Latest / Lab to Lives / Tomorrow's Cures Are Hiding in Today's Patient Records with Vish Srivastava

Transcript

Tomorrow's Cures Are Hiding in Today's Patient Records with Vish Srivastava · Lab to Lives

• 0:18
  Speaker 1: Welcome to another episode of Transformation in Trials.
• 0:21
  Speaker 1: I'm your host, Ivana Rosendahl.
• 0:23
  Speaker 1: In this podcast, we explore how clinical trials are currently transforming so we can identify trends that can be further accelerated.
• 0:30
  Speaker 1: We want to ensure that no patient has to wait for treatment and we get drugs to them as quickly as possible.
• 0:36
  Speaker 1: Welcome to another episode of Transformation in Trials.
• 0:40
  Speaker 1: I'm very excited to have Vish Srivastav in the studio with me, and we are going to focus on patient data from the real world.
• 0:48
  Speaker 1: Vish, could you give us a brief introduction to yourself before we continue to the content?
• 0:56
  Speaker 2: Absolutely Well.
• 0:57
  Speaker 2: First, ivana, thank you so much for having me on.
• 0:58
  Speaker 2: It's great to be here with you.
• 0:59
  Speaker 2: My background is I've been building software and healthcare for a little over a decade and now building Century Health.
• 1:07
  Speaker 2: And what we do at Century is we accelerate breakthrough treatments by applying AI to clinical data.
• 1:13
  Speaker 2: And specifically what that means is we unlock messy, siloed and unstructured data locked away in EHRs to enable real world evidence for pharmaceutical companies.
• 1:19
  Speaker 2: To enable real-world evidence for pharmaceutical companies.
• 1:24
  Speaker 1: I am so excited to have you in the studio with me, especially because one of the first episodes we did three years ago was about the untapped potential of EHRs.
• 1:34
  Speaker 1: But this was kind of before AI had stepped into the scene and it was quite cumbersome back then.
• 1:40
  Speaker 1: So I'm very interested to hear how this might have changed now.
• 1:46
  Speaker 2: Definitely.
• 1:46
  Speaker 2: Well, in some ways not too much has changed.
• 1:49
  Speaker 2: Ehr data, for all the talk about interoperability, still remains quite siloed.
• 1:54
  Speaker 2: But I think really what has happened the last two years is we now have a whole new set of technologies that allow us to break down some of those silos.
• 2:01
  Speaker 2: And historically, ehr data in general, but also specifically patient registries, have been an incredibly cumbersome process.
• 2:12
  Speaker 2: I was just talking to a researcher in rheumatology a couple of weeks ago and she was telling me about how she's been running a registry for 10 years and they have a little under a thousand patients in this registry just to give a sense for how long it takes to even get a relatively small patient population into a registry.
• 2:30
  Speaker 2: And it is now possible to automate a lot of those manual workflows, right the recruitment of the well, first recruitment of the site, then the recruitment of the patient to participate in the registry, then the recruitment of the patient to participate in the registry and then, of course, all of the manual processes of entering that data sometimes still on paper forms into into CRFs.
• 2:51
  Speaker 2: That all get recorded.
• 2:52
  Speaker 2: And if all that sounds familiar, you know that's pretty similar to running a clinical trial and you know.
• 2:58
  Speaker 2: And of course the difference is there isn't an intervention for a registry, it's observational research Essentially.
• 3:04
  Speaker 2: Essentially we're pretty obsessed with, you know, in real world evidence.
• 3:08
  Speaker 2: A lot of people talk about fit for purpose data.
• 3:11
  Speaker 2: We like to obsess about fit for purpose data infrastructure.
• 3:14
  Speaker 2: So how do we essentially, from the ground up, build the infrastructure for building registries, for creating these valuable longitudinal data sets and starting with, like, what is the data that's required and how do we create the most efficient workflow surrounding that data, versus just taking what might be a preconceived notion about data collection from you know historical ways of collecting data and just kind of shoehorning that into registry study design?
• 3:40
  Speaker 1: Well, Vish, tell us a little bit more about the lifecycle of a registry.
• 3:43
  Speaker 1: Who usually gets the idea to create a registry and which parties are involved?
• 3:50
  Speaker 2: Yeah, great question.
• 3:51
  Speaker 2: You know there are a lot of different types of registries out there and if you just go to clinicaltrialsgov, which is a pretty small sample of the overall set of registries that are created, those are just the ones that of course get recorded into the overall database.
• 4:07
  Speaker 2: But even in that sampling of registries there's quite a mix.
