Bottom line: A new GLP-1 safety study published in 2026 found associations between weight-management treatment with semaglutide or liraglutide and several categories of medical outcomes, including psychiatric disorders, anxiety, depressive disorders, gastrointestinal dysmotility or obstruction, and vision impairment. Liraglutide was also associated with hepatic and pancreatobiliary outcomes. The findings are important because they come from routine electronic-health-record data across 13 South Korean hospitals, but they do not prove that either drug caused the events.
The study should be read as a real-world safety-signal report, not as a definitive risk calculation. Its results sit alongside, rather than automatically replacing, randomized-trial evidence, product labeling, and regulatory reviews. In particular, the findings about psychiatric diagnoses should not be misreported as evidence that GLP-1 receptor agonists cause suicidal behavior. In January 2026, the U.S. Food and Drug Administration said its own review did not identify an increased risk of suicidal ideation or behavior with GLP-1 receptor agonists.
This distinction matters as semaglutide, liraglutide, tirzepatide, and newer incretin medicines move from specialist use into broad population-level treatment. Large exposure creates more opportunities to detect uncommon events, but it also creates more opportunities for confounding, diagnostic overlap, and statistical noise. The central question is therefore not whether one study has settled GLP-1 safety. It has not. The question is which signals deserve confirmation and how researchers should design the next studies.
What the GLP-1 safety study 2026 actually examined
The study, published in Diabetes, Obesity and Metabolism, used electronic health records from 13 South Korean hospitals covering the period from 2018 through 2025. Investigators used the OMOP Common Data Model and a new-user design, meaning the analysis focused on people initiating semaglutide or liraglutide for weight management rather than simply counting everyone who had ever received one of the medicines.
After propensity-score matching, the semaglutide analysis included 2,357 initiators compared with 22,602 matched non-initiators. The liraglutide analysis included 6,953 initiators compared with 68,001 matched non-initiators. The researchers assessed 10 safety outcomes and estimated site-specific hazard ratios before combining the results through meta-analysis.
This design has several strengths. New-user analyses reduce some of the bias that occurs when long-term users are mixed with people who have only recently started treatment. Propensity-score matching can also balance measured differences between treatment groups, such as age, diagnoses, medications, and healthcare history. The multicentre structure provides a broader evidence base than a single hospital or one clinician’s case series.
However, electronic-health-record studies remain observational. The records were not created primarily for research, diagnoses may be incompletely coded, and important factors may not be captured with enough detail. People starting a weight-management medicine may differ from non-initiators in motivation, healthcare access, screening frequency, baseline disease severity, diet, physical activity, mental-health history, and willingness to report symptoms. Matching can address measured variables; it cannot fully remove unmeasured differences.
The full peer-reviewed report is available through PubMed.
The main findings: psychiatric, gastrointestinal, and vision signals
For semaglutide, the study reported associations with five outcomes. The strongest estimate involved gastrointestinal dysmotility or obstruction, with a hazard ratio of 3.91 and a 95% confidence interval of 1.42 to 10.82. Semaglutide was also associated with psychiatric disorders overall, anxiety disorder, depressive disorder, and vision impairment.
The reported hazard ratios for semaglutide were:
| Outcome | Reported hazard ratio | 95% confidence interval | Interpretation |
|---|---|---|---|
| Psychiatric disorders overall | 2.02 | 1.30–3.14 | Association with higher recorded risk in the matched analysis |
| Anxiety disorder | 2.39 | 1.22–4.71 | Association, with substantial uncertainty around magnitude |
| Depressive disorder | 3.42 | 1.51–7.74 | Association, not proof of drug causation |
| Gastrointestinal dysmotility or obstruction | 3.91 | 1.42–10.82 | Largest reported estimate, with a wide interval |
| Vision impairment | 1.58 | 1.04–2.41 | Modest association that was not consistent in every sensitivity analysis |
The confidence intervals are as important as the headline estimates. A hazard ratio of 3.91 may sound precise, but an interval extending from 1.42 to 10.82 indicates that the underlying effect could be materially smaller—or considerably larger—within the study’s statistical uncertainty. That is one reason why the result should prompt confirmation rather than immediate causal conclusions.
The vision finding also requires caution. The investigators noted that the semaglutide association with vision impairment was not statistically significant in some sensitivity analyses. A signal that changes under alternative modeling assumptions is not useless, but it is less robust than a finding that remains stable across multiple prespecified analyses.
What the study found for liraglutide
Liraglutide showed a different pattern. It was associated with psychiatric disorders overall, anxiety disorder, depressive disorder, hepatic impairment, pancreatobiliary disorder, pancreatitis or cholangitis or cholelithiasis, and vision impairment.
The reported hazard ratios included 1.66 for psychiatric disorders overall, 1.68 for anxiety disorder, and 1.51 for depressive disorder. The study reported a hazard ratio of 1.46 for hepatic impairment, 1.33 for pancreatobiliary disorder, 1.40 for the combined pancreatitis, cholangitis, or cholelithiasis outcome, and 1.34 for vision impairment.
