Glucagon-like peptide-1 receptor agonists and risk of depression: a systematic review and meta-analysis

Abstract

Background

Glucagon-like peptide-1 receptor agonists (GLP-1RAs) have proven effectiveness in stimulating weight loss, as well as enhancing metabolic parameters. Nevertheless, the information available on the possible effects on mood and the risk of depression is not consistent, and it is possible that further research should evaluate the psychiatric safety of GLP-1RA usage.

Objective

To conduct a systematic review and quantitatively assess whether the usage of GLP-1RA is associated with the increased risk of having depression-related adverse events and to learn more about the sources of heterogeneity.

Methods

In accordance with the PRISMA guidelines, we systematically searched PubMed, Embase, Cochrane Library, Web of Science, Scopus, and LILACS for randomized controlled trials and observational studies published between 2000 and 2025. Studies were eligible if they enrolled adults with overweight or obesity and reported outcomes related to depression or mood. Random-effects models were used for meta-analysis, with heterogeneity, sensitivity analysis, meta-regression, and publication bias assessments conducted.

Results

A total of 19 studies were included in this systematic review. Combined findings revealed that the risk of depression was linked to the utilization of GLP-1RA (OR = 1.49, 95% CI: 1.18–1.88). Between-study heterogeneity was very high (I² = 99.21%) although sensitivity analysis revealed that the findings were robust. The heterogeneity was not explained by a single factor, indicating a multifactor interaction, as found in meta-regression. None of the publications exhibited any considerable bias.

Conclusion

The current evidence suggests a possible association between use of GLP-1RA and an increased risk of depression, but overall risk seems to be dominated by population characteristics, drug class, and how mood outcomes are assessed. Given the metabolic benefits of GLP-1RAs, caution is advised in patients with a history of depression or psychiatric vulnerability and a heightened vigilance for these events is warranted. High-quality studies are warranted with psychiatric outcomes as primary endpoints in order to determine whether there is a causal relationship and to identify high-risk subpopulations.

Introduction

Depressive diseases are one of the most widespread and problematic mental illnesses in the world, and their spread is growing. They are closely related to critical functional impairment, metabolic anomalies and unfavorable cardiovascular incidents [1, 2].Massive studies point towards two-way relation between obesity and depression: the obese individuals is highly predisposed to the manifestation of depression whereas the depressed individuals is as well susceptible towards accumulation of weight, lifestyle, and metabolic syndrome [3, 4]. Amid the escalating obesity rates and the increasing numbers of metabolic disorders, the use of therapeutics that can build the body weight and metabolic status has become a priority of the medical practice and the overall health of the world population.

The agonists of glucagon-like peptide-1 receptor (GLP-1RAs) have been used in recent years to treat diabetes and obesity because they have remarkable effects of hypoglycaemia and weight-reduction ability [5, 6]. It has been established through clinical trials that GLP-1 receptor agonists (liraglutide and semaglutide) have proven to be consistently useful in lowering body weight, enhancing glycaemic control, and reducing cardiovascular risk [7–9].Nevertheless, since GLP-1 receptor agonists have the capability to work on several central nervous pathways, including the one controlling appetite, reward circuits, and the pancreatic-enteric-brain axis, the potential psychological impact of such agonists has increased with time [10–12]. Over the last few years, several regulatory authorities, such as the FDA and EMA, reported the incidence of mood alterations, depressive symptoms, and even suicidal thoughts with the use of GLP-1 receptor agonists during the pharmacovigilance surveillance process, which raises the issue of their psychiatric safety [13–15].

Though there has been some rudimentary research work indicating that the GLP-1 receptor agonists can indirectly provide mood-enhancing effects by regulating the dopamine reward system, lessening inflammatory responses, or improving metabolic status, adverse events that have been observed in real life studies and case reports can suggest that their action at the central nervous system is more complicated [11, 16, 17]. Furthermore, it is common to prescribe GLP-1 receptor agonists to obese people; obese people in turn fall into the high-risk category of depression sufferers, thus making measurement of the actual psychiatric impact on the drugs still more problematic [18, 19]. The literature used shows a great variation in terms of sample size, research design, research populations and manner of depression diagnosis and thus presents inconsistent evidence on the relation between GLP-1 receptor agonists and the risk of depression [20–22].

Currently, the majority of systematic reviews are devoted to the impact of GLP-1 receptor agonists on the effects on weight and metabolism causes, although there is still no systematic evidence of their psychiatric adverse effects [23].Specifically, since depression is a mental consequence with the low incidence rate and is affected by a variety of determinants, its evaluation demands higher level of statistical power and analysis of studies across different studies [24, 25]. Even though there are few studies that have tried to investigate the safety of GLP-1 receptor agonist (GLP-1RA) in patients with psychiatric illnesses or who have comorbid metabolic issues, a systematic review or a meta-analysis of studies that focused on the main research question of whether GLP-1RAs are a risk factor in the development of depression has not been conducted. Since the use of GLP-1RA is increasing in a wider range of obese and metabolic diseases patient populations, inadequate disclosure of possible psychiatric risks might severely affect clinical decision-making.

It is against this background that there is an immediate need to carry out a systematic review and quantitative synthesis of the existing evidence to explain the actual relationship between the use of GLP-1RA and the likelihood of developed depression. Therefore a systematic review and meta-analysis of the current published randomized control trials and observational studies were performed on this topic to assess whether GLP-1RAs expose the participants to the increased risk of the depressive events and the possible sources of the heterogeneity. The results of the research presented will be used as a contribution to the safer nature of GLP-1RA, provide valuable standards in clinical practice, and guide future fields of research.

