Journal of Clinical Oncology (JCO) Podcast

Host Dr. Davide Soldato interviews Dr. Sana Raoof to discuss the JCO article Turning the Knobs on Screening Liquid Biopsies for High-Risk Populations: Potential for Dialing Down Invasive Procedures.

TRANSCRIPT

Dr. Davide Soldato: Hello, and welcome to JCO After Hours, the podcast where we sit down with others from some of the latest articles published in the Journal of Clinical Oncology. I am your host, Dr. Davide Soldato, Medical Oncologist at Ospedale San Martino in Genoa, Italy. Today, we are joined by JCO author Dr. Sana Raoof, Physician at Memorial Sloan Kettering, to talk about her article, “Turning the Knobs on Screening Liquid Biopsies for High-Risk Populations: Potential for Dialing Down Invasive Procedures.”

Thank you for joining us today, Dr. Raoof.

Dr. Sana Raoof: Thank you so much. It's lovely to be here.

Dr. Davide Soldato: So, Dr. Raoof, I just wanted to start a little bit about the theme of your article, which is really centered around multi-cancer early detection tests. And this comes from the results of several studies that showed their reliability and efficacy in identifying cancer in the average risk population. But I just wanted to ask you if you could give us and our readers a brief overview of how these tests work and how they were designed for this specific population.

Dr. Sana Raoof: Of course. Well, there's an interesting story. The origin of multi-cancer early detection tests actually begins with insights that come from the field of obstetrics and gynecology. So about six or seven years ago, in the peripheral blood of pregnant women, we discovered that you can actually find fetal DNA floating around. And that was an early discovery of cell free DNA coming from the baby into the mother's bloodstream. But in some of those young, otherwise healthy women, we also discovered that there's another clonal signal, unfortunately not coming from the fetus, but coming from an undiagnosed tumor. And that led to the entire field of circulating tumor DNA and all of its applications.

Of course, scientists in the last six or seven years have harnessed the fact that DNA and the methylation patterns on the circulating tumor DNA, as well as other analytes like glycosaminoglycans, proteins, and other analytes, are secreted by tumors into the peripheral blood in order to try and screen for tumors, hopefully at early stages, when there are still curative, definitive interventions that are available. There's several different tests now that are providing the ability to detect cancers at many stages, including early stages. They're in different phases of preclinical to clinical development, and one is even commercialized and available by prescription in the United States.

Dr. Davide Soldato: Okay. So I think that in most of these tests, they really look at the tumor DNA, so they identify mutations or, for example, methylation patterns. But do we also have some tests that integrate some other type of biomarkers that we can identify in the blood? Like, are they integrated all with the others, or are we just relying on circulating tumor DNA?

Dr. Sana Raoof: It's a great question. There's a lot of really fascinating biology that different companies predominantly are using in order to find signs of early cancer. One of the analytes that I find really interesting, other than looking for small variants in circulating tumor DNA and looking at methylation patterns, as you mentioned, is looking at fragment length. So, for example, the company DELFI looks at the different patterns of the length of DNA fragments that are floating around in the peripheral blood. And not only is fragment length tissue specific, so in theory, a fragmentomics based multi-cancer early detection test could tell us what is the tissue that this aberrant signal is coming from, but they can also tell you if there's likely a cancer present, because there's a difference in fragment length patterns in cancer versus non cancer.

There are also other analytes. I mentioned glycosaminoglycan. There's another company that doesn't yet have prospective data, to my knowledge, that is making a test that looks at these analytes instead. There are other companies, again, without prospective data yet, that are looking at circulating tumor cells. And I'm sure that in the next few years, we're going to start getting prospective data from all of these players and also hear about other analytes that scientists have found can predict cancer from non cancer and maybe even protect tissue of origin based on artificial intelligence.

Dr. Davide Soldato: So you mentioned artificial intelligence. So, basically what you're suggesting, but correct me if I'm wrong, is that when we use this test, we are actually measuring something in the bloodstream, but at the same time, we are actually applying some type of artificial intelligence to actually interpret these results and then give us the definitive results, or what we would call like a positive and a negative of the tests, is that right?

Dr. Sana Raoof: Yeah, absolutely. And it's an important distinction that you're making, we are measuring something in the blood, but we're not just measuring it. We're using machine learning algorithms that have been trained on thousands and thousands of patients with cancer and thousands and thousands of patients without cancer, and have measured various analytes and analyzed the patterns, for example, of DNA sequence, or bisulfite sequencing of methylation patterns of patients with and without cancer, and have been trained to look for the differences between them. And so the analyte that we're looking for is not a specific mutation per se, but is a pattern that looks like patterns that you typically find more so in cancer patients.

