Characteristics, and Radiologist with Risk of Recall

Note.—Data in parentheses are 95% CIs. Unadjusted odds ratio refers to simple logistic regression analysis without adjustment for confounding variables. Adjusted odds ratio refers to mixed-effects logistic regression analysis of the association of screening methods with the risk of recall after adjustment for other confounders.

According to the results of the mixed-effects logistic regression analysis (Table 3), personal history of breast cancer, BRCA mutation, and personal history of a breast biopsy with a benign result were not associated with the risk of recall. The results also indicated that there was no association between examination type and risk of recall after the other confounders were controlled for (odds ratio = 0.98; 95% CI: 0.84, 1.13; P = .7459). Similar to the simple logistic regression analysis, the mixed-effects logistic regression analysis revealed that patients with greater breast density, patients with a family history of breast cancer, and patients who were undergoing a baseline examination were more likely to be recalled. Furthermore, patients younger than 40 years of age (odds ratio = 1.96; 95% CI: 1.50, 2.57), patients between 40 and 49 years of age (odds ratio = 1.67; 95% CI: 1.50, 1.87), and patients between 50 and 59 years of age (odds ratio = 1.23; 95% CI: 1.11, 1.37) were more likely to be recalled compared with patients older than 70 years of age. There was no difference in risk of recall between patients between 60 and 69 years of age and patients older than 70 years of age (odds ratio = 0.94; 95% CI: 0.91, 1.16).

Additionally, the effects of individual radiologists on the risk of recall did not vary significantly for the two types of examinations (P = .4707). Radiologist 2 was less likely to recall patients compared with the more experienced radiologist (radiologist 1) (odds ratio = 0.32; 95% CI: 0.19, 0.52). There were no statistically significant differences in risk of recall between the other radiologists and the more experienced radiologist (radiologist 1).

Figures 1–4 present the adjusted odds ratios and the associated 95% CIs for examination type, family history of breast cancer, personal history of breast cancer, personal history of a breast biopsy with a benign result, first/baseline mammographic examination, patient age at screening, breast density, and radiologist. Note that where the 95% CI does not contain 1, the implication is that the associated effect was statistically significant. Patients were more likely to be recalled if they had a family history of breast cancer or if it was the patient’s first/baseline examination (Fig 1), if they were 59 years of age or younger (Fig 2), and if they had breasts that were more dense than fatty (Fig 3). Radiologist 2 was less likely to recall patients than the other radiologists (Fig 4), possibly because radiologist 2 interpreted a larger percentage of DBT examinations.

Figure 1:

Graph shows adjusted odds ratios for RR according to examination type, family history of breast cancer, personal history of breast cancer, personal history of a breast biopsy with a benign result, and first/baseline mammographic examination. Patients undergoing baseline (first) mammography and patients with a family history of breast cancer were more likely to be recalled.

Figure 4:

Adjusted odds ratios for RR according to radiologist. Radiologist 1, who interpreted the largest number of studies, was the reference radiologist. Radiologist 2, who interpreted the most DBT studies, was also significantly less likely to recall patients than the other radiologists.

Figure 2:

Graph shows adjusted odds ratios for RR according to patient age at screening. Age of 70 or greater was the reference condition. Patients younger than 60 years of age were more likely to be recalled.

Figure 3:

Graph shows adjusted odds ratios for RR according to breast density. Predominantly fatty density was the reference condition. Patients with scattered fibroglandular densities, heterogeneously dense breasts, or extremely dense breasts were more likely to be recalled.

Discussion

This prospective internally funded investigation with a retrospective cohort of 2D mammography examinations demonstrated that the addition of DBT to a 2D mammography screening program in a U.S. medical center resulted in a significant increase in the CDR, as well as a decrease in the screening RR overall, for patients with heterogeneously or extremely dense breasts, and for patients in their 5th or 7th decades. Our study contributes to the limited available prospective data regarding the implementation of DBT in breast cancer screening centers.

Variations in the screening RR with 2D mammography have been observed to occur on the basis of the patient’s breast density and age and interpreting radiologist factors (28–31). Our study documents that DBT is associated with similar variations in the RR on the basis of the patient’s breast density, age, and certain risk factors and interpreting radiologist factors.

Previous studies associated DBT with decreases in the RR such as from 10.4% to 8.8% (32), from 12.0% to 8.4% (21), from 12.3% to 7.8% (33), and from 10.7% to 9.1% (23). Notably, in our study, the 2D mammography RR (7.44%) was lower than in the described retrospective studies, yet our study still demonstrated a significant decrease in RR with DBT (6.08%).

Given the high costs of additional imaging evaluation and the potential emotional toll on a woman who fears she has cancer, we believe that calling back 18% fewer of these women is clinically meaningful. Extrapolating from the overall RR reduction observed with DBT, implementation of one DBT unit into a 2D mammography screening program may result in approximately 98 fewer patients being recalled each year.

Despite one study that demonstrated no significant change in CDR with DBT (21), several other studies have shown DBT to increase the CDR by 1.9 per 1000 (18), 1.5 per 1000 (20), 1.4 per 1000 (22), and 2.7 per 1000 (33). A meta-analysis (23) found an increase in CDR of 1.2 per 1000. Our investigation demonstrated a significant increase in the CDR (+1.9 per 1000) with DBT, and this difference is similar to or greater than these previously reported trends.

This study documents variations in RR with DBT among breast densities, patient ages, and individual interpreting radiologists. The significantly larger decrease in the RR for patients with heterogeneously or extremely dense breasts suggests that these patients may gain a larger benefit from DBT.

