Glucagon-like peptide-1 receptor agonists and risk of suicidality among patients with type 2 diabetes: active comparator, new user cohort study
BMJ 2025; 388 doi: https://doi.org/10.1136/bmj-2024-080679 (Published 26 February 2025) Cite this as: BMJ 2025;388:e080679Linked Editorial
GLP-1 receptor agonists and suicidality
- Samantha B Shapiro
, doctoral student1 2, - Hui Yin, statistician1,
- Oriana Hoi Yun Yu, endocrinologist and associate professor2 3 4,
- Soham Rej, psychiatrist and associate professor5 6,
- Samy Suissa, professor1 2 7,
- Laurent Azoulay, professor1 2 8
- 1Centre for Clinical Epidemiology, Lady Davis Institute, Jewish General Hospital, Montreal, QC, Canada
- 2Department of Epidemiology, Biostatistics, and Occupational Health, McGill University, Montreal, QC, Canada
- 3Division of Endocrinology, Jewish General Hospital, Montreal, QC, Canada
- 4Division of Endocrinology & Metabolism, McGill University, Montreal, QC, Canada
- 5Department of Psychiatry, Jewish General Hospital, Montreal, QC, Canada
- 6Department of Psychiatry, McGill University, Montreal, QC, Canada
- 7Department of Medicine, McGill University, Montreal, QC, Canada
- 8Gerald Bronfman Department of Oncology, McGill University, Montreal, QC, Canada
- Correspondence: L Azoulay laurent.azoulay{at}mcgill.ca (or @profLAzoulay on X)
- Accepted 24 January 2025
Abstract
Objective To determine whether the use of glucagon-like peptide-1 (GLP-1) receptor agonists is associated with an increased risk of suicidal ideation, self-harm, and suicide among patients with type 2 diabetes compared with the use of dipeptidyl peptidase-4 (DPP-4) inhibitors or sodium-glucose cotransporter-2 (SGLT-2) inhibitors.
Design Active comparator, new user cohort study.
Setting Primary care practices contributing data to the UK Clinical Practice Research Datalink linked to the Hospital Episodes Statistics Admitted Patient Care and Office for National Statistics Death Registration databases.
Participants Patients with type 2 diabetes.
Exposures Two cohorts were assembled, with the first composed of patients who started and continued on GLP-1 receptor agonists or DPP-4 inhibitors between 1 January 2007 and 31 December 2020 and the second composed of patients who started and continued on GLP-1 receptor agonists or SGLT-2 inhibitors between 1 January 1 2013 and 31 December 2020. Both cohorts were followed until 29 March 2021.
Main outcome measures The primary outcome was suicidality, defined as a composite of suicidal ideation, self-harm, and suicide. Secondary outcomes were each of these events considered separately. Propensity score fine stratification weighted Cox proportional hazards models were fitted to estimate hazard ratios and 95% confidence intervals (CIs) to estimate the average treatment effect among the treated patients.
Results The first cohort included 36 082 GLP-1 receptor agonist users (median follow-up 1.3 years) and 234 028 DPP-4 inhibitor users (median follow-up 1.7 years). In crude analyses, GLP-1 receptor agonist use was associated with an increased incidence of suicidality compared with DPP-4 inhibitors (crude incidence rates 3.9 v 1.8 per 1000 person years, respectively; hazard ratio 2.08, 95% CI 1.83 to 2.36). This estimate decreased to a null value after confounding factors were accounted for (hazard ratio 1.02, 95% CI 0.85 to 1.23). The second cohort included 32 336 GLP-1 receptor agonist users (median follow-up 1.2 years) and 96 212 SGLT-2 inhibitor users (median follow-up 1.2 years). Similarly, GLP-1 receptor agonist use was associated with an increased risk of suicidality compared with SGLT-2 inhibitors in crude analyses (crude incidence rates 4.3 v 2.7 per 1000 person years; hazard ratio 1.60, 95% CI 1.37 to 1.87) but not after confounding factors were accounted for (0.91, 0.73 to 1.12). Similar findings were observed when suicidal ideation, self-harm, and suicide were analysed separately in both cohorts.
Conclusions In this large cohort study, the use of GLP-1 receptor agonists was not associated with an increased risk of suicidality compared with the use of DPP-4 inhibitors or SGLT-2 inhibitors in patients with type 2 diabetes.
