A 1.0cm Invasive Ductal Carcinoma, Grade 3, discovered on screening MRI. Even in retrospect, mammography and ultrasound were both negative.
In a recent issue of the Journal of General Internal Medicine (Hill et al, 2018; 33:275-283), five regional imaging registries were analyzed, and more than 80% of patients undergoing screening MRI did not meet guidelines. The medical journalists then stretched the issue by converting “not meeting guidelines” into “unnecessary,” a much more provocative word that equates this violation to the knife-happy surgeon.
My response is two-fold: 1) Kudos to those using rational thought as opposed irrational guidelines (civil disobedience), and 2) How are they getting insurance coverage for their patients? (In OKC, we are chained to pre-authorizations since the “cash option” is still too high for truly low cost MRI screening).
After the oddly-conceived guidelines for screening MRI were introduced in 2007 by the American Cancer Society, few addressed the flaws. Instead, other organizations like the NCCN patterned similar recommendations, maintaining nearly all of the original flaws. And as one would predict, guidelines were converted over time to canon, whereupon all those who didn’t believe were labeled as heretical.
We have lost our way. And those who keep claiming that there is “no data” to support screening women at lower levels of risk have simply chosen not to look at available data. And certainly, critics don’t think through MRI performance characteristics applied to various sub-groups based on disease prevalence and incidence, where outcomes can be accurately predicted.
One could start with the 15% lifetime risk threshold (Claus model) utilized for entry to The Netherlands study (Dutch MRISC), by far, the largest of the 7 international MRI screening trials that led to the 2007 guidelines. In that study, both the modest risk patients and high risk patients identified twice as many invasive cancers as mammography.
But far more incriminating as to the weakness of our current “20% lifetime” guidelines is the work of Dr. Christiane Kuhl in Germany where she has screened a general population risk cohort (under 15% lifetime) with MRI (Kuhl CK et al. Supplemental Breast MR Imaging Screening of Women With Average Risk of Breast Cancer. Radiology 2017; 283:361-370).
Participants in Dr. Kuhl’s study are “cleared” with normal mammograms and, for most, screening ultrasound as well. Her ongoing series of more than 2,000 women is now larger than any of the international MRI screening trials based on high risk protocols. After excluding those cancers found on mammography, ultrasound, and clinical exam, MRI was still able to generate a cancer detection rate (CDR) of 15.5%, or 15.5/1,000, roughly triple the mammographic yield for cancer detection. And, if one looks only at the prevalence (first) MRI screen, then the CDR was 2.26% or 22.6/1,000, comparable to what one finds in the MRI screening studies performed in high risk population. No cancers were found with mammography alone, or ultrasound alone. And what about the old bugaboo of mammographic screening – the 25% interval cancer rate? Zero in Dr. Kuhl’s study. That’s right. The “inherent” problem of screening, that is, the more aggressive tumors that pop up in between screens…well, they are relegated to historical interest only in this average risk population.
What should be our goal for breast cancer screening with MRI? If we are shooting for cost-effective yields, we’re out of luck. Studies of cost-effectiveness for MRI screening require 3% CDRs at a minimum, and there is no possible way to maintain that number over the long-term. Who has a 100% risk of breast cancer over the next 33 years (3% X 33 years)? No one. Even BRCA-positivity only generates long-term risks in the 2%/year range, which is what you will find on long-term incidence screens with MRI. And for the women who barely squeeze above 20% risk for MRI justification, long-term screening will yield under 1%/year, nowhere close to being cost-effectiveness. So, unless we can get the cost of screening MRI down to $300 or so (or $600 and perform the MRI biennially) forget cost-effectiveness. Breast MRI will never be cost-effective for long-term screening if we rely on risk levels alone.
So, with cost-effectiveness currently impossible with MRI, are there alternative goals? What is the best way to offer MRI to the correct patients, opening up the possibility to all women, but without having to screen all women with MRI? How can we best recommend MRI to get as close to cost-effectiveness as possible without excluding the majority of candidates, as we currently do with our guidelines?
