Digital breast tomosynthesis: the future of mammography screening or much ado about nothing?

Digital breast tomosynthesis (DBT) or 3D mammography has been the subject of technically oriented research for many years; however, its evaluation in clinical studies has been relatively recent [1–4]. The publication of two pivotal screening trials of DBT in 2013 [5,6] has generated interest in this mammographic technology based on remarkable evidence of its detection capability, and another trial is currently in progress in Malmö [2,7]. This editorial presents the authors’ views on the current state of evidence on DBT for breast screening, and highlights key issues that should be addressed in future studies. Understanding various areas that merit evaluation in the application of DBT will help guide future research and assist in a timely translation of this technology into practice, while ensuring that adoption into breast screening practice is based on high-quality evidence.

The technologic derivation of tomosynthesis breast imaging from digital mammography platforms, which has evolved considerably since it was pioneered by a team at Massachusetts General Hospital [8], has been comprehensively described in recent literature [1–4]. In brief, DBT uses conventional x-rays and a digital detector to create cross-sectional slices of the breast; a series of individual low-dose images are acquired with the x-ray tube rotating over a limited arc above the compressed breast [1–4]. Data from these multiple low-dose frames are reconstructed parallel to the detector plane into a series of thin-slice images that the radiologist can scroll through, providing a quasi-3D mammogram. All standard projections used in conventional mammography can also be obtained with DBT. Active technical research in DBT is progressing to help optimize acquisition parameters and reconstruction algorithms, including the use of DBT acquisitions to reconstruct standard mammography views [2].

Summary of the evidence

Reviews of clinical studies of DBT

Reviews of the evidence on the clinical application of DBT [1,2] published prior to the recent reporting of two DBT screening trials [5,6], have indicated that DBT has the capability of improving the accuracy of mammographic interpretation. Several key issues were highlighted through these evidence reviews [1,2]. First, the addition of DBT to standard mammography generally improved interpretive accuracy through improved cancer detection and/or reductions in false mammogram reports. Second, the evidence was less consistent as to whether DBT alone (without standard 2D mammograms) was more accurate than standard digital mammography alone, only a few studies indicated that two-view DBT has equal or better accuracy than standard mammography [2]. These issues appear to be partly influenced by whether two-view or single-view DBT was used [2]; however a recent study has shown that the sensitivity of one-view DBT was higher than that of digital mammography (90 and 79%, respectively) [9] while the average false-positive fraction did not significantly differ between the two methods. Third, subjective interpretation of cancer conspicuity (which has been examined using qualitative or quantitative methods) indicates that cancers appear equally conspicuous or more conspicuous on DBT relative to standard mammography [2]. Important findings from these reviews were that the majority of clinical studies of DBT were based on generally small and selected series, or were test-set observer (reader) studies, that included a high proportion of breast cancer cases [1,2]. In addition, these studies used cases already suspicious or detected at standard mammography. These limitations mean that some studies were unlikely to have been powered to conduct meaningful comparisons between DBT and standard mammography, and overall these methodologies will have biased against showing an effect from DBT in breast screening [1,2].

Screening trials

The more recent publications of the STORM study [5] and the interim Oslo study results [6] have generated much interest in DBT, due to the reported study outcomes and also due to the improved study design relative to earlier clinical studies of DBT. These population-based screening trials have avoided the methodology limitations of earlier clinical studies of DBT (discussed above) by recruiting consecutive screening participants consenting to having their routine mammographic screen inclusive of DBT acquisition (meaning that combined 2D and 3D imaging was obtained), and were statistically powered for comparative analyses. Despite differences in the exact screen-reading and recall methods, both trials showed that integrated standard and DBT mammography screening significantly increased breast cancer detection (relative to standard mammography alone), and has the potential to reduce false-positive recalls. Based on the interim Oslo results [6] (the largest of the DBT screening studies), the estimated incremental cancer detection rate attributable to adding DBT to standard mammography was 1.9/1000 screens, and the estimate from the STORM study was 2.7/1000 screens [5]. Because the Oslo study randomized screen-reading strategies, whereas STORM applied a sequential screen-reading design, the former may represent the more realistic estimate for incremental cancer detection attributable to the addition of DBT. The final results from the Oslo study are awaited, as are those from other large-scale screening studies, such as the Malmö Breast Tomosynthesis Screening Trial (MBTST) [7]. The MBTST will include 15,000 women in population-based screening, imaged with standard digital mammography and DBT. The major difference compared with the Oslo and STORM trials is that one-view DBT is used and compared with standard two-view digital mammography. Preliminary results from the MBTST seem promising [2] and an interim analysis is underway during 2013. In the meantime, the issue that may be debated relates to whether and how the findings from the two published DBT screening studies, in the context of the overall evidence on DBT, affect screening practice and future research.

