Update from the Breast Cancer Surveillance Consortium (April 2017 Radiology)


[music] Deborah Levine, MD Hello. I’m Debbie Levine.
I’m the Senior Deputy Editor for Radiology and I’m here today having a round table
podcast with experts in breast imaging screening. The Breast Cancer Surveillance Consortium
has two articles in our April Radiology issue. One of these is on national performance benchmarks
for modern screening digital mammography, and Dr. Connie Lehman Professor of Radiology
at MGH will be discussing that. A second is an update on diagnostic mammography with Dr.
Diana Miglioretti who’s Professor of Biostatistics at the University of California Davis. And
then because benchmarks are such an important topic we have accompanying editorial in this
issue of radiology that’s co-written by Dr. Ed Sickles and Dr. Carl D’Orsi. They’re
both Professor Emeritus of Radiology; Dr. Sickles at UCSF and Dr. D’Orsi at Emory.
I’d like to thank you all for participating in this round table discussion with me today.
Let’s start with the update on screening mammography. Dr. Lehman can you please tell
us a little bit about what your group did and what you found? Constance D. Lehman, MD, PhD Sure so the Breast
Cancer Surveillance Consortium members we really wanted to address performance of modern
digital screening mammography. So we had a database of over 1.7 million all digital screening
mammograms from 360 radiologists at almost 100 facilities. And we measured the performance
and compared it back in time to film screen and earlier performance measures. And we were
excited to see the results. What we found was really great news. More cancers are being
detected with modern mammography programs than historically. The vast majority of radiologists
in U.S. community practice are doing a fantastic job in finding cancers. The sensitivity of
mammography is higher than previously reported. So this was really exciting to see. At the
same time, we found that radiologists were struggling more with keeping false positive
rates low. This increased cancer detection and increased sensitivity seemed to be at
the cause of more false positives. And we also saw there is a lot of variation. We have
a lot of radiologists that keep low recall rates and high cancer detection rates, we
have others that struggle more with that. So we thought this could really help inform
the future directions that we can take in screening mammography in the U.S. D.L. Terrific, and Dr. Miglioretti can you
tell us a little bit about the study on diagnostic mammography and what those main results were? Diana L. Miglioretti, PhD Sure. We looked
at over 400,000 diagnostic mammograms performed by over 400 radiologists at almost 100 facilities.
And similar to the results from the screening benchmarks, we found that compared to our
prior benchmarks that included film mammography, that we are now detecting more cancers or
the cancer detection rate has increased but that has come at a cost of increasing the
abnormal interpretation rate. So it seems that we’re now doing more biopsies that
may not be necessary in order to detect those extra cancers. We’ve also found wide variability
across radiologists in these performance metrics. So some radiologists are finding a lot more
cancers and others may be not meeting current benchmarks. D.L. So Dr. Sickles in your editorial you
had some rather pointed comments regarding using the breast cancer surveillance consortium
data as bench marks, basically saying that many people in other groups are going to have
difficulty linking to cancer registries and getting data on sensitivity and specificity
of their own screening populations and I was just wondering in this digital age shouldn’t
we work to solve that problem of getting the data rather than saying that these metrics
can’t work? Edward A. Sickles, MD The metrics work for
each facility. What’s important for people to realize is that the real strength of the
BCSC data and why the two articles you’re publishing are so important is that because
the BCSC is able to link to tumor registries, they get near 100% cancer ascertainment. Cancer
ascertainment is really important with outcomes because it helps reliably determine whether
a positive is a true positive or a false positive, whether a negative is a true negative or a
false negative. So that’s important and that’s why the BCSC data show what’s really
going on in the United States. The problem is that benchmarks for individual radiologists
and individual mammography facilities have to reflect the data collection that these
facilities do; and most facilities the vast majority of facilities in the United States
don’t link with regional tumor registries, they’re not large HMOs that have captive
populations so they have sub-optimal, often not very good cancer ascertainment so the
data they get from their own audits won’t match the completeness of the data in the
BCSC and that creates a problem because then those practices are hoping to match BCSC benchmarks
when in fact they may not be able to. D.L. So Dr. D’Orsi moving on in your editorial
and just talking about breast cancer screening in general, how do you recommend that we combine
the goals for reasonably low recall rate while preserving a high cancer detection rate and
how can we use these ideas for screening and diagnostic thresholds that are potentially
different from our current recommendations? Carl J. D’Orsi, MD I think one of the things
you have to realize is remember the price you pay to detect cancer basically is going
to be the false positive. Obviously if you cut the false positive rate down too low you’re
going to start missing malignancies. That doesn’t mean you should have a false positive
rate of 80%, but just keep that in mind. So the biggest thing I think to have radiologists
perform better in the false positive arena is education and really what we do at many
