Emily Castellanos: Thank you to everyone for joining us today.
My name is Emily Castellanos, and I'm a medical director at Flatiron Health. I'm joined today by a few esteemed guest speakers, Dr. Razelle Kurzrock from the University of California, San Diego, Dr. Simon Papillon-Cavanagh from Bristol Myers Squibb, and Dr. Michael Taylor from Roche. You will hear from each of them shortly.
We will also have Dr. Jeff Venstrom, Senior Vice President and head of clinical development from Foundation Medicine, joining us for Q&A at the end of the webinar. Some quick housekeeping items before we dive in. First, I'd like to draw your attention to the Q&A option available throughout the webinar. If you hover over your screen with your mouse, you'll see an option for Q&A towards the bottom middle of your screen within the black bar.
At any point during today's session, please feel free to submit a question through this feature, as everyone will be muted for the entire webinar. You don't need to worry about confidentiality, it's only you'll be able to see your own questions. We'll be excited to have an open dialogue during the Q&A at the end of this presentation. And we'll do our best to answer as many questions as we can. Please know we will follow up afterwards via email if we don't get to your question. And also, please do feel free to reach out to us after the webinar if you would like to discuss any questions in more detail.
Now, if you notice your screen is zoomed in too far on this slide, please know you can click on view options at the top of the screen, and click fit to window to zoom back out. And then finally, if you have any technical questions or issues, please let us know via the Q&A tool, and we'll do our best to assist.
And then lastly, before we jump in, please do excuse any interruptions from our pets or loved ones. As like most of you, many of us continue to work from home. So during our time today, we will be discussing how advances in precision oncology are enabling novel research, starting first with a discussion of precision medicine in the clinic from Dr. Kurzrock, followed by our bio pharma speakers, Dr. Papillon and Dr. Taylor to talk about their experience using the Flatiron FMI clinico-genomic research database in support of pharmaceutical R&D.
Now before we get started, though, we would like to learn a bit more about you, our audience, you should see a poll pop up on your screen momentarily. If you have two monitors, please do check those monitors, it may appear on your other screen. You can select the answers that you believe are the most appropriate, and your responses will not be seen individually by other attendees.
The question for our webinar is, to what extent has the use of real-world data or RWD for precision oncology been considered or incorporated into your organization? Please select all that apply, you can choose more than one answer. First, RWD has been used for identification of prognostic or predictive biomarkers. Second, RWD has been used for drug target prioritization. Third, RWD has been used to characterize the natural history or unmet medical need of rare or biomarker defined patient populations.
Next, RWD has been used to provide contextual evidence supporting trials in rare or biomarker defined patient populations. Or last, my organization is not considered using RWD for the above use cases. So we'll give it about 15 seconds and then we'll share the results. Okay, let's send the poll and share the results.
All right. Very interesting. Looks like we have a pretty good spread across our different answer choices. Looks like most popular answer was that we've been using RWD to characterize natural history or unmet medical need and rare biomarker defined patient populations. And it's great to see that the vast majority of you here are interested in using RWD and precision oncology. Great to see all that interest.
So without further ado, let's go ahead and dive in. So I would like to turn the presentation now over to our first speaker, Dr. Razelle Kurzrock, Director of the Center for Personalized Cancer Therapy and the Rare Tumor Clinic at UC San Diego's Moores Cancer Center. Welcome Dr. Kurzrock.
Razelle Kurzrock: Thank you for having me here. Can you hear me okay? Yes. So I'm going to talk about impacting complex genomic and immune landscape, and doing this by looking at precision medicine in the real world cancer clinic. And these are my disclosures. And this is what I do. I'm very passionate about the Center for Personalized Cancer Therapy at UCSD. And this center has been devoted to developing clinical trials and precision medicine.
Before I was at UCSD, I was at MD Anderson Cancer Center. And I founded and built the department with early phase clinical trials. And then these patients that were really end stage metastatic patients, we began to have the first insights into what precision medicine could do for our patients.
Next slide. So these are first generation studies and first generation technology. We started at MD Anderson Cancer Center about 2007. And we started collecting really real-world data from patients that were navigated to clinical trials. And this started with a master protocol, which we call PREDICT, Profile Related Evidence Determining Individualized Cancer Therapy. And this included any type of patient, was an umbrella trial, and that included individual histologies, regardless of their genomics, and it was a basket trial, and that it included cross tumor type genomic subsets.
