Generative AI Transforms Clinical Study Report Development

This presentation was created based on an article by John April, Senior Director, Adaptive Content Strategy, Global Scientific Communications, Eli Lilly and Company; Inger Ødum Nielsen, Team Lead, Content Management Team, Clinical Reporting at Novo Nordisk; and Vanessa de Langsdorff, Vice President, Customer Success at Yseop.

Generative artificial intelligence (AI) has absolutely exploded onto the scene. It's become a top strategic priority in the life sciences, and honestly, for good reason. It holds this incredible promise to dramatically speed up drug development and really shake up the regulatory processes we've known for years. Let's dive into this.

Right at the heart of this whole transformation is one single critical document, the clinical study report, or CSR. These reports, they're the absolute backbone of regulatory submissions, and generative AI is set to completely change the game for how they're created, cutting down on manual work, and we hope, shortening those timelines. But this leads to a really fascinating and kind of frustrating puzzle. With all the tech we have now, with all the advances we've made, why has the time it takes to produce these crucial reports pretty much stayed the same? For over a decade, it's a question that really gets to the core of the problem here. Back in 2013 a CSR took somewhere between 8-15 weeks to get done. Now you think that would have improved, right? Well, a recent survey shows that today, that ranges between 6-15 weeks. We've seen a decade of innovation with almost no change in the final output. Something is clearly broken.

This brings us right to the main point. The real bottleneck isn't the technology we're using, it's the process itself. If you just automate a flawed, inefficient process, you don't actually fix it, you just get faster at creating content that's still going to need weeks, maybe even months, of painstaking manual review. This really highlights a massive risk. If you take an AI and you train it on a bloated, inefficient process, what's it going to learn? It's just going to learn to replicate those same bad habits, so you won't get true efficiency. You'll just get faster at being inefficient, and that's not the goal at all, is it?

To really get this right, organizations have to focus on two fundamental pillars before they even think about deploying AI at scale. First up is data readiness. Think of that as the solid foundation you're building on. And second is content readiness. That's the framework that shapes what the AI actually creates.

Let's dig into that first pillar, data readiness. You can think of this as the high-quality fuel that any successful AI project absolutely has to run on. Here's where we hit our first major snag. While our patient level data often follows really clear industry standards like Study Data Tabulation and Analysis Data models, the summary level data, the stuff that's essential for building these CSRs is often just all over the place. It can be wildly inconsistent from study to study across different therapeutic areas, even within the same company, and all that inconsistency, it just forces a bunch of extra complicated steps before any AI can be reliably put to work.

Once you've got that data foundation solid, we can move on to the second pillar, which is just as critical, getting the content itself prepped and ready for automation. This is basically like cleaning the house before you let the robots in to help. Optimizing your content really boils down to three key tasks. First, you've got to clearly define the scope of each document, and then come the two big ones. You have to systematically get rid of all that subjective, opinion-based content, and then you have to hunt down and eliminate redundant, repeated information.

Let's just break down the difference here. Objective content is all about the facts. It describes pre specified outcomes. This is what preserves scientific integrity and lets you compare apples to apples across different studies. Subjective content, on the other hand, that reflects opinions, interpretations, and it introduces potential bias and creates all these inconsistencies that regulators then have to spend time sorting through. For AI to work, and frankly, for good science, the focus has to be on objective, fact-based reporting.

Now let's talk about redundancy. This is truly the silent killer of efficiency when you're creating documents. Think about it, every single time information is repeated, you're creating a risk for transcription errors, and you're adding to the massive burden of manual review and all those painful accuracy checks. It's just waste, plain and simple. This really highlights a fundamental shift we need in our thinking. The old way was to just reuse content; rewriting protocol details into the CSR or copying and pasting numbers from a table right into the text. The new, much smarter way is to refer. Instead of rewriting everything, just link to the protocol appendix. Instead of repeating data, just refer the reviewer directly to the validated tables and figures. It's cleaner, it's safer, and it's so much faster.

What does this all lead to? When you get this right for the CSR, it creates this amazing positive ripple effect. It's not just about improving one document. It's about transforming the entire content ecosystem for a whole regulatory submission. In this ideal ecosystem, every single part of the common technical document, the common technical document, has a crystal-clear purpose. The CSRs, they serve as the objective data foundation, just the facts. The module two summaries then use that data to weave the narrative story, and finally, the module two overviews provide that crucial analysis and interpretation. There's no overlap, no redundancy, just a clean, logical flow of information.

The benefits here are huge, and they go both ways. For the authoring teams, this approach saves an incredible amount of time by just getting rid of duplication, and for the regulatory reviewers, their experience is dramatically better. They get a submission that's clearer, more logical, and so much easier to navigate. It's a total win-win.

All of this brings us to the real value of AI in this space, and it's so much more profound than just making things go a little faster. At the end of the day, the real story here isn't just about speed. The greatest value of generative AI is that it's a catalyst. It's forcing the entire industry to stop and fundamentally rethink and redesign these outdated processes, making them leaner, more consistent, and ultimately a lot more valuable for everybody involved.

The path forward is pretty clear. To really unlock what AI can do, professionals in clinical operations and regulatory affairs need to standardize their data, streamline their content, fully embrace objective reporting, and build a culture that actually values efficiency over those old, outdated habits. Generative AI isn't coming, it's already here, so the real question for every organization isn't if you're going to adopt it, it's whether your foundational processes are actually ready for the transformation that's about to happen.