What If The Real Blocker Is Coordination Not Data w/ Anastasia Christianson

AI is everywhere in pharma right now, yet drug development still feels slow, fragmented, and full of avoidable rework. We sit down with Anastasia Christianson, former SVP and Global Head of AI Data and Analytics at Pfizer and now Managing Principal at EPAM Life Science Consulting, to get specific about what actually blocks progress in clinical trials and what AI can realistically change.

We dig into the uncomfortable pattern behind many “successful” AI deployments: a model fixes one bottleneck, then the system simply jams up somewhere else. Anastasia makes the case for end to end thinking across trial design, feasibility, patient recruitment, engagement, site operations, and data flow. We talk about why niche AI products are easier to build, why coordination is harder than modelling, and how digital health’s app explosion offers a warning for today’s AI tooling. Along the way, we unpack digitalisation, data standards, FAIR data, and why “perfect data” is not the same as “useful data”.

The most practical segment centres on patients: how AI can help build cohorts, find eligible participants, answer questions, reduce fear, improve convenience, and prevent dropouts that cost time and money. We also turn FDA denials into a quick quiz to highlight a key truth of life sciences innovation: approvals can fail due to control arm choices, surprising trial requirements, or manufacturing site inspection findings even when the science looks promising.

If you care about AI in clinical trials, R&D productivity, and how to turn data into faster access to medicines, you’ll get both a reality check and a playbook. Subscribe for more, share this with a colleague in clinical operations or data science, and leave a review telling us: where do you see the biggest bottleneck right now?

Visit Labtolives.com

Do you want to try the quiz from this episode? Find the link on our LinkedIn profile: https://www.linkedin.com/company/labtolives/ 

Would you like to join the show as a guest or collaborator? Find out how on our website: https://www.labtolives.com/ 

Transcript

SPEAKER_01 0:00

Hello everybody, welcome to another episode of Lab2Lives. Today in the studio with me I have co-host Alex and I have Anastasia Christensen. Welcome to the show. Anastasia, you got introduced as a potential guest to this show because you have some very relevant experience within AI, data, and analytics. And I'm very curious to unfold some of your experience later in this show. But perhaps to start us off, could you tell us a little bit more about yourself?

SPEAKER_02 0:31

Absolutely. First of all, thank you so much for inviting me to join you here. I've been listening to your podcast and very intrigued

Welcome And Guest Background

SPEAKER_02 0:38

by the format and love the topics that you discuss. So I'm currently serving as managing principal at EPAM Life Science Consulting and also executive in residence at Columbia Technology Ventures. This is after 30 years in the pharmaceutical industry, four major pharma companies. Most recently, I was senior vice president and global head of AI data and analytics at Pfizer. Prior to Pfizer, I held executive roles at Johnson ⁇ Johnson, Bristol My Scrib, and spent 20 years in drug discovery and development at AstraZeneca, driving data excellence, digital transformation, innovation across RD and for a significant period of time in clinical development.

SPEAKER_01 1:17

That sounds awesome. To be honest, Anastasia, I would like to be you when I grow up. That's been quite the career you've had.

SPEAKER_02 1:24

It's been a lot of fun, yes. Thank you.

SPEAKER_01 1:26

Maybe like from a very low practical standpoint.

SPEAKER_02 1:29

Where are you right now, physically? I'm physically standing in a room in our home in Media, Pennsylvania. So I'm in a room on the side of the house, hoping the dog won't bark while we're having this conversation.

SPEAKER_00 1:41

Well, Alex, I will ask you the same. Where are you right now? Physically, I am in my workspace here in Copenhagen that has a very nice, well-equipped podcast studio, which I like to hire out for this so my voice is nice and crisp rather than sounding terrible and tanning through my actual uh headphones.

SPEAKER_01 1:59

Awesome. Well, I am in an office with kind of subpar lighting. For those of you who will be watching this on YouTube, you can kinda guess that something is not quite right. It is more yellow than usual. Anastasia, you mentioned that you have been working with drug discovery and the development of new medicines. On this show, we kind of like to place our guests on these five value streams that at least I postulate exist in Pharma. There's the innovation stream, the manufacturing stream, the compliance stream, the commercial stream, and more the corporate stream. That's all about how do we build a company that can fulfill the drug promise. Where would you place yourself on one of these value streams?

SPEAKER_02 2:42

Definitely the innovation, RD. So I've been in RD through and through, although I have also supported manufacturing and commercial for a short period of time. But my heart is definitely in the innovation and RD.

SPEAKER_01 2:55

Yeah, well, I completely relate to your heart being in that space. It's always funny with you can say further downstream that the commercial organization and the pharmaceutical company are always proud of how much they drive the sales

The R&D Productivity Problem

SPEAKER_01 3:10

and create the revenue for the company. I always get a little bit annoyed by this because there would be no revenue whether no medicine in the first place. Okay, well, I'm going to start us off with the first round of this show. I would be curious both from you, Anastasia, and you, Alex, uh, what is a problem in life sciences that is occupying uh your mind right now? Well, I can start.

SPEAKER_02 3:32

I think RD productivity, how to increase productivity, shorten the time frame. Maybe you just put it as how to accelerate RD and especially clinical development. That's where I'm focusing most of my time at EPAM now, is how to use AI effectively to solve a bottleneck, not just use AI because it's there, but rather applied to some of the challenges that are in clinical development. We have digitized a lot, we, as in the industry, has digitized a lot of pharmaceutical RD. In fact, every step of the drug discovery and development process is digitized, which means we automatically pull data from instruments or collect data from patients and make it available to the people who will analyze it to ensure the drug is safe and effective. And despite using technology, the acceleration has been minimal. There is some acceleration, but not enough yet. So, Alex, I don't know whether you are struggling with the same or looking at a different problem.