Introduction
The glucagon-like peptide-1 (GLP-1) receptor agonist drug class is widely prescribed to manage type 2 diabetes.1 In large trials of cardiovascular outcomes, GLP-1 receptor agonists were highly effective in achieving glycaemic control, showing beneficial cardiorenal effects and reducing all cause mortality.2 Despite these benefits, recent reports linking GLP-1 receptor agonists to suicidal ideation and self-harm have raised significant concerns.3
The initial signal regarding this possible safety problem emerged from the Icelandic Medicines Agency in July 2023, prompting regulatory bodies worldwide, including the European Medicines Agency and the US Food and Drug Administration, to launch independent investigations into the potential risk of suicidality among GLP-1 receptor agonist users.3 Despite these efforts, regulatory agencies have yet to reach definitive conclusions owing to limitations in the available evidence.45 Clinical trials of GLP-1 receptor agonists have not reported any imbalances in suicidal ideation or self-harm,67 whereas several observational studies have generated varied findings on of the effect of GLP-1 receptor agonists on the risk of suicidal ideation, self-harm, and suicide, which may be a result of methodological limitations in these studies.8910111213 From a biological standpoint, GLP-1 receptor agonists may lead to suicidality through different mechanisms, including hyperactivity of the hypothalamic-pituitary-adrenal axis,14 sudden weight loss leading to elevated suicide risk,1516 and under-expression of brain derived neurotrophic factor linking low body weight and increased anxiety, potentially contributing to higher suicidality.1718 However, the exact manner in which GLP-1 receptor agonists could increase the risk of suicidal ideation and self-harm remains unclear.
Owing to the increasing popularity of GLP-1 receptor agonists and the inconsistency of observational studies on this potential association, we did a large, observational study to estimate the average treatment effect of continuous GLP-1 receptor agonist use compared with continuous dipeptidyl peptidase-4 (DPP-4) inhibitor or sodium-glucose cotransporter-2 (SGLT-2) inhibitor use on suicidal ideation, self-harm, and suicide among patients with type 2 diabetes.
Methods
Data sources
We used data from the UK Clinical Practice Research Datalink (CPRD) GOLD and Aurum, which was linked to the Hospital Episode Statistics Admitted Patient Care (HES APC) and Office for National Statistics (ONS) Death Registration databases. The CPRD is a large primary care database encompassing 60 million patients from more than 2000 general practices, which accurately reflects the demographics of the UK population in terms of age, sex, and ethnicity.1920 Diagnoses in the CPRD are recorded using the Read and SNOMED-CT classifications. The CPRD contains data on prescriptions written by general practitioners.21 The HES APC database archives hospital admission records of patients treated within English National Health Service hospitals.2223 The ONS Death Registration database is a vital statistics database used as the gold standard to identify patients who died from suicide and other causes during follow-up.222425 Diagnoses in the HES APC and ONS are recorded using the ICD-10 (international classification of diseases, 10th revision) classification.
Study population
We implemented two active comparator, new user cohorts.26 The first cohort was composed of patients who started and continued on either a GLP-1 receptor agonist (dulaglutide, exenatide, liraglutide (except the 3 mg/0.5 mL formulation indicated for weight loss), lixisenatide, semaglutide) or a DPP-4 inhibitor (alogliptin, linagliptin, saxagliptin, sitagliptin, vildagliptin) for the first time between 1 January 2007 (the year these medications entered the UK market) and 31 December 2020. We selected DPP-4 inhibitors as the comparator because they are not associated with the outcome and are used at a similar disease stage to GLP-1 receptor agonists16827; furthermore, DPP-4 inhibitors and GLP-1 receptor agonists entered the UK market in the same year. As both the GLP-1 receptor agonist and DPP-4 inhibitor drug classes exert their effects through the incretin system,28 we selected SGLT-2 inhibitors as a secondary comparator drug. This cohort was composed of patients who started and continued on either a GLP-1 receptor agonist or an SGLT-2 inhibitor (canagliflozin, dapagliflozin, empagliflozin, ertugliflozin) for the first time between 1 January 2013 (the year that SGLT-2 inhibitors entered the UK market) and 31 December 2020. We chose SGLT-2 inhibitors as the secondary comparator because they are also used at a disease stage similar to GLP-1 receptor agonists and have not been associated with the outcome.16
For both cohorts, we defined cohort entry as the date of the first prescription for either a GLP-1 receptor agonist or the comparator drug during the study period. At cohort entry, we excluded patients who were less than 18 years of age, had less than one year of medical history in the database, had no diagnosis of type 2 diabetes, had contraindications such as end stage renal disease or multiple endocrine neoplasia, or had concurrent prescriptions for both a GLP-1 receptor agonist and the comparator drug.
Follow-up
For both cohorts, we followed patients by using an on-treatment exposure definition, whereby we considered patients to be continuously exposed to the study drugs if consecutive prescriptions overlapped. We applied a 90 day grace period to bridge consecutive prescriptions that did not overlap. Thus, we followed patients from cohort entry until the outcome (detailed below), switching or discontinuation of study drug (defined as the lack of a prescription by the end of the 90 day grace period), death from any cause other than suicide, end of registration with the CPRD, or end of the study period (29 March 2021), whichever occurred first. Patients who switched from a GLP-1 receptor agonist to the comparator drug, or vice versa, did not re-enter the cohort after switching.
