Ep. 3: AI in Medical Communications: Why Human Expertise Still Matters

You're listening to a sponsored episode of The Top Line.

Welcome everyone, you're listening to the top line brought to you by Fierce Pharma and Fierce Biotech. I'm your host, Irene Lowell. In this episode, we're tackling one of the biggest shifts in medical communications, the rise of artificial intelligence and what it really means for scientific rigor, accuracy, and the way evidence is created and shared. Today's conversation is powered by experts from RTI Health Solutions.

A global research organization dedicated to generating the evidence and insights life science stakeholders need to make better decisions. RTI Health Solutions unites experts in HEOR, real-world evidence, medical affairs, and communications to help clients demonstrate product value through research and effective storytelling. Joining us are three communication and information specialists who sit at the intersection of science, communication, and intelligent technology.

First, we have Gabrielle Dardis, a scientific and communications strategist at RTI Health Solutions with more than eight years of experience and a special interest in integrated evidence and early communication strategy. While recognizing AI's transformative role in scientific communication, she emphasizes the ongoing importance of human expertise for rigor and trust. Welcome, Gabrielle. Thank you. It's really nice to be here. We're also joined by Francis Ramsey.

Senior Information Services Specialist at RTI Health Solutions. With more than 10 years of experience in scientific evidence identification, Francis brings an expert lens into how AI is reshaping information discovery and risks that come with OPEC algorithm. Welcome, Francis. Hello, thanks for having us. And finally, we have Margaret Mathis, Director of Medical Editing at RTI Health Solutions, who has over 20 years of experience in scientific editing and communication.

As a professional editor, she understands the importance of both details and overall messaging for the intended audience, which gives her a unique view of how AI must be used with thought and care. Welcome, Margaret. Nice to be here. Together, this group represents many of the facets of medical communications that AI touches: writing, editing, information discovery, and strategic storytelling. So we're gonna just dive right in. As AI becomes more embedded across life science,

Where is it genuinely strengthening medical communications and what are the limitations that still most evident? I'll get us started here. If you remember in episode two, Sarah talked about the importance of understanding your audience. So we may have various audiences for one communication piece. AI is good at adjusting one communication piece for multiple audiences.

You can write, say, a plain language summary or have AI write the draft and have a subject matter expert like Gabby review it. And then you can have AI adjust that summary to a different audience. So for example, let's say you're writing a plain language summary describing the prognosis for AOS. AI can draft that for your initial audience. In this case, an adult who is newly diagnosed with ALS and has little knowledge of the disease.

And then you can take that same plain language summary and have AI target it to a different audience, say the teenage child of a parent who was recently diagnosed, they need to know about the prognosis too, but in a different way. So rather than reinventing the wheel, AI can tailor a medical communication to a different audience. And then the subject matter experts can strategically assess it and revise that output as needed.

I totally agree with that, Margaret. I think AI is particularly strong when it comes to handling those early drafting stages and considering your audience and what language might be useful for that audience, as well as there's more repetitive or time consuming tasks that are sort of the low hanging fruit for AI applications. And I think that the flip side of that is where the limitations become really evident. And as you mentioned, having that subject matter expertise review.

any content that might be generated by AI is incredibly important in any application, but especially when you are talking about medical communications. Yeah. But from the perspective of library services, we are looking to the writers and our clients as our audience, not necessarily a broader audience, because we're doing the initial process of information retrieval, information discovery. And

AI tools in that regard, like Gabby just mentioned, are definitely important nowadays as really the the size of the body of literature of certain topics has grown to such an unmanageable size that being able to use AI tools with literature search screening in particular has made it a much more efficient process in terms of

time being spent, human man hours necessarily being dedicated to complete a process of gathering the relevant literature for a project and our work at Health Solutions in particular. But there's also the limitations of within that process, we don't have the transparency and the reproducibility that has always been really important in scientific work. We cannot provide that

when AI is involved, we're not able to specifically document exactly how this works and thus how it can be reproduced. So we've got pros and cons, but we're working to of course to deal with all of it together. Yeah. And there's a growing appetite in the pharma industry for this magic button AI tools, instant insights, instant manuscripts, instant graphics.