Future application & research

Although the STORM and (interim) Oslo reports provide convincing evidence that integrating DBT with standard mammography could be used for breast screening [5,6], it will be important to consider the final Oslo study results, and further evidence from additional screening trials that will provide relevant information. Data from several trials would be valuable to support the likelihood that adding DBT to (or potentially substituting DBT for) standard mammography in population breast screening would improve outcomes across different screening settings and screen-readers. There is likely to be relatively polarized views on how mammography screening practice should evolve based on the current state of evidence; one perspective would consider evidence of improved detection measures as sufficient and hence that DBT should be used in mammography screening. In this regard, further data are needed on how DBT affects recall rates and false-positive recall outside the context of the two screening trials (which applied recall rules that might not broadly reflect practice) [5,6]. The effect of DBT on false-positive recall is expected to vary according to the ‘baseline’ recall rates using conventional mammography, for example, readers or services with high recall rates might benefit more from using DBT than those who have low recall rates. Furthermore, in view of both the additional radiation dose and screen-reading time from adding DBT to standard mammography, it will be particularly important to perform further research to explore and identify the most efficient imaging combination for screening practice.

An opposing perspective would point out that detection measures alone do not constitute screening benefit; hence to assess whether the observed increase in breast cancer detection from DBT is likely to translate into additional screening benefit, a randomized controlled trial (RCT) comparing DBT-based mammography screening (DBT in combination with standard or with synthetic mammograms, and potentially DBT alone) with standard mammography would be the ideal evidence. A RCT allows comparison of interval cancer rates, a surrogate end point for screening benefit [10] that can be measured within a relatively short follow-up time; however this would require very large numbers to be powered to show differences in interval cancer rates. Therefore, a RCT might only be feasible through multicenter or even international collaborative efforts between screening programs. This perspective considers the possibility that improved breast cancer detection from using DBT might contribute to overdiagnosis, hence evidence that improved detection from DBT-based screening will effectively reduce interval cancer rates might be necessary to recommend a change to screening practice.

Other issues that should be factored into deciding future application of DBT include the increased costs and radiation resulting from combining DBT with standard digital mammography, although the latter of these concerns is addressed with newer DBT technology or by using one-view DBT. Cost–benefit evaluation, including incremental cost–benefit analysis, will be critical to address in future studies. Ideally, programs contemplating the use of DBT should plan to do so as part of planned evaluations whereby research trials are embedded into population screening delivery.

Conclusions

Breast cancer screening with mammography is, in many countries, one of the major secondary prevention strategies involving large populations. DBT represents the most significant improvement in mammography development in recent years, if not decades, and shows extremely promising results in population screening. However, changing well-established screening practice would be expected to have a major impact and should be based on strong and consistent evidence of benefit. Therefore, more evidence is needed from large-scale trials before a change in mammography screening practice can be recommended.

Financial & competing interests disclosure

N Houssami is supported by a National Breast Cancer Foundation (Australia) Practitioner Fellowship. The authors have no other relevant affiliations or financial involvement with any organization or entity with a financial interest in or financial conflict with the subject matter or material discussed in the manuscript apart from those disclosed.

No writing assistance was utilized in the production of this manuscript.