Study outcomes
Our primary outcome was suicidality, which we defined as a composite of suicidal ideation (identified in the CPRD), hospital admission for self-harm (identified in HES APC), or completed suicide (identified in ONS). We identified self-harm and completed suicide on the basis of ICD-10 codes X60-84 and Y87.0 (a complete description of codes can be found in supplementary table S1). Our secondary outcomes were suicidal ideation, hospital admission for self-harm, and suicide, each considered separately. Thus, for a given secondary outcome, patients were not censored if they had another event that was part of the composite outcome (other than suicide).
Potential confounders
We considered a comprehensive list of potential confounders measured at or before cohort entry, which we chose on the basis of expert and a priori knowledge. They included age (modelled as a cubic spline with five interior knots), sex, socioeconomic status (based on the Index of Multiple Deprivation), body mass index (<30, 30-34.9, 35-39.9, and ≥40), proxies for severity diabetes (haemoglobin A1c, duration of diabetes, microvascular and macrovascular complications, previous use of antihyperglycaemic drugs), mental health disorders (depression, anxiety, bipolar disorder, psychosis, schizophrenia, obsessive-compulsive disorder, attention deficit/hyperactivity disorder, borderline personality disorder, eating disorder, post-traumatic stress disorder, history of self-harm, history of suicidal ideation), and other variables previously associated with suicidality (insomnia, epilepsy, dementia, chronic pain, history of cancer, substance use).29303132 Additionally, we considered drugs prescribed for common comorbidities, indicators of health seeking behaviour, and the calendar year of cohort entry. Variables with missing information (body mass index, smoking, haemoglobin A1c, Index of Multiple Deprivation) were modelled with an unknown category. A description of the confounders and their measurement periods can be found in supplementary table S2. Supplementary figure 1 illustrates the relations between the exposure, outcome, and covariates through a directed acyclic diagram.
Statistical analysis
Within each cohort, we used propensity score fine stratification weighting to account for confounding.33 We achieved this by using multivariable logistic regression to predict the probability of receiving a GLP-1 receptor agonist versus the comparator drug (a DPP-4 inhibitor or SGLT-2 inhibitor) conditional on the previously listed covariates. We trimmed patients in non-overlapping regions of the propensity score distributions and generated 50 strata based on the propensity score distribution of the GLP-1 receptor agonist users. GLP-1 receptor agonist users were assigned a weight of one, and users of the comparator drug class were reweighted to be proportional to the number of GLP-1 receptor agonist users in each stratum.3334 Further details on propensity score fine stratification weighting are available in the supplementary methods. The estimand was the average treatment effect among the treated patients, representing the effect of the GLP-1 receptor agonists when exposed patients used a GLP-1 receptor agonist versus a comparator drug.33 We used descriptive statistics to assess the balance between the exposure groups by using absolute standardised mean differences, with differences of less than 0.10 being considered acceptable.
We calculated crude and weighted incidence rates for the primary and secondary outcomes with 95% confidence intervals based on the Poisson distribution. We used weighted Kaplan-Meier curves to show the cumulative incidence of the outcome for each exposure group over the follow-up period. Finally, we fitted weighted Cox proportional hazards models to estimate hazard ratios of the outcomes and 95% confidence intervals by using robust variance estimators to compare the use of GLP-1 receptor agonists with the use of the comparator drug class. We visually examined Schoenfeld residuals and included an interaction term with the natural logarithm of time to ensure no violation of the proportional hazards assumption.35
We did several secondary analyses based on the primary outcome for both cohorts. Firstly, we assessed whether the relation between GLP-1 receptor agonist use and suicidality differed between the individual drugs in the class by doing separate analyses for each GLP-1 receptor agonist (exenatide, liraglutide, lixisenatide, dulaglutide, and semaglutide). Secondly, we determined whether the association between GLP-1 receptor agonist use and the risk of suicidality is modified by age at cohort entry (18-30, 31-50, 51-70, and >70 years), sex, history of self-harm, history of suicidal ideation, history of depression, body mass index (<30, 30-34.9, 35-39.9, and ≥40), or socioeconomic status (categorised according to fifths of the Index of Multiple Deprivation36) by including an interaction term between the exposure variable and the effect modifier of interest in our regression models. Thirdly, we assessed for heterogeneity of treatment effects across the range of the propensity score by categorising propensity score into tenths based on the distribution among GLP-1 receptor agonist users and including an interaction term between the exposure variable and propensity score tenth in our regression model.37
To assess the robustness of our findings, we did six sensitivity analyses. Firstly, we varied the grace period between consecutive prescriptions to 60 and 120 days. Secondly and thirdly, we did two analyses to assess the potential effect of informative censoring—one analysis using time varying inverse probability of censoring weighting to account for censoring due to drug switching or discontinuation between the two exposure groups, mortality, and administrative censoring during the follow-up and one analysis using an intention-to-treat exposure definition (that is, patients were followed regardless of treatment discontinuation or switching) with up to one year of follow-up (which was implemented to minimise the effect of exposure misclassification in this analysis). Fourthly, we used multiple imputation with chained equations to evaluate the influence of missing data on our findings for the following variables: smoking, body mass index, haemoglobin A1c, and socioeconomic status (measured by the Index of Multiple Deprivation). Fifthly, we expanded the outcome definition also to include self-harm events recorded in the CPRD. Finally, we did a sensitivity analysis using an array approach to determine the necessary conditions for an unmeasured variable to substantially confound the observed association. This analysis sets the prevalence of an unmeasured confounder among unexposed patients to a given value and allows for the prevalence of the confounder among exposed patients and the strength of the confounder-outcome association to vary across an array of values.38
This study emulates two target trials, one comparing GLP-1 receptor agonists with DPP-4 inhibitors and the other comparing GLP-1 receptor agonists with SGLT-2 inhibitors. Supplementary table S3 compares the characteristics of the target trials and the emulated trials using our observational datasets.
