Use of GLP-1 receptor agonists and risks of suicide attempts or self-harm in patients with type 2 diabetes: a multicountry self-control case series study

Abstract

Background

Inconclusive findings regarding the association between suicidal ideation/suicide attempt and glucagon-like peptide-1 receptor agonists (GLP-1RAs) have been recently revealed in a small number of studies.

Methods

This was a multinational self-controlled case series analysis using Hong Kong’s Clinical Data Analysis and Reporting System (2008−2023), Taiwan’s National Health Insurance Research Database (2012−2020) and the UK’s IQVIA Medical Research Database with The Health Improvement Network (2000−2021). A total of 642 suicide attempt or self-harm cases with GLP-1RA use were included to assess pooled incident rate ratios (IRRs) of suicide attempts or self-harm associated with GLP-1RA treatment versus non-treatment with their 95% CIs.

Results

The pooled IRR (95% CI) of suicide attempts or self-harm associated with GLP-1RA treatment versus non-treatment was 0.67 (0.51 to 0.88). The suicide attempt or self-harm risk varied with the time window of GLP-1RA use, with pooled IRRs (95% CIs) of 1.94 (0.86 to 4.37), 0.61 (0.23 to 1.63), 0.72 (0.37 to 1.41), 0.60 (0.32 to 1.09) and 0.63 (0.49 to 0.87) for the pretreatment period and Days 1−30, Days 31−90, Days 91−180 and Days>180 of GLP-1RA treatment, respectively. Subgroup analyses by age, sex and individual GLP-1RAs and sensitivity analyses showed no significant increase in the suicide attempt or self-harm risk associated with GLP-1RA use. The point estimate and CI of the E-value for suicide attempts or self-harm were 2.35 and 1.53, respectively.

Conclusions

We found no increase in the risks of suicide attempts or self-harm following GLP-1RA treatment, and even in the long-term use of GLP-1RAs. Close monitoring of potential suicide attempts or self-harm and ensuring treatment tolerability during treatment initiation are required, and well-controlled or pragmatic trials remain warranted to validate our findings.

WHAT IS ALREADY KNOWN ON THIS TOPIC

• Case reports have raised concerns about rare but serious psychiatric adverse events such as suicidal ideation and suicide attempt following glucagon-like peptide-1 receptor agonist (GLP-1RA) treatment, which triggered the European Medicines Agency to initiate a safety review in July 2023.

• Inconclusive findings from current evidence were revealed. Several disproportionality analyses using the adverse drug reaction systems showed the increased reporting OR of suicide attempts associated with GLP-1RAs versus metformin or within GLP-1RAs, while some cohort studies showed either neutral or protective effects of GLP-1RAs on suicidal ideation or suicide attempts.

WHAT THIS STUDY ADDS

• In this self-controlled case series analysis of data obtained from Hong Kong, Taiwan and the UK, no significant increase in the suicide attempt or self-harm risk in the GLP-1RA treatment period was observed. Compared with the non-treatment period, lower suicide attempt or self-harm risk following GLP-1RA treatment was observed, especially after 6 months of treatment.

HOW THIS STUDY MIGHT AFFECT RESEARCH, PRACTICE OR POLICY

• Use of GLP-1RAs was not associated with an increased risk of suicide attempts or self-harm. Close monitoring for suicide attempts or self-harm is still suggested during the initial GLP-1RA treatment phase.

Background

Emerging evidence has shown the remarkable benefits of glucagon-like peptide-1 receptor agonist (GLP-1RA) treatment for cardiovascular,¹ kidney¹ and liver diseases² and weight loss³ beyond its apparent glycaemic control effect. GLP-1RA thus has an increasingly important role in the clinical management of type 2 diabetes (T2D), overweight and obesity. Despite its growing popularity, case reports have raised concerns about rare but serious psychiatric adverse events such as suicidal ideation and suicide attempt following GLP-1RA treatment, which triggered the European Medicines Agency to initiate a safety review in July 2023.⁴ Hence, it is crucial to investigate the potential risk of suicide attempt associated with the use of GLP-1RAs to help prevent any adverse events that may occur among patients receiving this treatment.

