Transformation in Trials
A podcast about the transformations in clinical trial. As life science companies are pressured to deliver novel drugs faster, data, processes, applications, roles and change itself is changing. We speak to people in the industry that experience these transformations up close and make sense of how the pressure can become a catalyst for transformation.
Transformation in Trials
Revolutionizing Clinical Trials: The Promise of AI and Predictive Analytics with Vinodh Balaraman
When nine out of ten clinical trials don't cross the finish line, it's time to ask tough questions and seek innovative answers. That's exactly the journey we embark on with Vinodh Balaraman from KolateAI in our latest podcast episode. Together, we unravel the complex knot of challenges that lead to clinical trial failures, from flawed study designs to enrollment obstacles and the unexpected curveballs thrown by the approval of competitor drugs. But it's not all doom and gloom; we shine a light on the potential of AI to bring about much-needed predictive capabilities, paving the way for trials that are as efficient as they are effective.
Ever wondered how big data and predictive analytics could be the game-changers in drug development? Our conversation takes a deep dive into the role of AI in forecasting how patients will respond to treatments, potentially transforming the hit-and-miss nature of clinical trials into a targeted approach. We discuss the transformative power of a SaaS business model and shared knowledge bases, considering how they could overhaul the pharmaceutical landscape. Despite the industry's tentative steps toward data sharing, we explore the gradual shift as medical professionals begin to wield these cutting-edge tools to make smarter, data-driven decisions.
Bringing a dose of humanity to the tech talk, I share my personal motivation stemming from a family member's battle with rheumatoid arthritis—a catalyst for my commitment to pushing the envelope in clinical development. The episode concludes with a reflection on how AI stands on the brink of revolutionizing medical affairs and real-world studies, but is often met with skepticism. By fostering a culture open to change and innovation, we envision a future where clinical trials become more successful and efficient. Join us as we discuss the landscape of trials and the untapped potential that KolateAI brings to this critical field.
Guest:
Vinodh Balaraman
________
Reach out to Sam Parnell and Ivanna Rosendal
Join the conversation on our LinkedIn page
Welcome to Transformation in Trials. This is a podcast exploring all things transformational in clinical trials. Nothing is off limits on the show and we will have guests from the whole spectrum of the clinical trials community and we're your hosts, Ivana and Sam. Welcome to another episode of Transformation in Trials. Today in the studio with me I have Binod Balaran, who is the co-founder and CEO of Collate AI, and Collate AI is on a mission to accelerate and streamline drug development leveraging AI, and today we're going to focus on the topic of why clinical trials fail. And, Binod, just to get us started, can you take us through some of the typical reasons why clinical trials fail?
Speaker 2:Yeah, thank you so much, Ivana, for the opportunity, really appreciate it. Yeah, addressing the question you just brought up, there's a lot of reasons why clinical trials fail. Unfortunately, there's also a very high failure rate, to the tune of almost 90% in clinical development. Wow, these drivers okay. So let's talk about some of the core ones. Efficacy and safety, okay, are among the top reasons. Among the top reasons, obviously, one of the focuses of the trial is to make sure that drug is producing the effect that is desired. If that's not achieved relative to the standard of care, that's an issue. Similarly, safety goes without saying. You do not want adverse events and side effects unrelated to the issue you're trying to cure. So that's another big one.
Speaker 2:There's other design-related issues like study design flaws, insufficient sample size, regulatory issues that were missed that can come into play. There's logistical issues that can happen. Oftentimes. Enrollment is a challenging, challenging dimension, you know. So you got to get to the right sites and you got to have the sites and you need to find people that fit the criteria you're looking for. Protocol violations, data quality issues those are additional ones. And, last but not the least, I would say there are company-related issues like financial difficulties or organizational changes and things of that sort that may come into play as well. Right and external factors. Right, I mean you may have competitor drug approvals that happen, which sort of you know inhibit, you know stop you from moving forward with this, you know in the direction you're thinking of um. So those are some of the, some of the drivers, and there are some commonalities to these right uh yeah, I'm surprised to hear that not many of these regions have anything to do with the drug itself.