Let me mention an idea that I proposed during that short interval after the clinical introduction of breast MRI and before guidelines were handed down from on high. My breast radiology counterpart (Dr. Rebecca Stough) and I developed a simple point system that obviates all the current problems with our guidelines, based on that old principle known as “common sense,” or in the world of philosophy – rational thought. Two parameters were used – breast density and risk. We had 3 levels of risk (baseline, high and very high, taken from a prior publication of a working group – Am J Surg 2004; 187: 349-362), and this was combined with the 4 levels of breast density as commonly defined by the BIRADS system, yielding a single score. Importantly, this score defined the interval as well – annual, biennial or triennial MRI – rather than the bizarre all-or-nothing annual MRI (such that 21% lifetime risk tells us to perform MRI annually, but for 19%, no MRI at all, or else it will be “unnecessary”).
We published our first 6 MRI cancer discoveries (Semin Breast Dis 2008; 11:67-75) using this method, but we were too late by a few months, our article coming out on the heels of the American Cancer Society 2007 guidelines. Of those 6 cancers, using the 3 mathematical models suggested by the ACS, 0/6 would have been identified with the Gail, 0/6 using Claus, and only 3/6 using Tyrer-Cuzick. Clearly, we were targeting a different population for MRI success that was being missed by the ACS guidelines. When we updated our experience in 2014 with 33 MRI discoveries (The Breast Journal 2014; 20:192-197) , the Gail model would have selected only 9/33, Claus 1/33, and Tyrer-Cuzick 12/33. Combining all 3 models, and using the model that calculated the highest risk, only 16 of 33 cancers would have been identified had we followed ACS guidelines.
Today (after 50 MRI-detected cancers), with our spirit crushed by insurers that are enamored by the “magical” 20% lifetime risk, nearly all of our patients officially qualify for MRI as we were forced to surrender to the guidelines-turned-canon. At the same time, our total number of MRI-detected cancers has dropped precipitously as MRI is closed to women who don’t meet the “risk-only” guidelines. Where are the women we used to diagnose? Their discoveries come later, perhaps much later, in the form of mammographic screening with its attendant 25% interval cancer rate. Or, they have mammographically invisible tumors and are discovered well on down the pike with palpable cancers and node positivity rates far in excess of our node-negative MRI discoveries.
What did our formula do to identify so many MRI-discovered cancers missed by current guidelines? It took into account a major variable – every bit as powerful as risk – that is, a feature that should be considered in all women – breast density. This is in sharp contrast to the ACS guidelines that relegated density to some sort of quasi-risk issue in the “needs more research” category, treating it as an isolated issue. It is not isolated. It is an integral feature of every woman considering auxiliary imaging. By tossing it into the 15-19% risk grouping in the ACS guidelines, it was turned into “just another risk factor.”
Again, what is our goal in proper patient selection to improve efficiency? We are trying to identify those patients who, on the day of the planned MRI, have a heightened risk for developing breast cancer (best reflected by short-term risk, not lifetime), AND – with equal importance – are more likely to have that cancer missed by mammography (density level).
Why is the Level D patient denied MRI without additional risk factors? They certainly anticipated this issue in The Netherlands where the Level D Trial is underway, using density as the SOLE entry criteria for breast MRI. And does the Level A patient with 21% risk really need annual MRI? Come on. Admit it. Our current guidelines are in a bad state of affairs, predicated on so-called empirical data that has nothing to do with true empiricism.
And when it comes to pseudo-empiricism, how did we get locked into lifetime risks anyway? Well, it was lifetime risks that were mostly used in the MRI screening trials, so that defect simply got transferred into the 2007 guidelines, a la canon.
But if one wants to use a surrogate for higher CDRs on screening MRI, why would it be important to include the breast cancer risk 20, 30, 40 or even 50 years from now? Wouldn’t you really want to know the short-term risk, calculated let’s say, for the next 5 years? Does it make any difference in patient selection whether you use short-term or lifetime? Actually, it makes an enormous difference.
–Lifetime risks are not as accurate as short-term risks, often extrapolations of known risk beyond what has ever been observed.
–Lifetime risks have a wide range of calculations, depending on the models used
–Risk models might be accurate for populations, but are poor predictors at the individual level.
–The most widely utilized model for MRI screening (Tyrer-Cuzick) has the narrowest application, that is, white Western European ethnicity only.