Yet, several studies⁵,¹⁶ on this topic have reported inconclusive results (online supplemental table 1). Some of these studies using the adverse drug reaction systems showed the increased reporting OR of GLP-1RAs versus metformin⁷ or within GLP-1RAs,⁵ and no elevated reporting OR of GLP-1RAs versus non-GLP-1RAs.^{6 8} However, these results rely on the spontaneous reports of adverse events and thus self-harm and associated behaviours (eg, suicidal attempts) are likely under-reported. Moreover, the disproportionality analyses conducted in these studies are suitable for only signal detection; they could not determine association or causality.⁵,⁸ Other meta-analysis¹⁵ or post-hoc analysis¹² using trial data, cohort studies with relatively short follow-up periods (ie, less than 1 year,^{9 10} around 1 year^{13 14}) or using target trial emulation design,¹¹ or case-time control studies¹⁶ also yielded inconclusive results, namely either neutral⁹¹¹,¹⁵ or protective^{10 16} effects of GLP-1RAs on suicidal ideation, self-harm or depression, which might be due in part to different patient populations of interest (ie, patients with diabetes⁹¹¹,¹⁵ or overweight/obesity^{10 12 15}) and study comparators (ie, other glucose-lowering agents^{9 11 13 14} or anti-obesity medications¹⁰). Uncertainty regarding between-treatment group (ie, GLP-1RAs vs anti-obesity medications) comparability in unmeasured patient baseline characteristics (eg, lifestyle, health awareness, and healthy behaviours of people seeking anti-obesity treatments) in these studies is of concern and affects their validity. Lastly, existing evidence was obtained from the USA^{6 7} and the UK,⁹ with no data from other countries. However, the treatment pattern of GLP-1RAs might vary across countries with different healthcare systems and levels of cultural adoption of injectable treatment, possibly affecting the risks of mental-related adverse effects of GLP-1RA use. Hence, studies that include data from multiple geographical regions and ethnicities are urgently needed.

In light of the above, this study assesses the association of GLP-1RA treatment with suicide attempt risk by using multicountry data from Taiwan, Hong Kong (HK) and the UK. We implemented the self-controlled case series (SCCS) approach to examine whether the risk of suicide attempt during GLP-1RA treatment periods was higher than that during non-treatment periods. The SCCS design directly compares the incidence rate of suicide attempt between treatment and non-treatment periods for each person, thus removing inter-person variability and implicitly controlling for time-invariant confounders. Such a within-person comparison design overcomes many of the issues encountered with traditional inter-person comparison designs (eg, cohort studies). This study thus provides timely evidence for health authorities to increase their awareness regarding the safety of patients using GLP-1RAs.

Methods

Data sources

This multicountry study used three large databases from HK, Taiwan and the UK, respectively, namely the Clinical Data Analysis and Reporting System (CDARS) for 2004–2023, Taiwan’s National Health Insurance Research Database (NHIRD) for 2012–2020, and the IQVIA Medical Research Database (IMRD) for 2000–2021. Briefly, CDARS contains anonymised individual-level electronic medical records regarding clinical services provided by the HK Hospital Authority for over 7.4 million people,¹⁷ including medical records of diagnoses, prescriptions, accidents, emergency room and outpatient visits, hospitalisations and laboratory tests. Taiwan’s NHIRD contains anonymised individual-level claims-based data of healthcare services (eg, outpatient and emergency room visits, hospitalisations, and medication prescriptions) of beneficiaries enrolled in the National Health Insurance programme, which includes over 99% of Taiwan’s population.¹⁸ The IMRD (incorporating data from The Health Improvement Network, a Cegedim database) contains anonymised person-level electronic medical records of more than 12 million individuals (representing nearly 6.2% of the UK population¹⁹) in over 587 primary care practices in the UK.

Study populations

First, given the different time points of GLP-1RA introduction into markets across countries (ie, 2008 for HK, 2012 for Taiwan and 2007 for the UK), patients who received any prescriptions of GLP-1RAs and had suicide attempts or self-harm (a composite outcome) during the study period specific to each country (ie, 1 January 2008 to 30 November 2023 in HK, 1 January 2014 to 31 December 2020 in Taiwan and 1 January 2008 to 2 November 2021 in UK) were included. The study start date was defined as 1 January 2008, 1 January 2014 and 1 January 2008 for study subjects in HK, Taiwan and the UK, respectively. In particular, for subjects identified in the IMRD, only those with at least 1 year of up-to-standard data recorded with a patient flag of ‘A’ or ‘C’ in the database were analysed. Furthermore, the remaining years before the start date served as washout periods to exclude those who had been exposed to GLP-1RAs or had prior suicide attempts. Of note, all GLP-1RAs were used for patients with T2D, given that the indication of GLP-1RAs for obesity was not approved yet during the study periods across study sites. All study populations were followed from the study start dates or 1 year after the patient’s registration to a general practice (UK only) until death, the date of transferring out from the general practice (UK only) or the end of the study period, whichever came first. A flowchart of the selection of the study populations is shown in online supplemental figure 1.

