Consecutive Screening Rounds with Digital Breast Tomosynthesis Enable Detection of Breast Cancers with Poor Prognosis

See also the article by Conant et al in this issue.

Dr Moy is professor of radiology at NYU. She is the senior deputy editor for Radiology and deputy editor of Breast Imaging for Radiology. Her research focuses on diagnostic oncologic imaging, with an emphasis on breast cancer. She is an NIH-funded investigator with applications on multiparametric breast MRI and artificial intelligence. She collaborates with the NYU Center for Data Science to investigate deep learning tools for multitask learning across modalities.

Dr Heller is an associate professor in the breast imaging section of the Department of Radiology at the NYU Grossman School of Medicine. She is the fellowship director for breast imaging at NYU. Her research interests focus on improving screening outcomes and on screening utilization. Her experiences working in the U.S. and U.K. allow her to harmonize multiple national screening mammography guidelines. She is a fellow of the Society of Breast Imaging.

In digital breast tomosynthesis (DBT), multiple mammographic images are acquired to produce a volumetric reconstruction of the breast. DBT has been shown to decrease recall rates and to increase cancer detection rates (CDRs) across multiple studies, with effects likely dependent on the screening environment. Decreased recall rates are more likely to be observed in the United States, and increased CDRs are more likely to be observed in Europe (1). However, despite the apparent advantages of DBT regarding screening outcomes, several critical questions remain insufficiently answered by the current literature.

First, much of the tomosynthesis literature is based on data from a single initial screening round only (1). Thus, the sustained benefits of tomosynthesis with respect to the CDR, recall rate, and interval cancer rate over multiple screening rounds remain unproven. It is possible that the increased CDR reported in multiple studies is a function of an overreliance on evidence from single DBT prevalent round results, when a greater number of preexisting cancers would be expected to be identified.

Second, even if DBT helps detect more cancers than digital mammography (DM), we do not have a good understanding of the kinds of breast cancers detected with tomosynthesis. We also do not understand how and if these malignancies differ in type and biologic activity from those detected at DM. In other words, is tomosynthesis simply identifying relatively greater numbers of indolent nonaggressive cancers? Or is it able to improve detection of aggressive cancers with a poorer prognosis, for which early detection would be of highest benefit to patients? The answers to both of these questions are key to establishing an evidence-based justification for the use of tomosynthesis in breast cancer screening. The importance of these questions is augmented given some drawbacks of DBT (ie, greater cost than DM, increased radiologist reading time, and increased radiation if performed in conjunction with two-dimensional mammography).

In this issue of Radiology, Conant and colleagues (2) address these two critical questions. The authors compared the outcomes from 5 years of consecutive DBT screening with those from 1 year of DM screening alone in a population of 29 310 women who underwent 67 350 examinations. They found lower recall rates with DBT versus DM (8.0% vs 10.4% overall) and higher CDRs (6.0 vs 5.1 per 1000 women screened overall) for DBT compared with DM over multiple years of DBT use. In addition, odds ratios for performance across screening rounds were significantly different for biopsy recommendation rate, positive predictive value of recall, and recall between the two technologies at the patient level.

The study provides data on a greater number of screening rounds than any other work to date. Few investigations have explored incident versus prevalent round DBT outcomes over multiple screening years. In a recent article from the same authors, McDonald et al (3) analyzed screening outcomes after up to 3 years of DBT implementation. Another study by Hovda et al (4) assessed 35 736 Norwegian women screened with at least two consecutive screening examinations in four groups (DM after DBT, DBT after DM, DM after DM, and DBT after DBT). Both studies also found sustained relatively low recall rates for all study years and increased CDR relative to DM (8.3 per 1000 women screened in the DBT followed by DBT group and 4.6 per 1000 women screened with DM after DM in the Norwegian study). Conant and colleagues’ current work therefore expands on the scant body of literature addressing whether DBT confers sustained improvement with respect to DM over time. Analysis of outcomes over even later screening rounds for the cohort will continue to be of interest.

