GLP-1 receptor agonists and risk of all cause and cause specific acute pancreatitis: target trial emulation

Abstract

Objectives

To assess whether initiation of glucagon-like peptide-1 receptor agonists (GLP-1RAs) rather than sulfonylureas is associated with all cause acute pancreatitis and cause specific acute pancreatitis and to characterise temporal risk patterns of the various causes of acute pancreatitis associated with GLP-1RA use.

Design

Target trial using the electronic healthcare databases of the US Department of Veterans Affairs healthcare system.

Setting

US Department of Veterans Affairs.

Participants

333 687 users of the Veterans Affairs healthcare system with type 2 diabetes initiating GLP-1RA (n=132 551) or sulfonylureas (n=201 136) between 1 January 2017 and 31 December 2023.

Exposure

Initiation of GLP-1RA or sulfonylureas, and separately, continued use of GLP-1RA or sulfonylureas during follow-up.

Main outcome measures

Risks of acute pancreatitis during 12 months’ follow-up including all cause acute pancreatitis and cause specific acute pancreatitis (eg, pancreatitis that is suspected drug induced, alcohol induced, hypertriglyceridaemia associated, biliary, and idiopathic or other cause). The risks were measured through discrete time survival models after application of inverse probability weighting.

Results

In intention-to-treat analyses, participants who initiated GLP-1RAs had similar rates of all cause pancreatitis at one year compared with participants who initiated sulfonylureas (rate difference −3.64, 95% confidence interval (CI) −30.76 to 23.48 per 100 000 persons at one year). However, the overall null finding was the net result of countervailing effects across various causes of acute pancreatitis. GLP-1RA was associated with an increased risk of suspected drug induced acute pancreatitis (23.45 (14.27 to 33.85)) and decreased risks of hypertriglyceridaemia associated (−16.96 (−27.41 to −7.34)) and alcohol induced pancreatitis (−10.32 (−18.12 to −3.17)). Similar rates of biliary and idiopathic or other causes of acute pancreatitis were observed in the two groups. In per protocol analyses, drug induced acute pancreatitis attributable to GLP-1RA clustered early (42% of GLP-1RA attributable events over one year of follow-up occurred in the first three months), whereas reductions in alcohol related acute pancreatitis concentrated between months four and six and hypertriglyceridaemia associated acute pancreatitis concentrated between months 10 and 12. Cumulatively, these divergent temporal trends resulted in a net increase in risk of all cause pancreatitis during the first two months of treatment, with similar monthly risk observed for the remainder of follow-up.

Conclusions

In this nationwide cohort, similar rates of all cause acute pancreatitis in GLP-1RA and sulfonylurea users were observed at one year. Increased risk of suspected drug induced pancreatitis during the early period was offset by later reductions in alcohol induced and hypertriglyceridaemia associated acute pancreatitis. Clinicians should counsel patients on early risk while recognising the overall neutral long term effect.

WHAT IS ALREADY KNOWN ON THIS TOPIC

• Randomised trials have not shown a significant increase in acute pancreatitis with use of glucagon-like peptide-1 receptor agonists (GLP-1RAs), but these studies may be underpowered with low event counts

• Results from pharmacovigilance and observational studies are mixed, with reports of increased, null, and reduced risks of acute pancreatitis

• GLP-1RAs may have a complex, opposing association with various causes of pancreatitis (increasing risk of acute pancreatitis from some causes and decreasing risk from other causes), thus analysing all causes of pancreatitis together can mask possible opposing effects

WHAT THIS STUDY ADDS

• GLP-1RA use was not associated with a higher one year risk of all cause acute pancreatitis

• This null finding was the net result of an increased risk of suspected drug induced pancreatitis being offset by decreased risks of alcohol induced and hypertriglyceridaemia associated pancreatitis

• The risk of drug induced pancreatitis manifests early after treatment initiation, while the reductions in alcohol related and hypertriglyceridaemia associated pancreatitis accrue later

HOW THIS STUDY MIGHT AFFECT RESEARCH, PRACTICE, OR POLICY

• Pancreatitis risk with GLP-1RAs is cause specific and time dependent; research should adjudicate and report pancreatitis by cause and onset (early suspected drug induced v later reductions in alcohol associated and hypertriglyceridaemia associated events)

• Clinical and policy guidance should emphasise early stage counselling and monitoring after initiation while recognising that later risk reductions may offset early drug induced risk

Introduction

Glucagon-like peptide-1 receptor agonists (GLP-1RAs) are widely used across an expanding list of indications, including obesity.¹ As GLP-1RA use has grown, longstanding concerns about a possible association between GLP-1RA and pancreatitis have resurfaced.² This renewed attention has been accompanied by several new case reports and pharmacovigilance signals linking GLP-1RA and pancreatitis, prompting public and regulatory calls to investigate this link.³,⁸ In June 2025, the UK Medicines and Healthcare products Regulatory Agency announced an investigation into a possible link between GLP-1RA and pancreatitis,^{6 9} underscoring the need for robust epidemiological evidence.