Speaker 2:These are more like issues surrounding the drug true, true and uh, there are, but you know the probably the bigger, I mean percentage wise, I think, yeah, the drug related issues are up there, right. So, in terms of contribution to the eventual number for many of these is the fact that you know, pharmas don't have great visibility into upcoming future events, clinical events and future trial trajectory, right. So they kind of they're operating off of retrospective data, which is somewhat limited, of course, in scope. They don't have great tools to project the direction things are going or a crystal ball to see the direction it may be going. So oftentimes you're moving in a certain direction and you find too late that that direction is not the direction you should go in, whether it is in terms of an endpoint that is very difficult to reach. You know the effectiveness not being at the level that should be because it's too dilutive. You're looking at a very broad set of patients versus focusing where the effect may be stronger.
Speaker 1:This issue, a broad set of issues relate to this poor visibility that you have, due to lack of tools and prospective data, of the direction things may be going. That makes sense there be some way that we could have more visibility in some of the information missing as we plan and conduct the trial so that we can correct some of these.
Speaker 2:Indeed, Indeed, there's ways we can leverage the information that's out there and put that together in a cogent way to uh, uh, get to the get to the type of visibility you need, Right, so, uh.
Speaker 2:So there are, uh, for instance, there are uh approaches coming into play where, uh, you can sort of simulate the entire trial, uh, as a macro event, right, and try to find a probability of success at that macro level.
Speaker 2:And there is another way which we're focusing on, which I will talk about, which is we put together retrospective data around completed clinical trials and real-world evidence in that therapy area, across the body of work that's been done, trying to look at various patient medical histories and how they reacted to all the different drugs that have been out there and how they reacted to all the different drugs that have been out there.
Speaker 2:And we pre-built these AI foundation models that predict major clinical events at a patient level for the trial in question. So, specifically, what I'm talking about is adverse events and serious adverse events, drug response and efficacy metrics and trial endpoints, which are combinations of these things I talked about. So we predict these events out at a patient level and leveraging machine learning and deep learning techniques, leveraging machine learning and deep learning techniques, and we make these predictions and insights available via a custom LLM co-pilot, which provides an interface for the medical staff to interact with these predictions so that in turn becomes a great decision support tool that assists the trial medical staff in making these critical medical decisions to design the protocol the right way, include the right set of patients and also then move into running the trial.
Speaker 1:I would be curious to understand this a little bit better. I would be curious to understand this a little bit better. For the tool to be able to predict some of these outcomes, what kind of data would you need to feed it with?
Speaker 2:Indeed. So, as I was just briefly touching on, we look at a combination of previously completed clinical trial data sets available from public and private sources and also real world data. And these are all data sets that show how you know it shows the full set of patient medical history, shows the conditions they've had, shows the different drugs they've taken, how they responded to that. They're obviously the patient characteristics you know demographics, psychographics, all of those. And then we have a model, the foundation model for the therapy area in question, that has already looked at all the different drugs and patient interactions that have happened. When we get into a target trial, we typically try to identify the closest analog or proxy in the data set that we've got. I mean, that's how the model automatically works, right, it's trained to understand all of those data points. It's picked up. So it will do the matching and pick up signals from there and it'll try to generate predictions for the for the drug in question, right, for the trial in question. Uh, and these predictions um are probably I mean, there's there's probably three, uh, three to four dimensions in which you could use this. One is, uh, identifying the best performing patient segments or clusters in advance, predicting those. So what I mean here is we can identify the patient groups you know which have the best drug response rate as well as the minimum side effects right. Number two we can help identify the trial direction in advance by comparing out the projected performance of the treatment arm with the control arm. Right, so we, and we try to do this starting from you know before starting the trial. Right, so you get a sense of, directionally, which way it's going to go and which metrics are likely to trend in what direction. So that's a way to get an early warning and deep dive and figure out the source of the issues.
