Finding the Right Patients Faster: AI and Real-World Data in Practice

Hi, I’m Daniel Chancellor and welcome to Real-World Data After Dark. This is our playful take on a late night talk show, and I’m your host, Daniel Chancellor. I’m back with my two very special guests. We have Ted Search, Chief Real-World Data and Analytics Officer at Norstella, and Kris Kaneta, Chief Product and Innovation Officer at Norstella. From both of these gentlemen, you’re going to be getting the inside track on how we see AI really adding a whole new dimension of value that we can provide to the pharmaceutical industry through our real world data. We hope you enjoyed our first episode where we tried to keep our feet on the ground and bust a few myths, but maybe we got a little bit carried away with ourselves. We’re going to be continuing on without excitement.

And to get things going, I’m going to suggest we start with a bit of a game. Red light, green light. You guys heard of that before?

Sure. Let’s do it.

Sounds great. Not the squid game version. So I mean, I don’t think there’s enough space in this industry to do that. Rather, think of it more of as a set of traffic lights. So what do you want to stop doing or what do you want our clients and industry to stop doing? And what do you want to know? Full steam ahead, what’s our big green light? So red light, green lights, spending three weeks building a patient cohorts. Who wants to go?

I’d say red light. I can jump on that one. Absolute red light. This is why I say that. What AI should do is be able to surface insights, cohort generation on very well-designed and linked data very quickly. What took three weeks before should literally be done in maybe an hour, maybe minutes, depending on how complex the cohort that you had. Historically, why it took that long is if you look at a partner in pharma that we would work with, they would have to go and decide and source which types of data they want. Do they want claims, lab, EMR? They would then have to take that data, get their tech team involved to bring it in to an infrastructure that could set up. Their clinical team, their data science team would have to work in generating a cohort. They would work with partners like Norstella to help generate that.

Literally now AI with the right prompts, with the right support from clinical services, you could generate a cohort on the fly. There’s tools that we’re generating specifically that allows them to do that. And the point of being able to do that then is you can get into the evidence generation much quicker. So absolute red light.

All right. Anything to come in?

I would simply build on this notion that, yes, with the right prompt, I can get to an understanding of what a cohort looks like in minutes. But equally as important, I can understand analytically what are the specific aspects defining that cohort coming from that prompt. So it’s not just about looking at millions or billions of rows and columns. It’s actually seeing in an analytical perspective, this is what you’ve created. Does this meet the requirements of your evidence generation study, your clinical trial, what have you?

And if I could add just one more thing to that. Historically, it’s not even the time right now that it completely takes. It’s the back and forth you have to go through. Understanding, because really generating a cohort is a bunch of inclusion exclusion criteria. If you have AI in a tool and the right individuals that understand the data, understand the therapeutic area that you’re working on, you can literally collaborate in live time letting the AI agent with the right prompts generate what the next question is. What is the answer? What do you have to work through right away in a workflow? You can ask it 10 more questions than you could before because of how fast this technology works. That’s the exciting piece behind it. Not

Just building a cohort then, but building a better cohort. It’s rating and really getting to a real true representation of what your patient group is. Absolutely. Okay, let’s move on. Red light, green lights, identifying, recruiting patients into clinical trials, what do you think?

Yeah, definitely green light. When you think about the importance of the dimension of time as it relates to clinical trials, you just don’t have that much time to be taking that a lot of guesswork and over analysis. So the idea that through AI, looking at the structured data or unstructured data with LLMs, understanding what’s happening from recent lab tests, being able to bridge all of those aspects together to say, “Hey, here’s a patient that is a high likely candidate for your clinical trial.” And being able to see that within 24 hours, 48 hours, and actually reaching out to that healthcare provider to have a conversation about your trial. Yeah, absolutely green light. We don’t have the time to waste when it comes to these vulnerable patient populations.

