Guest hosted by Victoria Hart, this episode features industry experts Kelly Soldavin and Jon Viney discussing the current landscape and future implications of AI in medical publications from a publisher and agency perspective. From enhancing accessibility and readability through plain language summaries to addressing concerns such as data security and accuracy, the conversation delves into the potential benefits and challenges of integrating AI into content development processes. The discussion also highlights the importance of evolving policies and guidelines to navigate the ethical and practical considerations of AI adoption in medical communications. Tune in to gain insights into how AI is shaping the future of medical publishing and communication.
Victoria Hart is a Medical Associate at Open Health, Kelly Soldavin is a Senior Editor for the publisher Taylor & Francis Group, and Jon Viney is Senior Scientific Director at OPEN Health.
Produced by ISMPP (International Society for Medical Publication Professionals), in partnership with Evergreen Podcasts. The views expressed in this recording are those of the individuals and do not necessarily reflect the opinions of ISMPP or the companies or institutions with which they are currently or past affiliated. This presentation is for informational purposes only and is not intended as legal or regulatory advice.
Thank you for listening to InformED! Please subscribe to the show on your favorite podcast app and rate our show highly if you enjoyed it. ISMPP benefits medical publications and medical communications professionals by providing members with knowledge, community, and professionalism. Consider becoming a member! Visit ismpp.org.
[00:00:04] .
[00:00:35] I'm joined today by our guest speakers, Kelly Suldivan and John Beine.
[00:00:39] Kelly and John, would you like to introduce yourselves?
[00:00:42] Sure. As Victoria said, I'm Kelly Suldivan.
[00:00:45] I'm a senior editor with Taylor & Francis, a publisher.
[00:00:48] I work on publication development and manage a portfolio of original research journals.
[00:00:54] My passion projects are plain language summaries and patient collaboration.
[00:00:58] And I'm really keen to better understand how generative AI can be used
[00:01:02] to improve PLS and engagement with patients, as well as make research more accessible overall.
[00:01:08] As a long-term patient with a chronic illness, these initiatives are really close to my heart.
[00:01:12] I've been involved in publishing for over 20 years, and the first 14 were spent in veterinary medicine publishing.
[00:01:18] So having moved over to the human side, I'm always amazed that we are able to communicate with patients
[00:01:23] in a language that everyone understands.
[00:01:25] And I just want to keep improving upon how we do that. Thank you.
[00:01:29] Hi, my name is John Beine.
[00:01:31] I started as a medical writer in medical communications about 10 years or so ago,
[00:01:36] and now I'm managing a team of medical writers in open health.
[00:01:39] I've been in publication specifically probably for three or so years now.
[00:01:43] And I recently went to the ISMAP EU meeting, and I was very interested to hear all about
[00:01:48] how people are starting to use AI, their concerns, their questions, their excitement.
[00:01:54] And I suppose I'm interested in seeing how the use for medical writers in particular changes over the next few years.
[00:02:00] Awesome. Thank you.
[00:02:02] So John, since you were at ISMAP EU in January,
[00:02:06] what were some of the key takeaways regarding the use of AI for content development?
[00:02:11] As you might expect, there was a lot of discussion about AI. It was quite a hot topic.
[00:02:17] The meeting was two days, and I think probably every session mentioned AI at some point.
[00:02:23] In discussions, in workshops, in the breaks, there's AI everywhere.
[00:02:27] There's a lot of posters on the use of AI.
[00:02:29] With the takeaways, I think at the minute we're probably, from the discussions I saw
[00:02:34] and the kind of presentations we had, it's almost like an early stage, I would say, in terms of adoption.
[00:02:39] People are excited, but I would say people aren't sure what to do just yet.
[00:02:43] I co-led a roundtable with some colleagues from open health,
[00:02:46] and the main feeling from that was that people are excited but aren't sure how to use it yet or if they can.
[00:02:51] So there were, I suppose, more questions than answers about how the AI tools might be used.
[00:02:55] The potential was considered large.
[00:02:57] There were concerns about hallucinations, so that's obviously, you know,
[00:03:01] chat GPT making up information and not perhaps providing factual answers.
[00:03:05] Data security was a big concern.
[00:03:07] It's obviously inputting sensitive clinical trial information into a tool like chat GPT.
[00:03:12] Obviously it can't be done. Obviously confidentiality concerns.
[00:03:15] So those caused some hesitation amongst attendees about implementing the use of AI tools.
