Sylvia Plevritis: Better cancer treatment through data
Biomedical data scientist Sylvia Plevritis is an expert in computational modeling of cancer risk and treatment options hidden in the remarkable quantity of data available today.
Rarely is a tumor made up of a single mutation, she says, but more commonly of a mix of different mutations. Such heterogeneous tumors may require complex combinations of drugs to produce the most effective treatments. That’s where computers can help.
Using mathematical simulations, Plevritis is helping patients and their doctors understand the genetic makeup of a given cancer for the purpose of identifying drug combinations that stand a better chance of success. Some of the models Plevritis works with can be run in an hour or less and yet return invaluable guidance that can save a patient’s life.
Plevritis says these computational approaches can even help those without cancer understand their inherent genetic risks to assess whether and when additional screening or risk-reducing interventions are warranted.
Join host Russ Altman and biomedical data scientist Sylvia Plevritis as they dive into the promising intersection of computers and cancer care. You can listen to The Future of Everything on Sirius XM Insight Channel 121, iTunes, Google Play, SoundCloud, Spotify, Stitcher or via Stanford Engineering Magazine.
Russ Altman: Today on The Future of Everything: the future of cancer. Now I don’t think I need to tell anybody cancer is a feared diagnosis, it kills millions of Americans and people worldwide every year. Breast cancer, lung cancer, prostate cancer, brain cancer, these are things that people worry about and fear, and unfortunately too many people suffer from.
Now certain areas of cancer have seen amazing progress in the treatment, detection and treatment. Hodgkin’s Disease, some forms of leukemia, even these common ones: breast cancer, prostate cancer, have seen some progress. But many of them remain difficult to treat.
And the things that make cancer difficult to treat are protean. First of all, you need to kill pretty much every cancer cell because they can continue to grow if not. They are very good at avoiding the immune system, so the degree that your immune system can help kill your cancer, they’re diabolically smart at avoiding that. They take advantage of surrounding cells to create little walls around themselves. They even can direct the creation of new blood vessels to help get the nourishment that they need to grow. Of course, when you do treat cancer with either radiation or chemotherapy, they can develop resistances because the DNA mutates and figures out a way, again, to evade the chemotherapy. And this makes it hard for physicians who are trying to use combinations of drugs, it makes it hard for them to know which combinations will work the best.
Now believe it or not, and I bet you you’ll believe this, computers have become an important part of the fight against cancer. I don’t think we should be totally surprised because we’ve been seeing computers revolutionizing so many fields. But you can use computers to model the molecular and cellular changes associated with cancer, figure out what’s going wrong, and then simulate treatments to see what might work. You can use them at the clinical level to say who should we screen, what would be the - if we had this policy, policy X, what would that policy mean for the number of cases we would capture and the number of people we would cure? Would it be worth the investment of society to do that kind of screening program? And you can also use these computer algorithms to make tough decisions about treatments, screening, all of these kinds of things.
Well my guest today is Professor Sylvia Plevritis. She’s a professor of biomedical data science and radiology at Stanford, she’s also the chair of biomedical data science. Sylvia, how is big data, as people have been calling it, how is that transforming cancer research. Is it like the way it’s transforming so many other areas of society, where the availability of data is bringing new opportunities?
Sylvia Plevritis: Oh, absolutely. Russ, first of all, thank you for having me. This is really exciting to be here. Absolutely, so big data is transformative in cancer research in many different ways. First of all, we are collecting, through clinical records, data on patients longitudinally. And it’s kind of a first ever kind of experiment in a way to do this on patients who are going through usual care, not in a clinical trial setting. And so we are sort of recording observations and linking them in ways we’ve never done before. And the other thing we’re doing is, for example, for cancer patients their tumors, we’re profiling them more deeply. So on an individual tumor, for an individual patient, we’re collecting what we call multiomic data.
Russ Altman: Multiomics?
Sylvia Plevritis: Multiomics.
Russ Altman: And what does that mean?
Sylvia Plevritis: Yeah, so it’s different types of omics. And so we have the genomics, which could be mutational, transcriptional, epigenomic, methylation level. We have proteomic data. And we’re also collecting high throughput data on post-translational modifications. They’re not quite at the omics level but they’re getting there.
Russ Altman: So when you say post-translational, this means after a protein is made in the cell there’s modifications made to it that might effect its function or it’s importance for the cancer growth.
Sylvia Plevritis: Exactly. Phosphorylation, glycosylation, many things that we’re really just observing with new technologies now.
Russ Altman: Are there areas of cancer where this is standard of care, or is this still early days of research but it’s not reaching the patients out in the community?