Yeah. And I’d say it’s a wonderful green light from a patient’s perspective because you can leverage the AI now to really understand from clinical notes, does this patient have a specific biomarker that’s right for this therapy that’s in a clinical trial? That’s one example. And then we can utilize that data to work with their prescriber to give them another care option, let them know where the clinical trial is taking place at the point of care which you couldn’t before. So it’s wonderful from a patient perspective.

And that clinical trial could well be standard of care that patients are eligible for, for a very finite period of time. And if we don’t act quickly, then that’s a window of opportunity loss for everyone.

Absolutely, Dan. All

Right. Red light, green lights. I’ll put this one to you, Kris. Safeguards in AI models.

100% has to be a green light. And I think this is true regardless of industry, but we’ll keep it focused on life sciences and pharma for a moment. I mean, we talked in the previous episode how AI can sound very compelling, very credible, but I’m not about to, or at least you shouldn’t. I advised against automating actions and triggers from AI that haven’t been appropriately vetted either at the time the insight is surfaced or in terms of ensuring that a workflow that you’ve automated, whether that’s, I don’t know, an email trigger or an HCP recruitment outreach, whatever it might be, you’ve got to make sure that that workflow has the right clinical context being applied by the right expert in that loop. So we’ve absolutely got to guard against that. And also just from an overall security and privacy, I mean, at the end of the day, we have to never forget that these are real patients whose lives we are trying to improve and benefit, but we also can’t take for granted that there are a lot of bad actors out there.

So 100%.

Yeah, I completely am aligned with Kris, and I’ll even take it from this perspective on safeguards. Bad in is bad out, right? So meaning the data itself. As a company, you really need to understand what data are you working with. The models that you’re applying, you have to understand the precision, the accuracy of the models on top of, I’d say a linked longitudinal, very clinically relevant data set. Because at the end of the day, if you can understand that there’s safeguards in place to assure that, yes, this AI and the models that are created can provide analytics cohorts out quicker that then allows specialists to go in and really generate high value evidence. If you don’t have those safeguards in our world of healthcare, you can never validate what the actual result is.

Okay. I mean, it’s similarly related, but AI black box is red light, green light. So I guess having results without necessarily knowing where they come from. Yeah. I

Mean, I think this is really a no-brainer when you zoom out here. When we think about the importance and the magnitude of the decisions being made in pharma, whether it’s a $10 billion acquisition to hundreds of millions on a clinical trial, to launching a new drug in the market, these are not decisions that our clients take lightly, nor can they go off of a black box of analysis. So as you look at the models that we are introducing into the market, one of the things that has been so crucially important for us at Norstella is providing traceability and site ability as to where did this information come from and how did it inform the output of that model? And at Norstella, we’re obviously super fortunate that over the last couple of decades, we have been a major aggregator and structurer, is that a word, structure of this data?

And really getting it in a place where it is AI ready to be consumed through APIs and MCPs. But at the end of the day, that model has to be able to say, “I got to this output because of these following 20, 30, 50, 100 inputs, and here’s where they came from.” And I think that’s going to be super important because it’s not only citability, but it’s also recency. I don’t want to be triggering an output off of something that came out from a finding 10 years ago that has nothing to do with what’s actually happening. So 100% a green light.

Yeah, I agree completely with Kris and he said it so well, I won’t repeat everything he said. I’ll build on it. I think a big area that we’re getting into and also our partners and life sciences are getting into are predictive analytics. There’s no way that you can build predictive analytics if you don’t have a solid base solid evidence, everything that Kris just said. So the importance of being able to get into that predictive analysis, predictive analytics, which I think are going to be immensely valuable in the future with these technologies, all start with exactly the foundation that Kris talked

About. And it’s reputational too. It’s not just our clients that need that assurance, but it’s the regulators, it’s the physicians, it’s the other partners that they’re engaging with. It’s their investors and sponsors and whoever else. There’s so many stakeholders that are going to be put into motion based on these very critical decisions. This is not something we can assume or be comfortable with 80% right. I mean, that last 20% is a long 20%. You’ve got to be confident.

