Olivier Humblet: So our agenda for today, we're going to focus on the role of real-world evidence in tumor-agnostic drug development, starting first with the genomics perspective from Jeff Venstrom. Next, Chelsea Shao will share a pre-approval use case --- a study on the prevalence of high tumor mutational burden and its association with survival in patients with less common solid tumors. This will be followed by a use case in the post-approval setting presented by Tamara Snow which assessed characteristics and outcomes of real-world patients with MSI-H solid tumors treated with pembrolizumab. Lastly, we'll hear from Mike Spencer, who will present the trends with tumor-agnostic approvals, but also the challenges that still lie ahead. And lastly, before we jump in, please excuse any interruptions from our pets or loved ones. As, like most of you, we are all still working from home.
One last thing before we get started, we'd love to learn a little bit about our audience, so you'll see a poll pop up on your screen momentarily. And if you have two screens, please check both, it could pop up on either. And other attendees will not be able to see what option you select. So the question is, in which of these areas are you most interested to learn about uses for tumor-agnostic, clinico-genomic real-world data? Discovery and translational research, clinical development, supportive regulatory evidence, market access, or medical/commercial use of therapeutics. We'll give it about 10, 15 seconds, and then we'll share the results. All right, let's end the poll and share the results. All right, let's see, highest is supported regulatory evidence, 34%. Followed by clinical development, 27% and then the rest 12 to 14%. Very interesting. Thank you. So with that, I'll pass it over to Dr. Jeff Venstrom to kick things off.
Jeff Venstrom: Yeah, thank you, Olivier. And thanks to the Flatiron organizers for the opportunity to speak on this really important topic. Next slide. In the brief 10 minutes that I have for my component, I'll just give a high level overview of a clinico-genomic perspective on tumor site agnostic drug development, a topic that I'm particularly passionate about. Next slide.
The obligatory disclosure information. I am an employee of Foundation Medicine. And I will not be addressing or be able to answer any questions related to any pharmaceutical products. The major reason why I'm so excited about talking about this particular topic around tumor-agnostic drug development is because I really think it's a major breakthrough in the 21st century for really making biology actionable.
The human genome project and our ability to sequence all of our genomes, and particularly the genomes of cancer patients came with a lot of promise and a lot of hope. And I think to some extent, there's been a disconnect between the promise of mapping the genome and the specific actionability of genetics in the clinic, and in the clinic particularly for cancer patients. By and large, we still march forward treating patients in a tumor site-informed paradigm where a patient has lung cancer, breast cancer, non-small cell lung cancer or any other subtypes. And we decide on therapies based upon the histology.
What the recent tumor-agnostic approvals have really done is added a fundamental and foundational new paradigm to oncology decision-making, that really brings in the biomarker, that brings into genomics, that complements the established and the continued role of tumor site-informed treatment decisions with biology. And with really important mechanistic biology driven really by insights around how recent drugs work, and how the biology is driven by different drivers. Focusing more on this topic for drug development, I've aggregated four key points or perspectives from a clinico-genomic perspective for this audience to consider as we're thinking about doing more of this really challenging drug development pathway for tumor-agnostic approvals.
Next slide. The first core points that I wanted to bring to this audience and have a conversation around is the importance of knowing your biomarker. These are three key considerations as we're thinking about doing more tumor-agnostic drug development is one: understanding the biologic hypothesis. I think tumor mutation burden is a great example of understanding the role and the ability in a probabilistic manner to predict immunogenicity of a specific tumor. And to understand that connection between this biology and the mechanism of checkpoint blockade, for instance, was really foundational for driving forward the success ultimately of TMB as a biomarker and as tumor-agnostic approval.
TMB of course is an imperfect biomarker. It doesn't capture all components of a tumor's immunogenicity, but it's fit for purpose. It was ultimately successful for getting the approval and I think can actually help patients. There's still a lot more that we need to understand around the immunogenicity of tumors, and the ability to appropriately select patients for checkpoint blockade. But I still feel like the foundational understanding of a biologic hypothesis for TMB and tumor approvals is really fundamental to success.
Once you have that mechanistic hypothesis, making sure that you have an accurate definition of the biomarker if it's a genetic biomarker of clarifying how you're defining a specific alterations as actionable, and thinking about the potential need to update that definition of actionability, based on things like updates of known and unknown variants within the genome. There's a fundamental role for big data and real-world data to understand both the prevalence of the biomarker that you're chasing for these tumor-agnostic drug development opportunities, in addition to the potential prognostic effects.