There's many different companies, they are trained on different types of cancer. So some companies, like GRAIL, have a test that looks for a very expanded list of over 50 cancer types. Other tests have a narrower focus and were trained and validated on a smaller list of cancer types. So there's just a great diversity in this space. These tests are trained to look for different types of cancer. They're trained and validated on different populations of interest. So, for example, some of the populations that these tests were trained on are predominantly white, and that will have impacts, potentially on how these tests perform in non-white populations. And that's a really interesting area of future research. These tests may or may not have included cancer survivors in their populations, and that could ultimately impact how these tests perform in those populations.

So there's just so much to learn, so much data that's going to be coming out in the next few years from all of these different key players in the multi-cancer early detection space. But one thing that I'm sure of is between all of the different analytes, all of the different training and validation studies, and all of the different prospective studies, we're going to learn a tremendous amount about the potential clinical utility of using multi-cancer early detection tests to complement the few standard of care surveillance cancer screening tests that we have recommended today.

Dr. Davide Soldato: So just taking a step back and going back to the fact that we actually use machine learning algorithms to identify a pattern that can give us an idea of whether cancer is present or not, I believe that there is also some room for calibration of these types of tests. And I think that this is one of the key arguments that you make in your paper where you say that we can actually personalize a little bit more these types of tests to understand and then to decide what we are looking for. Is that correct and can you expand a little bit on that?

Dr.Sana Raoof: Yeah, absolutely. This is the central concept of the paper that we're discussing. Because these tests are machine learning based, as I said, they're trained to say cancer versus not cancer, and some of them are further trained to say, coming from this organ or coming from that organ. But what does it mean to say cancer or not cancer? There are specific thresholds that are defined to say, above this threshold of signal detection, we're going to say this is a positive cancer signal detected, and below it we're going to say negative. And so right now, these tests are kind of designed to have this binary output, and the concept that I wanted to put forth in the paper is it doesn't necessarily have to be binary, and the thresholds don't have to be static. So, for example, you can imagine that in an average risk population where the pretest probability of cancer in your lifetime for Americans, it's pretty high, roughly 40% for lifetime. But at any given moment in time when you're getting a test, it's lower. For example, in Americans, 50 to 80, the chance of having cancer at any given moment is just under 3%. So you don't necessarily want a test that is very nonspecific, you don't necessarily want to tell a lot of perfectly healthy people that are asymptomatic screening populations that they have cancer if they don't. And so these tests were designed to have very high specificity, predominantly across the board, across the different companies making them at the cost of, in some cases, having lower or moderate sensitivity in early stages.

And it's important to keep in the back of your mind that we cannot ever expect the types of early stage sensitivities from multi-cancer early detection tests that we're used to thinking about for single cancer screens that are just optimized for one single organ. They work in a completely different way. So I don't expect a future where the sensitivity of a mammogram, which is only for breast cancer, is going to be analogous to the sensitivity of a blood-based test that's looking for all cancers in your entire body. I don't think it's fair to expect that. But I do think it's possible to imagine a future where we do change the thresholding of these tests that were trained and validated in average risk screening populations, and say, “Let's turn the knob on the dial and let's take the sensitivity a little bit higher, even if it means the specificity drops from 99%, for example, which is the very high number of the gallery test, down to 98%, down to 97%. Let's see how this affects the positive predictive value and the negative predictive value of the test.” And how having a higher negative predictive value by having a higher sensitivity may or may not make it more clinically useful for higher risk populations that have higher pretest probabilities, in which case we are kind of more interested in being sure that we're ruling out cancer.

Another concept that I talk about in the paper, aside from just turning the knobs, is to make it a continuous variable rather than a binary report. Rather than saying signal detected or not signal detected, I can also imagine a future where we personalize the output of multi-cancer early detection tests to return a score, for example, from 1 to 100 or 1 to 10, and give physicians the ability to use that continuous variable in addition with other clinical findings, physical exam findings, other labs, symptoms, patient’s past medical history, family history, all of that together to make decisions about should we pursue further workup, should we do an invasive biopsy. This is kind of the way that we use other scoring tests in oncology, like the oncotype tests for breast cancer, decipher test in prostate cancer. And I think physicians like having continuous variables to work with and to help them make very personal decisions for patients' diagnostic workups.

