Semaglutide Linked to Lower Risk of Worsening Depression and Anxiety in Large Swedish Cohort

A national register study using a self-controlled design finds GLP-1 receptor agonists, especially semaglutide, associated with reduced psychiatric hospitalisation and sick leave in nearly 96,000 people with depression or anxiety, but the observational design cannot establish causality and time-varying confounding remains a significant concern.

Why This Matters

Depression and anxiety disorders are among the most prevalent and disabling conditions worldwide, and their co-occurrence with metabolic disease, particularly type 2 diabetes and obesity, is well documented and clinically challenging. Patients living with both metabolic and psychiatric illness face compounding burdens: each condition worsens the prognosis of the other, treatment adherence suffers, and clinicians often must weigh psychiatric side effects when selecting metabolic therapies. GLP-1 receptor agonists have become one of the fastest-growing drug classes globally, driven by robust evidence for glycaemic control, cardiovascular benefit, and weight reduction, yet regulatory agencies have flagged potential psychiatric safety signals, creating genuine uncertainty about whether these drugs help or harm mental health. Preclinical work has identified GLP-1 receptors in brain regions implicated in mood regulation and reward, providing a biologically plausible basis for neuropsychiatric effects, but human evidence has remained sparse and inconsistent. This study arrives at a moment when millions of patients with comorbid metabolic and psychiatric conditions are either already taking or considering GLP-1 receptor agonists, and clinicians urgently need data to inform those conversations.

Clinical Summary

Depression and anxiety are disproportionately common in people with type 2 diabetes and obesity, with prevalence rates roughly twice those of the general population. The relationship is bidirectional: metabolic dysfunction promotes neuroinflammation and hypothalamic-pituitary-adrenal axis dysregulation, while psychiatric illness impairs self-care, diet, and physical activity. Clinicians treating this overlapping population face a persistent challenge: many metabolic therapies are psychiatrically neutral at best, and some carry their own neuropsychiatric risk profiles. The possibility that a metabolic drug could simultaneously improve psychiatric outcomes would represent a meaningful clinical advance, but previous evidence has been limited to small trials, retrospective claims analyses, and preclinical models.

This national register-based cohort study, published in JAMA Internal Medicine in 2025 by Markku Lähteenvuo and colleagues, examined whether GLP-1 receptor agonist use was associated with changes in the risk of worsening mental illness among people with pre-existing depression or anxiety. The mechanistic rationale draws on the presence of GLP-1 receptors in the hippocampus, amygdala, and prefrontal cortex, regions central to mood regulation, fear processing, and executive function. GLP-1 signalling has been shown in animal models to modulate neuroinflammation, promote neuroplasticity, and influence dopaminergic reward pathways. Beyond direct central nervous system effects, GLP-1 receptor agonists could plausibly improve psychiatric outcomes through indirect routes: weight loss reducing stigma and physical limitation, improved glycaemic control reducing fatigue and cognitive dysfunction, or enhanced gut-brain axis signalling altering neurochemical milieu. The study leveraged a within-individual, or self-controlled, design that compares each person’s outcomes during periods of GLP-1 receptor agonist use against their own outcomes during non-use periods, thereby eliminating stable between-person confounders such as genetics, childhood adversity, or baseline disease severity.

The cohort comprised 95,490 Swedish adults identified through linked national registers between 2009 and 2022. All had ICD-10-coded depression or anxiety and used at least one non-insulin antidiabetic medication during follow-up. Among them, 22,480 used a GLP-1 receptor agonist at some point. The primary outcome was a composite of psychiatric hospitalisation, sick leave exceeding 14 days for psychiatric reasons, hospitalisation for self-harm, or death by suicide. Semaglutide use was associated with a 42% lower hazard of this composite outcome compared with the same individuals’ non-use periods (adjusted hazard ratio 0.58, 95% confidence interval 0.51 to 0.65). Liraglutide showed a more modest 18% reduction (aHR 0.82, 95% CI 0.76 to 0.89). Critically, neither exenatide (aHR 1.01, 95% CI 0.69 to 1.46) nor dulaglutide (aHR 1.01, 95% CI 0.85 to 1.20) was associated with any change in risk, demonstrating substantial heterogeneity within the GLP-1 receptor agonist class. Semaglutide was additionally associated with reduced risk of worsening depression, worsening anxiety, and worsening substance use disorder when these outcomes were examined individually.

Despite the rigour of the within-individual design, several important limitations constrain interpretation. The self-controlled approach eliminates stable confounders but cannot address time-varying confounders: patients initiating semaglutide may simultaneously be receiving intensified psychiatric care, adjusting other medications, losing weight through dietary changes, or experiencing a natural remission of metabolic or psychiatric symptoms. The study’s reliance on register-based outcomes means that only severe episodes, those leading to hospitalisation, prolonged sick leave, or self-harm, were captured, while outpatient symptom improvement or worsening would be missed entirely. Ethnicity data were unavailable, limiting the generalisability of findings beyond the Swedish population. The heterogeneity across agents raises the question of whether the semaglutide signal reflects pharmacological specificity, differential dosing, or simply the era effect of semaglutide being prescribed more recently, during a period of potentially different psychiatric care practices. The authors appropriately conclude that randomised controlled trials are needed before any causal claims can be made, and they note that the findings should not alter current prescribing guidance.

