GLP-1 and mood drugs: reading the signal

On 30 July 2026, Clinical Pharmacology & Therapeutics published an Australian analysis reporting that people who started a GLP-1 receptor agonist were less likely to go on to start an antidepressant (adjusted sequence ratio 0.85, 95% CI 0.76 to 0.94) or a medicine for substance use disorder (aSR 0.70, 95% CI 0.51 to 0.88)^[1]. Semaglutide was the most common drug in the sample. The finding is worth knowing. The study design is worth knowing better, because the design sets the ceiling on what the finding is allowed to mean.

We have already written up the GLP-1 neuropsychiatric datasets themselves, in our review of the tirzepatide and semaglutide psychiatric safety comparison and our write-up of the 2026 obesity cohort study. This piece is about the method underneath: prescription sequence symmetry analysis, what it is built to catch, and what it structurally cannot see.

What a sequence symmetry analysis actually measures

The design dates to a 1996 paper by Jesper Hallas in Epidemiology, which examined 11,244 people who had started an antidepressant and asked whether certain cardiovascular medicines tended to come first^[2]. The logic is simple enough to state in a sentence. Take only the people who ended up on both medicines, then count how many started drug A before drug B, and how many did the reverse. If the order is symmetric, there is no signal. If it tilts, something about drug A may be prompting drug B.

The appeal is that every person acts as their own comparison. As Hallas put it, confounders causing two drugs to be co-prescribed would rarely be expected to affect the symmetry^[2]. Anything fixed about a patient (sex, genetics, long-standing diagnoses, baseline mental health, socioeconomic position) sits on both sides of the comparison and cancels out. That is a real advantage over the propensity-matched cohort studies that dominate this literature, where the matching only covers the variables somebody thought to record.

The raw count is then corrected. Because prescribing volumes shift over time, a drug that launched recently will look as though it always arrives second. Researchers therefore divide the crude sequence ratio by a null-effect ratio built from background prescribing trends, producing the adjusted sequence ratio. An aSR of 1.0 means perfect symmetry and no signal^[3].

What this study reported

The authors, based at Monash University, used a 10% random sample of Australia's Pharmaceutical Benefits Scheme covering 1 July 2013 to 31 December 2024. They included 2,033 individuals who had an incident dispensing of both a GLP-1 receptor agonist and a neuropsychiatric medicine, applied a one-year exposure window, and ran sensitivity analyses across other time windows^[1]. Semaglutide accounted for 50.6% of the GLP-1 prescriptions, exenatide 27.5%, and dulaglutide 21.9%.

Two of the eight outcome classes moved. Antidepressants and substance use disorder medicines both showed inverse associations. Antipsychotics, psychostimulants, antidementia medicines, antiparkinsonian medicines, antiepileptics and antimigraine agents showed nothing statistically significant^[1].

It is worth being literal about what an aSR of 0.85 says. Among people who took both a GLP-1 receptor agonist and an antidepressant, the antidepressant tended to come first more often than the reverse, after adjusting for prescribing trends. That is a statement about sequence, not about disease. It is not a statement that the drug prevented depression, and the authors do not make one.

What the design controls, and what it cannot

The strength is narrow and genuine: confounding that stays constant across a person's timeline is handled by construction. The weakness is the mirror image. Anything that changes during the observation window is not handled at all. Losing 15% of body weight on semaglutide, sleeping better, or simply being seen more often by a clinician monitoring a new medicine are all time-varying, and any of them could tilt the order in which two prescriptions appear.

Two further limits matter here. Dispensing records are a proxy for diagnosis, not a diagnosis. And the result moves with the parameters chosen: the methods guide most often cited for this design shows that changing the washout window, the exposure window or the blackout period changes the strength of the association, which is why its authors ask researchers to state every parameter choice and run sensitivity analyses^[3].

Scale is the other caveat. At 2,033 individuals this is a small study by pharmacoepidemiology standards, and the drug mix reflects Australian subsidised prescribing between 2013 and 2024 rather than a 2026 obesity clinic. Exenatide and dulaglutide together made up nearly half the sample, and tirzepatide does not appear in the reported breakdown at all^[1]. The authors close by calling for further work to clarify the nature and magnitude of what they found.

Where the regulators have landed

The reason any of this gets attention is the 2023 case reports of suicidal and self-injurious thoughts in people taking GLP-1 medicines. Both major regulators have since reviewed the question. In April 2024 the EMA's Pharmacovigilance Risk Assessment Committee concluded that the available evidence does not support a causal association between GLP-1 receptor agonists and suicidal and self-injurious thoughts and actions, covering dulaglutide, exenatide, liraglutide, lixisenatide and semaglutide^[4].

The FDA went further on 13 January 2026, asking application holders to remove the suicidal ideation and behaviour warning from the labelling of Saxenda (liraglutide), Wegovy (semaglutide) and Zepbound (tirzepatide). Its review pooled 91 placebo-controlled trials covering 107,910 patients and added a Sentinel System cohort of 2,243,138 people, comparing GLP-1 starters against SGLT2 inhibitor starters, and found no increased risk of intentional self-harm^[5]. Country-level status for semaglutide and tirzepatide sits on the peptide pages.

How to read the next headline like this one

Sequence symmetry analysis is a signal-detection tool. It is fast, it is cheap, it runs on data that already exists, and it is good at flagging a question worth a proper study. It was never designed to answer one. When the next database analysis of GLP-1 medicines and mental health appears, the useful questions are the same three every time: which confounders could have changed during the window, how large was the sample, and does the effect survive when the authors move the time windows around.

This article is for educational and journalistic purposes and is not medical advice. Semaglutide and tirzepatide are prescription-only medicines everywhere we cover. If you are taking one and notice a change in your mood, that is a conversation for your prescribing clinician, not something to settle from a sequence ratio.