Chelsea will actually get into a very specific and fascinating study, looking at the prognostic role, specifically for TMB. The real-world prevalence for specific biomarkers will help you design your statistical analysis plan, anticipate challenges for accrual, understand putative mechanisms of resistance, and really help inform the design and execution of your tumor-agnostic drug development program. And finally, trying to get a sense of any type of disease specificity or where potentially the biologic hypothesis might break down and specific context is also important, as you're thinking about embarking on this adventure for tumor-agnostic drug development.
Next slide. A second key consideration as you're planning these drug development pathways is the platform or platforms, plural. And the importance of trusting that platform, the importance of the performance and the reliability of a platform that you're choosing for this drug development opportunity. Making key decisions around for instance, is there an opportunity to use a liquid genomics platform, perhaps in addition to a tissue genomics platform? Or what are the trade offs, the pros and cons of using one versus the other? What type of analytical validation does the platform that you're implementing into your tumor-agnostic drug development trials... What degree of analytical validation do we have specifically in specific diseases and specifically for different genetic alterations?
Is there transparency in your platform? Is there ability to understand when the calls are erroneous? When the calls are complicated, and need further follow up? How much clinical utility has been established for the platform that you choose for your tumor-agnostic drug development platform? Are there opportunities to leverage the clinical utility that's already established, perhaps in other tumors or for other drugs to complement the success of the tumor-agnostic drug development trial that you're planning?
Is there external validation? Does the platform work in multiple different hands? Is it published by multiple different groups? And is your platform or the partner that you're choosing for this drug development, adventure committed to continuous improvement, committed to updating the definition when necessary, and as our understanding of the target evolves, and in the definition process for your biomarker? Is it adaptable, and do you have reproducible algorithms for instance, if you are reliant upon a computational biomarker? Next slide.
Third is to explore nuance. This is more of a nice to have. I feel like the first two features for your drug development program are critical success factors. Exploring nuance might be something that we plan as part of a post-marketing study, for instance. It might actually be fundamental to the success of the drug, particularly if there's significant disease nuance that you discover through assessing prevalence and prognostic data in the real-world setting as you're planning your specific trial.
Is there a need or an opportunity to think about disease-specific combinations, again, that you're learning because of these studies that you're doing with big data, with prevalence information for understanding the biomarker and understanding the co-mutations that come along with the biomarker that you're targeting and the need or opportunity to combine therapies?
Are you assessing performance and reproducibility throughout the drug development pathway? In phase one and phase two, are there opportunities to reassess the performance and reproducibility of your biomarker definition, of your cut off, of the assay performance? And are there opportunities through the drug development lifecycle, to assess interim results, and to think about the opportunity to update the definition or pivot? Next slide.
And then fundamentally, and finally, the importance of collaboration is so critical. This is a really challenging drug development paradigm. It's really important for patients, particularly patients that have rare disease. And getting this right and doing this carefully, collaborating both on the design and execution. From a statistical standpoint, regulatory standpoint, obviously, the genetics and clinical standpoint, bringing in your genomicist early to really help with that biomarker definition and knowing your biomarker, leveraging real-world data like the CGDB, the clinico-genomic database that fuses Foundation Medicine information with Flatiron, real-world outcome data, and granular clinical characteristics.
Again, to understand the prognostic and the prevalence information is really important for planning for success. And then thinking creatively and globally about the execution and often the challenge of targeting these rare populations and these rare patients for accruing into your clinical trial. Next slide.
And with that, I'll end and just remind the audience of the summary of what I talked about really briefly, again, from a high level around a clinic-genomics perspective. Fundamentally, I think, number one, know and trust your biomarker. Trust the platform that you're using for assessing your biomarker. Explore nuance, and then collaborate broadly. Thank you, Olivier, again, for the opportunity to speak.
Olivier Humblet: Thank you so much, Jeff. It's really great to get the genomic perspective. With that, I'll now hand it over to Chelsea Shao who will share her research, a study which utilized tumor-agnostic real-world data in the pre-approval setting.
Chelsea Shao: Thank you, Olivier, for your introduction, and much appreciative of Flatiron organizers for their invitation. It's my great pleasure to present our tumor mutation burden, prevalence and prognostic study. This is truly a collaboration through Flatiron, Foundation Medicine and Merck. So thank you to all the colleagues who contributed to this great study. Next slide please.
This is my conflict of interest. I'm working in Merck. Next please. We all know that tumor mutation burden is an emerging clinical biomarker for response of immune checkpoint inhibitors. As you can see, from this bar graph, different tumor types, they have a different prevalence of a tumor mutation burden. And also you can see from the right hand at the plot, the higher tumor mutation burden associated with the higher overall response rates for immune checkpoint inhibitors in multiple variety clinical trials. Next please.