Sylvia Plevritis: It’s beginning to reach the patients in the community. There’s a number of investigators who are now taking patients who are kind of late stage in their disease course and throwing the book at them.
Russ Altman: In a good way?
Sylvia Plevritis: In a good way. Like literally profiling with every technology we have and creating these really comprehensive new tumor boards that have computational scientists in them, too. To try to leverage what we know about signaling pathways and try to be more strategic. Not that people aren’t strategic, but more creative in their strategic approach to try to understand what this tumor is really doing, how it’s setting up its defense mechanisms.
Russ Altman: So let me go back, ’cause you said a few things and I wanna make sure it’s clearly defined. So first of all, tell me about a tumor board. What is a tumor board? And tell me what they used to be like and what they’re like these days.
Sylvia Pleviritis: Yeah, that’s a great question. So a tumor board is composed of a variety of clinicians that manage a patient. So you have everyone from the oncologist, radiologist, even anesthesiologist, every point of care that a cancer patient sort of experiences. You have the professional who comes into a room with everyone else and goes through a comprehensive evaluation.
Russ Altman: So this is to make sure that nothing squeaks through the cracks and that everybody’s on the same page with respect to the treatment plan for this patient?
Sylvia Pleviritis: Right, and this is integrated care model. So the coordination of care is really important through, and also the discussion and different perspectives, and what are we gleaning from different fields that we may bring into the management of an individual patient. Maybe there’s new trial evidence that somebody brings into the discussion. And these tumor boards are changing, they’re becoming also very molecularly oriented.
Russ Altman: I see.
Sylvia Pleviritis: And so we’re bringing in genomics, for sure, into the conversation. And more recently, we’re bringing in, sort of, even some new ways of looking at the individual tumor through computationally derived understanding of the tumor.
Russ Altman: So you might actually have a computer scientist at your tumor board making contributions?
Sylvia Pleviritis: Absolutely.
Russ Altman: So that sounds like a change to me. And that’s because this data is not the kind of data that the traditional clinician can just look at a simple graph or a simple chart and say oh it’s getting better or it’s getting worse. These are pretty complicated analyses that have to be done?
Sylvia Pleviritis: Right, and so you’re thinking about what’s really happening in the tumor from sort of a cellular pathway and a cell-cell interaction perspective, and so it’s a more complex phenotyping of the individual tumor to understand how you really want to go about and try to eradicate that tumor.
Russ Altman: This is The Future of Everything, I’m Russ Altman. I’m speaking with Professor Sylvia Plevritis about cancer and the clinical treatment of cancer, and the new ways of forming teams to make sure that patients get the best treatment. Now another thing that I wanted to ask you about is you mentioned the multiomics, and you said genomics, transcriptomics, all of these kind of molecular measurements. Those are all kind of measurements, but how does it help the patient? So what do you learn from those measurements that might lead to a change in the treatment?
Sylvia Plevritis: Right now what we’re really using is a mutational profile of the tumor. And that’s really actionable, so there’s drugs that go after specific mutations.
Russ Altman: So you see a mutation and say: that means drug X is more likely to work than drug Y, let’s go with X?
Sylvia Plevritis: Right. But we also know that the tumor is really heterogeneous, and that not maybe all the cells have that mutation, or maybe that have more quote unquote driving mutations. And that the mutational profile itself is not the only indicator of the responsiveness of a tumor to a given drug. And so on a more basic science side, we are looking at other mechanisms that the tumor sort of infers. There’s resistance induced, even by the drug. And it may not be driven by the mutation, it may be more epigenetic or through these post-translational modifications. And so we recognize that there is a complexity beyond the mutational landscape and that we need to understand that complexity to understand the mechanisms of drug resistance and then to think about how to combat those mechanisms of drug resistance.
Russ Altman: Yes, and I’m glad you mentioned this because I wanted to ask. I noticed that you had published papers about drug combinations, and so in many areas of medicine, and cancer is definitely one of them, we have learned that sometimes it’s better to hit the disease with more than one medication at a time. You can kind of get two or three shots at the same time. So tell me, what are the challenges for combination therapy in cancer and how does your work kind of help inform it?
Sylvia Plevritis: Well combination therapy is a way we think that we can use existing drugs in a more effective way. And so when you put together FDA approved drugs in a combination that have not been tested under clinical trials setting, they’re still considered off-target use. And so we don’t often understand what the combination is going to do, even though the individual drugs have been approved.
Russ Altman: So it might not be approved in combination, even though each individual one? But doctors are allowed to make these combinations?