These results do not establish that liraglutide is more dangerous than semaglutide. The drugs were not simply placed into a single head-to-head safety ranking in the primary analysis. The populations, prescribing patterns, observation periods, clinical indications, and baseline characteristics may differ. A larger hazard ratio for one drug and one outcome cannot be converted into a general statement that one GLP-1 medicine is safer overall.
The pattern does, however, illustrate why pharmacovigilance needs to examine individual agents rather than treating all GLP-1 receptor agonists as interchangeable. Semaglutide and liraglutide share a broad pharmacologic class, but they differ in molecular structure, half-life, dosing schedules, exposure profiles, clinical populations, and the historical periods in which they were used. Those differences can influence both biology and how outcomes are detected in medical records.
Why the psychiatric findings need careful interpretation
The psychiatric findings are likely to generate the greatest public attention. Headlines that translate the study into “GLP-1 drugs cause depression” would go beyond the evidence. The study identified recorded diagnoses of psychiatric disorders, anxiety, and depression. It did not establish a direct mechanism, demonstrate a dose-response relationship, or prove that the medication preceded every clinically relevant change in mental health.
People seeking weight-management treatment may already have elevated rates of depression, anxiety, disordered eating, sleep problems, chronic pain, social stress, or metabolic disease. They may also have more contact with clinicians after beginning treatment, increasing the probability that a psychiatric condition is recognized and coded. This is a form of detection bias: the treatment group may be watched more closely than the comparison group.
The study also does not overturn the FDA’s January 2026 safety communication. After reviewing 91 placebo-controlled GLP-1 trials involving 107,910 participants, along with FDA Sentinel data covering more than 2.2 million users, the FDA reported no increased risk of suicidal ideation or behavior. The FDA also said the totality of available evidence did not support a causal relationship between GLP-1 receptor agonists and suicidal ideation or behavior.
The two findings can coexist. A real-world study may detect an association with broad psychiatric diagnoses while a regulatory review finds no increased risk of suicidal behavior. They are measuring different outcomes, using different comparators and data sources, and asking different questions. Broad psychiatric diagnoses are not equivalent to suicide attempts or completed suicide.
This article does not provide individual medical guidance. New or worsening mental-health symptoms and possible emergencies are clinical matters that require timely assessment through appropriate local health or emergency services.
Why the gastrointestinal signal matters
The gastrointestinal finding is biologically plausible because GLP-1 receptor agonism affects gastric emptying, appetite, nausea pathways, and intestinal motility. But biological plausibility is not the same as proof that the observed association was caused by semaglutide.
The study combined gastrointestinal dysmotility and obstruction into one outcome. Those conditions are clinically related but not identical. A database code for dysmotility may represent a different level of severity from a confirmed mechanical obstruction requiring imaging, hospitalization, or surgery. Without detailed chart review or adjudication, the analysis cannot always distinguish the underlying clinical circumstances.
The wide confidence interval also matters. The association was statistically notable, but the estimate was imprecise. Future studies should separate confirmed obstruction, ileus, gastroparesis, severe constipation, emergency-department visits, hospitalization, and outpatient symptom codes. They should also account for prior gastrointestinal disease, opioid exposure, abdominal surgery, diabetes duration, and other medications that influence motility.
This is a useful example of why a safety signal is not automatically a contraindication. It is a prompt to ask better questions: Which patients are at greatest baseline risk? Does risk vary by dose or duration? Are events concentrated during initiation or escalation? What proportion is reversible after treatment interruption? Are there differences between semaglutide, liraglutide, and tirzepatide? A single database study cannot answer all of those questions.
How the findings fit with existing GLP-1 evidence
Randomized clinical trials remain the strongest tool for estimating treatment effects because randomization balances known and unknown confounders. They are not perfect for rare adverse events, however. Trial participants may be healthier, more closely monitored, and less diverse than people treated in routine practice. Trials may also be too short or too small to identify uncommon outcomes.
Real-world studies fill part of that gap. They can include older adults, people with multiple conditions, patients who discontinue treatment, and treatment patterns that would not fit a tightly controlled protocol. Their weakness is that treatment is not assigned at random. The most reliable safety picture comes from triangulation: randomized trials, active-comparator observational studies, regulatory databases, medical-record review, and mechanistic research pointing in the same direction.
Readers can use the same evidence hierarchy when evaluating claims about semaglutide or liraglutide. A single adverse-event report may identify a possibility but cannot estimate frequency. A large database study can identify associations but may retain confounding. A randomized trial can provide stronger causal inference but may lack power for rare events. A regulatory review can integrate multiple sources, although its conclusions depend on the data available to the agency.
GLP1Place’s existing review of peptide side effects and anecdotal claims provides useful background on this evidence hierarchy. For a broader explanation of endpoints, controls, and statistical interpretation, see the guide to reading a GLP-1 clinical trial.
