GLP-1 Agonist Adverse Events: What Systematic Reviews Show
Semaglutide Long-Term Safety
1.46x gallstone risk
A meta-analysis of 76 randomized controlled trials found GLP-1 receptor agonists increased the relative risk of cholelithiasis by 1.46 compared to placebo, translating to approximately 2 additional cases per 1,000 treated patients.
He et al., JAMA Internal Medicine, 2022
He et al., JAMA Internal Medicine, 2022
View as imageGLP-1 receptor agonists have become among the most prescribed drug classes in the world, used for type 2 diabetes and obesity. With tens of millions of patients now taking semaglutide, liraglutide, tirzepatide, and related compounds, their safety profile is under intense scrutiny from researchers and regulators. This article synthesizes what systematic reviews and meta-analyses have found about GLP-1 agonist adverse events across organ systems, distinguishing signals supported by pooled data from those that remain uncertain. For a broader overview of how the class works, see every GLP-1 receptor agonist compared. For long-term semaglutide-specific data, see our pillar article on semaglutide long-term safety.
Key Takeaways
- A 2025 Gastroenterology meta-analysis of 39 RCTs confirmed GLP-1RAs increase nausea, vomiting, diarrhea, and constipation as a class effect, with symptom onset at a median of 1 day after initiation (Chiang et al., 2025)
- GLP-1RAs increased cholelithiasis risk by 46% (RR 1.46) across 76 RCTs, translating to ~2 extra cases per 1,000 patients treated (He et al., JAMA Internal Medicine, 2022)
- Pooled cardiovascular outcome trial data showed no significant increase in acute pancreatitis with GLP-1RA treatment (Cao et al., Endocrine, 2020)
- A 2026 JCI review found no statistically significant association between GLP-1RAs and suicidality in large pharmacoepidemiologic studies (Jalleh et al., 2026)
- FAERS data from 2004-2021 identified disproportionate reporting of thyroid and pancreatic tumors with GLP-1RAs, though reporting bias limits causal interpretation (Yang et al., 2022)
- A semaglutide-specific meta-analysis found the drug was not associated with increased risk of serious adverse events overall, but GI events were dose-dependent (Rivera et al., 2024)
Gastrointestinal adverse events: the dominant signal
Gastrointestinal symptoms are the most common and most studied adverse events across all GLP-1 receptor agonists. Chiang et al. (2025) published a systematic review and meta-analysis in Gastroenterology examining 39 RCTs of GLP-1RAs, establishing these as a class-wide effect.[1]
The four most frequently reported GI adverse events: nausea, vomiting, diarrhea, and constipation. These occur at significantly higher rates than placebo across all approved GLP-1RAs. The median time-to-onset is 1 day after initiation, and symptoms are most pronounced during dose escalation. Most GI events are mild to moderate and decrease over weeks as patients titrate upward.
Ismaiel et al. (2025) conducted a network meta-analysis specifically in non-diabetic patients with overweight or obesity. Their findings showed that GI adverse event profiles differ between agents: tirzepatide showed the highest risk of nausea and diarrhea, dulaglutide and lixisenatide the lowest nausea risk, and exenatide the highest vomiting risk.[2] This agent-specific variation matters clinically because switching between GLP-1RAs can sometimes resolve intolerable GI symptoms.
Serious gastrointestinal events are rarer. Chiang et al. (2025) found GLP-1RAs increased the risk of gastroesophageal reflux disease (RR ~2.19 vs placebo) and cholelithiasis, but did not find statistically significant increases in gastroparesis, bowel obstruction, or ileus in the pooled RCT data.[1] For a deeper dive into the gastroparesis question specifically, see gastroparesis and GLP-1 drugs.
The clinical reality is that GI side effects are the primary driver of treatment discontinuation. Rivera et al. (2024) found in their semaglutide-specific meta-analysis that GI events were dose-dependent, with higher doses (2.4 mg weekly) producing more nausea and vomiting than lower doses (1.0 mg weekly).[3] Gradual dose escalation remains the standard mitigation strategy.
Gallbladder and biliary disease
He et al. (2022) published what remains the most cited meta-analysis on GLP-1RAs and gallbladder risk in JAMA Internal Medicine. Pooling data from 76 RCTs with over 103,000 participants, they found GLP-1RA treatment increased the relative risk of gallbladder and biliary diseases by 1.46 (95% CI 1.22-1.74) compared to placebo.[4]
The breakdown by specific condition: cholelithiasis (gallstones) RR 1.27, cholecystitis RR 1.36, and biliary disease overall RR 1.46. The absolute risk increase was small, approximately 2 additional cases of cholelithiasis per 1,000 patients treated. Risk was higher with weight-loss doses compared to glucose-lowering doses, and increased with longer treatment duration.