So the recent KEYNOTE-158 trial evaluated the TMB and outcomes after patients are treated with pembro monotherapy among these 10 tumor types as listed in this bar graph. You can see that for total populations, there are about 13% of patients that has a TMB-H across these 10 uncommon tumors and the objective response rate was 29% in the TMB-H group. However, only 6% in the non TMB-H group. However, there are limited real-world data on the prevalence of TMB across these uncommon 10 tumor types. Also very little data about the prognostic effects of the TMB across these tumor types, when the patients do not get treated with immunotherapy
Next slide please. So, for our studies, we have two main objectives. One is to evaluate the prevalence of TMB across these 10 tumor types from this real-world dataset. And another main objective is to look at the association between TMB and the real-world overall survival across these 10 tumor types who do not receive IO treatment. So for this study, we are using the Flatiron and the Foundation Medicine clinico-genomic dataset.
As you can see from the figure, the clinical data is coming from Flatiron data and linked with the genomic data, including the TMB measurements from Foundation Medicine. By July 2018, there are about 28,000 patients identified from this CGDB dataset. To be eligible for our study, we require the patients to be adult at the time of FMI report date and they had to have a valid TMB measurement. Also, they need to have one of the solid tumors from these 10 tumor types as listed below.
Next slide please. So for the TMB-H, we predefined that at least 10 mutation per megabase assessed by FMI. For all this analysis, we pre-specify them and for the primary populations we use the pan-tumor population. The patient who has confirmed the MSI-H cancers will [be] excluded from the primary analysis. However, we have the analysis in the sensitivity analysis including this MSI-H population. OS was analyzed by Kaplan-Meier method and also Cox Model was implicated. One thing to call out for this study, because we try to evaluate the TMB prognostic effects to eliminate the impact of IO treatments on the clinical outcomes, the patients with IO treatments were excluded if the start of the IO treatment is early or equal to the index date, and the patients who were censored if the start of IO day is after the index date.
Next slide please. So here is our high level results. Among these pan-tumor CGDB datasets, we identified about 2,600 patients who had the TMB measurements, and also had one of the 10 tumor types. These patients were included in the study. As you can see from the left hand table, this is the patient characteristic. You can see about 65% of patients were female. On average, the patients aged 64 years, and majority of today's patients come in from the community setting. And on average, there are about eight months follow up from the Foundation Medicine report date. If you see it from the pie chart, this is the distribution of the tumor types across this cohort. The top three tumor types are biliary, endometrial and also small cell lung cancer.
Next slide, please. So here is one of our main study results. When we use the 10 mutation per megabase as the TMB-H cut points, there are about 13% of patients who were TMB-H. As you can see from this bar graph, different tumor types, they have different TMB-H prevalence. Small cell lung cancer has a high TMB-H prevalence. And also mesothelioma has the lowest TMB prevalence. These data are very consistent with the KEYNOTE-158 findings and also consistent with the limited reports from the published papers. Next slide please.
So another important finding from this study is from the OS analysis. As you can see from the Kaplan–Meier curve showing at the right hand, those orange color represents the patients with non-TMB-H and the blue-ish color represents the patients with TMB-H group. So, from this analysis, we can see that patients with TMB-H have similar real-world overall survival compared to those patients with non-TMB-H tumors when they are not receiving IO treatment. The adjusted hazard ratio was 0.94. We adjusted for age, gender, cancer types, etc. Next please.
We also did a variety of different sensitivity analysis by looking at a different study population and to including or excluding MSI-H tumor patients and also using different index dates. So all of these sensitivity analysis is showing very consistent results. These are our primary analyses, as you can see from the forest plot, all of the 95 confidence interval cross one. Next please.
For this study, we conclude that the prevalence of TMB-H in this real-world dataset was about 13%. We used 10 mutation per megabase across the 10 uncommon tumors. And this prevalence varied across the tumor types. There was no association between TMB-H status and the real-world overall survival across these tumor types. These findings suggest that TMB-H do not have prognostic associations among these tumor types when they're not receiving IO treatment.
The studies suggest that the difference observed from the temporal trial across these 10 tumor types is likely coming from treatments and not necessarily coming from the TMB prognostic value. Next slide, please. So for this study, I like especially to thank our colleagues from different companies for their great contributions. This is an excellent collaboration. Thank you for your attention and the interest. Thanks.
Olivier Humblet: Thank you so much, Chelsea, those are really amazing results. So with that, let's switch gears a bit and explore an example of how tumor agnostic clinico-genomic real-world data can be useful in the post-approval setting. So I'll hand it over to Flatiron's Tamara Snow.
Tamara Snow: Great, thanks so much, Olivier. Hi, all, my name is Tamara Snow. And I manage the clinico-genomic data product for the Flatiron team. So today, I'm excited to talk through one of our earlier explorations into tumor-agnostic analyses using our clinico-genomic, pan-tumor dataset, and how this research not only demonstrated how a novel treatment approval is being used in the real-world setting, but also help guide how we're hoping to continue to improve our datasets, and analytical guidance to more easily unlock these tumor-agnostic analyses in the future.