We used SAS version 9.4 for all statistical analyses. Data visualisation was done in R version 4.4.0, with the ggplot version 3.5.1 and dagitty version 0.3-4 packages,3940 and in Microsoft Excel.
Patient and public involvement
Our study was a secondary data analysis and did not include patients as study participants. No patients were involved in setting the research question or the outcome measures; nor were they involved in the design and implementation of the study. This is because no specific funding had been allocated for this purpose. Moreover, CPRD data are not publicly available, and the analysis plan required specialised training.
Results
GLP-1 receptor agonists versus DPP-4 inhibitors
The cohort included 36 082 and 234 028 new users of GLP-1 receptor agonists and DPP-4 inhibitors, respectively (fig 1). We followed these patients for a median of 1.3 (interquartile range 0.6-2.8) years and 1.7 (0.7-3.7) years, respectively.
Study flow diagram of patients in UK Clinical Practice Research Datalink starting glucagon-like peptide-1 (GLP-1) receptor agonist or dipeptidyl peptidase-4 (DPP-4) inhibitor between 1 January 2007 and 31 December 2020
Table 1 and table 2 show the characteristics of the exposure groups before and after weighting. Before weighting, GLP-1 receptor agonist users were younger and more likely to have obesity compared with DPP-4 inhibitor users. GLP-1 receptor agonist users had a longer duration of diabetes and higher levels of haemoglobin A1c; they were also more likely to have a history of depression, anxiety, schizophrenia, self-harm or suicide attempt, and insomnia. All covariates were well balanced after weighting. The propensity score distributions of the exposure groups before and after weighting are shown in supplementary figure S2A.
Baseline characteristics (demographics, lifestyle, and socioeconomic factors) of glucagon-like peptide-1 (GLP-1) receptor agonist and dipeptidyl peptidase-4 (DPP-4) inhibitor users before and after weighting. Values are numbers (percentages) unless stated otherwise
Baseline characteristics (clinical characteristics and medical history) of glucagon-like peptide-1 (GLP-1) receptor agonist and dipeptidyl peptidase-4 (DPP-4) inhibitor users before and after weighting. Values are numbers (percentages) unless stated otherwise
Primary outcome
Three hundred and one suicidality events occurred over 77 377 person years among GLP-1 receptor agonist users (crude incidence rate 3.9 per 1000 person years), and 1087 events occurred over 599 271 person years among DPP-4 inhibitor users (crude incidence rate 1.8 per 1000 person years) (table 3). In crude analyses, the use of GLP-1 receptor agonists generated an elevated hazard ratio compared with the use of DPP-4 inhibitors (crude hazard ratio 2.08, 95% confidence interval (CI) 1.83 to 2.36). After weighting, the use of GLP-1 receptor agonists was not associated with an increased risk of suicidality compared with DPP-4 inhibitors (weighted incidence rate 3.9 (95% CI 3.5 to 4.4) v 3.7 (3.6 to 3.9) per 1000 person years, respectively; weighted hazard ratio 1.02, 0.85 to 1.23) (table 3). The cumulative survival curves overlapped and crossed during the follow-up period (fig 2). Visual inspection of Schoenfeld residuals showed no violation of the proportional hazards assumption, and the P value for interaction with the natural logarithm of time was 0.83.