The proposed mechanism is pharmacologically straightforward. GLP-1RAs slow gallbladder motility and emptying. Reduced gallbladder contraction allows bile to concentrate, promoting stone formation. Rapid weight loss itself (regardless of mechanism) is an independent risk factor for gallstones, so disentangling the drug effect from the weight loss effect is methodologically challenging.
Pancreatitis and pancreatic cancer
The pancreatitis question has followed GLP-1RAs since the class emerged. Early pharmacovigilance signals raised concern, but pooled trial data tells a different story.
Cao et al. (2020) pooled data from seven large cardiovascular outcome trials (LEADER, SUSTAIN-6, PIONEER 6, REWIND, HARMONY, ELIXA, and EXSCEL), which together enrolled over 56,000 patients followed for a median of 2-5 years. They found no statistically significant increase in acute pancreatitis (RR 0.93, 95% CI 0.65-1.34) or pancreatic cancer (RR 0.94, 95% CI 0.52-1.70) with GLP-1RA treatment.[5]
Wen et al. (2025) updated this analysis with additional RCTs, examining pancreatitis and pancreatic cancer rates across the full GLP-1RA class. Their meta-analysis confirmed the Cao findings: no statistically significant increase in either outcome across the pooled dataset.[6]
These null findings carry important caveats. Cardiovascular outcome trials typically enrolled patients for 2-5 years. Pancreatic cancer has a long latency period, and a decade or more of exposure data may be needed to detect a true signal. Additionally, trial populations may not reflect the broader population now using these drugs for obesity. The question is not fully settled; it is simply that available RCT data does not support the signal.
For the full evidence breakdown on the pancreatitis question, see the sibling article on GLP-1 agonists and pancreatitis: what the meta-analyses conclude. A related article from a different cluster also covers the broader pancreatitis evidence.
Thyroid cancer: rodent signal, uncertain human relevance
GLP-1RAs carry a boxed warning for medullary thyroid carcinoma (MTC) based on rodent data showing thyroid C-cell tumors in rats and mice exposed to liraglutide and semaglutide. The critical question is whether this signal translates to humans.
Yang et al. (2022) analyzed the FDA Adverse Event Reporting System (FAERS) database from 2004 to 2021 and found disproportionate reporting of thyroid cancer with GLP-1RAs compared to other drug classes.[7] However, FAERS data is subject to reporting bias, notoriety bias (more reports after media attention), and cannot establish causation.
Jalleh et al. (2026) addressed the thyroid question in their comprehensive JCI review, noting that the rodent signal relates to species-specific differences in GLP-1 receptor expression on thyroid C-cells. Rodents have much higher C-cell GLP-1 receptor density than humans. To date, no RCT or large observational study has demonstrated a statistically significant increase in MTC in humans treated with GLP-1RAs.[8] The boxed warning remains appropriate as a precaution, particularly for patients with a personal or family history of MTC or Multiple Endocrine Neoplasia type 2, but the evidence base for human thyroid cancer risk is rodent-derived.
The sibling article on GLP-1 agonists and thyroid cancer: rodent signal vs human evidence covers this topic in full detail.
Psychiatric safety and suicidality
Reports of suicidal ideation in patients taking semaglutide generated significant media attention in 2023-2024. This prompted multiple systematic analyses.
Jalleh et al. (2026) reviewed the available evidence in the JCI and concluded that large-scale pharmacoepidemiologic studies have not demonstrated a statistically significant association between GLP-1RAs and suicidality. A meta-analysis of randomized controlled trials found no increase in suicides or suicidal behavior with GLP-1RA treatment.[8]
Chen et al. (2025) conducted a specific meta-analysis of GLP-1RA impact on suicide behavior based on randomized controlled trials. Their pooled analysis did not find a significant increase in suicidal ideation or behavior with GLP-1RA use compared to placebo.[9]
Alansari et al. (2025) published a separate meta-analysis in Medicine that also assessed the association between GLP-1 agonists and suicidal ideation. Their findings were consistent with the Chen analysis: no statistically significant signal in the pooled RCT data.[10]
Three independent analyses reaching the same conclusion strengthens confidence in the null finding. However, these meta-analyses rely on RCT data where psychiatric events were secondary endpoints collected through standard adverse event reporting, not through validated psychiatric instruments. Patients with active psychiatric illness are typically excluded from obesity and diabetes RCTs. Whether there is a subpopulation at genuine risk (e.g., patients with pre-existing mood disorders) is a question that RCT meta-analyses cannot definitively answer.