Treatment period definitions and outcome measurements

Exposure to GLP-1RAs, namely liraglutide, dulaglutide, semaglutide, exenatide and lixisenatide, was measured according to the WHO Anatomical Therapeutic Chemical Classification System. A 30-day gap was used as the grace period to define the status of treatment continuation. The study periods (see figure 1) were classified into treatment and non-treatment periods (which served as reference periods in the analysis). Multiple treatment periods per person were allowed. The study outcome of interest was a composite outcome, including suicide attempts and self-harm. That is, only the cases with suicide attempts or self-harm but still survived were included to avoid the event-dependent censoring. The operational definitions of a suicide attempt or self-harm across the study databases are detailed in online supplemental table 2. Considering the possibility of the dependence of multiple suicide attempts or self-harm for a given person, only the first occurrence of a suicide attempt or self-harm in the study period was included in the analysis.²⁰

Figure 1. Study scheme and definitions of treatment periods for individual study sites. ^*Given the different data availability of the databases (cells highlighted in light yellow) and different GLP-1RA introduction dates in different countries (ie, 2007 for Hong Kong, 2008 for the UK and 2011 for Taiwan), the length of the washout period was specific to each database. ^†The grace period was 30 days after the discontinuation of prescriptions (for all three databases). GLP-1RA, glucagon-like peptide-1 receptor agonist.

Statistical analyses

In the descriptive analyses, we obtained the means and corresponding SDs for continuous variables (eg, age at cohort entry and follow-up time) and the frequencies and associated percentages for dichotomous variables (ie, gender). The incident rates of the composite outcomes of suicide attempts and self-harm are presented as the number of events per 100 person-years. Conditional Poisson regression was employed to calculate the incident rate ratio (IRR) of the composite outcome associated with GLP-1RA treatment versus non-treatment and the corresponding 95% CI. For potential confounders, the time-invariant variables (eg, gender) were implicitly controlled for by the SCCS design, where the comparisons are made only within individuals,²⁰ and the study observation periods were split according to the years after the start date²⁰ to control the potential effects of time-varying variables (eg, age as a categorical variable indicating the year (ie, 1st, 2nd, 3rd …) to stratify the cases’ observation time) or temporal effect on the study outcome (ie, a suicide attempt or self-harm) arisen from the ageing or the nature of disease progression. After the IRR estimates were obtained from the three databases, a meta-analysis with a random effect model was conducted to generate the pooled IRRs and corresponding 95% CIs and the heterogeneity of the study estimates was assessed using the statistic I².

Several sensitivity analyses were performed to test the robustness of the primary findings. First, to consider the possibility of increased or decreased risks of suicide attempts or self-harm before GLP-1RA exposure, a 14-day pretreatment period before the date of initiation of GLP-1RAs was specified. In addition, to quantify whether the effect of GLP-1RAs on suicide attempts or self-harm varied with the duration of GLP-1RA treatment, the treatment periods were separated into the following time intervals: Day 0, Days 1–30, Days 31–90, Days 91–180 and Days>180. The pretreatment period and these specific treatment periods were considered in the analysis. The risks in different periods were compared. Second, the deceased cases from any cause of death during the study period were excluded to minimise the impact of the event-dependent censoring, given that the subjects might be relatively vulnerable to death. Third, given that the risk of suicide attempts or self-harm is sensitive to age²¹ and gender,²² subgroup analyses were performed for subjects aged ≥50 and <50 years and male and female individuals. Fourth, considering that the difference in molecular structure, size and action duration across individual types of GLP-1RA might affect their transportation to the central nervous system,²³ the primary analysis was repeated with the treatment periods further separated by the specific GLP-1RA use (ie, liraglutide, dulaglutide, semaglutide, exenatide and lixisenatide). Fifth, to evaluate the risk of suicide attempts or self-harm for subjects newly initiated on GLP-1RA treatment, only the first records of GLP-1RA exposure were included and the exposure duration was further classified into (1) the length of the first prescription with a 30-day grace period in the follow-up and (2) a 730-day period from the start date of the first prescription, which was determined from the mean duration of GLP-1RA use in the UK’s clinical practice. The primary analysis was then repeated with different definitions of the treatment period. Lastly, to consider the possible unmeasured confounding effects for a given patient in our study, the E-value was estimated to determine the minimum effect of unmeasured confounders on either treatment or outcome that can nullify the association observed in our analysis. All statistical analyses were performed using SAS software V.9.4 at the Taiwan site and R with the ‘SCCS’ package at the HK and the UK sites.

