Comparative pharmacovigilance analysis of suicidality-related adverse events among GLP-1 and non-GLP-1 anti-obesity drugs in the FDA Adverse Event Reporting System

To compare disproportional reporting of suicidality-related adverse events among GLP-1/dual incretin versus non-GLP-1 anti-obesity drugs in the FDA Adverse Event Reporting System.

We performed a retrospective pharmacovigilance study using data from the U.S. Food and Drug Administration Adverse Event Reporting System (FAERS), a spontaneous reporting database collecting adverse drug reaction (ADR) notifications from healthcare professionals, consumers, and manufacturers worldwide. Reports from January 1, 2012, through February 28, 2025, were extracted via the openFDA application programming interface (API; https://open.fda.gov↗).

The analysis followed recommendations from the European Medicines Agency (EMA) and the International Society of Pharmacovigilance for disproportionality analyses [18–20].

All FDA-approved anti-obesity medications indicated for chronic weight management were included:

Setmelanotide (Imcivree®) was excluded due to its orphan indication for genetic obesity syndromes, which involves a distinct patient population and safety profile. Reports were restricted to those submitted from the United States and identifying the study drug as the primary suspect (patient. drug. Drug characterization = 1) to reduce confounding by concomitant therapies [21]. Duplicate cases were automatically removed by FAERS based on unique case identifiers.

These restrictions were applied to enhance internal validity. Limiting analyses to U.S. reports reduces heterogeneity in MedDRA coding practices, reporting completeness, and submission workflows across regions. Similarly, restricting to primary suspect drugs minimizes attribution noise when multiple drugs are listed in the same report, thereby improving the specificity of the signal detection process.

Suicidality-related adverse events were defined according to MedDRA version 27.0, encompassing the following Preferred Terms (PTs): suicidal ideation, suicide attempt, and completed suicide. Each PT was analyzed separately to avoid aggregation bias, following the approach recommended in previous FAERS-based suicidality studies [11, 22].

Suicidality-related adverse events were retrieved at the level of MedDRA Preferred Terms (PTs). All FAERS reports mapped to these PTs, irrespective of the underlying Lowest Level Terms (LLTs), were included in the analysis. A list of the suicidality-related PTs and representative LLTs used in our queries is provided in Supplementary Table. S1

Given the markedly unequal post-marketing exposure between agents—particularly for tirzepatide, which entered the obesity market in late 2023—we anticipated substantial variability in cumulative report volume. This imbalance limits the interpretability of disproportionality estimates for recently approved drugs, where low case counts produce wide confidence intervals and immature pharmacovigilance signals.

A within-class and between-class comparative disproportionality design was used. For each anti-obesity drug, the frequency of suicidality-related reports was compared with the combined frequency of the same events reported for the remaining agents within the same therapeutic class. In addition, aggregated comparisons were performed between incretin-based (GLP-1 receptor agonists and dual GLP-1/GIP agonists) and non-GLP-1 anti-obesity drugs to explore potential class-level differences. Counts were obtained directly from open FDA queries using specific MedDRA filters and manually verified for plausibility. Analyses were restricted to reports originating from the United States to ensure data homogeneity and minimize heterogeneity related to regional reporting practices.

Reporting Odds Ratios (RORs) and 95% confidence intervals (95% CIs) were calculated to quantify disproportionality between the target drug and its comparators. When zero counts were present in any cell, the Haldane–Anscombe correction was applied to stabilize the variance [23].

A disproportionality signal was considered statistically significant when the lower bound of the 95% CI exceeded 1.0, consistent with prior pharmacovigilance standards [18, 24].

All statistical calculations were performed in Microsoft Excel (Microsoft Corp., Redmond, WA), and the results were cross validated with outputs from OpenVigil 2.1 [25].

The ROR reflects disproportionality in reporting and should not be interpreted as a measure of incidence, absolute risk, or causality.

FAERS data are publicly available and de-identified. Therefore, this study did not require institutional review board approval or patient consent [26].

Between 2012 and 2025, a total of approximately 78,000 FAERS reports involving FDA-approved anti-obesity drugs were retrieved, of which less than 0.5% included suicidality-related Preferred Terms (PTs), such as suicidal ideation, suicide attempt, or completed suicide. Most reports originated from the United States and were submitted after 2021, coinciding with the approval and widespread use of semaglutide (Wegovy®) for weight management.