[00:03:20] That being said, there was a lot of research presented around the use.
[00:03:23] For example, generating plain language summaries,
[00:03:26] and these were showing that perhaps they're better than humans at the moment at providing readable text.
[00:03:31] Obviously, although there are caveats regarding the indices used to assess readability,
[00:03:36] but it's quite interesting that already these tools are getting quite good.
[00:03:40] A big takeaway from me was ensuring all publications professionals adhere to standards
[00:03:45] and guidelines regarding use of AI.
[00:03:47] And there's a good position statement from ISMAP that was highlighted in the meeting that's available online.
[00:03:52] And that has a list of standards it encourages us to adhere to,
[00:03:56] you know, being responsible, being accountable for the information we put in,
[00:04:00] you know, being mindful of confidentiality and such.
[00:04:03] So I think that was a really big one for me just to try and look at what people in the industry,
[00:04:07] other professionals are doing and try and be mindful of concerns or any issues that might come up.
[00:04:13] Yeah, I think that's really helpful to have that position statement
[00:04:17] to kind of help guide writers and others in MedCamps.
[00:04:22] Did the issue of proprietary models for content development arise in discussion?
[00:04:27] Yeah, it did a fair bit.
[00:04:29] So I think because of the data security concerns,
[00:04:32] I think that proprietary models is going to be the way to go for a lot of industry uses of AI tools.
[00:04:38] They're not going to want their staff or agency staff,
[00:04:41] anyone else to be putting confidential information into a publicly available tool.
[00:04:45] So there's a number of companies developing in-house tools from the discussions I saw.
[00:04:49] There was a poster from some authors at Ipsen,
[00:04:52] which were kind of demonstrating the kind of initial capabilities of a proprietary model.
[00:04:56] This is in kind of filtering and summarizing articles and doing kind of literature screening.
[00:05:01] And they had some way to go, but it's interesting kind of a first step in such a model.
[00:05:05] A lot of discussion in workshops around the use of these proprietary models for plain language summaries
[00:05:10] so we could perhaps imagine a future in which companies have their own model
[00:05:14] which is trained up on PLS text development and they plug in their clinical trial data
[00:05:20] and it produces out a plain language summary ready to be published once the main article is live.
[00:05:27] So I think we could probably see that happen in the next few years.
[00:05:30] And there was a big topic was on systematic literature reviews.
[00:05:33] And again, these would require like a custom model
[00:05:36] because there's a lot of investment required to kind of train these, update these,
[00:05:39] but the potential for kind of time saving for them is quite enormous.
[00:05:42] And there's already a number of companies trying to develop kind of off the shelf systems
[00:05:47] for SLR screening and kind of quality control as well.
[00:05:51] So definitely an avenue for proprietary models in the future, I would say.
[00:05:54] Great. Thank you, John.
[00:05:56] Of course, journal editors have a unique perspective on the use of AI in MedComs.
[00:06:01] Kelly, could you share your thoughts on the pros and cons of AI as it currently stands?
[00:06:07] Yeah, absolutely.
[00:06:08] I think the main pro of generative AI or I'll probably refer to it as GenAI,
[00:06:13] in particular to the medical publications community on our audiences,
[00:06:17] is that it helps increase equity and health literacy.
[00:06:20] It makes research more accessible to more people in a variety of ways.
[00:06:24] It also increases the quality of research as well as its discoverability.
[00:06:29] Overall, the biggest and most acknowledged con obviously is the hallucinations that GenAI produces
[00:06:35] that create false and inaccurate data.
[00:06:37] GenAI is really still in its infancy, as John noted, and it's not really that intelligent yet.
[00:06:43] It's only as good as the information that's being fed into the large language models
[00:06:47] and even securing that information is resulting in copyright infringement and data appropriation.
[00:06:53] When discussing the pros and cons, I like to break it down into three categories.
[00:06:57] Content development, the publication process, and research accessibility.
[00:07:02] With regard to pros and content development, GenAI can provide assistance for a number of things.
[00:07:07] Data collection and analysis, writing, figure generation.
[00:07:11] The most common example here in this category is for systematic literature reviews.
[00:07:16] It can also help with journal identification and selection.
[00:07:20] During the publication process, GenAI powers tools that detect plagiarism,
[00:07:24] fabricated research from paper mills, and image manipulation.
[00:07:28] Dimensions has a tool that can identify text that indicates the integrity and reproducibility of research.