And this is what we began to see. Next slide, please. What we began to see with our patients, these are waterfall plots. And any patient below the horizontal line is a responder, and any patient above the horizontal line is a progression. So the response rate in these patients with end stage disease, median of about four or five prior therapies when given phase one clinical trials was about 5%.
But what we saw was, when we began to match patients, and this is first generation matching, starting in 2007. So we did not have next generation sequencing yet, but when we began to match patients based on their alterations, the response rates really increased very dramatically.
Next slide. But this was just the beginning. And we realized that things have moved really quickly in the last, approximately 13 years. And we now have a new era with tissue agnostic approvals that are driven by genomics. The first one, and this was really key. So I put up the date, was when the FDA approved pembrolizumab for solid tumors. And there were three key features of those approval that made it really transformative.
The first was that it was tissue agnostic, it was the first time, but certainly not the last time that we had a tissue agnostic approval. The second key feature was that it was approval based on the genomic marker, mismatch repair gene defect. And then the third part of this approval, which I think was even more important, was that the approval was based in part on retrospective, real-world data. And I think this is the first time that the FDA, but not the last time, began to think about looking at approvals based on real-world data.
Next slide. But as we began to do next generation sequencing, we were really excited to have this powerful, new tool. And then, very soon afterwards, we realized that there was a challenge that came with this new tool. And that challenge was what if every patient with metastatic disease is different? I live in San Diego, but I was born in Canada, and know a little bit about snowflakes. And we began to think about tumors as metastatic, the equivalent of metastatic snowflakes, every snowflake has a unique crystal structure, and every snowflake is very complicated.
Next slide. And this can be shown in this slide. We've actually done thousands of patients, but this is what fits on the slide. There is several patients with breast cancer. And you can see that patient 42, and patient 25 both have ERBB2 amplification. And so what we've normally done is given both of these patients ERBB2 targeted agents, and of course, that's the right thing to do.
But when you look at the rest of their genomic aberration portfolio, it's different. And so now the question arises, do these people, both of these patients really belong on the same regimen? Or do we need to individualize the therapy to each of these patients? And I think the answer is the latter. But this is very different than anything we've done before.
Next slide, please. So, genomics has unveiled disruptive complexity. If there's 300 drugs for oncology, there's almost 4.5 million three drug combinations. And how are we going to go do all these clinical trials, individualizing all this therapy? Next slide. And then we need to layer on this all the single nucleotide polymorphisms, and all the snips. And then of course, we've talked about genomics, but there's transcriptomics, proteomics, epigenomics, and so forth.
Next slide. So we need real-world data and machine learning to sort all of this out. Next slide, please. So at UCSD, we created a real-world data master protocol, which is really an umbrella protocol over the entire institution. And, next slide, what helps us with this is our molecular tumor board. And I want to talk a little bit about what we concentrated on. We concentrated on the pillars of precision medicine, which includes genomics and immunotherapy. And people think of them separately, but they're really a couple, and they're married to each other.
And this marriage, next slide, is exemplified by the fact that the higher the tumor mutational burden, the better the response to immune checkpoint blockade. So overall, if you have a low mutational burden, your response rate is 4%. If you have a very high mutational burden, based on our real-world data, the response rate can go up to 67%.
Next slide. So I want to give you some examples of what we're seeing and what we're doing in the clinic. And so this is one patient, this patient came to see me in January 2016, she had a hybrid neuroendocrine cervical cancer, this is a rare tumor, I run a rare tumor clinic, she came from the Middle East, she'd had prior surgery radiation therapy, and chemotherapy. Next slide. And she was very end stage with bowel obstruction and kidney obstruction.
Next slide. And this is, we didn't really have a lot of time. So what we did is we did profiling on her blood. And what we saw, the usual patient has three or four abnormalities. She had about 20 abnormalities. And we assumed that this was hyper-mutated DNA. Next slide, please. And this meant, we assumed that she might have a high tumor mutational burden, remember, this is January 2016. And we began to treat her with immunotherapy, with checkpoint blockade. And you can see her CAT scan before and after therapy. And essentially, she is now, I guess, about four and a half years now with an ongoing response.
Next slide, please. I also wanted to say that real-world data can tell us not only who we should be treating, but who we shouldn't be treating. And one of the things that we've seen is this phenomenon of hyper progressors. These are patients who have accelerated progression after immune checkpoint blockade, and this is a patient. You can see in the left that this patient's tumors were stable from March till June. And then she started on immune checkpoint blockade, and six weeks later, she just exploded.