SPEAKER_00 4:33

Yeah, well, I mean I guess there's there's there's an ongoing question in general around how to get the promise from AI embedded in the industry. Right, and I I mean my exposure is more on the on the on the product side uh in terms of diagnostics and things like that. But it's very interesting. I was at a conference earlier last year that was talking about there are quite a few AI products on the market on the diagnostic side, but they're very much tend to be within one niche category, which is the the image analysis, right? You take your X-ray or your CT scan or or whatever else, and you do some clever visual analysis and spot these deeper patterns that are maybe earlier earlier indications of disease. And I think that's from what I remember, 80 or 90% of the market right now. It's just an interesting question. Why is that? Why why are we what are the barriers to expanding more into other areas of healthcare? And and maybe that's where the two tie up, right? Which is on the RD side, there's this question of well, what's preventing us from really utilizing this stuff in RD, and also what are the barriers to utilizing it on market as well. So I'm interested in both sides of that equation.

SPEAKER_02 5:47

Yeah, so definitely on the image analysis and diagnostics, there there are new better ways to diagnose disease earlier. As you mentioned, imaging is a big part of it. And in the RD space, the in discovery, well, as I mentioned, technology and AI is being used in every step of the drug discovery and development process at this point, and for several years now. In the discovery space, we're seeing new drugs emerging that we think are going to be beneficial, are going to be changing the trajectory of disease. They still have to go through the clinical trials, which takes nine to twelve years. The earliest ones are in phase two clinical trials right now, so we don't yet know if there's going to be a higher success rate for drugs that were developed using AI in discovery. So part of the reason why I'm actually focusing in the clinical development space to say we are doing a lot of operational improvements using AI, including, you know, accelerating designing better trials and recruiting patients and so on. We still have a ways to go. My theory is that part of the challenge, or maybe the opportunity, I should say, is in going beyond solving one bottleneck at a time, but rather looking at the process more holistically and seeing if we can apply AI in a way to accelerate and resolve multiple bottlenecks in one go. And so looking at patient recruitment and engagement, for example, if we engage patients from the very start where they learn about a potential therapy that might be beneficial to them and keep them engaged and informed, use AI to identify the right patient, to engage them, to keep them engaged, to answer their questions, to help them through

Solving Bottlenecks Across The Trial

SPEAKER_02 7:29

the process of finding the right site or the best site for them, scheduling, talking to the healthcare provider, and accelerating all of that, keep them engaged throughout the process. That should reduce the number of patients that from the trial. So then you don't have to replace them. You don't have delays because you're replacing them. And if you've picked the right patients from the very beginning because you've engaged the right patients, then that should significantly contribute to some of the delays that we experience in clinical trials.

SPEAKER_01 8:00

I think you're hitting the nail on the head there with optimizing the end-to-end process and not just like the individual workflows. From my experience right now, where I'm trying to understand what potential AI actually holds, it does seem to be making the biggest difference like in the niche areas. We're seeing these amazing service providers that have developed models that work for one specific example because they have the data volume, they've built models that are really intelligent in understanding a specific area. And that's great. But I think we risk accelerating like one process, and then we just have a traffic jam with that process just like spewing out a higher volume output without the rest of the value chain being able to kind of take that on.

SPEAKER_02 8:45

Absolutely. Absolutely. Kick the can down the road, I guess the bottleneck becomes elsewhere. And if we look at all the bottlenecks or the potential bottlenecks and resolve them at the same time wherever possible. So, in a way, we're not failing at discovery or innovation, but at speed and coordination. And we are doing quite a bit in coordinating this. So I want to be positive about just how much is being done. In fact, similar to you, Alex, I was at a conference, a couple of them earlier in the year. And the talk was all about optimizing clinical trials and using AI to optimize different steps of the clinical trial process. So it's like, let's put some of these together.

SPEAKER_00 9:23

Yeah, well, and I am in the very comfortable position on this episode of being the person who knows the least about the subject because I'm not a data guy, I'm not an AI guy, I'm not particularly a clinical trial guy either. So I get to ask all of the dumb questions, which is much, much easier than answering the dumb questions. And I that strikes a dumb question for me then, which is okay, we're in this situation where there's a lot of niche applications coming through to support trials where a holistic approach would be more useful. What is it that we think is driving that? Is it just is it simply easier for an individual company, right, to address this niche aspect? Or is there something more fundamental about how we're approaching these systems that that needs to change? Do you think?

SPEAKER_02 10:08

Yeah, maybe I'll start because I teach courses at uh Penn, I teach one drug discovery and development course. There's no such thing as a dump question. They're all good questions that someone else has thought about and hasn't raised it. And you do ask a really good question. I think we're um we're sort of failing at speed. I think we're so busy trying to accelerate that we find one bottleneck and we just go, you know, heads down on that bottleneck to try to resolve it, which is an approach. It's not a bad approach. It's the taking a step back though and looking holistically. I my daughter is a musician scientist,