Randomised controlled trials generate valuable safety data, but pancreatitis is relatively rare, limiting the power of these trials to detect possible increases in risk. Individual trials adjudicated pancreatitis events yet observed few cases; for example, in the LEADER trial, use of liraglutide was associated with increases in serum lipase and amylase but adjudicated acute pancreatitis occurred in 18 liraglutide versus 23 placebo recipients over 3.8 years.^{10 11} Meta analyses of large, placebo controlled GLP-1RA trials have not shown a significant increase in acute pancreatitis, though imprecise estimates show some uncertainty.²¹²,¹⁴ These trial data reduce concern about a large average effect and suggest that, if a drug attributable risk exists, it is small enough to evade detection in trials with limited events; they do not, however, exclude a rare idiosyncratic signal.

Observational data are mixed. Pharmacovigilance analyses have reported disproportionate signals of acute pancreatitis associated with use of GLP-1RAs.^{3 8} Results from real world studies (which mostly evaluated all cause pancreatitis) are mixed: some report increased risk, some report no statistically significant association, and some report lowered risk of pancreatitis among people using GLP-1RA.⁷¹⁵,¹⁷ Heterogeneity in study design, methodologies, and outcome adjudication that did not distinguish drug induced pancreatitis from other causes of pancreatitis likely contributed to these mixed results, leaving the causal relationship unresolved.

Two key limitations of the current evidence exist. Firstly, acute pancreatitis is serious but uncommon. Even large randomised controlled trials and pooled analyses are underpowered to detect increases in risk when the outcome rate is low. Secondly, most studies evaluated all cause acute pancreatitis and did not distinguish drug induced pancreatitis from other causes (eg, alcohol induced pancreatitis). This matters because GLP-1RAs may have opposing associations with different types of pancreatitis. Because use of GLP-1RAs reduces alcohol use and triglyceride levels, GLP-1RAs may reduce risk of pancreatitis from these causes,¹⁸,²⁵ but may increase the risk of drug induced pancreatitis. Analyses that aggregate all causes of pancreatitis can therefore yield a net null result that masks countervailing drug specific signals.

The US Department of Veterans Affairs operates the largest integrated health system in the US, providing the population size needed to study rare adverse events. Its longitudinal electronic health records link medication exposure with clinical data (ie, diagnoses, laboratory information, pharmacy information, clinical notes, hospital notes, hospital discharge summaries), enabling classification of acute pancreatitis by cause (eg, alcohol induced, biliary, or drug induced). This depth and breadth of information allows robust estimation of the relationship between GLP-1RA use and various causes of acute pancreatitis.

In our study, we used the Department of Veterans Affairs healthcare databases to emulate a pragmatic target trial in patients with type 2 diabetes who initiated GLP-1RA (n=132 551) or sulfonylureas (n=201 136). We followed participants for a year to evaluate risks of all cause and cause specific acute pancreatitis (suspected drug induced, hypertriglyceridaemia associated, alcohol induced, biliary, and idiopathic or other causes of acute pancreatitis).

Methods

Emulation of target trial

We designed and conducted this study following the target trial emulation framework.^{26 27} We first specified the protocol for a randomised pragmatic trial that evaluates the risk of acute pancreatitis after initiation of GLP-1RAs compared with a control arm of initiation of sulfonylureas (online supplemental table S1). The target trial would enrol Department of Veterans Affairs users with type 2 diabetes who had not used GLP-1RAs or sulfonylureas within the previous year and had no contraindication to GLP-1RAs or sulfonylureas. Participants would be randomly assigned to initiate use of GLP-1RA or sulfonylureas after enrolment. We then emulated the target trial using data from electronic healthcare databases of the US Department of Veterans Affairs.

