Leveraging AI for Medical Affairs
Speaker: Lori Klein
Speaker: Sonal Bhatia, MD
In Medical Affairs, our eyes are open to the vast possibilities of Artificial Intelligence — not only to make our current work more efficient, but to expand our capabilities in areas we never considered possible. Here, MAPS speaks with experts from Pfizer and Putnam Associates about the most exciting current uses of AI and also where we may be headed in the future. Join us for a wild ride through the possibilities of AI.
Following is an automated transcription provided by otter.ai. Please excuse inaccuracies.
Garth Sundem 00:01
Welcome to this episode of the Medical Affairs Professional Society podcast series: “Elevate”. I’m your host, Garth Sundem, Communications Director of MAPS. And today we’re speaking with experts from Putnam and Pfizer about what AI tools we are implementing now in which we are exploring for Medical Affairs. Joining us are Lori Klein, Partner and Medical and Scientific Affairs Practice Lead with Putnam, and Sonal Bhatia, Chief Medical Officer for Rare Disease at Pfizer. So first, I’m going to say if listeners haven’t already read the MAPS vision for AI in Medical Affairs, white paper, Putnam contributed along with many other organizations, you can find it at MedicalAffairs.org. And it’s a nice place to start. We have picked some topics that we think were under discussed in that paper as the focus of our chat today. So Lori, let’s get started with some, I guess we would call them use cases. But what is your current experience with AI and Medical Affairs? What what are you implementing now? And then maybe next, we’ll get into what we’re exploring, but are you using it? And if so, what in the world are you using it for?
Lori Klein 01:22
Yeah, thanks for having us, Garth, it’s very exciting to talk about AI. And we’re using it in a lot of places with Medical Affairs. You know, whether it’s machine language learning, machine learning, natural language processing, generative AI, we’re implementing it across the board for different use cases in with clients. Things like using generative AI to create plain language summaries from full length publications, we’ve used AI for medical insights, analyzing the insights, we can also use AI to do classify data and conduct sentiment analysis so that we can help optimize omni channel strategies. We’ve also been able to chain to use it as an impact metric, because we can look at changing clinical behavior within claims data. And then, you know, I’d say also, we’re doing things like aI landscape assessments, where we’re looking at tools that are out there being developed and identifying tools that might be good for partnering to, like improve an asset that’s in the pipeline. So I think there’s lots of different places to get started with AI.
Garth Sundem 02:33
That’s interesting. I thought for the first half of that answer that we were going to stay in the land of condensing meaning from large amounts of unstructured language data. But then you took us into many other uses, including looking at changes in physician behaviors through claims data. Those will be interesting to unpack a little bit more as we move forward in this conversation. But Sonal, are you using AI? No. And if so, how?
Sonal Bhatia 03:09
Yeah, no, I think we’re all delving into the world of AI and trying to figure out the how the how is, is really the Holy Grail, right. But when we think about all the wonderful examples, Lori provided, when I think about large hospital networks, we want to be able to implement AI tools into electronic health records to help physicians be able to, for example, detect patients earlier in the patient journey, or make sure that the patients are diagnosed accurately while in advance, so that they have a shorter journey, and they get on treatment even faster. So that is the big impact, I think in healthcare that we can have, as industry works collectively to create these tools. Obviously, there’s generative AI as well. So we can actually use generative AI to solve for a lot of things. As simple as helping us draft papers and get them through the approval process much faster. What does that mean to the patient into the physician, it means physicians have access to data that’s going to help them inform them on the efficacy and safety of medicine much faster than traditionally has been done. So those are two very different modalities of the role of AI.
Garth Sundem 04:29
Yeah, those are very different modalities. You know, it seems like both kind of center on sort of pattern recognition or recognition of meaning within large data sets, you know, for the patients, I guess, if you’re identifying patients earlier in the what diagnosis journey or treatment journey that would be looking into a massive database of claims data and trying to identify the features of people that would, you know, maybe warrant fall follow up for in your case, rare diseases, is that what how you’re using AI in that situation?