Methods

Study protocol and registration

The design, conduct, and reporting of this systematic review and meta-analysis adhered to the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) guidelines.

Literature search

We conducted a comprehensive search across multiple international databases, including PubMed, Embase, the Cochrane Library, Web of Science, Scopus, and the Latin American and Caribbean Health Sciences Literature database (LILACS). In addition, we searched the U.S. National Institutes of Health clinical trial registry, including ClinicalTrials.gov. The search period was restricted to January 1, 2000, through May 6, 2025, as GLP-1 receptor agonists began entering clinical research in the early 2000s.

The search strategy combined text words (tw) and Medical Subject Headings (MeSH). Keywords included terms related to glucagon-like peptide-1 receptor agonists (e.g., “Glucagon-Like Peptide-1,” “GLP-1”), mental health outcomes (e.g., “Depression,” “Mental Disorder*,” “Psychiatric Disease*,” “Depressive Symptoms”), and weight-related conditions (e.g., “Obesity,” “Overweight,” “Weight Gain”). The final search strategy was developed with assistance from an experienced medical librarian. An example of the PubMed search strategy is as follows: ((“Glucagon-Like Peptide 1 Receptor Agonists“[Mesh] OR “GLP-1 receptor agonist*” OR “GLP1 agonist*” OR liraglutide OR semaglutide OR dulaglutide OR exenatide OR lixisenatide OR tirzepatide) AND (“Depression“[Mesh] OR “Depressive Disorder*” OR depression OR depressive symptoms OR “Patient Health Questionnaire” OR PHQ-9 OR “Mental Disorders“[Mesh] OR suicidal OR “suicidal ideation” OR “suicide attempt” OR “Suicide“[Mesh]) AND (“Cohort Studies“[Mesh] OR cohort OR observational OR “Randomized Controlled Trial“[Publication Type] OR randomized OR RCT)). To ensure the completeness of included studies, we also performed a manual search of the reference lists of all eligible English-language articles.

Study selection

The study selection process adhered to the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) guidelines. Two investigators independently screened the titles and abstracts, followed by a full-text review of studies deemed potentially eligible. Any discrepancies were resolved through discussion and, if necessary, by consulting a third investigator.

The inclusion criteria were the following: (1) The population was adults (age 18 years or older); (2) The study population needed to have reported depression or depression-related adverse events (depression new-onset, worsening of depressive symptoms, change of negative mood, events related to suicide, etc.); (3) The population of the study was to include overweight or obese individuals (Body Mass Index > 25 kg/m²); (4) The intervention involved any Glucagon-like peptide-1 receptor agonist (GLP-1 RA); (5) Study design was a randomized controlled trial (RCT) or an observational study; (6) The control group received either a placebo or an active comparator drug.

The exclusion criteria were: (1) Conference abstracts, case reports, case series, and other studies had inadequate data; (2) Such studies were not obtained and their final figure and the selection flow diagram were provided as an additional figure.

Outcomes

The main consequence was cases of adverse events that were related to depression. This included treatment-emerging depression, exacerbation of depressive symptoms, fluctuations in depressive mood, and other clinically pertinent depressive-related events expressly covered in the studies.Considering the possible expanded effect of the GLP-1 RAs in mood and the central nervous system, the secondary outcomes were anxiety, mood disorder, suicidal thoughts, and mood-related adverse events.To increase the comparability and exhaustiveness of our analysis and investigate possible mediating variables about psychiatric adverse events, we also released data about metabolic parameters including body weight, BMI, fasting blood glucose, and lipid profiles.

Data extraction

Data were independently extracted by two investigators using a predesigned, pretested standardized data extraction form, with any inconsistencies resolved through discussion and adjudication by a third investigator. Extracted variables encompassed four domains: (1) study characteristics, including first author, publication year, country, study design, and follow-up duration; (2) participant characteristics, comprising sample size, age, sex, baseline BMI, history of psychiatric disorders, and prior or concomitant medications; (3) intervention details, encompassing the type, dosage, and duration of GLP-1 RA use, as well as comparator group specifications; and (4) outcome data, including incidence rates of depression and other psychiatric events, changes in metabolic parameters (body weight, BMI, fasting blood glucose, and lipid profiles), and loss to follow-up information (proportion, timing, and reasons). This comprehensive extraction approach ensured thorough assessment of study quality and finding robustness.

Risk-of-bias assessment

Methodological quality appraisal was performed independently by two reviewers. For randomized controlled trials, the Cochrane RoB 2.0 tool was applied across seven domains: random-sequence generation, allocation concealment, blinding of participants and personnel, blinding of outcome assessment, incomplete outcome data, selective reporting, and other sources of bias. Observational studies were evaluated with the Newcastle–Ottawa Scale. When at least ten studies were available, funnel plots were generated to examine potential publication bias. Disagreements during the appraisal were resolved by consultation with a third reviewer.

Statistical analysis

A random-effects model was used to pool the overall effect of GLP-1 receptor agonist use on depression risk. For dichotomous outcomes (e.g., incident depression), risk ratios (RR) or odds ratios (OR) with 95% confidence intervals were calculated; for continuous outcomes (e.g., changes in body weight or metabolic parameters), mean differences (MD) and their standard deviations were reported. To ensure statistical comparability across studies, transformations followed the methods recommended in the Cochrane Handbook for Systematic Reviews of Interventions (e.g., converting medians and inter-quartile ranges to means and standard deviations).