Dr. Davide Soldato: To summarize a little bit, what you're arguing in the paper is that we could potentially modify a little bit these tests as they fit the type of population that we are looking for. For example, if we are looking at the average risk person in America, there we just want to be sure that we are just doing additional workout and additional follow ups and additional invasive procedure, for example, biopsy, when we have a very high probability of finding that cancer. At the same time, if we have someone who has a baseline risk which is higher, like cancer survivors, in that case, we are more interested in seeing if there is really cancer at that point, and so we can increase the sensitivity and go down on specificity, but still looking at the overall outcome that we want to have for that specific patient.

One thing that I was wondering is, do you also see a future where we personalize a little bit more also including additional information that comes from risk factors, environmental or behavioral patterns, type of diet, or these types of risk factors that we already know from epidemiology are associated with a higher risk? So could we potentially customize this test even more, saying, this patient has a higher risk of developing colorectal cancer, so could we look more specifically to that specific cancer type and that specific risk compared to tobacco associated cancers, that for that specific patient, they are not so relevant?

Dr. Sana Raoof: What you're saying is actually a fascinating and really compelling idea, and it reminds me of the way that noninvasive prenatal testing works. So, again, back to the world of obstetrics and gynecology, you have a woman at the end of her first trimester having fetal DNA testing to look for chromosomal abnormalities. And when you order that test, you actually do put in various features about the woman to help you understand her baseline risk for carrying a fetus that has chromosomal abnormalities, including her age, the status of her other children, and other things in order to help you calculate a pretest probability. And so after that, the non invasive prenatal test takes that into consideration and returns a probability of carrying a fetus that might have those aberrations, and it's not a binary risk. It's, as I said, a continuous variable.

So I think what you're proposing actually goes beyond what I wrote about in the article. I think it's a fabulous idea. And I think that in the near future, I can imagine that as natural language processing is exploding, and in general, large language models and the ability to extract features about a patient from the EMR are exploding, we might have a better stratification in general of patients into average risk, low risk, high risk, and really high risk, using EMR data, using real world data that could help us feed a really accurate picture of a patient's pre-test probability into this test, so that these tests could be further refined and further trained and validated on patients, taking into consideration more factors and help us improve the predictive power of the tests as they're returned in a report to the physician. So I think maybe you should even write an article about the idea just proposed. It's a great idea.

Dr. Davide Soldato: So another aspect that I was really interested in is I've looked at one of the papers that you cited, and I wanted to discuss this with you as you are an expert on the topic. In one of the articles that you cited that used this type of test, they identified some of the cancers that we also normally identified with standard screening procedures, like breast or lung or colorectal. So for those cancers, we add a certain proportion, or like, for example, for breast cancer, a higher proportion identified with conventional screening. But still we had some other cancer that eluded those types of screening and were identified using liquid biopsy tools. So do you envision a strategy where we would use the screening methods that we already add as a complement to those liquid biopsies, or do you think that someday liquid biopsy could potentially completely substitute standard screening procedures?

Dr. Sana Raoof: I think we're too far from a day where liquid biopsies are going to replace standard of care screens. The scope scans and smears that the United States Preventive Services Task Force has recommended are gold standard screening interventions because, number one, for all of them, except for cervical cancer screening, we have randomized data with definitive endpoints that tell us that there are mortality benefits from doing those screens. We don't have that type of data yet from the world of multi-cancer early detection. And as we talked about earlier in this podcast, those tests are kind of designed with a different approach where they have higher sensitivity and much lower specificity than multi-cancer early detection tests.

So I think that the molecular cancer screening companies have done a very careful job of creating tests that are really more optimized to be complementary tests rather than a standalone catch all test, to have higher specificity at the cost of lower sensitivity. So I don't imagine a near future, at least not in my career, where we're going to stop doing colonoscopies and mammograms and pap smears. I don't think that that's going to happen. But I do think that whereas right now 75% of cancers that Americans die from, we lack cancer screening mechanisms for them, I think that that number has the potential to really drop. If in the next few years, one of these multi-cancer early detection tests is ultimately approved and covered, then I think that a lot more cancers could be detected by screening rather than by symptoms, and we might ultimately see a big stage shift.

Dr. Davide Soldato: Yeah, I think you're absolutely right. In the same article that I was mentioning before, there were several of those cancers which can be lethal if diagnosed at an advanced stage, that were diagnosed at an early stage, for example, ovarian cancer, bladder cancer. So I really think that we really have potentially the way to screen, or at least have a signal for cancer that currently we just diagnosed when symptoms associated with higher stage appear.