Why is expertise, curated data, and human judgment still essential behind the scenes? This is a great question. And it's coming up more and more as AI becomes increasingly universally seen and appreciated for these sorts of amazing things that it can do in the public domain. And that trickles down into the pharma domain and medical communications as well. And that even without explicit requests for AI as a tool or a method.

is becoming evident in the types of project requests or queries that we're seeing on the medical communication side. Like you mentioned, an instant manuscript or instant insights. And AI may make some of these things possible, but it's a lot more nuanced than a magic button. It can support parts of the process. We like to talk about an augmented process where humans are largely involved and have to be involved for a lot of those

insights and checks and balances and things that we already just mentioned, but considering where AI can support parts of that process. Yeah. And AI, it doesn't bring into consideration the context for research questions to be fully developed. It's not the case that a research question that we're working with is black or white just one answer kind of research question. There's

a lot of nuance and context that an AI tool is definitely not going to necessarily comprehend from a query being provided. It is the case that the expertise has to be somebody who can understand, comprehend, and work with the AI so that all of the relevant topics are being looked at and covered. But then there's also issues of where you're looking for

what isn't already out there. You're maybe looking for gaps in the literature. And AI can't necessarily just automatically know that. And so you're needing to find out what is out there so that then you can identify what isn't out there. There's budgets, there's timelines, there's so much that AI doesn't have any concept of the human does. And then from a practical standpoint, projects very least static, as anyone who's worked on a research question before knows.

The question you start with is not necessarily the direction that you end up going for the whole duration of the project. And that's where, again, that human team behind the scenes is really critical. No, absolutely. It's the human touch cannot be replaced with AI. So when AI assists with complex tasks, whether that's summarizing studies, checking consistency, shaping narratives, or visualizing data, how do you ensure the output remains scientifically accurate and compliant? I have a good example of this.

So Francis and I did a lit review recently that was in an area where there were not a lot of data. And we came back with what was it, Francis? Maybe 15 publications. Yeah. Small number. And most of them were abstracts. So I decided to use AI to try to connect the publications with certain clinical trials. And I decided to see how two different tools did it. So Copilot produced this very organized and succinct summary. I was like,

Wow, that's beautiful. It's amazing to read. And then Gemini's summary was a bit more in-depth and overwhelming to me. And maybe that's because I did not prompt well. But that being said, they gave two different results. Gemini linked several of the publications to one clinical trial. Copilot didn't make those connections and actually didn't even address some of the publications. So I had to take a step back and

Do additional research, I engage with the subject matter expert to get some more information so I could then figure out how to connect those dots. It showed me that AI is a good starting point when you're trying to summarize something, but it also showed me how different tools can provide different results that may not be right. We still need a human to quality check the output. Because had I sent either of those summaries directly to the client, they would have been.

wildly misinformed and wildly disappointed in the work we've provided. I know that Gabby has a lot of experience in this area, so I'd love to hear what she has to say about this too. Yeah, that example really highlights, again, quality control is a principle that, you know, at RTI Health Solutions and just personally is really important when doing any sort of research project, and especially with AI where hallucinations are still very much possible or research questions might get

answered to different levels of rigor. And this actually also links pretty closely to one of the previous episodes about communicating HEOR and real world evidence with impact, where credibility and context were central themes of those conversations and apply here too. And just like any other process, quality control is really important and AI outputs should never be accepted just at face value. So what skills would distinguish high value medical communications in an AI embedded environment and

Which legacy skills will diminish in importance? So effective communication is really the most critical skill set that someone in the medical communications field needs to excel at. And that is included when working with an AI tool. You have to understand the research questions. And from my experience, when you are prompting AI for information.

to be gathered and summarized, there is a learning curve. It is not necessarily the case that it is a simple Google type search box. There is the need to develop how you communicate even with AI, because it's not just that you are identifying what it is that you would like to know about. You also perhaps need to say what you don't want included.

So there's there is the need to still understand and be able to effectively communicate about a topic as the human who is driving the AI tool. Absolutely. And I think that point really reinforces the value of skilled medical communicators, where AI might reduce the effort or time required for tasks to derive data or summarize data.

That actually emphasizes the importance of those interpretation, coordination, and judgment steps. You still need a skilled medical communicator to connect those dots. And that's where writers, editors, library specialists, designers, everybody working in medical communication still has to coordinate together to connect those dots and turn those insights and discrete pieces of evidence into something meaningful.

So I've actually been polling my team of editors over the past few weeks about this because we're looking at AI applications that will assist us with editing. And my question to them has been if you had an ideal AI editing software, what editorial task would it do for you? Now, one person said everything. And I was like, Do you still want a job one day? Because that's not what we're after, right?