Speaker 2:Number three we have our co-pilot, which is a custom LLM trained to work in this domain, that has been trained against all of these predictions that have come up through our deep learning approaches. And the co-pilot also works with trial actual data. So you can use it to manage the clinical trial by asking simpler queries as well. So you can query the predictions, manage the clinical trial right by asking simpler queries as well. So you can query the predictions that we came up with, but also other things like what has happened so far, how is site A done relative to site B, and so on. And, lastly, we we can extend this clinical outcome prediction into the real world, which is the phase four of the clinical trial right For real world studies, to provide further insight with a much larger real-world population, into how the drug itself should be shaped where it's working, where it's not working.
Speaker 1:How would a pharmaceutical company usually interact with your tool? Would they I don't know get a license for a specific trial? To answer some questions, would this have to become integrated into their clinical ecosystem? What is the model?
Speaker 2:Yeah, very good question. It would typically be a SAS model. So this is a software as a service and we it's typically based on, you know, we work with them to structure it based on the it's sized by tier, based on the number, work with them to structure it based on the it's sized by tier, based on the number of patients in the trial. So we have different thresholds based on the size of the trial in question and it's a monthly fee-based approach. So when we integrate this into a client's ecosystem, uh, yeah, we it's a fairly seamless deployment. I mean, we bring this pre-built model to bear. We would ingest any relevant uh target trial data, uh and any, any. We'd also look at any pre-clinical data that they have, so those are easily uh ingestible into our framework and and then we can be off and running with predictions on the trial that they want to target, starting from the design stage.
Speaker 1:And would it be possible for one company to utilize your tool with the data that is already ingested in the tool from other companies within the same drug area, or would the data need to be ingested from scratch every time?
Speaker 2:Yeah, I know that's a very good question. So the approach we're taking is we have created a central SAS-based model for each therapy area so that model leverages data that we have built in from all those sources. I talked about the completed private and public clinical trial data sets that we were able to pull in, as well as real-world data. So if a company is bringing its own data to the equation, there's different models we can work with them on. So we can certainly just have an instance just for that company right where we're not going to share that, we're not going to take their data and make it part of our central knowledge base, if you will. There's also another option possible where those that contribute to the knowledge base you know also get to leverage it. They get to leverage everybody else's contribution to the knowledge base in addition to just our central contributions.
Speaker 1:I am curious how this has been received, because data sharing is typically a contested area in life sciences. What has your experience been so far?
Speaker 2:No, it's definitely. That's why it's this is the second model that I mentioned is an optional. I mean it's not something we try to enforce in any fashion at all. So the initial clients have been more typically on the first model that I talked about, but we are looking to build momentum behind some way that this consortium that works with us can benefit collectively from everything. So it's areas that they're comfortable working in and want to learn more. That's where that approach could comfortable in working in and want to learn more. Right, that's where that approach could work.
Speaker 1:That makes sense. Who would the typical users be in a life science company of this tool Clinical development, medical, yeah.
Speaker 2:Where are we? Yeah, clinical development is a very core area, and medical affairs is the other area that would benefit as well, because these extend into the real-world studies as well, so you can generate these insights on how a drug is doing there as well.
Speaker 1:And how would the data usually be presented to the users? Are there graphs? Is it tables?
Speaker 2:Yeah, great question. So we have canned reports or dashboards that come out of this. So once the prediction models run and they keep getting updated on a regular basis quasi real-time I would say not truly real-time, but very frequently we produce dashboards which obviously provide a consistent, simple view. But the other important dimension is that we have the co-pilot. So the co-pilot allows the medical staff, the trial medical staff, to have a conversational English language interface by which they can access all these predictions and then query in different fashions around it, right. So that provides a great degree of flexibility, right, and additional ways to deep dive in different ways that those core dashboards may or may not. We try to cover some scenarios but we like to provide flexibility, right that makes sense.