But moving on to turning the knobs on this type of test, and so going to the higher risk population, for example, cancer survivors, which is something that you speak a lot about in the manuscript. So you also discuss a little bit the question of whether we should use multi-cancer testing versus single cancer testing. So are we looking at a specific recurrence from that specific tumor, or are we looking at a general risk of cancer in a population that has a common risk factor, like tobacco? And so I was wondering if you think, and this is probably just your perception or just your opinion, that that is another way that the physician should turn the knob. Should we evaluate the risk of those cancer survivors and say, in this specific patient right now, the risk of recurrence is higher so I should use or I should be more in favor of a test that is more centered on the risk of recurrence versus I have a general risk of several cancers that could appear, and so should I use something that is more multi-cancer? This, of course, is merely speculative because we still don't have definitive data regarding the efficacy of this test. But it is just your perspective on this type of approach in the near future or not so near future.

Dr. Sana Raoof: Well, I think if we're speculating, then I think that the fantasy situation for any oncologist is that you have two types of liquid biopsies. One is a multi-cancer early detection liquid biopsy. And it would be great if you could select whether you want it to be optimized for highest NPV, negative predictive value, or highest PPV, positive predictive value. And then you also have a host of single cancer screening liquid biopsies that can help you specifically figure out if there's a recurrence of a single cancer type that you're suspicious about.

So, for example, in the article, I talk about how there will be clinical gray areas, and it's not always going to be obvious which test you should reach for. But one example that I think we can all relate to in the oncology community is you have some indeterminate imaging finding, and you don't know what to do about it. So, for example, you have a woman that has a history of breast cancer, has had no evidence of disease for a few years, now, has back pain. You do a spine MRI, you see a lesion. Maybe it's an atypical hemangioma that's causing pain, maybe it's a breast cancer metastasis. You're not sure. What should you do? Should you do a biopsy of that lesion in the spine? Should you wait and see if it grows and do another MRI in two or three months? What are your options? And so in this situation, I think we can all agree that if you had a liquid biopsy that was optimized for really high sensitivity, specifically for breast cancer, and had a very high negative predictive value, and if it came back negative, then in that setting, it might help you avoid an invasive test, like a biopsy in the spine, and give you a little bit more comfort as a physician to say, “You know what? I'm going to come back in two or three months and do another spine MRI. I'm going to see how this woman is feeling, and I don't need to biopsy this right now. Maybe it really is just hemangioma.”

Dr. Davide Soldato: And in this specific setting, let's take the same patient. So it's a female patient, she had a previous diagnosis of breast cancer. Do you think that there is a difference between tumor-informed tests, really based on the molecular aberration that the primary tumor had for these women, versus just a standard test that gives us information regarding the presence of breast cancer cells or not? And if you think that there is a difference, what would you think would be the advantage of one? And the disadvantages, for example, is a tumor informed essay more complex to obtain? Do we need more time? Is it more expensive versus a commercial test that is already available or something like this? This is my understanding as someone who's not so much in the topic, but I think that this is a point that many oncologists probably wonder about, and probably we should speak a little bit more about with someone who is an expert on the topic.

Dr. Sana Raoof: Absolutely. And I think that you've actually hit all of the major points on the head. So comparing a tumor informed versus a tumor agnostic test is like really comparing apples and oranges. A tumor informed test where you're starting with a patient's pathology and you are looking specifically for mutations and other molecular features that you know the patient has in their tumor, is going to, of course, result in a test that is, number one, more expensive and harder to make, but also, number two, more sensitive, more specific, more predictive, and in every way probably just more powerful than a test that is, in general, optimized for a single cancer type, but is almost certainly going to be trained and validated on people with a mix of histologies, a mix of molecular features, and will not be as sensitive or specific as a test that is actually informed by that single individual's tumor. One of the things that matters a lot to me is health equity in oncology. There are just huge disparities in outcomes in patients that are advantaged and disadvantaged. And it stems from lots of different things. In no small part, it stems from later stages of diagnosis in disadvantaged patients, and then even once you have a diagnosis, delays to confirmatory workup, delays to starting treatment, disparities in the treatments offered.

I don't imagine a world where everyone on earth is going to have access to tumor-informed liquid biopsies. I do imagine a future where tumor agnostic liquid biopsies, both for single and multi-cancer screening, should be a lot more economical than they are now, and should be more available for multiple cancer types, and should be more available to patients that aren't at just the Memorial Sloan Ketterings and the Dana-Farbers of the world. And so I do think that those types of off the shelf tests have the potential to really revolutionize the way that we work up suspicion of cancer, not just in advantaged patients, but also in patients that are diverse, in patients that are not at academic cancer centers, but at other cancer centers around the world. And I think it's a really exciting prospect.

