Semaglutide and Panic Attacks: Active-Comparator Disproportionality Analysis, Temporal Assessment of Notoriety Bias and Evidence Triangulation in Post-Marketing Pharmacovigilance.

Semaglutide and Panic Attacks: Active‐Comparator Disproportionality Analysis, Temporal Assessment of Notoriety Bias and Evidence Triangulation in Post‐Marketing Pharmacovigilance

ABSTRACT

Purpose

To evaluate how confounding by indication and notoriety bias shape the apparent semaglutide–panic attack signal in the FDA Adverse Event Reporting System (FAERS) and to interpret this signal using an evidence‐triangulation framework integrating randomized and observational studies.

Methods

We conducted an active‐comparator restricted disproportionality analysis of FAERS (2018Q1–2024Q4), including individual case safety reports in which semaglutide, dulaglutide or empagliflozin was the primary suspect drug. Panic attack was identified using the MedDRA preferred term. Crude and adjusted reporting odds ratios (RORs, aRORs) with 95% confidence intervals (CIs) were estimated using logistic regression, adjusting for age, sex and indication (Type 2 diabetes, obesity/weight management). Notoriety bias was assessed using prespecified temporal analyses (analyses restricted to reports received before July 1, 2023 and a drug‐by‐period comparison of 2018–2022 with 2023–2024) and reporting proportions per 10 000 reports. A sensitivity analysis used all other primary‐suspect drugs in FAERS as the reference set. In parallel, we conducted a systematic review (MEDLINE, Embase, CENTRAL to November 24, 2025) of randomized trials, cohort and cross‐sectional studies, pharmacovigilance analyses and case reports reporting panic attacks or acute anxiety outcomes with semaglutide.

Results

Among 33 267 semaglutide, 54 282 dulaglutide and 25 380 empagliflozin reports, panic attacks were recorded in 87 (0.26%), 28 (0.05%) and 13 (0.05%) cases, respectively. Crude RORs were 5.08 (95% CI 3.32–7.78) versus dulaglutide and 5.12 (95% CI 2.86–9.17) versus empagliflozin, attenuating after adjustment to 2.82 (95% CI 1.74–4.56) and 2.75 (95% CI 1.41–5.36). In the pre‐warning period crude RORs were 2.24 (95% CI 1.23–4.07) and 2.15 (95% CI 1.00–4.58), attenuating to 1.38 (95% CI 0.69–2.75) and 1.14 (95% CI 0.48–2.68) after adjustment. Panic‐attack reports per 10 000 semaglutide reports rose from 6.7 in 2018–2022 to 42.0 in 2023–2024 (drug‐by‐period interaction p = 0.007). Against all other primary‐suspect drugs, the aROR was 1.48 (95% CI 1.18–1.87) overall and 0.76 (95% CI 0.48–1.18) before July 1, 2023. Across 12 studies, randomized trials and active‐comparator observational cohorts did not show a consistent increase in panic attacks or anxiety‐related outcomes.

Conclusions

Crude FAERS analyses suggested a semaglutide–panic attack signal that attenuated after adjustment was no longer significant before the July 2023 regulatory communication and rose sharply thereafter in a pattern consistent with notoriety bias. Triangulation with randomized trials and active‐comparator observational studies does not support a consistent pharmacological association. Disproportionality analyses are hypothesis‐generating and cannot estimate incidence or establish causality. This case study illustrates how active‐comparator design, temporal assessment of notoriety bias and evidence triangulation can improve the interpretation of post‐marketing safety signals.

Keywords: confounding by indication, FAERS, GLP‐1 receptor agonists, notoriety bias, panic attack, pharmacovigilance, semaglutide

Key Points

Crude FAERS analyses suggested increased reporting of panic attacks with semaglutide, but the association was substantially attenuated after adjustment for age, sex and indication.

In analyses restricted to the period before July 2023, reporting was elevated in crude analyses, but the adjusted estimates were no longer statistically significant after adjustment for age, sex and indication. This attenuation is compatible with confounding, although a pharmacological effect cannot be excluded.

Triangulation with randomized trials and active‐comparator observational studies did not support a consistent increase in panic attacks or related anxiety outcomes with semaglutide.

This study shows how active‐comparator restricted disproportionality analysis, temporal assessment of reporting dynamics and evidence triangulation can improve the interpretation of post‐marketing safety signals in pharmacoepidemiology.

Plain Language Summary

This study investigated whether the diabetes and weight‐loss drug semaglutide (commonly known as Ozempic or Wegovy) increases the risk of panic attacks. Researchers analyzed patient safety data from the FDA Adverse Event Reporting System (FAERS) spanning 2018 through 2024. While the raw numbers initially showed that panic attacks were reported more often with semaglutide than with other medications (like dulaglutide or empagliflozin), this connection weakened significantly once researchers adjusted for patient age, sex and the specific reason for taking the drug. Crucially, the surge in panic attack reports occurred only after mid‐2023, coinciding with heightened media attention and regulatory reviews regarding the mental health impacts of GLP‐1 drugs. This pattern is consistent with ‘notoriety bias’, meaning that greater public awareness may have led to more reports rather than the drug itself causing more panic attacks. However, this type of data can neither prove nor rule out a true effect.

Furthermore, a review of 12 studies, including pooled clinical trial data and real‐world observational studies, found no consistent evidence linking semaglutide to increased panic attacks or anxiety. Overall, the available evidence does not show a consistent link between semaglutide and panic attacks, although an association with panic attacks cannot be excluded.

1.Introduction

1.1.Background

Semaglutide is a long‐acting glucagon‐like peptide‐1 receptor agonist (GLP‐1 RA) approved for Type 2 diabetes and chronic weight management [1, 2]. Its rapid uptake reflects robust glycemic control, sustained weight reduction and cardiovascular benefit demonstrated in the SUSTAIN, STEP and SELECT trials, alongside increasing off‐label use and widespread public visibility [1, 2, 3, 4]. In pivotal randomized trials, psychiatric adverse events were rare and balanced between semaglutide and placebo arms; however, these trials were not designed or powered to detect rare, acute neuropsychiatric outcomes such as panic attacks [5, 6].

Post‐marketing pharmacovigilance analyses have reported signals of disproportionate reporting of acute psychiatric events, including panic attacks, with GLP‐1 RAs, particularly semaglutide [7, 8]. These observations emerged in parallel with heightened regulatory attention, including an Icelandic Medicines Agency alert and a subsequent European Medicines Agency Pharmacovigilance Risk Assessment Committee (PRAC) review initiated in July 2023, Report Number: EMA/PRAC/258271/2023, with a parallel evaluation by the US Food and Drug Administration [9, 10]. The FDA's preliminary evaluation in January 2024 did not identify evidence of a causal association with suicidality. Following a comprehensive review, the FDA requested removal of the suicidal behavior and ideation warning from affected weight‐management products in January 2026 [9]. The specific signal for acute panic attacks remains incompletely characterized in post‐marketing surveillance.

Spontaneous reporting data are difficult to interpret for this question. Crude disproportionality estimates are vulnerable to confounding by indication, because GLP‐1 RA users differ systematically from comparator populations in cardiometabolic risk, sex distribution and age structure [11, 12]. They are also vulnerable to notoriety bias, the rise in reporting that follows regulatory communications and media attention [13, 14, 15]. Most prior analyses have compared against heterogeneous all‐drug reference groups, which limits the interpretability of indication‐relevant comparisons [16].

For rare adverse drug effects, well‐conducted observational studies using active‐comparator, new‐user designs provide the most appropriate basis for risk quantification [12]. Randomized trials may be underpowered for rare acute neuropsychiatric outcomes, whereas spontaneous‐reporting systems lack exposure denominators and detailed clinical information. Disproportionality analysis therefore serves a narrower, hypothesis‐generating purpose and cannot quantify risk or establish causality [17]. Accordingly, our interpretation triangulates evidence across these sources while giving the greatest weight to studies whose designs and data are best suited to support causal inference.