Heterogeneity was assessed with Cochran’s Q test and the I² statistic; I² > 50% together with Q-test P < 0.10 indicated substantial heterogeneity. When high heterogeneity was detected, subgroup and sensitivity analyses were conducted to explore its sources. All analyses were performed in R software (version 4.3.3).

Result

Study selection and characteristics

Of 821 records identified, 297 were excluded after title/abstract screening, leaving 524 articles for full-text review. Ultimately, 19 studies were included in this systematic review, comprising 11 RCTs and 8 observational studies (Fig. 1).

Fig. 1.

PRISMA flow-chart of literature identification and selection

Table 1 Summary of studies evaluating GLP-1 receptor agonists and risk of depression.

Table 1.

Summary of studies on the risk of GLP-1-induced depression

Author_Year	Country	Study_Design	Sample_Size	Population	GLP1_RA	Control_Group	Follow_Up
Reaney et al., 2013	Europe (6 countries)	Prospective Observational Study	2,388	Type 2 diabetes patients initiating injectable therapy	Exenatide twice daily	Insulin	24 months
Kornelius et al., 2024	Taiwan	Retrospective Cohort Study	324,506(162,253 vs. 162,253)	Obese patients on GLP-1RA	Liraglutide, Semaglutide	Non-GLP1 treatment group	5 years
Shapiro 2025	UK	Active comparator, new user cohort study	270,110	Type 2 diabetes	Dulaglutide, exenatide, liraglutide, lixisenatide, semaglutide	DPP-4 inhibitors or SGLT-2 inhibitors	1.3 years
Chung et al., 2020	Multinational(Argentina, Austria, Belgium, Egypt, Norway, South Korea, Switzerland, Turkey, USA)	Randomized controlled trial	504 (253 patients were randomly assigned to take semaglutide orally, 185 entered expansion; switching part: 198 patients were re-assigned to take semaglutide or sitagliptin orally)	Type 2 Diabetes	Oral semaglutide	Sitagliptin	2 years
Chang et al. 2025	Taiwan	Retrospective Cohort Study	51,408 (we compared 25 704 new GLP-1 RA users to 25 704 SGLT2i users)	Overweight or obese adults with type 2 diabetes	Tirzepatide, semaglutide, or liraglutide	SGLT2 inhibitors (canagliflozin, dapagliflozin, or empagliflozin)	1 year
Bode et al. 2010	Mexico and USA(participants were recruited from Mexico (12 sites) and the United States (126 sites))	Randomized, double-blind, controlled trial	732	Patients with type 2 diabetes	Liraglutide	Glimepiride	52 weeks
Ji et al. 2021	China, Korea, Brazil, South Africa, Ukraine (multinational)	Randomized, double-blind, phase 3a trial	868	Patients with type 2 diabetes inadequately controlled on metformin	Semaglutide	Sitagliptin	30 weeks
Unger et al. 2022	Multinational (9 countries)	Randomized, open-label trial	1991	Patients with type 2 diabetes uncontrolled with metformin alone	Liraglutide	Oral antidiabetic drugs (SGLT2is, DPP-4is, SUs, etc.)	Up to 104 weeks
Wadden et al. 2025	Sweden and Denmark	Active-comparator new-user cohort study	298,553 (124,517 GLP-1 RA users + 174,036 SGLT2 inhibitor users)	Type 2 diabetes patients	liraglutide, semaglutide	SGLT2 inhibitors	2.5 years
Ueda or Wadden et al., 2024	US	Post hoc analysis of randomized, double-blind, placebo-controlled trials	3,681 (STEP 1–3: 3,377; STEP 5: 304)	overweight or obesity; STEP 2 type 2 diabetes	Semaglutide, Tirzepatide	Placebo	STEP 1–3: 68 weeks; STEP 5: 104 weeks
Husain et al. 2019	Multinational	Randomized, double-blind, placebo-controlled trial (PIONEER 6)	3,183 (1,591 oral semaglutide + 1,592 placebo)	Patients with type 2 diabetes at high cardiovascular risk	Oral Semaglutide	Placebo	Median 15.9 months (range 0.4–20.0 months)
Kelly et al., 2020	USA, Belgium, Mexico, etc.	RCT	251	Adolescents with obesity	Liraglutide		56 weeks + 26 weeks
Tronieri et al.2019	US	RCT	113	obesity	Liraglutide	IBT-alone	52 weeks
Gerstein et al., 2019	Multinational	RCT	9901	Type 2 diabetes with CV risk	Dulaglutide	Placebo	68 weeks
Bezin 2025	France	Case-time-control study	6596		Exenatide, dulaglutide, liraglutide, semaglutide	Time-controls matched on age, sex, psychiatric history, obesity, calendar time	180 days prior to event (2013–2021)
de Wit 2014	Netherlands	Randomized controlled trial	50	Type 2 diabetes patients with insulin-associated weight gain	Liraglutide 1.8 mg/day	Standard therapy (continuation of insulin)	26 weeks
Tagliapietra_2024	USA	Retrospective cohort study	139,608	Veterans with type 2 diabetes	GLP-1 receptor agonists	DPP-4 inhibitor	1 year
Best_2011	Multinational	Randomized controlled trial	491	Type 2 diabetes patients on metformin monotherapy	Exenatide once weekly	Sitagliptin or Pioglitazone	26 weeks

Following a systematic literature search, a number of records were identified from various databases. After deduplication and a stepwise screening process, a series of studies were included for the final analysis. This review ultimately incorporated 19 studies, comprising prospective observational studies, large-scale retrospective cohorts, active-comparator new-user cohort designs, and multiple randomized controlled trials (RCTs). The included research encompasses both large-sample, long-term follow-up real-world databases and rigorously designed double-blind controlled trials, ensuring a diversity of study types and a rich hierarchy of evidence.