• 4:11
  Speaker 2: Sometimes registries are created for quality management purposes, so payers, for example, would require that certain outcomes safety and efficacy outcomes are recorded by clinicians and by providers.
• 4:25
  Speaker 2: In other cases, registries are created for research purposes, right?
• 4:29
  Speaker 2: So how do we understand the natural history of the disease?
• 4:31
  Speaker 2: How do we understand the subtypes of a specific disease?
• 4:34
  Speaker 2: This is particularly important complex and chronic conditions like MS and lupus and even IBD, for example.
• 4:40
  Speaker 2: Those are the ones that we're very interested in, where the primary purpose is to understand the progression of the disease and the application of existing treatments in that patient population and seeing in the real world what safety and efficacy and comparative effectiveness looks like.
• 4:59
  Speaker 2: And we think that there should be a lot more of those registries.
• 5:02
  Speaker 2: That the barriers to create them and the time it takes to create those is a solvable problem with the current state of AI.
• 5:12
  Speaker 1: That is very exciting and I see that there are many different applications and it can help us bridge some of those data gaps that we don't quite get covered in clinical research but that are important for actually marketed products also in clinical research, but that are important for actually marketed products also.
• 5:28
  Speaker 1: So maybe what has happened technologically that makes it easier today.
• 5:32
  Speaker 2: Yeah.
• 5:33
  Speaker 2: So stepping back from technology for just a second, so like what's the problem that we're trying to solve and then thinking about how does technology solve that problem?
• 5:41
  Speaker 2: So when we've looked at protocols for existing registries and observational studies, we ended up finding that there's a pretty significant overlap between the data that's collected in the sort of old school way of doing it like a CRF, a form that's filled out by a clinician or a clinical abstractor.
• 6:01
  Speaker 2: There's a pretty significant overlap between the data that's collected for the registry and data that's collected as a part of the standard of care and that's recorded in the EHR.
• 6:10
  Speaker 2: And you know that includes some pretty basic things like, of course, diagnoses, for example ICD-10 codes, but certainly you know symptoms, medication, history, demographics.
• 6:24
  Speaker 2: You know a lot of this information that you kind of need to collect to provide high quality care to the patient anyway.
• 6:31
  Speaker 2: The problem, of course, is that two things.
• 6:34
  Speaker 2: One, the structure of the data is quite different.
• 6:37
  Speaker 2: Again, we like to say lots about interoperability and kind of standardized data in healthcare and in EHRs, and it's true FHIR and other kind of data standards have moved the needle towards data being interoperable.
• 6:53
  Speaker 2: In reality there's still a fair amount of manual work that's required to map what data is available in the EHR and what's the data model that's required for analysis.
• 7:04
  Speaker 2: And at Century we're big believers in OMOP, which is a data model that has been quite widely adopted and has a fantastic and active community around it.
• 7:14
  Speaker 2: But OMOP is different from FHIR, so that simple mapping of fields and input to fields and output for analysis is one challenge.
• 7:23
  Speaker 2: The other challenge is and well, for me this is the exciting problem space so much of the most valuable insight about disease progression and about what treatments are actually safe and efficacious for patients.
• 7:36
  Speaker 2: That's locked away in clinical notes.
• 7:38
  Speaker 2: So, just to give an example here, we do a fair amount of work in neurology and in MS in particular, to treat a patient.
• 7:48
  Speaker 2: You will ask a series of questions about what's often thought of as kind of functional areas, right?
• 7:53
  Speaker 2: So what's your visual acuity, for example?
• 7:55
  Speaker 2: What are sort of any cognitive symptoms you might be experiencing?
• 7:59
  Speaker 2: The neurologist or the nurse is jotting down all of those symptoms into the clinical note, in just free text, and transforming that free text to a structured data point that can be analyzed.
• 8:15
  Speaker 2: That takes well, one an understanding of natural language, but two clinical context.
• 8:20
  Speaker 2: So that's the second problem that you know that we need to solve that and for both of those AI can be useful.
• 8:28
  Speaker 2: Ai is useful for doing that sort of mapping of hey.
• 8:32
  Speaker 2: This field has some, you know, very archaic label in the database for race and ethnicity, for example.
• 8:42
  Speaker 2: Map that to the OMOP field.
• 8:44
  Speaker 2: But the current state of LLMs means that we can fine tune existing LLMs to do that abstraction, right Reading the clinical note and turning it into a high quality structured data point that can now be analyzed.
• 9:01
  Speaker 1: That is incredible and that is definitely new and was not possible just those three years ago.
• 9:06
  Speaker 1: I'm curious what has been some of your favorite insights that have been able to generate this way?
• 9:14
  Speaker 1: Has there been some moments where you were like, wow, this is what this is all about.
