AI + Real-World Data: What the Next Five Years Look Like for Pharma

Hi, I’m Daniel Chancellor and welcome to Real-World Data After Dark. This is a series of conversations on real-world data and its importance to the pharmaceutical industry, all mixed with what we hope you can see as late night talk show vibes. I’m your host, Daniel Chancellor. Once again, I’m joined by Kris Kaneta, our chief product and innovation officer at Norstella. Welcome, Kris.

Thanks, Dan.

And Ted Search, Chief Real-World Data and Analytics Officer at Norstella.

Thank you, Dan, for being here.

Episodes one and two, we busted some myths about AI and real world data, but then we also turned to think about what value it’s delivering today to our clients and our industry. To wrap up with this episode three, we’re going to be getting our crystal balls out. I hope you’ve been polishing them and studying. Ready

Yo

Go. I’m hoping some big predictions for both of you. No sitting on the fences here. So first up, I’m going to put this question to you both. Actually, I’ll go first to you, Ted. Sure. We launched NorstellaLinQ two years ago. What’s the next two years going to hold for us?

Tell you what, it’s going to be an exciting two years, that’s for sure. I think the depth of which we’ll get into clinical data, EMR data, unstructured data, we’ll just get deeper. The first two years was exciting because it was all about surfacing data that really never was seen before. At least never seen in an impactful way that it could be impacted and acted on within a matter of days. And our whole work of creating and working with AI even two years ago in large language models to surface out information from billions of clinical notes has created data sets that now really go all the way along a patient journey. So historically you had diagnosis from your claims. You would get the different results coming from lab data. As you get in the EMR, you really understand the patient, right? The unstructured data brings in information like biomarker, tumor staging, physician sentiment.

So you can really understand that patient journey. And that was really the first two years. Now we’re just further building out on that. What these AI technologies are allowing us to do now that we’ve established a very clinical, relevant, linked, real world data set that understands the full patient journey, just what we’ll be able to do for clients with predictive analytics as a result, getting even deeper in the analysis on what that unstructured data brings to the patient journey and ultimately then to use cases like getting the patient on the right therapy at the right time, having a patient get access to an option of a clinical trial at the right time. I think the next two years is just going to be exciting with what these technologies will drive and will just further arm our partners with the way to really impact the patient even more.

Okay. How do you feel about that?

Two years is a long time, right? Two years is a long time. I agree with what Ted said, but I think it was a long time ago Bill Gates popularized this notion. I’m not sure if he said it first, but essentially is that we tend to overestimate the events to come over the next two years and underestimate the events that come over the next 10. I think with the advent and complete transformation of AI, that two in 10 is really more like six and two. So that’s my way of copying out saying two years isn’t a possibly difficult thing to predict. But I’ll try and say, I think first and foremost, in the context of real world data, your traditional structured data sets like claims really just become table stakes for a variety of reasons. One, they’re structured, they’re predictable, but there’s also lag. There’s lag and there’s a lacking of timely clinical context.

So what we’re going to see over the next several months, and we’re already seeing, is a prioritization for timely, novel, real world data insights, whether that’s EMR, prescribing behavior, unstructured data that really help paint a complete picture around a patient journey or a patient population. And I think the second thing to build on what Ted said is bringing all that together, having that timely contextual context is going to create models that are just going to be insanely incredible in terms of their predictive capabilities, in terms of driving next best action. So as we think about commercial use cases, for example, from a pre-call planning to triggering approved emails, to driving marketing analytics and marketing programming, all of these things are going to be completely turned upside down and transformed in a really positive way. And by that I mean, I hope less noise for the physicians and the patients on the receiving end, far more targeted, far more relevant, far more contextually appropriate.

And that can only mean better things for the patients at the end of the

Day. Yeah. So latency, you started to talk about there, how it’s going, really reducing the latency. I’ve heard you speak as well about predicting things before they happen. Is that something we are going to be expecting to be bringing to our customers in the next two years?

