Episode 427: Dr. Jan Herzhoff, President of Elsevier Health, Implants AI into Medical Research

Thanks so much for having me here, Josh. This is my first visit to the NYSE. Thoughts as I entered the building? I love it. It's a very impressive building. I like the small items – the basketball, the cars. It's a very special place. We need to put some issues of the Lancet downstairs. What would be the iconic thing to represent RELX? Clearly, one from the Lancet would be amazing, probably one of the recent articles around COVID that triggered the whole development of the COVID vaccine. That could be one. How often does the Lancet publish? I think it's on a weekly basis.

Well, that's a good question. Our team at the Lancet is based in London. I don't interact that much with them, but they always bring up very interesting content. You gave a quick overview of the company in the intro, but delve a little deeper. Beyond what comes to mind – the Lancet or the Journal of Pediatrics – what is Elsevier Health? When you look at Elsevier Health as part of Elsevier, our mission is to support students, faculty, and clinicians to be the best version of themselves to improve patient outcomes. We do that throughout the life journey of a clinician, starting with the first book they would have – Gray's Anatomy, for example – and moving into practice. We help them create a scaffolding over the years to become a very good physician or nurse. When they move into practice, we help them make sense of the latest medical knowledge, which accelerates at a rapid pace. We help them bring evidence-based knowledge to the point of care, and finally help with patient communication. There is my dad's edition of Gray's Anatomy somewhere in my mom's house. It has to be one of those first tomes that an aspiring doctor gets and begins to pore through.

That's an excellent question. In the past, it started with the printing press, when documents and books were printed, moving to journal distribution. In the late 1980s, early 1990s, Elsevier created the platform ScienceDirect, which was the first database for medical insights. From there, we moved into clinical decision support, into workflows, and now most recently into artificial intelligence.

That's a great question. When you look at the needs per country, various differences stand out. It can be the language, drug information (FDA approved here vs. other agencies), different metric systems, or different evidence based on environmental factors. Our local teams in all major geographies localize the content and look at the latest evidence. For example, our team in Japan many years ago worked with the former head doctor of the former emperor of Japan to create a network of top 1,000 physicians, producing evidence-based content now used by most physicians in Japan.

One interesting point in a business like ours is how you scale certain ideas and products. You need to understand different cultures and how countries operate – the difference between developing, emerging, and established countries. This variety of experience helps you see links you would otherwise not see.

For me, personally, it developed much more over time, and actually even more within Elsevier. One of the reasons I moved to Elsevier was its global footprint, operating in many different countries and cultures. I learned this here in this company.

In the division I lead, which covers all our health businesses, there are different groups. We have a content organization that engages with our authors, a technology organization that works with product teams and customers to co-create products, and a sales and marketing organization that thinks about go-to-market approaches and scaling. It's a broad range of teams and expertise. What's really interesting is that the people are here because of the strong purpose of the company – to help medical students, nursing students, faculty, and future doctors be the best version of themselves.

I think very similar to what you shared about your dad. I was born in a small village in rural Germany, just north of Frankfurt. My family comes from a background of health professionals and entrepreneurs. At the dinner table, discussions were always about the healthcare system. My dad is a pharmacist in Germany, and it was always a big topic. That's where it all started.

It all started with a startup with a friend during school, where we worked on solutions for pharmacies in Germany, linking technology with healthcare. At LSE, they looked at technology through a social lens – thinking about how technology links up with people. In healthcare, it's about how a clinician can bring their expertise combined with new technology to help patients. Also, the focus on large-scale infrastructures helped me think about creating platforms that support our customer groups.

We have used AI for more than a decade in many of our solutions, especially for data extraction and information extraction in our operations. What is different now is the element of generating answers using generative AI. But we have been using AI for a very long time.

One very big differentiator, especially in healthcare, is the evidence-based, high-quality content. If you don't combine technological solutions with the highest quality content, it becomes an empty shell. We curate that content over many years and build networks with physicians globally. This content powers our platforms. For example, textbooks like Gray's Anatomy have moved from print to databases to 3D. Now, in the world of large language models, these textbooks are incredibly valuable for providing answers and training models because of their structure. We have the biggest medical content base of this structured content, which sets us apart.

A couple of interesting things. RELX has different verticals, including our sister company LexisNexis, which launched Lexis+ AI about a year ago. We learn from each other constantly, sharing insights about platform creation, AI products, go-to-market approaches, costs, revenue, and pricing models. This helps the whole group move much faster. We have links on engineering, product, and go-to-market sides, with regular exchanges. There are many parallels between legal search and medical search, so we do a lot of exchanges on multiple levels.