Why AI Stays Stuck In Niches

SPEAKER_02 10:45

and she she always would say you can't rush creativity and innovation. And my mother had a saying of, you know, those who rush trip. So if you rush, you trip, you pick yourself up again and you're actually no better off. And is another way of looking at it. I had another colleague who'd say, lazy people work twice as hard because you take a shortcut and then you have to do it again. So whichever one of these you know resonates with you. I think speed is of the essence for sure, because we're all focusing on getting the best medicine to the patients who are waiting. So absolutely. But coordination in execution and taking a step back real quick to look at the full process, anticipate while you're working on resolving one bottleneck, having some coordination with others who are going to look at what's the effect going to be of resolving this bottleneck further down, and can we start working on that additional bottleneck before we get there? I I think it's something that we are maybe not doing as well as we could, and clearly we we did it very well during the COVID period because we, multiple companies, delivered vaccine in record time under two years. So could we do that again? There's there were a lot of things that had to go in, other programs had to be stopped during that time and so on. But I think coordinating and working together across the process and looking at the entire process. And now that we have AI, we can actually simulate the operations, the process end to end and use some of the synthetic data or legacy data that we have, real world data that we have, to simulate the trial. Again, we do it in in bits and pieces and fits and spurts. Then we actually look at it from design all the way through to patient recruitment at different sites, to coordination of getting the drug there, to coordination of the data coming off of lab tests and so on, and look at all the potential bottlenecks and start resolving them one at a time, but together at the same time.

SPEAKER_01 12:40

I we'll have a go at this uh question too, Alex. Yes, no, please do. Personally, I'm a very proud second mover on technology. I am not uh the first mover. I prefer that other people make uh all the big mistakes and then I take whatever actually seems to be working and start using that. And I think many of the first movers in our industry for AI solutions are specialized because of this anecdote about a guy looking for his keys like on a parking lot uh under a lamp where someone asks him, like, well, did you drop them here? No, I dropped them over there where it's dark, but it's easier to look over here. I forget how that story goes exactly, but you you both seem to know what I mean. I I think we are specializing uh in some areas because we have the data and we have the expertise, and it's easier to solve one problem than it is to solve the end-to-end of the entire clinical uh process all in one go. So hopefully we can combine things uh over time.

SPEAKER_00 13:37

It reminds me actually of another situation a few years ago, even in uh healthcare, if you'll bear with me, which was maybe six, seven years ago, we're all getting very excited about digital health. Not this was pre-AI, just digital health in general, sort of through apps and all of this. And I think that ended up in kind of a similar situation, which is that there was a lot of people developing a lot of apps, both companion apps for drugs or tracking apps or all of this. And I feel like what eventually happened there, one of the reasons that petered out is that there were just simply too many niche apps, none of which were built on a common backbone, all of which needed to interface with an already complex healthcare

Digital Health Lessons And Tool Choice

SPEAKER_00 14:15

IT system. And so it feels like are we once again walking into underutilizing a technology due to a lack of building infrastructure behind it? What do you think?

SPEAKER_02 14:25

Yeah, I I think so. And I'm familiar with the digital health space. In fact, we are using digital health measures now better than we were at the time when we were measuring and uh building hundreds of them, right, everywhere. I think you'll learn from doing that, but it's a little bit of the I've got this technology, this shiny object, where can I apply it, right? As opposed to looking at what is the problem and which of the technology, which tool in your toolbox are you going to use? And it may not always be AI, but it could well be AI. And if I use another analogy of uh if you need to put a nail in the wall, you could use a hammer or you could use a sledgehammer if that's what you have. But if you use a sledgehammer, you run the risk of it's gonna go too deep and you won't be able to hang your picture, or you might break the wall and put a hole in the wall because you hit it too hard, right? So using the right tool for the right problem. And if you had a screw, you want to use a screwdriver. But if you use a hammer, a sledgehammer might be better, but you might also, again, you know, drop the wall or break the wall. So I think we sometimes take a sledgehammer approach to any problem because more is better, and more is not better because it's more expensive and it doesn't necessarily solve the problem. So I think that was a little bit of what we were doing with digital health. It was, hey, we're able to use these nanotechnology now. Let's apply it everywhere, let's develop new ways of measuring whatever you want to measure. And they're turning out to be useful, but later on, right? So at the time, there was a lot of innovation, exploration, and so on. And as you said, way too many tools for anybody to use. And we went through that also, or have been going through that with AI. So and then you leave the burden on the physician who is gonna not look at all of them and not be able to test all of them, and will go with whichever one he or she happened to come across if they use it at all. So I think there's a little bit of let's find the appropriate tool for the appropriate problem. But maybe that's a secondary person that does it. The innovation folks are just gonna innovate and keep developing new things, and then you need someone else who's gonna look at all those new approaches, new technologies, and think about the process along that you're trying to uh resolve and pick the most appropriate tool, and chances are there's more than one. So then you have the challenge of deciding which one.

SPEAKER_01 16:42

I I feel like there's an analogy here that health tech innovation requires an editor. Kind of like when you write a book and it's very long and you have some complicated, convoluted narrative with many characters. You kinda need an editor to sit down and say, Well, you know what? This character, he's not really doing anything out with this chapter, you can kill the whole thing, you don't need it, and then you end up with a hero narrative after uh good editing. So perhaps it's also a question of, well, how do we trim and edit all this innovation that comes out now that we can even get the innovation produced faster?

SPEAKER_02 17:17

That's a great idea, and I I love the analogy. I suppose in some way an editor has a job and it's a fun job and it's easier to edit than to write from scratch. I think in the scientific space, it's a lot more fun innovate innovating than then reviewing and cutting things out. So that may be part of the challenge, but we also I don't think we have that type of role of editors. So what we have is the people who run the processes now and are, you know, might be looking at new technology, and then there's the challenge of, well, I have to change the process. I'm not sure I want to change the process, or what's gonna be the problem with changing the process. And so we we take that tool and try to squeeze it into the existing process, and we're not accelerating enough or not excelling as much as we would if we actually looked at what is the process and develop the tool for the process, or rather change the process, right? So we have a new tool, let's change the process because that's gonna accelerate how we use it. So we do need editors, you're absolutely right. We need to think about what we call them. Maybe we're solving a big problem here.