GLP‐1 receptors are expressed in hypothalamic and limbic circuits involved in stress and affect regulation [18, 19, 20]. Preclinical work suggests that acute GLP‐1 receptor activation may engage hypothalamic–pituitary–adrenal and sympathetic pathways and induce anxiety‐like behaviour in animal models [21, 22, 23], whereas human data suggest neutral or potentially beneficial neuropsychiatric effects with chronic exposure [24, 25].

No study has yet evaluated panic attack as a discrete outcome using an active‐comparator restricted design with adjustment for indication and prespecified temporal restriction to address notoriety bias. We therefore conducted such an analysis of FAERS, complemented by a systematic review of randomized and observational evidence, using the semaglutide–panic attack pair as a case example of the difficulty of interpreting high‐profile pharmacovigilance signals in widely used medicines.

2.Methods

To address this objective, we conducted an active‐comparator restricted disproportionality analysis (ACR‐DA) using the FDA Adverse Event Reporting System (FAERS) and a systematic literature review.

2.1.Active‐Comparator Restricted Disproportionality Analysis

2.1.1.Data Source

The ACR‐DA was reported in accordance with the Reporting of a Disproportionality Analysis for Drug Safety Signal Detection Using Individual Case Safety Reports in PharmacoVigilance (READUS‐PV) guidance (Tables S1 and S2) [17]. FAERS is a publicly available spontaneous reporting database maintained by the FDA. Reporting is mandatory for pharmaceutical manufacturers and voluntary for healthcare professionals and patients, with submissions made through standardized MedWatch forms or electronic reporting systems [26]. Quarterly data files were obtained from the publicly accessible FAERS dashboard for the period January 1, 2018, to December 31, 2024, corresponding to the period of increasing global uptake of semaglutide [27].

2.1.2.Exposed and Active‐Comparator Groups

The exposed cohort comprised ICSRs in which semaglutide (marketed as Ozempic, Wegovy or Rybelsus) was identified as the primary suspect drug. Two active comparators were selected to reflect overlapping clinical indications and mitigate confounding by indication at the design stage. Dulaglutide is a within‐class GLP‐1 receptor agonist that shares semaglutide's mechanism of action and has overlapping use in Type 2 diabetes. This comparison was intended to provide class‐relevant context and help assess whether the reporting pattern was specific to semaglutide or shared within the class. Empagliflozin is a sodium‐glucose cotransporter‐2 inhibitor used in a broadly overlapping Type 2 diabetes population but without GLP‐1 receptor activity. This comparison was intended to provide context relative to a pharmacologically distinct diabetes treatment while improving comparability with respect to indication. Because the direction and magnitude of bias in a disproportionality analysis can depend on the comparator selected [12], both contrasts are reported. Orlistat, approved for weight management and linagliptin, a dipeptidyl peptidase‐4 inhibitor used for Type 2 diabetes, were also considered but were excluded because each had fewer than three panic‐attack reports, precluding sufficiently stable estimation. Labelled indications are presented in Table S3.

2.1.3.Outcome Definition and Data Cleaning

The primary outcome was the MedDRA preferred term ‘panic attack’. Related MedDRA terms (e.g., ‘panic reaction’) were excluded to ensure outcome specificity and case definition consistency. Duplicate cases were identified and removed using the unique case identifiers provided by the OpenFDA platform [28]. Adverse events were analyzed at the individual report level. Only ICSRs in which the medication of interest was classified as the primary suspect were retained. Drug names were normalized using regular expressions and mapped against standardized vocabularies, including RxNorm and the Observational Health Data Sciences and Informatics framework, with RxNorm Concept Unique Identifiers, generic names and Anatomical Therapeutic Chemical codes added to enrich the dataset [29, 30]. Clinical characteristics extracted from each ICSR included age, sex, indication for use, time to panic attack, reporting country and serious outcomes.

2.1.4.Statistical Analysis

Crude reporting odds ratios (RORs) with 95% confidence intervals were estimated using logistic regression. Bayesian disproportionality was assessed using the information component (IC) and its 95% lower bound (IC025) [31, 32]. A safety signal was defined when three criteria were jointly met: (i) the lower bound of the 95% CI for the ROR exceeded 1.0; (ii) IC025 exceeded 0; and (iii) at least three panic attack reports were available [33, 34].

Multivariable logistic regression was used to estimate adjusted reporting odds ratios (aRORs). Models included adjustment for covariates serving as proxies for confounding by indication, including sex (male as reference), age category (0–18, 19–44, 45–64 and ≥ 65 years; reference ≥ 65) and indication‐related variables (Type 2 diabetes and obesity/weight management) [35]. Because age and sex were unrecorded for a substantial proportion of reports, reports with unknown values were retained as separate categories in the multivariable models to preserve the full analytic sample. Indication was modelled using indicator variables for Type 2 diabetes and obesity or weight management; reports with neither indication recorded, including reports with another or an unrecorded indication, formed the reference category.

To mitigate confounding by indication, we conducted the primary analyses within a restricted active‐comparator framework, comparing semaglutide with dulaglutide (a within‐class GLP‐1 receptor agonist) and empagliflozin (a sodium‐glucose cotransporter‐2 inhibitor).

To account for potential notoriety bias following the July 2023 European Medicines Agency Pharmacovigilance Risk Assessment Committee investigation, we pre‐specified three complementary temporal strategies. First, pre‐warning analyses restricted to reports received before July 1, 2023, the prespecified analytic cut point marking the beginning of the month in which the review was initiated. Second, a drug‐by‐period comparison of 2018–2022 with 2023–2024, using whole calendar years because annual reporting proportions are the natural unit for that comparison. Third, we reported annual proportions of panic attack reports per 10 000 total reports for semaglutide and each comparator, stratified by those periods.

As a sensitivity analysis, we repeated the disproportionality analysis using all other primary‐suspect drugs reported in FAERS during 2018Q1–2024Q4 as the reference set. This analysis was intended to assess whether the temporal pattern observed in the active‐comparator analyses was robust to the use of a broader, heterogeneous reference set. We estimated crude and adjusted RORs for the full study period and separately before and on or after July 1, 2023. Adjusted models included age category, sex, Type 2 diabetes and obesity or weight‐management indication. We tested semaglutide‐by‐period interactions using both the July 1, 2023, cut point and calendar periods of 2018–2022 versus 2023–2024.

All analyses were performed in SAS version 9.4 (SAS Institute Inc., Cary, NC, USA), with visualization in R version 4.5.3.

2.2.Systematic Review

2.2.1.Protocol Registration and Search Strategy

The systematic review was conducted in accordance with the Preferred Reporting Items for Systematic Reviews and Meta‐Analyses (PRISMA) 2020 statement (Tables S4 and S5) [36]. The protocol was prospectively registered in the International Prospective Register of Systematic Reviews (PROSPERO; CRD420251183110).

MEDLINE (Ovid), Embase (Ovid) and the Cochrane Central Register of Controlled Trials (CENTRAL) were searched from inception, with the searches completed between November 21 and November 24, 2025. The search combined controlled vocabulary and free‐text terms for semaglutide and GLP‐1 receptor agonists with terms for panic attacks, panic disorder and anxiety. We applied no study‐design or language filters in the database searches. As prespecified in the registered protocol, we excluded studies not published in English during screening. The complete search syntax, database coverage and execution dates and numbers of records retrieved are provided in Tables S6–S8.

2.2.2.Inclusion and Exclusion Criteria

Studies were eligible if they enrolled human adults aged 18 years or older who were exposed to semaglutide by any route, formulation or indication and reported panic attacks or acute anxiety outcomes. Eligible designs included randomized controlled trials, observational cohort and cross‐sectional studies, pharmacovigilance analyses and case reports or case series. Studies evaluating GLP‐1 receptor agonists collectively were eligible only when semaglutide‐specific findings could be extracted. We excluded reviews, editorials, grey literature, in vitro and animal‐only studies, studies restricted to pregnant populations, ongoing trials without publicly available results and studies not published in English. Two reviewers independently screened titles, abstracts and full texts against the prespecified criteria, with disagreements resolved by consensus or third‐party adjudication.

