1991. I’d just returned from a breast conference in Dallas where a young radiologist, Steve Harms, MD, creator of “RODEO MRI” for the breast, was the showstopper with images of breast cancer that were invisible on mammography. Over and over and over, the narrative was the same: “Here’s the cancer on MRI, and here’s the negative mammogram.” (We were still 10 years away from commercialization of breast MRI at the time.)
Once back in OKC, I walked into my Friday morning research group at OU (a multidisciplinary team I’d assembled, looking for a purpose), and one of the senior scientists at the Oklahoma Medical Research Foundation, Dr. Paul McKay, showed me an article proposing a blood test to detect early breast cancer (not the old tumor markers for advanced disease). He asked, “Do you see any potential for this?” At the time, I was preparing my first academic paper on why cancers were being missed on mammography (more often than was being alleged), while still reeling from the MRI presentation I’d seen in Dallas.
When I looked at the title of the article, the proverbial light bulb flickered above my head – “Yes, you could use a blood test to select patients for MRI screening.” 28 years later, I’m saying the exact same thing, while extending such a test for other purposes as well: 1) women with dense tissue on mammography who need a prompt to undergo ultrasound, 2) women who opt out of routine mammography, 3) women under 40 who don’t qualify for high-risk screening, and 4) women in countries without a screening infrastructure. In all these cases, a positive blood test would prompt a diagnostic work-up with breast imaging. It all sounds great, but when it comes to developing a blood test for early breast cancer, It’s Harder Than You’d Think.
Dr. McKay and I sponsored the “inventor” of the blood test, Chaya Moroz, PhD, on a trip to the U.S. (from Petah-Tikva, Israel) in 1993, and we wined and dined her in Oklahoma City, an obscure place she’d never heard of until I contacted her.
One-half of her expenses were covered by the Oklahoma Medical Research Foundation, while the other half was covered by a small army of Oklahoma fund-raisers I had as supporters back in the day.
Not long after her visit, I was invited by Dr. Chaya Moroz, developer of the test, to travel to New York to meet with the largest intellectual property firm in the world (or so they said), Penny & Edmonds. Founded in 1883, Penny & Edmonds included 100 or so lawyers, nearly all of them holding advanced degrees in science and technology.
A limo picked me up at the airport and took me to the top floor of a Manhattan skyscraper where I walked into a board room full of attorneys who were gathered to discuss the various aspects of a breast cancer blood test. There was only one other medical clinician in the room with me – Gerald Dodd, MD, an icon from MD Anderson who had been instrumental in the introduction of mammography to the U.S. Dr. Dodd was once President of the American College of Radiology, then served as Chairman of the Board of the ACR.
The chair of the meeting at Penny and Edmonds was an individual I’d be working with for the next 7 years – Leslie Misrock, a senior partner, and the first name listed on the firm’s stationery.
Needless to say, with a touch of megalomania, I was ready to revolutionize breast cancer screening after a mere two years in academics. Once MRI became available, a blood test would direct us when to use this highly sensitive tool for breast cancer detection. This blood test would be available to all women, not a limited high-risk group (like we currently do in selecting patients for MRI screening, wherein the majority of potential beneficiaries of MRI are excluded).
Long story short. Everything fell through in the end. Later, Leslie Misrock died of his prostate cancer in 2001 after 27 years of battling the disease, and the 120-year-old law firm of Penny & Edmonds was dissolved in 2003.
Oh yes, I should mention that we never actually tested anyone. That’s right. Realizing the explosive potential of such a test, we were completely bogged down by securing funding, starting a biotechnology company and assuring patent protection (that’s “patent,” not “patient”). We worked long (7 years) and hard on everything except validating the test. After all, the bar of evidence was set much lower back then. And, Dr. Moroz, developer of the test, in aligning her discovery with Penny & Edmonds, reminded everyone up front that her test needed “a little more work.” (Dr. Moroz is still active as a professor of immunology at Tel Aviv University.)
Lesson learned. Validate the test first.
At the same time that MRI was introduced to mainstream breast medicine, I began the arduous process of storing blood samples from women who had undergone breast MRI, under an IRB-approved protocol. The Internet was relatively new, and at first, I sought out collaborators online. However, once word got out that I had samples with a database linked to MRI results, I no longer had to seek opportunities. I’ve now shared information with 30 or so groups, sent 10,000 aliquots to 9 different groups, served on 3 Scientific Advisory Boards, and am currently ramping up for my 3rd formal prospective clinical trial.
And yet, we’re not much further along than when we flew Dr. Moroz from Tel Aviv to Oklahoma City 28 years ago. Why is it so hard?