SPEAKER_00 18:17

Yeah, well, I mean, also what it puts in my head maybe is is communication. I mean, my version of this is slightly different in that I think nominally there's editors built into the system, which is us as as healthcare innovators, right? By which I mean any any model of innovation. It's very often emphasized that the first thing you do is understand the need. What's the the need out there, what's pulling in your technology and your solution rather than you pushing it out. In innovation, that's our first job and continual job to make sure that that alignment is there. So maybe what's happening here is some of that's breaking down and and I'm sure there's people in there trying to do that role, but maybe there's a communication element here. It's either the messaging's being misunderstood or not fully explored, or those knees aren't being captured effectively somehow.

SPEAKER_02 19:04

Or maybe oversimplification of the problem because I know the technology, I understand enough about the process. But the reality of it, even I was talking to a physician friend over the weekend, and I don't know, I don't remember why we got onto the topic of AI and how AI is helping physicians, but not really, because it's creating more work for them. And I never imagined that it's actually creating more work for them because it does resolve some bottlenecks. So when you use it to take notes, for example, with a physician or his wife is an imaging

When Tech Creates More Work

SPEAKER_02 19:35

specialist, it does resolve the immediate bottleneck, but then it actually creates more work for them when they have to review some of the less clean data that comes from using AI to capture notes and so on. So it's resolving one bottleneck, creating another one. The net is not less time for them actually spent. So that was enlightening. And much of that people feel they understand the process, but until you're living it and you actually a physician who meets with patients every day doing the job, and then you don't have the time to look at technology, but also looking at technology, and so that working together and co-developing is absolutely essential.

SPEAKER_01 20:21

And there's like uh two incentives uh competing there, because when we have a good technology that creates just a bunch of notes, for example, that's great. We have achieved success with implementing a useful piece of technology, but we're still failing with actually getting the outcome that we want, which is more clear decision structure because we have more historical context. Because uh the many notes may be more difficult to kind of read through than the three sentences you could reasonably write down after a conversation with a patient. So the technology succeeded, but the process is still failing.

SPEAKER_02 20:56

Yes, and the one of the examples that he gave is that many healthcare systems and healthcare providers allow patients to email. So yeah, that's great. It's great from a patient's perspective because you don't have to wait to get an answer if you're on the phone or what have you. But then at the other end, the physician has many, many, many more queries because it's easy to send the question and then they have to go through all of those emails and respond to them. And technology isn't yet resolving that bottleneck. So uh yeah, you resolve one bottleneck, create another one. You need to look at the whole process. You need to actually test that. If we allow patients now to email, yes, it resolves one problem, but what does it do for the healthcare uh providers? And how do we resolve that before we implement the first one?

SPEAKER_01 21:42

Yeah, at each like point where we have congestion, perhaps there is a trade off to be made. Like, do we want more of this, or is it actually better that we move slower but we have higher quality and maybe the overall process is slower but we get to the results we want faster? After like there there are decisions to be made by humans about what is it that we're trying to achieve and what is the best aim to actually get that outcome. Absolutely.

SPEAKER_02 22:09

And I guess we digress a little bit from life science to healthcare, but it is very relevant in the clinical trial setting. So you've got the healthcare providers who are addressing patient questions, and it's great for the patient to have a way to ask their questions as long as they can get answers quickly and not drop out of the trial, and that puts the burden then on the healthcare provider to respond to all those questions. So we can use technology there very thoughtfully actually to help with that, not completely resolve it, but to help.

SPEAKER_01 22:42

And maybe going back to the clinical space a little bit, Anastasia, I am genuinely curious. In your experience, have you seen AI help us identify you can say the right target patient population for treatments? Because in my experience, that is one of the bottlenecks for starting a trial. Which patient profile should we be recruiting in the first place?

SPEAKER_02 23:03

It is, and it helps in different ways. So for example, starting from the design of the trial, when you're designing the trial, you can use AI to build cohorts, what we call cohorts, so patients who have those characteristics to

AI For Cohorts Recruitment Retention

SPEAKER_02 23:19

identify where are those patients geographically, and then use those patients to simulate like the feasibility of your trial, how they're going to respond to the trial. And by respond, I don't necessarily mean to the drug, but to the actual process of the trial. And you can also reach out to patients more easily using current technology by getting ads in their feeds and waiting for them to respond to them or giving them useful information for their condition to go and explore clinical trials, explaining for them what a clinical trial is, kind of alleviating some of their fears, because many patients are a bit leery of getting involved in a clinical trial. So once you've identified them, you can then have a have technology help them get comfortable with it and help them with which site is closest to them and most convenient for them. If the trial is enabling doing the trial at home, getting the drug delivered to them at home, or going to their nearest pharmacies, so you can start looking at patients who are in different geographies than where the site that's running the trial is, because you can engage them in other ways. So there are several different ways that you can use AI and technology to identify patients, identify where they are, identify the stage, help to get answers to questions to see who qualifies, and then look at if you are as restrictive as you want to be, is that gonna give you the best outcome on the trial? Are you gonna be able to get the number of patients that you need in order to achieve the outcome? So those are all things that you can test in silico, as I like to call it, or with technology and AI before you start the trial.

SPEAKER_00 25:04

No, that's interesting. I mean what strikes me there from a lot of that is how I'm gonna use this term soft some of those factors are, right? In terms of it's about how much can you help patients understand and how much can you help them manage their own fears and insecurities about being involved in the trial and and sort of willingness to to continue, right? And and convenience in terms of engaging in the trial. Are those then these softer factors quite big issues then that are impacting trials right now?