• 9:18
  Speaker 1: This is the change that we can make for our patients?
• 9:22
  Speaker 2: Yeah, I think our work in MS is a place where we get quite excited about the opportunity to unlock new subtypes or understand new subtypes of the disease.
• 9:36
  Speaker 2: So a lot of the drug development activity in the last handful of years builds on the current foundation of therapeutics for MS, which have really focused on controlling relapses, it turns out, once you've really controlled the relapses over time, that gets you to the next set of clinical questions, which is there is a progression of disability over time, even if you've controlled relapses, and how do we improve the quality of life for these patients over the duration of this disease?
• 10:11
  Speaker 2: And this gets back to our earlier discussion about what is that data that's locked away in the clinical note and what we've been, and we're kind of just working through that data now.
• 10:22
  Speaker 2: So there's going to be a lot more of, I think, very valuable insights as we move further down the path.
• 10:27
  Speaker 2: Um, but what we're able to see is that there is there.
• 10:32
  Speaker 2: There are these records of symptoms and an understanding of the progression of that disability over time that you can only understand if you look at 10 plus years of data for patients.
• 10:45
  Speaker 2: And there is a clear interest from researchers in MS as well as pharma companies that are developing treatments for these unmet needs, these underserved populations in MS, and we can really inform those questions by looking at this longitudinal rich data for MS patients.
• 11:04
  Speaker 1: That is an incredible example and also shows well, even though this is a new technology, we can actually make a difference for patients by looking back at the data that they already have provided over the long term.
• 11:17
  Speaker 1: How about the access to the data in these health records?
• 11:24
  Speaker 1: What has been some of the challenges in finding a good collaboration with the providers of the data or, potentially, the healthcare system?
• 11:35
  Speaker 2: What we've found is that providers are incredibly excited to unlock value from their EHR data sets.
• 11:43
  Speaker 2: That includes unlocking insights and advancing research, advancing that horizon of understanding any particular disease or therapeutic area, but also facilitating collaborations with life sciences.
• 11:58
  Speaker 2: And it's been remarkable how often we hear from individual clinician researchers or CMIOs, for example, that they've always had in the back of their head that they want to create registries, but it's just been cost prohibitive to do so.
• 12:13
  Speaker 2: I will say the biggest challenge of working with providers to create these registries is really about IT infrastructure.
• 12:21
  Speaker 2: It can take a significant amount of time to work through the approvals, for example, to integrate with an EHR, and I think this is for good reason.
• 12:34
  Speaker 2: Data security and patient privacy is a very important value of Century and obviously is a very important value for the stewards of these data, the providers themselves.
• 12:42
  Speaker 2: So it's very important that you know a provider will work through the process of ensuring that our platform is SOC 2 compliant, for example, that is, HIPAA compliant, that we are appropriately de-identifying this data to protect patient privacy.
• 12:57
  Speaker 2: So those are important checkpoints along the way in these collaborations.
• 13:01
  Speaker 2: But it does take time and it's important that all the stakeholders involved are comfortable with the way in which this data is used and it's respecting the wishes of the patients you know, for whom, ultimately, we're doing all of this work in the first place.
• 13:18
  Speaker 1: Yeah, and I completely agree.
• 13:20
  Speaker 1: We often talk about the barrier of access to data, but it is there for a good reason.
• 13:24
  Speaker 1: We are trying to make sure that patients consent to the usage of their data and ultimately do get to benefit from the uses.
• 13:32
  Speaker 1: I would be curious about now collecting registry data.
• 13:37
  Speaker 1: Have you experienced, you can say, how that data, those insights created, can be used back into the clinical area where potentially I don't know drugs can be used for new purposes or we find gaps in treatment?
• 13:55
  Speaker 2: Yeah, you know we we've seen so many interesting use cases for RWE and those use cases just continue to grow.
• 14:03
  Speaker 2: There were, I believe, three guidances from FDA just last year about applying RWE in drug development.
• 14:12
  Speaker 2: So this is a rapidly expanding field.
• 14:15
  Speaker 2: You know at a super high level.
• 14:17
  Speaker 2: Expanding field.
• 14:21
  Speaker 2: You know at a super high level.
• 14:22
  Speaker 2: You know these are stats that get thrown around a fair amount.
• 14:23
  Speaker 2: But you know the average drug takes $2.6 billion to develop and 12 years on average.
• 14:27
  Speaker 2: So once that drug is approved there's a mandate to get that drug to the patients that could benefit from it as quickly as possible.
• 14:36
  Speaker 2: So many of the initial use cases that we collaborate with pharma companies around, kind of center around.