The distribution by agent was as follows: tirzepatide (Zepbound®; approximately 23,000 reports), semaglutide (Wegovy®; approximately 20,000), liraglutide (Saxenda®; approximately 5,000), naltrexone/bupropion (Contrave®; approximately 6,000), and phentermine/topiramate (Qsymia®; approximately 5,000).

As shown in Table 1, the updated disproportionality analysis revealed distinct patterns across anti-obesity agents.

For semaglutide, the Reporting Odds Ratios (RORs) were 1.39 (95% CI, 0.99–1.94) for suicidal ideation, 1.38 (0.46–4.15) for suicide attempt, and 1.72 (0.62–4.74) for completed suicide, none reaching statistical significance. Liraglutide displayed a comparable ROR for suicidal ideation (1.01; 0.66–1.55) and suicide attempt (0.91; 0.21–3.94), but a markedly higher signal for completed suicide (18.11; 6.96–47.15), based on a small number of cases. In contrast, tirzepatide showed consistently low disproportionality estimates: 0.19 (0.13–0.28) for suicidal ideation, 0.12 (0.03–0.53) for suicide attempt, and 0.03 (0.00–0.43) for completed suicide—all well below unity, suggesting no safety signal. These low values should be interpreted with caution, as tirzepatide’s short post-marketing history and comparatively small number of FAERS reports result in very wide confidence intervals and likely under-ascertainment of rare events. Among non-incretin combinations, naltrexone/bupropion exhibited elevated RORs for suicidal ideation (3.84; 2.89–5.12) and suicide attempt (4.11; 1.62–10.45), whereas phentermine/topiramate showed a moderate increase for suicidal ideation (1.57; 1.09–2.27) but not statistically significant results for suicide attempt or completed suicide.

At the class level, GLP-1 receptor agonists and dual incretin therapies showed disproportionality estimates below unity for suicidal ideation (ROR 0.29; 95% CI 0.22–0.38) and suicide attempt (ROR 0.22; 95% CI 0.09–0.56), but a not statistically significant elevation for completed suicide (ROR 5.85; 95% CI 0.78–43.67). In contrast, non-GLP-1 anti-obesity drugs exhibited higher reporting for suicidal ideation (ROR 3.50; 95% CI 2.67–4.61) and suicide attempt (ROR 4.48; 95% CI 1.80–11.13), while completed suicides remained infrequent (ROR 0.17; 95% CI 0.02–1.28) (Table 2).

Figure 1 displays RORs for suicidal ideation, Fig. 2 for suicide attempt, and Fig. 3 for completed suicide. Table 2 shows class-level comparisons between GLP-1/dual incretin and non-GLP-1 agents, while Fig. 4 illustrates the overall number of FAERS reports included in the analysis.

Our findings align with recent pharmacovigilance investigations conducted using both FAERS and WHO VigiBase® database. McIntyre et al. (2025) analyzed over 25,000 reports from VigiBase and confirmed that no causal link could be established between GLP-1receptor receptor agonists and suicidality [13]. Similarly, Wang et al. (2025) and Guirguis et al. (2024) reported that while semaglutide displayed isolated numerical elevations in the reporting of depression or self-injury terms, these signals disappeared after adjustment for concomitant medications and indication [11, 12].

A U.S. Food and Drug Administration review (2024) and the European Medicines Agency (EMA) Pharmacovigilance Risk Assessment Committee also concluded that a causal relationship was unlikely, despite the increased media attention in 2023 [38, 39].

In real-world evidence studies, including large retrospective cohorts and electronic health record analyses, most reports found no increase in suicidal ideation or behavior among patients treated with GLP-1 receptor agonists compared with other antidiabetic agents. However, one exploratory pharmacovigilance study detected preliminary suicide-related signals for semaglutide in weight-loss populations, warranting further confirmation in large prospective datasets [27–29].

Data from randomized controlled trials also consistently demonstrates the absence of psychiatric safety signals, both in diabetes and obesity populations [3, 30, 31].

For instance, pooled analyses from the PIONEER and SUSTAIN programs found that rates of depression, anxiety, or suicidality were comparable to placebo, reinforcing the biological plausibility of a null association [32].