And one last piece on that, and the best part about the technologies now and even having access to this data in a quick way to analyze it is what you can understand on the backend, like the reporting on it. So if we don’t have the strong foundation, we’re not benefiting any of the patients. The companies that we work with can measure that very quickly. Are the analytics, is the patient data that we’re giving them to leverage with physicians to engage with them? Is it delivering results? And in a marketing analytics standpoint, are they getting the return on the investment they’re putting in because it’s driving a positive behavior adoption? So the ability then to understand and analyze a closed loop of not only data coming out, analytics going to an opportunity, but the reports that come back in really can assure that all this foundation is very solid.

We’ll stop our game of red light green like that. But let’s just imagine that we’re in a world where we’ve hit stop on everything that we wanted to stop and we’re full steam ahead with all of these wonderful initiatives that we’re talking about. What difference is this going to make to our clients, to our industry? Ted, do you want to come in first on that one?

So say it one more time, Dan, what you’re looking for

There. So if we can really stop some of these behaviors and stop some of these things which are adding time and cost and complexity into how our clients are bringing medicines through to patients and proceed with all of the best practices that we’re talking about, really open up our models and our explainability, regulatory grade data, all of those kind of things. What does it mean to our clients? What can they do that they haven’t been able to? What does success look like?

To our clients, we just make them more impactful as a result. And ultimately what we are all in this for is to benefit the patient, whether it’s to benefit patients by giving them access to another option in clinical trials, whether it’s benefiting the patient to getting them on the right therapy at the right time. So what these technologies are doing is allowing us to take this massive patient data that’s available to all of us right now. And what historically would’ve taken months upon months to understand which data sources do we bring in? How do we set the tech in order to understand this data, then get into analytics? You’re months out before you’re actually running programs that ultimately benefit the patient. These technologies are short in that gap. We talked about earlier having a very high integrity data set, running models, analytics on top of a very high integrity data set to get results out quicker.

What this just does for our partners is allow them to do really the objectives they’re looking to achieve much quicker with a very valuable data set and then be able to analyze what results they’re getting and pivot in real time to be impactful ultimately to the patient at the end of the day.

Yeah, I characterize it in two areas. One, building on what Ted said is we’re going to see this massive compression, and we already are, massive compression in terms of time to insight and time to action. So our discussion on patient recruitment into clinical trials is a perfect example where time is everything. So compression of time and action is part of it. But I think ultimately the second thing here that we’re going to see is, and you’ve been championing this for a long time, Dan, is we look at the rate of return in pharma pipelines. And what we’re going to realize is that the insights that are going to drive the most impactful outcomes are further upstream in the decision-making process. Understanding the benefit and the return on investment for an acquisition of a small molecule or understanding the cost and the benefit analysis of clinical trial in certain geographies or in certain inclusion exclusion criteria.

All of these decisions can be informed so much earlier in the journey. And this is what I think excites me most. Less clinical trial rescue activities and more thinking really purposefully about what’s my pool of investigators? Where am I going to have the highest likelihood of recruiting the right patients into these sites? And how do I get to a more accurate trial duration and successful enrollment? I think all of these things are going to keep moving and shifting to the left of the decision-making process.

You’re right. It speeds it all up. To Kris’s point, a lot of times you have to make a call or decision and then wait another month to see how it carries out. What this allows us to do is ask 10, 20 questions, analyze them very quickly and then be impactful. So I agree with you completely, Kris. Just the speed at which we’ll be able to move now is really exciting.

And we take all of this downstream, and I don’t think it’s going to be hyperbolic to say that ultimately better medicines treat patients more quickly at lower costs, which benefits access for everyone. So that’s the ultimate goal. That’s 100% the goal. If that’s not what we’re aiming at, then we’re falling short.

Yes, absolutely.

Absolutely. Well, fantastic. That was a leading question into some of what we’re going to talk through in episode three, which will be our final episode of season two of Real-World Data After Dark. But thank you for joining us. Thank you to Ted and Kris for today. Thank you. We look forward to seeing you next time.