Sylvia Plevritis: Yes, they are. It’s off-target use, but when you put drugs in combination, to be certain that it’s an effective combination, you wanna run clinical trials. And now think about the combinatorics of this question. Every combination in a clinical trial setting, people say that we don’t have enough patients to do all of these clinical trials to test all the combinations, especially now with immunotherapy. So the question that is posed to computational scientists is: how can we come up with the more likely combinations, the combinations more likely to be effective, so that when we do these trials we’re more likely to be successful, or maybe we could do the trials in a more efficient way.
Russ Altman: So make them pass the bar computationally before you start exposing them to patients?
Sylvia Plevritis: Exactly. And so I have been very interested in this question, and I’ve also been interested in sort of thinking about what combination to put together while you take into account the heterogeneity of the tumor. Oftentimes, the combination is there to directly counter the fact that the tumor is heterogeneous. So you might think: okay, the tumor has maybe different driving mutations, so that makes sense. But I think about it slightly from a different perspective. I think that the tumor has different cells, and different cell types, and so that the combination is maybe targeting different cell types within the tumor. And so the work that my lab does is characterize what are the different cell types in the tumor and how are these different cell types responding to the individual drugs? And so, of course, when you treat the tumor you’re treating all the cells at the same time, but when you account for the heterogeneity in terms of what the different cell types are in the tumor, we can show that that’s actually a more accurate way of estimating the response of the tumor to a given set of drugs or combination of drugs.
Russ Altman: This is The Future of Everything, I’m speaking with Professor Sylvia Plevritis about cancer and right now, cancer chemotherapy combinations. So you have your results, you worked very hard in your computational models, it kinda makes sense, you go after these different cell populations, how hard is it to convince the practicing oncologist to take up your ideas? Do they demand a clinical trial of a certain size or do they sometimes say: okay this makes sense, and I’m going to give it a try because I don’t see it too much downside? What is the process of translating your basic research inspirations and insights into clinical practice?
Sylvia Plevritis: Wow, that’s a great question. ’Cause I’m in the middle of that right now. And I’m very excited about trying to translate that. So this particular study that you’re referring to where we’re thinking about identifying combinations for the individual based on single cell analysis. I’ve been trying to do a real proof of concept. Like what we did is we kinda did a theoretical proof of concept, where we took archived samples. And this was in the context of pediatric ALL, and these samples were screened across a host of drugs. And the data was collected.
Russ Altman: So these are in dish, literally? Are they in a dish?
Sylvia Plevritis: Yeah so it’s a hematological malignancy. You culture the cells, you then expose it to the drug, and then you look at the signaling profile of the cells before they’re exposed to the drug and then after they’re exposed to the drug. And in our case, we’re really looking at a short-term stimulation, so like a 30 minute or one hour, maybe a couple hour, stimulation. So we get this short-term stimulation, and we look at the changes in the signaling at the individual cell level for each of the drugs individually, and we take that information to see: are these drugs targeting what we want them to target? Are they having the desired intercellular effect? And so we develop a computational model that identifies which combination of drugs, or actually more specifically, which minimum combination of drugs, has the maximum desired intercellular effect.
And so we found in this kind of sort of theoretical study, and I call it theoretical because we took these patient samples of samples that were archived. The patients got other drugs, you know, this is their five year old plus patient samples. And we exposed them to these new drugs that are just coming out onto the market now. And we made an assessment of which combination would be best for these patients, and interestingly, we found that for the vast majority of the patients, for half of them, it would be these two drugs. For the, almost the other half, it would be one of those two drugs that worked. But then for a really small group, like 10 percent, it would be a totally different drug. So it was quite interesting that you had this spectrum. You know, in a clinical trial…
Russ Altman: And going into this, you would have said these kids all have ALL, we can’t see that many differences between them. But then at the molecular level, you’re getting them —
Sylvia Plevritis: At the signaling level.
Russ Altman: Partitioning into these very different groups.
Sylvia Plevritis: Yeah, so at the signaling level we’re seeing the partitioning. So now we are working with colleagues at different institutions to actually to test combinations that they’re testing. To do a real proof of concept. So we’re not going quite into the patient yet to make decision making for the individual patient, but we are actively, we’re setting up a study right now that a certain group of patients, AML patients in this case, are getting combination therapy and we’re going to do this in parallel. So we’re going to guess what we think is the right, like in this case, if they would have been more responsive to one drug, the other drug, or this combination, but we’re also going to test other drugs that the patient isn’t getting.