conferences is a face-to-face with trainees with a set of mammograms. The other thing
that’s important is you have to find out or you have to know what type of cancer you’re
finding. You many have a very high or very low false positive rate and a good true positive
rate but you may be finding stage II cancers. So the point is that these benchmarks do not
work in isolation, they work as a cluster and you have to look at the product you’re
getting from these metrics to ensure that you’re doing what a screening exam should
do. So the bottom line is let’s be a little bit wary about knocking the head of false
positives. D.L. Great. Thank you. So Dr. Miglioretti
in their editorial Dr. Sickles and Dr. D’Orsi suggest that we might want to move to use
of a larger database such as the National Mammography Database since it has more accrual
than the BCSC does and has cancer ascertainment that could potentially reflect current practice
of mammography better that the BCSC. What do you think about that idea and is it possible
or helpful for these two databases to somehow combine forces? D.L.M. Yeah I think the BCSC and the NMD both
have unique strengths and they’re complementary in many ways. The BCSC has very high quality
data collection, work very hard to standardize across the facilities and that’s what we’ve
been talking about we linked cancer registries to have complete cancer ascertainment. The
NMD is very valuable in its large size and a number of participating facilities. We do
need to be careful about the variability in those facilities because some will have very
good cancer capture. I know several that link to state tumor registries as well like the
BCSC where others might have very poor cancer capture if the women have a lot of opportunities
at places to receive their follow-up care. So what we need to do is figure out a way
to measure how complete cancer capture is at a facility level so that we can correct
for that in the performance measures and comparing apples to apples. I would also suggest that
we find ways to make it easier for facilities to link to cancer registries so that everyone
can have high quality audit data. D.L. Great. I’d like to move now not to
just talking generically about screening benchmarks, but move to what we can actually do to improve
performance. Dr. Lehman this question is for you, in both of the BCSC papers your group
suggests that one method to improve performance could be to have programs to support second
reviews of mammograms recalled by radiologists who have these high or over call a lot of
mammograms. That would entail potentially 11 to 20 percent of mammograms overall and
I’m just wondering if that’s practical in today’s busy environment where most radiologists
already feel overworked. C.D.L. I totally agree. We really need to
think of options that are feasible. The idea of double reading has been implemented in
lots of different ways across the country. There are some big practices where every mammogram
is read by two radiologists. In others none of the mammograms are. What we’re proposing
is that sites might consider if they have a terrific mammographer who finds cancers
but their recall rate is 18% which we all agree is just too high, to have that person
review their recalls with a partner that has a much lower recall rate. That could really
support the radiologist who’s struggling with a high recall rate to bring that recall
down. I love when Dr. D’Orsi said also our educational programs. What can we do more
at our CMEs to support reducing false positives? We’re looking for that area where we can
maintain a high cancer yield but also keep our false positives as low as reasonable and
as possible. D.L. Thank you. So one of the issues that
was brought up in the editorial was that of the United States FDA regulations requiring
interpretation of only 480 examinations per year which is lower than that required by
other national screening programs. I’m wondering if any of you have a comment about whether
we should require higher minimum volumes in order to better our individualized statistics
for assessing performance. So Dr. D’Orsi would you like to comment on that? C.J.D. Yeah this was an issue that was discussed
for many years at the European population based screening centers and much higher qualification
rates to read mammography. The big argument here against that was, and I’m not saying
was a good argument, was the issue of actually having enough facilities here and people not
dropping out and reading mammograms because of that number issue. That was one of the
issues that was or one on the problems that was discussed. Personally I think it should
be increased. Perhaps not to some of the levels, I think and you may know this, but I think
in Europe it’s like 5,000 and it’s 940 here. I think there’s some happy medium
over here. And I think Connie’s point is excellent that to review your false positives
with someone who is more experienced and perhaps has better metrics at the facility. So yes
I think it could be increased without altering access to mammo facilities. D.L. Thank you. And then Dr. Sickles I have
a question that I’m not sure you can answer, but when we talk about all of these different
problems with false positives and cancer detection rate, how do you think we should best set
the performance bar that’s a minimum expectation for practicing radiologists? E.A.S. I think looking national performance,
seeing where it is, and trying to get as close as possible to what’s averaged there is
a good target. I would mention also in addition to the suggestion that Dr. Lehman gave about
double reading and mentoring radiologists, there’s one thing that all radiologists
can do, it’s very simple, it’s make sure or at least increase the likelihood that when
you read a mammogram you have the prior exams available for comparison. We’ve known for
decades that comparison with priors dramatically reduces recall rate without adversely affecting
