GLP-1 Pancreatitis Risk: What the 2023 to 2026 Evidence Actually Shows

When the 2023 headline went one way and the cohorts went the other

For two years the headline on GLP-1s and pancreatitis was Sodhi 2023. A 16-million-patient PharMetrics Plus analysis put the pancreatitis hazard ratio at 9.09 against a naltrexone-bupropion comparator, the popular press ran with it, and "Ozempic pancreatitis" became a permanent search term (Sodhi et al., JAMA 2023, PMID 37796527). What followed did not get the same coverage. A 258,238-patient TriNetX cohort of people with prior acute pancreatitis found tirzepatide users had a 6.2 percent recurrence rate, semaglutide users 11.7 percent, and non-users 40.9 percent (Nassar et al., Diabetes Metab Syndr 2024, PMID 39332263). A 969,240-patient comorbidity-free propensity-matched analysis found no excess pancreatitis between GLP-1 users and non-users at 6 months, 1 year, 3 years, or 5 years (Ayoub et al., J Clin Med 2025, PMID 39941615). A 62-trial RCT meta-analysis from Wen and colleagues reported a relative risk of 1.44 that lost statistical significance after stratifying for background diabetes medications (Wen et al., Endocrinol Diabetes Metab 2025, PMC12457091).

The shorter version: the 2023 alarm has not been replicated in the larger 2024 to 2026 active-comparator literature. The narrower clinical risk that did survive is sitting in case reports and in a handful of patient profiles where the absolute pancreatitis baseline was already elevated.

Bottom line: Across the largest 2024 to 2026 datasets, GLP-1 receptor agonists are not associated with a class-wide increase in acute pancreatitis risk in the general type 2 diabetes or obesity population. Several propensity-matched cohorts now show lower recurrence in patients with a history of pancreatitis. The residual safety concern lives in individual case reports, including one fatal necrotizing pancreatitis report on first-dose tirzepatide.

Why pancreatitis is the canonical GLP-1 safety question

Pancreatitis has been the safety question on this class since 2009, when exenatide's post-marketing reports surfaced acute pancreatitis cases that the FDA flagged in label updates. Two things drive that history. First, the pancreas is the tissue the molecule actually targets. The β-cell is where GLP-1 originally earned its place in diabetes pharmacology, and if a drug is engineered to signal in islet tissue, it is mechanistically reasonable to ask what it is doing to the surrounding exocrine pancreas. Second, acute pancreatitis is a relatively rare event in any single dataset but a common cause of GI hospitalization across the population, so a small relative risk shift becomes detectable in any cohort large enough.

The early signal was concerning enough to land on every GLP-1 label as a warning, not a contraindication. The 2013 FDA and EMA reviews concluded the data did not establish causation, and the regulatory bar at that time was a "monitor and disclose" labeling decision rather than a withdrawal. The regulatory posture has not shifted since those reviews. What has shifted is the size, scope, and methodological quality of the cohorts we use to argue about the same question.

What Sodhi 2023 actually showed

Sodhi and colleagues used the PharMetrics Plus claims database to identify new users of semaglutide (n = 613), liraglutide (n = 4,144), and naltrexone-bupropion (n = 654) for weight loss from 2006 to 2020 (Sodhi et al., JAMA 2023;330(18):1795-1797, PMID 37796527). The endpoints were biliary disease, pancreatitis, bowel obstruction, and gastroparesis. The headline pancreatitis hazard ratio was 9.09 (95% CI 1.25 to 66). Bowel obstruction was 4.22 (95% CI 1.02 to 17.4). Gastroparesis was 3.67 (95% CI 1.15 to 11.9).

Three things matter about that result. First, the confidence interval upper bound of 66 is the tell. That is a small-event analysis with too few outcomes to give a precise estimate. Second, the comparator is naltrexone-bupropion, a centrally-acting appetite drug with its own GI tolerability profile that is not the standard-of-care active comparator a clinician would actually pick from. Third, the pancreatitis comparison rested on a small absolute number of events in the semaglutide arm.

That is not a reason to dismiss the paper. It is a reason to read it as a hypothesis-generating signal that needed replication in larger cohorts with cleaner comparators. The 2024 to 2026 evidence is exactly that replication.

The Wen 2025 meta-analysis: 62 RCTs, 66,232 patients

Wen and colleagues published the largest RCT-only meta-analysis of GLP-1 pancreatitis risk in Endocrinology, Diabetes and Metabolism in 2025 (Wen et al., 2025, DOI 10.1002/edm2.70113, PMC12457091). The synthesis included 62 randomized controlled trials totaling 66,232 patients, mean age 58.3 years, mean follow-up 43.5 weeks, covering dulaglutide, exenatide, liraglutide, semaglutide, beinaglutide, retatrutide, and tirzepatide.

The headline pooled result was a relative risk of 1.44 (95% CI 1.09 to 1.89) for pancreatitis on GLP-1 versus comparator. That is a real signal at the top line and the reason this paper is the strongest argument for the cautious read on this question.