The final set of studies includes multiple randomized, double-blind, placebo- or active-controlled clinical trials (e.g., Bode 2010, Ji 2021, Husain 2019, Gerstein 2019, Kelly 2020, etc.), as well as several large-scale retrospective or prospective cohort studies from regions including Europe, Taiwan, the United States, and Nordic countries (e.g., Kornelius 2024, Chang 2025, Tagliapietra 2024, Ueda/Wadden 2024, and multinational database analyses). Additionally, a case-time-control study carried out in France (Bezin 2025) was used to give supportive evidence on the relationship between short-term event triggers and windows of drug exposure in a real-world aspect.

The geographical distribution of the studies is extensive, covering multiple continents, including Europe, North America, Asia (e.g., Taiwan, China, South Korea), Latin America (e.g., Mexico, Brazil), and South Africa. This wide coverage enhances the generalizability of the findings across diverse populations. The overall sample size is substantial, with several individual studies encompassing thousands of participants. For instance, the Taiwanese database in Kornelius 2024 included 324,506 subjects, the Nordic (Sweden and Denmark) active-comparator new-user study enrolled 298,553 patients, and the UK study by Shapiro 2025 included 270,110 patients with type 2 diabetes.

Regarding participant characteristics, the majority of studies enrolled overweight or obese adults with type 2 diabetes, or those receiving glucose-lowering treatments such as metformin, SGLT2 inhibitors, DPP-4 inhibitors, or insulin. A limited number of studies extended their inclusion to non-diabetic obese populations (e.g., the adolescent obesity trial by Kelly 2020 and the obesity intervention trial by Tronieri 2019). Some studies specifically included patients with high cardiovascular risk (e.g., PIONEER-6; Husain 2019), while others focused on specific subgroups, such as individuals experiencing weight gain associated with insulin therapy (e.g., de Wit 2014).

A variety of GLP-1 receptor agonists were evaluated across the studies, including exenatide (daily or weekly formulations), liraglutide (1.8 mg or 3.0 mg), dulaglutide, lixisenatide, oral and subcutaneous semaglutide, as well as the novel dual receptor agonist tirzepatide. The comparator groups differed among studies, with some employing placebo controls (e.g., PIONEER-6, Gerstein 2019), others using active comparators such as sitagliptin, glimepiride, DPP-4 inhibitors, or SGLT2 inhibitors, and some utilizing standard care or intensive lifestyle interventions as the control.

Follow-up durations varied widely in the studies as it took several weeks up to several years. The follow-up period of RCTs was 26 to 68 weeks, which was significantly shorter than the period of real-world cohort studies. For example, the follow-up up to 5 years in Kornelius 2024, and the study by Reaney 2013 lasted 24 months.

The large Nordic cohort had a median follow-up of approximately 2.5 years, and the Chang 2025 study followed patients for 1 year. This range in follow-up duration provides a robust data foundation for investigating the temporal relationship between GLP-1 RA use and depression risk.

Outcome analysis: risk of depression

Pooled effect estimate

Across the 19 included studies, all authors reported an effect estimate for incident or worsening depression-related adverse events associated with GLP-1 receptor agonist (GLP-1 RA) exposure. A random-effects meta-analysis yielded a pooled odds ratio (OR) of 1.49 (95% CI 1.18–1.88), indicating a 49% relative increase in the risk of depression-related outcomes among GLP-1 RA users compared with controls (Fig. 2).

Fig. 2.

Forest plot showing the significantly increased pooled risk of depression associated with GLP-1 receptor agonists

Heterogeneity was extreme (Cochran’s Q = 886.46, df = 7, p < 0.001; I² = 99.2%, τ² = 0.11), implying that the between-study variance in true effect sizes far exceeded what could be attributed to random error. Nevertheless, the direction of effect was consistent: 15 of 18 individual point estimates lay on the “increased-risk” side of the null, and the pooled result remained statistically significant under both the DerSimonian–Laird and restricted maximum-likelihood estimators.

Forest plot

Figure 2 presents the individual effect sizes and their corresponding 95% confidence intervals for each included study, visually illustrating their consistency with the pooled effect estimate. Overall, most studies had point estimates located to the right of the null line (OR = 1), indicating a higher incidence of depressive events in the GLP-1 RA group compared with the control group. Some studies, and especially those that had larger effect sizes, had significantly higher risk, others had estimates of effects that were similar to 1; however, overall direction of the effect was the same across studies.

On matters of distribution of the confidence intervals, the majority has not crossed the null line implying some level of statistical strength in the results. Nevertheless, the variability between individual estimates across the studies was quite large: some had mentioned a significantly high risk, whereas others had mentioned a slightly greater alteration. Such heterogeneity is consistent with the large total levels of heterogeneity (I² = 99.21%), indicating large differences among studies in terms of design, population of the study, types of GLP-1 RAs, or duration of treatment.