I have a long-term relationship with AI. The funniest example was in one of my first lectures at LSE about neural networks and genetic algorithms – that lecture is where I met my future wife. Over the years, the critical moment was when ChatGPT launched. We always thought about how to improve search for clinicians to find the right information in a huge ocean of information. The aha moment was that this technology could help us find the needle in the haystack. With current solutions, simple questions took five to ten minutes to answer; 30% of medium-hard questions would never be answered. With vector search powered by large language models, we can find information much faster. That was the aha moment, early last year.

I would consider a physician more as a detective. They have a situation and need to look at all the evidence, including the patient's vitals and interactions. AI is extremely effective in helping retrieve information from databases. But at the end of the day, the physician must bring all these pieces together and assess them. We believe it's not the right approach to replace; with responsible AI, we can enhance the capability of doctors and nurses, making them much more effective. They play the absolute critical role.

So starting with the problem we are trying to solve: the problem we're trying to solve is to help the physician in a moment when they need to figure out an answer or have a specific question in mind, and to figure out an answer to that question quickly. This is exactly when you have a very complex medical case as well, where you need to have follow-up questions. That's exactly where Clinical Key AI shines. So what we are doing here is we have created this platform where we combine our content with a cutting-edge vector search, together with a partner startup called Open Evidence. We have our curated content. The clinician can ask the question in normal conversational tone. We ensure that the content that has been initially indexed is only curated content so it's not coming from public internet sources, to ensure it's trusted. All the answers are then based on that content from journal articles or clinical overviews that have been created. The vector search finds these content pieces, these specific paragraphs, and places them back. These are all from trusted sources that physicians would use and have learned from over decades, based on textbooks or journals they have been using. It is trusted content. Specific paragraphs are directly linked to the deep content so you can always go back to the evidence and double check that it's accurate. In addition, we have an evaluation framework where we constantly evaluate that the results from Clinical Key AI are accurate. If we see any challenges or issues, we can immediately intervene.

So there are two separate topics here. One is about increasing overall capacity in the system for both physicians and nurses. One big challenge especially here in the United States is we don't have enough nurses. There's a nursing shortage. It's actually crazy when you think about it: there are enough candidates who would like to join a nursing program, but they can't be supported because there aren't enough faculty or enough clinical placements for them. It's all about how we can help nursing schools or medical schools to increase the number of placements and also have the faculty there to support that. There we are actually helping quite a lot, with specific AI products that support this educational use case. On the clinical side, it's similar to when you think about your own day-to-day work: when you use your normal tools, you want them to work, you don't want frustrations. If the computer shuts down or whatever happens, that creates frustration. We want to remove all of that. One big thing is when you search and it takes a long time and you don't find the answer, that creates additional anxiety and pressure. We want to remove that. That's one of the ways we help to reduce that. But we have a lot of other things we are doing in this space. This is especially for generative AI where we work on the press release that was distributed when Elsevier Health launched its global launch of Elsevier Health and Open Evidence.

So we have various design partners we work with. We are constantly getting feedback from them about the design of the product, the responses to the questions, the accuracy level. All of that helped initially in that very first step. Specifically around the 30,000 testers, it was all about the underlying technology, the search technology, where they provided feedback and input that it actually works and is highly effective. That was one of the key insights we gained earlier last year when we made the decision around partnering with Open Evidence. We tested it ourselves, we tested our own approach, we tested the Open Evidence approach, and we tested a combined approach with both. The accuracy level was so high that we believe okay, this is now ready to go for prime time.

So the underlying content of Clinical Key AI comes from our journals, our books, many peer-reviewed, many peer-reviewed content from clinical overviews that have been curated over a very long time. It takes an incredible amount of effort to get to that level of quality. Usually we have people who check on that. It's a very sophisticated process, similar to all the other content we have. It also doesn't only come from Elsevier sources, but also from other publishers as well. We have the rights to use the content. So it's a very broad range of content assets, but only the highest quality content that's out there.

We co-create everything with our partners and our customers to really understand where the biggest pain points are. In the case of Clinical Key AI, the big pain point was around search and how quickly you find information. We look at what is the biggest challenge. We also always look at the big focus on monetizable innovation. We can't solve everything; we need to look at what areas have the biggest value attached from our customers. That's a good proxy for the biggest challenges they have. So we look at both combined: the problems, how big they are, the value attached. We also look at it from a global scale perspective. Is it something that only helps in one specific country or setting, or something we can scale up? One of our key capabilities is scaling from something small to global. Great example: we had a couple of products in the education space, launched them in 2018, six years ago from zero, and they are now in 1,600 institutions globally. That's our capability: to scale up something when we have found a problem, understood the value is there, and then work with our customers and our sales and marketing teams to bring it to all markets across the world.

There are multiple use cases. One very interesting one is when you are a researcher at a very early stage, you want to understand a specific problem domain, it's still new to you, and you want to understand that domain. You can use Scopus AI to get a very quick overview of the entire domain, and it would show you the most important papers, etc., and where to then advance your research into it. It's a very powerful tool for exactly that.