Setting

We conducted the study using the electronic healthcare databases of the US Department of Veterans Affairs. The Department of Veterans Affairs operates the largest nationally integrated healthcare system in the US, providing care to more than 9 million veterans at 1380 healthcare facilities, including 170 medical centres and 1193 outpatient sites. The Department of Veterans Affairs provides a comprehensive package of medical benefits: inpatient and outpatient services, preventive and primary care, specialty care, geriatric care, mental healthcare, home care, extended care and rehabilitation services, prescription medication coverage, medical equipment and prosthetics coverage, and prescription coverage including prescription for GLP-1RAs to veterans enrolled in the system.

Data sources

We accessed data from the Department of Veterans Affairs Corporate Data Warehouse. The data domains used in the study were outpatient encounters, inpatient encounters, outpatient pharmacy, laboratory results, vital signs, patient, Department of Veterans Affairs vital status, health factors, and text integration utilities.²⁸ Medicare data provided by the Department of Veterans Affairs Information Resource Center were collected to account for healthcare encounters outside of the Department of Veterans Affairs system. We collected the Area Deprivation Index (scored from 1 to 100) on participants' residential location.^{29 30}

Cohort construction

We identified 620 168 participants with type 2 diabetes who received at least one prescription for GLP-1RA or sulfonylureas between 1 January 2017 and 31 December 2023, had at least two visits to the Department of Veterans Affairs healthcare systems and used the Department of Veterans Affairs outpatient pharmacy within one year before receiving the prescription (figure 1). The date of the first GLP-1RA or sulfonylureas prescription received by a participant was defined as time zero (T₀). Of the 620 168 participants, we selected 351 463 participants who had not received a prescription for GLP-1RA or sulfonylureas in the year before T₀. Participants with contraindications for initiating GLP-1RAs or sulfonylureas, including those with a history of medullary thyroid carcinoma, multiple endocrine neoplasia type 2, gastroparesis, estimated glomerular filtration rate <30 mL/min/1.73 m², dialysis or kidney transplant, hypoglycaemia with coma, hypoglycaemia requiring hospital admission, pancreatitis, and pancreatic cancer were not eligible for cohort entry, resulting in a final study population of 333 687 participants. Cohort participants were then separated into GLP-1RA group (n=132 551) and sulfonylureas group (n=201 136) based on the prescription they received at T₀. Participants were followed until the first of: occurrence of outcome, death, one year elapsing after T₀, or 31 December 2024.

Figure 1. Cohort flow diagram. VA=Department of Veterans Affairs.

Exposures

We defined GLP-1RA and sulfonylureas exposures based on outpatient pharmacy records. The sulfonylureas class comprised glimepiride (13.94%), glipizide (85.72%), and glyburide (0.33%). The GLP-1RA class comprised albiglutide (0.96%), dulaglutide (13.41%), exenatide (0.98%), liraglutide (18.48%), lixisenatide (0.01%), semaglutide (66.13%), and tirzepatide (0.03%) (online supplemental table S2). In addition to the GLP-1RA class effect, we evaluated the effect of semaglutide on risk of acute pancreatitis outcomes. In this study, we evaluated both the intention-to-treat effect and also the per protocol effect, where the protocol was specified as continued use of the initiated medication during follow-up. Participants were considered adherent to the protocol if they had a refill of the initiated medication class within 90 days after the end of earlier prescriptions.

Outcomes

We defined outcomes using definitions previously validated for use in the electronic health records.³¹,⁴⁷

The primary outcome was incident, clinically significant acute pancreatitis. We first identified a cohort of potential cases using ICD-10-CM (international classification of diseases, 10th revision, clinical modification) diagnosis codes K85.0X to K85.9X. To increase the positive predictive value and ensure clinical significance, a case was confirmed only if the diagnosis code was recorded during either an emergency department encounter or a hospital admission where acute pancreatitis was a listed diagnosis. We defined the event date as the date of the qualifying emergency department visit or hospital admission.

Cause specific acute pancreatitis was classified into a single, mutually exclusive aetiologic category using a predefined hierarchical algorithm, applied in the following order: biliary acute pancreatitis (a confirmed case of acute pancreatitis with a diagnosis code K85.1X and an accompanying diagnosis code K80.X-K83.X for cholelithiasis, cholecystitis, or other biliary tract diseases in the 180 days before or including the event date). Alcohol induced acute pancreatitis (a non-biliary case of acute pancreatitis with diagnosis code K85.2X and an accompanying diagnosis code F10.X for alcohol use disorder, abuse, dependence, or K70.X for alcohol related liver disease in the 365 days before or including the event date). Hypertriglyceridaemia associated acute pancreatitis (a non-biliary, non-alcohol induced case of acute pancreatitis with diagnosis codes K85.0X, K85.8X, or K85.9X recorded within 30 days after a triglyceride measurement >500 mg/dL). Suspected drug induced acute pancreatitis (a non-biliary, non-alcohol, non-hypertriglyceridaemia associated pancreatitis with a recorded diagnosis code of K85.3X which was confirmed through a review of clinical notes). Idiopathic or other acute pancreatitis (any confirmed case of acute pancreatitis not meeting the criteria for the causes listed above with a diagnosis code K85.X not otherwise classified into any of the above categories). We additionally examined all cause acute pancreatitis, defined as the composite of all acute pancreatitis categories above.