SPEAKER_02 25:32

They are, because if a patient has started in a trial and then drops out because of a side effect they didn't expect, or they can't get to the site to get the next drug. They're not sure if they're on placebo or an actual drug, they're not seeing the difference, or they have a side effect that nobody's responding or they don't know how to resolve it. All those different reasons and more probably why patients will drop out of trials. Now you have the cost of bringing another patient in and the delay of now you're starting with another patient from the very first dose and not, you know, the patient, the other patient may have dropped out months or even a year or two after they had started. So now there's a cost in reinitiating, there's a delay, and that affects how fast you can get the drug to other patients that are waiting. So absolutely, the better we address and support patients to go through trials, find the right patients, engage them, keep them engaged, answer their questions, support them with scheduling a visit or getting an Uber to the site or whatever it is, and there's different ways and therefore different pieces to the technology that can help support them, the more likely you are to keep that patient through to the end of the trial. And that avoids, like I said, in fact, it's tens of thousands of dollars, and sometimes hundreds of thousands, uh, depending on at what stage the patient dropped out and how many patients drop out. And it can shorten the time significantly because you don't have to now write an amendment to the trial design in order to accommodate this.

SPEAKER_01 27:07

It's kind of like it's uh cheaper to retain the same employee and then go out and find a new one. Don't get me started on that.

SPEAKER_02 27:18

There's actually a lot of churn that's happening right now in everywhere. But yeah, that's another topic for where you know we're we're looking to uh new blood. New blood is fantastic and great to do, and we absolutely need to do that. But the rate with which we are changing and we're restructuring and giving people new roles or getting new people in and so on is hard to keep up with, for sure.

SPEAKER_01 27:41

That that is a whole separate topic, I agree. But it is a topic that is uh quite present in life sciences uh right now. I think many big companies are restructuring after restructuring after restructuring. And trying the good thing is trying all sorts of things that are interesting, but also like quite whoplashy for the employees.

SPEAKER_02 27:59

Well, and a little bit of attention span to see if the first change had an impact before you make the next change and the next change. And everywhere. It's not any one company or any one industry, but I of course I'm noticing it a lot in the pharmaceutical industry and in biotech as well. And I can understand it a little bit more in the biotech space. It's harder to see the value and the benefit in in the pharmaceutical industry right now with the changes being literally monthly.

SPEAKER_01 28:24

And it's uh kind of back to the clinical trial. If we were to change the trial design on a monthly basis, like we would never get to see whether it works or not.

SPEAKER_00 28:33

You're absolutely right. Absolutely right. That brings us back to a point we were discussing earlier then. I mean, uh whatever you want to call it, this sort of more haste, less speed, I guess. When we're having the conversation earlier, I was also thinking about you know these drag racing cars, right? Where they they sort of floor the accelerator immediately at the start of the rate. Then you see the wheel spin and spin and spin, but no one's going anywhere. And I think that maybe we've got a lot of wheel spin in the in the industry as a whole, in various different places, of which this AI trial thing is one. And if we can get better at understanding what it is that we're trying to achieve, how we intend to achieve it, and measuring that success, you know, measure twice, cut once, then could we perhaps be making some more progress?

SPEAKER_02 29:14

Absolutely. In fact, I like your analogy, and I wonder if anyone has tried this. So flooring it, your wheel spin, eventually they get traction, they go. What we're actually doing is not having the attention span to keep the foot on the gas though. So we're taking it off, putting it on, putting it, right? That's what we're doing. And is that actually gonna get you there faster, or are you gonna stay in the wheel spinning longer? We should probably test that out with the cars.

SPEAKER_01 29:37

That is something where we can find out. And then maybe at least like you have the experience, like you are uh being fast, but in fact you're probably not it's it's that illusion that you're getting somewhere faster, right?

SPEAKER_02 29:49

That's something that I've noticed is when you're stuck in traffic and you take a shortcut or you get yourself out of traffic and take a long way to somewhere, do you actually uh get there any faster? You're maybe happier because you're moving, but you're not gonna necessarily get there faster.

SPEAKER_01 30:10

Anastasia, you were talking about digitalization in the clinical trial space. I have the word digitalization in my current title. And very often people ask me, what does that actually mean? What is digitalization? So I'm gonna put you on the spot here and ask you, what is digitalization?

SPEAKER_02 30:27

In the simplest way, it's turning paper into digital. So in the lab, it would be when you're running your experiments, you're collecting the data, moving them from your instrument directly to your lab notebook or moving them into whatever their final resting place is, instead of taking notes and then you have you know handwritten

What Digitalisation Really Means

SPEAKER_02 30:48

notes and you have to go type them somewhere to turn them digital. So this is automatic digitization. And instead of keeping them as a paper result, which only a few people have access to, and you can't reuse the data. This way you're able to reuse the data and an easier way collect data. So you can collect more data also, which is an added advantage.

SPEAKER_01 31:10

That that's absolutely true. One of the ways I've been trying to explain it is maybe a little bit as a joke. Before, as a functional leader, yeah, I had to walk around and talk to people. Now I can just message them on Teams.

SPEAKER_02 31:23

Yes, which is a double-edged sword as an advantage for sure, disadvantage as well. Conversation and communication is a good thing. Yeah, absolutely. Do you remember an example in the past where the wrong outcome or the wrong thing was done because of misinterpreting a shorthand?