The 2023–2024 regulatory reviews and extensive media coverage of GLP-1 receptor agonists and suicidality are also likely to have contributed to notoriety bias. In our dataset, suicidality-related reports were concentrated in the most recent years, coinciding with the rapid uptake of incretin-based anti-obesity therapies and the period of heightened public and regulatory attention. Such stimulated reporting may inflate disproportionality estimates for older or more widely used drugs, while underestimating potential signals for newer agents with shorter post-marketing histories.

Interpretation of the elevated ROR for completed suicide with liraglutide requires particular caution. The 2023–2024 regulatory reviews and extensive media attention regarding GLP-1 receptor agonists and suicidality likely contributed to stimulated reporting and notoriety bias. Liraglutide, as one of the earliest anti-obesity incretin agents, has substantially greater cumulative post-marketing exposure than newer therapies, increasing the probability that rare events are reported disproportionately. Importantly, the absolute number of completed-suicide reports was extremely small (n = 14), generating wide and statistically unstable confidence intervals. Misclassification or heterogeneity in FAERS coding may further inflate this estimate. Taken together, these considerations indicate that the observed signal is statistically fragile and should be regarded as hypothesis-generating, requiring cautious interpretation rather than being taken as definitive evidence of a pharmacological effect.

Given the extremely low number of completed suicide reports and the wide confidence intervals, this finding should not be interpreted as evidence of increased risk. Instead, the signal reflects statistical fragility and stimulated reporting rather than a reproducible pharmacological phenomenon.

Interpretation of tirzepatide disproportionality estimates is inherently limited by pronounced exposure and reporting imbalance. As the newest anti-obesity therapy, tirzepatide has accumulated far fewer FAERS reports than semaglutide or liraglutide, producing immature pharmacovigilance data with wide confidence intervals. Accordingly, the very low RORs observed for tirzepatide should not be interpreted as evidence of a protective effect, but rather as a reflection of insufficient follow-up and lower reporting volume.

It is also important to acknowledge that the potential explanations for the liraglutide signal—greater cumulative exposure, longer time on the market, and heightened regulatory/media attention—are bidirectional and cannot be used to adjudicate causality. Increased exposure may inflate reporting through stimulated reporting or selection bias, but it may equally increase the likelihood of detecting true drug-event associations. FAERS does not provide the necessary information to disentangle these possibilities. Therefore, the elevated ROR for completed suicide cannot be dismissed as non-causal on the basis of exposure history alone; rather, it represents a pharmacovigilance signal that requires cautious interpretation and external validation in controlled settings.

Interpretation of class comparisons requires careful consideration of the distinct neuropharmacological profiles of non-GLP-1 anti-obesity drugs. Naltrexone/bupropion combines an opioid antagonist with a dopaminergic–noradrenergic reuptake inhibitor, a mechanism known to modulate mood, reward processing, and impulsivity [16]. Bupropion-containing regimens have been associated with neuropsychiatric adverse events in both clinical trials and post-marketing settings, which may partly account for the markedly higher RORs observed in our analysis for suicidal ideation and suicide attempt.

Phentermine/topiramate also carries theoretical psychiatric implications due to topiramate’s associations with cognitive slowing, mood changes, and irritability [17]. These pharmacodynamic effects differ fundamentally from the incretin-based mechanism of GLP-1 and dual GIP/GLP-1 agonists, which do not engage monoaminergic pathways linked to suicidality.

As a result, the higher disproportionality detected among non-GLP-1 agents may reflect drug-specific neuropharmacology and differences in baseline psychiatric risk rather than evidence of a class-wide safety contrast [35–37].

Because suicidality-related events were rare, a class-level comparison between GLP-1/dual incretin and non-GLP-1 therapies was included as a complementary analysis. However, this aggregation must be interpreted with caution. Combining semaglutide, liraglutide and tirzepatide into a single category may conceal substantial heterogeneity in post-marketing exposure, reporting volume, and time on the market. Class-level RORs therefore do not indicate a protective effect of the incretin class; rather, they underscore the importance of evaluating each active substance individually, as class aggregation can dilute or obscure drug-specific patterns. This principle is consistent with established pharmacovigilance methodology—where analyses are ideally conducted at the active-substance level—and reinforces that class-level estimates are hypothesis-generating and should not guide clinical.

Although several mechanistic hypotheses have proposed links between GLP-1 signalling and central dopaminergic or stress-related pathways, preclinical findings remain inconsistent [33, 34]. Some models describe modulation of stress reactivity or mood, whereas others suggest anxiolytic rather than pro-depressive effects. Clinical studies assessing neuropsychiatric outcomes—such as cognition, addiction-related behaviour, or appetite regulation—have likewise shown neutral or beneficial central effects of GLP-1 receptor agonists [35, 36].