2.2.3.Outcomes of Interest

The primary outcome was panic attacks, defined as abrupt surges of intense fear peaking within minutes according to DSM‐5 criteria [37]. Because panic attacks are frequently captured under broader anxiety‐related terminology in clinical and pharmacovigilance datasets, studies reporting acute anxiety or anxiety‐related outcomes were also included.

2.2.4.Study Selection and Data Extraction

Records were managed in Covidence (Veritas Health Innovation, Melbourne, Australia). After de‐duplication, two reviewers (B.M. and M.P.‐Z.) independently screened titles, abstracts and full texts against predefined eligibility criteria, with disagreements resolved by consensus or third‐party adjudication (F.T.M.). Data were extracted using a standardized form capturing study characteristics, population details, exposure and comparator definitions, outcome measures and effect estimates with corresponding 95% confidence intervals.

2.2.5.Risk of Bias Assessment

Risk of bias and methodological quality were assessed independently by two reviewers (B.M. and M.A.O.), with disagreements resolved by consensus. The Cochrane Risk of Bias 2 tool was used for randomized trials [38]. ROBINS‐I was used to assess risk of bias in estimates of comparative intervention effects from non‐randomized comparative cohort studies [39]. For each study and outcome, we specified a target trial and the effect estimate of interest before making domain‐level risk‐of‐bias judgements. For single‐arm before‐after studies and cross‐sectional surveys that did not estimate comparative intervention effects within a target‐trial framework, methodological quality and design limitations were assessed using the modified Downs and Black checklist [40]. The Naranjo scale was used to assess causal attribution in case reports. For pharmacovigilance analyses, we considered the choice of reference group, covariate adjustment, duplicate‐report management, outcome definition and susceptibility to underreporting and stimulated reporting.

2.2.6.Data Synthesis

Given heterogeneity in study designs, comparators and outcome definitions, findings were synthesized narratively by study type. Where sufficient data were available, effect estimates were recalculated and expressed as risk ratios with 95% confidence intervals. Analyses were performed in R (version 4.5.3; R Foundation for Statistical Computing, Vienna, Austria).

3.Results

3.1.ACR‐DA

3.1.1.Study Population

After de‐duplication and restriction to primary‐suspect reports, 33 267 semaglutide, 54 282 dulaglutide and 25 380 empagliflozin individual case safety reports (ICSRs) were included from 2018Q1 to 2024Q4 (Table 1).

TABLE 1Baseline demographic and clinical characteristics of reports associated with semaglutide, dulaglutide and empagliflozin from the FDA Adverse Event Reporting System (FAERS) database (2018Q1 to 2024Q4).

Characteristic Semaglutide Dulaglutide Empagliflozin

Total reports 33 267 54 282 25 380

Panic attack reports, n (%) 87 (0.26) 28 (0.05) 13 (0.05)

Age (years)

Median (IQR) 61.0 (51.0–70.0) 63.0 (55.0–71.0) 66.0 (57.0–74.0)

≤ 18, n (%) 17 (0.1) 10 (0.0) 1 (0.0)

19–44, n (%) 2565 (7.7) 1548 (2.9) 1019 (4.0)

45–64, n (%) 7246 (21.8) 8692 (16.0) 5323 (21.0)

≥ 65, n (%) 6837 (20.6) 8439 (15.5) 7665 (30.2)

Unknown, n (%) 16 602 (49.9) 35 593 (65.6) 11 372 (44.8)

Sex

Male, n (%) 10 784 (32.4) 20 401 (37.6) 11 723 (46.2)

Female, n (%) 20 497 (61.6) 27 828 (51.3) 9998 (39.4)

Unknown, n (%) 1986 (6.0) 6053 (11.2) 3659 (14.4)

Indication of use

Obesity/weight, n (%) 4099 (12.3) 232 (0.4) 17 (0.1)

Diabetes, n (%) 10 441 (31.4) 20 236 (37.3) 10 840 (42.7)

Serious outcomes

Death, n (%) 397 (1.2) 773 (1.4) 1109 (4.4)

Disabling, n (%) 594 (1.8) 216 (0.4) 243 (1.0)

Hospitalization, n (%) 4710 (14.2) 3459 (6.4) 6263 (24.7)

Life‐threatening, n (%) 583 (1.8) 221 (0.4) 1603 (6.3)

Other/not serious, n (%) 26 983 (81.1) 49 613 (91.4) 16 162 (63.7)

Reporting country

United States, n (%) 26 449 (79.5) 51 294 (94.5) 16 493 (65.0)

Great Britain, n (%) 803 (2.4) 214 (0.4) 1111 (4.4)

Germany, n (%) 165 (0.5) 230 (0.4) 1144 (4.5)

Canada, n (%) 604 (1.8) 41 (0.1) 761 (3.0)

Other, n (%) 3299 (9.9) 2272 (4.2) 4993 (19.7)

Unknown, n (%) 1947 (5.9) 231 (0.4) 878 (3.5)

Note: Total reports for each drug are indicated. Panic attack reports are shown as n (%) for each drug.

Abbreviation: IQR, interquartile range.

Panic attacks were reported in 87 semaglutide ICSRs (0.26%), 28 dulaglutide ICSRs (0.05%) and 13 empagliflozin ICSRs (0.05%). Baseline demographic, indication and reporting characteristics are summarized in Table 1.

3.1.2.Signal Detection (2018–2024)

Compared with dulaglutide, semaglutide was associated with increased reporting of panic attacks in crude analyses (ROR, 5.08; 95% CI, 3.32–7.78; IC025, 0.57) (Table 2; Figure 1).

TABLE 2Crude and adjusted reporting odds ratios for panic‐attack reports with semaglutide compared with two active comparators (dulaglutide and empagliflozin) and with all other primary‐suspect drugs in FAERS (full reference set), 2018Q1–2024Q4.

Drug of interest Comparator Crude IC (IC025) Crude ROR (95% CI) aROR Model 1 (95% CI)a
aROR Model 2 (95% CI)b

Semaglutide Dulaglutide 0.99 (0.57) 5.08 (3.32–7.78) 2.82 (1.74–4.56) 1.38 (0.69–2.75)

Semaglutide Empagliflozin 0.62 (0.03) 5.12 (2.86–9.17) 2.75 (1.41–5.36) 1.14 (0.48–2.68)

Semaglutide All other primary‐suspect drugs 0.88 (0.58) 1.85 (1.50–2.29) 1.48 (1.18–1.87) 0.76 (0.48–1.18)

Abbreviations: aROR, adjusted reporting odds ratio; CI, confidence interval; IC, information component; ROR, reporting odds ratio.

A positive disproportionality signal was defined as a lower bound of the 95% confidence interval for the ROR > 1.0, an IC025 > 0 and at least three panic‐attack reports.

Full‐reference analyses compared semaglutide with all other primary‐suspect drugs reported in FAERS during 2018Q1–2024Q4. Active‐comparator and full‐reference estimates depend on different reference distributions and address complementary potential sources of bias. Estimates for reports received on or after July 1, 2023, are reported in the Section 3.

aModel 1 (full study period): adjusted for age category, sex and indication (indicator variables for Type 2 diabetes and for obesity or weight management) across the entire 2018–2024 study period.

bModel 2 (pre‐warning period): restricted to reports received before July 1, 2023, to reduce the potential influence of notoriety bias and adjusted for the same covariates as Model 1.

Compared with empagliflozin, the crude ROR was 5.12 (95% CI 2.86–9.17; IC025 0.03) (Table 2; Figure 1).

After adjustment for age, sex and indication, the adjusted ROR was 2.82 (95% CI, 1.74–4.56) versus dulaglutide and 2.75 (95% CI, 1.41–5.36) versus empagliflozin (Table 2; Figure 1).