SPEAKER_01 31:40

Ooh, yeah. To be honest, often when people they send me these two or three-letter acronyms, like in in chat language, I have to Google them. I often have to also.

SPEAKER_02 31:49

Does that mean we're all fashioned?

SPEAKER_01 31:51

Yes. I think yes.

SPEAKER_00 31:53

I like your double-edged sword talk because it's I guess it's the two sides of digitalization also, which is that you have all of this data, you have all of these voluminous data that you can then do things with. But I guess the sheer fact of that volume makes it actually quite difficult to do something useful with it. Because you've got to wade through so much and make sure it's all the right format or different ways of measuring something, or are kind of equalized the same way and and and all of this. So we've just got so much at our fingertips, it's actually really difficult to work out how best to use it.

SPEAKER_02 32:27

So I think the fact that we have a lot of data, so part of the reason why I made the shift from being a lab scientist, bench scientist, to data science and technology is the sheer passion of doing more with our data. It used to be the temptation to run a new experiment every time. So you design the experiment, you run the experiment, you use it once, and then you move on, you design the next experiment. But in reality, those experiments and the results can be used more. If you have all the context right and you use it appropriately, AI and predictive science, as we used to call it, has been around since like the 40s. But it was much more of a the experts, you know, bottom-up using predictive capabilities, using AI and trying to convince others that it's okay, get safe. Part of the reason for accelerating in 2022 was, of course, the transformer models that were developed, but because we had a lot of data. So you have a lot of data, you need a lot of data to really use AI effectively. And so it took us a while to really make good use of big data. I don't know if you remember that term back in the 2000s. I'm dating myself now, right? In the early 2000s, big data, we're generating big data. Clinical trials, it's not so big data, but there are a lot of data that you're collecting from different trials and with the right context and the right consent, of course, from the patients. You can use that data to understand patients better, to understand disease better. And all of the pre-clinical data and all the genomic data and uh human genome project data that we've collected, all the imaging data that we've collected, we're able to use all that data much better now because of the technology that we have. Having a lot of data is a good thing. Collecting data just for the sake of collecting data is not the best thing. So you still have to design what are you trying to achieve to collect the data and actually standardize the data and store it appropriately so you can reuse it. You may or may not have heard the term fair data, right? Findable, accessible, interoperable, and reusable. So you've got to be able to do all of that, and therefore the data you must use some standards with the data so you can compare one type of apple to another type of apple, sometimes even to a different kind of fruit.

SPEAKER_01 34:33

Anastasia, I I find it super interesting that well, both the the the lineage that you outlined that we started with predictive sciences and we've had like this whole movement towards data analytics in pharma, and now it's kind of tipping over to the AI part. But you kind of need the like clean usable data to be able to advance in this value chain. And I'm just curious, did each stage of our data evolution as an industry add on, but kind of encompass itself? Or did we actually shed some of the original ideas and replace them with newer ones? Or are we just kind of maturing gradually into what data science can do for us?

SPEAKER_02 35:17

I think we're maturing gradually for sure. What I find intriguing is that we're still talking about data silos. We were talking about data silos again in the early 2000s. Twenty years later, we're still talking about data silos. We're still generating data silos, but partly because we are generating new data and we have new measures, new ways to measure data. We're collecting more data, say, from devices, right? So that's new data, new type of device. Uh you put it in its own place. Now you have to standardize it and somehow connect it to other data. We also

Data Silos Standards And Cleanliness

SPEAKER_02 35:50

evolve from warehousing, putting all the data in one place. You don't need all the data in one place. You just need to be able to access it. We have better ways to connect the data. You do need the data standardized. You need to have use some common language, right? Common ontologies to standardize the data. But you don't actually need to clean all the data. You don't need completely pristine data, because if you think about what ChatGPT does or did, right, it just scraped all the data from the internet. How clean was that? How clean is it still, right? And part of the reason why it works is because you want to train your model to recognize the difference between good data, bad data, clean data, not so clean data, standardized, not standardized data. We don't have to spend all the time in the world to actually clean all the data if we use the current technology to actually learn from the data that we have. You're almost doing yourself a disservice if you only give a model clean data, it then doesn't know if it got one piece of data that's not clean, it probably won't know what to do with it. Whereas if it's trained, it won't make and you give it some feedback. And as you're uh training the model, it can make better use of the data in whatever shape it's in. I think at one point in also in the early 2000s, we talked about self-organizing data. I don't think we're there yet, but wouldn't that be great? Self-organizing, self-standardizing data. I think we probably have the technology now to get closer to that.

SPEAKER_01 37:15

That would be fascinating. I mean, we're still trying to get the self-organizing teams to work if our data can also self-organize.

SPEAKER_02 37:22

Well, except you have the personalities in the teams, so I don't know. You might have less personality in the data.

SPEAKER_00 37:28

Just thinking back to what we're talking about earlier and going back to this point about there being a lot of niche AI applications. Then I wonder, is some of that due to simply collecting the data? I would imagine that for these niche applications, it's much easier to get your hands on sufficient data to utilize and to do something with, whereas if you're trying to get data over a whole clinical trial process with lots of different elements going on and different types of communication needing to be done, then that's a much bigger task. Does that roll into this at all?