A “blood test” could be used in several ways. Of course, the old tumor markers like CA27-29 were designed for detecting metastatic disease. As it turns out, one of the first problems we encountered is that the markers for early disease are likely to be different than advanced disease.
Yes, a single marker for early disease would be nice (yet, look at the confusion surrounding PSA for prostate cancer after many years). My hope here was dashed when the NMP66 (nuclear matrix protein) test failed in a clinical trial in the early 2000s.
Adding more complexity, it can be assumed that the different biologic types of cancer are going to generate different biomarker profiles, making it very difficult to develop a single test for Luminal A, Luminal B, HER2-positive, triple negative, etc.
Dr. Laura Esserman has suggested that we focus right away on a blood test for the more aggressive types of cancer, and leave the indolent types alone, to be discovered by screening mammography or by palpation. I won’t venture a guess as to whether or not this more exclusive target will be reached before or after a “general” breast cancer blood test has been developed.
Whenever a “blood test for breast cancer” is mentioned, many conceive of its use based on the challenges unique to one’s specialty. The medical oncologist is likely to consider a blood test for long-term follow-up purposes (like the old markers). Or, biomarkers in the blood might even take on a predictive component, helping to guide therapy. Meanwhile, the radiologists might consider a test with a strong NPV to be of invaluable help, allowing them to choose short interval follow-up rather than biopsy. But my intent is still the same – to select patients for supplemental imaging (MRI or ultrasound) in the screening setting. It would be especially helpful if the performance characteristics of the blood test hold up through all levels of breast density.
The basic tenet from which I operate is simple – early detection saves lives, as proven in the screening mammography trials, and it does so with relatively low sensitivity (less than 50% when compared head-to-head with MRI). That is, mammograms miss more cancers than we appreciated in the past, yet still reduce breast cancer mortality rates in clinical trials. If we are finding only half of detectable cancers with mammography (ignoring the nice boost with 3-D), then imagine what we’d do if we found the other half. It’s conceivable we could double the impact of current screening methodologies.
Some critics like to point out, “We don’t really need earlier detection, where we’ll only make overdiagnosis worse.” I agree, but that’s a different question. Ultrasound screening does not find cancers “earlier,” that is, smaller tumor size. It finds the camouflaged cancers missed on dense mammograms. Adding ultrasound to mammography in dense breasts “ought” to lower mortality more than mammograms alone, but this will have to be proven in prospective RCTs, just as the impact of a blood test in selecting patients for US (or MRI) will eventually have to be proven in RCTs.
Supplemental breast MRI is a bit more complicated when predicting mortality reductions as this modality has two effects: 1) finding cancers lost in a dense background, much like ultrasound, and 2) lowering the threshold of detection, such that you see “next year’s cancers” on this year’s MRI. This earlier detection (#2) might not be as important as #1, but my beef is that critics love to treat the benefit of MRI as entirely due to #2, i.e., “earlier.” In reality, it’s a combination. We know there’s a benefit to finding cancers in the 1.0 to 1.5cm range on screening studies, which is actually the majority of cancers found on screening MRI.
Ergo, mortality rates will likely be decreased accordingly, and a blood test will facilitate the process of patient selection for maximal cost-effectiveness. But despite these various approaches, the only proven screening method that confirms a mortality reduction is mammography, so the U.S. Preventive Services Task Force et al don’t even want to hear about our theoretical benefits, without a prospective RCT.
Those answers won’t emerge for another 20-30 years, so in the meantime, we could use some help, based on rational thought, a scary precept for the pure empiricist. Yet, it might surprise you to know that there are some screening experts in epidemiology who admit that the only acceptable surrogate for improved mortality reduction is the Sensitivity of the screening tool, an endpoint vastly more accessible and practical than 20- and 30-year studies of screening (and, in 30 years, we might not need to screen at all, if cures for metastatic disease become the norm).
Consider this – a perfect blood test (100% sensitivity/100% specificity) would pre-empt radiologic screening entirely. You’d only undergo breast imaging if the test were positive. We’re not going to see that anytime soon, however. Instead, we’re going to be dealing with lower accuracy, but perhaps enough to select patients for supplemental imaging. And, perhaps enough accuracy to use in women under 40 who are not being screened at all.
No one – in spite of some bold claims – has a test with reliable performance characteristics, ready for general use in the detection of early stage breast cancer. Yes, there’s a great deal of excitement about “liquid biopsies” (technically, this term applies only to cell-free DNA) and other tests, so the question remains: “How Good is Good Enough?”