3.1.3.Temporal Analyses

3.1.3.1.Pre‐Warning Analysis (Before July 1, 2023)

In analyses restricted to reports received before July 1, 2023, crude RORs for semaglutide were 2.24 (95% CI, 1.23–4.07) versus dulaglutide and 2.15 (95% CI, 1.00–4.58) versus empagliflozin. After adjustment for age, sex and indication, these estimates attenuated to 1.38 (95% CI, 0.69–2.75) and 1.14 (95% CI, 0.48–2.68), respectively and were no longer statistically significant (Table 2; Figure 1). Panic‐attack reports accounted for 0.108% of semaglutide reports received before July 1, 2023 and 0.262% of all semaglutide reports in the full study period, whereas the corresponding proportions were 0.048% and 0.052% for dulaglutide, 0.051% in both periods for empagliflozin and 0.142% and 0.140% for all other primary‐suspect drugs (Figure 2).

3.1.3.2.Drug‐By‐Period Interaction Analysis

A significant drug‐by‐period interaction was observed in the comparison of 2018–2022 with 2023–2024 (p = 0.007) (Figure 3).

For semaglutide, the odds ratio for panic attack reporting in 2023–2024 versus 2018–2022 was 6.31 (95% CI, 3.26–12.20). For dulaglutide, the corresponding odds ratio was 2.95 (95% CI, 1.39–6.23). For empagliflozin, the odds ratio was 0.65 (95% CI, 0.18–2.35).

3.1.3.3.Annual Reporting Patterns

Annual reporting proportions of panic attack ICSRs per 10 000 total reports for semaglutide were near zero in 2018.

Annual proportions for dulaglutide ranged from 0 to 11.8 per 10 000 across the study period and those for empagliflozin ranged from 0 to 5 per 10 000. Aggregated by calendar period, the proportion rose from 6.7 per 10 000 reports in 2018–2022 to 42.0 in 2023–2024 for semaglutide and from 3.7 to 10.9 for dulaglutide and fell from 5.8 to 3.7 for empagliflozin (Figure 3). In the full reference set of all other primary‐suspect drugs, the corresponding proportions were 14.3 and 12.9 per 10 000 reports (Figure 3).

3.1.4.Full‐Reference Analysis

The full‐reference analysis included 33 267 semaglutide reports and 9 518 893 reports involving all other primary‐suspect drugs. For the full study period, the crude ROR for semaglutide was 1.85 (95% CI, 1.50–2.29; IC, 0.88; IC025, 0.58) and the adjusted ROR was 1.48 (95% CI, 1.18–1.87) (Table 2; Figure 1). Before July 1, 2023, the crude and adjusted RORs were 0.76 (95% CI, 0.49–1.18) and 0.76 (95% CI, 0.48–1.18), respectively (Table 2; Figure 1). On or after July 1, 2023, the corresponding estimates were 3.50 (95% CI, 2.74–4.46) and 2.15 (95% CI, 1.57–2.93), respectively. Semaglutide‐by‐period interactions were statistically significant using both the July 1, 2023, cut point and the comparison of 2018–2022 with 2023–2024 (both p < 0.001) (Figure 3).

3.2.Systematic Review

3.2.1.Study Selection

Database searches identified 1002 records that met the eligible publication types (Embase n = 552; CENTRAL n = 342; MEDLINE n = 108). The CENTRAL search returned 496 records across all Cochrane Library segments, of which 342 were trial records; Cochrane Reviews, protocols and clinical answers were not eligible publication types and were not exported. After removal of 111 duplicates, 891 records underwent title and abstract screening, of which 845 were excluded.

Forty‐six full‐text articles were assessed for eligibility; 34 were excluded for incorrect outcome (n = 17), study design (n = 5), intervention (n = 1), language (n = 1) or ongoing studies without results (n = 10).

Twelve studies met eligibility criteria and were included in the qualitative synthesis (Figure 4; Tables S9–S12): one post hoc randomized trial analysis, six observational or cross‐sectional studies, four pharmacovigilance analyses and one case report [6, 8, 41, 42, 43, 44, 45, 46, 47, 48, 49, 50]. Risk of bias assessments are provided in Tables S13–S16.

The included studies were heterogeneous in design, population, comparator and outcome ascertainment. Only two of the 12 studies reported panic attacks as a discrete outcome and one of these included only five panic‐attack events. The remaining studies assessed broader outcomes, including anxiety diagnoses, anxiety‐related healthcare utilization or changes in anxiety symptom scores. These outcomes are not interchangeable with panic attacks. Given the clinical and methodological heterogeneity across studies, quantitative synthesis was not considered appropriate.

3.2.2.Randomized Controlled Trials

Wadden et al. pooled data from the STEP 1, 2, 3 and 5 randomized, double‐blind, placebo‐controlled trials (n = 3377; semaglutide n = 2116; placebo n = 1261). Panic attacks occurred in 3 of 2116 participants receiving semaglutide and in 2 of 1261 receiving placebo (risk ratio, 0.89; 95% CI, 0.15–5.34). Anxiety disorders and symptoms (MedDRA high‐level group term) were reported in 75 of 2116 semaglutide‐treated participants and 50 of 1261 placebo‐treated participants (risk ratio, 0.89; 95% CI, 0.63–1.27). Risk of bias was assessed as low (RoB 2).

3.2.3.Cohorts and Cross‐Sectional Studies

Six observational or cross‐sectional studies evaluated broader anxiety‐related outcomes—rather than panic attacks specifically—across administrative claims databases and survey cohorts.

In a propensity‐score matched cohort of 46 772 adults with Type 2 diabetes (TriNetX), De Giorgi et al. reported a hazard ratio of 0.92 (95% CI, 0.84–1.01) for incident anxiety disorders comparing semaglutide with sitagliptin (ROBINS‐I, moderate risk of bias).

In a difference‐in‐differences analysis of 774 968 adults with Type 2 diabetes (Komodo Healthcare Map), Do et al. reported a semaglutide‐specific incidence rate ratio of 0.81 (95% CI 0.76–0.87) for anxiety‐related office visits compared with dipeptidyl peptidase‐4 inhibitor use. Corresponding semaglutide‐specific estimates were 0.98 (95% CI 0.85–1.12) for emergency department visits, 1.26 (95% CI 0.90–1.77) for inpatient visits and 0.99 (95% CI 0.97–1.02) for outpatient hospital visits. This outcome reflected healthcare utilization for a broad group of anxiety diagnoses rather than panic attacks specifically (ROBINS‐I, moderate risk of bias).

In a cohort of 324 506 adults with obesity (TriNetX), Kornelius et al. reported a hazard ratio of 2.36 (95% CI 2.23–2.49) for incident anxiety in the semaglutide 2.4 mg subgroup compared with untreated obese controls (ROBINS‐I, serious risk of bias).

Two cross‐sectional surveys reported lower self‐reported anxiety among semaglutide users. A Polish survey (n = 1105) reported lower GAD‐7 scores among semaglutide users compared with non‐users and a Kuwaiti survey (n = 468) reported lower anxiety risk compared with liraglutide (risk ratio, 0.38; 95% CI, 0.22–0.67) and no difference versus tirzepatide (0.80; 95% CI, 0.41–1.55). Both surveys were susceptible to volunteer selection and self‐report biases; Witaszek et al. used the GAD‐7 scale, whereas Alali et al. used an unvalidated self‐report instrument (modified Downs and Black scores, 15/28 and 14/28, respectively). A German single‐arm before‐after study of 122 patients reported a reduction in HADS‐A scores over 1 year without a concurrent comparator (modified Downs and Black score, 14/28).

3.2.4.Pharmacovigilance Studies

Four pharmacovigilance analyses evaluated spontaneous reporting databases (FAERS, VigiBase, DAEN).

In FAERS (2010–2024), Lu et al. reported crude reporting odds ratios (RORs) of 1.30 (95% CI 1.12–1.50) for anxiety in diabetes indications and 0.49 (95% CI 0.40–0.60) for weight‐control indications.

Chen et al. reported a ROR of 1.41 (95% CI, 1.23–1.63) for grouped psychiatric adverse events, with a median onset of 12 days (IQR, 4–45.5).

