Top 10 PubMed Studies on GLP-1 Agonists and Mental Health (2026)

Wang and colleagues used the TriNetX multi-institution EHR network to compare 240,618 patients prescribed semaglutide for obesity or type 2 diabetes against propensity-matched cohorts on non-GLP-1 antidiabetics or anti-obesity drugs. Over 6 months, incident suicidal ideation was 49% LOWER on semaglutide in the obesity cohort (HR 0.27 vs other anti-obesity drugs after multiple matching) and 56% lower in the T2D cohort. Recurrent suicidal ideation was also reduced. This was the first large real-world dataset to push back against the EMA signal — observational, not randomized, but using the strongest active-comparator design available outside of an RCT.

PMID 38182782 ↗DOI 10.1038/s41591-023-02672-2 ↗

Ueda and colleagues exploited Sweden's and Denmark's nationwide health registers to follow 124,517 new users of GLP-1 receptor agonists vs 174,036 new users of SGLT2 inhibitors (an active-comparator chosen because both classes target the same diabetes population). Over a median 2.5 years, suicide death occurred in 0.13 vs 0.18 per 1,000 person-years — a non-significant 17% relative reduction (HR 0.83, 95% CI 0.59-1.16). Non-fatal self-harm hazard ratios were also null. The register-based design largely eliminates the confounding by mental-health history that limits FAERS/VigiBase signal-detection studies and is the strongest observational evidence against a causal suicide signal.

PMID 39226030 ↗DOI 10.1001/jamainternmed.2024.4369 ↗

Wadden and colleagues pooled the four STEP weight-management trials of semaglutide 2.4 mg (3,377 participants total) to evaluate prospectively-collected psychiatric outcomes. Suicidal ideation or behavior on the Columbia Suicide Severity Rating Scale was reported by 1.0% of semaglutide-arm participants vs 0.7% on placebo (OR 1.40, 95% CI 0.66-2.94, non-significant). PHQ-9 depression scores improved slightly more on semaglutide than placebo. This is the highest-quality randomized psychiatric-safety evidence available for the obesity dose of semaglutide and is the basis for the FDA's view that the trial data do not support a causal suicidality signal — though enrolled patients excluded major psychopathology.

PMID 39226070 ↗DOI 10.1001/jamainternmed.2024.4346 ↗

Ebrahimi and colleagues pooled 27 randomized clinical trials of GLP-1 receptor agonists (semaglutide, liraglutide, tirzepatide, dulaglutide, exenatide, lixisenatide) across diabetes and obesity indications, capturing approximately 78,432 randomized participants. Pooled odds ratios were non-significant for suicide (OR 1.20, 95% CI 0.49-2.93), self-harm (OR 1.36, 95% CI 0.71-2.62), and suicidal ideation (OR 0.97). Depression-related adverse-event reporting was numerically lower on GLP-1 arms. The meta-analysis is the largest randomized-only synthesis to date and is the most-cited 2025 reference point for the absence of a randomized causal signal — although individual trials were not statistically powered for rare psychiatric events.

PMID 40105856 ↗DOI 10.1001/jamapsychiatry.2025.0091 ↗

McIntyre and colleagues analyzed the FDA's FAERS spontaneous-reporting database following the European Medicines Agency's July 2023 signal-detection alert. Disproportionality analyses showed elevated reporting odds ratios for suicidal ideation with semaglutide (ROR 1.45, 95% CI 1.18-1.77) and liraglutide (ROR 1.96, 95% CI 1.66-2.31) vs other drugs in FAERS. Importantly, the FAERS signal was attenuated but persisted after restricting to reports without concomitant antidepressant use. FAERS data cannot establish causality (denominator unknown, reporting bias dominant), but the paper documents the formal pharmacovigilance signal that drove the subsequent cohort and meta-analytic work.

PMID 38087976 ↗DOI 10.1080/14740338.2023.2295397 ↗

McIntyre and colleagues replicated their FAERS analysis using the WHO's global VigiBase pharmacovigilance database, which aggregates spontaneous reports from over 130 countries. Disproportionality signals for suicidal ideation persisted with semaglutide (IC 0.79, 95% credible interval 0.63-0.91) and liraglutide. The replication argues against the original FAERS finding being a US-reporting artifact, although it shares the same fundamental limitations: spontaneous-report data with no denominator, no time-to-event analysis, and heavy confounding by indication and prior psychiatric history. The paper is the strongest pharmacovigilance evidence for the persistence — though not causality — of the signal across populations.

PMID 39433133 ↗DOI 10.1016/j.jad.2024.10.062 ↗

Hendershot and colleagues randomized 48 adults with alcohol use disorder (not actively seeking treatment) to semaglutide weekly (titrated to 1.0 mg) or placebo for 9 weeks. The semaglutide arm showed greater reductions in lab-paradigm alcohol craving and drinks per drinking day (b = -0.48 drinks, 95% CI -0.85 to -0.11) and a 30% greater reduction in heavy-drinking-day frequency. Cigarettes per day fell in the smoking subgroup as a secondary finding. This is the FIRST randomized clinical trial of a GLP-1 receptor agonist for AUD and provides the proof-of-concept that the observational Wang/Volkow cohort signals reflect causal pharmacology — although it is small (n=48), short (9 weeks), and pre-dates phase 3 confirmation.

PMID 39937469 ↗DOI 10.1001/jamapsychiatry.2024.4789 ↗

Wang and colleagues used TriNetX to emulate a target trial comparing 222,942 patients prescribed semaglutide vs propensity-matched patients on seven other antidiabetic comparators (insulin, metformin, DPP-4 inhibitors, SGLT2 inhibitors, sulfonylureas, thiazolidinediones, and other GLP-1s) in T2D patients with tobacco use disorder. Semaglutide was associated with substantially lower risk of new TUD-related medical encounters (HR 0.68 vs insulin), more smoking-cessation prescriptions, and more cessation counseling visits. The signal was consistent across all seven comparators. Strongest real-world evidence for a nicotine-craving effect of semaglutide, although confounding by motivation-to-quit and indication bias cannot be excluded.

PMID 39074369 ↗DOI 10.7326/M23-2718 ↗

Wang and colleagues extended their substance-use cohort series to cannabis use disorder, identifying 85,223 patients with obesity and 1,000,066 patients with T2D who initiated semaglutide vs non-GLP-1 antidiabetics. Incident CUD was 44% lower on semaglutide in the obesity cohort (HR 0.56) and 60% lower in the T2D cohort. Among patients with pre-existing CUD, recurrent diagnoses fell 36-66% depending on comparator. The signal pattern matches the tobacco and alcohol cohort findings and supports a general dopaminergic-reward mechanism. As with all TriNetX cohort studies, residual confounding by health-seeking behavior and unmeasured cannabis exposure is a key limitation.

PMID 38486046 ↗DOI 10.1038/s41380-024-02498-5 ↗

Wang and colleagues identified 33,006 patients with T2D and pre-existing opioid use disorder in the TriNetX network and propensity-matched semaglutide initiators to users of seven other antidiabetic comparators. Twelve-month incident opioid-overdose risk was 42-68% lower on semaglutide depending on comparator (HR 0.32-0.58, all statistically significant). The effect was robust across sex, race, and OUD severity strata. The finding extends the substance-use hypothesis to the most severe addiction outcome (overdose), although it cannot exclude residual confounding by adherence to medical care. No randomized GLP-1 OUD trial has reported results as of May 2026.

PMID 39320894 ↗DOI 10.1001/jamanetworkopen.2024.35247 ↗