Covariates

Covariates were specified based on earlier knowledge⁴⁸,⁶¹ and aided by a directed acyclic graph (online supplemental figure S1). Baseline covariates associated with treatment assignment for GLP-1RAs or sulfonylureas, and risk of pancreatitis were ascertained within one year before T₀ unless otherwise specified. For variables with repeated measurements, we used the measurement closest to and preceding T₀.

We adjusted for sociodemographic variables including age, race (white, black, and other), sex, rurality (urban, rural, highly rural, and isolated island), and area deprivation index. The area deprivation index represents a composite of socioeconomic disadvantage based on education, income, employment, and other poverty measures and is measured at the participants' location. Covariates also included vital measurements including systolic and diastolic blood pressure and body mass index, laboratory measurements including estimated glomerular filtration rate, albuminuria, serum albumin, low density lipoprotein, triglycerides, serum calcium, aspartate aminotransferase and alanine aminotransferase, alkaline phosphatase, and total bilirubin.

In consideration of associations between diabetes status and antihyperglycaemics initiation, and also associations between diabetes and risk of pancreatitis, we adjusted for haemoglobin A_1c at baseline, average haemoglobin A_1c within one year before T₀, and separately within five years before T₀, duration of diabetes from October 1999 until T₀, use of metformin, sulfonylureas, thiazolidinediones, dipeptidyl peptidase-4 inhibitors (DPP4i), sodium-glucose cotransporter-2 inhibitors (SGLT2i), and insulin, as well as the duration of metformin, sulfonylureas, thiazolidinediones, DPP4i, SGLT2i, and insulin use within five years before T₀.

We adjusted for a comprehensive array of comorbidities. Circulatory comorbidities comprised stroke, transient ischaemic attack, atrial fibrillation, arrhythmias, acute coronary disease, myocardial infarction, ischaemic cardiomyopathy, angina, heart failure, and non-ischaemic cardiomyopathy. We also adjusted for gastrointestinal comorbidities including nausea, diarrhoea, abdominal pain, dyspepsia, constipation, gastroesophageal reflux disease, gastroparesis, gastritis, non-alcoholic fatty liver disease, ulcerative colitis, Crohn's disease, intestinal obstruction and ileus, diverticulosis and diverticulitis, biliary tract disease, hepatic failure, gastrointestinal haemorrhage, gallstone and cholecystitis, cholecystectomy, primary sclerosing cholangitis, other cholestatic liver disease, liver cancer, pancreatic cancer, pancreatic cysts and other pancreatic anatomic abnormalities, cancer of the gallbladder or biliary system, endoscopic retrograde cholangiopancreatography, bariatric surgery, and other abdominal surgeries.

We also adjusted for conditions including acute kidney injury, urinary tract infections, fluid and electrolyte disorders, hypoglycaemia, diabetic ketoacidosis, HIV, primary hyperparathyroidism, cystic fibrosis, other cancers, peripheral artery disease, peripheral neuropathy, and autoimmune disorders.

Health behaviour and healthcare use that may be associated with treatment initiation and risk of pancreatitis were also adjusted for. Health behaviours we adjusted for were smoking status and alcohol use disorder; the healthcare use variables adjusted for were long term care, number of outpatient encounters, number of hospital admissions, number of prescriptions, number of blood panel tests, number of haemoglobin A_1c measurements, and number of outpatient encounters and hospital admissions accessed through Medicare. The calendar week of treatment assignment was also adjusted for.

We also adjusted for history of medication: use of statins, other antihyperlipidaemic agents, angiotensin converting enzyme inhibitors, angiotensin receptor blockers, beta blockers, diuretics, calcium channel blockers, anti-obesity medications including bupropion, naltrexone, orlistat, phentermine, topiramate, and medications that may increase risk of pancreatitis (valproate, azathioprine, 6-mercaptopurine, isotretinoin, tetracyclines, trimethoprim/sulfamethoxazole, thiazides, and fibrates).