Thinking about the chance of recurrence in the breast cancer patient is a perfect example of when you want to test that is optimized just for breast cancer, because you see something in the spine, you know her history, and you're less worried about a new primary and a new MET from that primary. But there are other situations that are also interesting to consider. For example, patients that have had lung cancer and have a history of smoking, because they've had a history of smoking, they're actually at risk for a dozen different cancers, not just lung cancer. And when you think about what we do to follow lung cancer survivors, we're just doing CTs of their chest and of course, physical exams. But the vast majority of cancers that people with lung cancer history will get may not be present in the field of view of a CT of the chest. They may also get renal cancer, bladder cancer, they might get leukemias, they might get pancreatic cancer. So there are a lot of things that you're not going to catch in a CT of the chest. And so in that situation, you care not only about recurrences, which in thoracic oncology, it's kind of a gaussian probability distribution, where the tail is almost close to 0 after five years, but also a uniform distribution of roughly 3% per year of a second cancer, a new primary cancer that goes on for the rest of their life. And so in that clinical setting, you can imagine that having an off the shelf multi-cancer early detection test may be dialed up for higher negative predictive value, would be extremely useful.

Dr. Davide Soldato: Yeah, I totally agree, but thank you for clarifying these points, because I think that there is a little bit of confusion also in the oncology community, as this type of tests, they're also based on very complicated molecular biology, sometimes could be potentially integrated, and we could potentially integrate them in the clinic.

And so I wanted to close up with kind of a personal question. I was wondering how you came to be so interested in this field of molecular screening or early diagnosis and prevention associated with molecular data.

Dr. Sana Raoof: Well, it's an interesting story. I did my MD PhD at Harvard Medical School, and my PhD was in the opposite world from molecular cancer screening. I was designing drug combinations that could be used in advanced oncogene mutant lung cancers. And I thought I would become a medical oncologist and spend my life designing new systemic therapies for advanced malignancies. And what I saw every day in the lab during my PhD is drug resistance emerges and it's a process of evolution by natural selection happening on a cellular level. And although we have some really great slam dunk drugs that come to mind, for example EGFR inhibitors in certain lung cancers, immunotherapy in melanoma, on average, the median overall survival gain from all of the FDA approved drugs in the last 10 years is roughly two months.

By the end of my PhD, I really started feeling like, is the best use of my life to continue fighting a battle against natural selection in cancer cells, or is it a better strategy, to me, it seemed like a more sensical strategy to just try and find cancers in these patients earlier, when you don't have to engage with the complex signaling mechanisms of a cancer cells biology, and instead can just provide a definitive local intervention, like surgery or radiation, which already is curing many patients with non metastatic cancers. And as I looked around the world, I just didn't see that many people investing heavily in early detection research at the time. It was the very early days of multi-cancer early detection. And so I became involved with all of the groups, the companies, the organizations that were developing these tests, and really fell in love with, number one, just the concept of the tests, the concept of multi-cancer early detection, rather than single cancer screening alone, because no one knows what cancer they're ultimately going to get. But I also really fell in love with methylation biology, fragmentomics. I fell in love with the types of clinical trials that were being designed and the new types of endpoints that we have to think about when we're designing clinical trials for a multiverse of single cancer screening. And it's just such an exciting time in that community, it's the early days. So that's how I came to this space, and it's just the perfect time to be in this space, because everything is exploding.

Dr. Davide Soldato: Thank you very much. And thank you also for sharing the personal side of the story.

Dr. Sana Raoof: Thank you so much. I'd like to thank Razelle Kurzrock, who's an amazing medical oncologist who's worked with me on two really fun papers so far, one on real world data, and this one on turning the knobs on liquid biopsies. It's always great to bounce ideas around about multi-cancer early detection with friends and collaborators, and Razelle did an absolutely amazing job helping write this piece.

Dr. Davide Soldato: So this brings us to the end of the episode. Thank you Dr. Raoof, for joining us and sharing more on your JCO article titled, ”Turning the Knobs on Screening Liquid Biopsies for High-Risk Populations: Potential for Dialing Down Invasive Procedures.”

If you enjoy our show, please leave us a rating and review, and be sure to come back for another episode.

The purpose of this podcast is to educate and to inform. This is not a substitute for professional medical care and is not intended for use in the diagnosis or treatment of individual conditions.

Guests on this podcast express their own opinions, experience, and conclusions. Guest statements on the podcast do not express the opinions of ASCO. The mention of any product, service, organization, activity, or therapy should not be construed as an ASCO endorsement.

Disclosures:

Dr. Raoof

Stock and Other Ownership Interests Company name: Illumina Radiopharmaceuticals Honoraria Company name: AstraZeneca Consulting or Advisory Role Company name: Verily Company name: GRAIL Company name: Exact Sciences Travel, Accommodations, Expenses Company name: Grail