Assessment of heterogeneity

There was a significant heterogeneity in the effect sizes used in this meta-analysis. Cochran’s Q test revealed that the variation between the studies is statistically significant (Q = 886.46, df = 7, p < 0.001) indicating that the style of variations are not the same. The I² statistic was 99.21%, showed that about 99% of total variability in effect estimates must be due to real differences in studies as opposed to the sampling error, which is very high amount of heterogeneity.Besides, the approximate τ2 standard of 0.1074 in the random-effects model is a reflection of the extent of this true variance in the effect sizes of the studies. The great heterogeneity could indicate significant differences among the included studies, which could be due to different population aspects, a study design, the type of the GLP-1 RAs that were used, or intervention durations.

Exploration of heterogeneity sources: meta-regression analysis

The meta-regression analysis was carried out on pre-specified variables in order to investigate possible causes of heterogeneity. Potential covariates that were taken into account as they might have affected the effect size were: study design (RCT vs. observational study), sample size, specific GLP-1 RA agent (liraglutide, semaglutide, dulaglutide, etc.), follow-up period, geographical region where the study was conducted (Asia, Europe/North America, International/Multi-center), and type of control group (placebo vs. active comparator).

The meta-regression failed to show any individual covariate that had statistically significant moderating effect on the pooled effect size. All the coefficient regression of the covariates that were included lacked statistical significance (p-value over 0.05). This means that the high level of heterogeneity was not possible to be attributed to any particular characteristic of the studies under investigation. This observation indicates that the heterogeneity is most probably due to the complicated interrelation of various factors. These can be the differences in the characteristics of the populations, the risks of depression at its baselines, the differences of drug dosages and treatment time, and the differences between the definition and ways of reporting the depression-related events in the different studies.

Although the meta-regression did not significantly reduce the overall heterogeneity, it confirms that the high heterogeneity is driven by a combination of multiple factors rather than a single dominant study feature.

Assessment of publication bias

Funnel plot

The funnel plot used to assess potential publication bias is presented in Fig. 3. The plot shows a degree of scattering of points, even in the region of higher precision (towards the top). The overall distribution appears slightly asymmetrical. This asymmetry may suggest the potential presence of publication bias or small-study effects. However, it is also important to consider that the high level of inherent heterogeneity among the studies itself can contribute to asymmetry in the funnel plot.

Fig. 3.

Funnel plot of publication bias analysis

Egger’s test

To quantitatively assess the presence of publication bias, we further performed Egger’s regression test for funnel plot asymmetry. The results showed that the intercept estimate was 0.6031 (standard error = 9.518), with a t-value of 0.063 and a corresponding P value of 0.952, indicating that the intercept was not statistically significant. In other words, no evidence of small-study effects or publication bias was observed.

Moreover, standardized effect sizes were regressed on their precision, where slope (precision_egger) = 0.4229 (standard error = 0.267), and t = 1.582 and P = 0.175 were not statistically significant. This once again highlights the idea that the curves of asymmetry of the funnel plot are not likely attributable to systematic publication bias and tend to be more indicative of actual variation in the studies.

The model fit was quite poor (R² = 0.334; adjusted R²= 0.200), which is expected of Egger regression in a data with much heterogeneity. In general, the Egger test did not offer any statistically significant evidence of the publication bias, which is consistent with the visual image analysis of the funnel plot.

Sensitivity analysis

To check the strength of the pooled effect estimate, a leave-one-out sensitivity analysis was performed by successively dropping out each study. These findings indicated that the pooled log risk ratio (log RR) was similar irrespective of the study dropped, and the effect was all in a similar direction which suggested that there is a relationship between the use of GLP-1 RA and the risk of developing depression.

Across the leave-one-out analyses, the pooled log RR values ranged from 0.287 to 0.516. The corresponding 95% confidence interval lower limits ranged from 0.017 to 0.238, and the upper limits ranged from 0.557 to 0.794. For example, when excluding Reaney et al. (2013), the pooled log RR was 0.516 (95% CI: 0.238–0.794); when excluding Kornelius et al. (2024), the pooled estimate slightly decreased to 0.368 (95% CI: 0.076–0.659). Excluding Shapiro (2025) yielded a pooled log RR of 0.287 (95% CI: 0.018–0.557); excluding Chang et al. (2025) produced a pooled estimate of 0.473 (95% CI: 0.195–0.750); and excluding Ueda/Wadden et al. (2024) resulted in a pooled log RR of 0.480 (95% CI: 0.202–0.758). All variations were modest in magnitude.

Figure 4 shows that the blue points, which are the estimates of the leave-one-out pooled log RR, all three of them were in close proximity to the overall effect estimate (red marker). Each of the confidence intervals was within the range of the overall pooled confidence interval and there was no reversal of direction of effect or complete attenuation of the association. These results confirm the strong stability of the pooled effect size, and show that there was no study, which had a disproportionate effect on the overall results, in support of the strength of the association between GLP-1 RA use and risk of depression.

Fig. 4.

Leave-one-out sensitivity analysis assessing the robustness of the pooled effect estimate

Quality of evidence assessment

The evidence quality of the outcomes of interest was assessed systematically based on the GRADE framework that takes into consideration the risk of bias, inconsistency, indirectness, imprecision, and magnitude of effect.

