A scan of the FDA's adverse-event database flagged 308 drug-pair combinations that looked like they raised the risk of pancreatitis in people taking GLP-1 or related diabetes drugs. After a stricter test, none of them held up.

That is the result of a pharmacovigilance study ↗ published June 10 in Frontiers in Drug Safety and Regulation, and the more useful finding is the method, not the molecules.

The old fear

Pancreatitis, a sudden inflammation of the pancreas, has shadowed the incretin drugs since the first ones reached the market. The class covers GLP-1 receptor agonists such as semaglutide (Ozempic, Wegovy ↗) and dulaglutide (Trulicity ↗), plus the older DPP-4 inhibitors, the "gliptins" such as sitagliptin (Januvia), which raise the body's own incretin levels. Regulators have watched the pancreatitis question for over a decade, because the pancreas is where these drugs do part of their work, and because early case reports raised a flag that large trials have never cleanly confirmed or dismissed.

What the new study asked is narrower and not much studied. Not whether the drugs alone raise pancreatitis risk, but whether combining them with some other common medication makes it worse. A statin? A blood-pressure pill? That is a drug-drug interaction question, and the obvious place to look is the FDA Adverse Event Reporting System, known as FAERS, the public repository for side-effect reports.

Why FAERS lies to you

FAERS is huge, voluntary, and uncontrolled. Anyone can file a report, the reports are not verified, and there is no denominator telling you how many people took the drug without incident. So the standard tool is disproportionality: you ask whether a drug and a side effect show up together more often than chance would predict across the whole database. That comparison produces a number called the reporting odds ratio. A high ratio is a signal, not a fact. It can be created by reporting bias, by the disease itself, or by the simple truth that sicker people take more drugs.

Run that analysis across every pairing of an incretin drug with another medication, and 308 pairs lit up as positive signals for acute pancreatitis. Taken at face value, that is 308 reasons to worry.

The filter

The authors did not take it at face value. Using a conservative signal measure called Omega shrinkage, they split the database into six multi-year periods running from the fourth quarter of 1997 to the first quarter of 2024 and asked which signals showed up consistently, not just once in a lucky window. A real interaction should be there year after year. A statistical accident should flicker and vanish.

Of the 308, only 11 pairs held a signal across three or more of the six periods. The most persistent was dulaglutide combined with atorvastatin, a common cholesterol drug, which appeared in four of the six periods. Even that one did not clear the bar the authors set for a robust association. Their conclusion: no drug combination involving these diabetes medicines was reliably tied to extra pancreatitis risk.

Why this matters more than a null result usually does

A clean null on a scary question is worth something, but the transferable lesson is about the data source. FAERS-based disproportionality studies are cheap to run and easy to publish, and they routinely generate headlines about a drug "linked to" some alarming outcome. This study shows what happens when you make the same data prove itself over time. The 308 alarms became 11 weak ones, then zero robust ones. The signal that survives that filter is the only kind worth acting on.

For a drug class that keeps throwing off real-world associations ↗, most of which shrink or disappear under scrutiny, that discipline is the whole game. The drugs may yet carry a small pancreatitis risk on their own, and the large trials are the place to settle that. What this analysis rules out is the next layer of fear, that some everyday companion pill quietly multiplies it.