So to start, I'll give some real context on the rationale for this study. So in May 2017, pembrolizumab monotherapy received the first tumor-agnostic, biomarker-based FDA approval in oncology for patients with mismatch repair deficient or microsatellite instability high or MSI-H tumors. This really was a groundbreaking approval based on a meta analysis of five single arm trials, which include approximately 150 patients diagnosed with 15 different cancer types.
Given this new tumor-agnostic approach for therapy approvals, we thought it would be important to evaluate how this is translating into the real-world setting, specifically the use and effectiveness of this novel treatment paradigm. So for the study, we sought out to examine the characteristics and outcomes of real-world patients with solid tumors, identified as MSI-H via a FMI test, who had received pembrolizumab and monotherapy, after the FDA approval in May 2017. And to do so we leveraged our pan-tumor clinico-genomic data set to observe pembro use across the various histologies.
For this analysis, we used our Q1 2020 pan-tumor clinico-genomic dataset, which included 54,000 patients. To align with the approval, we limited it to patients that had an MSI-H solid tumor, and had started their first pembrolizumab treatment after their FMI test. And then within this cohort, we identified 148 patients with 36 unique FMI tumor types that received their first pembrolizumab monotherapy treatment after May 2017.
So one thing to note is that typically Flatiron derives line of therapies on a tumor specific basis. However, given the pan-tumor nature of this analysis, we had to derive a treatment block heuristic using our structured treatment data across tumor types. And this logic was really essential for us to identify patients with the pembrolizumab monotherapy line, as well as to describe the treatment sequencing in this pan-tumor cohort. And then in terms of the outcomes we evaluated were time to treatment discontinuation and overall survival.
Time to treatment discontinuation was defined as the time from first pembrolizumab administration or non-canceled order to last pembrolizumab treatment. And then overall survival was assessed from first time pembrolizumab administration or non-canceled order to the date of death or last activity date. Both of these outcomes were evaluated with Kaplan-Meier analyses across all patients in the cohort, as well as the largest tumor groups, which were colorectal cancer, endometrial cancer, and gastric or gastroesophageal junction cancer.
So moving on to the results, we looked at the baseline characteristics across the full 148 patient cohort, as well as by the largest tumor groups. So the overall cohort had a median age of 69 years, 65% were female and 78% had an ECOG performance status of zero to one around the time of pembrolizumab monotherapy start. This was slightly different from what we saw in the trial meta analysis, which had a median age of 55 years. 46% were female, and all patients had an ECOG of zero to one.
And then with regards to the genomic characteristics, 42% of patients had concurrent MMR alteration. And then the median tumor mutational burden was 32.2 mutations per megabase across all patients. Having this higher median TMB in this cohort was definitely expected, as prior research has shown a high overlap in high TMB and MSI-H tumors. But it was really interesting to see how it did vary a bit across the different tumor groups.
And then last thing to mention is that the median follow up time from pembrolizumab monotherapy start was about eight months. And not highlighted on here but just want to note that the median number of therapies prior to pembrolizumab monotherapy, was one across the cohort with approximately 60% receiving that pembrolizumab monotherapy line, in either the second or third line setting.
Great. So then in terms of the outcomes, the 12-month overall survival in the clinico-genomic cohort was 62% across all cancer types. And then when looking at the larger tumor groups, it ranged from 47 to 70%. And this is actually similar to the 12-month overall survival data that was subsequently published for the clinical trial cohorts, which show that for two of the five trials, patients demonstrated a one year overall survival greater than 70% for those with CRC and then greater than 60% for those with a non-CRC tumor type.
So what this really shows is that while the real-world cohort had slightly different baseline characteristics, such as being older or coming from a wider variety of tumor types than the pembrolizumab monotherapy trial cohort. It was really interesting to see that the one year overall survival rate across all patients in the largest tumor groups in the real-world cohort was consistent with the trial outcomes.
So in terms of the impact, from the clinical perspective, we really hope this study helps to inform how this first tumor-agnostic biomarker driven FDA approval in oncology is being translated into routine clinical practice and really helps validate the adoption of more precision medicine in cancer care. It was also a really great opportunity for us to provide supportive real-world evidence of the clinical trial data in a rare cohort where additional trials would be challenging to conduct.
And then from a real-world data perspective, this study really demonstrates the ability of the clinico-genomic database to evaluate these complex genomic biomarkers in a pan-tumor population, and also helps advance our understanding of how to design and execute these real-world pan-tumor analyses. So lastly, I just want to note some limitations about the study, and really our plans to address it through future work.