Absolutely. So Ted’s team, for example, is working on predicting therapy switching behaviors. The notion that I can, six, eight weeks before the actual event, based on disease progression, based on what’s happening in the unstructured data and everything else, be able to predict the likelihood of a patient switching therapies, for example, and understanding more importantly the why behind that. That’s right. That’s incredibly powerful in terms of being able to think about targeting, in terms of thinking about how I’m bringing new therapies to market. And absolutely. So predicting things such as therapy switching, predicting things such as trial duration, predicting things such as responsiveness to marketing programming, all of these things is really on the forefront, on the cusp of what’s to come. And I think from a Norstella standpoint, we’re really not that far off from a lot of these things becoming reality.

And Kris, you made a great point there, and I’ll tie in what you even said earlier of where I think we’re going and where the industry’s going. We’re going from having the largest counts of data to go into the most depth of data. And that’s really important for everything Kris just talked about. What I mean by that is historically, you had to get big counts of data. You had to get Rx data, diagnostic data, so you could understand a large patient population and try to drive your analytics off of that. Everybody wanted to move downstream in the patient journey and understand the EMR data, but technologies a few years ago weren’t there to surface the data at the right point of time. As Kris talked about lag, being able to even bring this data into a point that it could be used for commercial use cases, impact the patient at the time of care.

But those technologies are changing. Also, the access to the EMR data as a result is driving larger patient numbers. So when you look at the depth of the data versus just count, I believe we’re going to really move to really a full patient journey. Because when you have that, when you have the depth of the data versus just large diagnostic counts, you can really understand that, apply these technologies with the right rigor, right clinical inputs into them, and leverage them in a way that then when understanding that complete patient journey, you can then predict different results that you went after it, actions that would take place and looking at the historical patient journey, and then applying that to real world events that may happen at a very high prediction rate.

So we’re speaking on a two-year timeframe, and apologies. Yeah, I get your analogy completely. And you’re going to hate me now because you can answer like 10 years. But don’t be worried. We all make predictions and comes back and haunts us. So we’ll just scrub it from the internet or add it to our blooper reel or something like that. But can you make any. Deep inside you, do you feel any really big predictions where you really aspire to be heading on a longer timeframe than two years for us to really be transforming the industry for our clients?

There’s so many areas that I can see transforming, whether or not it’s two years, five years. My gut is going to be sooner than later. But nonetheless, I think about digital twins and simulation, whether that’s in a clinical trial or whether that’s thinking about the probability of covered lives being exceeding a certain threshold from a payer access perspective. Being able to simulate that based on what’s in the market, based on what’s happening in the real world data. All these things are, well, possible today and being done at a predictive modeling perspective, but now being able to simulate it against a variety of what ifs and potential scenarios I think is very real and very probable. I think the other thing is really ultimately how far do we go down the adoption of agentic AI? And for me, that really comes down to so much of the work that is done today, whether you’re thinking about real world data or clinical trial data or market access data.

It is so much of that is heavy manipulation, rows and columns and getting to an insight. And don’t get me wrong, we still need really smart people to understand and validate, but agentically getting to the insight and really leaning on your subject matter experts to make a determination of validity and feasibility based on that output and then triggering a whole slew of activities to follow. And that’s true for all business and enterprise, not just in life sciences. But I definitely, given the broad variety of stakeholders pipeline to patient when it comes to our industry, I think there’s massive potential there.

Okay.

Yeah. Kris said it really well. I would just add into it. My bold prediction is that real world data, the evidence, the analytics, the predictions that come off of it are going to be a massively pivotable point to say the FDA to do approvals for clinical trials, for new indications on therapies already out in market right now. So right now it can be used as a proxy, it can use to be a supporting data. But as we’re leveraging these technologies to really understand that patient journey at a very detailed and in-depth way, and we can create models with very high rigor, with very high propensity to understand results and prove that, I think these data sets and the evidence that will be generated because of these technologies will actually be one of the main points that are being used by the FDA for clinical trial approvals, for new indications when it’s in market as well, and even HEOR evidence studies as well.

It feels to me though, I mean, in all of my working life, we’ve had the internet, but there was a pre-internet era and working in that and how that might have felt like to try and imagine what it’s like today, or even the pre-desktop computer era, we’re certainly at a cusp, the tip of a tipping point really where how we design drugs today, how we generate evidence is really just going to be transformed completely. It’s like you guys used to do that. Right. As clients or as we recommend clients future-proof their AI and robot data strategies as they navigate this tipping point, what advice would you give to in terms of future-proofing?

