GLP-1 safety cohort: what it found

A 2026 South Korean electronic-health-record study found associations between starting weight-management semaglutide or liraglutide and several recorded safety outcomes. This was a retrospective cohort, not a randomized trial, so it cannot show that either medicine caused those outcomes. After matching, the analysis compared 2,357 semaglutide initiators with 22,602 non-initiators and 6,953 liraglutide initiators with 68,001 non-initiators^[1]. Readers can use our semaglutide overview for class background, but the paper's estimates need to be read within its observational design.

What the study compared

The researchers analyzed de-identified records from 13 secondary and tertiary hospitals mapped to a common data model. The semaglutide analysis covered October 2024 through September 2025 at five hospitals; the liraglutide analysis covered March 2018 through September 2025 at 12 hospitals. Adults were classed as initiators after a first qualifying prescription with no use of that same agent in the prior 90 days and at least 180 days of recorded history^[2]. Our semaglutide evidence summary explains why a new-user design is useful but does not make a database study randomized.

Non-initiators had no recorded use of the study drug at baseline or during follow-up. Their index date was their first encounter involving another prescription after the same history requirement. The team used propensity scores, a statistical method for balancing measured differences, and matched as many as 10 non-initiators to each initiator. Age, sex, comorbidities and recent hospital use were included. Most measured characteristics balanced after matching, although sleep apnea retained a small imbalance in the semaglutide comparison^[2]. The semaglutide safety section provides broader context, not a substitute comparator.

The reported associations

For semaglutide, the study reported higher hazards for the broad psychiatric-disorder outcome, anxiety, depression, gastrointestinal dysmotility or obstruction, and vision impairment. The estimates were HR 2.02 (95% CI 1.30 to 3.14), 2.39 (1.22 to 4.71), 3.42 (1.51 to 7.74), 3.91 (1.42 to 10.82), and 1.58 (1.04 to 2.41), respectively^[1]. A hazard ratio compares event rates over follow-up. It does not tell an individual reader their absolute risk, and it should not be read as evidence that semaglutide produced the event.

For liraglutide, higher hazards were reported for psychiatric disorders overall, anxiety, depression, hepatic impairment, pancreatobiliary disorder, a narrower pancreatitis-cholangitis-cholelithiasis composite, and vision impairment. The respective HRs were 1.66, 1.68, 1.51, 1.46, 1.33, 1.40, and 1.34^[1]. No significant association with all cancers or thyroid cancer was found for either agent. The two drug analyses used different hospitals and follow-up periods, so differences between their estimates are not head-to-head evidence. See the semaglutide research record for that distinction.

The median follow-up was 163 days for semaglutide initiators and 301 days for their comparators. It was 1,051 days for liraglutide initiators and 1,110 days for comparators^[2]. That large difference limits direct comparisons between medicines and leaves less time to observe uncommon or delayed events in the semaglutide cohort. Readers checking the semaglutide timeline should keep this exposure window in view.

Why psychiatric results need care

The paper's psychiatric-disorder endpoint was a broad diagnosis-code composite that included anxiety and depressive disorders. It was not the same as suicidal behavior or ideation. In the semaglutide analysis, the three self-harm or suicidal-ideation events included in that composite occurred among non-initiators, not initiators. In the liraglutide analysis, three such events occurred among initiators and 15 among non-initiators^[2]. Those small counts do not support a drug-specific conclusion about suicidality. The semaglutide overview does not alter that limitation.

This distinction also matters because the US Food and Drug Administration reported in 2026 that its review did not identify increased suicidal ideation or behavior with GLP-1 receptor agonists and requested removal of that warning from the relevant weight-management labels^[3]. The FDA finding and this cohort are not interchangeable: they used different evidence sets and endpoints. Anyone with new or worsening depression, suicidal thoughts, or unusual mood changes should seek professional help rather than trying to interpret a database estimate through the semaglutide profile.

Sensitivity analyses changed some signals

The authors repeated analyses after restricting the cohort to people meeting obesity-related criteria, expanding the washout for other weight-management medicines, requiring at least two study-drug prescriptions, and extending selected look-back windows. Semaglutide associations for psychiatric outcomes and gastrointestinal dysmotility or obstruction persisted across those checks. Its vision-impairment estimate was no longer statistically significant in several sensitivity analyses. Liraglutide's hepatic-impairment estimate was not significant after restriction to obesity-related criteria^[2]. These changes are a reason to avoid turning the primary estimates into settled safety claims on a semaglutide safety page.

The vision outcome was also broad. It covered diagnosis codes involving visual function, visual pathways, eye movement, and related structures while excluding selected congenital and diabetes-related disorders. It was not a study of one specific eye disease. A higher estimate for that composite therefore should not be rewritten as proof that semaglutide causes a named eye condition^[2].

What the cohort cannot establish

Matching can balance recorded characteristics, but it cannot remove unmeasured differences in why people start treatment, how their health is monitored, or which diagnoses get entered into the record. Body mass index was too incomplete to include in the main propensity model. The comparator index depended on another prescription, and those medicines could have affected outcomes. A 180-day look-back may also have missed older conditions^[2]. These limits leave residual confounding, meaning differences other than treatment may explain some or all of an association. Our semaglutide methods context should be read with the same caution.

Other limits include small counts for rare outcomes, reliance on diagnosis codes, and the short semaglutide follow-up. Because weight-management GLP-1 medicines were not reimbursed under South Korea's national insurance system, prescribing was concentrated in secondary and tertiary hospitals. Generalizability to primary care or health systems with different access patterns is uncertain^[2]. Korean public agencies funded the study; one author reported research funding from Pfizer, Johnson & Johnson, and GSK outside the submitted work. Other authors declared no conflicts.

What we do not yet know

The cohort raises questions that need replication in other databases and designs with better control of body mass index, weight change, treatment persistence, healthcare-seeking behavior, and baseline mental-health status. Longer semaglutide follow-up is especially important. Research should also separate broad diagnostic composites from clinically specific outcomes and report absolute risks alongside relative estimates. Until then, this paper is a safety signal for further study, not a reason to start, stop, or switch a prescription without a clinician. The semaglutide regulation overview does not change that boundary.

Medical disclaimer: This article is for educational and journalistic purposes only and does not constitute medical advice. Peptides discussed may be classified as prescription medicines or research chemicals depending on your jurisdiction. Always consult a qualified healthcare professional before using any peptide product. PeptideMethods.com does not sell, distribute, or facilitate the sale of any peptide product.