Hazard ratios for suicidal ideation, self-harm, and completed suicide comparing glucagon-like peptide-1 (GLP-1) receptor agonists with dipeptidyl peptidase-4 (DPP-4) inhibitors
Weighted cumulative incidence curves for composite of suicidal ideation, self-harm, or completed suicide among patients taking glucagon-like peptide-1 receptor agonist (GLP-1 RA) versus patients taking dipeptidyl peptidase-4 (DPP-4) inhibitor
Secondary outcomes
Table 3 shows the results for suicidal ideation, self-harm, and suicide separately. Overall, the crude analyses generated elevated hazard ratios for all outcomes when we compared GLP-1 receptor agonists with DPP-4 inhibitors. After weighting, the hazard ratios were close to the null value for suicidal ideation (1.00, 95% CI 0.79 to 1.26) and self-harm (0.94, 0.70 to 1.25); the hazard ratio for suicide was below the null value but with a wide confidence interval (0.77, 0.19 to 3.03).
Secondary and sensitivity analyses
In our secondary analyses, we did not observe evidence of effect measure modification by individual drug, age, sex, body mass index, history of suicidal ideation, history of depression, or socioeconomic status, as indicated by overlapping 95% confidence intervals in all comparisons (supplementary tables S4-S7). Overall, a history of self-harm or suicide attempt was associated with a lower hazard ratio than no history, but with overlapping confidence intervals that encompassed the null (hazard ratio 0.74 (95% CI 0.54 to 1.01) v 1.17 (0.93 to 1.48)). We also observed no clear patterns of treatment effects across tenths of the propensity score, with wide and overlapping confidence intervals and point estimates largely close to the null (supplementary table S8).
Our results remained consistent across all sensitivity analyses (fig 3; supplementary tables S9-S10). The array sensitivity analysis indicated that the two exposure groups would need to have at least a 15 percentage point imbalance in an unmeasured confounder with a confounder-outcome risk ratio of 3.5 or a 30 percentage point imbalance with a confounder-outcome risk ratio of 2.0 to bias our results by at least 25% (supplementary figure S3A). In other words, a hypothetical unmeasured confounder could substantially bias our results only if it was both highly imbalanced between the exposure groups and strongly associated with the outcome.
Hazard ratios and 95% confidence intervals (CIs) from sensitivity analyses for composite outcome of suicidal ideation, self-harm, or completed suicide among patients taking glucagon-like peptide-1 receptor agonist (GLP-1 RA) versus patients taking dipeptidyl peptidase-4 (DPP-4) inhibitor. CPRD=Clinical Practice Research Datalink
GLP-1 receptor agonists versus SGLT-2 inhibitors
Overall, we included 32 336 GLP-1 receptor agonist users and 96 212 SGLT-2 inhibitor users in the cohort and followed them for a median of 1.2 (interquartile range 0.5-2.3) and 1.2 (0.5-2.5) years, respectively (fig 4).
Study flow diagram of patients in UK Clinical Practice Research Datalink starting glucagon-like peptide-1 receptor agonist (GLP-1) or sodium-glucose cotransporter-2 (SGLT-2) inhibitor between 1 January 2013 and 31 December 2020
Table 4 and table 5 show the characteristics of the exposure groups before and after weighting. Before weighting, patients with prescriptions for GLP-1 receptor agonists were more likely to be female and to have obesity, a longer duration of diabetes, a higher level of haemoglobin A1c, more microvascular complications, and a history of depression compared with those treated with SGLT-2 inhibitors. The covariates were all well balanced after weighting. Supplementary figure S2B shows the propensity score distribution of the GLP-1 receptor agonist and SGLT-2 inhibitor cohorts before and after weighting.
Baseline characteristics (demographics, lifestyle, and socioeconomic factors) of glucagon-like peptide-1 (GLP-1) receptor agonist and sodium-glucose cotransporter-2 (SGLT-2) inhibitor users before and after weighting. Values are numbers (percentages) unless stated otherwise
Baseline characteristics (clinical characteristics and medical history) of glucagon-like peptide-1 (GLP-1) receptor agonist and sodium-glucose cotransporter-2 (SGLT-2) inhibitor users before and after weighting. Values are numbers (percentages) unless stated otherwise
Primary outcome
Two hundred and forty suicidality events occurred over 55 620 person years among patients with prescriptions for GLP-1 receptor agonists (crude incidence rate 4.3 per 1000 person years) and 454 events over 168 384 person years among patients treated with SGLT-2 inhibitors (crude incidence rate 2.7 per 1000 person years) (table 6). The hazard ratio was elevated in the crude analysis comparing GLP-1 receptor agonists with SGLT-2 inhibitors (hazard ratio 1.60, 95% CI 1.37 to 1.87) but null after adjustment for confounding (weighted incidence rate 4.3 (95% CI 3.8 to 4.9) v 4.6 (4.3 to 4.9), respectively; weighted hazard ratio 0.91, 0.73 to 1.12). Visual inspection of the cumulative incidence curves (fig 5) and Schoenfeld residuals showed no violation of proportionality, and the P value for interaction with the natural logarithm of time was 0.56.