Results

As shown in online supplemental figure 1, this SCCS study included a total of 642 study subjects (mean age (SD): 43.04 (15.15) to 43.58 (11.51) years, male: 47.9%–51.4%) across the three study sites. The mean durations (SD) of GLP-1RA treatment/follow-up time were 1.64 (1.52)/14.42 (2.06) years, 0.40 (0.34)/6.84 (0.54) years and 2.12 (2.31)/8.71 (3.57) years in HK, Taiwan and the UK, respectively. The patients’ characteristics for the three study sites are detailed in online supplemental table 3.

Table 1 shows the IRRs of suicide attempts or self-harm associated with GLP-1RA treatment versus non-treatment in the pooled estimate and across the three study sites. Overall, the pooled IRR (95% CIs) was 0.67 (0.51 to 0.88) with an I² value of 1%, indicating non-heterogeneity in treatment effects across the three countries. The IRRs (95% CIs) of suicide attempts or self-harm were 0.56 (0.32 to 1.00), 1.49 (0.44 to 5.06) and 0.67 (0.49 to 0.92) for GLP-1RA treatment versus non-treatment in HK, Taiwan and the UK, respectively. Table 2 shows the variation of suicide attempt or self-harm risk in different time intervals of GLP-1RA exposure during the study period. The pooled IRRs (95% CIs) were 1.94 (0.86 to 4.37), 0.61 (0.23 to 1.63), 0.72 (0.37 to 1.41), 0.60 (0.32 to 1.09) and 0.63 (0.49 to 0.87) for the pre-treatment period and Days 1–30, Days 31–90, Days 91–180 and Days>180 of GLP-1RA treatment, respectively.

Table 1. Incident rate ratios of suicide attempts or self-harm associated with GLP-1RA treatment versus non-treatment across countries obtained from primary analyses.

	No. of events		Total follow-up time (pys)		Incident rate of suicide attempts or self-harm (no. of events/100 pys)		Incident rate ratio (95% CI) of suicide attempts or self-harm*
	Non-treatment	Treatment	Non-treatment	Treatment	Non-treatment	Treatment
All sites (pooled)	556	86	2031.69	416.53	27.37	20.65	0.67 (0.51 to 0.88)†
Individual sites (study period, year)
Hong Kong (2004–2023)	245	14	988.24	132.90	24.79	10.53	0.56 (0.32 to 1.00)
Taiwan (2012–2020)	25	5	174.78	30.29	14.30	16.51	1.49 (0.44 to 5.06)
UK (2000–2021)	286	67	868.67	253.34	32.92	26.45	0.67 (0.49 to 0.92)

^*Bold incident rate ratios indicate statistically significant treatment effect associated with GLP-1RA use versus non-use.

^†I2 value of 1% for the pooled incident rate ratio suggests a low level of heterogeneity in treatment effects across the individual sites.

GLP-1RA, glucagon-like peptide-1 receptor agonist; pys, person-years.

Table 2. Incident rate ratios of suicide attempts or self-harm in different time windows of GLP-1RA treatment (pooled from all study sites).

	No. of events	Total follow-up time (pys)	Incident rate of suicide attempts or self-harm (no. of events/100 pys)	Incident rate ratio (95% CI) of suicide attempts or self-harm*
Pretreatment (14 days before initiation of GLP-1RAs)†	6	9.29	64.61	1.94 (0.86 to 4.37)
GLP-1RA treatments in different time intervals†
Days 1–30	4	19.65	20.36	0.61 (0.23 to 1.63)
Days 31–90	9	38.04	23.66	0.72 (0.37 to 1.41)
Days 91–180	11	54.99	20.00	0.60 (0.32 to 1.09)
Day 181 and beyond	59	277.98	21.22	0.63 (0.46 to 0.87)
Non-treatment†	523	1781.92	29.35	Ref.

^*Bold incident rate ratios indicate statistically significant treatment effect associated with GLP-1RA use versus non-use.