• 14:42
  Speaker 2: Once that drug is approved, how do we work with, for example, commercial and medical affairs teams to really understand in the real world what is the comparative effectiveness of this treatment?
• 14:51
  Speaker 2: For example, what is that medical unmet need that perhaps we had an idea about during the clinical trials and during clinical development but need a more kind of real world and richer assessment of now that we are in market.
• 15:06
  Speaker 2: But, to your point, there are many use cases within clinical development as well that RWE can be valuable and you know this includes, for example, protocol design.
• 15:18
  Speaker 2: Can we have a good sense for you know what are the patient subpopulations that you know could maybe most benefit from a particular treatment and design the protocol based on what we already see in the real world?
• 15:33
  Speaker 2: And then also to your point, label expansion.
• 15:36
  Speaker 2: You know a drug might already be approved for one indication.
• 15:40
  Speaker 2: Of course, the classic example that everyone's talking about these days is you know GLP-1s.
• 15:47
  Speaker 2: You know currently the label is diabetes and obesity and you know, in some cases, a couple others.
• 15:50
  Speaker 2: We are already seeing very interesting real world data about, for example, how GLP-1s can benefit patients with MASH, about, for example, how GLP-1s can benefit patients with MASH.
• 16:03
  Speaker 2: And there are a whole host of other interesting indications, including Alzheimer's disease, for drugs like GLP-1s.
• 16:08
  Speaker 2: So can we look at the real world, identify patients that might be benefiting from some of these treatments and include that in kind of prioritizing pipelines, for example, and can we de-risk a particular clinical trial to expand the label?
• 16:26
  Speaker 2: So I get really excited about the current and, I think, soon-to-be-discovered use cases for RWE across the drug lifecycle.
• 16:35
  Speaker 1: Yeah, I think there are multiple ways where it could help us lead, just by providing more data and helping us target the treatments that we're developing better, and we haven't really had that opportunity before.
• 16:47
  Speaker 1: That is amazing.
• 16:49
  Speaker 1: I would be curious, vish, are you mainly focusing on the US market right now, or are there any other markets that seem interesting in this regard?
• 17:12
  Speaker 2: It's a great question.
• 17:13
  Speaker 2: So initially we're focused on the US and on three specific specialties gastroenterology, neurology and rheumatology and the reason for that is, as we were discussing earlier, one of the biggest barriers to creating these AI-powered registries is the data infrastructure and data integration itself.
• 17:28
  Speaker 2: So how do we form that collaboration with the providers that are the stewards of this data?
• 17:32
  Speaker 2: So we have to be pretty focused on how do we essentially start with a population of providers and patients where we can have a replicable technology, and you know, our platform first has to be HIPAA compliant and then there are a series of other regulations, for example in the EU and GDPR, that we would need to kind of start to implement some of those requirements in order to expand, start to implement some of those requirements in order to expand.
• 18:00
  Speaker 2: So we are starting here and having sort of initial traction and initial success in those specialties, but absolutely the goal is to expand across therapeutic areas and certainly to expand globally from here.
• 18:14
  Speaker 1: That'll be exciting to follow.
• 18:16
  Speaker 1: Vish, you're obviously very passionate about this area.
• 18:19
  Speaker 1: I would be curious what led you into this specific space.
• 18:23
  Speaker 1: What was your journey to this area?
• 18:28
  Speaker 2: You know, there are really two things that motivate my work today and really motivate why we created Century Health.
• 18:35
  Speaker 2: The first is I've been building software and healthcare for a little while now, but specifically for the last five years or so I've been working in healthcare data and in particular, prior to Century, I worked at a company called Evidation Health, which really has been at the forefront of integrating wearable device data into clinical trials but, more broadly, clinical research, and I was head of product for a platform that had about 5 million patients on it and it was extraordinary work with extraordinary people and I learned so much through that.
• 19:11
  Speaker 2: And part of what I learned was that, you know, there is now pretty ready access to a few sources of data, including claims, for example, as well as, in many cases, wearable device data.
• 19:23
  Speaker 2: But pretty often I heard about this desire for clinical data that comes from EHRs.
• 19:29
  Speaker 2: So there was a bit of an initial spark for hey, is now the time to create a company that can really unlock that EHR data.
• 19:36
  Speaker 2: So that's maybe, on the front, what motivates my work.
• 19:41
  Speaker 2: On a more personal note, a couple of years ago my grandfather's Alzheimer's disease progressed to the point where he couldn't really recognize his grandkids anymore, and this is heartbreaking to see firsthand.
• 19:55
  Speaker 2: I'd just gotten married and my wife would reintroduce herself to him every few days, and I know it was.