Russ Altman: So this really ties everything together, ’cause now I can see that all of these omics measurements allow you to characterize these patients and then based on the subsets that you observe, you can do basically very personalized treatments based on the features that you see in that patient and not just based on overall statistics.
Sylvia Plevritis: Exactly.
Russ Altman: And it sounds like you have found some clinical collaborators who are at least willing to consider doing this in real time on real patients with real disease.
Sylvia Plevritis: Yeah, it’s very exciting.
Russ Altman: This is The Future of Everything, I’m Russ Altman. More with Dr. Sylvia Plevritis about cancer and moving on to screening, next on SiriusXM Insight 121.
Welcome back to The Future of Everything, I’m Russ Altman. I’m speaking with Professor Sylvia Plevritis about the future of cancer treatment and detection. So in the first part of our discussion we talked about this molecular stuff and how you were using these amazing ways to get personalized treatments. Another area that you’ve worked in, however, is computational models of the impact of screening and early diagnosis. So tell me, what are the big challenges? It seems obvious, you know. Of course we wanna screen, of course we wanna find cancers as soon as possible, because then we can get to the doctor and start treating them. Is it more complicated than that?
Sylvia Plevritis: On one level, no it’s not more complicated than that. We wanna find cancer early. It’s a much less complex disease, it’s easier to cure, it’s more localized. So it’s not more complicated than that. But it’s a complicated question, because screening is a population-level intervention. And introducing intervention at the population level in itself is a complex process. So as we think about what are the best screening guidelines in terms of who should we screen, when should we screen them, like what age should we start screening them, when should we stop screening them, how often we should screen them? Every year, every two years? It becomes sort of a policy question. Like what’s the best thing on a population level?
Russ Altman: And I would guess you’re balancing expense, somebody’s paying for all this screening, but also the benefit in health to the population, and those might not always be going in the same direction.
Sylvia Plevritis: That’s part of the complexity in the discussion. So what my lab tries to do is try to quantify the benefit. And so we have been almost for 20 years now modeling the impact screening interventions have had on reducing mortality rates. So we have simulation models that we can answer questions such as: if you didn’t have screening, what would the mortality rates have been?
Russ Altman: Oh wow. That really would be informative for the policymakers and hopefully it hasn’t been bad news. So what have you been finding?
Sylvia Plevritis: So screening goes hand in hand with treatment. So when you screen, you also have to measure what treatments people would get at the time of, let’s say, screen detection. Or if they were missed at the time of screening, what treatments they’re getting at the time of what we call symptomatic detection. And so as you think about the benefits of screening, you have to account for the fact that treatments are advancing as well. Because screening itself is beneficial after you treat the patient.
Russ Altman: Right, you’re not just doing it for fun.
Sylvia Plevritis: Well it’s not that you’re not doing it for fun, the effects of treatment are changing over time, too. We have new treatments that are molecularly specific, as we talked about kind of in the first part. And so as you think about the benefits of screening, the point I’m trying to make is that: great, you screen detect a patient early but now you have to treat the patient and how effective is that treatment? So these things go hand in hand in overall mortality reduction.
Russ Altman: So what I’m hearing is it could be that a screening program makes perfect sense for a certain period of time, but as the capabilities for treatment change, that screening program may or may not change in its cost-effectiveness or whether it’s really desirable, is that a possibility?
Sylvia Plevritis: So some people are having those conversations right now, and those are complex conversations: the interactions between screening and treatment. They’re complex conversations. Our computer models what we try to do is actually tease those apart at the population level. When we answer the question what impact screening had on mortality reduction, we’re also answering the question what impact treatment had. What if we run these simulation models, let’s say you didn’t screen the population but you had all these advances in treatment, what would mortality reduction have been? Or what if you never had the advances in treatment and you really had only these advances in screening, what mortality reduction would have been. Through these type of counterfactual analyses we’re able to tease apart the contributions and the interactions between screening and treatment.
Russ Altman: And then that information goes to who? Who are the deciders who then say: thankfully Dr. Plevritis has done this study, and we’re now gonna make this decision? Is that the Feds or is it individual healthcare institution organizations? Who acts on these advice?
Sylvia Plevritis: Yeah, so that’s a great question. So I’m part of a consortium, and this consortium is called CISNET. It stands for Cancer Intervention Surveillance Network Modeling. I’ve been part of this consortium for 20 years. You know, a model is just a model. And so the purpose of consortium is to have multiple modelers sort of address this question that we’re talking about, so then we look at robustness in our results. We often get asked by a number of policy organizations to model specific guidelines that they may be interested in exploring. We often work, for example, with the US Preventative Service Task Force, which is the Feds. And, you know, we model scenarios for them to help as it’s sort of what they call secondary evidence that they use in their deliberations as they look at primary clinical trial data, other observational data, they also look at our simulated outcomes under different scenarios that they may be considering. Oftentimes they ask us to model screening policies, like a thousand different screening policies. We could do that with our simulation models, and then we could, you know, start ranking the results.