^†Given the limited cases at the Taiwan site, which could not be further stratified by treatment duration window, only data from Hong Kong and the UK were analysed.

GLP-1RAs, glucagon-like peptide-1 receptor agonists; pys, person-years.

The results of the sensitivity analyses (table 3) were consistent with those of the primary analyses (table 1). No significant difference in the suicide attempt or self-harm risk with GLP-1RA treatment compared with the reference period was observed in the subgroups of subjects aged<50 years, female and male (pooled IRRs (95% CIs): 0.76 (0.55 to 1.05), 0.60 (0.28 to 1.31) and 0.81 (0.39 to 1.66), respectively). Significantly reduced risks were found in those aged≥50 years (ie, 0.52 (0.32 to 0.84)) and in the analysis where deceased cases during the study period were excluded (0.66 (0.45 to 0.98)). There was no suicide attempt or self-harm risk found in the analysis that restricted the treatment window to the first GLP-1RA prescription; that is, the pooled IRRs (95% CIs) in the first prescription with a 30-day grace period and the 730-day follow-up from the first date of GLP-1RA prescription were 0.64 (0.33 to 1.24) and 1.06 (0.68 to 1.69), respectively. Regarding the risk of suicide attempts or self-harm associated with individual types of GLP-1RA, the pooled IRRs (95%CIs) were 0.64 (0.34 to 1.19) for liraglutide, 0.64 (0.36 to 1.15) for dulaglutide, 0.28 (0.07 to 1.16) for semaglutide, 0.63 (0.25 to 1.61) for exenatide and 2.44 (0.67 to 8.87) for lixisenatide compared with non-treatment. The point estimate and CI of the E-value for suicide attempts or self-harm were 2.35 and 1.53, respectively, suggesting that the results of the primary analyses would be nullified if the relative risk of an unmeasured confounder between exposure and outcome was greater than 2.35 or 1.53.

Table 3. Pooled incident rate ratios of suicide attempts or self-harm associated with GLP-1RA treatment versus non-treatment stratified by different characteristics and settings from subgroup and sensitivity analyses.

	No. of events		Total follow-up time (pys)		Incident rate of suicide attempts or self-harm (no. of events/100 pys)		Incident rate ratio (95% CI) of suicide attempts or self-harm*
	Non-treatment	Treatment	Non-treatment	Treatment	Non-treatment	Treatment
Primary analyses	556	86	2031.69	416.53	27.27	21.13	0.67 (0.51 to 0.88)†
Subgroup and sensitivity analyses
Age groups
<50 years	364	60	1317.18	266.83	27.63	22.49	0.76 (0.55 to 1.05)
≥50 years	192	26	661.47	141.33	29.03	18.40	0.52 (0.32 to 0.84)
Gender
Female	286	44	1018.88	211.80	28.07	20.77	0.60 (0.28 to 1.31)
Male	270	42	959.77	196.36	28.13	21.39	0.81 (0.39 to 1.66)
Individual types of GLP-1RA‡
Liraglutide	522	30	1846.68	151.25	28.27	19.83	0.64 (0.34 to 1.19)
Dulaglutide	522	15	1846.68	89.84	28.27	16.70	0.64 (0.36 to 1.15)
Semaglutide	497	2	1671.89	25.94	29.73	7.71	0.28 (0.07 to 1.16)
Exenatide	497	17	1671.89	71.36	29.73	23.82	0.63 (0.25 to 1.61)
Lixisenatide	273	4	771.14	5.93	35.40	67.49	2.44 (0.67 to 8.87)
Excluding deceased cases during study period	515	77	1850.60	381.40	27.83	20.19	0.66 (0.45 to 0.98)
Analysis of only first record of GLP-1RA exposure with different exposure durations§
Total day supply of first prescription with 30-day grace period	227	10	847.32	77.49	26.79	12.90	0.64 (0.33 to 1.24)
A 730-day follow-up from start date of first prescription	227	23	887.41	110.24	25.58	20.86	1.06 (0.68 to 1.69)

^*Bold incident rate ratios indicate statistically significant treatment effect associated with GLP-1RA use versus non-use.

^†I2 value of 1% for the pooled incident rate ratio suggests a low level of heterogeneity in treatment effects across the individual sites.