So both products are part of our nursing education business. What's interesting about Shadow Health is it helps the future health professional build their communication skills with the patient. It's extremely good at bringing very different diverse patient situations to the nursing students, engaging with these simulations, with these digital characters, and asking the right questions. There's a multitude of questions you can ask; it's a very free-flowing conversation you can have. When you ask these questions, the patient reacts very differently, and you have to find the path which helps you identify the right challenges or the right evidence to then think about the right treatment plan for the patient. It's a great tool, developed by a company in Florida that we acquired a few years ago.

When you're talking about the thing inside you, yeah. This is a very exciting technology, actually going much more beyond traditional simulations. When Apple launched the new Apple Vision Pro, we were one of the very first apps that were there. Enterprise customers and the press considered it one of the most advanced simulation products on this new device. What's incredible about it: you might remember the movie The Matrix. This supports you with that type of learning experience. It's about understanding a concept. You go in, you see the heart in 3D through the glasses, you engage, you understand it. When you have understood it, we push you into a completely virtual simulation where you then have to deal with a patient who has a heart situation. You learn in an incredible way. You help students engage with specific, sometimes very challenging concepts. Many years ago, they were only able to look into a 2D book to see the heart, or later in a 3D version. Now they can engage with it, and it's a much richer learning experience possible with a device that can move from an extended or mixed reality setting into a virtual spatial situation.

I think it's incredible, especially in a medical setting. Other devices we have tried in the past would not give you that type of resolution and quality needed in a medical setting. For us as a company, it's a game changer for how we can bring the latest insights and knowledge in a way that is extremely engaging and gives our customers completely new insights they didn't have in the past. You could imagine in a clinical setting, if there's a situation, you could directly push the latest journal articles specifically on that situation. It's a very powerful tool, and from our perspective, extremely fun to use.

If you asked me what would be one of the proudest moments of our entire team, of what we have created over the years, I would say this is exactly it. Just a few years ago, when we acquired Complete Anatomy in December 2009, it was basically a white male, and you would learn the anatomy on that. Most of the other solutions on the market were very similar. We put a lot of focus to ensure we had a female model and also different diverse characters to have a much better representation. What's interesting about this technology: this application, Complete Anatomy, is now used by a third of all future doctors globally. Every third future doctor is trained on this tool. In the past, it was really just one single way of representation. Now our latest data showed that of these future doctors, 77% chose a different representation that really helped them understand different anatomy, female anatomy. Also when you think about skin cancer, how it would look on a different type of skin. It's a big change we were able to drive through this offering, with global impact.

For me, one important thing is it helps them when they see a patient. For example, in the context of heart diseases, they would be sensitized to think that a female might have very different reactions to a heart disease than a male does. If you're not sensitized, you would not pick it up. Fifty percent of the population are females, so you would really risk a lot of hard cases if you don't have that. In fact, it's one of the biggest, if not the highest, reason for death across the United States: heart diseases, especially for female patients. So that's one of the things we believe we can really help to change, always together with our customers. It's a big thing.

It's a very substantial input. It's a global survey looking at different countries and geographies. One of the key insights was how open clinicians are to the use of AI as part of clinical decision support. Last year, about 7 to 8% globally of physicians were using AI technology to help with clinical decision support, but up to 50% said over the next two to three years they would actually switch to this type of technology. That helped us understand the time horizon and the share of customers who are very open to it. The other half is less open, and we need to understand the blockers. It's also interesting by geography. For example, in China, physicians are much more open to using AI technology than in the United States or the UK. So we see geographical differences, which are very helpful when thinking about resourcing and which markets to go into.

About a year ago, I attended a program at Harvard where we had Professor Tushman, who has a very interesting concept called the ambidextrous organization. This is actually one of the biggest challenges I can see for leaders in this time, especially in global organizations: how do you ensure you move fast on innovations while also driving the core and ensuring the core works really well? You need to balance it, and it's not only at the leadership level; it's across the entire organization. How we do it within Elsevier Health is we have a lot of mechanisms that bring both the innovation teams and the core teams very closely together. But it's a constant readjustment. You constantly need to look at how the team is working together, how we can ensure we move fast but not break the system. It's a big balance and a challenge that never stops for an ambidextrous organization.

For us, the big thing is ensuring we can leverage and identify the right use cases for artificial intelligence to bring our insights in a very responsible way to our customers, giving them an alternative to offerings that are out there and really helping them on their journey. That's the number one priority. It's also about how we can bring these technologies into different countries and scale them. I'm very excited about technologies like spatial computing and how we can bring very cutting-edge technology into the education space, again very responsibly. These are some of the key challenges and things that excite me a lot for the next few years: helping doctors and nurses on their professional journeys, no matter where they may be practicing around the world.

Thank you so much for having me, Josh.