Russ Altman: This is The Future of Everything, I’m Russ Altman. I’m speaking with Professor Sylvia Plevritis about computational models for screening. So let’s go, I’m sure people are wondering, where are the most questions these days about screening? Which ones have been established and everybody should be doing it, and which are the ones where people are still trying to figure out what the best practices should be?
Sylvia Plevritis: Yeah, that’s a great question. So I would say one of the most active areas is getting personalized screening guidelines. So instead of this sort of one size fits all, really thinking about individuals that may need to have more screening or different types of screening tests that may be a little more invasive than others. And others who don’t really need a lot of screening or maybe more infrequent screening.
Russ Altman: And this is based on like a combination of family history and their genetics?
Sylvia Plevritis: Exactly.
Russ Altman: Which says you’ve had a lot of cancer in your family, we think we should keep an eye on you, versus there hasn’t been cancer in two or three generations, you should worry about your heart attack, not about your cancer or whatever?
Sylvia Plevritis: Yeah and it’s also behavioral, like if they’re smoking. So there are behavioral risk modifiers that we also consider in addition to the ones that you mentioned.
Russ Altman: So I’m sure that, you didn’t mention this explicitly, but implicit in your comments were when you do screening the one, I don’t know if we want to call it a disaster scenario, but the one bad thing would be what we call a false positive. So you get a screening test and it comes back positive. And typically then there’s a follow-up test that’s usually more expensive and more precise to see if you really have that problem. But for anybody who’s ever had a positive screening test, the period between when you find out about this positive and then when you get the follow-up can be incredibly stressful. And I know that you’ve worked to try to inform patients, you even created some online resources, to kind of help educate patients about their choices and the situation. Can you tell me a little bit about these kind of online resources to help patient education?
Sylvia Plevritis: Sure. So just a little bit about false positives. That is a very complex area. So there is, when we create these models, we associate some disutility with the false positives. But that’s very patient specific, too. And some patients will be willing to undergo false positives if they know that because they had the screening exam another patient benefited and their life was saved. So how we characterize false positives is very complex, and I think we really need to talk to patients and understand how their experience is with dealing with false positives.
Russ Altman: It’s kind of the same theme of personalization.
Sylvia Plevritis: Exactly.
Russ Altman: It’s not one size fits all.
Sylvia Plevritis: Yeah, not everybody reacts to the false positive the same way. And that’s something we have to take into account. The online tools that I’ve developed, one tool I’ve developed at Stanford, really addresses a different kind of question, not necessarily the false positive question. It’s a decision tool for women who are at high risk for developing breast cancer because they carry a mutation in the BRCA1 or BRCA2 gene. And they are looking at a number of different ways of managing their risk. So screening is one of them, but there are also prophylactic procedures that they can do. And they can also delay everything, right? They could just say: look, I’m young, I don’t wanna think about it.
So we give them a tool to help them, at given moment in time, think about they’re at a given age, they have a given risk, factors such as a BRCA mutation, and they have different options ahead of them. Sort of wait, sort of do nothing, screen, do prophylactic procedure. And then we show them what are the consequences of that decision in terms of their downstream risks. And in some cases, they can wait a little longer or they’re not incurring in the next five years, their risk isn’t gonna change significantly. But in other cases, they may need to take action, ‘cause in the next five years what’s ahead of them has more risk associated if they don’t take action.
Russ Altman: And this is out there being used?
Sylvia Plevritis: It is out there being used.
Russ Altman: Can you give us the URL?
Sylvia Plevritis: Yes, so it’s BRCAtool.Stanford.edu.
Russ Altman: BRCAtool.Stanford.edu?
Sylvia Plevritis: Yeah.
Russ Altman: And have you gotten any feedback?
Sylvia Plevritis: Lots of feedback. We’ve had to date about 45 thousand users, and we have people sending us emails from all over the country and even internationally that this has been very helpful to them, that this has helped them understand their risk and manage their risk. So it’s very satisfying to get these emails and this feedback from individuals who feel that this complex information was distilled in a way that they can make sense of it.
Russ Altman: Well there you have it. From molecules to screening to emails of thanks from grateful patients. Thank you for listening to The Future of Everything, I’m Russ Altman. If you missed any of this episode, listen anytime on demand with the SiriusXM app.