A study of 63,215 patients with pre-existing neuropsychiatric conditions, published on 1 September, found semaglutide associated with broadly lower rates of neuropsychiatric events over two years than metformin, SGLT2 inhibitors or DPP-4 inhibitors. Within the semaglutide group, patients who reached higher doses had fewer such events afterwards — the inverse of the dose-toxicity pattern implied by the suicidality scare of 2023. It is observational data, it reports statistical significance without effect sizes, and it cannot tell any individual what their dose should be.
How the comparison was built
The analysis used de-identified electronic health records and propensity matching, tracking 24 incident neuropsychiatric outcomes after treatment began. The dose question used a landmark design: the authors measured the highest dose each patient attained during the first two years after starting semaglutide, then counted new diagnoses during the following two years. Separating the exposure window from the outcome window this way prevents an event being credited to a dose the patient only reached after the event.
Higher attained dose was associated with significantly lower incidence in the later period across seven categories: substance-related disorders (P < 0.001), mood disorders (P < 0.001), anxiety- and stress-related disorders (P < 0.001), central nervous system atrophies (P < 0.001), neuromuscular disorders (P = 0.013), eating, sleep and behavioural disorders (P = 0.022), and personality and impulse-control disorders (P = 0.028). Dementia and CNS degenerative disease showed no difference between the higher- and lower-dose groups (P = 0.15), which matches what randomised trials have reported.
Two outcomes broke the pattern. Incident cognitive symptoms and speech or language symptoms tracked more closely with weight loss than with dose. The authors also ran bulk and single-cell transcriptomic analyses and found low-level, regionally restricted GLP1R transcript signals in central and peripheral nervous tissue. They label that hypothesis-generating, and that is what it is.
What the abstract does not report is a single effect size or confidence interval — only P values. A P value indicates that a difference is unlikely to be chance. It says nothing about whether the difference is big enough to matter to a patient, and that distinction does real work here.
What it settles, and what it does not
This lands on a specific and still-live worry. In 2023, European and US regulators opened reviews after reports of suicidal ideation among people taking GLP-1 drugs; both concluded the available evidence did not support a causal link, and several large observational studies since have pointed the same way. This one goes further by describing an inverse dose relationship — more drug, fewer events — which is difficult to reconcile with the drug causing them.
The obvious competing explanation is that attained dose is a proxy for something else. Patients who titrate up and stay at higher doses are, by definition, patients who tolerated the drug, kept their appointments and remained in care. Healthy-adherer bias produces findings shaped exactly like this one, and propensity matching on baseline characteristics does not remove it. The landmark design handles reverse causation; it does not handle that.
The authors are employees of nference, a company that runs research collaborations with pharmaceutical manufacturers whose products appear in the study. They state that no collaborator funded, designed, reviewed or approved the work, and that all analyses were run by the authors on de-identified records. That is the correct disclosure to make, and it is still worth knowing when reading a result this favourable to the drug.
The practical reading is narrow: this is reassurance about a class-level safety question, not an instruction to anyone. Dose is set by the clinician who wrote the prescription, weighing tolerability, response and cost alongside evidence like this. Our summary of what the GLP-1 evidence base currently supports does not move on a single observational study, however many patients it covers.
Sources: Murugadoss K, Venkatakrishnan AJ, Soundararajan V. Higher semaglutide dose is associated with lower neuropsychiatric event incidence independent of weight loss. npj Metabolic Health and Disease, 1 September 2026, 4(1):36, doi:10.1038/s44324-026-00131-3 · PubMed record, PMID 42680805