SPEAKER_02 37:59

Yeah, absolutely. Clinical trial data have unique challenges that we need to be aware of, not least that you have personal identifiable data actually that you want in some cases. So for the trial outcome, you need to, you know, the people who are running the trial need the identifiable data. But for any other post-trial analysis, you need to remove any personal identifiable data when you want to do some secondary analysis. The methods right now, so you can use large language models, you can use small language models, you can build small language models based on a specific trial. But there's advantages, of course, to looking at cross-trial

Privacy Protocols And Cross Trial Analysis

SPEAKER_02 38:37

data. You can do that appropriately, and we have the methodologies right now to do that appropriately. So I think your question was does it create unique challenges? Yes. Can anyone use and analyze clinical data? No. So you have to have you have to understand the trial. And it's not just a matter of analyzing the data. The data are based on the protocol that was used. So whenever you're analyzing the data, they're dependent on the protocol and the method that you use to collect the data. So you can't just take the data and analyze it. You have to relate it back to the protocol. And when you're doing that across trials, it gets more complex. Yeah, that makes sense.

SPEAKER_01 39:16

I think it is time for our game show round. And this game show round is a little bit inspired by the introduction video that you sent to us, Anastasia, which was also about drug approvals and how do we optimize the clinical value chain so that we can actually get the drug approved. And that made me think, I wonder how drug approvals are doing these days. And I looked up how uh the FDA has been farying for the first quarter of this year, and they have been uh denying a lot of drugs. So that's the context for this quiz. The FDA is denying a record amount uh of drugs.

SPEAKER_02 39:55

More so this year, I actually haven't looked at this.

Game Show FDA Denials And Why

SPEAKER_02 39:57

So are they denying more this year than in the past?

SPEAKER_01 40:00

Yes. At least in this quarter. So I don't know if it is like a seasonal thing or there's some other reason why this is happening. But I'm going to ask you three questions about a drug that has been denied by the FDA in the first quarter of 2026. And I'll give you three options for why it was denied and I'll have you guess. Okay, I didn't do my homework for this, so let's go. Let's see.

SPEAKER_00 40:25

Are we allowed to ask clarifying questions or do we only get the information we're given? You can attempt to ask clarifying questions.

SPEAKER_01 40:32

Okay, great. Use Google to search. Exactly. Gemini, help me. Yeah. Yeah, right? Okay, well, first question. So in February 2026, FDA denied approving Moderna's mRNA fluvaxy. And why did they do this? I'll give you three options. Was it manufacturing concerns? Was it concerns about the control arm of the trial? Or was it insufficient data? And this is Moderna's mRNA fluvaxy.

SPEAKER_02 41:01

Alex, I'm gonna give you the first chance. I should know this actually, because I remember reading about it, but I don't. I don't remember. I'm gonna say manufacturing. For some reason, I was thinking there was something not quite right in the manufacturing process for this one, but I'm kind of guessing. I'm not remembering correctly.

SPEAKER_00 41:16

Alex? I'll go for control arm of the trial, so often difficult to keep straight.

SPEAKER_01 41:21

Well, I am happy to announce that Alex for Wunth, you get a question right? I handed that to you, Alex.

SPEAKER_00 41:28

Yeah, yeah, yeah, exactly.

SPEAKER_02 41:30

I was being a little too much of a drug development person to say, for sure we did the design right for this.

SPEAKER_01 41:36

Ah, but this is this is actually a fun case because it was in the control arm of the trial. They did not use the highest possible dose of the alternative treatment. And that was uh why it was denied. So this was highly contested and all over the news. All over the news, and I missed it. Wow. Okay. Well, all over the to be honest, pretty niche news that I read about life sciences. So maybe it didn't get to the BBC. Yeah. All right. Then in sometime this quarter, the FDA denied approving Regen X BIOS R. And why did they do that? Was that due to manufacturing concerns? Was that due to a requirement for a placebo trial that involves placebo surgery? Or was it a change in standard of care?

SPEAKER_02 42:27

Can you tell us what the drug was?

SPEAKER_01 42:29

Yes, yes, I can. Let me just make sure that I have the right information here. Because it is the uh treatment for the ultra-rare disease known as uh Hunter syndrome. And it is a brain implant.

SPEAKER_00 42:44

Then I'm gonna go for the need for surgery, surgical intervention. But I'm slightly metagaming here because then I'm uh questioning why why that element was in there if it's not that. Maybe you're trying to double block me this time around.

SPEAKER_02 42:56

What were the three options? Because the other one was similar, right? It was changing uh standard of care or something, right?

SPEAKER_01 43:01

Yeah. Manufacturing concerns, a requirement for a placebo trial that would involve placebo surgery, or a change in the standard of care. Oh yeah, it's gotta be that middle one.

SPEAKER_02 43:11

You're not gonna have a placebo arm implanting in patients who don't need something. But you would think that the FDA would be able to account for that and not require something that's not gonna happen. But I'm gonna go with Alex on this one.

SPEAKER_01 43:25

Well, you are both correct. This was in fact what the FDA requested: a placebo trial that would involve a placebo surgery, which was unexpected for everyone involved.

SPEAKER_00 43:38

Talk about a problem of recruiting patients. I was gonna say, talk about trials you'd be scared to sign up to. Placebo uh surgical interventions. That doesn't sound like a lot of fun.

SPEAKER_01 43:48

Yeah, so the the drug company has appealed because that does not seem reasonable for the patient population, and also the patient association has been quite concerned. By this outcome.

SPEAKER_02 44:01

Right.

SPEAKER_01 44:01

Okay. But interesting news. Like, wow, really? Placebo surgery. So Alex, you're two for two.

SPEAKER_00 44:07

Fascinating one to dig into the detail of, I'm sure.

SPEAKER_02 44:09

Yeah, no, for once, right? I maybe it's my lucky week. The second one was a device problem, though, so it's true.

SPEAKER_00 44:16

Yeah, well, that's that's more my area. Yeah, yeah, exactly.