Hazard ratios for suicidal ideation, self-harm, and completed suicide comparing glucagon-like peptide-1 (GLP-1) receptor agonists with sodium-glucose cotransporter-2 (SGLT-2) inhibitors
Weighted cumulative incidence curves for composite of suicidal ideation, self-harm, or completed suicide among patients taking glucagon-like peptide-1 receptor agonist (GLP-1 RA) versus patients taking sodium-glucose cotransporter-2 (SGLT-2) inhibitor
Secondary outcomes
Crude analyses generated elevated hazard ratios for suicidal ideation, self-harm, and suicide separately when we compared GLP-1 receptor agonists with SGLT-2 inhibitors. After control for confounding, the hazard ratio for self-harm was around the null value (1.03, 95% CI 0.70 to 1.51), and those for suicidal ideation and suicide were under the null value but with wide confidence intervals (0.84 (0.65 to 1.08) and (0.14 (0.02 to 1.31)), respectively (table 6).
Secondary and sensitivity analyses
In secondary analyses, we found no evidence of effect measure modification by individual drug, age, history of self-harm or suicide attempt, history of suicidal ideation, or socioeconomic status (supplementary tables S11-S14). With respect to sex, the hazard ratio was lower among men (0.74, 95% CI 0.54 to 1.01) than women (1.10, 0.82 to 1.48), but with overlapping confidence intervals. Similarly, the hazard ratio was lower among patients without a history of depression (0.57, 0.32 to 1.00) than among those with such a history (1.01, 0.80 to 1.29). Finally, hazard ratios were elevated for patients with lower body mass index and decreased with higher body mass index (<30: 1.72, 1.02 to 2.89; 30-34.9: 1.34, 0.91 to 1.98; 35-39.9: 0.61, 0.41 to 0.91; ≥40: 0.78, 0.53 to 1.16). Point estimates for the treatment effect by tenth of propensity score did not follow any clear pattern and were largely close to the null with wide and overlapping confidence intervals for most tenths (supplementary table S15). However, we observed slight non-overlap between the lower limits of the confidence interval for the first and second tenths with the upper limit of the confidence interval for the last tenth.
All sensitivity analyses yielded results consistent with our main findings (fig 6; supplementary tables S16 and S17). The array sensitivity analysis showed that an unmeasured confounder would again need to have at least 15% imbalance and a confounder-outcome association of 3.5 or 30% imbalance and a confounder-outcome association of 2.0 to bias results by 25% or more (supplementary figure S3B).
Hazard ratios and 95% confidence intervals (CIs) from sensitivity analyses for composite outcome of suicidal ideation, self-harm, or completed suicide among patients taking glucagon-like peptide-1 receptor agonist (GLP-1 RA) versus patients taking sodium-glucose cotransporter-2 (SGLT-2) inhibitor. CPRD=Clinical Practice Research Datalink
Discussion
In this large active comparator, new user cohort study, the use of GLP-1 receptor agonists was not associated with an increased risk of suicidality compared with DPP-4 inhibitors or SGLT-2 inhibitors. We observed similar findings when investigating suicidal ideation, self-harm, and suicide as separate outcomes. Overall, our findings remained consistent across multiple sensitivity analyses.
Comparison with other studies
Overall, events of suicidal ideation and self-harm have not been found to be imbalanced in clinical trials between randomised groups.67 However, these trials were limited by their sample sizes and thus could not detect these rare but potentially important adverse events. Several pharmacovigilance analyses assessed a potential association between GLP-1 receptor agonists and suicidality, several of which found significantly increased reporting odds ratios for at least one safety signal.41424344 Although useful for signal generation, pharmacovigilance analyses are subject to incomplete reporting and confounding and have limited use in establishing causal associations. Additionally, these studies had much heterogeneity owing to the inclusion of different outcomes and different comparator drugs.