[00:07:34] Finally, for research accessibility, one of the other popular uses that we've already talked about
[00:07:39] is in writing plain language summaries and making research accessible to a much wider group of people.
[00:07:44] It also provides really improved translation over traditional AI because it's continuously learning
[00:07:49] and adapting language to allow relevant up-to-date translations.
[00:07:53] That's the highlights of some of the pros.
[00:07:55] Then when we come to the cons, that's kind of the flip side of the same coin.
[00:07:59] In content development, it's also being used to create research images, citations, other content,
[00:08:05] but this is where it's being fabricated and being used by paper mills,
[00:08:08] and we've seen a huge increase in paper mill papers being submitted to our journals.
[00:08:12] For authors, editors, and peer reviewers, another con is how easily confidentiality and patient privacy can be breached
[00:08:19] if we're not using proprietary models and using LLM or GenAI that's available to the public.
[00:08:25] And when it comes to readers, because it's a digital resource for regions that lack the infrastructure to support it,
[00:08:31] we're going to just continue seeing a widening information gap.
[00:08:35] So I believe we're at a place now where we're determining the many ways that GenAI can assist us in our work
[00:08:41] and benefit our authors and readers, but we as humans need to remember that it's for assistance.
[00:08:46] We still need to write our original content.
[00:08:48] We always fact check it and validate data that is generated by GenAI,
[00:08:53] and then be careful about the AI tools that we're using for confidentiality and private information.
[00:08:59] Yeah. And then the rapid progression of AI tools makes developing guidance particularly difficult.
[00:09:05] We also know the divergence in guidance between publishers who are more open to AI use in publications
[00:09:11] and those who are more hesitant.
[00:09:13] What do you think about the current policies on AI use in publications?
[00:09:18] When you look at all the publisher policies on AI, they do vary quite a bit.
[00:09:23] But there's two consistent messages we're seeing for medical research.
[00:09:27] One is that publishers are consistently prohibiting GenAI from being an author,
[00:09:32] mainly because it cannot take responsibility for the content created,
[00:09:36] which is one of the standards of ICMJE criteria for authorship.
[00:09:41] And ICMJE does specify that AI cannot be named as an author or co-author.
[00:09:47] The other thing that we're seeing across all the guidelines is that publishers are encouraging disclosure of GenAI use,
[00:09:54] and that also aligns with ICMJE recommendations as well as the ISMAP position statement.
[00:10:00] However, there isn't a consistent template for how this information should be disclosed.
[00:10:05] I think Elsevier has come out with one, but across publishers there's no consistent guidance.
[00:10:10] So this is an area where we need to find better consensus as we move forward.
[00:10:14] Because of the current differences in publisher policies, I would just really recommend to authors and their teams
[00:10:20] that if you have a target journal, to check with that publisher's guidance
[00:10:23] and make sure you understand how you can use GenAI because it does differ.
[00:10:28] And what you don't want to use is start using it and then have it not be allowed by the journal
[00:10:33] that you're planning on submitting your manuscript to.
[00:10:35] With regard to ISMAP's guidance that came out late last fall,
[00:10:39] it's a position statement and a call to action, and it's kind of a prerequisite to more detailed
[00:10:44] and defined recommendations that their AI task force is currently working on.
[00:10:48] Their guidance focuses on several key concepts that come up in the pros and cons that we just talked about.
[00:10:54] So mainly verifying that the data generated is accurate to prevent mis- and disinformation,
[00:11:00] protecting confidentiality of business scientific and patient data,
[00:11:04] disclosing its use in content writing, creation, editing, peer review,
[00:11:08] and using it to address accessibility gaps such as using plain language and translation to further the reach of research.
[00:11:15] And then the other half of the position statement is asking us as medical publication professionals
[00:11:21] to take responsibility and accountability in our approach to AI.
[00:11:25] For example, by educating ourselves, respecting it as a legitimate academic field,
[00:11:30] and addressing misconceptions about GenAI itself.
[00:11:33] Thank you, Kelly. Thanks for sharing some insight into ISMAP's statement
[00:11:38] and as well some advice about how to kind of handle the use of AI
[00:11:44] when considering what journal you're going to publish on.
[00:11:48] Since this is, you know, AI is just rapidly evolving,
[00:11:52] we're looking forward to seeing the advancements in this area presented at the annual ISMAP meeting next month.