Overwhelmingly though, the response has been around those tedious tasks that editors do so well but can take a lot of time. And this can relate to all aspects of medical communication, writing, graphic design, literature searching. So if we're editing like abbreviation checks, style guide points, here's something that Gabby can relate to, like aligning manuscripts to journal specifications, even taking care of basic spelling, grammar, and punctuation errors.

Editors are great at all these things, but they take a lot of time away from the true value add that editors bring to medical communications. That is editing for the intended audience, editing to create the story and messaging for the communication piece, strategically thinking about how the words are going to land with the end user. Human editors are the first line of defense for any document or communication, because if it doesn't make sense to me,

It is likely not going to make sense to someone in your audience. So as AI does more of the tedious editorial tasks, I think human editors are going to grow in their ability to partner more with other medical communicators, writers, graphic designers to achieve the goals of the document. We're going to have more time to think strategically about the messaging for a human audience.

So looking ahead, what do you see as the most meaningful future applications of AI in medical communications and what guardrails will the industry need to keep quality and trust at the center? I think as Margaret just that example she just gave illustrates how I think the future sort of might look with AI and Medcoms. I think AI and these low-hanging fruit, sort of tedious tasks, time consuming tasks that AI is particularly strong with assisting with.

I think that'll become relatively the norm. And I think clear guardrails are still going to be very important from industry and professional societies and journals on transparent use and standards of quality and disclosure, of course. But in addition to that, I think I'm hoping that the future looks more hybrid where AI tools are augmenting again, these human medical, skilled medical communicators and what will be important.

for those humans using AI tools is to have their own guardrails on generally using AI. Looking at a project, looking at a task and thinking, do I really need to use AI here? It introduces all of these pros and cons that we've been discussing and discretion from the start about whether AI even needs to be involved or limiting its application to just a certain task or a certain single aspect of a project will really help.

maintain the integrity and the robustness of a project as you move forwards through it. Yeah, definitely, Gabby. I I think all of that is true. And then in terms of the human having the quality control approach, there's always the potential for bias based on whatever AI tool you're using, because there isn't the transparency, as you mentioned, and there isn't an equivalent of a peer review of information coming out of AI. It is

the user, the human user who receives the output who has to serve as that reviewer to ensure that you're getting quality output even from an AI tool. So those are important guardrails, but I do see a potential in the future for where AI could be extremely helpful in a situation that hopefully we don't encounter anytime soon. But looking back, even just

approaching six years ago when the pandemic began, the amount of literature being published all of a sudden on the topic of COVID was absolutely unbelievable. It was unprecedented and it was way too much to manage. But an AI tool would absolutely be able to help with that in the future for any other

pandemic like occurrences that we encounter. That's right, Francis. I'm gonna sidetrack a little bit as I answer this and then get back to the heart of your question. One of the things I want to think about is like Word and PowerPoint are still the main applications with which medical communications are developed. And so I think it will be useful for AI tools to be embedded within these applications. And I know there's some that are

But I've not been as impressed with their functionality and their quality as say when I put some text into Chat GPT and have it edited for grammar, spelling, and punctuation. So I think looking forward, AI needs to think about being embedded in the applications that we use because it's not efficient to take our communications from Word into Chat GPT or Gemini, pull them back. That's a loss of efficiency.

So we need AI within Word and within PowerPoint. And when the day comes that we're no longer using Word and PowerPoint, which I know some people are looking forward to, AI needs to be embedded in the new applications as well. Now, back to your question about quality and trust. We need to remember that healthcare decisions are made by people and should continue to be made by people. In episode one, Mark talked about.

the living human perspective as part of the HTA submission process. I loved that. Living human perspective. Therefore, we need that human medical communicator in the loop creating the messaging with AI as the assistant. If AI does all the work, I personally feel like that's the day when humans are no longer making their own decisions when it comes to their health care because

We're letting AI do it all for us. And that's not where I think people want to be. So as we're navigating, integrating AI more into our workflows, we need to remember that living human perspective because that is the foundation of everything that we do in medcoms. Absolutely agree. I love that. AI has opened up to so many possibilities, but we all still need that human touch to really fully function in today's society.

Thank you, Gabrielle, Margaret, Francis. Thank you for joining us on this episode of the Top Line. I'm your host, Irene Oll. For more information, you can learn more in our show notes at fearsperma.com. And that's the bottom from the top line.