Speaker 1:Uh, the company itself is about a year and a half old, as far as I understand. What prompted you to bet on AI within this area?
Speaker 2:Yeah, you know there is a very, very pertinent question, right. So both me and my co-founder have seen the pain in this clinical trial space, you know professionally, and I also have some personal reasons that got me going in this area. So I'll start with the latter. You know so, one of my close family members you know has been suffering for a long time, you know, with RA and you know she got on to. I mean, it's been a very difficult journey where you know she uh from a young age she uh was diagnosed with this and uh started getting affecting her joints. Uh, and it's been difficult to um, you know there's been nothing great that's been coming out. I mean, the trials that she's leveraging are not. You know they had issues, they didn't work well, you know they went the wrong way and then it it had direct wrong effects on her Right. So so we just really wanted something that got to got to a successful conclusion sooner, right, I mean they look promising on paper but they turned out not to be right, leading down the wrong path. So I've been burned, on the personal side, I think. But professionally, we've seen in our work with life sciences companies, particularly in trials, we've just seen these huge failure rates and huge costs in a very long time. It takes 10 to 15 years to bring a drug to market successfully right, and so we see this operating in the blind problem quite a bit, and the time we believe is now to do what we're doing, and the reason for that.
Speaker 2:There's a number of factors coming together that we think make this right. Specifically, you have this problem the systemic drug development cost getting worse and worse I mean it used to be like a billion dollars per drug, now it's approaching 2 billion Continues to be at a very high failure rate in clinical development. Those things are not getting better and the low-hanging fruit are gone right. So, drug development, you're left with tough candidates to find. Uh, that's one.
Speaker 2:You have, uh, things like adaptive trial designs coming into play and which potentially added some level of flexibility uh, you have. Most importantly, you have the maturation of technology and data right. You have, uh sets, all these electronic health records and various other digital data sources coming into play right On the real world side. And you have technology, you know, deep learning and machine learning and gen AI all powerful tools to leverage all of this information in a better way. And I think, lastly, you have this AI being used quite a bit on the preclinical side for drug discovery, which has led to almost like a bottleneck of sorts at the clinical stage. So you got this big body of preclinical drugs discovered which are waiting to clinic and get approved. So all of these reasons, I think, make this the time in our mind to actually create a disruptive innovation that takes us to the next level.
Speaker 1:Well, I hope that this is the right time, because we definitely need the sort of acceleration that you two could promise. I would be curious how there's a lot of hype around AI, and a lot of hype around AI in the clinical area too. When you are talking to customers, what kind of attitude are you being met with? Kind of attitude are you being met with? Is it? Oh, let's do this, we believe in this, or is it?
Speaker 2:oh, this is going to blow over in a couple of months. That's again a fantastic question. I think that this is sort of the hype cycle going on in a way where the it sort of hit a crescendo, you know, probably at the end of 2022, into 2023, right, once, once open, ai came into the picture, uh, with chat, gpt and everything. So now so I mean now you got the issue of sort of separating the wheat from the shaft, if you will, right, so you have to, because of people are, you know, they like to slap on the AI tag, whether it's directly related or not, and then for the average observer, it's, you know, it's being more discerning to figure out. You know that this promising technology is indeed being leveraged in a true way, and that's what this really is that this promising technology is indeed being leveraged in a true way, and that's what this really is. So my core point on this is there is big value in using AI in different ways in clinical trials. Right, it's against, like I said, separating out what is real from a discerning perspective. That is one of the things that comes into play.
Speaker 2:But I certainly think the potential for AI in the medical space, and particularly in drug development is enormous and it's not quite been tapped to the extent it could be.