^‡Due to no suicide attempts or self-harm occurred in the time intervals of lixisenatide use in the UK, and semaglutide, exenatide and lixisenatide use in Taiwan, the pooled analyses of liraglutide and dulaglutide included the results from Hong Kong, Taiwan and the UK, and the pooled analyses of semaglutide and exenatide only included the results from Hong Kong and the UK, and the result of lixisenatide was only from Hong Kong.

^§Given the limited cases at the Taiwan site, only data from Hong Kong and the UK were analysed.

GLP-1RAs, glucagon-like peptiode-1 receptor agonists; pys, person-years.

Discussion

To respond to the safety concerns regarding GLP-1RA treatment raised by health authorities worldwide, we performed a multicountry study based on several large-scale population-based databases from HK, Taiwan and the UK to assess the risk of suicide attempt or self-harm associated with the use of GLP-1RAs. No increased suicide attempt or self-harm risk with the use of GLP-1RAs versus non-use was found, irrespective of age, gender and the type of GLP-1RA. This was confirmed by sensitivity analyses that excluded deceased cases and focused on the first GLP-1RA prescription. We also found that the suicide attempt or self-harm risk of a patient was highest before exposure to GLP-1RA treatment and gradually declined throughout the GLP-1RA treatment, with a significantly reduced risk observed following the use of GLP-1RAs for 6 months (181 days). These findings show no increase in the risk of suicide attempts or self-harm following GLP-1RA treatment, and the safety of long-term GLP-1RA use is also assured, thereby supporting rational use of GLP-1RAs in routine care settings.

Comparison between results of present study and previous studies

Previous studies investigated the association of GLP-1RA treatment and self-harm, suicide ideation, suicide attempt and completed suicide events, reporting inconclusive results (online supplemental table 1). Specifically, some of these studies were pharmacovigilance studies that used data collected from spontaneous reporting systems for adverse drug events.⁵,⁸ They showed that the signs of composite risks of suicide, suicide attempt, suicidal ideation, self-injurious ideation, suicidal behaviours and self-injurious behaviours associated with GLP-1RA use varied by the comparator of interest. That is, an increased reporting OR of composite suicide-related events between individual GLP-1RAs (eg, reporting OR (95% CI) of liraglutide vs dulaglutide: 3.98 (2.73 to 5.82))⁵ and increased reporting ORs of GLP-1RAs compared with those of metformin/insulin (eg, liraglutide: 3.26 (2.53 to 4.22)/3.62 (2.78 to 4.72))⁷ were found, but there was no statistically significant signal of disproportionate reporting of GLP-1RA versus other drugs (eg, liraglutide: information component 025 = −1.18).⁷ Despite the observed signal, the causality between GLP-1RA exposure and suicide-related events cannot be corroborated through pharmacovigilance studies. A meta-analysis of clinical trials reported non-significant composite outcome (completed or attempted suicide, occurrences of suicidal ideation or self-harm) rate ratios (95% CIs) of 0.76 (0.43 to 1.32) and 0.77 (0.34 to 1.75)¹⁵ for patients with T2D and obesity, respectively, and a post hoc study using the STEP 1, 2, 3 and 5 trials observed the limited suicide ideation events during study periods (ie, semaglutide vs placebo: 0.4%–0.7% versus 0.6%–1.4%).¹² The remaining cohort studies were from the USA^{10 11} and the UK.^{9 13 14} They reported either a protective or neutral effect of GLP-1RA treatment on suicide-related events or self-harm. The HRs (95% CIs) of incident suicidal ideation associated with the use of semaglutide versus other glucose-lowering agents/other anti-obesity drugs were 0.27 (0.20 to 0.36)/0.36 (0.25 to 0.53),¹⁰ and that of depression, self-harm events or suicidal behaviours with GLP-1RAs versus other oral glucose-lowering agents (oGLAs; that is, sulfonylureas, dipeptidyl peptidase-4 inhibitors and sodium-glucose cotransporter-2 inhibitors) ranged between 0.94 (0.71 to 1.24)¹¹ and 1.25 (0.63 to 2.50),⁹ and that of suicidality with GLP-1RAs versus oGLAs ranges from 0.91 (0.73 to 1.12) to 1.02 (0.85 to 1.23).¹⁴ Neutral findings for using semaglutide were also revealed,¹³ showing that the HRs (95% CIs) of semaglutide versus sitagliptin/empagliflozin were 0.75 (0.52 to 1.07)/0.96 (0.66 to 1.40), except for the comparison with glipizide (0.54 (0.36 to 0.83)). However, such studies⁹,¹¹ are prone to bias due to unmeasured confounding that arises from unmeasured baseline characteristics (eg, lifestyle, health awareness and healthy behaviours), affecting the validity of their findings. In addition, the previous findings do not indicate the period during which a patient treated with GLP-1RAs is at risk for suicide attempts or self-harm, which is of clinical importance to avoid potential undesirable side effects during treatment. Although a previous case-time control study,¹⁶ a type of within-subject studies accounting for the time-invariant confounders, indicated the protective findings, with OR (95% CI) of 0.62 (0.51 to 0.75) for risk (30 days before the suicide-related events) versus reference periods (either 90–120, 121–150 or 151–180 days before the suicide-related events), which assured no increased risks of suicidal behaviours with the short-term GLP-1RA use, the long-term effects of GLP-1RAs on suicide-related events remain unclear. Against these previous studies, we applied the SCCS design, which also allows for the adjustment of time-invariant confounders (both measurable and unmeasured characteristics) and a comparison of the risk in various time windows of treatment for a given person (ie, before and after treatment). Importantly, due to cross-national variations in coding practices, differences in healthcare documentation and the low frequency of specific suicidal behaviours, we adopted a composite outcome definition encompassing both suicide attempts and self-harm. This approach, while aligned with prior pharmacoepidemiologic research,¹¹¹⁴,¹⁶ enabled harmonised cross-site analyses while acknowledging limitations in disentangling individual suicidal event types. In the mean follow-up periods of 6.84–14.42 years across three study sites, our study found no increased risk of suicide attempts or self-harm during GLP-1RA treatment and a 33% risk reduction based on the pooled IRR estimates in the primary analyses (table 1), ensuring the safety of GLP-1RA use in a long-term period.

