What Stuck With Us at DIA 2026
We went to Philly to give a talk and came home with a notebook full of everyone else's. Here are a few things we're still thinking about.
This is a guest post from Nick Capman, President and CEO of The FDA Group.
This year, I went to DIA to present on a panel, but the honest truth about any conference is that the useful part is the rooms you sit in, not the one you talk in. So here’s what I’ve been chewing on since.
Two themes ran through almost every session. The first was AI, which surprises no one, though what struck me was how uniform the qualifiers were. Speakers kept saying the same thing in different words: AI can cut risk, add speed, and show you things you’d otherwise miss, but only alongside real data governance, validation, transparency, and someone accountable at the end.
The second theme was quieter and, I think, more important. The science is now moving faster than the systems built to handle it. That showed up in drug repurposing, GLP-1s, rare disease work, trial modernization, regulatory technology, and precision medicine. The opportunity is real here, but so are the constraints around access, affordability, trust, data quality, and regulatory alignment.
DIA framed itself this year as neutral ground where regulators, industry, academics, patients, payers, investors, and technologists can actually meet. The theme was finding hidden potential in today’s science and turning it into tomorrow’s therapies through collaboration that’s organized rather than accidental.
They also announced a couple of global efforts worth watching, a DIA and MHRA Global Summit and an Africa Access Summit.
The session I keep telling people about
David Fajgenbaum’s keynote on drug repurposing was the one that stayed with me the most. His group, Every Cure, is using AI and knowledge graphs to scan something like 75 million drug-disease pairs, built out of roughly 4,000 approved drugs and about 18,000 diseases. They’ve cut the time to run that kind of scan from around 100 days to about 17 hours.
The sort of uncomfortable part is why nobody had done it. A lot of the promising matches involve generic drugs, which often carry less commercial incentive to develop. So the treatments may already exist, sitting in plain sight, and the reason they haven’t reached patients isn’t scientific. It’s that the money doesn’t line up.
His ask to the room was practical: share your repurposing signals, contribute data, bring forward abandoned assets, and work with regulators on non-commercial pathways. I’d like to see our corner of the industry take that seriously.
GLP-1s and the mess a breakthrough exposes
The obesity plenary, on the past, present, and future of care in the age of GLP-1 therapies, did two things at once. Robert Califf called GLP-1s a genuine scientific breakthrough, with benefits reaching well beyond weight into cardiovascular outcomes and other organ systems. Then he used the same breakthrough to point at everything it exposed: high cost, supply shortages, unregulated compounding, misinformation spreading online, and access that isn’t remotely equal.
Scott Butsch of Novo Nordisk made the case that obesity is a physiological disease, not a lifestyle choice or a failure of character, and walked through how the newer therapies change the body’s biology, including its weight set point, with some benefits that seem to run past weight loss itself.
Brad Jordan of Eli Lilly stayed on access and stigma and the real danger of unapproved compounded products, and argued that industry and regulators have to deal with quality, safety, misinformation, pricing, comorbidities, and prevention together rather than one at a time.
Where AI is earning its place in development
Audrey Greenberg of Mayo Clinic Ventures moderated a panel on where AI is actually creating value in drug development and where people need to calm down. The framing I liked was AI as a biological risk reduction engine, useful for target selection, biomarker discovery, patient stratification, toxicology, drug-drug interaction analysis, and enrollment.
The nearest-term wins looked like discovery, reducing risk before you ever get to the bench, toxicology, and trial optimization.
They also got into money (which conferences usually avoid). One idea was clinical trial insurance, where investor capital gets reimbursed if a trial fails, which could free up financing for mid-stage assets that are hard to fund under normal risk math. The regulatory side covered sandboxes, data infrastructure, rare disease pathways, and AI-assisted submission review.
The panel also kept circling back to the same non-negotiables: transparency, reproducibility, third-party vetting, published evidence, and a human being who is accountable. You’ll notice that list shows up everywhere at this conference.
Quality is moving from sampling to surveillance
For years, “AI in quality” was mostly slideware and hype. This year it looked like something people are actually running now.
One of the core arguments was that traditional QA, leaning on manual sampling and point-in-time site audits, can’t keep up with the complexity and data volume of modern trials, and that E6(R3) is pushing everyone toward continuous, proactive, risk-based oversight built on analytics. The examples were specific enough to believe!
Merck runs study-level audits off analytics dashboards that watch critical-to-quality factors across every site and steer resources toward the systemic risks and the sites that need it.
Roche’s RAPID methodology uses fully remote audit packages aimed at particular risk areas, and one of them, SIM AE REP, reads adverse event reporting patterns across sites to flag likely under-reporting.
Bayer built a Quality Insights Engine that mines historical data from CTMS, QMS, 483s, and audit reports to catch recurring problems and fix protocol design before those problems happen. One case study cited a 50% drop in severe quality events!
There was a business-case point aimed squarely at quality leaders, and it’s one I’d repeat. Stop justifying analytics on financial ROI alone. The stronger frame speakers used was value of investment, which counts time saved, cost reduced, inspection readiness, data maturity, workforce development, and regulatory risk taken off the table. Those are harder to drop into a spreadsheet, and they’re the real reasons to do this.
The agencies seem to be behind the direction.
Staff from the FDA described moving toward directed inspections built around critical-to-quality factors and leaning on Remote Regulatory Assessments.
MHRA was encouraging about innovation but firm that anything new has to be fit for purpose, independent, documented, piloted, and validated before you rely on it.
How to read a 483
A separate quality panel took on Form 483s, and I wish more people had been in the room for it. The core message was that a 483 is not automatically a critical finding or a serious breach. It’s a formal communication tool, and it calls for a proportional, evidence-based response with strong root cause analysis and interpretation that accounts for context, not a company-wide panic every time one lands.