–Picking a number like 20% turns risk assessment into gamesmanship to use the highest number possible for insurance coverage (even it means using inappropriate models)
–Lifetime risk is actually remaining lifetime risk which decreases over time, while at the same time, short-term incidence increases. By ignoring incidence as distinct from long-term risk, lifetime risks are highly age-discriminatory. Younger women qualify for MRI, even though an older patient with the same risks won’t qualify in spite of the fact the older patient might be triple the risk of the younger patient in the short-term. Another way age discrimination manifests itself is the patient who barely qualifies for MRI at age 45 with “21%” will no longer qualify at age 55 (19%) when she is at peak short-term probability of breast cancer. Remaining lifetime risks are deeply flawed, yet have become ingrained as the only option.
–Another indirect way lifetime risks discriminate against older patients is the “needs more research” category that includes ADH, ALH and LCIS. The young patient with these findings on biopsy will bypass the “needs more research” designation because the Gail and/or the Tyrer-Cuzick model will elevate them to a “greater than 20%” status. In contrast, the 65 year-old patient with ADH won’t make it without other risk factors, even though her chance of getting breast cancer is roughly the same at 1%/year.
–The (remaining) lifetime risk approach being used in current guidelines gives us a very wide range, from 20% to 85% lifetime, and everyone in that group is advised to consider annual MRI. This is analogous to the original BIRADS 4 lesion on mammography where the risk of malignancy was 3% to 90%. Once the clinical futility of such a range was recognized, the system was modified to 4a, 4b and 4c where there was some clinical help in PPV estimates. And so it is with our very wide range of risk for MRI, with the lower end of the spectrum clustering most patients around 20-25% where they are given the same recommendations as the patients at 85% risk.
The unfortunate part of all of this is that our simple point system proposed at Mercy-OKC 10 years ago erases all of these issues, yet will never make it to prime time. It opens up the possibility of MRI screening to all women. It takes away age discrimination totally. It uses categorical risk levels rather than the bizarre “20% lifetime” and all its attendant problems. And incidentally, it was created without any data whatsoever, simply applying performance characteristics of MRI in sub-groups of patients where 1) disease prevalence and incidence is known, coupled to 2) the probability that a cancer will be missed on mammography.
We were not the only ones thinking along these lines. If you want to see how rational thought, unfettered by pseudo-empiricism, can create a very nice model for selecting patients for auxiliary imaging, review the entry requirements for ACRIN 6666. This was a study primarily for proving the benefit of screening ultrasound. However, one-fourth of the participants underwent a screening MRI at the end of the study. This single MRI found more cancers than 3 screening sessions at 0, 12 and 24 months using double-modality screening. It’s an incredible testimony to the power of MRI over mammography and ultrasound combined.
But I digress. My purpose here is to show how ACRIN 6666 used well-informed rational thought to design accrual criteria that are very well-conceived. Granted, mathematical models were used, but included a short-term risk calculation that, in effect, erases the age discrimination problems.
In its simplest form: ACRIN 6666 included women with mammographic density, plus one additional risk factor.
In its actual form: Five levels of BREAST DENSITY were used (not correlating exactly with BIRADS), such that one could have relatively low density overall if there was significant density in at least one quadrant – a qualitative aspect in addition to the usual quantitative attempts to define density. One can see the sophistication used here, in that the goal is to identify those women most likely to have a cancer missed on mammography, rather than the simplified dense vs. non-dense.
RISK FACTORS for accrual:
Mutation in BRCA1 or 2
History of prior chest, mediastinal or axillary radiation
Personal history of breast cancer treated with conservation (over 50% of trial participants, btw, for those who believe there is no data to support MRI after breast conservation)
Lifetime risk greater than, or equal to 25% (Gail or Claus)
5-year risk using the Gail model – greater than or equal to 2.5% (or 0.5%/year)
5-year risk greater than, or equal to 1.7% + extremely dense breasts (note the interplay of risk and density — the greater the density, the lower the required risk. More than any other accrual feature, this adjustment reflects the principle I’m espousing in this article — it’s the interplay of risk and density that should be used currently for auxiliary imaging selection.)