Single imputation based on multivariate imputation by chained equations was used to assign values to missing data including 2.86% of haemoglobin A_1c, 4.93% of estimated glomerular filtration rate, 7.85% of low density lipoprotein and triglyceride, 2.36% of blood pressure, 6.18% of BMI, 7.99% of aspartate aminotransferase and alanine aminotransferase, 10.51% of alkaline phosphatase, 11.25% of serum albumin, and 11.03% of total bilirubin. We adjusted continuous variables in the form of restricted cubic splines with knots at the fifth, 35th, 65th, and 95th centiles.

For per protocol analyses, we updated the covariates at every 30 day interval during follow-up, and captured the occurrence of any contraindication for GLP-1RA or sulfonylureas including medullary thyroid carcinoma, multiple endocrine neoplasia type 2, gastroparesis, dialysis or kidney transplant, hypoglycaemia, and pancreatitis during follow-up. Covariates updated by time were used along with baseline covariates to estimate the probability of continued GLP-1RA use and, separately, continued use of sulfonylureas during follow-up.

Statistical analyses

Baseline characteristics of the sulfonylureas group and GLP-1RA group are presented as mean (standard deviation) or frequency (percentage) as appropriate. Differences in characteristics between the two groups are evaluated through absolute standardised mean differences, where a value of <0.1 was considered evidence of good balance.⁶²

To evaluate the intention-to-treat effect of GLP-1RA and sulfonylureas on risk of acute pancreatitis, we balanced the baseline characteristics between the two groups through inverse probability weighting. Logistic regression was constructed to evaluate the probability of being assigned to the GLP-1RA group (the propensity score). We then constructed the standardised mortality ratio weighting as one for the GLP-1RA group and as propensity score/(1−propensity score) for the sulfonylureas group to estimate the average treatment effect within the treated group.⁶³ Truncation for the weighting was set to be 0.5th and 99.5th centile, and we used discrete time survival analysis based on pooled logistic regression. Spline function of time and interaction between group and time were included to estimate the incidence rate of outcome at one year, where death was treated as a competing risk. Rates within each group and absolute rate differences per 100 000 persons at one year were reported. We also applied similar analyses to evaluate risk of semaglutide compared with sulfonylureas, where the overall GLP-1RA group served as the target population in the standardised mortality ratio weighting.

We then evaluated the per protocol effect of GLP-1RA and sulfonylureas on risk of acute pancreatitis, where the protocol was defined as continued use of the assigned medication during follow-up. To account for the informative censor of not continuing treatment in both arms, time varying inverse probability of treatment adherence weights were constructed for each group during the follow-up period. We used pooled logistic regression including both baseline and time varying variables to estimate the probability of continued treatment at every 30 day interval within those adherent to protocol at the beginning of the interval. The time varying inverse probability of treatment adherence weighting was then computed as 1/P(treatment) for those who remained adherent to the protocol within the interval. These weights were stabilised using the probability of treatment conditional on only baseline variables. We then constructed the cumulative inverse probability of treatment adherence weights as the product of weights from treatment initiation until each time point: $S{W}_k^Z=\prod _{k=1}^t\frac{\mathrm{P}\mathrm{r}\left(Z_k=1|{\stackrel{-}{\mathrm{Z}}}_{k-1}=1,V\right)}{\mathrm{P}\mathrm{r}\left(Z_k=1|{\stackrel{-}{\mathrm{Z}}}_{k-1}=1,V{\stackrel{-}{,Q}}_{k-1}\right)}$,where Z is an indicator of treatment, V is a vector of covariates at baseline, and ${\stackrel{-}{Q}}_{k-1}\mathrm{}$is a vector of covariates at time k-1 (online supplemental table S3). We then generated summary weights as the product of baseline standardised mortality ratio weights and time varying inverse probability of treatment adherence weights. These summary weights were then applied to a discrete time survival analysis to estimate the rate at every time point and the cumulative incidence rate. Risks (per month and cumulatively) on the absolute and relative scale within participants who continued use of the assigned treatments were further computed based on the rates.