Sonal Bhatia 05:06
That could be one way, which is retrospectively looking at the data that’s within the charts and claims data and assessing the proper diagnosis. But even when you think about prospectively, you could have an algorithm placed into into the tool being placed into the health network. And that tool would then for example, pop up on a physicians workflow and say, Have you considered based on these risk factors that we’re seeing in this patient profile? Or, you know, that identify any identifying characteristics, but the risk factors themselves could give the physician an insight of, Oh, am I thinking through the diagnosis accurately? Or is there something else, and it could then take notes for a different diagnostic journey, to finally find what their patient’s problem is.
Garth Sundem 06:00
Okay, and let me know,
Lori Klein 06:05
I was just gonna say, similarly, you know, you can develop computational phenotype, mental models that can be applied across, a lot of systems are now, you know, connected from, you know, there’s things like P kids, which is a peds net that has, you know, eight or more institutions that are all connected, so you can look across their electronic records and look for patients who may have some characteristics of a particular rare disease. So it becomes a very good way to identify patients earlier.
Garth Sundem 06:37
It’s interesting. So I’m interested to know how these tools are being developed. Are these being developed, you know, obviously, over a platinum, but then are they deployed by Medical Affairs? To physicians who are, you know, HCPs, more broadly defined, that we would usually be interfacing with in Medical Affairs? Or are these kind of rolled out from the commercial side of the organization? Who’s developing these? And how do we actually get them out there into the provider landscape?
Sonal Bhatia 07:15
Yeah, that’s a that’s a very good question. And I don’t think there’s one size fits all, as you think about companies, different companies. I know, what we’re doing is medical is developing these tools that the research arm off, let’s say this, this landscape, right, there’s a research arm, and that is being done by the Medical Affairs in partnership with a host of clinical team members as well. So it’s not just medical. And once we develop these tools, we work with external partners and vendors who are helping us validate them. Once they’re validated, then in the next phase, which is where we’re going to go. So this is all in process right now, is going to require partnership with commercial to deploy them into the space, and I think commercial and our field medical and field outcomes and analytics teams, all of them will play their particular role within their standard operating processes to ultimately get it into the hands of the provider, or the hospitals.
Garth Sundem 08:25
Okay, I don’t want to take us too much into the weeds of implementation. But we’ve spent so much time on the vision side of AI that it is interesting, I think, to dig in a little bit on what do we do with these tools physically, mechanically, in the real world, to put them into the hands of people who use them. So okay, all that said, Now, let’s get back to the vision side, what cool things are you looking forward to on the horizon of AI that you’re not implementing yet, but that you might be exploring or are excited about? And either Lori or Sonal, if you want to start us out? I’m sure there’s a lot.
Sonal Bhatia 09:02
I mean, I think I’ll start and then Lori, I’ll pass it to you. But I think I’m very excited to already see how we’re taking the data and the insights from our fields and bringing them in, and actually having some outcomes based on those insights. But it’s a very quick process. So unlike the weeks and months it took to bring an inside look, see it on some, I don’t know, PowerPoint or something. Now you can take a lot of data and within minutes, you can pull insights. What I’m curious to see in the future is how those insights can potentially shape clinical trials and, and make those trials more relevant to the patient and the regulators and ultimately, the physicians and actually much faster with the right changes in those trials so that we are welcome Entering more positive trials at a faster rate. And now waiting to work completed phase three to say, If only we knew X, I think that if only we knew will get solved now that we’re taking these insights at a much faster pace. So we’ll have more studies going through Phase Three at a faster rate and more studies that are being positive, which will be wonderful for patients.