• 20:03
  Speaker 2: I know it was really challenging for him to recognize in himself that he was losing his, his memory and his cognitive abilities A remarkable man, a poet and a playwright and a professor, and that that sort of sent me down a rabbit hole to.
• 20:19
  Speaker 2: You know, I'm a technologist, not a researcher or a clinician.
• 20:22
  Speaker 2: But I did sort of a quick literature review to understand you know what is our, what's the current, you know, state of the science around understanding Alzheimer's disease.
• 20:30
  Speaker 2: And I found that, you know, even for a disease that afflicts millions of people around the world, we don't have a really deep understanding of how this disease progresses and what all the biomarkers and drivers of it are.
• 20:42
  Speaker 2: And now we have two therapeutics approved, which is incredible.
• 20:47
  Speaker 2: But we have so much more to do.
• 20:48
  Speaker 2: And it felt like one of the bottlenecks in understanding neurodegenerative diseases but across many chronic diseases, is access to good longitudinal clinical data.
• 21:01
  Speaker 2: And that's another part of what kind of motivates my work and motivates what we do at Century.
• 21:07
  Speaker 1: That makes a lot of sense, and especially for some of these diseases where we have been trying to crack them for a long time.
• 21:13
  Speaker 1: But it is hard to recruit patients, especially for some of the diseases that you mentioned, and to gather longitudinal data.
• 21:22
  Speaker 1: But it's already there.
• 21:22
  Speaker 1: That is a brilliant approach.
• 21:26
  Speaker 1: Well, vish, as we start rounding off, I always ask our guests the same question towards the end, and that is if I gave you the transformation trials magic wand that can change one thing in the life sciences industry, what would you wish to change?
• 21:47
  Speaker 2: I love this question.
• 21:48
  Speaker 2: We, as a quick aside, our AI model is called CHARM, the Sensory Health Abstraction and Retrieval Model, but we like that word CHARM because we think we're applying some magic in real-world evidence and life sciences.
• 22:03
  Speaker 2: But to answer your question directly, you know, if I could wave that magic wand, it would be to make patient data accessible and actionable.
• 22:12
  Speaker 2: This data to your point earlier, it exists.
• 22:15
  Speaker 2: There's so much important insight that's locked away in our understanding of disease and potentially, that breakthrough treatment that we can get to patients.
• 22:25
  Speaker 2: Instead of taking 12 years, maybe it'll take five years.
• 22:27
  Speaker 2: Maybe instead of once that drug is approved, you have to wait another five years for your insurance company to cover that treatment.
• 22:35
  Speaker 2: Then it could be available the next day.
• 22:37
  Speaker 2: The next day because the real world evidence is there and it's compelling about the economic value.
• 22:42
  Speaker 2: You know the cost effectiveness of that treatment.
• 22:45
  Speaker 2: So it feels like we can benefit patients like I'll start again there.
• 22:52
  Speaker 2: It feels like we can create dramatically better outcomes for patients if we can just make the data that's already collected from their care accessible and actionable.
• 23:04
  Speaker 1: I think that's a great wish, and it seems like we're actually getting closer and closer to having that wish granted.
• 23:10
  Speaker 1: That's amazing.
• 23:13
  Speaker 1: Well, vish, where can our listeners find you to have more questions answered, to learn more about your company or yourself?
• 23:23
  Speaker 2: Well, I always love to compare notes with other folks in the space, so please do reach out.
• 23:27
  Speaker 2: You can find me on LinkedIn.
• 23:30
  Speaker 2: You can also find us at centuryhealth, and just hit that contact us button.
• 23:36
  Speaker 1: Thank you so much, Vish.
• 23:37
  Speaker 1: It was amazing having you on the show.
• 23:38
  Speaker 1: I loved our conversation.
• 23:39
  Speaker 1: This was a fantastic conversation.
• 23:39
  Speaker 1: Thank you so much, Mish.
• 23:39
  Speaker 1: It was amazing having you on the show.
• 23:40
  Speaker 1: I loved our conversation.
• 23:41
  Speaker 2: This was a fantastic conversation.
• 23:42
  Speaker 2: Thank you so much, Ivana.
• 23:47
  Speaker 1: You're listening to Transformation in Trials.
• 23:50
  Speaker 1: If you have a suggestion for a guest for our show, reach out to Sam Parnell or Ivana Rosendahl on LinkedIn.
• 23:58
  Speaker 1: You can find more episodes on Apple Podcasts, spotify, google Podcasts or in any other player.
• 24:04
  Speaker 1: Remember to subscribe and get the episodes hot off the editor.