So first, unfortunately, at the time we ran this study, we didn't have our tumor-agnostic abstracted data model at scale, so we relied purely on structured EHR clinical data. And as such, some unstructured data elements like confirmation of date of diagnosis and stage were not available. In addition, our structured treatment block heuristic was definitely a minimalist approach to try to derive lines of therapy across this heterogeneous cohort, making it likely less robust than our tumor-specific line of therapy rules.
So in the future, what we'd love to do is rerun this analysis using our tumor-agnostic, enhanced or abstracted data model, which is now fully scaled as of last year. So we can leverage those additional clinical data elements to learn more about this cohort. And then another big goal for us in 2021, is to really advance our thinking on how to conduct these pan-tumor analyses and develop tumor-agnostic elements as needed to support them.
For example, further refining and validating our approach to pan-tumor lines of therapy. Great. Thank you so much for listening. I really appreciate your taking the time to listen to this study, and just noting that this was presented at ASCO 2020. So if you'd like a copy of the work, we're happy to share. Thanks.
Olivier Humblet: Fantastic. Thank you so much, Tamara. So now for our final perspective, let's hear from Mike Spencer, who will share some of the challenges as well as the opportunities associated with using real-world data for tumor-agnostic research, particularly from the health authority and payer decision-making perspective. Over to you, Mike.
Mike Spencer: Thanks, Olivier, and thanks for the opportunity to speak today. So despite my accent, I am based in the US but I'm going to be taking a little bit of an ex-US view and there's been lots of really exciting signs and opportunities presented. But I'm going to be focusing a little bit more on the challenges, I guess, than some of the previous speakers. So just want to go to the next slide. Thanks, Olivier. So this is some data from EFPIA, the European Federation of Pharmaceutical Industry Associations, where IQVIA collects for them the number of days between regulatory approval, and access to medicines in Europe.
This is specifically the oncology data and what you can see is a bit of a depressing picture here of the gap between regulators making a call about the safety and efficacy and the HTA bodies, and payers making the product available to patients in the countries. I think what we're seeing is there's increasing... From the regulators, increasing flexibility, forward looking approach, and taking into account real-world data, and novel approaches, like tumor-agnostic approvals, but still having the HTA bodies taking a much more conservative view.
I think the tumor-agnostic approach is really an extreme example of that, where this is a really very novel approach to drug development, and regulatory approval and the HTA bodies and payers are somewhat behind. On the next slide please. There's three areas that I wanted to pick up on today. The first was in terms of the testing challenges, both from a practical point of view, but also from the way our HTA bodies, particularly look at that.
The second one was in terms of the health technology assessment methods and the principles by which they work and how that impacts tumor-agnostic development. The final one was around evidence generation and acceptance to support demonstrating the value of these tumor-agnostic indications. So I just want to go on to the next slide please.
So in terms of testing, so the first thing really is and I'm sure there are people on the call who know more about this than myself, but just in terms of the diagnostic pathways, outside the US, having those pathways which ensure that the testing data is in the hands of the physician at a timely manner to make those treatment decisions is a challenge in itself. When it comes to how the health technology assessment agencies and the payers are thinking about this, then the question of cost per identified patient becomes very relevant. So if you have a rare marker that you're looking for, and a cost associated with that, obviously, to find one patient can be pretty expensive if you're having to test 100 patients to identify one with the relevant marker.
And if you look over on the right hand side of this slide, this is from the entrectnib HTA assessment through the UK's NICE. And you can see there just the wording on the bottom that says, "In the base case scenario, 100% of the incremental screening costs are applied to entrectnib." What you have here really is a catch 22 --- in the less treatments that are available that can leverage the testing, the more those costs get piled on to the treatments that are coming through. And that then becomes a block to those treatments being approved and the uptake of those treatments so you're caught in this cycle.
And then, what that leads to is this issue with logistics and infrastructure, where there's really a tragedy of the commons, where if there's not the treatments being approved, that need the testing, then no one's investing in it. And it's all the cost of input on to the companies, or being at least wrapped into those HTA assessments and cost effectiveness assessments. So you go on to the next slide.
So in terms of the methods and principles, health technology assessment agencies really have grown from the evidence-based medicine paradigm. And so, randomized clinical trials really take primacy there. We have some specific challenges, I think, here with tumor-agnostic approach, both in terms of equipoise. So, do we have acceptance from the health technology assessment agencies that even though these patients by definition, normally, from an indication point of view, have no other options, that an option that is, maybe exquisitely targeted to a driving mutation in their tumor means that there's no equipoise to run a randomized trial?
That's something I think that seems maybe obvious from a development point of view. But it's been severely challenged, when we go to the health technology assessment agencies, and often is seen as, well, that's the company's problem to try and work around that, rather than that being a fundamental part of this kind of drug development. Rarity of course, becomes another challenge there, where, again, the agencies are not going to relax their standards on the basis that it's difficult to run a study, when there's this level of rarity.