What most readers miss is the subgroup result. The 1.44 relative risk lost statistical significance after the authors stratified the analysis by background medications. That stratification matters because patients on a GLP-1 in these trials were typically continuing other diabetes drugs (metformin, sulfonylureas, DPP-4 inhibitors), some of which carry their own pancreatitis associations. The pancreatic cancer pooled estimate in the same analysis was 1.30 (95% CI 0.86 to 1.97), crossing one.

Read together with the cohort literature below, the Wen meta-analysis frames the pancreatitis question as a small, possibly real, but not robustly demonstrable signal that does not survive cleaner stratification. It does not deliver a clean acquittal, which is what makes the cohort evidence next.

The TriNetX cohorts that reframed the question

The cohort evidence base is dominated by TriNetX-platform analyses. TriNetX is a federated electronic health record dataset that pools de-identified records from health systems across multiple countries. Three TriNetX papers carry the most signal on this question, all propensity-matched and all running active comparators.

Recurrent acute pancreatitis in patients with prior AP. Nassar and colleagues at the University at Buffalo ran a propensity-matched five-year retrospective on 258,238 individuals with type 2 diabetes or obesity and a prior episode of acute pancreatitis (Nassar et al., Diabetes Metab Syndr 2024;18(9):103116, PMID 39332263). The recurrence rate was 6.2 percent on tirzepatide, 11.7 percent on semaglutide, 13.6 percent on dulaglutide, 27 percent on exenatide, and 40.9 percent on non-users. The recurrence numbers move in the opposite direction from what the 2023 alarm predicted.

The 672,069-patient confirmation. The same Buffalo group ran a parallel TriNetX analysis on 672,069 patients with type 2 diabetes and a history of acute pancreatitis, comparing GLP-1 receptor agonists against SGLT2 inhibitors and DPP-4 inhibitors (Nassar et al., Diabetes Metab 2025;51(2):101613, PMID 39826595). GLP-1 receptor agonists showed lower AP recurrence at every time point from one to five years. The active-comparator design is the methodological upgrade: SGLT2 and DPP-4 are the metabolic alternatives a prescriber would actually consider, not insulin or naltrexone-bupropion. The relative protection persisted in those head-to-head comparisons.

Comorbidity-free general T2D population. Ayoub and colleagues at Charleston Area Medical Center propensity-matched 969,240 type 2 diabetes patients with 9.7 percent on GLP-1 receptor agonists (Ayoub et al., J Clin Med 2025;14(3):944, PMID 39941615). After matching 81,872 pairs and explicitly excluding patients with gallstones, alcohol use, hypertriglyceridemia, and other known pancreatitis risk factors, the pancreatitis rate at 6 months was 0.1 percent in both groups, at 1 year 0.1 versus 0.2 percent, at 3 years 0.2 versus 0.3 percent, and at 5 years 0.3 versus 0.4 percent. All four time points showed numerically lower pancreatitis on GLP-1, not higher.

Multicenter confirmation in T2D patients who developed AP. A 2025 multicenter analysis using the same TriNetX backbone reported that GLP-1 receptor agonist use did not increase acute pancreatitis risk and was associated with lower complications and lower all-cause mortality in T2D patients who developed AP (2025 multicenter analysis, PMID 40358430).

The methodological point is that TriNetX-platform studies of this question converge on the same direction. The Sodhi 2023 result has not been replicated in larger active-comparator real-world data. The patients more likely to develop acute pancreatitis in the first place do not get more pancreatitis on a GLP-1, and in several cohorts they get less.

Evidence hierarchy: where each study sits

The pattern is consistent. The trial-level meta-analytic signal is weak and condition-dependent. The active-comparator real-world signal is null or favorable. The pharmacovigilance and case-report literature catches individual severe events that the population-level statistics cannot.

What the case reports actually tell us

Grennan and colleagues at the Endocrine Society published the most-cited individual event in the 2025 to 2026 literature in JCEM Case Reports (Grennan et al., JCEM Case Reports 2025, PMID 40270999). The patient was a 64-year-old female with class 1 obesity and hyperlipidemia, no type 2 diabetes, and no prior pancreatitis. She received her first 2.5 mg tirzepatide dose for weight loss. Four days later she presented with severe abdominal pain. CT showed acute pancreatitis with hypoperfusion of multiple organs, escalating within 24 hours to severe necrotizing pancreatitis and acute hypoxic respiratory failure requiring intubation. She died.

A single case report does not establish population-level causation. What it does establish is that individual catastrophic events can happen on first-dose tirzepatide, the timing can be very early in therapy, and the absence of obvious risk factors (no diabetes, no prior AP, no gallstones at presentation) does not eliminate the possibility. Class-level safety statistics work in aggregate. They do not protect any individual patient.

The Guo 2024 FAERS pharmacovigilance analysis catalogued 39 case reports across the GLP-1 class and found liraglutide accounted for 48.7 percent of those reports, dulaglutide for 23.1 percent, exenatide for 10.3 percent, and semaglutide for 10.3 percent (Guo et al., Front Pharmacol 2024, PMID 39605914). FAERS is a passive surveillance system with well-known reporting biases (older drugs accumulate more reports, severe events are over-reported, and prescription volumes are not in the denominator). It is useful for hypothesis-generation, not incidence estimation. The proportional pattern in FAERS is not a statement about which GLP-1 is safest.