Katranski et al. reported a FAERS ROR of 1.46 (95% CI 1.16–1.82) for panic attacks and a DAEN ROR of 3.56 (95% CI 1.14–11.12).

In VigiBase, Nishida et al. reported an adjusted ROR of 1.26 (95% CI 1.18–1.35) for anxiety disorders, with a median onset of 7 days (IQR 1–33).

Of the four included pharmacovigilance analyses, three used unrestricted reference sets without adjustment. Nishida et al. used an active‐comparator design within antidiabetic and anti‐obesity medications with adjustment for age, sex and WHO region. None restricted the analysis temporally around the July 2023 regulatory communication and none reported panic attack as a discrete outcome adjusted for indication.

3.2.5.Risk of Bias and Methodological Quality

The pooled randomized trial analysis was judged to have a low overall risk of bias using RoB 2, although the small number of panic‐attack events limited precision. Among the three non‐randomized comparative cohort studies assessed using ROBINS‐I, De Giorgi et al. and Do et al. were judged to have a moderate overall risk of bias for the relevant comparative effect estimates, whereas Kornelius et al. were judged to have a serious overall risk of bias, principally because of concerns regarding confounding, selection of participants into the study and measurement of outcomes. The single‐arm before‐after study and the two cross‐sectional surveys had important design and methodological limitations, as summarized using the modified Downs and Black checklist. The pharmacovigilance analyses were limited by spontaneous‐reporting biases and their inability to estimate incidence; three of the four used unrestricted reference groups without covariate adjustment. These methodological limitations reduce certainty in the observational evidence (Tables S13–S16; Figure S2).

3.2.6.Case Report

A single case report described a male patient in his late 70s with Type 2 diabetes who developed acute anxiety, inner restlessness and distress 4 weeks after initiation of oral semaglutide 3 mg. Symptoms resolved within 9 days of discontinuation and initiation of sertraline and quetiapine. The Naranjo score was 7 (probable).

4.Discussion

4.1.Summary of Findings

We combined an ACR‐DA of FAERS with a systematic review of randomized and observational evidence to evaluate how confounding by indication and notoriety bias may shape the semaglutide–panic‐attack signal. Crude analyses suggested increased disproportional reporting. This association was substantially attenuated after adjustment for age, sex and indication and was no longer statistically significant in analyses restricted to the pre‐warning period before July 2023. Randomized trials and propensity‐score‐matched active‐comparator cohort studies did not suggest an increased risk of panic attacks or related anxiety outcomes with semaglutide, although event counts were low, trial estimates were imprecise and residual confounding may persist in observational cohorts. Taken together, our findings are more readily explained by confounding by indication and notoriety‐driven reporting than by a pharmacological effect, though an association with panic attacks cannot be excluded.

4.2.ACR‐DA

In the ACR‐DA, the initial disproportionality signal for semaglutide was reduced after adjustment for demographic and clinical characteristics and after restriction to the pre‐warning period. These findings are consistent with the possibility that differences in underlying patient characteristics, including age, sex distribution and indication (obesity versus diabetes), contributed to observed reporting patterns in spontaneous reporting data.

Several features of the post‐2023 reporting pattern suggest that it is more consistent with stimulated reporting than with a pharmacological effect. First, in the calendar‐period comparison, the proportion of semaglutide reports describing panic attacks increased from 6.7 per 10 000 reports in 2018–2022 to 42.0 per 10 000 reports in 2023–2024, corresponding to an OR of 6.31 (95% CI, 3.26–12.20). This increase coincided temporally with the 2023 EMA review and heightened public attention. Second, the increase was more pronounced for semaglutide than for the comparators: the corresponding OR was 2.95 (95% CI, 1.39–6.23) for dulaglutide and 0.65 (95% CI, 0.18–2.35) for empagliflozin, with evidence of a drug‐by‐period interaction (p = 0.007). Third, before July 2023, the adjusted RORs were 1.38 versus dulaglutide and 1.14 versus empagliflozin and were not statistically significant. Together, the timing and drug‐specificity of the increase are compatible with stimulated reporting [13, 14, 15]. However, FAERS lacks exposure denominators and these analyses cannot exclude contributions from changes in utilization, patient mix, event occurrence or other time‐varying factors. The temporal pattern should therefore be interpreted as suggestive rather than definitive evidence of notoriety bias.

In the sensitivity analysis using all other primary‐suspect drugs as the reference set, the temporal pattern was consistent with that observed in the active‐comparator analyses. No statistically significant disproportionality was observed before July 2023, whereas reporting was elevated thereafter, with statistically significant semaglutide‐by‐period interactions. This finding indicates that the observed post‐2023 temporal pattern was not dependent solely on the use of dulaglutide or empagliflozin as restricted active‐comparator reference groups.

The magnitude of the full‐period disproportionality estimate differed by reference set. The crude ROR for semaglutide was lower when compared with all other primary‐suspect drugs (ROR, 1.85) than when compared with dulaglutide (ROR, 5.08) or empagliflozin (ROR, 5.12). Panic attacks were reported in 14.0 per 10 000 reports in the full reference set, compared with approximately 5.2 per 10 000 dulaglutide reports and 5.1 per 10 000 empagliflozin reports. These differences demonstrate that the reporting profile of the selected comparator can materially influence the magnitude of a disproportionality estimate [12].

Thus, although the broader reference set attenuated the overall disproportionality estimate, the post‐2023 temporal pattern persisted and was compatible with, but not specific to, stimulated reporting. Changes in semaglutide utilization, patient characteristics, clinical recognition or coding of panic attacks, co‐reported conditions or medications, background event occurrence, media attention and other time‐varying factors may also have contributed. Conversely, because the full reference set includes drugs used across diverse indications and patient populations, it provides less control for confounding by indication and other between‐drug differences than an active‐comparator design. The active‐comparator and full‐reference analyses therefore address complementary concerns and should be interpreted jointly rather than as interchangeable estimates.

4.3.Systematic Review

The systematic review identified 12 studies with heterogeneous designs, populations and outcome definitions. The most consequential source of heterogeneity is the outcome itself. Only two studies reported panic attacks as a discrete outcome and one of these contributed five events in total. The remaining studies reported broader constructs, including anxiety diagnoses in claims or electronic health records, anxiety‐related healthcare utilization and self‐reported symptom scores, which are not interchangeable with panic attacks. Randomized trials from the STEP program showed no imbalance in either outcome but were not powered for rare events. Active‐comparator cohort studies generally reported null or inverse associations, whereas the study using an untreated comparator reported an elevated hazard ratio. Under ROBINS‐I, the latter study was judged to have serious risk of bias, particularly because of concerns about confounding, selection of participants into the study and outcome measurement. Compared with an active‐comparator design, an untreated‐comparator design may be more susceptible to residual confounding when treatment initiation is associated with baseline clinical characteristics, healthcare‐seeking behaviour or other factors related to anxiety risk. In addition, anxiety ascertainment based on diagnostic codes requires healthcare contact, which may differ between treated and untreated individuals. Consequently, this design was less able to distinguish a potential drug effect from differences in baseline risk, healthcare utilization or outcome ascertainment.

Pharmacovigilance studies using spontaneous reporting databases reported variable disproportionality estimates; however, only one used an active‐comparator design and none adjusted for indication or accounted for temporal reporting effects. Cross‐sectional survey data were subject to selection and reporting bias and uncontrolled cohort data did not permit causal inference. Substantial heterogeneity in outcome definitions limited comparability across studies and precluded quantitative synthesis. Overall, we interpreted these heterogeneous findings within a triangulation framework, evaluating the coherence of estimates across trials, cohort studies and pharmacovigilance analyses with FAERS patterns, while accounting for confounding by indication and notoriety bias. Interpretation prioritized evidence from randomized trials and active‐comparator cohort studies over pharmacovigilance estimates.

