Breast tomosynthesis: a fine balance between benefits and harms in breast cancer screening

Many questions need to be discussed before 3D‐mammography is adopted for standard screening

In this issue of the MJA, Houssami and colleagues report the first prospective trial of breast tomosynthesis screening in the Australian BreastScreen program. They found that tomosynthesis has a higher cancer detection rate (CDR) than standard mammography; the trade‐offs were a higher recall rate, higher radiation dose, and longer reading time.1

The availability of breast tomosynthesis (3D‐mammography) has increased exponentially in recent years. In tomosynthesis, digital technology acquires an image of the breast that is converted into a series of slices in each of the two standard mammographic views, allowing the screen reader to scroll through an image stack or to view a cineloop. Tomosynthesis is now the standard of care in many private radiology practices around Australia, where it is used for both diagnosis and screening, despite limited evidence for its superiority over standard 2D‐mammographic imaging.

Tomosynthesis is not currently used by BreastScreen for routine screening. Its introduction across the program would require a major investment in equipment, data storage, and radiologist time. Evidence of significant benefit and lack of harm is therefore essential before considering using tomosynthesis for population screening.

Houssami and colleagues present data that could guide further research and decision making in this regard. Their real world trial introduced the option of tomosynthesis in an established BreastScreen service, with feasibility one of the outcomes examined. This prospective, semi‐randomised pilot study included 10 184 mammographic screens, half each with 3D‐ and 2D‐mammography. The CDR of 9.8 detected cancers per 1000 screens for the tomosynthesis group was impressively high compared with the 6.6 per 1000 screens for 2D‐mammography, especially as the mean age of the women in the tomosynthesis group was lower (58 v 62 years). It is consistent with the incremental CDR for tomosynthesis (1.6 per 1000 screens) derived by a recent meta‐analysis of international studies.2 However, a higher CDR is only beneficial if the additional detected cancers do not constitute overdiagnosis.3,4

In the study by Houssami and colleagues, nine of 49 malignancies in the tomosynthesis group (18%) were ductal carcinomas in situ, compared with four of 34 (12%) in the 2D‐mammography group; eighteen of 40 invasive cases in the tomosynthesis group (45%) were grade 1, compared with six of 30 (20%) in the 2D group. It is therefore possible that many of the additional abnormalities detected by tomosynthesis were not clinically significant. The proportions of invasive cancers that were larger than 15 mm in size were identical in the two groups (40%), as were the proportions of lymph node‐positive cancers (15–20%). The higher proportions of in situ and grade 1 invasive tumours in the tomosynthesis group data suggest that tomosynthesis detects some larger tumours occult to standard mammography, and detects them early, but is also detecting more small, early stage lesions. The proportion of invasive lobular cancers was lower with tomosynthesis group (13% v 23% of invasive cancers), despite earlier suggestions that tomosynthesis might be superior to standard mammography for detecting these cancers.

Previous tomosynthesis studies have raised concerns about an increased recall rate (as recall mostly involves false‐positive findings), although the published evidence is conflicting. Tomosynthesis seems to reduce the recall rate in programs with higher baseline recall rates, but to increase recall when the baseline rate is low (as in the Australian program).2 In the study by Houssami and colleagues, the overall recall rate was higher in the tomosynthesis group (4.2% v 3.0%), but the difference may not be significant for BreastScreen; the recall rate of 7.6% for first round and 3.4% for subsequent round screens with tomosynthesis is well within the corresponding BreastScreen national accreditation standards of < 10% and < 5%.5

The authors conclude that tomosynthesis screening was feasible, based on the ability to implement this option in an existing screening service and the low opt‐out rate by screened women. They acknowledged limitations in their evaluation of screen reading time, but their overall impression was that tomosynthesis screens took three times as long to read as standard mammographs. This would have long term resource implications for a population screening program. While it is likely that radiologists would become faster should tomosynthesis be adopted for routine screening, it is also possible that reading time could increase during subsequent screening rounds, as earlier tomosynthesis files would be available for comparison.

It is concerning that the mean glandular radiation dose per tomosynthesis image was 1.9 times the standard mammography dose. The tomosynthesis dose of 2.55 mGy per image is still low, but the cumulative exposure from repeated screenings must be considered, as well as women with a family history of breast cancer, who may commence screening in their 30s and have annual rather than biennial imaging.

The amount of information required to ensure that consent by women in screening programs is genuinely informed has been the subject of debate.6 This is particularly important when weighing up the potential benefits (the magnitude of any survival benefit is not always certain) and harms of screening (including the risks of overdiagnosis and overtreatment).7 Should tomosynthesis become routine in BreastScreen, these discussions will become even more complicated. The study by Houssami and colleagues provides local data that could inform BreastScreen research. It also provides some reassurance that it is reasonable to wait for the results of international randomised trials before making decisions about incorporating tomosynthesis screening into BreastScreen. It is clear that any benefits from tomosynthesis will come at a cost.