Who should actually think harder about this

The honest read on the population data is that pancreatitis risk on a modern GLP-1 is not elevated for the average user. The patients where the calculation looks different fall into specific buckets:

• Prior idiopathic acute pancreatitis without an identified trigger. The TriNetX cohorts show lower recurrence on GLP-1 than off, but those analyses lump idiopathic and non-idiopathic recurrence together. A patient whose original AP cause was never identified should have the GLP-1 decision made by the clinician who managed the original event.
• Active or recent biliary disease. GLP-1-driven rapid weight loss increases symptomatic gallstone events, and gallstone disease is the single most common cause of acute pancreatitis. The cancer-risk discussion overlaps here: the Wang 2024 gallbladder cancer signal is favorable, but the gallstone-associated pancreatitis pathway is mechanistically separate.
• Severe hypertriglyceridemia. Triglyceride-driven pancreatitis is rare overall, but a baseline triglyceride above 1,000 mg/dL is its own pancreatitis trigger independent of any drug. GLP-1 receptor agonists improve triglycerides on average, but the patient-level decision belongs with the prescriber, not the population estimate.
• Chronic pancreatitis or prior pancreatic cancer. These patients are excluded from most trials, and the active comparator cohort literature does not extend cleanly into them. The benefit-risk discussion in those populations should be individualized.
• First dose to first month. The Grennan case report and the temporal pattern in pharmacovigilance data are weighted toward early-therapy events. Symptom literacy in that window matters more than at month six.

For everyone else, the modern evidence base does not support delaying or declining a GLP-1 over pancreatitis risk. The discussion looks similar to the parallel safety conversations we covered in the GLP-1 cancer risk synthesis, the vision loss NAION review, and the mental health 2024-2026 evidence: a once-loud signal that softened as the comparator-controlled cohorts grew.

Symptoms that should stop a dose

The clinical picture of acute pancreatitis on a GLP-1 is the same as acute pancreatitis off one. The signs that should trigger a same-day medical evaluation:

• Severe epigastric pain that radiates straight through to the back
• Pain that worsens within an hour of eating
• Nausea or vomiting that does not resolve over several hours
• Fever
• Abdominal tenderness with guarding on light palpation

Mild nausea during titration is not pancreatitis. The class causes nausea in 15 to 25 percent of new users, and most of it resolves over the first four to six weeks. Severe, sudden, central epigastric pain that does not match the gut-rest nausea pattern is the symptom worth taking seriously. If pancreatitis is on the differential, the correct response is to hold the next dose and call the prescriber, not to push through.

Warning: Acute pancreatitis is a clinical emergency. If you develop sudden severe central abdominal pain on a GLP-1, stop dosing, do not take the next scheduled injection, and seek immediate medical evaluation. Population-level statistics do not protect any individual patient against an individual episode.

What the 2026 evidence base means for the prescription decision

The pancreatitis question on semaglutide and tirzepatide in 2026 looks closer to the thyroid cancer story than to the NAION story. The early signal that earned the warning has not been validated in the largest active-comparator cohorts. The propensity-matched evidence points in the opposite direction in the population most likely to be at risk. The residual safety concern lives in individual case reports rather than class-level epidemiology. Population-level data does not rule out individual pancreatitis events on a GLP-1. What it does say is that the comparator-controlled prior in 2026 sits at or below the active-comparator class rate, not above it.

For the average obesity or type 2 diabetes patient, the modern population data does not support declining a GLP-1 on pancreatitis grounds. The reasonable default in 2026 is matching the molecule to the metabolic goal, monitoring for warning symptoms, and acting on them immediately if they appear.

Looking at the broader safety conversation that frames this one, our GLP-1 hair loss evidence audit, bone density and fracture risk review, and pre-surgery aspiration risk guide cover the parallel non-pancreatic safety questions. The thyroid cancer evidence audit covers the closest analogue, an early rodent-derived warning that did not survive larger human follow-up. Our semaglutide page and tirzepatide page carry the compound-specific dosing and side effect summaries. The tirzepatide vs semaglutide SURMOUNT-5 head-to-head review frames the molecule selection question on its actual efficacy axis.

Researchers exploring GLP-1 protocols can source semaglutide and tirzepatide for research-only use from Ascension Peptides with 50% off using code ENHANCED.

Disclaimer

This article is for research and educational purposes only. It is not medical advice and is not a substitute for consultation with a qualified clinician. Acute pancreatitis is a serious medical condition. Anyone experiencing symptoms suggestive of acute pancreatitis should seek immediate medical care regardless of whether they are taking a GLP-1 receptor agonist. The studies cited are real and published; the pancreatitis evidence base for this class continues to evolve. Anyone with a personal history of pancreatitis, chronic pancreatitis, pancreatic cancer, severe hypertriglyceridemia, or active gallbladder disease should discuss GLP-1 therapy with their prescribing clinician before starting, switching, or stopping.