Rates of Pancreatitis and Pancreatic Cancer Linked to Different GLP-1 Receptor Agonist Drugs

On May 6, 2024, a systematic search was performed following the guidelines from the Preferred Reporting Items for Systematic Reviews and Meta‐Analyses (PRISMA) framework. The following search query was utilised in PubMed, Embase, and Cochrane Library: ((pancrea*) OR (pancreatitis)) AND (((((((((glucagon‐like peptide‐1) OR (glp‐1)) OR (lixisenatide)) OR (liraglutide)) OR (semaglutide)) OR (tirzepatide)) OR (albiglutide)) OR (exenatide)) OR (dulaglutide)).

The Patient, Intervention, Comparison, Outcome, Time (PICOT) methodology was utilised to guide article selection. The patient population and intervention encompassed patients of any age taking a GLP‐1 RA. As all studies were randomised controlled trials (RCTs), comparisons with active comparators or placebo were included. The main outcomes assessed were rates of acute pancreatitis and pancreatic cancer among different agents within the GLP‐1 RA class. Studies with a minimum follow‐up of 1 week were included. The following inclusion criteria were utilised: (1) RCTs with patients taking GLP‐1 RAs; (2) active GLP‐1 RAs or discontinued GLP‐1 RAs (e.g., albiglutide) that were compared against another active GLP‐1 RA; (3) reported on adverse events (AEs) with a focus on pancreatitis or pancreatic cancer. Although we excluded studies that compared albiglutide or lixisenatide against placebo, we included studies that compared those agonists against other GLP‐1 RAs. These head‐to‐head trials were retained to provide comparative safety data for the active GLP‐1 RAs that are still in use, which remain clinically relevant for clinicians and patients. Exclusion criteria consisted of studies that did not meet the three inclusion criteria, discontinued GLP‐1 RAs (e.g., albiglutide) when compared against placebo, or if they were study designs such as non‐RCTs, case reports, review articles, expert opinions, abstracts, commentaries, animal studies, and studies not in English. This systematic review protocol is registered with PROSPERO as CRD42024550993.

Two independent authors participated in the title/abstract and full‐text screening of each article to determine study eligibility for inclusion. If there was a discrepancy between authors, the lead author (J.W.) reviewed the article for final study eligibility. Upon completion of full‐text screening, a rigorous reference search was performed for the included studies to assess for further studies to add to this review.

The risk of bias (RoB) and study quality was assessed utilising the Cochrane RoB tool, given that all included studies were RCTs. Two independent authors assessed the risk of bias and methodological quality, and any disagreements were resolved via rigorous re‐review until consensus was achieved. The Cochrane tool contains seven domains: sequence generation, allocation concealment, blinding of participants and personnel, blinding of outcomes, incomplete outcome data, selective outcome reporting, and other sources of bias. Each domain is evaluated as high, low, or unclear RoB. Two independent authors analysed each article, and any discrepancies were resolved via rigorous re‐review or by consulting a third reviewer until agreement was reached.

The study variables extracted and analysed included title, author, study year, number of patients, GLP‐1 RA agent and dosages, mean age, mean follow‐up time, pancreatitis/pancreatic cancer rates, related lab values (amylase/lipase), and those taking background anti‐diabetic drugs. Notably, the diagnostic criteria were not consistently reported across the studies and were reported as investigator or an independent committee–determined AE. The data was compiled using Google Sheets (Google Drive; Google, Mountain View, CA). The mean, median, percentages, ranges, and standard deviations were reported if available and applicable.

Meta‐analysis was conducted using the fixed effects model to evaluate the risk ratio (RR) of developing acute pancreatitis or pancreatic cancer. Chronic pancreatitis was not specified for pooling, and pancreatic cancer was an exploratory outcome, given that trial durations were not expected to be long enough to assess the risks for these two conditions. Sensitivity analyses were conducted with RCTs that had at least 24 weeks of follow‐up to test for differences in pancreatitis and pancreatic cancer findings. Heterogeneity was calculated with the I‐squared statistic, with p‐values less than 0.05 considered statistically significant. Forest plots were generated using Cochrane's Review Manager application (RevMan, version 5.4; The Cochrane Collaboration, London, UK). Studies with double‐zero events (treatment arms that had zero events) were included in the forest plot for transparency.

Across the 62 studies, there were a total of 66,232 patients (34,941 male/32,291 female) with a mean age of 58.3 years (14.4 to 68.0), mean duration of diabetes of 9.5 years (1.5 to 21.6), mean body mass index (BMI) of 32.6 kg/m² (24.4 to 46.2), and mean follow‐up of 43.5 weeks (1 week to 198 weeks). (Table 1).