In case of depression-related outcomes, the general quality of evidence was moderate. Various observational studies were categorized down as having a risk of bias, and this was mainly because of the possibility of residual confounding and selection bias. Although there was significant statistical heterogeneity (inconsistency), some of the studies in which a significant effect size (e.g. HR = 2.9) was seen were considered to be upgraded. As a result, there was a moderate level of the evidence of the outcomes of depression, which illustrates a steady direction of the mutagenic effect with no significant difference in its strength.

On the contrary, suicidal ideation and outcomes associated with suicide were rated as low to very low. This grading was mainly fuelled by serious imprecision since the event rates are low. Also, evidence was typically based on post hoc analysis or findings not determined a priori in study designs, and thus risk of bias was downgraded. Additionally, for the outcome of completed suicide, the generalizability of findings to the broader population of interest was limited (indirectness), as the evidence largely originated from studies involving overweight or obese patients with severe metabolic comorbidities.

A detailed summary of the quality assessment for each outcome is provided in Table 2. The evaluation of publication bias is presented in Table 3.

Table 2.

Research quality assessment form

Study			In the melancholy ending		In the suicide-related outcome
	Type	Bias risk (-1 point) Observational studies (retrospective cohort) carry potential confounding bias and selection bias.	Inconsistency (-1 point) Kornelius et al. demonstrated a substantial increase in risk (HR = 2.9), whereas Chang et al. showed a modest increase (HR = 1.1). Although effect sizes varied considerably, their directionality was consistent. This suggests potential high heterogeneity.	Large effect size upgrade (+ 1 point) Kornelius et al. demonstrated HR = 2.9 (> 2), indicating a large effect size.	Inconsistency (1 point) Kornelius et al. demonstrated increased risk (HR = 2.05), whereas Shapiro et al. showed no effect (OR = 1.05), indicating inconsistent results.	Large effect size upgrade (+ 1 point) Kornelius et al. demonstrated HR = 2.05 (> 2), indicating a large effect size.	Risk of bias (-1 point) Post-hoc analysis of RCTs may not have been pre-specified for outcomes, presenting a risk of reporting bias.	Intermittent (-1 point) The cohort experiencing suicide-related mortality comprised overweight/obese individuals with type 2 diabetes; this outcome is rare, resulting in limited external validity.	Imprecision (-1 point) The confidence interval is wide and encompasses 1 (HR = 1.25, CI [0.83,1.88]), indicating imprecision.	Bias risk (-1 point) Although case-control studies mitigate some sources of bias, they remain observational designs and carry residual confounding risk.	Total score (9 points)
Reaney et al. (2013)	Prospective Observational Study										9
Kornelius et al. (2024)	Retrospective Cohort Study	-1	-1	1	-1	1					8
Shapiro (2025)	Active comparator, new user cohort study	-1			-1						7
Chung et al. (2020)	Randomized controlled trial	-	-	-	-	-	-	-	-	-	NA
Chang et al. (2025)	Retrospective Cohort Study	-1	-1								7
Bode et al. (2010)	Randomized, double-blind, controlled trial	-	-	-	-	-	-	-	-	-	NA
Ji et al. (2021)	Randomized, double-blind, phase 3a trial	-	-	-	-	-	-	-	-	-	NA
Unger et al. (2022)	Randomized, open-label trial	-	-	-	-	-	-	-	-	-	NA
	Active-comparator new-user cohort study	-	-	-	-	-	-	-	-	-	NA
Ueda or Wadden et al. (2024)	Post hoc analysis of randomized, double-blind, placebo-controlled trials						-1	-1	-1		6
Husain et al. (2019)	Randomized, double-blind, placebo-controlled trial (PIONEER 6)	-	-	-	-	-	-	-	-	-	NA
Kelly et al. (2020)	Randomized controlled trial	-	-	-	-	-	-	-	-	-	NA
Tronieri et al. (2019)	Randomized controlled trial	-	-	-	-	-	-	-	-	-	NA
Gerstein et al. (2019)	Randomized controlled trial	-	-	-	-	-	-	-	-	-	NA
Bezin (2025)	Case-time-control study									-1	8
de Wit (2014)	Randomized controlled trial	-	-	-	-	-	-	-	-	-	NA
Kornelius (2024)	Retrospective cohort study										9
Tagliapietra (2024)	Retrospective cohort study										9
Best (2011)	Randomized controlled trial	-	-	-	-	-	-	-	-	-	NA

Evaluation criteria

1. Risk of bias: Observational studies (retrospective cohorts) have potential confounding bias and selection bias, reduced by 1 level;

2. Inconsistency: In the depressive outcome, Kornelius et al. showed a significant increase in risk (HR = 2.9), and Chang et al. showed a slight increase in risk (HR = 1.1). The size of the effect varied greatly, but the direction was consistent. Due to the possible high heterogeneity, it is downgraded by one level

3. Large effect size upgrade: Kornelius et al. showed HR = 2.9 (> 2), with a large effect size upgrade of 1 level

4. Inconsistency: In the outcome of suicidal ideation, Kornelius et al. showed an increased risk (HR = 2.05), while Shapiro et al. showed no effect (OR = 1.05). The results were inconsistent, with a reduction of 1 level

5. Large effect size upgrade: Kornelius et al. showed HR = 2.05 (> 2), with a large effect size upgrade of 1 level

6. Risk of bias: Post hoc analysis of RCTS may not be the pre-specified outcome, and there is a risk of reporting bias. Downgrade by 1 level

7. Indirect: The population with suicidal death outcomes was overweight/obese with type 2 diabetes, a rare outcome, and limited external validity, downgraded by one grade

8. Imprecision: The confidence interval is wide and includes 1 (HR = 1.25, CI [0.83,1.88]), imprecise, downgraded by 1 level;

9. Risk of bias: Although the case-time control study managed to control for partial bias, it was still an observational design and had a risk of residual confounding, which was downgraded by one level

Table 3.