Confidence intervals (CIs) of the results were estimated based on 1000 iterations of parametric bootstrapping. Point estimates and 95% CIs were reported. Analyses were conducted using Statistical Analysis System Enterprise Guide 8.3 and data visualisations were produced by R 4.3.3. This study was approved by the institutional review board of the Department of Veterans Affairs St Louis healthcare system, which granted a waiver of informed consent (protocol No 1606333). The study was reported following both the strengthening the reporting of observational studies in epidemiology (STROBE) statement and the transparent reporting of observational studies emulating a target trial (TARGET) statement.^{64 65}

Patient and public involvement

We did not directly involve patients or the public in developing the research question, or in the study design or implementation owing to a limited timeline. We have no specific plans to disseminate the findings to study participants. The results of the study, however, will be disseminated by press release with mainstream media and will also be promoted on X from the corresponding author’s account. The results will be shared with advocacy groups and will be presented at scientific meetings and at academic institutions.

Results

We recruited 132 551 people to the GLP-1RA group and 201 136 to the sulfonylureas control group. We followed the cohort for a mean of 359.74 days (standard deviation (SD) 35.54 days) and a median of 365 days (interquartile range (IQR) 365-365). The personal and clinical characteristics of the two groups before and after weighting are presented in online supplemental table S4 and S5, respectively. We evaluated the standardised mean differences for covariates before and after weighting in the two groups. All standardised mean differences were below the conventional threshold of 0.1, suggesting that adequate covariate balance was achieved (online supplemental figure S2 and S3).

GLP-1RA and cause specific pancreatitis

In intention-to-treat analyses, compared with sulfonylureas, GLP-1RA had similar rates of all cause pancreatitis (rate difference −3.64 (95% CI −30.76 to 23.48) per 100 000 people at one year). However, this overall null result represented countervailing effects across various causes of acute pancreatitis.

GLP-1RA was associated with an increased risk of suspected drug induced acute pancreatitis (23.45 (14.27 to 33.85)), and reduced risks of hypertriglyceridaemia associated acute pancreatitis (−16.96 (−27.41 to −7.34)) and alcohol induced acute pancreatitis (−10.32 (−18.12 to −3.17)). GLP-1RA and sulfonylureas groups had similar rates of biliary acute pancreatitis and idiopathic or other causes of acute pancreatitis (−2.66 (95% CI −12.82 to 7.40) and 2.85 (−19.56 to 26.44), respectively) (figure 2 and online supplemental table S6).

Figure 2. Intention-to-treat effect of GLP-1RA compared with sulfonylureas on acute pancreatitis over one year's follow-up. Upper panel shows absolute rate difference. Lower panel shows incidence rates in GLP-1RA group (blue) and sulfonylureas group (yellow). Rates and rate differences reported as per 100 000 people at one year with 95% confidence intervals. GLP-1RA=glucagon-like peptide-1 receptor agonists; CI=confidence interval.

The null association between GLP-1RA and all cause pancreatitis represents the net effect of an increased risk of suspected drug induced pancreatitis which was offset by reductions in alcohol induced pancreatitis and hypertriglyceridaemia associated pancreatitis.

Per protocol analyses showed results directionally consistent with those derived from intention-to-treat analyses (online supplemental figure S4 and table S7).

Semaglutide and cause specific pancreatitis

Because semaglutide represents the most widely used GLP-1RA globally, we conducted a prespecified agent specific analysis. Results of intention to treat analyses recapitulated the class effect, with similar risk of all cause acute pancreatitis in GLP-1RA and sulfonylureas group. Analyses of causes of pancreatitis showed increased risk of suspected drug-induced pancreatitis, lower risks of hypertriglyceridaemia associated and alcohol induced pancreatitis, no association with biliary, idiopathic or other causes (online supplemental figure S5 and table S8).

Temporal distribution of the various causes of acute pancreatitis

To understand the timing of different causes of acute pancreatitis, we evaluated the monthly risk over 12 months of follow-up (figures3 4). We observed distinct temporal patterns by cause (figure 5, figure 6). The risk of suspected drug induced pancreatitis was highest immediately after treatment initiation, with the first three months alone accounting for over 40% (15.8%, 14.1%, and 12.1% during the first, second, and third months, respectively) of the total one year rate difference. Conversely, the risk of alcohol induced pancreatitis showed its steepest decline during months 4-6, with these months accounting for approximately 33% of the total one year risk reduction. We observed a similar trend for hypertriglyceridaemia associated pancreatitis, where the risk declined most sharply in the final quarter (months 10-12). In contrast, and as expected for causes with no significant association with GLP-1RA use, the risks for biliary pancreatitis and idiopathic pancreatitis remained stable throughout the follow-up period (online supplemental table S9). Cumulatively, these divergent temporal trends resulted in a net increase in all cause pancreatitis risk during the first two months of treatment, with similar monthly risk for the remainder of follow-up.

