Does Ozempic Cause Depression? What the Evidence Actually Shows

Few GLP-1 safety questions generate more anxious searching than this one, and the honest answer requires holding two real but very different pieces of evidence at once — a spontaneous-report signal that looks alarming in isolation, and a more rigorous comparative study that found the opposite. Neither should be quietly dropped in favor of the other.

The pharmacovigilance signal

A 2025 study analyzed VigiBase, the World Health Organization's global pharmacovigilance database, covering spontaneous adverse event reports from inception through January 2024¹. It calculated reporting odds ratios (ROR) — a measure of how disproportionately often a drug appears in reports of a given event, relative to all other drugs in the database — and found elevated RORs for suicidal ideation with semaglutide (5.82), liraglutide (4.03), and tirzepatide (2.25), with an even higher figure for reports combining depression and suicidality specifically for semaglutide (14.74)¹. Read in isolation, those numbers look concerning. But the same analysis found something that complicates a simple "GLP-1s cause suicidality" reading: reporting odds ratios for completed suicide and suicide attempts were significantly decreased across the same drugs — semaglutide's ROR for suicide attempts was 0.11 and for completed suicide was 0.01¹. The study's authors were explicit about what a pharmacovigilance signal like this can and can't establish: "Causation between GLP-1 RA exposure and suicidality (either increased or decreased) cannot be ascertained from ROR data"¹ — a spontaneous-report database captures what gets reported, not a controlled comparison of what actually happens to similar people on and off the drug.

The more rigorous comparative study

A separate 2024 study took the harder, more reliable approach: a propensity-weighted, population-based cohort study using Spanish population-wide health databases covering five million people, comparing new users of GLP-1 receptor agonists against new users of SGLT-2 inhibitors (a different diabetes drug class, chosen as an active comparator rather than no treatment) among people with type 2 diabetes and obesity². This design — propensity score weighting to balance the two groups on measurable characteristics, and an active comparator rather than a raw population baseline — is specifically built to reduce the confounding that a spontaneous-report database can't control for. Among 3,040 GLP-1 users and 11,627 SGLT-2 inhibitor users, the hazard ratio for suicidal ideation and self-injury was 1.04 (95% CI, 0.35-3.14) in the per-protocol analysis and 1.36 (95% CI, 0.51-3.61) in the intention-to-treat analysis² — both confidence intervals wide and crossing 1.0, meaning no statistically significant difference was detected. The authors' conclusion was direct: "Our findings do not support an increased risk of SIS when taking GLP-1RA in individuals with type 2 diabetes and obesity"², while appropriately cautioning that the rarity of the events and the wide confidence intervals mean a small true effect can't be fully ruled out.

Why these two studies point in different directions, and which one to weight more

This isn't a case of one study being "right" and the other "wrong" — they're measuring different things with different levels of reliability. A pharmacovigilance database captures raw reporting volume, which is shaped heavily by media attention, prescriber awareness, and how motivated a patient is to report a symptom once GLP-1s and mental health became a widely-discussed topic — all of which can inflate reporting for a drug in the news without reflecting a true increase in the underlying event rate. A propensity-weighted cohort study with an active comparator is specifically designed to correct for that kind of reporting bias by comparing actual outcomes between two similar, real-world patient groups. That's why regulatory bodies and researchers generally treat a pharmacovigilance signal as a prompt for further, more rigorous study — exactly what happened here — rather than as proof on its own. The cohort study is the more reliable of the two for answering "does this actually happen more" specifically because of its design, even though the raw reporting signal is real and worth taking seriously as a reason the question got studied properly in the first place.

What this means if you have a personal or family history of depression

Neither study is a reason to avoid a GLP-1 if you have a history of depression, but it is a reason to disclose that history to your prescriber before starting, so any mood changes get tracked against a known baseline rather than attributed to guesswork after the fact — the same disclosure principle that applies to every other item on your medical history, covered in GLP-1 drug interactions: what to tell your prescriber. If you experience new or worsening depressive symptoms after starting, that's worth reporting to your prescriber regardless of what the population-level data shows, since population averages don't rule out an individual reaction.

Where this fits alongside the class's other safety signals

This sits alongside the other rare, harder-to-pin-down safety questions covered in GLP-1 boxed warnings: what the label actually says — genuine signals worth knowing, evaluated with the same rigor rather than either dismissed or overstated. If alcohol use or mood are part of why you're considering or hesitant about a GLP-1, see GLP-1 and alcohol: what the evidence shows for the separate, and considerably more positive, evidence on GLP-1s and alcohol cravings.

Where this fits your provider decision

A program that never asks about mental health history as part of intake is skipping a legitimate part of informed consent here, even though the best current evidence doesn't support an increased suicidality risk. See how we score real medical oversight in our Value Score methodology, and compare pricing across verified programs on our semaglutide and tirzepatide price boards.

This piece sits in our research index alongside every other sourced explainer we publish — the trials, the compounding rules and the pricing mechanics, grouped by the question each one answers.