Depressed mood and suicidal thoughts reporting with GLP-1 receptor agonists in type 2 diabetes: A WHO VigiBase study
Aboukaoud M, Hoch B, Weiser M, Amiaz R
Journal of Affective Disorders (2026) · n=1,183,817
Across 1,183,817 GLP-1RA adverse-event reports in the WHO VigiBase pharmacovigilance database, semaglutide, liraglutide and tirzepatide all showed elevated reporting of depressed mood (adjusted ROR0.25: 2.13, 1.52, 1.07) and suicidal thoughts (6.76, 2.43, 3.39) — but with no signal for suicide attempts or completed suicide, and with concomitant antidepressant use and comorbid depression 25–120% higher than in comparator drug users, suggesting affective vulnerability drives the signal rather than a uniform drug-specific effect.
This 2026 Journal of Affective Disorders paper from the Sheba Medical Center psychiatry group is the most comprehensive pharmacovigilance analysis to date of the GLP-1 receptor agonist depression and suicidality signal that has driven regulatory attention, media coverage and patient-counselling complexity since the European Medicines Agency opened its 2023 safety review. The investigators conducted a disproportionality analysis of the WHO VigiBase pharmacovigilance database (2010-2024), restricted to type 2 diabetes patients, comparing GLP-1RA reports to reports from other glucose-lowering medications.
Across 1,183,817 GLP-1RA adverse-event reports, the authors identified elevated reporting of depressed mood for semaglutide (adjusted ROR0.25 2.13), liraglutide (1.52), and tirzepatide (1.07), and elevated reporting of suicidal thoughts for the same three drugs (ROR0.25 6.76, 2.43, 3.39 respectively). Crucially — and this is the load-bearing finding for clinical framing — no signal was identified for suicide attempts or completed suicide. Median time-to-onset was 96 days. Survival analysis showed an early increase in reporting followed by stabilisation, consistent with either a true early-phase effect or with detection-bias as patients and physicians attribute mood symptoms in the weeks following drug initiation.
The interpretive shift the authors make is to focus on what differentiates GLP-1RA users from comparator users in terms of pre-existing psychiatric vulnerability: concomitant antidepressant use was 2.3 to 5 times more frequent, and comorbid depression diagnosis was 25-120% higher, in the GLP-1RA cohort than in comparator drug users. Adjustment for antidepressant use modestly attenuated the associations. The authors conclude that the signal is more consistent with affective-vulnerability-driven reporting than with a uniform drug-specific causal effect, and they recommend psychiatric monitoring during early treatment in patients with pre-existing affective vulnerability.
This is a pharmacovigilance disproportionality analysis, not a controlled cohort study or RCT — VigiBase reports are spontaneous notifications subject to severe selection bias, reporter-attribution bias, and detection bias from media coverage of the GLP-1RA mood signal during the study window. The reporting odds ratio (ROR) is a relative-frequency-of-reports metric, not a population-incidence estimate; an elevated ROR does not establish causation and does not establish the absolute increment in risk per treated patient.
The differential ROR pattern across drugs (semaglutide > liraglutide > tirzepatide for depressed mood, but semaglutide > tirzepatide > liraglutide for suicidal thoughts) is not consistent with a simple class effect, and is likely confounded by differential prescription volume, marketing pattern, media-coverage attention, and the differing approval timelines of the three drugs. The Weibull time-to-event model and Bradford-Hill causality framework are appropriate tools but cannot overcome the fundamental selection-bias problem with VigiBase data.
The negative signal for suicide attempts and completed suicide is the most reassuring finding in the paper but should be read with caution — completed suicide is rare in absolute terms, time-to-event detection in spontaneous reporting is poor for severe outcomes, and the negative signal is consistent with both genuine absence of a hard-outcome risk and with under-reporting of attempted/completed suicide compared to reported ideation.
The conclusion that affective vulnerability drives the signal rather than a drug-specific effect is consistent with the underlying epidemiology — patients prescribed GLP-1RAs have higher baseline rates of depression, anxiety, and antidepressant use than the general T2D population — but the residual confounding cannot be fully resolved by adjustment in VigiBase data. A properly designed prospective cohort or active-comparator new-user pharmacoepidemiology study would settle the question more definitively; this paper is the strongest VigiBase analysis but does not replace such a study.
The authors declare no conflicts of interest. The analysis is conducted to disproportionality-analysis methodological standard but inherits the underlying limitations of the data source.
07·Member discussion
No member discussion yet.
Member-only conversation lives here — cycle notes, practitioner commentary, pattern-matching. Be the first paying member to start the thread.