Several observational studies have investigated the potential association between GLP-1 receptor agonist use and different outcomes related to suicidality.8910111213 In a study published in 2018, GLP-1 receptor agonists were not associated with a composite of depression and self-harm (hazard ratio 1.25, 95% CI 0.63 to 2.50) in patients with type 2 diabetes. However, that study had few GLP-1 receptor agonist users (n=501), leading to an inability to assess self-harm separately and imprecise estimates.8 In a more recent study, semaglutide was found to be strongly associated with a decreased risk of both incident and recurrent suicidal ideation separately among patients with type 2 diabetes and with overweight or obesity (hazard ratios 0.27 (95% CI 0.20 to 0.36) for incident suicidal ideation and 0.44 (0.32 to 0.60)for recurrent suicidal ideation among patients with overweight or obesity; 0.36 (0.25 to 0.53) for incident suicidal ideation and 0.51 (0.31 to 0.86) for recurrent suicidal ideation among patients with type 2 diabetes).9 A third study found protective effects of GLP-1 receptor agonists compared with DPP-4 inhibitorson the risk of suicide attempts among patients with type 2 diabetes (odds ratio 0.46, 95% CI 0.37 to 0.58),10 and a fourth study found no effect of GLP-1 receptor agonists compared with SGLT-2 inhibitors or DPP-4 inhibitors on the risk of suicidal ideation (hazard ratios 1.05 (95% CI 0.77 to 1.43) and 0.96 (0.72 to 1.27), respectively) or suicidal behaviours (0.82 (0.41 to 1.67) and 0.92 (0.42 to 2.01), respectively) among older adults with type 2 diabetes.11 Another study conducted among patients with both type 2 diabetes and obesity observed no association between GLP-1 receptor agonist use and a composite of suicidal ideation and self-injury (hazard ratio 1.04, 95% CI 0.35 to 3.14),12 and a final study observed a reduced risk of self-harm (hazard ratio 0.77, 0.65 to 0.91) but a non-significant increase in the risk of suicide (1.25, 0.83 to 1.88) among a diverse group of patients on GLP-1 receptor agonists compared with SGLT-2 inhibitors.13 Overall, some of these observational studies had methodological shortcomings, including prevalent user bias,9 immortal time bias,9 important confounding,9101113 limited generalisability,11 exclusion of patients at high risk (such as those with a history of psychiatric conditions),811 and limited sample sizes generating imprecise estimates.812
Biological mechanisms
The absence of an independent association between GLP-1 receptor agonists and suicidality suggests that the effects of these drugs on the hypothalamic-pituitary-adrenal axis, sudden weight loss, or brain derived neurotrophic factor dysregulation are not sufficient to induce suicidal ideation, self-harm, or suicide.1415161718 Although these biological mechanisms may contribute to the complex interplay of factors influencing suicidality, they do not seem to act independently in causing such outcomes in GLP-1 receptor agonist users. We did not observe associations even among patients with a history of psychiatric disorders, such as self-harm, suicidal ideation, and depression.
Strengths and limitations of study
This study has several strengths. Firstly, using the CPRD and its linked databases allowed us to account for potentially important confounders, such as body mass index, and laboratory tests, such as haemoglobin A1c, which are often absent in administrative databases. Secondly, we used an active comparator, new user design, an approach that minimises biases related to the inclusion of prevalent users. Contrary to most of the previous studies,89101112 our outcome definition also included suicide, a relevant outcome when assessing the safety of GLP-1 receptor agonists. Additionally, unlike several other studies,811 our study population did not exclude patients with a history of self-harm or suicidal ideation, who could be those at the highest risk of experiencing adverse psychiatric events associated with GLP-1 receptor agonist use.
The study also has some limitations. Firstly, as with all observational studies, residual confounding is possible. Our models were weighted to account for a wide range of potentially important confounders. This weighting dropped the crude hazard ratios of 2.08 and 1.60 to 1.02 and 0.91 in the comparisons with DPP-4 inhibitors and SGLT-2 inhibitors, respectively. The array sensitivity analyses indicated that this association would be confounded only under implausible conditions, in which the unmeasured variable is highly imbalanced between exposure groups and strongly associated with the outcome. That such a variable exists beyond those we considered in our propensity score models is unclear. Secondly, the CPRD does not include prescriptions written by specialists as it is a primary care database. However, general practitioners are responsible for the management of type 2 diabetes in the UK, meaning that this is unlikely to be an important source of misclassification.45 Additionally, exposure misclassification is possible given that prescriptions in the CPRD are those written by general practitioners and not those filled by the patient; however, our on-treatment exposure definition based on overlapping prescriptions likely minimised this potential bias. In the primary analysis, we identified self-harm by using hospital admission events, thus restricting it to the most serious events to maximise the specificity of the outcome definition, but leading to the possibility of outcome misclassification. Reassuringly, we obtained nearly identical results in a sensitivity analysis that also expanded the outcome definition to include self-harm events recorded in the CPRD. Additionally, the rates of suicide and self-harm observed in our study are similar to those observed in the general population in England.31 Finally, given the rarity of the outcomes, some secondary analyses generated wide confidence intervals, so these should be interpreted with caution.
Conclusions
In summary, the results of this large active comparator, new user cohort study indicate that GLP-1 receptor agonists are not associated with an increased risk of suicidal ideation, self-harm, and suicide when compared with DPP-4 inhibitors or SGLT-2 inhibitors among patients with type 2 diabetes. These findings should provide some reassurance with respect to the psychiatric safety of these drugs.
