What the new semaglutide cognitive side effects study actually looked at
The largest of the three papers is a retrospective cohort study published in BMJ Mental Health by De Giorgi, Lipunova, Mathias and co-authors. Its stated population is 13,007 people in the United States who already carried a psychiatric diagnosis and who were prescribed semaglutide, the GLP-1 receptor agonist marketed as Ozempic for type 2 diabetes and Wegovy for weight management. The outcome under examination is described in the title as cognitive signs and symptoms.
A retrospective cohort study works backwards through records that already exist — typically electronic health records or claims data — identifying people who took a drug and people who did not, then counting how often a defined outcome was recorded in each group. It is fast and can reach very large sample sizes, but the researchers did not assign the treatment, so any difference between groups can reflect why people were prescribed the drug rather than what the drug did.
That distinction matters more than usual here. People with an existing psychiatric diagnosis are the exact population in which cognitive complaints — difficulty concentrating, memory lapses, mental fatigue — are most likely to be documented for reasons unrelated to a metabolic medication. Choosing that population is a deliberate design decision, and it makes the study more relevant to a real clinical question while making the confounding harder to untangle.
Why cognitive symptoms are hard to attribute to a GLP-1 drug
Cognitive complaints do not arrive in isolation during incretin therapy. Rapid weight loss, reduced food intake, altered sleep, dehydration from gastrointestinal side effects and changes in glycaemic control can each plausibly show up in a chart as brain fog or poor concentration. Depression and anxiety, which by definition are common in a cohort selected for psychiatric diagnosis, carry cognitive symptoms as part of their own diagnostic criteria.
There is also the problem of what gets written down. Symptoms recorded in a health record are symptoms someone reported and a clinician coded. Heightened public attention to the mental-health effects of GLP-1 drugs can raise both reporting and coding rates without any change in underlying biology — a phenomenon epidemiologists call ascertainment bias. Studies of this type usually attempt to control for it, and how well they succeed is what peer reviewers argue over.
Cognitive symptoms are among the easiest to under-report, which is why side-effect logging with dates attached is more useful to a prescriber than recalling a rough impression at an appointment.
Naunyn-Schmiedeberg's Archives of Pharmacology published a pharmacology analysis of psychiatric and eye disorders associated with use of GLP-1 receptor agonists.
What a Bradford Hill review of GLP-1 psychiatric side effects adds
The second paper, a systematic review by Schifano, De Luca, Bonaccorso and colleagues in Current Psychiatry Reports, addresses the psychopharmacology of GLP-1 and dual GIP/GLP-1 receptor agonists and their putative neuropsychiatric effects. The dual agonist class includes tirzepatide, which acts at both the GLP-1 receptor and the receptor for glucose-dependent insulinotropic polypeptide, or GIP.
What makes the review methodologically interesting is its explicit use of the Bradford Hill framework. Proposed by the epidemiologist Austin Bradford Hill in 1965, it is a set of considerations — strength of association, consistency across studies, a plausible biological mechanism, a dose-response relationship, correct temporal ordering, and others — used to judge whether an observed statistical association is likely to reflect causation. It is a structured argument rather than a statistical test, and reasonable specialists can weigh the same evidence differently.
Applying that framework to neuropsychiatric signals is a step up from simply cataloguing case reports. GLP-1 receptors are present in regions of the brain involved in appetite, reward and mood regulation, so a mechanism is at least conceivable; whether the reported human signals meet the rest of the criteria is precisely the question the review sets out to answer.
Readers following a provider-directed semaglutide protocol often pair a symptom log with the semaglutide dosage calculator so that timing of any change is recorded against the schedule they were given.
Psychiatric and eye disorders in the same analysis
The third paper, by Ammendolia, Mondello, Esposito and co-authors in Naunyn-Schmiedeberg's Archives of Pharmacology, is titled as an examination of psychiatric and eye disorders associated with use of GLP-1 receptor agonists. Grouping those two categories together is characteristic of pharmacovigilance work, where adverse events are sorted into standardised organ-system classes — psychiatric disorders and eye disorders are each such a class in the MedDRA dictionary used by regulators worldwide.
Pharmacovigilance databases collect spontaneous reports from clinicians, patients and manufacturers. They are useful for spotting unexpected patterns early, and they are the weakest form of evidence for quantifying risk, because there is no denominator: nobody knows how many people took the drug without reporting anything. A disproportionality signal is a prompt to investigate, not a measured rate.
Ophthalmic outcomes in this class have drawn scrutiny before, which is part of why a paper pairing eye and psychiatric findings is worth noting even before its numbers are public.
Because symptoms are frequently reported around schedule changes, it helps to understand what titration means before trying to interpret when something started.
What these three papers can and cannot establish
None of the three is a randomised controlled trial. Randomisation is what breaks the link between the reasons a person receives a drug and the outcomes they later experience, and without it, an association between semaglutide and a recorded cognitive symptom remains an association. The authors of all three papers work within those limits, and the peer-reviewed literature treats observational findings as hypothesis-generating unless replicated across designs.
There is a further caveat specific to this moment. At the time of writing, what is publicly indexed for all three papers is the bibliographic record — journal, authors, design and, in the BMJ Mental Health case, the cohort size. The direction, magnitude and statistical precision of the findings are not yet in the public abstracts. Anyone summarising these studies as showing that semaglutide does or does not affect cognition is going beyond what is currently available.
Taken together, the cluster is still meaningful. Three groups in three different journals — a mental-health journal, a psychiatry review journal and a pharmacology journal — are converging on the same question in the same window, which is usually how a safety topic moves from anecdote to formal investigation.
The BMJ Mental Health cohort study is indexed as covering 13,007 patients in the United States who carried a psychiatric diagnosis and were prescribed semaglutide. the BMJ Mental Health cohort study.
What happens next
The realistic next steps are replication in independent datasets and, eventually, prespecified analysis of neuropsychiatric endpoints within randomised trials. Large cardiovascular and weight-management trials of semaglutide and tirzepatide already collect adverse event data systematically; pooled analyses of those datasets carry considerably more weight than any single retrospective cohort.
Regulators generally act on this kind of literature only when signals are consistent across sources. Until then, the practical output for readers is a sharper vocabulary: knowing that a cohort study, a Bradford Hill review and a pharmacovigilance analysis answer different questions makes the next round of headlines easier to read accurately.
The Bradford Hill-informed systematic review in Current Psychiatry Reports covers the psychopharmacology of both GLP-1 and dual GIP/GLP-1 receptor agonists. the Bradford Hill-informed systematic review.