Variation of suicide attempt risk with treatment duration and study site

A decline in the risk of suicide attempt or self-harm following GLP-1RA treatment compared with the baseline before treatment initiation (ie, pretreatment period) was found in our analysis that stratified the treatment period into multiple time intervals. That is, compared with the non-treatment period, the suicide attempt or self-harm risk of a patient was the highest (ie, IRR (95% CI): 1.94 (0.86 to 4.37); table 2) in the pretreatment period (ie, in the 14 days before GLP-1RA initiation) and then gradually decreased throughout the treatment. The suicide attempt or self-harm risk was significantly reduced after approximately 6 months (181 days) of GLP-1RA treatment (ie, decrease of 37%; 0.63 (0.46 to 0.87)). Nevertheless, in contrast to the results from HK (ie, 0.56 (0.32 to 1.00); table 1) and the UK (ie, 0.67 (0.49 to 0.92)), those from Taiwan showed an insignificant increase in the risk of suicide attempt or self-harm (ie, 1.49 (0.44 to 5.06)). These findings might be related to different treatment durations of GLP-1RAs across the individual study sites. The mean treatment duration (SD) of GLP-1RAs in Taiwan was 0.40 years (0.34), which is shorter than those for HK (1.64 years (1.52)) and the UK (2.12 years (2.31)). The heterogeneous treatment effect of GLP-1RAs with treatment duration has been previously discussed,²⁴ with increased anxiety-like and decreased depression-like behaviours caused by short and long central GLP-1 activation times, respectively.²⁴ This biological mechanism might explain our findings, suggesting a possibly increased risk of suicide attempt or self-harm from a shorter treatment duration (ie, in Taiwan) but a reduced risk following a longer treatment duration (ie, in HK or the UK). Hence, the overall results imply that the long-term use of GLP-1RAs is associated with lowering depression-like behaviours such as suicide attempts or self-harm. The potential psychological or emotional side effects should be closely monitored at the start of treatment or during short-term use, and mental health assessment and care are encouraged while undergoing long-term GLP-1RA treatment, especially for vulnerable patients at risk of suicide.

