Twelve years ago, the FDA approved Hologic’s digital breast tomosynthesis (DBT) system – sometimes incorrectly referred to as 3D mammography – as an alternative to 2D full-field digital mammography (FFDM) to screen for breast cancer. Since then, other companies have introduced DBT systems, with the FDA reporting there are now 24,707 accredited mammography systems in the USA, and FFDM units (13,332) slightly outnumber DBT units (11,375). However, a key question still needs to be answered: Does DBT perform better than FFDM to justify the significant extra cost?
This is an important question, considering that this week the FDA set national standards to protect women with dense breast tissue, and one of the claimed benefits of DBT is that it detects cancers not seen with FFDM. A paper published earlier this week in Radiology (click here) has provided evidence to suggest that DBT is perhaps slightly superior to FFDM. Dr Emily Conant, chief of breast imaging at the University of Pennsylvania (seen above right) compared FFDM and DBT for four key parameters: recall rate; cancer detection rate per 1,000 women screened; positive predictive value (PPV) of recall; and biopsy rates (per 1,000).
She and her colleagues, who included two Hologic employees, completed a retrospective cohort study that included women aged 40 to 79 years who underwent FFDM or DBT screening over a 7-year period, beginning January 2014. There were five study sites and a total of 2.5 million examinations (1.7 million DBT, and 0.8 million FFDM) of 1.1 million women. The main findings are summarised in the table at left, where the differences between FFDM and DBT are all statistically significant at the 0.001 level.
As seen, the recall rate for DBT is lower than for FFDM, while the cancer detection rate is slightly higher. The PPV for recall is greater for DBT, as is the biopsy rate for 1,000 women screened. These findings are presumably why Conant told AuntMinnie.com: “This study further substantiates that DBT should be the standard of care for mammographic screening.”
The authors did not reference the TMIST trial, funded by the National Cancer Institute to the tune of $100 million, that is yet to be completed, and was designed to see if DBT is better than FFDM for reducing advanced breast cancer. They also argued the inclusion of false-negative results for both imaging modalities rarely overestimates cancer detection rates, but we know this is not the case. In fact, both DBT and FFDM in particular, have sensitivities less than 100%, especially for dense breasts. So, is DBT there yet? Maybe.
Dr. Conant and her group have done excellent work on validating DBT. However, this latest study still suffers from the fact that there are almost certainly biases in the data. The women were not randomly assigned to DBT or FFDM and we don’t know the selection criteria used for each. Early on in comparing a new test to an old it is often not possible to randomly assign women to one test or the other to create identical groups. Consequently, it is almost certain that the groups were not comparable. It is possible and likely that the “prior probability” of cancer differed between the two groups. Since women were likely selected for the DBT arms, I am fairly certain that the tissue patterns in the DBT group were more complex and more likely to obscure cancers.
Anyone who has used DBT for routine screening (without selection) knows that there are cancers that are visible on DBT that, even in retrospect, are not evident on 2D FFDM. Given that every DBT study includes FFDM MLO and CC projections, it is impossible for DBT to not be better at detecting cancers than FFDM alone. What has been needed for years is a study in which everyone gets DBT, but the FFDM images are interpreted first, without that planar information. Then the study is reviewed with the planar images. This is the best way to compare FFDM to DBT (same patients, same day, no selection bias) and there is no doubt that DBT will be superior. Just as with MRI, a randomized, controlled trial with death as the endpoint should be done to prove that the earlier detection with these systems saves lives.