4.4.Comparison With Prior Studies

Our findings are broadly consistent with recent large observational studies using active‐comparator and longitudinal designs, which have not demonstrated an association between GLP‐1 receptor agonist use and serious psychiatric outcomes [51, 52]. In contrast, earlier spontaneous reporting analyses suggested potential signals; however, these studies generally relied on broad or ‘all‐drug’ comparators and did not apply active‐comparator restricted or temporally stratified designs, making them susceptible to confounding by indication and notoriety bias [8, 47, 48]. Differences in study design, comparator selection and temporal context likely account for discrepancies across studies.

4.5.Biological Considerations

GLP‐1 receptors are expressed in central nervous system regions involved in autonomic and affective regulation, including hypothalamic and limbic structures. Preclinical studies have shown that acute GLP‐1 receptor activation may engage stress‐related neuroendocrine pathways and induce anxiety‐like behavior in animal models [18, 19, 20]. However, clinical trial and observational data do not suggest consistent increases in anxiety‐related outcomes with therapeutic GLP‐1 receptor agonist use [24, 25]. The clinical relevance of mechanistic findings remains uncertain, particularly given differences between experimental exposure conditions and chronic clinical use.

4.6.Strengths and Limitations

This study has several strengths, including the use of an active‐comparator restricted design, prespecified temporal analyses to evaluate reporting dynamics and triangulation with a systematic review of randomized and observational evidence. The incorporation of temporal sensitivity analyses strengthens the interpretation of potential reporting bias in pharmacovigilance data. Importantly, we did not assume a single definitive reference standard for semaglutide‐associated panic attacks. Instead, we interpreted the FAERS signal alongside randomized trials and observational studies, placing greater emphasis on evidence from stronger causal inference designs, while recognizing that each data source addresses a different estimand.

Disproportionality analysis of spontaneous reports is a hypothesis‐generating method [17]. It compares the reporting of a particular event for a drug of interest with reporting for a specified comparator drug or group of drugs or with all other drugs in the database. It cannot estimate the incidence of panic attacks among people taking semaglutide because the number of exposed individuals is unknown and no population denominator is available. It also cannot establish that semaglutide caused any reported event and a reporting odds ratio should not be interpreted as a relative risk.

Spontaneous reporting systems are subject to underreporting and incomplete clinical information. Although underreporting does not necessarily bias a reporting odds ratio, bias may arise if panic attacks are preferentially reported for one drug relative to the comparator or if reporting patterns change over time, such as following regulatory communications or heightened public attention [29, 53].

Panic‐attack reports in FAERS are not routinely clinically verified and MedDRA preferred terms may not correspond to episodes meeting DSM‐5 criteria. Publicity surrounding a possible psychiatric safety concern could also influence how symptoms are recognized, described or coded, potentially introducing differential outcome misclassification. However, the direction and magnitude of any such misclassification cannot be determined from FAERS.

Missing data warrant particular emphasis. Age was unrecorded in approximately half of semaglutide reports and two‐thirds of dulaglutide reports; sex was missing in a smaller but differing proportion across drug groups and indication was recorded in fewer than half of reports in each group. The extent of missingness therefore differed between semaglutide and the comparators. Retaining unknown age and sex values as separate categories preserved the full sample but does not eliminate bias if covariate recording was related to the drug, the reported event, reporter characteristics or reporting route. Moreover, the limited recording of indication constrained our ability to address confounding by indication. Consequently, adjustment for age, sex and indication may have been incomplete and the adjusted reporting odds ratios should be interpreted cautiously.

Neither comparator is approved for weight management. Orlistat, the candidate comparator approved for weight management, was considered but could not be used because fewer than three panic‐attack reports were available. Obesity or weight management was recorded as the indication in 12.3% of semaglutide reports but in only 0.4% of dulaglutide and 0.1% of empagliflozin reports (Table 1). Adjustment for this indication therefore relies on very few comparator reports and residual confounding by indication among patients treated for weight management cannot be excluded.

The evidence identified in the systematic review provides limited certainty regarding panic attacks specifically. Only two studies evaluated panic attacks as a discrete outcome: a pooled trial analysis with five events in total and a pharmacovigilance analysis, which cannot estimate risk. Most observational studies assessed broader anxiety diagnoses, anxiety‐related healthcare utilization or symptom scales, which may not capture panic attacks and cannot be assumed to represent the same clinical outcome. The included studies also differed substantially in design, study population, comparator, follow‐up, outcome definition and risk of bias. Accordingly, the lack of a consistent increase in risk across this broader evidence base should be interpreted cautiously and does not exclude an association with panic attacks specifically.

Residual confounding by disease severity, psychiatric history and healthcare utilization may persist despite comparator restriction and covariate adjustment [29, 54, 55].

5.Conclusions

In this pharmacovigilance analysis, semaglutide was associated with increased reporting of panic attacks in crude FAERS analyses; however, this association was attenuated after adjustment for demographic and indication‐related factors and the adjusted estimates were not statistically significant in analyses restricted to reports received before July 1, 2023. Evidence from randomized trials and active‐comparator observational cohort studies did not suggest an increased risk of panic attacks or related anxiety outcomes with semaglutide.

Taken together, these findings are more consistent with confounding by indication and notoriety‐driven reporting within spontaneous reporting data than with a robust pharmacological signal. Nevertheless, given the inherent limitations of spontaneous reporting systems and observational evidence, residual uncertainty remains and further controlled pharmacoepidemiological studies are warranted.

This case study illustrates how active‐comparator disproportionality analysis, temporal assessment of reporting dynamics and evidence triangulation can improve the interpretation of post‐marketing safety signals for widely used medications.

Author Contributions

Ben Melamed: conceptualization, formal analysis, investigation, methodology, visualization, writing original draft, writing review and editing. Mahak Parsi‐Zahabi: investigation, data curation. Mohammad Ali Omrani: methodology, investigation, data curation, writing review and editing. Bala Swetha Baskaran: investigation, data curation. Lujain Ez Eddin: investigation, data curation. Nazgol Kermanshahi: investigation, data curation. Niaz Chalabianloo: data curation, software. Flory Tsobo Muanda: conceptualization, supervision, writing review and editing, project administration.

Funding

The authors have nothing to report.

Conflicts of Interest

The authors declare no conflicts of interest.

Supporting information

Acknowledgements

The authors have nothing to report.

Data Availability Statement

The data that support the findings of this study are publicly available at https://open.fda.gov.

References

1.
Davies M., Færch L., Jeppesen O. K., et al., “Semaglutide 2·4 mg Once a Week in Adults With Overweight or Obesity, and Type 2 Diabetes (STEP 2): A Randomised, Double‐Blind, Double‐Dummy, Placebo‐Controlled, Phase 3 Trial,” Lancet
397, no. 10278 (2021): 971–984, 10.1016/S0140-6736(21)00213-0.

2.
Marso S. P., Bain S. C., Consoli A., et al., “Semaglutide and Cardiovascular Outcomes in Patients With Type 2 Diabetes,” New England Journal of Medicine
375, no. 19 (2016): 1834–1844, 10.1056/NEJMoa1607141.

3.
Ryan D. H., Lingvay I., Deanfield J., et al., “Long‐Term Weight Loss Effects of Semaglutide in Obesity Without Diabetes in the SELECT Trial,” Nature Medicine
30, no. 7 (2024): 2049–2057, 10.1038/s41591-024-02996-7.

4.
Wilding J. P. H., Batterham R. L., Calanna S., et al., “Once‐Weekly Semaglutide in Adults With Overweight or Obesity,” New England Journal of Medicine
384, no. 11 (2021): 989–1002, 10.1056/NEJMoa2032183.

5.
Kushner R. F., Ryan D. H., Deanfield J., et al., “Safety Profile of Semaglutide Versus Placebo in the SELECT Study: A Randomized Controlled Trial,” Obesity
33, no. 3 (2025): 452–462, 10.1002/OBY.24222.

6.
Wadden T. A., Brown G. K., Egebjerg C., et al., “Psychiatric Safety of Semaglutide for Weight Management in People Without Known Major Psychopathology: Post Hoc Analysis of the STEP 1, 2, 3, and 5 Trials,” JAMA Internal Medicine
184, no. 11 (2024): 1290–1300, 10.1001/JAMAINTERNMED.2024.4346.