Then one thing we have to remember is that the incentives for these kind of agencies are maybe different too from the regulators. They are looking to find gaps in the evidence based on the company, because whilst there really should be a separation, good HTA principles would say there's a separation between the scientific assessment and the pricing negotiations, that's often not the case in reality.
The other point, in a similar vein is around marginal decision-making. From a payer HTA body point of view, it's not just saying, well, this treatment is effective, or this is something that's worth doing it and in which patients do I get the best bang for my buck. And so, that really challenges the idea of a tumor-agnostic approval where you're looking across multiple tumor types.
Now, of course, within the standard paradigm, you're always going to have heterogeneity within studies. In this case, you maybe have heterogeneity of tumor types for homogeneity around the driver. In normal drug development, you may have multiple biomarkers, but within a single tumor, but this is a novel approach. And you can see on the right here again from entrectnib but this time looking at the German HTA agency. If you can read German you can see the verbatim but on the right, in English, no tumor individual assessment and that's relevant and required. No relevant comparative data because there was no randomized trial.
There was some real-world data brought into this process with the HTA body, but they didn't consider that the effect size was big enough to overcome their concerns about confounding and bias, and no data on pediatric indications. And so, they basically assess that with no relevant data and so forth they couldn't assign any benefit to treatment. Move to the next slide. Thanks. And then evidence generation and acceptance. I have referred to this already.
One of the key principles for health technology assessment agencies, is the idea of what's the appropriate comparator? What's the cost of that? What are the benefits of that comparator and how do you compare? Well, again, if you're looking at an indication statement that says, who have no satisfactory treatment options, well, what is that appropriate comparator and can you even run a clinical study?
When it comes to real-world evidence availability, where we just heard the work that the Flatiron [team] are doing on this, but it's difficult to have access to biomarker data for this specific market, and may be relevant to your approval in the real-world to do these indirect comparisons, external controls, if no treatment exists, because why are people going to be testing for something if there's no treatment? The thing that I think that Flatiron are trying to address here is that most data sources that we have access to are organized by tumor, so vertically from a tumor point of view rather than transversally from the biomarker genomic point of view.
And then, finally, just going back to the previous point I had was, how acceptable is RWE as a decision-making evidence base? So some HTA bodies simply won't accept RWE, and some set the bar impossibly high, both in terms of their expectations of how data is collected, but also the effect size they're wanting to see. Next slide, thanks. So what do we need? Well, investment in testing, I think we have to somehow break this tragedy of the commons outside the US where the incentives for testing are low. And therefore, it's difficult then to have an interesting paradigm built that can support it.
A pragmatism with regards to evidence, but also, I think, a need for us as industry to show the validity of these kind of external controls, particularly when we are looking at these rare tumor types at the very end of the treatment line, so that maybe we can accept a lower bar when it comes to the effect size we need to see. So building the validity around that, but also pragmatism from health energy investment agencies and payers.
And then flexible access models. We're often coming to market with relatively limited amounts of data. A lot of the data that we have on the treatment is going to be in the real-world post approval. So how can we build access models that leverage that data, the increasing certainty we have over time to make sure that patients can get access as quickly as possible, but then the data can support the continued approval and the pricing? So I will stop there, Olivier. And hopefully, we can help with the questions directed to me.
Olivier Humblet: Sounds great. Thank you so much, Mike. So before we move on to the Q&A, we have one final poll for the audience. So the question is, how would you describe your organization's interest level in utilizing real-world data for tumor-agnostic drug development? First, we are actively pursuing it, we're extremely interested, we're moderately interested, we have low interest, it's not strategically aligned at this time. Again, let's give it 10, 15 seconds and then share the results.
All right, let's take a look. So let's see, 44% actively pursuing, 34% extremely interested, 19% moderately interested. Well, very interesting. Thank you very much. All right. So now we have some time for Q&A so that's exciting. Thanks, everybody, who's posted the great questions. So as we go through, I'll read the questions. I'm going to repeat all of them twice so that the speakers and the attendees on the line can understand the question being asked.
Our first question, this one is for Chelsea. Chelsea, could you comment on your team's decision to use real-world data for this study as opposed to from another data source? Do you see it becoming a more regularly considered option to supplement learnings from basket trials for rare tumors or multi-tumor cohorts? One more time. Chelsea, could you comment on your team's decision to use real-world data for this study as opposed to data from another source? Do you see it becoming a more regularly considered option to supplement learnings from basket trials for rare tumors or multi-tumor cohorts?