Consistent findings across subgroup and sensitivity analyses

Among the series of subgroup and sensitivity analyses, only the analysis of cases aged≥50 years and the analysis that excluded deceased cases during the study period showed consistent results with the primary findings. Compared with non-treatment, the use of GLP-1RAs was associated with a significantly reduced suicide attempt or self-harm risk, namely 48% (ie, IRR (95% CI): 0.52 (0.32 to 0.84)) and 35% (0.65 (0.45 to 0.92)), respectively. The remaining subgroup/sensitivity analyses showed a neutral effect of GLP-1RA treatment. In contrast to the increased risks of suicide attempt and suicide with age have been reported in systematic reviews,²⁵ which emphasises the importance of close monitoring and timely intervention for elderly populations. Our findings revealed a decrease in the risk of suicide attempt or self-harm with GLP-1RA use for cases aged≥50 years, relieving concerns regarding the potential of additional risk of suicide attempt or self-harm contributed by GLP-1RA treatment. In addition, a neutral effect of GLP-1RAs on suicide attempts or self-harm was observed for young, male and female populations. Also, while targeting different types of GLP-1RAs, consistent findings were revealed with the pooled IRRs (95% CIs) of 0.63 (0.25 to 1.61)/2.44 (0.67 to 8.87)/0.64 (0.34 to 1.19) for exenatide/lixsenatide/liraglutide (daily use), respectively, and 0.28 (0.07 to 1.16)/0.64 (0.36 to 1.15) for semaglutide/dulaglutide (weekly use), respectively. Though the pooled IRR of lixsenatide seemed higher than others, its wide CI covering 1.0 indicated the non-significant association. Hence, the safety of GLP-1RA treatment corroborated in this study supports the rational use of GLP-1RAs, irrespective of types, for real-world populations with diverse clinical characteristics. Furthermore, to avoid the potential censoring as a result of event occurrence (ie, censoring due to an increased mortality risk following suicide attempts or self-harm) and allow the likelihood of repeated exposure (ie, multiple GLP-1RA prescriptions) as two assumptions of SCCS, a sensitivity analysis that excluded cases that deceased during the study period was performed. Its results were consistent with the primary findings, thereby supporting the robustness of the study results.

Limitations

First, although the SCCS design was employed to carefully handle the effects of intra-person variability and eliminate time-invariant confounding effects for a given patient, the concern of measured or unmeasured time-varying confounders (eg, depression) may remain.²⁶ With this regard, the calendar time has been measured (quantified using the ‘age’ variable) and adjusted in the Poisson regression to account for the potential of time-varying confounding effect. Yet, the estimated E-value suggests that the study results would be nullified if the association of depression and suicide attempts or GLP-1RA use was greater than 2.35 or 1.53. However, the OR of having depressive symptoms versus no depressive symptoms on suicide attempts or self-harm reported in previous meta-analyses was only 1.30.²⁷ In addition, existing evidence suggests a statistically significant decrease in the depression rating scale associated with GLP-1RA use.²⁸ Low proportions of the study patients had major depressive disorder, namely 4%, 10% and 3% in HK, Taiwan and the UK, respectively. Considering the above, the potential impact of unmeasured depression on suicide attempts or self-harm in the present study is likely negligible. Second, given that the baseline suicide attempt or self-harm risk is affected by comorbidities, such as alcohol abuse,²⁹ major depressive disorder²⁹ and microvascular or macrovascular diseases, further research on these subsets of patients is warranted. Third, given that only the databases in relatively early years were available across study sites, the overall number of cases was small (ie, 642) and limited to exenatide, liraglutide and dulaglutide. Particularly, the cases in the Taiwan study site were few (ie, 30), leading to the wide CIs in the primary analyses. Future research using more recent data and including newer GLP-1RAs (ie, semaglutide and tirzepatide) is warranted to corroborate our findings. Fourth, as in previous studies,¹¹¹⁴,¹⁶ the use of a composite outcome for suicidal events in this study may limit the specificity of our findings for any individual type of suicidal behaviour or self-harm. Nevertheless, conducting stratified analyses is challenged by the limited case counts for specific event types, cross-country variability in operational definitions and clinical documentation, and the lack of validated measures for each event category. Future large-scale studies, designed to overcome these constraints, are warranted to delineate the effects of GLP-1RA therapy on specific suicidal or psychiatric outcomes. Fifth, the lower precision and higher susceptibility of the SCCS design should be acknowledged.³⁰ Lastly, given the SCCS design, only relative incidence ratios of suicide attempts could be estimated, and thus, the findings should be interpreted with caution. Future well-controlled clinical trials or pragmatic trials remain warranted to corroborate our findings and determine the causal inference.

Conclusions

The long-term use of GLP-1RAs was not associated with an increased risk of suicide attempt or self-harm in this multicountry (HK, Taiwan and the UK) study. Close monitoring at the start of treatment or in the initial treatment phase is suggested to avoid potential undesirable side effects such as suicide attempts or self-harm.

Data availability statement

Data may be obtained from a third party and are not publicly available.