As part of the discussion, you’ll hear that Specificity is more important, then you’ll hear that Sensitivity is more important, or you’ll hear the only thing we need to know is NPV (negative predictive value, which is another reflection of sensitivity). Although there is no confirmed data yet indicating a useable test, I took the liberty of writing a “what if” editorial on the subject, to demonstrate how a blood test would be incorporated into existing algorithms. You can access that article HERE: https://forum.breastcarenetwork.com/wp-content/uploads/2019/06/The-Breast-Journal-BLOOD-TESTING-final-copy-2019.pdf
In short, good specificity “rules in” supplemental imaging in patients who are relying entirely on mammography. Alternatively, good sensitivity “rules out” supplemental imaging in patients who are already utilizing breast MRI (or ultrasound) for screening where MRIs could be done less frequently with strong NPV. My article goes on to demonstrate how a relatively low accuracy can still do a better job of cost-effective patient selection than what we are currently doing now with risk stratification and density as our guides.
Great enthusiasm exists for the “liquid biopsy” technique being used in the GRAIL trial for multiple cancer types. So far, however, it looks like good specificity, but not-so-good sensitivity for cfDNA in breast cancer. This puts one in the position of missing a breast cancer on both mammography and a “liquid biopsy.” The money and power and prestige of star-studded supporters of GRAIL will be capturing headlines for years to come, even more so with the launch of STRIVE, focused on breast cancer (120,000 blood samples is the goal). But in evaluating their results, we’ll be hearing mostly about Specificity. We might have to read between the lines on Sensitivity – not just a percentage, but these 2 items as well: 1) what is the stage and biology of those tumors identified?….and 2) how will they know when cancer has been missed? (this being the great bugaboo when it comes to measuring Sensitivity, and the reason for the inflated Sensitivity that has been attributed to mammography for 40 years).
Handling raw data from blood test research is a formidable task. Each biomarker used in the algorithm has its own range of normal, rather than a dichotomy of positive vs. negative. Thus, minimal tweaking can change outcomes for each individual marker being studied, and then neural networks combine 8 or so tweaked biomarkers into a binary outcome – “positive” vs. “negative.”
I once tabulated performance characteristics for 6 different models used for each participant in a clinical trial of blood testing. Of course, one model had excellent specificity (but very low sensitivity) while another model had excellent sensitivity (but unacceptably low specificity). The other 4 models were somewhere in between. One proposal to overcome this problem is to create a “Score,” like Oncotype DX, then dump the problem back in the clinician’s lap, with guidance as to the probability of an underlying cancer. Most are going to prefer a binary outcome, but it might not be practical with circulating biomarkers. And this will be a factor that might give “liquid biopsies” with cfDNA the edge. Sensitivity might be only so-so, but the presence of cfDNA specific for breast cancer might give a more definitive test result, that is, a binary outcome.
But that’s not where the confusion ends. The performance characteristics in a blood test – sensitivity, specificity, accuracy, NPV and PPV — have to be merged with the characteristics found in our 3 primary methods of breast imaging. And for mammograms especially, the performance characteristics are highly dependent upon breast density. So, you have to combine blood test performance to that of mammography, or mammography combined with US, or mammography plus MRI, or merging blood test performance with US or MRI alone.
The questions don’t end there. How do you classify DCIS? Cancer or not? (My solution is that you must generate DCIS-specific data.) Do you even want to find subclinical DCIS? And what about the borderline results in those patients with significant proliferative disease? Should certain patients be excluded from blood testing due to florid or atypical hyperplasia, which tends to give blurry results on blood testing? What if the sensitivity of a blood test is so good that the breast cancer is not yet seen even on MRI? In this case, the sensitivity will be underestimated as the patient will be called “cancer-no” even though an early cancer is present. And what if the test is reflective of increased future risk? In this case the blood test devolves to a risk assessment tool, which I call “the blood test graveyard.” The list goes on and on.
Pop sociologist Malcolm Gladwell has claimed that practicing a task for one hour, 10,000 times, is the difference between excellence and not-so-excellence. Well, my team spends one man-hour (actually a woman-hour) for each specimen of blood obtained in this line of research (consent form, blood draw, specimen processing, data entry and shipping). Yet, after 10,000 specimens (or hours) we’re no closer to a blood test than before. Bottom line: it’s harder than you think.
For my Top Ten Lessons Learned when it comes to Blood Testing Research, visit: https://www.breastcarenetwork.com/news/developing-a-blood-test-for-early-breast-cancer
PS – My final shot at the moon (I’m getting too old for this) will hopefully begin later this year, with a company based in Alberta, Canada, (Syantra) for whom we’ve been sending samples for the past year or so, while a prospective, blinded trial is in the works. Preliminary results look good, so fingers crossed. Hope springs eternal.