What is already known on this topic
Concerns exist about a potential association between glucagon-like peptide-1 receptor (GLP-1) agonists and suicidality
This has prompted regulatory agencies to conduct comprehensive reviews of this drug class
Observational studies investigating this safety concern remain inconclusive.
What this study adds
GLP-1 receptor agonist use was not associated with an increased risk of suicidal ideation, self-harm, or suicide compared with use of dipeptidyl peptidase-4 inhibitors or sodium-glucose cotransporter-2 inhibitors
GLP-1 receptor agonists do not seem to increase the risk of suicidality among patients with type 2 diabetes
Ethics statements
Ethical approval
The study protocol was approved by the Research Data Governance of the Clinical Practice Research Datalink (protocol 24_003735) and by the Research Ethics Board of the Jewish General Hospital, Montreal, Canada. The study was a secondary data analysis and did not include patients as study participants. Hence, the Research Data Governance of the CPRD and the Research Ethics Board deemed it exempt from the need to obtain patient consent.
Data availability statement
ICD-10 codes to identify the outcome are included in supplementary table S1. Example code lists used to identify covariates are included in supplementary table S2. Codes used to used to identify antidiabetic drugs, suicidal ideation, and self-harm in the CPRD, as well as reproducible codes for cohort construction, propensity score generation, propensity score fine stratification weighting, and primary analyses are available on Open Science Framework and are linked in the supplementary methods. No other data are available. This study is based on data from the CPRD obtained under licence from the UK Medicines and Healthcare products Regulatory Agency. The data are provided by patients and collected by the UK National Health Service as part of their care and support. The interpretation and conclusions contained in this study are those of the author/s alone. Because electronic health records are classified as “sensitive data” by the UK Data Protection Act, information governance restrictions (to protect patient confidentiality) prevent data sharing via public deposition. Data are available with approval through the individual constituent entities controlling access to the data. Specifically, the primary care data can be requested via application to the CPRD (https://www.cprd.com).
Acknowledgments
SBS is the recipient of a doctoral training award from the Canadian Institutes of Health Research. OHYY holds a junior 1 clinical research scholar award from the Fonds de Recherche du Québec – Santé. SR holds a provincial clinician scientist award from the Fonds de Recherche du Québec – Santé. SS is the recipient of the distinguished James McGill award from McGill University. LA holds a distinguished research scholar award from the Fonds de Recherche du Québec – Santé and is the recipient of a William Dawson scholar award from McGill University.
Footnotes
Contributors: SBS wrote the manuscript, and all authors edited, critically reviewed, and approved the final version of the manuscript. All authors conceived and designed the study, analysed and interpreted the data, approved the final version of the manuscript, and are accountable for its accuracy. SBS, HY, and LA did the statistical analyses. LA acquired the data and supervised the study. LA is the guarantor. The corresponding author attests that all listed authors meet authorship criteria and that no others meeting the criteria have been omitted.
Funding: This study was funded by a Foundation Scheme grant from the Canadian Institutes of Health Research (FDN-143328). The funders had no role in considering the study design or in the collection, analysis, or interpretation of data, the writing of the report, or the decision to submit the article for publication.
Competing interests: All authors have completed the ICMJE uniform disclosure form at https://www.icmje.org/disclosure-of-interest/ and declare: this study was funded by the Canadian Institutes of Health Research; SR is a shareholder of Aifred Health; SS attended scientific advisory committee meetings or received speaking fees from AstraZeneca, Atara, Boehringer Ingelheim, Bristol-Myers-Squibb, Covis, Novartis, Panalgo, Pfizer, CSL Behring, and Covispharma, all unrelated to this study; LA received speaking and consulting fees from Pfizer, Boehringer Ingelheim, and Roche for work unrelated to this study; no other relationships or activities that could appear to have influenced the submitted work.
Transparency: The lead author (the manuscript’s guarantor) affirms that this manuscript is an honest, accurate, and transparent account of the study being reported; that no important aspects of the study have been omitted; and that any discrepancies from the study as planned (and, if relevant, registered) have been explained.
Dissemination to participants and related patient and public communities: We will disseminate the results through our general media communications plan by organising a press release through the Public Relations Department of the Lady Davis Institute. Additionally, we will directly communicate our results to diabetes associations and regulatory agencies.
Provenance and peer review: Not commissioned; externally peer reviewed.
This is an Open Access article distributed in accordance with the Creative Commons Attribution Non Commercial (CC BY-NC 4.0) license, which permits others to distribute, remix, adapt, build upon this work non-commercially, and license their derivative works on different terms, provided the original work is properly cited and the use is non-commercial. See: http://creativecommons.org/licenses/by-nc/4.0/.