And what we saw in this patient is that she has MDM2 amplification. When we did an analysis of our data, next slide, we found that this was very common in our patients with hyper progression. They had MDM2 amplification, and the other alteration that they had was EGFR alterations. Now, that's not to say that every patient with MDM2 or EGFR will hyper progress, or that every hyper progressor will have MDM2 or EGFR.
But my point here is that some of these patients are at risk of hyper progression, and that genomics can not only tell us who may respond to therapy, but it may tell us who may be resistant or do poorly on therapy, and who in whom we shouldn't choose these therapies. Next slide, please. So I'll just talk a little bit about our study, the I-PREDICT Study, next slide, that we published in Nature Medicine a few months ago, where we gave customized combinations to newly diagnosed patients with lethal malignancies.
And this is really an n-of-1 study. And, next slide, this is the outcomes on this study. So what you can see here is with these n-of-1 combinations, we found that patients that were highly matched to therapy had a response rate of 45%. Patients with low matches had a response rate of about 16%. And progression free survival and overall survival was also increased, with overall survival being about 60% at two years in patients with high match, and about 20% in patients with low match. And this tells us that is not just about, yes, patients are matched or no, they're not matched to therapy, but it's the degree of matching that is important.
Next slide, please. So this is an example of a patient. This patient had a carcinoma of unknown primary, had a KRAS mutation and then ARID1A mutation, KRAS, we targeted with the MEK inhibitor trametinib, and ARID1A, there's literature suggesting that it can be targeted by the PARP inhibitor, this patient has had a beautiful response as you can see in this slide. And therapy is ongoing with very little in the way of side effects that 17 plus months.
Next slide, please. I also wanted to mention some new trial types such as the master observational trial that we've been involved with. And this, we recently published in Cell. This is structuring real-world data within the concepts of a clinical trial. Next slide. So thank you for listening, I want to give special thanks to our precision medicine team, and especially to our patients because they do all the work. And thank you again for inviting me.
Michael Taylor: So really appreciate the opportunity to spend a few minutes to share a bit about how we're using the Flatiron/FMI clinico-genomic database at Roche/Genentech to support research and development.
Initially, I'll share some thoughts on some high level opportunities, and follow up with a quick overview of one project that we did using the clinico-genomic database at Roche. And then finally close with a bit of a view toward the future. Ultimately, I see many opportunities to leverage the clinico-genomic database to support research and development. As we progress further along the journey to achieve personalized health care for patients.
Well, it might be seen basic, there are a number of incredibly impactful opportunities to use innovative data sources, such as the clinico-genomic database, to address fundamental questions on the natural history of molecularly defined populations to inform the research and development of new medicines. It was really encouraging to see in the poll results at the beginning of this presentation that so many of you are thinking along these lines.
Often, these are relatively basic, you can go back to the prior slide, basic epidemiology questions, such as the prevalence of a mutation within a given cancer type, or across all cancer types. Well, historically, we might have gone to the literature when looking for this data. Often, when we're talking about a new novel biomarker, be it a RET, or NTRK or an ALK, this data doesn't exist in the literature.
The unique value of the clinico-genomic database in our work with it has been that it really allows us to look at the prevalence of these biomarkers that weren't routinely tested before their inclusion on next generation sequencing panels. Data on the prevalence of these mutations granted, we need to think about the biases of who gets tested with an NGS panel, and that needs to play into our interpretation of the results. But ultimately, these results and understanding this prevalence can really be quite informative, informing our research and development plans for a new molecule.
Likewise, as a result of the linking of the molecular data from Foundation Medicine, with the clinical and the outcomes data from Flatiron, we can get deeper insights into that prognostic significance of a novel biomarker, which is a common question that we often come across from stakeholders, be they researchers, from an R&D perspective within our companies, regulators, payers or healthcare providers.
Continuing on this pursuit of a deeper understanding of a potential target population defined by a molecular alteration, we can analyze the clinico-genomic database to get insights into the outcomes that patients have, and stratify these outcomes based on a number of different factors. Things such as tumor type, the prior treatments a patient has received, patient characteristics, etc, all of which is really informative when we're working to create, and then to refine the clinical development plan for a molecule and a new indication, particularly a new indication that we don't know quite a bit about.
Another opportunity we are starting to explore is the area of genomic patterns of resistance. And it was interesting to hear some of this in the prior presentation. And just to see the way that clinicians are using this information, it's really encouraging and exciting for patients. So as patients are treated with new targeted medicines, it's likely that unfortunately, many of these patients, especially if they have metastatic disease, are going to develop resistance mechanisms.