7.
Tobaiqy M. and Elkout H., “Psychiatric Adverse Events Associated With Semaglutide, Liraglutide and Tirzepatide: A Pharmacovigilance Analysis of Individual Case Safety Reports Submitted to the EudraVigilance Database,” International Journal of Clinical Pharmacy
46, no. 2 (2024): 488–495, 10.1007/S11096-023-01694-7/TABLES/3.

8.
Chen W., Cai P., Zou W., and Fu Z., “Psychiatric Adverse Events Associated With GLP‐1 Receptor Agonists: A Real‐World Pharmacovigilance Study Based on the FDA Adverse Event Reporting System Database,” Frontiers in Endocrinology (Lausanne)
15 (2024): 1330936, 10.3389/FENDO.2024.1330936/TEXT.

9.
US Food and Drug Administration
, “FDA Requests Removal of Suicidal Behavior and Ideation Warning From Glucagon‐Like Peptide‐1 Receptor Agonist (GLP‐1 RA) Medications,” 2026, https://www.fda.gov/drugs/drug‐safety‐communications/fda‐requests‐removal‐suicidal‐behavior‐and‐ideation‐warning‐glucagon‐peptide‐1‐receptor‐agonist‐glp.

10.
Ruggiero R., Mascolo A., Spezzaferri A., et al., “Glucagon‐Like Peptide‐1 Receptor Agonists and Suicidal Ideation: Analysis of Real‐Word Data Collected in the European Pharmacovigilance Database,” Pharmaceuticals
17, no. 2 (2024): 147, 10.3390/PH17020147/S1.

11.
Yoshida K., Solomon D. H., and Kim S. C., “Active‐Comparator Design and New‐User Design in Observational Studies,” Nature Reviews Rheumatology
11, no. 7 (2015): 437–441, 10.1038/nrrheum.2015.30.

12.
Gravel C. A., Bai W., and Douros A., “Comparators in Pharmacovigilance: A Quasi‐Quantification Bias Analysis,” Drug Safety
47, no. 8 (2024): 809–819, 10.1007/s40264-024-01433-5.

13.
Neha R., Subeesh V., Beulah E., Gouri N., and Maheswari E., “Existence of Notoriety Bias in FDA Adverse Event Reporting System Database and Its Impact on Signal Strength,” Hospital Pharmacy
56, no. 3 (2021): 152–158, 10.1177/0018578719882323.

14.
Pariente A., Gregoire F., Fourrier‐Reglat A., Haramburu F., and Moore N., “Impact of Safety Alerts on Measures of Disproportionality in Spontaneous Reporting Databases: The Notoriety Bias,” Drug Safety
30, no. 10 (2007): 891–898, 10.2165/00002018-200730100-00007.

15.
de Boissieu P., Kanagaratnam L., Abou Taam M., Roux M. P., Drame M., and Trenque T., “Notoriety Bias in a Database of Spontaneous Reports: The Example of Osteonecrosis of the Jaw Under Bisphosphonate Therapy in the French National Pharmacovigilance Database,” Pharmacoepidemiology and Drug Safety
23, no. 9 (2014): 989–992, 10.1002/pds.3622.

16.
Hammad T. A., Naylor M., Ely D. M., and Davies S., “Exploring the Complexities of Disproportionality Analysis in Pharmacovigilance: Reflections on the READUS‐PV Guideline and a Call to Action,” Frontiers in Pharmacology
16 (2025): 1573353, 10.3389/fphar.2025.1573353.

17.
Fusaroli M., Salvo F., Begaud B., et al., “The REporting of A Disproportionality Analysis for DrUg Safety Signal Detection Using Individual Case Safety Reports in PharmacoVigilance (READUS‐PV): Explanation and Elaboration,” Drug Safety
47, no. 6 (2024): 585–599, 10.1007/s40264-024-01423-7.

18.
Trapp S. and Brierley D. I., “Brain GLP‐1 and the Regulation of Food Intake: GLP‐1 Action in the Brain and Its Implications for GLP‐1 Receptor Agonists in Obesity Treatment,” British Journal of Pharmacology
179, no. 4 (2022): 557–570, 10.1111/BPH.15638.

19.
Drucker D. J., “Mechanisms of Action and Therapeutic Application of Glucagon‐Like Peptide‐1,” Cell Metabolism
27, no. 4 (2018): 740–756, 10.1016/j.cmet.2018.03.001.

20.
Holst J. J., “The Physiology of Glucagon‐Like Peptide 1,” Physiological Reviews
87, no. 4 (2007): 1409–1439, 10.1152/physrev.00034.2006.

21.
Anderberg R. H., Richard J. E., Hansson C., Nissbrandt H., Bergquist F., and Skibicka K. P., “GLP‐1 Is Both Anxiogenic and Antidepressant; Divergent Effects of Acute and Chronic GLP‐1 on Emotionality,” Psychoneuroendocrinology
65 (2016): 54–66, 10.1016/J.PSYNEUEN.2015.11.021.

22.
Kinzig K. P., D'Alessio D. A., Herman J. P., et al., “CNS Glucagon‐Like Peptide‐1 Receptors Mediate Endocrine and Anxiety Responses to Interoceptive and Psychogenic Stressors,” Journal of Neuroscience
23, no. 15 (2003): 6163–6170, 10.1523/JNEUROSCI.23-15-06163.2003.

23.
Diz‐Chaves Y., Herrera‐Pérez S., González‐Matías L. C., Lamas J. A., and Mallo F., “Glucagon‐Like Peptide‐1 (GLP‐1) in the Integration of Neural and Endocrine Responses to Stress,” Nutrients
12, no. 11 (2020): 3304, 10.3390/NU12113304.

24.
Wilbon S. S. and Kolonin M. G., “GLP1 Receptor Agonists—Effects Beyond Obesity and Diabetes,” Cells
13, no. 1 (2023): 65, 10.3390/CELLS13010065.

25.
AbuAlrob M. A., Itbaisha A., Abujwaid Y. K., Abulehia A., Hussein A., and Mesraoua B., “Exploring the Neuroprotective Role of GLP‐1 Agonists Against Alzheimer's Disease: Real‐World Evidence From a Propensity‐Matched Cohort,” Journal of Alzheimer's Disease Reports
9 (2025): 25424823251388650, 10.1177/25424823251388650.

26.
Sakaeda T., Tamon A., Kadoyama K., and Okuno Y., “Data Mining of the Public Version of the FDA Adverse Event Reporting System,” International Journal of Medical Sciences
10, no. 7 (2013): 796–803, 10.7150/IJMS.6048.

27.
“FAERS US Food and Drug Administration FDA Adverse Event Reporting System (FAERS),” accessed May 12, 2026, https://open.fda.gov/.

28.
Kass‐Hout T. A., Xu Z., Mohebbi M., et al., “OpenFDA: An Innovative Platform Providing Access to a Wealth of FDA's Publicly Available Data,” Journal of the American Medical Informatics Association
23, no. 3 (2016): 596–600, 10.1093/JAMIA/OCV153.

29.
Chalabianloo N., Abdullah S. S., Omrani M. A., Jafari A., Sedig K., and Muanda F. T., “Enhancing Adverse Drug Reaction Data Quality in Canada: A High‐Precision Pipeline for Medication Name Standardization and Enrichment,” PLoS One
20, no. 9 (2025): e0331940, 10.1371/JOURNAL.PONE.0331940.

30.
Hripcsak G., Duke J. D., Shah N. H., et al., “Observational Health Data Sciences and Informatics (OHDSI): Opportunities for Observational Researchers,” Studies in Health Technology and Informatics
216 (2015): 574–578, 10.3233/978-1-61499-564-7-574.

31.
Alkabbani W. and Gamble J. M., “Active‐Comparator Restricted Disproportionality Analysis for Pharmacovigilance Signal Detection Studies of Chronic Disease Medications: An Example Using Sodium/Glucose Cotransporter 2 Inhibitors,” British Journal of Clinical Pharmacology
89, no. 2 (2023): 431–439, 10.1111/BCP.15178.