[00:11:58] What are your impressions of where we stand and where we need to develop in the future with AI?
[00:12:04] So I think at the moment there's a lot of interest and discussion.
[00:12:08] I think we both said that maybe there's more to be done in terms of implementation and guidance and practicalities.
[00:12:14] For me, I would like to see plain language summaries, for example,
[00:12:18] more research on beyond readability scores,
[00:12:21] like how do human reviewers rate the content generated by generative AI,
[00:12:25] looking at the accuracy of the content as well a bit more.
[00:12:28] I think on process, I would like to see how we work with authors better,
[00:12:33] for example, ensure that things are compliant or that we're meeting good publication practice guidelines.
[00:12:39] For example, if we've got a PLS that's generated by an AI tool,
[00:12:43] what stages are you getting authors involved? How compliant is that?
[00:12:47] I think it goes back to what Kelly was saying about ensuring we're kind of meeting those guidance points.
[00:12:52] And when I was finishing up the ISMAT meeting, the European meeting,
[00:12:56] I was looking on PubMed and there's a lot of research and it's so cutting edge.
[00:13:01] It's very, very technical and there's a lot of SLR development and how they train models and things.
[00:13:05] And I think that's got to filter down to us who are maybe doing the more practical side of things.
[00:13:10] So I think that filtering down of the real cutting edge of research
[00:13:14] and how they're developing the models and how they can be implemented is something that,
[00:13:18] as someone who's not an AI expert, I'd love to see that.
[00:13:22] So maybe more engagement with the kind of the tech leaders to kind of translate it to our field would be a bit more useful too.
[00:13:29] From the publisher perspective, I was talking to a colleague last week and preparing for this podcast
[00:13:35] and he gave me this really great analogy about kind of the current state of gen AI.
[00:13:40] He says we're having a Sputnik moment. Sputnik went out into space and it changed how the world looked at space travel.
[00:13:47] Even though the technology for the space race was really lagging and in its infancy at that point,
[00:13:52] that event drove innovation and it created the flow of resources to support the space race.
[00:13:58] And he said that's kind of where we're at with AI at this point.
[00:14:01] Chat GPT becoming available for general public use has everybody talking about AI and looking at how we use it.
[00:14:09] But we're still in that very early stage and we're moving into that driving of innovation,
[00:14:15] improving our technology, having more resources dedicated to it in our own companies.
[00:14:20] And we're also getting to where we're addressing its major limitations.
[00:14:24] The future landscape, I think specific to publishing and publishers themselves is that we need to have more
[00:14:30] homogeneous guidelines across publishers with better consensus and better clarity.
[00:14:35] We want to have a goal that we're ensuring gen AI is assisting in the development and publication of the highest quality
[00:14:42] research that is the most useful to the most amount of people.
[00:14:46] And then also there's a huge opportunity for education from publishers on how these tools should be used
[00:14:52] to help authors, editors, peer reviewers, our readers, our own staff on how we navigate AI generated
[00:15:00] content and understanding best practices for using it and how to transparently disclose its use.
[00:15:06] And John made a great point of I'm not an AI expert either.
[00:15:10] And I feel like there is a bit of a widening gap between those that are the experts and really understand it.
[00:15:16] And there's also a bunch of people, including me, that are trying very hard to catch up.
[00:15:21] And I think we need to provide more educational opportunities to help people understand just the basics
[00:15:27] of using gen AI and prompt engineering, things like that.
[00:15:31] What are the best practices?
[00:15:33] And then finally, publishers are also because of the content we publish.
[00:15:37] We can provide that verified high quality peer reviewed content for large language model development.
[00:15:42] And that ensures that the best possible outputs will emerge on the other side.
[00:15:46] It's only natural that governments, AI corporations, companies will look to scholarly publishers for high quality,
[00:15:52] verified content.
[00:15:54] So that's my future look to the landscape for AI with publishing.
[00:16:00] Thanks for listening to Informed for medical communication professionals.
[00:16:04] Please take a minute to subscribe to the show on your favorite podcast app, inform your colleagues
[00:16:10] and rate our show highly if you liked what you heard today.
[00:16:13] Join ISMAP today to become a part of our community, to participate in our webinars
[00:16:18] and to receive instant access to exclusive tools and resources.
[00:16:22] If you're interested, just go to ismap.org.
[00:16:25] That is ISMPP.org to learn more.
[00:16:29] I'm Victoria Hart.