The comprehensive safety picture: the Jalleh JCI review
Jalleh et al. (2026) published what may be the most thorough single review of GLP-1RA safety to date in the Journal of Clinical Investigation. Their synthesis covers adverse events across every organ system.[8]
Key findings from the review:
- GI events are the most common reason for dose reduction or discontinuation, affecting 30-50% of patients at some point during treatment
- Injection site reactions occur in 1-20% of patients depending on agent and formulation, and rarely lead to discontinuation
- Diabetic retinopathy worsening has been observed specifically with semaglutide in patients with pre-existing retinopathy, likely related to rapid glycemic improvement rather than a direct drug effect
- Acute kidney injury reports exist but are predominantly associated with severe GI events (dehydration from vomiting/diarrhea) rather than direct nephrotoxicity
- Cardiovascular effects are overwhelmingly positive: the major cardiovascular outcome trials showed reduced MACE (major adverse cardiovascular events) with GLP-1RA treatment
The review emphasizes that the risk-benefit calculation for most patients favors treatment, given the substantial cardiovascular and metabolic benefits demonstrated in large trials. The challenge is identifying and managing the subset of patients who experience intolerable or serious adverse events.
What the FAERS data adds (and its limits)
Yang et al. (2022) analyzed 16,923 GLP-1RA-related adverse event reports in the FAERS database from 2004 to 2021. They found disproportionate reporting signals for thyroid tumors and pancreatic tumors compared to background rates for other drug classes.[7]
FAERS data provides pharmacovigilance signals but cannot establish causation for several reasons:
- Reporting bias: adverse events for widely discussed drugs are more likely to be reported
- No denominator: FAERS captures reports, not incidence rates (the number of patients taking GLP-1RAs is unknown in the dataset)
- Confounders: obesity and diabetes are themselves risk factors for multiple cancers, and FAERS data cannot adjust for baseline risk
- Temporal distortion: media coverage of potential GLP-1RA cancer links likely increased reporting independent of any true signal
FAERS signals are best understood as hypotheses that require confirmation through controlled studies. The thyroid and pancreatic signals from Yang et al. have not been confirmed by pooled RCT data (Cao et al., 2020; Wen et al., 2025), which provide stronger causal evidence.
Limitations of the current evidence
The systematic review literature on GLP-1RA safety has structural limitations that affect all conclusions:
Follow-up duration. Most RCTs followed patients for 1-5 years. Cancer outcomes require longer observation windows. The safety profile at 10+ years of continuous use is unknown for any GLP-1RA.
Population selection. RCTs exclude patients with significant comorbidities, psychiatric illness, pregnancy, and advanced age. The real-world population now using these drugs is broader than trial populations.
Outcome ascertainment. Adverse events in most RCTs were captured through standard reporting, not through active surveillance protocols. Mild to moderate events and psychiatric symptoms may be underreported.
Publication bias. Systematic reviews can only synthesize published data. Negative findings (no safety signal) may be less likely to be published as standalone studies.
Class effects vs agent effects. Meta-analyses that pool all GLP-1RAs assume a class effect. Individual agents (exenatide, liraglutide, semaglutide, dulaglutide, tirzepatide) have different pharmacokinetic profiles, receptor binding affinities, and formulations. Agent-specific safety differences may be obscured by pooling. For how tirzepatide's dual agonism changes the risk picture, see the dedicated sibling article.
The Bottom Line
Systematic reviews and meta-analyses establish GI adverse events (nausea, vomiting, diarrhea, constipation) as the dominant and most consistently reported side effects of GLP-1 receptor agonists, with onset within days of initiation and dose-dependent severity. Gallbladder disease risk is modestly elevated (RR 1.46). Pancreatitis and pancreatic cancer have not shown significant increases in pooled cardiovascular outcome trial data, though follow-up duration may be insufficient. The thyroid cancer signal remains rodent-derived with no confirmed human translation. Three independent meta-analyses found no association between GLP-1RAs and suicidality. The overall safety profile supports the risk-benefit case for most patients, but 10+ year exposure data remains unavailable, and the real-world population is broader than RCT cohorts.