Dr. Caplan’s Take

This study represents exactly the kind of large-scale, methodologically thoughtful observational work that moves a clinical question forward without prematurely answering it. The within-individual design is a meaningful step above standard cohort comparisons, and the fact that the investigators found clear heterogeneity across GLP-1 receptor agonists, rather than a uniform class effect, adds credibility. If all four agents had produced identical results, I would be more suspicious of unmeasured confounding driving the signal. The specificity of the semaglutide and liraglutide findings, and the dose-response pattern with semaglutide, at least suggests a pharmacological explanation worth pursuing.

That said, I see patients regularly who have already absorbed the headline version of studies like this. They arrive asking whether semaglutide will treat their depression, or whether they should request it from their psychiatrist. The honest answer is that we do not know. What this study shows is an association between periods of semaglutide use and reduced risk of severe psychiatric outcomes within the same person over time, but it cannot tell us why. Was it the drug acting on GLP-1 receptors in the brain? Was it the weight loss improving self-image and physical function? Was it the metabolic stabilisation reducing fatigue and cognitive fog? Was it the fact that people initiating a new medication tend to be in a period of greater overall clinical engagement? We simply cannot distinguish these explanations with the data available.

The evidence gap that matters most here is the absence of randomised controlled trials designed with psychiatric outcomes as the primary endpoint. We have a biologically plausible mechanism, a large and internally consistent observational signal for semaglutide specifically, and some corroboration from smaller studies, but that constellation of evidence has been wrong before in medicine. Time-varying confounding is particularly difficult to rule out in this context because GLP-1 receptor agonist initiation is rarely an isolated clinical event; it typically coincides with a broader shift in metabolic management, lifestyle counselling, and clinical monitoring. Until we see trial data with psychiatric endpoints, the association remains unproven rather than disproven, an important distinction that must be communicated clearly.

In my own practice, I treat this evidence as clinically informative but not directive. For a patient with comorbid type 2 diabetes and depression who needs a GLP-1 receptor agonist for metabolic indications, this study offers some reassurance that semaglutide is unlikely to worsen their psychiatric condition and may offer ancillary benefit. I do not prescribe semaglutide as a psychiatric intervention, and I am cautious about patients who seek it primarily for that purpose. What I do is monitor psychiatric symptoms systematically in patients starting GLP-1 receptor agonists, particularly during the first six months when both metabolic and psychiatric effects are likely to emerge. I also pay close attention to whether apparent psychiatric improvement tracks with weight loss and metabolic improvement or appears independent of it, as this distinction will matter enormously as the field designs the trials that are needed.

Clinical Perspective

This study sits at a pivotal inflection point in our understanding of GLP-1 receptor agonist neuropsychiatric effects. Previous evidence has been fragmented: small pilot trials of GLP-1 receptor agonists in depression, pharmacovigilance signals from adverse event reporting databases raising concern about suicidality, and preclinical studies demonstrating antidepressant-like effects in rodent models. The current study, with its large cohort, long follow-up, and within-individual design, substantially raises the evidentiary bar. It does not close the question, but it shifts the prior probability in favour of benefit rather than harm for semaglutide specifically. The heterogeneity across agents is among the study’s most important findings, as it argues against a simple class effect and underscores the need for agent-specific evaluation in future research.

For clinicians fielding patient questions, the appropriate communication strategy is nuanced. The evidence supports telling patients that current data from a large, well-designed observational study suggest that semaglutide use is associated with lower rates of severe psychiatric outcomes in people with depression or anxiety, but that this does not prove the drug treats depression or anxiety. It is reasonable to say that, for patients who have metabolic indications for semaglutide, the psychiatric signal is reassuring rather than concerning. It is not reasonable to prescribe semaglutide off-label for depression based on this study, nor to suggest to patients that it substitutes for evidence-based psychiatric treatment. The distinction between “this drug may not harm your mental health and might even help” and “this drug treats depression” is clinically and ethically significant.

Several pharmacological and safety considerations merit attention. Semaglutide’s gastrointestinal side effects, including nausea, vomiting, and reduced appetite, can interact with psychiatric medications that also affect appetite and gastrointestinal motility. Patients on SSRIs or SNRIs may experience compounded nausea during the semaglutide titration period. Rapid weight loss can alter the pharmacokinetics of lipophilic psychiatric medications, potentially requiring dose adjustments. Clinicians should also be aware that the study excluded patients with schizophrenia-spectrum and bipolar disorders, so these findings cannot be extrapolated to those populations. The dose-response signal for semaglutide, where higher doses appeared more strongly associated with benefit, is pharmacologically interesting but also raises the possibility of confounding by treatment duration or metabolic response magnitude.

The most actionable recommendation for clinical practice at present is to integrate systematic psychiatric symptom monitoring into the care of patients with comorbid metabolic and psychiatric conditions who are starting GLP-1 receptor agonists. Validated instruments such as the PHQ-9 for depression and GAD-7 for anxiety can be administered at baseline and at regular intervals during GLP-1 receptor agonist treatment. Documenting these trajectories creates a clinical dataset that has value both for individual patient care and for contributing to the broader knowledge base as registries and pragmatic trials begin to address this question more directly. Clinicians should also encourage eligible patients to participate in any emerging randomised trials evaluating GLP-1 receptor agonists for psychiatric outcomes, as these trials represent the only path to the causal evidence the field needs.

Study at a Glance

Study Type

National register-based cohort study with within-individual (self-controlled) design

Country

Sweden