So the ability to look for potential mechanisms of resistance by analyzing sequencing results following the progression of a given medicine could provide new insights, right, for new targets, for future R&D. So as the use of sequencing and hopefully serial sequencing increases in the future, I see a number of opportunities where data sources such as the clinico-genomic database could really inform our research and development efforts.
Beyond gaining insights into the natural history of disease for molecular defined populations, I see great opportunity to use sources of real-world data that combine genomic, clinical, as well as outcomes data in addressing regulatory post marketing commitments, as well as data gaps for payers. In particular, this would involve increasing the size of the evidence base both from a safety perspective, as well as an evidence perspective, for medicines that get approved in some type of accelerated fashion, be it for a formal accelerated approval by the FDA or a full approval, we often are going to continue to need data on these populations, particularly given that they are small and often get approved on small populations to start with.
Lastly, what particularly excites me here is the opportunity to use a clinico-genomic database such as the Flatiron/Foundation Medicine database, to look at drug development from an indication agnostic manner, right. And so that was highlighted in the prior presentation as well, in terms of really the groundbreaking situation with pembro in the MSI high population in an agnostic way.
So we can really understand what are these significant unmet medical needs in these populations, when you combine both a rare biomarker or perhaps a common biomarker with an uncommon cancer or common cancer, right, but especially in those places, where it is the combination of both the uncommon cancer and the uncommon biomarker, that data is incredibly hard to find. And those populations are incredibly hard to understand and characterize to be able to design a drug development program for.
The clinico-genomic databases really exciting in this arena. So you're already seeing it in places such as the MSI-high, the ALK, the NTRK, the RET positive populations, regardless of tumor type. And I predict that we're going to see many, many more of those. Real-world data provides an opportunity here, not only to understand that natural history of these diseases, but actually if a molecule gets approved for one of these biomarker selected populations, and perhaps a relatively common cancer type, such as lung cancer, then there's also this opportunity that it could be used by physicians and their practice of medicine and other tumor types. And we can learn from this data that ends up in a clinico-genomic database to see how these patients do, and to design studies that actually use this data as well to gain a regulatory approval perhaps in the future.
If we go to the next slide, one of the key tactics that we're using really, to enable our scientists across the company to learn from the clinico-genomic database is through the use of dashboards such as the one you see on this slide. By enabling scientists who aren't programmers to gain high level insights, we're able to increase the overall use of the data. Within these dashboards, we also go and provide contact details for our data scientists so that if the clinical scientist or basic science researcher would like to do a deeper dive into the data beyond that, that's possible in the dashboard, they know who to contact to do a more rigorous retrospective analysis to gain insights.
Moving on, I'd like to speak briefly about one specific use case that we have, you can go to the next slide. So a recent evidence gap that our research and development teams have had was in the area of working on a PI3K inhibitor in developing that. And a key question they had is, well, what do patients look like? What are the patient characteristics, the outcomes, according to the number of PIK3CA mutations.
So this work was recently completed and presented by MK Downer and colleagues at AACR. And on the right, you'll see the Kaplan-Meier curves and the hazard ratios for overall survival, based on whether a patient has wild type or has a single or multiple PIK3CA mutations. As you can see, there's no difference in survival. And learning this was really important for our team as they work to design our phase III trial.
What was also interesting in this study was that we found that the majority of patients receive tumor genomic profiling results during or after first line treatment, suggesting that the genomic results may be used to guide clinical decision making after standard therapies have failed.
We'll go to the next slide. I'd like to finish by taking a step back and looking a little bit towards the future in ways that we could use clinico-genomic data to further enable research. So historically, our focus has been on the retrospective data, right? And that's what the Foundation Medicine/Flatiron database currently is a retrospective data source, and we get tremendous value from that.
There's also an opportunity to use it and perhaps do prospective research. So as such, we set out to do a low interventional prospective clinico-genomic study. An effort that was shared publicly earlier this year at ASCO as a trial in progress. The primary objective of this study is really to evaluate the feasibility of a scalable prospective research program, and to collect real world clinical and serial monitoring data.
A secondary objective is to explore the CTDNA levels, and how they might be associated or be a predictor of response to therapy. So in my last slide, I've included a schema of the study, in which you can see that a key element here involves the collection of multimodal data, which we think is really, really important. So having the clinical data, but also the serial genomic data, the digital path, the clinical imaging.
And critically, what I think is going to make this study successful, and perhaps one that we can repeat in the future, is that while it is prospective in nature, the vast majority of the data is actually collected through the same processes. Is for extracting the clinical and the treatment outcomes data that are used in the clinico-genomic database, rather than the more traditional ECRF approach. So stay tuned for findings from this study. I'm particularly excited to see what will come in the months and years to come. And with that, I think I'm going to pass it back to Simon.