Speaker 2:It's coming into play. Some of the reasons for that are, well, we've started seeing some of it in preclinical, but there's a lot more conservatism and regulatory slowness and traditional ways of working that are common in the clinical development space in particular. So I think that has left some untapped potential at this point. But I mean, the opportunity is just really out there because, I mean, at the end of the day, you're saving lives and big dollars are being spent in the space, but not, you know, with low success rates, right? So you have that huge gap to be filled. You know, getting that 90 percent to failure rate to something closer to 70 percent, or even an ideal world like 50 percent, right? So, yeah, so we think opportunity is enormous. It's about separating the, you know, obviously picking the true contenders from the pretenders, and you know, I mean AI. Native companies are a good way to start, right, I mean to truly tap into the promise of AI. I think that's a good way to look at it, right? Oh?
Speaker 1:absolutely Well, birat, I would be curious how did you get into the clinical trial space to begin with? What was your journey?
Speaker 2:It's really you know, as I said, you know, with the personal anecdote that I shared, that was one of the motivators that, seeing that you know these trials being so, uh, ineffective and could lead you down the wrong path.
Speaker 2:You know, in many cases, uh, for a long time, you know without any, without any sign of discernible success. Uh, on the professional side, I think you know working, working with pharmas and working with my co-founder as well. He's a longtime clinical trial leader Merck and Stryker and so on. He has been living this for the last 30 years. He and I have been brainstorming, we're brainstorming around ways to innovate in this space, and we felt, as I said, with the confluence of a few factors coming together, like the technology maturation with AI, machine learning and deep learning and Gen AI, and the problem getting worse and this backlog with preclinical development that is getting bottlenecked here, we believe the time is right to get into this space. So that's why I got into it. I had been working in this already, but getting into this startup that I'm talking about and getting this going in this fashion, that makes sense.
Speaker 1:Well, birat, we are going to start rounding off, and I always ask our guests the same question towards the end, and that is if we gave you the transformation trials magic wand that can change one thing in our industry, what would you wish to change?
Speaker 2:Very good one. You know, I want to have that magic wand. So what I want to change is I would, I would, um, I would want to change the improve, the openness, uh, and willingness to change, right. So a lot of that. You know, the traditional ways of working and sort of a conservative approach to it. I mean, some of it makes sense, don't get me wrong, but, but there's also a mindset that needs to be open to innovation, which I think can improve in this space, which I think can improve in this space, right, I think, because the tools are here, the data is here and the potential is enormous. I mean, there's other applications of AI. I mean we're doing these patient-level clinical event predictions, but you can apply it in other ways too. I mean you can have it write the protocol. I mean you can have it create summ protocol.
Speaker 2:I mean you can have it be summaries of the submission for FDA, et cetera. There's various ways you can leverage this in this space, right. But, like I said, I think we need innovation champions, yes, and we need a broader wave of that. It cannot just be one person out of 25 people, right, or 500 people. It has to be more of a mindset change that embraces this, owns this and wins together with the technology, right.
Speaker 1:No, absolutely. That's an excellent wish. That's an excellent wish. If people want to learn more about Kool-Aid AI or have any follow-up questions to anything you've said, where can they find you?
Speaker 2:Absolutely, they can contact me on LinkedIn, which is I can spell my name out it's V-I-N-O-D-H, space, B-A-L-A-R-A-M-A-N. They can also look at our website colateai H-C-T-P-S, colateai K-O-L-A-T-Eai, and there's links there to contact us from there.
Speaker 1:Yeah, so both of those will work. Awesome. Well, thank you so much for coming on the show today. This was a very enlightening conversation. I do hope that we can start accelerating our trials with the insights that your tool helps create.
Speaker 2:Thank you so much. I really appreciate the opportunity and the time and welcome you to contact me if you're interested.
Speaker 1:You're listening to Transformation in Trials. If you have a suggestion for a guest for our show, reach out to Sam Parnell or Ivana Rosendahl on LinkedIn. You can find more episodes on Apple Podcasts, spotify, google Podcasts or in any other player. Remember to subscribe and get the episodes hot off the editor.
.png)