CTTC Opens With A Reality Check On AI

By Dan Schell, Chief Editor, Clinical Leader

When I walked into the opening plenary of the Clinical Trials Technology Congress (CTTC) in London, I expected the usual AI conversations. New capabilities. New platforms. Predictions about where the technology is headed. Blah, blah, blah. So, you can imagine how relieved I was when I realized we weren't getting the same old messaging and presentations.

Sure, everyone talked about AI, but the discussion quickly moved past the familiar predictions that "AI will transform clinical trials!" Most of the speakers seemed to assume we'd already accepted that. (I mean, this was a clinical trial technology conference, after all.)

But, instead of extolling the miracles of AI, the presenters spent their time discussing what organizations have to do before AI delivers any type of meaningful value. (Cue me sitting up in my seat to pay closer attention.)

Over the course of three presentations, the discussion kept returning to the same challenges. Organizations still have to connect data scattered across hundreds of systems. Trial teams have to trust new workflows. Leaders have to help employees adopt technologies that may initially feel like one more thing added to an already overwhelming workload. And throughout it all, humans remain responsible for the decisions.

By the time the opening session ended, I realized I'd spent very little time thinking about AI itself. Instead, I was thinking about the people who will ultimately determine whether AI succeeds or fails inside their organizations.

It’s not AI … It’s You

Let’s start with that people problem. AI’s ultimate success in clinical trials is — at least partially — dependent upon everyone agreeing to change the way they work in certain situations. I know I added a few caveats to that statement, but they are important, and frankly, you know why.

Moderator Adama Ibrahim (former VP, R&ED Digital Transformation, DS&I, Novo Nordisk) started us off by stating she feels we don't have a problem with data, technology … or even people. “Instead, what we have is this new era of generative AI where AI is connecting us to systems that aren’t connected. So, in my view, we actually have a ‘decision latency’ problem, which, I'm sure, is a term you've all heard of. We need to shift to building more decision orchestration across systems.” Considering I had never heard of “decision latency,” I was all ears as to how this related to the bigger issue of the human’s role in AI use.

During his main presentation, Johannes Streffer, SVP of global clinical development at Lundbeck, covered a lot of disparate issues. One that stuck with me was his point that AI document creation is obvious opportunity. Just as people increasingly rely on ChatGPT to draft emails, Streffer believes sponsors should use AI to help generate and revise regulatory and clinical documents. At the same time, he noted that regulators and auditors are also beginning to use AI, meaning submissions will likely face more sophisticated scrutiny than ever before.

And, while AI ultimately reduces workload, he noted that organizations must first overcome the learning curve that makes adoption seem burdensome. “I find it very interesting that, last year, when we would talk about AI, we often were saying, ‘This is fantastic. What all can we do with it?’ But now, we're saying, ‘Oh, fantastic. How do we make sure people don't see this just as an extra burden and then not even adopt it?’”

That places as much responsibility on leadership as it does on individual users. Technical experts remain accountable for using AI appropriately, but leaders are responsible for creating an environment where new tools are supported, properly implemented, and viewed as genuine enablers rather than just another task added to an already full workload.

Presenter Jenn Embury, director of clinical and regulatory strategic services at Astrix, stayed on point when she argued that technology projects rarely fail because software doesn't work — they fail because people don’t adopt it. Using childbirth as an analogy, she stressed that trust, communication, and leadership matter as much as technology. "Training is not change management," she said. During the panel discussion she joked that users often admit, "Don't tell my boss, but I don't use that. I still use Excel," drawing laughs while highlighting a very real adoption challenge.

Embury went long why she believes AI education should begin well before people enter the workforce. Rather than discouraging students from using AI, she argues educators should teach them how to use it responsibly, critically evaluate its output, and still develop their own voice. "But, I don't fear that [kind of AI education] as much as I fear how folks that are already in our industry need to adopt and change to new ways of working with AI."

The AI Challenge Starts With Connected Data

Valdemar Borum Svarrer, head of strategic & digital development, data science, development at Novo Nordisk, took a different approach during his presentation, focusing on connecting data rather than accumulating more of it. When he stepped into his role, Novo Nordisk’s clinical development organization was working across roughly 600 tools and systems, and his team’s mission became to “make the complex simple” by building an integrated clinical data ecosystem that links study design, preparation, data collection, monitoring, analysis, and submission around connected systems and a single source of truth. That work includes reusing decades of historical clinical trial data through what he called “found data,” spanning more than 350 cardiometabolic studies, about 243,000 patient-years, and 28 billion data points, with AI helping users find, understand, and apply the information. But Svarrer emphasized that data usefulness depends on trust: Data must be harmonized before it can support reliable analysis, as illustrated by a pregnancy-related trial question that had been asked 37 different ways over 20 years. He also described the clinical trial protocol as a “holy grail” document for AI, highlighting how AI assistants can compare historical protocols, identify endpoint tables, trace outputs back to source material, and deliver value through practical, time-saving tasks. Not exactly flashy, but still super useful.

In fact, “super useful” is kind of how I felt after a lot of these CTTC sessions. Like I said, this show delivered a message I didn’t expect — here’s what organizations have to do before AI delivers on any promises. Better data. Better workflows. Better leadership. Better adoption. None of those topics generate the same excitement as the latest AI breakthrough, but after listening to these speakers, I'm convinced they're the issues that will determine which organizations actually realize AI’s potential in clinical trials.

Editor’s note: Yes, there are em dashes in this article. That doesn’t mean AI wrote it. I happen to love em dashes and have used them for nearly 30 years in my writing. I did, however, use AI to analyze some common issues discussed in this very long transcript, and then I used that “data” to help me organize how I wanted this piece to flow. So, this isn’t “AI slop,” it’s “Dan Slop.” Enjoy!