The systematic review includes 62 RCTs comparing various GLP‐1 RAs, including various dosages of dulaglutide, exenatide, liraglutide, semaglutide, beinaglutide, retatrutide, and tirzepatide. The review also divides studies between those that included patients who continued background medications to maintain glycaemic control and those that did not use background medications. These medications included any combination of metformin, sulfonylureas, insulin, dipeptidyl peptidase 4 (DPP‐4) inhibitors, thiazolidinediones, Sodium‐Glucose Cotransporter 2 inhibitor (SGLT‐2i), and other background anti‐diabetic medication except GLP‐1 RAs.

Across the seven domains, most studies were low risk of bias for sequence generation (50/62), incomplete outcome data (54/62), selective outcome reporting (60/62), and other sources of bias (33/62). However, allocation concealment had a substantial amount of unclear ranking (29/62), and high risks of bias were found for blinding of participants/personnel (20/62) as well as blinding of outcomes (23/62). These results are depicted in Tables and. S1 S2

Across all studies that included patients who continued background medications, the greatest number of pancreatitis and pancreatic cancer occurred in patients who received dulaglutide [28 (0.35%) and 19 (0.24%), respectively]. Across the studies that included patients who did not receive background medications, the greatest number of pancreatitis occurred in patients who received exenatide [42 (0.56%)] and the greatest number of pancreatic cancer occurred in patients who received dulaglutide [1 (0.93%)]. These results are illustrated in Table 2.

A comprehensive meta‐analysis was conducted to assess the risk of pancreatitis and pancreatic cancer associated with GLP‐1 RAs compared to controls. In the pooled analysis of all studies evaluating pancreatitis incidence (Figure 2), GLP‐1 RA use was associated with a statistically significant increase in risk (RR = 1.44; 95% CI: 1.09–1.89; p = 0.009; I² = 0%). When stratified by background medication use, studies that included background therapies (Figure S1) and those that did not (Figure S2) both showed nonsignificant associations (RR = 1.28, 95% CI: 0.87–1.87, p = 0.22, I² = 0% and RR = 1.37, 95% CI: 0.91–2.05, p = 0.13, I² = 0%, respectively).

For pancreatic cancer, the overall pooled analysis (Figure S3) demonstrated no significant association with GLP‐1 RA use (RR = 1.30, 95% CI: 0.86–1.97, p = 0.22, I² = 0%). There was also no significant association found in the subgroup analysis without background medications (Figure S4; RR = 0.81, 95% CI: 0.43–1.55, p = 0.53, I² = 0%); however, the results were significant with background medications (Figure S5; RR = 1.85; 95% CI: 1.05–3.26, p = 0.03, I² = 0%).

When subgrouped by studies that had a minimum 24‐week follow‐up, pancreatitis risk (RR = 1.43, 95% CI: 1.09–1.89, p = 0.01, I² = 0%) and pancreatic cancer (RR = 1.28, 95% CI: 0.84–1.96, p = 0.25, I² = 25%) remained statistically significant and non‐significant, respectively (Figures S6 and S7).

Incretins have been suspected to relate to pancreatitis, especially as T2DM is associated with the development of acute and chronic pancreatitis [12]. Although the mechanism of action is not yet elucidated, warnings and precautions of pancreatitis are included in the prescribing information for GLP‐1 RAs. Preliminary studies have found that GLP‐1 elicits a beta‐cell preserving effect, and these sustained proliferative actions may cause unintentional exocrine and endocrine effects (e.g., pancreatitis, pancreatic cancer) [13]. Acute pancreatitis, specifically fatal haemorrhagic and necrotising types, has been noted, but the causal relationship between GLP‐1 RAs and pancreatitis/pancreatic carcinoma is not yet confirmed [14]. GLP‐1 RAs, thus, are not recommended to be prescribed for patients with a history of pancreatitis and should be discontinued if pancreatitis symptoms develop. However, recent literature has suggested that GLP‐1 RAs can be used in patients with a history of acute pancreatitis, albeit on an individualised basis where the glycaemic, cardiovascular, and weight loss benefits outweigh the risks [15]. A retrospective review of 161 patients with a history of acute pancreatitis at a mean follow‐up of 28.2 months found a total of 16 (9.9%) secondary episodes, of which 6 (37.5%) were attributed to the GLP‐1 RA [16]. The 9.9% rate is lower than a pooled recurrent pancreatitis rate by Li et al. of 21% (95% CI: 18%–24%) and a pooled incidence rate calculated by Gagyi et al. of 5.26 per 100 person‐years (CI: 3.99–6.94) [17, 18].