So first of all, shameless plug, find the right partners. Find the right partners that really understand how this data comes together and gets linked. And to Ted’s point, it’s no longer about I need to acquire and consume and ingest the widest possible corpus of real-world data. It actually now comes down to what are my priorities and how do I get really deep? And so working with the right partners that have that depth and that can build something that’s going to scale for what are your priorities, I think is number one. And number two is let’s not fall in love with the promise of the great frontier AI models that are out there. Don’t get me wrong, these frontier models are incredible. They’re powerful, they’re sophisticated, they have broad analytical capabilities, but without the right context layer to feed these models or to create your own models, you are no better than you would be with a thousand interns.That’s literally how I try to explain the advent of what’s happening agentically.

Yes, the output potential is massive, but it’s like a thousand interns. And if I send a thousand interns out to go get coffee, but I don’t say I want a triple shot here and a double pump here and a latte here, I’m going to come back with a thousand black coffees and probably 900 people are not going to be happy with their coffee. So we’ve got to build the right context layer that is driving to the right outcome, which means really understanding what does good look like from an outcome perspective, which goes back to the first point of making sure you find the right partner that’s going to get you the right depth to inform and feed that context layer.

I agree completely. I’ll build on that. Find the right partner with the right link data set that can give you a full patient journey because that’s going to be so important with what these technologies can do. And then understand that every AI model, every prompt, every chat tool you’re using aren’t the same. So what I mean by that is it looks good, but there’s no real magic in typing in what you want a patient cohort to be and something coming out without proof or giving you the sequel that you need to then prompt it and get a result back. How do you know that result is right? Rather, make sure that you’re working with a partner that understands when you write this prompt in the backend, the prompt needs to understand, are you looking at claims data? Do you need to add lab data? Do you need to add EMR data?

Is unstructured data important? Make sure the model can do that and then source for you where it’s getting it from. Make sure it could provide you back with clinical documentation of why they choose these individual components that gets built into that cohort. Yes, it can give you the SQL code, which is great to do it. Yes, it can give you the output, but make sure you have the background both clinically and also to make sure that it’s pulling the right codes, whether it be IC-10 codes, lunch codes, so that when you get your output, you can reference back why it gave me this output. Because without that, you’re just trusting a model to generate some kind of cohort. And yeah, you get it instantly, but how good is that? How can we use that to generate evidence when we don’t know the sourcing behind it?

So make sure when you’re using these technologies that you’re getting that sourcing behind it. That’s the biggest thing I think I can tell

Partners. And know when good enough is to make a billion dollar decision, right? Yes. I mean, that to me is the most important thing. It can sound very credible, but if there’s a billion dollar decision of the other end of that, that’s going to trigger a variety of activities and investments, man, I better be really certain that the right information fed into that model and that that output understands the context with which I am operating.

That’s right. It’s going to allow whether it be the experts in your company, the clinicians in your company, or the clinicians and experts from your partners, just to be more involved in making sure that evidence is very high powered to generate the evidence you want. They’re just going to be able to do more. They’re going to understand the data sets, the therapy areas very specifically in order to understand when they use these tools that will move quick, quicker than we ever could before. They’ll ask the right 10, 20 questions to make sure now, not only are you getting a cohort out that you can utilize quicker, but you’re getting one that’s more accurate, more impactful for what you’re trying to do for your business.

Okay. What I’m hearing from you both for us to still it into two words, partnership and trust. So yeah, that’s certainly how we’d also engage. Well said. Great. Kris, Ted, you’ve been fantastic guests. It’s been a pleasure talking to you.

You’ve been a great host. You’ve been great. It’s always a pleasure.

Well, we’ll see about that. And thank you to the audience for tuning in. We’ve had our fun here. Hopefully that’s come across. We’ve very lighthearted take on clearly what is a very serious business in terms of decisions that are supporting the development of new medicines and getting them to patients as quickly as possible. If you have any feedback for us on things you’d like us to talk about, please let us know. We’ve done season one in London. We’ve done season two in New York. What’s next? Hopefully Paris, Tokyo. That’s going to be Boston, isn’t it? Boston’s all right. Yeah. Well, we’ll see you there. Hopefully you can join us. Thank you.