Research and regulatory implications
The immediate implication is methodological rather than a change in treatment standards. Drug developers, regulators, and health systems have another reason to invest in postmarketing surveillance that can distinguish genuine safety signals from differences in patient populations.
Studies that report only average adverse-event rates across a full trial can miss clinically meaningful timing patterns. Future programs could combine electronic health records with pharmacy claims, laboratory data, hospitalization records, imaging, ophthalmology records, and standardized psychiatric assessments. Such linked datasets would be better positioned to distinguish a coded diagnosis from a clinically confirmed event.
The study does not change FDA approval status, prescribing information, or official treatment recommendations by itself. It is a peer-reviewed observational publication, not an FDA safety communication or a labeling action.
The FDA’s separate January 2026 communication is especially important for interpreting the psychiatric findings. The agency requested removal of suicidal-behavior and suicidal-ideation language from the labeling of Saxenda, Wegovy, and Zepbound after concluding that its comprehensive review did not identify an increased risk. That conclusion should not be misrepresented as proof that every psychiatric outcome is impossible. It addresses a specific safety question—suicidal ideation and behavior—and is based on a separate evidence review.
Compliance teams, publishers, clinicians, and health educators should avoid converting observational hazard ratios into absolute individual risk. They should also avoid language implying that semaglutide or liraglutide are proven to cause depression, obstruction, liver injury, or vision loss on the basis of this study alone.
Researcher takeaways
- Replicate the psychiatric findings with active comparators. Comparisons with other evidence-based weight-management medicines may reduce confounding by indication more effectively than comparisons with non-initiators.
- Report absolute risks. Relative measures such as hazard ratios are difficult to interpret without event rates, person-time, and the number needed to harm or screen.
- Separate clinical outcomes. Gastrointestinal dysmotility, gastroparesis, ileus, mechanical obstruction, constipation, and emergency visits should not be treated as interchangeable endpoints.
- Use adjudication where possible. Chart review, imaging, laboratory results, and specialist confirmation can improve outcome validity.
- Study timing. Analyses should examine initiation, titration, maintenance, discontinuation, and re-exposure periods separately.
- Address detection bias. More frequent visits, medication counseling, and screening may increase diagnosis rates among treatment initiators.
- Compare molecules carefully. Semaglutide, liraglutide, and tirzepatide should not be treated as identical exposures simply because they affect incretin pathways.
What happens next
The next stage should be confirmation, not panic. Researchers will need to test whether the signals persist in other countries, healthcare systems, age groups, and treatment populations. The most informative designs will likely include active comparators, validated outcomes, longer follow-up, and transparent reporting of absolute event rates.
It will also be important to reconcile apparently different findings. The 2026 cohort study identified associations with broad psychiatric outcomes, while the FDA found no increased risk of suicidal ideation or behavior. Those results are not necessarily contradictory because they address different endpoints. A mature safety discussion should preserve that distinction instead of selecting whichever finding best fits a predetermined narrative.
For now, the study’s strongest contribution is that it expands the list of questions surrounding real-world GLP-1 safety. It supports continued monitoring of psychiatric diagnoses, gastrointestinal motility, vision, hepatic outcomes, and pancreatobiliary disease. It does not establish that semaglutide or liraglutide caused those outcomes, and it does not justify broad claims that GLP-1 medicines are either universally dangerous or universally risk-free.
Frequently Asked Questions
What did the 2026 GLP-1 safety study find?
The study found that semaglutide initiation was associated with higher rates of psychiatric disorders, anxiety, depressive disorders, gastrointestinal dysmotility or obstruction, and vision impairment compared with matched non-initiators. Liraglutide was associated with psychiatric, hepatic, pancreatobiliary, pancreatitis or cholangitis or cholelithiasis, and vision outcomes.
Does the study prove that semaglutide or liraglutide caused these problems?
No. It was a retrospective observational cohort study using electronic health records. Even with propensity-score matching and sensitivity analyses, residual confounding, differences in healthcare use, coding practices, and other biases can remain.
Did the study find an increased risk of suicide from GLP-1 drugs?
No. The study assessed psychiatric disorders, anxiety, and depressive disorders, not a definitive causal increase in suicidal behavior. Separately, the FDA reported in January 2026 that its review did not identify an increased risk of suicidal ideation or behavior with GLP-1 receptor agonists.
Were semaglutide and liraglutide compared directly with each other?
The primary analysis compared initiators of semaglutide or liraglutide with matched non-initiators. It did not establish that one drug was safer than the other across every outcome.
What gastrointestinal signal was reported for semaglutide?
Semaglutide initiation was associated with a hazard ratio of 3.91 for the study’s combined outcome of gastrointestinal dysmotility or obstruction. The confidence interval was wide, indicating uncertainty around the exact size of the association.
What should researchers take from this study?
The main value is signal detection. The findings support additional active-comparator, longer-duration, and adjudicated studies that separate new disease from medication-related events and report absolute risks alongside hazard ratios.