32.
Omrani M. A., Tsobo C. T., Chalabianloo N., Ahmadi F., Abdullah S. S., and Muanda F. T., “Fluoroquinolones and Risk of Nightmares: A Literature Review and Disproportionality Analysis Using Individual Case Safety Reports From Food and Drug Administration Adverse Event Reporting System Database,” Journal of Psychopharmacology
39, no. 8 (2025): 782–789, 10.1177/02698811251344684.

33.
Van Puijenbroek E. P., Bate A., Leufkens H. G. M., Lindquist M., Orre R., and Egberts A. C. G., “A Comparison of Measures of Disproportionality for Signal Detection in Spontaneous Reporting Systems for Adverse Drug Reactions,” Pharmacoepidemiology and Drug Safety
11, no. 1 (2002): 3–10, 10.1002/PDS.668.

34.
Rothman K. J., Lanes S., and Sacks S. T., “The Reporting Odds Ratio and Its Advantages Over the Proportional Reporting Ratio,” Pharmacoepidemiology and Drug Safety
13, no. 8 (2004): 519–523, 10.1002/PDS.1001.

35.
Lu Z., Suzuki A., and Wang D., “Statistical Methods for Exploring Spontaneous Adverse Event Reporting Databases for Drug‐Host Factor Interactions,” BMC Medical Research Methodology
23, no. 1 (2023): 71‐, 10.1186/s12874-023-01885-w.

36.
Page M. J., McKenzie J. E., Bossuyt P. M., et al., “The PRISMA 2020 Statement: An Updated Guideline for Reporting Systematic Reviews,” BMJ
372 (2021): 1–9, 10.1136/BMJ.N71.

37.
American Psychiatric Association
, “Anxiety Disorders,” in Diagnostic and Statistical Manual of Mental Disorders, 5th ed., text rev. ed. (American Psychiatric Association Publishing, 2022), 10.1176/APPI.BOOKS.9780890425787.X05_ANXIETY_DISORDERS.

38.
Sterne J. A. C., Savović J., Page M. J., et al., “RoB 2: A Revised Tool for Assessing Risk of Bias in Randomised Trials,” BMJ
366 (2019): l4898, 10.1136/BMJ.L4898.

39.
Sterne J. A., Hernan M. A., Reeves B. C., et al., “ROBINS‐I: A Tool for Assessing Risk of Bias in Non‐Randomised Studies of Interventions,” BMJ
355 (2016): i4919, 10.1136/bmj.i4919.

40.
Downs S. H. and Black N., “The Feasibility of Creating a Checklist for the Assessment of the Methodological Quality Both of Randomised and Non‐Randomised Studies of Health Care Interventions,” Journal of Epidemiology and Community Health
52, no. 6 (1998): 377–384, 10.1136/JECH.52.6.377.

41.
De Giorgi R., Koychev I., Adler A. I., et al., “12‐Month Neurological and Psychiatric Outcomes of Semaglutide Use for Type 2 Diabetes: A Propensity‐Score Matched Cohort Study,” EClinicalMedicine
74 (2024): 102726, 10.1016/j.eclinm.2024.102726.

42.
Do D., Lee T., Inneh A., and Patel U., “Glucagon‐Like Peptide‐1 Use and Healthcare Resource Utilization for Depression and Anxiety Among Adults With Type 2 Diabetes: 2019 to 2023,” Journal of Behavioral Health Services and Research
53, no. 4 (2025): 1–17, 10.1007/s11414-025-09950-6.

43.
Kornelius E., Huang J. Y., Lo S. C., Huang C. N., and Yang Y. S., “The Risk of Depression, Anxiety, and Suicidal Behavior in Patients With Obesity on Glucagon Like Peptide‐1 Receptor Agonist Therapy,” Scientific Reports
14, no. 1 (2024): 24433‐, 10.1038/s41598-024-75965-2.

44.
Witaszek T., Kłoda K., Mastalerz‐Migas A., and Babicki M., “Association Between Symptoms of Depression and Generalised Anxiety Disorder Evaluated Through PHQ‐9 and GAD‐7 and Anti‐Obesity Treatment in Polish Adult Women,” Nutrients
16, no. 15 (2024): 2438, 10.3390/nu16152438.

45.
Alali S., Al‐Otaibi W., Ashodian K., Achour N., Almulla A., and Mutlaq H., “A Comparison of Glucagon‐Like Peptide‐1 Receptor Agonists on Weight Change, Side Effects, and Quality of Life in Kuwait,” Frontiers in Nutrition
12 (2025): 1590379, 10.3389/fnut.2025.1590379.

46.
Balata M., Sugiura A., Hassan M., et al., “Impact of Semaglutide on Health Outcomes and Mood in Obese Heart Failure Patients: A Retrospective Analysis,” Clinical Research in Cardiology
115, no. 9 (2025): 1–10, 10.1007/s00392-025-02694-5.

47.
Lu W., Wang S., Tang H., Yuan T., Zuo W., and Liu Y., “Neuropsychiatric Adverse Events Associated With Glucagon‐Like Peptide‐1 Receptor Agonists: A Pharmacovigilance Analysis of the FDA Adverse Event Reporting System Database,” European Psychiatry
68, no. 1 (2025): e20, 10.1192/J.EURPSY.2024.1803.

48.
Katranski J., Liang S., Morris D., Suppiah V., and Lim C. X., “Psychiatric Adverse Events Linked to Glucagon‐Like Peptide 1 Analogues: A Disproportionality Analysis in American, Canadian and Australian Adverse Event Databases,” International Journal of Clinical Pharmacy
47, no. 6 (2025): 1739–1747, 10.1007/s11096-025-01943-x.

49.
Nishida K., Chrétien B., Dolladille C., et al., “Psychiatric and Psychological Adverse Effects Associated With Dulaglutide, Semaglutide, and Liraglutide: A Vigibase Study,” Clinical Nutrition
51 (2025): 252–265, 10.1016/J.CLNU.2025.06.011.

50.
Stojkovska A., Rus Prelog P., and Kokalj Palandacic A., “Association of Semaglutide Use With Depressive Symptoms and Suicidal Behavior in a Patient With Type 2 Diabetes: A Case Report,” Journal of International Medical Research
53, no. 7 (2025): 03000605251349393, 10.1177/03000605251349393.

51.
Shapiro S. B., Yin H., Yu O. H. Y., Rej S., Suissa S., and Azoulay L., “Glucagon‐Like Peptide‐1 Receptor Agonists and Risk of Suicidality Among Patients With Type 2 Diabetes: Active Comparator, New User Cohort Study,” BMJ
388 (2025): e080679, 10.1136/BMJ-2024-080679.

52.
Ueda P., Söderling J., Wintzell V., et al., “GLP‐1 Receptor Agonist Use and Risk of Suicide Death,” JAMA Internal Medicine
184, no. 11 (2024): 1301–1312, 10.1001/JAMAINTERNMED.2024.4369.

53.
van der Heijden P. G. M., van Puijenbroek E. P., van Buuren S., and van der Hofstede J. W., “On the Assessment of Adverse Drug Reactions From Spontaneous Reporting Systems: The Influence of Under‐Reporting on Odds Ratios,” Statistics in Medicine
21, no. 14 (2002): 2027–2044, 10.1002/sim.1157.

54.
Maignen F., Hauben M., Hung E., Van Holle L., and Dogne J. M., “Assessing the Extent and Impact of the Masking Effect of Disproportionality Analyses on Two Spontaneous Reporting Systems Databases,” Pharmacoepidemiology and Drug Safety
23, no. 2 (2014): 195–207, 10.1002/PDS.3529.

55.
Hazell L. and Shakir S. A. W., “Under‐Reporting of Adverse Drug Reactions: A Systematic Review,” Drug Safety
29, no. 5 (2006): 385–396, 10.2165/00002018-200629050-00003.

Associated Data

Supplementary Materials

Data Availability Statement

The data that support the findings of this study are publicly available at https://open.fda.gov.