Figure 3. Cumulative incidence of GLP-1RA and sulfonylureas on all cause pancreatitis and cause specific acute pancreatitis. Top left graph shows all cause acute pancreatitis; top right shows suspected drug induced acute pancreatitis; middle left shows hypertriglyceridaemia associated acute pancreatitis; middle right shows alcohol induced acute pancreatitis; bottom left shows biliary acute pancreatitis; bottom right shows idiopathic or other causes of acute pancreatitis. GLP-1RA group is shown in blue and the sulfonylureas group is shown in yellow. Shaded areas represent 95% confidence intervals. GLP-1RA=glucagon-like peptide-1 receptor agonists.

Figure 4. Monthly absolute rate difference of GLP-1RA compared with sulfonylureas on all cause pancreatitis and cause specific acute pancreatitis. Top left graph shows all cause acute pancreatitis; top right shows suspected drug induced acute pancreatitis; middle left shows hypertriglyceridaemia associated acute pancreatitis; middle right shows alcohol induced acute pancreatitis; bottom left shows biliary acute pancreatitis; bottom right shows idiopathic or other causes of acute pancreatitis. Monthly absolute rate difference per 100 000 people from one to 12 months after treatment initiation, with 95% confidence intervals. Positive values indicate higher rates in GLP-1RA compared with sulfonylureas, negative values indicate lower rates. Results are based on per protocol analyses where treatment protocol was defined as continued use of the initiated medication during follow-up. Rate differences are reported as per 100 000 people per month. GLP-1RA=glucagon-like peptide-1 receptor agonists.

Figure 5. Absolute rate difference and proportional contribution of acute pancreatitis associated with GLP-1RA compared with sulfonylureas over time. Upper graph shows rate difference by quarter of year (months 1-3, 4-6, 7-9, and 10-12); lower graph shows proportional contribution to one year rate difference by quarter of year. GLP-1RA=glucagon-like peptide-1 receptor agonists.

Figure 6. Proportional contribution to one year rate difference by month. Results are based on per protocol analyses where treatment protocol was defined as continued use of the initiated medication during follow-up. Proportional contribution was estimated as the absolute rate difference within the specified period divided by the difference during full one year of follow-up. Results are reported for suspected drug induced acute pancreatitis, hypertriglyceridaemia associated acute pancreatitis, and alcohol induced acute pancreatitis, where statistically significant associations with GLP-1RA were observed. For suspected drug induced acute pancreatitis, higher values indicate a greater increase in risk among the GLP-1RA group compared with the sulfonylureas group. For hypertriglyceridaemia associated and alcohol induced acute pancreatitis, higher values indicate a greater reduction in risk among the GLP-1RA group compared with the sulfonylureas group. GLP-1RA=glucagon-like peptide-1 receptor agonists.

Discussion

In this nationwide target trial emulation comprising more than 333 000 participants, we found that initiation of a GLP-1RA was not associated with an increased risk of all cause acute pancreatitis at one year follow-up compared with sulfonylureas. However, this neutral finding represents the net effect of a more complex and clinically important dynamic: a small, early onset risk of suspected drug induced pancreatitis was fully offset by a later reduction in pancreatitis related to alcohol use and hypertriglyceridaemia. These findings expand the evidence base and provide insights for a deeper understanding of the multifaceted association between GLP-1RA use and pancreatitis.

Our study answers a limitation of previous research by analysing pancreatitis based on its specific cause rather than as a single condition. Following a large cohort for one year with ample statistical power, our cause specific approach enabled us to characterise the bidirectional relationship between GLP-1RAs and various causes of pancreatitis. By separating acute pancreatitis by cause, we found that, while the drug's net effect on all cause pancreatitis appears neutral, it is actually composed of opposing signals that effectively cancel each other out. This masking effect may explain the null results seen in most earlier studies. Specifically, our work demonstrates the simultaneous presence of both a safety signal (an increased risk of drug induced pancreatitis) and a protective signal (a reduced risk of pancreatitis related to alcohol and hypertriglyceridaemia).

The distinct temporal patterns for each cause of pancreatitis provide clinically useful risk windows for predicting when a specific type of pancreatitis is most likely to occur during GLP-1RA treatment. The excess risk of suspected drug induced pancreatitis was front loaded, clustering within the first few months of therapy. This pattern is consistent with a drug effect, where susceptible individuals react soon after initial exposure. In contrast, the reductions in alcohol induced and hypertriglyceridaemia associated pancreatitis accrued more slowly over the course of the year. This timeline is consistent with the emerging mechanistic and epidemiological evidence suggesting that GLP-1RAs may curb alcohol consumption and ameliorate metabolic parameters (eg, by reducing triglyceride levels).¹⁸²¹,²⁵ The transient increase in all cause risk during the first two months is a direct result of these opposing timelines: the drug induced risk manifests before the protective benefits have had sufficient time to accumulate.