Chelsea Shao: Thank you for the question. So at Merck we actually always start with the strategy and the research questions. We will evaluate as a team to see what's the best approach to address the research question. Certainly in this case, because our question is the TMB prognostic effect across these 10 uncommon tumor types, these Flatiron pan-tumor data sets include every one of these 10 tumor types with reasonable sample size. And also Foundation Medicine, I say for the TMB testing is quite consistent with the methods used for our KEYNOTE-158 trial for the TMB measurement. So that's why we believe this is fit for purpose to understand the prognostic value for TMB across these 10 tumor types.
So actually, the second question, with the emerging real-world data sets, such as the clinico-genomic data sets, and also a lot of improvements of the data quality, methodology, technology, I do think that there will be a broader application of real-world data in this oncology arena, and especially for the rare tumors or understudied populations. Of course, we need to consider... The data needs to be fit for purpose and should have the sound study designs, close communications and etc. I do think the real-world data is an important component part with the totality of the evidence. Thank you.
Olivier Humblet: Thank you, Chelsea. The next question is from Mike. Mike, how do we support the dialogue around a more pragmatic approach to evidence with regulators and payers? And once again, how do we support the dialogue around a more pragmatic approach to evidence with regulators and payers?
Mike Spencer: I think part of it is obviously having those direct dialogues, whether it's through industry association, whether it's engaging with scientific debates at meetings, etc. As I refer to, I do think there's also a need for us to support the evidence base for these approaches to be able to hopefully do some of the work that I know is ongoing around validating what we see within clinical trials with what we see within real-world data sources. Being able to try to show that we can get out of the clinical trial paradigm in certain cases, and be able to show the limitations of that as well as to where it is appropriate.
And then I think that we can support that move to a more pragmatic approach. We've seen an enormous amount of movement, particularly from the regulators, who maybe have somewhat different incentives. But we are seeing also on the HTA and payer side a move to be more accepting of these methods, but we've got a way to go, I think.
Olivier Humblet: Absolutely. Thank you, Mike. Great. The next question is for Tamara. Tamara, what is Flatiron's approach to capturing clinical variables that are typically defined for a single tumor type in a tumor-agnostic cohort? Once again, what is Flatiron's approach to capturing clinical variables that are typically defined for a single tumor type, but in a tumor-agnostic cohort?
Tamara Snow: Great. That's an awesome question. So typically, Flatiron designs our data models in a disease-specific way. So for this, we really collaborated very closely with our clinicians and our clinical data experts in designing a different approach where we're capturing the same data element across a pan-tumor cohort. So I think there were a few things that enabled us to develop this novel approach to abstraction in both a scalable way that still upheld data quality.
So first was designing a flow of abstraction that just makes it easier for abstractors to digest and capture this information from the patient's chart regardless of their tumor type. For example, the first question we have an abstractor to answer is whether or not the patient has a solid versus a heme malignancy. And then pending their response to that, they get queued to the next set of questions specific for either solid or heme tumors.
So, we slowly filter them down into more nuanced data elements like stage at initial diagnosis to make it easier to capture those more complex data elements. And then I think another important point is that we're making sure that we're designing policies and procedures for abstractors in a way that allows us to capture that same data element in a harmonized way, without overlooking important tumor-specific nuances. As I mentioned, we're capturing stage on initial diagnosis across the tumor-agnostic data model.
We understand that how this is defined and captured in the EHR can vary across diseases. So we've developed a standardized approach that also calls out or counts for tumor specific nuances. For example, cancers of the CNS do not typically have a numerical group staging. So we note that in our modules, we train our abstractors to expect that when capturing information on those patients. And then the third point, and probably the most important is that what's really core to a lot of Flatiron's abstracted data models is having a robust process for both training and testing abstractors before they are clear to abstract any new module, as well as QA/QC measures to ensure that abstraction is done correctly. And we're continuously monitoring that data quality.
So, this approach to abstraction really makes sure that abstractors are comfortable with abstracting this novel data model before launching it and scaling it. And so I really think this methodology to data abstraction designed by our experts internally, really enabled us to abstract these clinical variables that are typically defined for single tumor type in a tumor-agnostic cohort.
Olivier Humblet: Fantastic. Thank you, Tamara. So let's see. The next question is for Jeff. So Jeff, what opportunities lie ahead to make genomic testing an essential element of treatment planning across more tumor types, and even when approved, targeted therapies are not available? I'll repeat that again. Jeff, what opportunities lie ahead to make genomic testing an essential element of treatment planning across more tumor types, and even when approved, targeted therapies are not available?