Previous studies utilised postmarketing reports and observational studies, which carry inherent limitations that decrease the strengths of those studies. Notably, among the database studies, it was found that there was remarkable notoriety bias for the AEs reported [19]. It is important to note that obesity increases the rates of pancreatitis and pancreatic cancer risk factors (e.g., gallstones, hypertriglyceridemia, biliary disease). Furthermore, since T2DM is a risk factor for pancreatic disease, it can be difficult to differentiate whether pancreatitis or pancreatic cancer is attributable to the underlying condition or the GLP‐1 RA. Thus, patient and treatment selection provide another source of imbalance and confounding across these studies. Other systematic reviews and meta‐analyses found no significant findings for pancreatitis or pancreatic cancer. Cao et al. collected data from seven cardiovascular outcome trials with 56,004 patients with T2DM and a median follow‐up ranging from 1.3 to 5.4 years, with no significant difference compared to placebo for acute pancreatitis (OR: 1.05, 0.78–1.40, p = 0.76) or pancreatic cancer (OR: 1.12, 0.77–1.63, p = 0.56) [19]. Similarly, a 2023 meta‐analysis with 43 trials found that pancreatitis (OR: 1.24, 0.94–1.64, p = 0.13) and pancreatic cancer (OR: 1.28, 0.87–1.89, p = 0.20) displayed no clear association for GLP‐1 RA usage [20]. Most recently, a 2024 population‐based cohort study following 543,595 patients at over 7 years of follow‐up supported no increased risk for pancreatic cancer (OR: 0.50, 0.15–1.71) [21]. However, an evaluation of the FDA Adverse Event Reporting System (FAERS) database from 2004 to 2020 found a significantly increased (proportional reporting ratio: 9.86) pancreatic cancer cases associated with GLP‐1 RAs compared to the other glucose‐lowering agents analysed in the study [22]. Notably, the authors stated that the combination of GLP‐1 RAs with DPP‐4 inhibitors may have led to an increased reporting of several tumours, including pancreatic cancer [22]. Interestingly, a 2024 study presented at the Endocrine Society's annual meeting evaluating 638,501 patients with a history of acute pancreatitis suggested that GLP‐1 RAs reduce the risk of recurrent acute pancreatitis [23]. However, these findings have not been published, and further research is required to detail these preliminary results. Regardless, this possibility will offer more options for patients that will help improve patient outcomes and quality of life.

However, these findings must be understood within the context of their limitations. First, as mentioned earlier, there was heterogeneity in the included studies in terms of GLP‐1 RA dosage/type, study design, age (range of 14.4 to 68 years), follow‐up time (range of 1 to 198 weeks), and background medication usage. Most importantly, this heterogeneity was seen in the definition of pancreatitis, which was not consistently reported and was determined through investigator or independent committees. Second, we included only RCTs, which provide high‐quality evidence but generally have a shorter duration, which may not be long enough to fully capture the risk of pancreatic cancer. In contrast, databases and pharmacovigilance, such as FAERS, suggest increased pancreatic cancer risk, but these data are subject to reporting bias, preventing the determination of causality. Thus, the discrepancy between FAERS and the findings of this study highlights the need for long‐term observational studies to further clarify this risk. Third, a key limitation of the meta‐analysis is the exclusion of studies with zero events in both arms from the pooled risk ratio calculations. This was done due to the risk ratio becoming undefined if no events occur in either arm; these studies would not provide data to the pooled estimate and were excluded. However, this limits the comprehensiveness of the pooled values and can overestimate the overall relative risk as the excluded trials likely could dilute the observed risk ratio. RevMan software prevented smaller decimal precision of very small effect sizes, as we originally intended to use risk difference calculation, which could better capture the low incidence across all included studies. Fourth, despite the high quality of RCTs, patients typically enrolled in RCTs are highly selected and may not be representative of the general T2DM or obese population, possibly limiting the external validation of these findings. Fifth, although we included trials with at least a 1‐week follow‐up to maximise the capture of acute events of pancreatitis, this would limit the detection of longer‐term outcomes such as chronic pancreatitis and pancreatic cancer. Long‐term observational studies, real‐world evidence, or pharmacovigilance are needed to reliably test these outcomes. To test this, we conducted a sensitivity analysis with trials that had at least 12 weeks of follow‐up to assess for differences in pancreatitis. Sixth, patients with a history of pancreatitis or risk factors for pancreatitis are usually excluded during the screening phase. It is important to note that acute pancreatitis secondary to drugs is a diagnosis of exclusion. Thus, cases that are deemed idiopathic may very well be due to drug exposure as well. Additionally, a lack of classic symptoms or a lack of evolution of symptoms may lead to further underdiagnosis.