Lori Klein 10:25
Yeah, and I think just adding adding on to that, I think, I was excited to see that the FDA put out a call for comments on using AI and drug development and really providing a lot of specific use cases, whether it’s generating real world data incorporating digital health technologies, from recruitment to data analysis. So I think it’s kind of laying the foundation for a lot that can be happening. And before that happened, I think a lot of people had questions around well, you know, we decide to use generative AI, what is that? Is the FDA going to be cool with that? And I think they’re laying the foundation for, there’s a lot of potential good things that can come out of leveraging AI.
Garth Sundem 11:12
Okay, let’s go back just quickly, Sonata insights, you said, something that caught my ear, and one is that you’re excited about the efficiency of insights. But that often is where the endpoint for AI, sometimes we get stuck just in seeing AI as an efficiency tool. You know, it helps us create standard response letters quicker, or, you know, plain language summaries quicker, or, you know, it helps us do these things more and more efficiently. But what you said is that the efficiency of insights created by AI is going to help us create more positive clinical trials. And that seems to be value creation, outside of just efficiency. So are there are there things, other things or other other things like that, that are not only advances in efficiency? But advances in value beyond what we’ve been able to do before? You got it by AI?
Sonal Bhatia 12:22
Yeah, yeah, absolutely. I think, you know, everyone is worried that AI is going to replace their roles. And I, I actually disagree with that statement, I think AI is going to make things more efficient. But at the end of the day, it is our brainpower that is going to utilize whatever’s coming out of these AI processes are however, we’re utilizing it to ultimately add to value. And one of the things is okay, so we collect all these insights. And I talked about our journey through clinical trials. But another one is often we get a lot of ideas on potentially a research component. Now, we may hear from different people in the field on research, but it could be at different parts of the year or in different parts of the organization. But with AI, you can put it all in one place. And when you put it all in one place, that research AI, concept or idea or unmet need will surface at a much faster. And sometimes it’s not even fast, it wouldn’t have happened without the ability to collate all these insights from all these different groups. Because insights are not just from one function, commercial has insights, regulatory gets insights, our clinical team gets insights, everyone’s co leading insights, but they’re staying in a silo. And my hope is that with AI, you bring them all together, and a research idea could surface or an unmet need, that could potentially lead to a new research or a new indication, or a new area to delve into, that we may have not considered before. And that would be value, tremendous value to our physicians or thought leaders and ultimately, our patients.
Lori Klein 14:10
Oh, you can also look at, you can look at sentiment of insights changing over time, and you can overlay like what you’re doing, what actions you’re taking, and show that, you know, your content strategy is moving the conversation. So there’s lots of value that can come out of things like that impact.
Garth Sundem 14:28
I was gonna say that previously, probably all we could handle was one of these silos, right? And we were stressed to make meaning from the data that was contained within an individual silo. But now with AI, we have the capability to put all these things together and much bigger data, maybe we can see meaning that we never would have seen before that gap analysis that leads to a research project that leads to an okay so another area that I know people are We’re exploring with AI. And we mentioned this very briefly previously was in evidence generation, you know, with RW E. Are you seeing AI as transformative in, you know, helping meet the higher evidence benchmarks now, on the on the edge Gen side? Are we using it? Is that still a future? Are we ditching that idea? What do you think?
Sonal Bhatia 15:30
Don’t think we’re ditching it I think as, as AI is coming more to the forefront and garnering respect by everyone that’s utilizing it, the FDA will also start to look at anything created, even on a real dividends evidence generation platform, if you’re using AI, it will be viewed as an acceptable platform. So that only means that you can get more data out quickly and do quicker analyses that are going to help the FDA make their decisions or feel comfortable with whatever medicine is out in the market. Because sometimes you’re utilizing real world evidence to support an approval that is happening in the in the form in the future. So I think it’s it’s not out it’s in it’s the question is going to be how do we do it in a way that FDA will recognize this and they will, they will wrap their arms around it and say, this is a good tool and an acceptable tool to develop the real world evidence that we hope to.