the cancer detection rate. So this is like a win-win. In many practices for a variety
of reasons are in a situation where they accept patients who walk in for screening exams and
don’t have the priors available and they wind up with higher recall rates. Other practices
try for patient convenience to read out the screening mammograms while the patient waits.
Often in that situation they don’t have the prior mammograms available. In our practice
where we make a great effort to have priors available, we pretty much insist, there are
always a few exceptions but we pretty much insist on having the priors available when
we read out which is why our whole practice has a very low recall rate. D.L. A very positive outcome from all of this
discussion regarding metrics is that the BCSC has agreed to post yearly performance metrics
and I think that this is a huge amount of work for the consortium but will be very important
as we move from digital mammography to tomosynthesis. So Dr. Miglioretti given how involved you
are in the BCSC do you have any ideas about how tomosynthesis will change these performance
metrics? D.L.M. Well we’re just starting to look
at tomosynthesis now in the breast cancer surveillance consortium as part of (inaudible)
grants. I don’t have data from our study but I’m really looking forward to seeing
whether we also find reduced recall rates and possibly improve cancer detection with
tomosynthesis. I think it’s a very promising technology. D.L. Terrific, and then just to sum up what
we’re talking about, obviously breast cancer diagnosis is a hot research topic and I was
wondering if each of you could just briefly mention a new project that you’re working
on in this arena. Let’s start with Dr. Lehman what are you working on? C.D.L. There’s so much going on right now
that’s incredibly exciting. One of these areas we’ve been talking about is how can
take big data to improve patient care? Dorothy Sippo is a recent recipient of a Giraffe Award
and she’s using informatics and big data to help radiologists during their regular
routine clinical day get feedback about their recalls, about their recommendations for biopsies
in a way that really supports them to learn more from their own cases. We’re excited
about that. We think we live in an exciting time with tomosynthesis. How can we make sure
with this advanced imaging technology that we’re teaching how to use it in a way that
we really benefit from its promised potential of reduced recalls and higher cancer detection?
We’re excited about this area. We feel like we’re just at the brink of the next revolution
in early detection through screening and can’t wait to see what the future holds. D.L. Terrific. Dr. D’Orsi what are you working
on? C.J.D. We’re currently evaluating a device
that an Israeli company is manufacturing. What it does is through various imaging parameters
it evaluates a woman’s risk of malignancy and doesn’t give a score but it suggests
which patients should be recalled and/or sent for further imaging. Interestingly we just
finished a little leader study on this devise and the ROC curve showed a 15% increase in
the AUC in the area under the curve which to me was very surprising because I’m very
skeptical about everything and I was very, very surprised at that. So that’s one device
if it proves in a larger study to be fruitful and would help to decrease false positives
in increased cancer detection rates. That part is very interesting and we’re right
now going to approach the NIH if they have enough money in their budget to try to do
a larger study. D.L. Terrific. Dr. Sickles what is your team
working on? E.A.S. One of the things that we’ve just
been working on we just completed is a study looking at prior exams as I spoke before,
but not just comparing with one prior but comparing with more than one prior. What we
found in that study was that when you compare with more than one prior you actually have
a lower recall rate and a higher cancer detection rate than when you compare with one prior.
For all of you out there remember don’t just look at the one, look at everything that’s
available if you want to get the best results. The second thing that we’ve been doing is
we’ve been working with the NMD, the National Mammography Database. Even though their data
are not anywhere near – have as anywhere near as complete cancer ascertainment as does
the BCSC. They do have valuable data and one of the things that their data has shown recently
is that when you look at screening outcomes in elderly women, that is women 75 and older,
for whom there is less than complete data from randomized controlled trials because
very few women, elderly women, were studied in randomized trials. When you look at the
NMD data it’s very strongly indicative that the benefits that have been proved for screening
in women up to the age of 75 continue up to at least the age of 90. D.L. Wow. And then finally Dr. Miglioretti
can you tell us something about what you’re working on? D.L.M. Sure. The BCSC is working towards trying
to come up with a more personalized approach to both breast cancer screening and surveillance
in women with a history of breast cancer. And so we hope to use personal risk factors
and breast density to identify women who may need more frequent screening or may benefit
from supplemental imaging with breast MRI or breast ultrasound or tomosynthesis or who
might want to start screening earlier than recommended by guidelines. D.L. Well I’d just like to thank all four
of you for participating today and for all of the work that you do for breast cancer
screening and for imaging our patients. Thank you very much and we’ll look forward to
getting some of those papers you just discussed submitted to Radiology.

You May Also Like

About the Author: Oren Garnes

Leave a Reply

Your email address will not be published. Required fields are marked *