The particularly low RORs observed for tirzepatide, including a near-zero estimate for completed suicide, likely reflect its recent market authorization and limited post-marketing exposure time. Zepbound® was only approved for obesity in late 2023, resulting in fewer cumulative reports relative to older GLP-1 agents such as liraglutide and semaglutide. Hence, these data may underestimate rare events due to incomplete pharmacovigilance accrual rather than true absence of risk. The emerging evidence on tirzepatide, a dual GIP/GLP-1 receptor agonist, also supports a neutral psychiatric profile, consistent with our findings [37].

To contextualise our findings, it is important to consider the relative frequency of suicidality-related reports historically observed across anti-obesity drug classes. Non-GLP-1 agents, particularly naltrexone/bupropion and phentermine/topiramate, have consistently generated a higher proportion of neuropsychiatric adverse event reports in both FAERS and other international pharmacovigilance repositories [12, 16, 17]. In contrast, incretin-based therapies have historically shown low reporting rates for suicidality-related Preferred Terms, with multiple FAERS, VigiBase and real-world cohort analyses reporting either no disproportionality signal or numerically lower rates compared with active comparators [13, 14, 27–29]. These historical patterns are consistent with the class differences observed in our dataset, suggesting that the relatively higher RORs among non-GLP-1 drugs reflect established pharmacovigilance trends rather than new or unexpected safety concerns. However, interpretation should remain cautious, as reporting frequencies may also be shaped by drug utilisation patterns, media attention, and evolving regulatory scrutiny over time.

The temporary EMA investigation in 2023, triggered by spontaneous reports of suicidal thoughts among GLP-1 users, likely contributed to reporting amplification rather than true incidence changes.

Both the EMA and FDA concluded that available data did not support label modifications or additional warnings [38, 39].

These findings reinforce the overall psychiatric safety of modern anti-obesity pharmacotherapy, particularly incretin-based therapies. While sporadic reports of self-harm or mood alterations continue to emerge, their absolute frequency remains extremely low compared with global exposure.

Confounding by indication remains a major limitation, as obesity and related comorbidities inherently increase baseline risks of depression and suicidality. FAERS lacks reliable information on psychiatric history, depression, socioeconomic status, substance use, and other determinants of suicidality; therefore, no adjustment for these factors was possible.

Confounding by indication is likely an important contributor to any disproportionality signal, particularly for liraglutide. Although we explored the feasibility of stratification using concomitant antidepressant or antipsychotic medications as proxy indicators of psychiatric burden, FAERS coding of concomitant therapies is incomplete, inconsistently structured, and affected by substantial missingness. As a result, no valid stratified disproportionality analysis could be performed.

Although we considered stratifying RORs by calendar period, the extremely low number of suicidality-related events—particularly completed suicides—produced highly unstable and non-interpretable time-specific ROR estimates. For this reason, we did not conduct formal temporal disproportionality modelling. References to notoriety bias in the Discussion therefore reflect conceptual mechanisms well described in spontaneous reporting systems during periods of regulatory or media attention, rather than empirical temporal testing.

Several structural limitations inherent to spontaneous reporting systems must also be acknowledged, including under-reporting, the absence of exposure denominators, and regional variability in reporting patterns. Restricting analyses to U.S. submissions and to primary suspect drugs may affect comparability with studies using broader definitions; however, these methodological choices also reduce heterogeneity and misattribution, improving internal consistency.

Misclassification is possible, particularly for completed suicide, a rare and heterogeneously coded event for which narrative details are often unavailable. Variable report quality, incomplete narratives, and inconsistent coding can further contribute to uncertainty.

Finally, additional forms of reporting bias—including competition bias, channeling bias, and stimulated reporting following regulatory or media attention—may influence the distribution of Preferred Terms across drug classes. For these reasons, FAERS-based disproportionality estimates should be interpreted cautiously and regarded as hypothesis-generating rather than evidence of causal risk.

interpretation.

The authors declare no competing interests.

Not applicable. This study used publicly available, anonymized pharmacovigilance data from the FDA FAERS database.

Not applicable.

Not applicable.

All data analyzed in this study are publicly accessible through the openFDA platform (https://open.fda.gov/data/faers/↗).