Garth Sundem 16:38
Okay, here’s another question that inside baseball question, do you think that FDA seeing AI studies as legitimate? Do you think that they need to see inside the black box of the methodology? Or is it just that, you know, valid results and validated results over time will start to show that this is a methodology that can be used? Like do they need to actually look inside the black box of AI? Before they’re gonna believe it?
Sonal Bhatia 17:14
Yeah, I think they will. I mean, I’d love to hear his thoughts. But I do think similar to how when we present them with real world evidence, they want to see what’s your methodology? What are your, what were the pros and cons? What are the risks? What are the mitigation factors to it, because nothing is perfect in real world. That’s why it’s the real world. And it’s not a control study. And similarly, in order for AI to be utilized and validated appropriately, they will want to see how, how did we get to that endpoint? Maybe in a few years after we do that, it will be more standardized, but we’re all starting new. And I think that how will be important, not just for us, but everyone who’s developing data so that we all feel confident on what’s coming out on the other end.
Lori Klein 18:05
Yeah, and I think like, just to add on that, I think, you know, depending on the type of AI tool you’re using, there are features that can be looked at, like, if you’re looking at machine learning, there’s explainability, where you can use models like line or shaft to like, kind of explain how you got to the the answer. But, you know, there’s also just like being transparent about what were the things that you were using for search terms, or, you know, things that were part of the model, anything that we can provide to help the user, the end user, whether it’s the FDA, or, you know, a physician, whatever, we can help explain what went into making the decision will help, even if it doesn’t explain fully what’s going on, it’ll give a little bit more confidence and in the output.
Garth Sundem 18:57
So maybe teams starting to track and systemize there, how would that be a good practice recommendation for folks listening, just just to sort of start formalizing your processes now?
Sonal Bhatia 19:10
Yeah, I think the what and the how, what are you trying to solve for? How did you do it? And then what were the results? Right, I think those are gonna be very important to always have.
Garth Sundem 19:25
So, not just the fact of the thing that AI gives you, but the steps along the way that can help to validate, you know, the results and also the process. All right, so now because we asked this of folks, when we chat about AI, okay, so how is AI going to evolve in Medical Affairs in the next years? And maybe, do you think that AI is influence is going to remain internal to Medical Affairs processes and uses or do you See AI actually helping to meet the needs of audiences outside of Medical Affairs like clinicians and patients? Is it just hours? Or are we going to use it and influence our stakeholders? What do we think?
Lori Klein 20:19
I think it’s, it’s going to be, it’s already, it’s already beginning to be out there, there are AI, clinical decision support tools, there are already tools to diagnose patients earlier, which is all great, because that’s gonna help help physicians in a lot of ways that we’re at the cusp of like implementation and getting more of them pulled in. But I think it’s a really exciting time, it’s also an opportunity to provide more precision, personalized medicine, just being able to customize what a particular patient needs. So I’m really excited about the possibilities. Yeah,
Sonal Bhatia 20:58
I think it’s definitely something that’s going to be front and center in hospitals and clinics and with physicians and patients. I mean, all the wearables are generating data and that data is going up and an AI is looking at it and and helping physicians think through earlier whether patients have risk factors. So it’s going to change the paradigm on how quickly again, speed is important for patients to make sure that a signal that they see maybe a CAT scan that has been taken on a patient with lung cancer, AI is going to help detect that with more precision, as Lori mentioned, right? And they’re going to get the care they need faster. So I always say time is life and for patients AI is going to give that life eventually when you think about how it’s going to be utilized in the healthcare ecosystem. And parents are going to utilize AI as well.
Garth Sundem 21:58
Why there’s about six topics for our next podcast episodes in there. One would be specifically AI in wearables and Medical Affairs use of wearable data. Another one is AI on the payer and access side. Okay, I’m keeping notes here, but for today, let’s leave it at that. So thank you, Lori and Sonal for joining us today. To learn more about how your organization can partner with Putnam, visit PutAssoc.com. MAPS members, don’t forget to subscribe, and we hope you enjoyed this episode of the Medical Affairs Professional Society podcast series: “Elevate”.