These findings may have implications for clinical practice and regulatory agencies.⁴ For clinicians, our results provide a basis for patient counselling. Patients initiating a GLP-1RA treatment regimen can be counselled about a small, early risk of drug induced pancreatitis, while also being reassured that the overall long term risk is not increased. Our study suggests that, while the signal for drug induced pancreatitis is empirically evident, it does not translate to an overall increase in pancreatitis burden at the population level because of the drug's countervailing benefits (reducing alcohol and triglyceride related pancreatitis). This distinction may be crucial for regulators, informed risk benefit assessments, and public communication.

This study had several limitations. The study participants were US veterans who are older and mostly male, which may not represent the general population and may limit the generalisability of results to other populations (eg, women, younger people). However, the study participants comprised 23 187 (6.95%) women, 7288 (2.18%) people younger than 40 years, 32 433 (9.72%) people younger than 50 years, and 95 382 (28.58%) people younger than 60 years. Despite adjusting for a comprehensive set of covariates from multiple data domains, we cannot completely rule out the possibility of residual confounding. While we leveraged the breadth and depth of Department of Veterans Affairs electronic health records to identify causes of pancreatitis based on diagnosis codes, laboratory test results, vital measurements, and healthcare encounters to enhance the specificity of our outcome definition, our approach may not capture all acute pancreatitis. For example, cases of undiagnosed acute pancreatitis, acute pancreatitis recorded only in outpatient setting and without the patient attending an emergency department, visiting urgent care, or being admitted to hospital may not have been captured. To fit our aim of emulating a target trial evaluating the comparative safety of GLP-1RA versus sulfonylureas, we prioritised specificity and positive predictive value when defining cause specific acute pancreatitis. Consequently, cases lacking sufficient evidence for a specific cause were classified into the idiopathic or other causes of acute pancreatitis category. Although we used a hierarchical algorithm and validated definitions to assign outcomes, drug induced pancreatitis is a diagnosis of exclusion, hence we refer to it throughout as suspected drug induced pancreatitis. We required a high acuity of care (emergency department visit or inpatient stay) to establish the diagnosis of acute pancreatitis. This approach mitigates but may not eliminate surveillance bias and ascertainment bias because of increased overall attention to the possible link between GLP-1RA and pancreatitis. We cannot rule out the possibility of misclassification bias. We examined the associations for class effect of GLP-1RA and for semaglutide, we did not examine other types of GLP-1RA. All cohort participants had received a diagnosis of type 2 diabetes and we did not examine risks by treatment indication.

The first strength of this study is use of the Department of Veterans Affairs health system, which provides a comprehensive medical benefits package, including prescription coverage for GLP-1RA. Consequently, initiation or discontinuation of GLP-1RA was less likely to be driven by financial considerations. We specified and emulated a randomised pragmatic target trial that would answer our research question and its corresponding estimand (including the use of an incident user, active comparator design) to reduce the risk of biases such as immortal time and confounding bias.⁶⁶,⁶⁸ We leveraged the breadth and depth of Department of Veterans Affairs data to account for a comprehensive array of covariates from multiple data domains such as demographics, diagnoses, laboratory test results, medications, vital signs, healthcare use, and contextual characteristics. We examined risks of all cause acute pancreatitis and all the major causes of acute pancreatitis, providing a comprehensive characterisation of the relationship between GLP-1RA and acute pancreatitis. In addition to intention-to-treat analyses, we conducted per protocol analyses to estimate the effect of continued GLP-1RA use during follow-up, which reduces the influence of non-adherence to the study protocol and treatment switching on effect estimates. To capture the dynamic changes in risk over time, we estimated cumulative risks from time varying incidence rates, avoiding reliance on the proportional hazards assumption.^{69 70}

In conclusion, initiation of GLP-1RAs was not associated with an overall increase in the one year risk of acute pancreatitis. This neutral overall effect is the net result of a small, early risk of drug induced pancreatitis that is counterbalanced by later reductions in pancreatitis associated with alcohol use and hypertriglyceridaemia. This work provides a nuanced, data driven framework for clinicians and regulators to assess the risk-benefit profile of this important class of medications.

Data availability statement

Data may be obtained from a third party and are not publicly available.