Conant et al report that a higher proportion of DBT- versus DM-identified cancers were invasive (70.0% vs 68.5%, respectively), a finding consistent with those in the prior literature. More thought-provoking: A greater percentage of tumors were advanced cancers (32.6% vs 25.0%, respectively). Advanced cancers were defined using the Tomosynthesis Mammographic Imaging Screening Trial criteria. A cancer was considered advanced if it had at least one of the following characteristics: metastatic disease, node positivity, human epidermal growth factor 2 receptor positive and size of at least 1 cm, hormone receptor negative and size of at least 1 cm, or hormone receptor positive and size of at least 2 cm (4). Age group analysis showed that a higher proportion of DBT-detected advanced cancers occurred in women ages 40–49 years and 50–74 years. This finding is worthy of note given that some current screening guidelines, such as those from the U.S. Preventive Services Task Force, encourage initiation of screening only at age 50 years.

Most strikingly, Conant et al found a higher percentage of poor-prognosis disease detected with DBT versus two-dimensional mammography—in contrast to multiple other studies. Findings from the multisite Population-based Research to Optimize the Screening Process trial, as well as other recent work, had suggested that tomosynthesis tends to either depict less aggressive smaller and good-prognosis early-stage luminal A cancers or helps find similar grade, hormone receptor status, and nodal status malignancies relative to two-dimensional mammography (5–8). Conant et al suggest their differing results may be due both to the diversity of their patient population and to possible prevalent screening bias in prior DBT literature, leading to preferential detection of less advanced and less aggressive cancers.

If DBT has the potential to identify more aggressive cancers with successive screening rounds, then it might prove to be the better mammogram, as has previously been touted. Ideally, however, we would want DBT to help us identify aggressive disease prior to metastatic spread. This will be an arena for future assessment in randomized controlled trials such as the Tomosynthesis Mammographic Imaging Screening Trial (9).

The efficacy of a screening modality should be evaluated not only according to what types of cancers are found, but also by what types of cancers are not detected. For this reason, the number and histopathologic characteristics of false-negative cancers are vital to understanding how well DBT performs in comparison to DM for breast cancer screening. Overall, no significant difference existed in false-negative rates between DM and DBT. However, Conant and colleagues report trends where false-negative rates were consistently slightly lower for DBT than for DM for all years, except the year 5 timepoint. Furthermore, at the year 5 timepoint, the sensitivity of DBT was equivalent to that of DM (85.7%, P = .17) for all cancers (in situ and invasive); in earlier years, it was higher.

In more recent years, Conant et al replaced two-dimensional DM with synthetic DM. The synthetic DM/DBT combination has the advantage of decreasing radiation dose engendered through combined two-dimensional DM/DBT imaging (10). Some studies have shown synthetic DM to be diagnostically equivalent to DM in terms of CDR (10). However, in the study by Conant et al, the dip in sensitivity for all cancers and the increase in false-negative rates in DBT year 5 compared with earlier DBT years raise the question as to whether the adoption of synthetic DM technology impacts some screening outcomes. This is worth further investigation, especially as practices move to adopt synthetic DM/DBT. In particular, analysis of the mammographic characteristics of all detected cancers (eg, calcifications vs masses vs architectural distortion) at synthetic DM versus DM would be helpful. Assessment of the molecular subtypes of false-negative cancers identified after synthetic DM/DBT screening would also aid in determining the strengths and weaknesses of this technologic approach.

In conclusion, Conant and colleagues address major and unresolved questions that must be answered as the breast imaging community continues to adopt DBT and synthetic DM. The authors’ current study illustrates the need for continued long-term evaluation of DBT over multiple screening rounds. In addition, we need continued analysis of the types of incident-round cancers detected—and not detected–with DBT and synthetic DM. Ultimately, such work may help us address the most critical consideration of all: the impact of tomosynthesis on mortality.