Risk of bias assessment for randomized controlled trials (Cochrane RoB 2 tool)

	Random sequence generation	Allocation concealment	Blinding of participants and personnel	Blinding of outcome assessment	Incomplete outcome data	Selective reporting	Other bias
Bode 2010	Low	Unclear	Unclear	Low	Unclear	High	Unclear
Ji 2021	Low	Unclear	Low	Low	High	Low	Low
Husain 2019	High	Low	Low	Unclear	Low	Low	High
Kelly 2020	Low	Unclear	Unclear	Unclear	Unclear	Unclear	Low
Tronieri 2019	Low	Unclear	Low	High	Unclear	Unclear	Low
Gerstein 2019	High	Low	Unclear	Low	Low	High	Unclear
de Wit 2014	Unclear	Low	High	Unclear	High	Low	Low
Best 2011	Unclear	Unclear	Low	Low	Unclear	High	Unclear

Discussion

This systematic review and meta - analysis integrated the results of 19 randomized controlled trials and observational studies to systematically evaluate the association between the use of glucagon-like peptide-1 receptor agonists (GLP-1RA) and the risk of depression. The comprehensive analysis of the studies revealed that, compared with the control group, the use of GLP − 1RA was significantly associated with a higher risk of depression onset. The direction of the pooled effect size was consistent across different studies, although there were variations in its magnitude. Meanwhile, there was extremely high heterogeneity among the studies. However, sensitivity analyses indicated that the overall effect was robust, suggesting that this risk signal was not driven by any single study but might reflect the true variable effects of GLP − 1RA on mental health across different populations and study designs.

This finding is consistent with the regulatory signals that have gradually drawn attention in recent years. In the databases of adverse drug events in multiple countries, the number of reports of mood changes, suicidal ideation and depression related to GLP-1RA has been on the rise, prompting regulatory authorities to request further monitoring [13, 15, 26, 27]. In previous clinical trials, since depression was not the primary prespecified outcome, most of them did not systematically track mood changes, which may lead to an underestimation of the real risk of adverse mental events during the clinical trial stage [28, 29]. Our results suggest that with the widespread use of GLP-1RA and the increase in long-term follow-up, its potential emotional impact has become more clearly manifested.

The association mechanism between GLP-1RA and the risk of depression has not been fully elucidated yet, but neurobiological studies around the GLP-1 signaling pathway have provided important clues. The GLP-1 receptor is widely distributed in key emotional regulation brain regions such as the limbic system, prefrontal cortex and hypothalamus [30, 31]. Its effect on the central nervous system is not limited to appetite control, but also involves reward processing, stress response and motivation regulation [10, 11, 32]. Some studies have revealed that the activation of GLP-1 may reduce the response of the midbrain dopamine system to pleasurable stimuli and decrease reward-driven behaviors. This mechanism can explain the potential benefits of GLP-1RA in treating anorexia and behavioral addiction [33]. However, the same neural regulation may manifest in some individuals as symptoms such as anhedonia, reduced motivation and low mood, which are precisely the important core manifestations of depressive disorders [34, 35]. Therefore, GLP-1RA may induce or exacerbate depressive phenotypes in some susceptible populations, while in others it may exert antidepressant effects by improving metabolism and inflammation. This bidirectional possibility has shown signs in basic research, but has not yet been systematically verified in human studies.

Beyond the neurobiological mechanisms of the drugs themselves, the pharmacokinetic characteristics of GLP-1 receptor agonists and the rapid weight loss observed during treatment may also exert effects on mood. Evidence suggests that rapid weight reduction can trigger neuroendocrine alterations, such as decreased leptin and elevated cortisol levels, which are associated with an increased risk of depression [36–38]. Concurrently, GLP-1 receptor agonists alter the diet-emotion regulation patterns in some individuals by enhancing satiety and reducing cravings for high-calorie foods [39–41]. For individuals who previously relied on eating to regulate their emotions, this shift may lead to difficulties in emotional adaptation. Previous research indicates that ‘emotional eating’ is closely associated with obesity and depressive symptoms, and may influence long-term weight changes [42, 43]. Furthermore, the study also reported that antidepressants may diminish the weight-loss effects of GLP-1 receptor agonists, potentially leading to a discrepancy between users’ expectations of treatment efficacy and their actual experience. This disparity could exacerbate negative emotions, thereby further increasing the risk of depression [44].