Jeff Venstrom: Yeah, thank you, Olivier, for that question. It's something that definitely keeps me up at night and is I think really relevant for patients as we think about what is the role of comprehensive genomic profiling for different diseases. I think one of the just fundamental opportunities include education and level setting on what is the added value of getting something like an NGS-based profile, rather than single biomarker testing in specific diseases that could inform treatment sequences, that could inform opportunities for clinical trial enrollment. So I do believe that first and foremost, there's a huge education piece around the role of broad genomic profiling and testing in the clinic that's really tumor-agnostic or across the horizon for multiple different tumors.
Two other areas from a research perspective that I think are also big opportunities is in the early disease setting. So asking the questions for de novo, patients that are potentially diagnosed with curable malignancies, a lot of our current testing patterns are in really late stage advanced metastatic disease, where you're looking for your last option. Thinking and asking the question, what is the role for immediate first testing of using genomic profiling for patients newly diagnosed is an active area of research and really important, I think as an emerging opportunity.
From a data perspective too, we're really excited and love to partner with others around understanding the role of bringing some of this clinical genomic data into molecular tumor boards and as innovative clinical decision support tools too. This is also another active area of research and questions that we're asking internally and love to identify partners too to really help what is the potential role in the clinic tomorrow for asking these questions using clinico-genomic data, particularly for potentially rare patients or patients that you're trying to make a decision that has a rare histology or rare genomic alteration and there's not much data available in the literature. So those are three active areas that I think are opportunities that we're excited about advancing.
Olivier Humblet: Awesome. Thank you, Jeff. Great, so we're almost at time. So let me move on to one good question to end on here. This is a question for the panel. So, ideally, what does the future of tumor-agnostic real-world data look like for you? Jeff, from a data availability perspective. Tamara, from a data development standpoint. Chelsea, for applying it in the drug development lifecycle. And finally, Mike, to overcoming market access hurdles. One more time. So ideally, what does the future of tumor-agnostic real-world data look like for you? Jeff, from a data availability perspective. Tamara, from a data development standpoint. Chelsea, for applying it in the drug development lifecycle. And finally, Mike, to overcoming market access hurdles. Jeff, maybe you could start and then followed by Tamara and Chelsea, and then Mike to close it out.
Jeff Venstrom: Sure, just really quickly, in the interest of time, I guess two areas is one really expanding the access to very different outcomes, and data components. I think some of the questions that have come up in the chat too around different imaging modalities, different capabilities, I'd love to see more patient reported outcome data available in the real-world setting for application, for decision-making. And then also the quality that gets you driving towards that regulatory grade real-world data products to really enable additional totality of evidence for regulatory drug submissions.
Tamara Snow: Great. I'll also try and keep it short. So on top of continuing to develop additional data elements to unlocking tumor analyses like a pan-tumor line of therapy approach, I think it'll be important for us to ensure that we're providing a data set that allows for both the tumor-agnostic view as well as the disease-specific deep dives, to really ensure that researchers can tease out how effectiveness might vary across the different sub cohorts, especially within a real-world setting.
And then the second thing that I think will be important is making sure that we're building out really robust analytical guidance for researchers using our data, not only clearly defining what the various data elements mean, and how they're defined across diseases, but also how to run a real-world data analysis in a tumor-agnostic cohort. So folks have lots of really interesting questions in the chat. I know some things like considerations for how to decide what index date you should use when looking at an overall survival analysis across a tumor-agnostic cohort is one important point for us to be thinking about moving forward.
Chelsea Shao: Great, so just to try to keep it in short. So yeah, the real-world data definitely is very helpful throughout the clinical development lifecycle including to understand the disease epidemiology, biomarker epidemiology, to identify the target population, to understand the current treatments and also outcomes, etc. So after the drug is on market, certainly the real-world data are very helpful for the safety effects we need to study and etc. But certainly, we need to keep in mind that everything should come from the strategy and what's the research question, what data is fit for that purpose and etc. Thank you.
Mike Spencer: So probably a good job to be last because I think everything has been said, Olivier, and we're over time. I think that last point around being able to follow from the pre-launch into the launch phase, and actually be able to see the objectives at scale across multiple countries in the real-world once the treatment's available, I think that would probably be the additional thing that I would just add to.
Olivier Humblet: Fantastic. Thank you so much, Mike, and all those speakers. That concludes the Q&A portion of our webinar. I want to give a huge thanks to all of our presenters for making the time to come and share their insights here today. Thank you all very much. And also thank you to everyone who joined. If we did not get to your question today, we will be sure to follow up with you via email. And if anyone has any questions about the content presented, please don't hesitate to reach out to your Flatiron Life Sciences contact or at firstname.lastname@example.org.
I'll also add, everybody who's registered will within a week or so receive a link to the recording and slides from this webinar. And also a friendly reminder to please take a survey upon closing out which will help us improve future webinars. So thanks again. Have a great rest of your day, everybody, and stay healthy and stay safe. Thank you.