SPEAKER_01 44:19

Right, let's go for the third and last one. Why did the FDA deny approval for Britty Fanly Map in March 2026? And before you even ask me what this is, I will look it up. But I will give you options while I look that up. Option number one is manufacturing concerns at the Phil Finish Manufacturing Plant. Then request for an additional clinical trial or request for additional trial endpoints. It is a treatment for cell carcinoma.

SPEAKER_02 44:50

And request for additional endpoints or request for additional trial? Additional trial endpoints. That was one. What was the what was the second?

SPEAKER_01 45:00

There was a request for additional clinical trial or a manufacturing concern at the fill finish manufacturing plant.

SPEAKER_02 45:07

So if you're requesting additional trial endpoints, you pretty much have to do another trial. So those two are related, so I could pick one or the other. Because if you're gonna have new endpoints, you're gonna need to run another trial.

SPEAKER_01 45:19

That is a good point. That is essentially just giving you one option. I'm giving you two options to pick between. Was it a manufacturing issue or was it an extra trial?

SPEAKER_02 45:28

And this was a cell therapy.

SPEAKER_01 45:30

Yes.

SPEAKER_00 45:32

I'm happy to follow my sword and and say they request an additional trial.

SPEAKER_01 45:35

Yeah. Because I'm so eager for an extra trial with the two options.

SPEAKER_02 45:39

I think an additional trial because they need new outcomes. Different outcomes.

SPEAKER_00 45:44

I mean, to be fair, I do know. I mean, health authorities are getting a bit tighter on endpoints these days. They're wanting to see more relevant, more clear endpoints around safety and efficacy. So I don't think it's impossible. I'm suspicious that there's two clinical trial answers though.

SPEAKER_02 46:00

Right. So I'm gonna go with needing a new trial because they need new endpoints, though. Because I I don't know if that messes up your uh your algorithm here for uh what's the right answer.

SPEAKER_01 46:10

I believe this just uh showcases my way of providing you options for this quiz. But in fact, the trial data was fine. Lack of approval for this treatment was due to an a manufacturing site inspection because of the inspection findings. Interesting.

SPEAKER_02 46:28

Which the hint was that two different answers, those were uh yeah, okay.

SPEAKER_01 46:33

That was just uh that was the trick in this question. But it's uh one of those sad situations where ah the drug actually works because we can't uh prove that it has been manufactured safely.

SPEAKER_02 46:47

It'll have to go another route. Or you have to find different manufacturing processor, a different manufacturer. Yeah.

SPEAKER_01 46:52

And since then, this manufacturing plant has actually been sold to another company. So hopefully that that that plant will be brought up to snuff.

SPEAKER_00 47:01

Yes, interesting decision to buy a manufacturing plant that's just failed in inspection. Hopefully it came at a discount. Yes.

SPEAKER_02 47:09

It's probably a bargain price. Yeah, exactly.

SPEAKER_01 47:13

All right. Well thank you for for for playing with me. Yeah, I find it interesting. The reasons why drugs aren't passing the FDA's approval process right now.

SPEAKER_02 47:24

I'm gonna start reading up on this. I usually look for approvals and celebrate approvals. I've not looked for when something does not pass, but there's a lot of learning in that for sure.

SPEAKER_01 47:35

It's been quite the rabbit hole. Fascinating. Okay, well, I'll start rounding off this episode. I would be curious to hear what have been your most interesting findings in our conversation. Any takeaways or new thoughts that our conversation has sparked that you might be thinking about the next couple of days? Who do you want to start with? Let's see who looks ready.

SPEAKER_02 47:59

I was gonna answer the question by saying uh the your regulatory questions, so why drugs are failing and the fact that it seems more are failing in this first quarter than in the past, perhaps. So I'm gonna be digging into this to understand that a little bit better.

Key Takeaways And Where To Connect

SPEAKER_02 48:15

And uh also we'll be keeping up with the news not only on successes, which I usually celebrate, but also failures. Why did things fail? So I'll be ready next time. You're gonna pick something different though though at that point, right? It'll be uh drug or Pokemon, but I think I would have done better with that one. So that that was uh that was the main learning. I also really enjoyed the conversation.

SPEAKER_00 48:37

Yeah, I think for me the the most interesting thing was again the need to really embed this lesson of although we can be very excited about a new technology and a new innovation and what we can do with it in a very immediate way, a lot of the real long-term effectiveness comes from taking that step back and really understanding what the need is that we're using that technology for and that fit between them. You know, if we do that, maybe we can shortcut this adoption pattern that we often see where you get a lot of different applications come and 95% of them fall by the waveside and five percent that that are well matched survive. So there's a lot of process stuff we can do to uh to be better here.

SPEAKER_02 49:16

Absolutely. And looking at the whole process, right? Not just a little bottleneck look at the whole thing.

SPEAKER_01 49:21

And for me, Anastasia, kind of the red thread that you painted for us from how the predictive sciences were something we talked about earlier in life sciences history and how that has evolved into data analytics and now also AI, that there is a clear red thread of evolution of how we mature as life scientists, and we need to be successful with what we have now. We also have to get uh the previous steps right. And I have extensive notes on that piece of our conversation. Excellent, fantastic. Well, Anastasia, if our listeners have forward questions to you or want to reach out, where can they find you?

SPEAKER_02 49:59

On LinkedIn. Please reach out. On LinkedIn. Love to meet people, answer more questions. This has been a fun conversation. Thank you so much for uh including me here in your podcast. I've enjoyed listening to it, and I hope the listeners will enjoy this episode as well. Well, thank you so much for joining us. Absolutely. Have a good rest of the day.