Results of ACRIN 6666 for the total cohort (n=2,321) undergoing both mammography and ultrasound screening at 0, 12 and 24 months:
Sensitivity of mammography = 52% (55% invasive)
Sensitivity of ultrasound = 52% (94% invasive)
Sensitivity of both modalities together = 76% (not that great when you think about it)
But for the 612 patients who opted for a single MRI at the end of the study:
The combined sensitivity of both mammography and ultrasound = 44% (ouch!)
That is, 7/16 cancers were detected with mammo + US performed at 0, 12 & 24 months. Then, at the study’s end, MRI detected an additional 9 cancers (8/9 MRI-detected cancers were invasive with a mean size of 0.85cm, all node-negative).
As the study was designed with the primary interest being ultrasound’s potential, it was counted as a victory for the concept of auxiliary screening using that modality. However, many saw it as an incidental victory — by a wide margin — for MRI screening.
Then, the Principal Investigator of ACRIN 6666 decided that perhaps she should undergo an MRI herself, given her family history. A 1.0cm mammographically invisible invasive cancer was identified on MRI, prompting this PI to form a new breast density organization to increase awareness about the hazard of breast density.
The answer to proper patient selection is not refinement of risk assessment. Add SNPs to the Tyrer-Cuzick model if you must in order to get insurance coverage for the MRI, but we shouldn’t deceive ourselves that this approach is the long-term solution for MRI screening. Using risks alone can never reach the state of cost-effectiveness (unless costs drop dramatically and intervals longer than annual are validated).
No, the answer for cost-effectiveness is R&D with the sole purpose being to develop a post-mammography tool (e.g., a blood test) that will properly select patients for screening MRI (or MBI, or contrast-enhanced mammography). Such a tool would convert the patient from screening to diagnostic status, and the purpose of the adjunct imaging would be to confirm cancer (yes or no), and if yes, then to locate the cancer.
This approach would obviate the need for asymptomatic screening using these expensive modalities at pre-determined intervals in only a select few, as is currently done. Only those patients at extremely high risk, such as BRCA-positivity, would maintain fixed intervals (unless such a post-mammography tool were found that was nearly flawless). Such an approach would be all-inclusive and would largely do away with fixed interval screening where CDRs drop to 2% or less once the steady state is reached. CDRs obtained through a post-mammography tool would remain high as they would always be based on disease prevalence, which for breast cancer, is roughly 3-fold disease incidence.
To that end, I’m involved with 3 approaches (and very little else):
–Ultra-CAD analysis of normal mammograms (currently under an R-01 grant)
–Artificial Intelligence applied to non-contrast breast MRI, with gadolinium only if AI signals a problem.
Most of my experience is with #1, starting in 1993. This is a research agenda once considered at the lunatic fringe (Why are you doing that? We already have mammography!). Today, many groups are after what is now called a “liquid biopsy,” which was originally applied to tumor DNA fragments in the circulation, but is now applied to any blood-based test for cancer.
As prospects and performance get better and better, with proprietary implications, I work with one group at a time, most recently aligning with Syantra, Inc., working in laboratories at the University of Calgary. A prospective study is planned for Canada and the U.K., with possible sites in the U.S. If retrospective results can be duplicated in the prospective trial, we will be looking at clinical utility that could make MRI screening cost-effective.
Our goal doesn’t have to be perfection, no matter what the tool. With peak MRI cancer yields at “below 4%,” even in the highest risk patients (and lower for incidence screens), a CDR of 5% will be a wild success as the cost-effective threshold will be crossed. If this sounds like a low bar, recall that the CDR for screening mammography in a mixed population of prevalence and incidence screens hovers around 5/1,000 or 0.5%. Thus, our target CDR of 5% must be 10-fold more effective than screening mammography, a very high bar indeed. But then imagine the ease with which cost-effectiveness will be achieved with CDRs of 10% or even 20%.
One thing for sure, we’re going nowhere if we conform to the outdated MRI guidelines under which we are currently mandated, lest we be accused of doing something “unnecessary.” So, here’s to Civil Disobedience that, hopefully, will lead to the revolution in how we properly select patients for multi-modality breast cancer screening.