Emily Castellanos: Okay, thanks so much, Dr. Taylor. As you all can see, I'm not Simon, but just wanted to let you know we are going to be heading back to Dr. Papillon's presentation. Looks like we got it. Perfect. All right, we're going to restart. Dr. Papillon, take it away.
Simon Papillon-Cavanagh: Right. So the challenge with assessing the prognostic versus predictive nature of STK or KEAP1 mutations and non-small cell lung cancer is twofold, which is a fairly low mutation prevalence and a modest effect size. So you really need a large cohort to really test the specific interaction terms between the mutation status and outcomes across different treatment paradigms.
So the key questions we try to address in this presentation. The first one, the major one is, are those mutations associated with poor outcomes across all first line treatment paradigms? And the second question is, is there a synergistic effect among those two mutations on outcomes?
Next slide, please. So as I was mentioning before, there are really a plethora of papers or publications that have detailed the negative effect of STK11 mutations on ICB treatment outcomes. But really, the big caveat here is that there are very few if any, in this case is none, control arms that are non ICB treated, so we can't really assess this, there is a prognostic biomarker, or does that have treatment specific effects on outcomes.
Next slide, please. And the debate rages on, this is just a snapshot of publications at ASCO 2020. So just back this summer. So it's a hot topic of conversation. So we hope that our study here leveraging the CGDB does not add to the noise. But really, given the depth of the cohorts and really the size of the cohort, we can help answer some of those questions.
Next slide, please. So cohorts in approach, basically, this kind of the funnel that is quite common, and how we filter out patients to make sure we have the highest quality data, again, leveraging the large cohort. So we filter the non-small cell lung cancer patients that have a non-squamous histology and first line treatments across those kind of five buckets, if you will, to have a total of just above 2000 patients. And we'll do our comparison across those five treatment paradigms against this first line.
But really, the kind of the killer comparison we want to look at is the Anti-PD-1 vs Platinum chemo. So this is what the cohort looks like. We separated here mutant wild-type. So mutant means if a patient has a mutation in either STK11 or KEAP1, just note that those mutations are frequently co-occurring. So if you have one, you very likely have the other. So enrichen male patients, they are younger, more likely to be smokers and high TMB.
And just one note on the, there is a statistical difference in the treatment paradigm, that is because EGFR mutations and STK11 and KEAP1 mutation are mutually exclusive. So if we repeat the same comparison, but we discard or exclude EGFR mutated patients, there's no difference in mutant versus wild-type first line treatment paradigms. And as previously reported, there is a lower PDL1 staining in the STK11 mutants sub cohort here.
Next slide, please. So the kind of workflow that we use is very simple here on patients that have complete data. The funnel I described before, we selected the CGDB data on two genes. So the mutation date on two genes, STK11 and KEAP1, and we tested for OS and real world PFS. And we tested the specific interaction term. I'll show you the model in a few slides to really says the IO specific or immune checkpoint specific effect of those mutations.
Next slide, please. So let's go right into the results. The effect of the two mutations, real world PFS [inaudible] except EGFR as for the reasons I just described earlier, basically those patients don't get EGFR treatment. Which kind of gets to our points that the negative outcomes associated with those mutations are [inaudible] to actually test statistically, on the B panel here, we're looking at the interaction term of real world PFS of either mutation and treatments. They're really comparing Anti-PD-1 versus Platinum based chemotherapy here.
And we see that there is no treatment specific effects. So the hazard ratio is not different for either mutations. So this suggests that the impact of those mutations is really prognostic and not predictive, or not specific for IO therapy. Next slide, please. And here, we looked at an either Anti-PD-1 on the left hand side or platinum based chemotherapy on the right hand side, just stratify by the different mutational status, just to test if there was a synergistic effect of either mutation, then you can clearly see that the double mutants does worse than KEAP1 or STK11 alone, which all of them do worse than the double wild-type. So wild-type for all those issues. That's consistent with the treatment paradigm.
So suggesting that those effects are additive, and that there's not a synergy. We also tested that mathematically. And the hazard ratio of the double mutation was not the sum of [inaudible], basically. Next slide, please. We profile the same analysis really [inaudible]. Thus, again, supporting our results that those biomarkers are prognostic and not specific for IO therapy.
Next slide please. Again, same thing, we tested the additivity of those mutations, it was also clear. No synergy of STK11 or KEAP1 mutations as per the overall survival outcome. And I think that's the end.