Patients in all age groups experienced a reduction in the RR, and this reduction was statistically significant for patients in their 5th or 7th decades. However, there is limited biologic explanation, to our knowledge, as to why patients in the 6th decade would not have a similar benefit. Further research is needed to understand the benefit that DBT may provide and how this benefit is affected by patient age. Further investigation may be useful to better understand the role of age in RR reduction.

There were variations in the radiologist-specific RR among radiologists interpreting both 2D mammographic studies and DBT studies, with some demonstrating a lower screening RR for DBT and others showing a higher RR with DBT. Furthermore, given that the introduction of tomosynthesis reflects the adoption of a new imaging technology, there is likely a learning curve that occurs at different rates for different radiologists. Future investigation is indicated to understand radiologists’ practice variations and learning curves as they interpret DBT examinations.

A significantly higher proportion of patients in the DBT subgroup in our study had a personal history of breast cancer, a history of breast biopsy with benign results, documented BRCA mutations, and/or a family history of breast cancer, but these patients were subject to fewer callbacks than the patients who underwent 2D mammography. This may have resulted from patients with an increased risk of breast cancer, having been more informed on current breast screening, being proactive in their screening preferences and seeking out DBT screening. It is conceivable that measured differences in RR and CDR could have been affected by differences in the populations screened with the two technologies.

Despite this increased prevalence of risk factors for breast cancer, there were still significantly fewer patients recalled after screening with DBT overall compared with the number of patients recalled after screening with 2D mammography. At the same time that patients with a higher risk of cancer were screened and fewer patients were recalled with DBT, DBT was also used to detect more cancers than 2D mammography. Given that DBT is a new technology, follow-up information is not yet available. Although some may consider this a potential limitation, it is notable that, when compared with 2D mammography, DBT resulted in a significantly lower percentage of patients being recalled, while at the same time a significantly larger percentage of cancers were detected with this new technology.

Recalling significantly fewer patients after screening has many benefits. Patients not recalled after screening may benefit by experiencing fewer missed days of work, less anxiety, fewer additional imaging evaluations, fewer unnecessary biopsies, and decreased health care and transportation costs. DBT could potentially also be associated with reduced health care costs related to screening mammography, but this is also clearly a function of the additional costs of purchasing DBT equipment, as well as the added physician time of interpreting DBT examinations. DBT has many potential benefits that are beyond the scope of this article.

The implementation of a DBT unit into an existing 2D mammography screening program resulted in there being far more 2D mammographic examinations than DBT examinations, and this could be considered a limitation of our study. Despite the low expected frequency (approximately five per 1000) of detecting cancers with screening mammography, we were able to detect a significant difference in the CDR after review of 5703 examinations. Furthermore, some subgroups had relatively small sample sizes, and this may have contributed to some differences not reaching statistical significance.

The exclusion of many high-risk patients with a personal history of breast cancer would be expected to reduce the CDR of the screening program, and, as a result, our 2D mammography CDR is on the lower end of the Breast Imaging Reporting and Data System recommended CDR of three to six cancers per 1000 (34). Yet when patients were screened with DBT in otherwise similar circumstances, the observed CDR moved close to the higher end of the recommended CDR.

The percentage of invasive cancers was lower with DBT than with digital mammography. The rate per 1000 was not statistically significantly different. It is not known which of the additional cancers detected by using DBT would have progressed to invasive malignancy and which would not have. For some, this could be considered a limitation of this investigation. The current trend in clinical practice is often to treat most malignancies. Further research is needed to determine which radiologic or pathologic information could be leveraged to mitigate the potential for overdiagnosis or overtreatment.

Our study was designed to be a prospective investigation with a retrospective cohort with as little bias as possible regarding which patients underwent which type of examination. Nonetheless, there were very few patients who requested to be changed from their scheduled examination type to another on the day of their examination, generally because they had heard of the potential benefits of tomosynthesis or because of concern about the additional radiation dose of the combination 2D and 3D protocol. These very few patients were not specifically tracked and are not expected to have introduced a substantive influence on this investigation or bias into our results. Furthermore, once there is no need for combination 2D mammography and DBT protocols resulting in double exposure, everyone could be screened with DBT alone.

Further research is indicated to discern whether the reduction in RR observed with DBT versus 2D mammography might also reduce the number of unnecessary diagnostic mammographic and ultrasonographic evaluations and unnecessary biopsies. Further cost-benefit and outcomes research is indicated to enable us to better understand the results of implementation of a DBT screening option.

Advances in Knowledge

• ■ Our prospective study with a retrospective cohort evaluated 85 852 patients, and digital breast tomosynthesis (DBT) was associated with a 54.3% increase in the cancer detection rate compared with two-dimensional (2D) mammography (3.5–5.4 per 1000; absolute change, +1.9 per 1000; relative change, +54.3%; P < .018).

• ■ DBT was also associated with a statistically significant reduction in the overall recall rate (RR), by 18.8%, for 5587 patients screened with DBT compared with 70 173 patients screened with 2D mammography (7.51%–6.10%; absolute change, −1.36%; relative change, −18.8%; P < .001).

• ■ DBT screening resulted in a decrease in the RR for all breast density subgroups, with significant decreases for patients with heterogeneously dense breasts (−1.98%; relative change, −21.3%; P < .007) or extremely dense breasts (−1.80%; relative change, −27.5%; P < .0429).

• ■ There was a reduction in the RR with DBT in all patient age groups, and this was statistically significant for patients in their 5th decade (−2.27%, P < .0075) or 7th decade (−2.2%, P < .0006).

Implications for Patient Care

• ■ DBT appears to represent a significant improvement over traditional 2D screening mammography, as it results in fewer recalls from screening and can be used to detect more breast cancers.

• ■ Some patients may benefit more than others from screening with DBT, such as those with heterogeneously dense or extremely dense breasts and those in their 5th or 7th decades.