It is noteworthy that the high heterogeneity observed in this study is not coincidental. Therefore, the pooled estimates should be interpreted with caution. The substantial heterogeneity suggests that the observed association may be influenced by differences in study populations, study designs, and outcome definitions rather than reflecting a uniform causal effect across all settings. Significant variations exist across studies in terms of subject characteristics, definitions of depression, GLP-1 receptor agonist (GLP-1RA) types, follow-up durations, and psychiatric histories, all of which may be associated with depression risk. For instance, different GLP-1RAs exhibit varying degrees of central penetration. Semaglutide is considered to exert stronger central effects, whereas liraglutide and dulaglutide possess distinct mechanisms of action, potentially leading to non-uniform emotional effects [41, 45]. Recent systematic reviews and meta-analyses on GLP-1RA and mental health outcomes have also shown that the subjects included in the trials, such as obese individuals, patients with type 2 diabetes, and subgroups with mental disorders, have significantly increased baseline depression and suicide risks themselves [29, 46, 47]. Moreover, most randomised controlled trials did not include depression as a primary pre-specified outcome, seldom conducted systematic baseline mental health screening, and frequently lacked standardised depression assessment tools. This resulted in considerable variation in the reporting and interpretation of psychiatric adverse events [23, 48]. All these differences collectively constitute significant heterogeneity statistically and also suggest that the psychological effects of GLP-1RA in the real world are highly dependent on individual metabolic status, mental comorbidities and specific drug properties.

Despite the aforementioned controversies and complexities, the results of this study should not be interpreted as indicating that GLP − 1 receptor agonists (GLP − 1RAs) universally cause depression. Instead, they should be regarded as a potential risk signal that warrants further attention. The benefits of GLP − 1RAs in improving body weight, metabolic health, and cardiovascular risk have been well - supported by a large number of high - quality randomized controlled trials and large - scale outcome studies [7, 9, 49, 50]. Based on this evidence, multidisciplinary guidelines have listed GLP − 1RAs as first - line or preferred treatment options for patients with type 2 diabetes and/or obesity with high cardiovascular risk [51, 52]. Therefore, in the overall benefit - risk assessment, the risk of depression associated with GLP − 1RAs is not sufficient to alter existing treatment recommendations [29, 50].

Limitations and future prospects

This study has several limitations that should be considered when interpreting the findings. First, substantial heterogeneity was observed across the included studies, likely reflecting differences in study design, participant characteristics, GLP-1 receptor agonist types, follow-up duration, and methods used to assess depression outcomes. Such variability may influence the magnitude of the pooled estimates and requires cautious interpretation of the results. Second, our meta-analysis included both randomized controlled trials and observational studies. While this allowed for a broader synthesis of available evidence, observational studies are more susceptible to residual confounding and selection bias, which may influence the observed associations. Third, the definition and assessment of depression varied across studies, with some reporting clinically diagnosed depression or depression-related adverse events, while others used self-reported symptom scales. This inconsistency may have contributed to the observed heterogeneity. Several domains in the risk-of-bias assessment were rated as unclear due to incomplete methodological reporting. In addition, pharmacovigilance studies have reported signals of neuropsychiatric adverse events associated with GLP-1 receptor agonists, particularly in vulnerable populations such as individuals with prior psychiatric disorders or rapid weight loss [17, 53–55]. These factors should be considered when interpreting the findings.

Further research needs to be deepened at multiple levels. On one hand, future clinical trials should incorporate mood outcomes as pre - specified endpoints and employ standardized measurement tools to avoid the low sensitivity resulting from reliance on self - reported adverse events. On the other hand, it is essential to explore potential differences among drugs by comparing the central action characteristics of different GLP − 1 receptor agonist (GLP − 1RA) formulations. In addition, studies on genetic susceptibility, neuroimaging indicators, and the metabolic - mood interaction mechanisms are expected to explain why some individuals are more sensitive to the mood effects of GLP − 1RAs. These studies will lay the foundation for precisely identifying high - risk subgroups and formulating safer treatment strategies. Additionally, while our study focused on depression outcomes, we acknowledge that the relationship between GLP-1RAs and suicidality appears more complex and inconsistent in existing literature. The discrepancy between our consistent depression findings and mixed suicidality reports underscores the need for future studies to employ standardized, prospective assessments of both outcomes to clarify their temporal and causal relationships.

Conclusion

In conclusion, the association between GLP − 1RAs and an increased risk of depression revealed in this study provides crucial evidence for research on the mental safety of this class of drugs and raises new concerns for clinical practice in the context of their increasingly widespread use. As the role of GLP − 1RAs in the prevention and treatment of obesity, diabetes, and even cardiorenal diseases continues to expand, clarifying their potential impact on mood and mental health holds significant public health implications. More rigorous and systematic future research is necessary to further elucidate the causality, mechanistic basis, and population differences of this association, ultimately achieving an optimal balance between safety and benefit.

Author contributions

Z.F.: conceptualization; Z.B.: data curation and formal analysis; Z.F.: funding acquisition; J.L.: methodology; Z.F.: project administration; B.Z., Z.J., and J.L.: validation; Z.B.: writing - original draft; Z.B. and Z.F.: writing - review and editing. All authors have given their final approval for the manuscript.

Funding

This study was supported by the National Natural Science Foundation of China (Grant No. 82474431), the Anhui Province Clinical Medical Research Transformation Special Project (Grant No. 2024112201), and the Scientific Research Project of Anhui Provincial Education Department (Grant No. 2025AHGXZK50099).

Data availability

All data generated or analysed during this study are included in this published article (and its supplementary information files).

Declarations

Ethics approval and consent to participate

Human Ethics and Consent to Participate declarations: not applicable. As this study is a meta-analysis based on previously published literature, no new human or animal data were collected. Therefore, this study was exempt from Approval Committee or Internal Review Board (IRB) approval, and patient consent to participate was not required. However, all original studies included in this meta-analysis were approved by their respective institutional review boards and were conducted in accordance with the Declaration of Helsinki.

Consent for publication

Not applicable.

Competing interests

The authors declare no competing interests.