Hostname: page-component-6898cfcb55-zkb6p Total loading time: 0 Render date: 2026-11-11T10:45:26.579Z Has data issue: false hasContentIssue false
  • English
  • Français

Sodium-glucose cotransporter 2 inhibitors and risk of self-harm in patients with type 2 diabetes mellitus: an active comparator, new user cohort study in two independent cohorts

Published online by Cambridge University Press:  09 November 2026

Yuhao LiOpen the ORCID record for Yuhao Li[Opens in a new window],
Huatang Zeng,
Liqun Wu,
Xiatong Ke,
Peifen Li,
Yongfeng Song,
Houyu Zhao,
Shengfeng WangOpen the ORCID record for Shengfeng Wang[Opens in a new window] and
Fang Du
Yuhao Li
Affiliation:
Department of Epidemiology and Biostatistics, School of Public Health, Peking University, Key Laboratory of Epidemiology of Major Diseases, Ministry of Education , Beijing, China
Huatang Zeng
Affiliation:
Shenzhen Health Development Research and Data Management Center , China
Liqun Wu
Affiliation:
Shenzhen Health Development Research and Data Management Center , China
Xiatong Ke
Affiliation:
Shenzhen Health Development Research and Data Management Center , China
Peifen Li
Affiliation:
Shenzhen Health Development Research and Data Management Center , China
Yongfeng Song
Affiliation:
Key Laboratory of Endocrine Glucose & Lipids Metabolism and Brain Aging, Ministry of Education, Shandong First Medical University , Jinan, Shandong, ChinaDepartment of Endocrinology, Central Hospital Affiliated to Shandong First Medical University, Jinan, Shandong, China
Houyu Zhao
Affiliation:
Research Center of Clinical Epidemiology, Peking University Third Hospital , Beijing, China
Shengfeng Wang*
Affiliation:
Department of Epidemiology and Biostatistics, School of Public Health, Peking University, Key Laboratory of Epidemiology of Major Diseases, Ministry of Education , Beijing, ChinaInstitute for Artificial Intelligence, Peking University, Beijing, China
Fang Du
Affiliation:
Shenzhen Health Development Research and Data Management Center , China
*
Corresponding author: Shengfeng Wang; Email: shengfeng1984@126.com
Rights & Permissions [Opens in a new window]

Abstract

Background

Although type 2 diabetes mellitus (T2DM) has twice the self-harm risk of the general population, the role of sodium-glucose cotransporter 2 inhibitors (SGLT-2i) is unclear; this study aimed to assess whether SGLT-2i use is linked to lower self-harm risk in T2DM.

Methods

Through the Shenzhen Public Health Data Platform, we constructed two cohorts (SGLT-2i vs. dipeptidyl peptidase-4 inhibitors [DPP-4i], and SGLT-2i vs. glucagon-like peptide-1 receptor agonists [GLP-1RA]) via a target trial emulation framework. Hazard ratios (HRs) and 95% confidence intervals (CIs) were estimated using overlapping weighting Cox regression models.

Results

The SGLT-2i versus DPP-4i cohort included 85,789 SGLT-2i users and 53,810 DPP-4i users. After a median follow-up of 1.85 years (interquartile range: 0.91–2.70), 567 (0.66%) cases of self-harm occurred in the SGLT-2i group and 567 (1.05%) in the DPP-4i group. After overlapping weighting, SGLT-2i was associated with a lower risk of self-harm compared with DPP-4i (HR: 0.86, 95% CI: 0.76–0.98), and this result was consistent across various sensitivity analyses. In the SGLT-2i (n = 130,330) versus GLP-1RA (n = 5,126) cohort, a lower risk of self-harm (HR: 0.76, 95% CI: 0.59–0.99) was associated with SGLT-2i compared with GLP-1RA. However, in some sensitivity analyses, no statistically significant difference was observed between SGLT-2i and GLP-1RA users.

Conclusions

SGLT-2i use is associated with a lower risk of self-harm compared with DPP-4i and GLP-1RA. These findings provide supportive evidence for the safety of SGLT-2i and highlight their potential role in the mental health management of T2DM patients.

Keywords

glucagon-like peptide-1 receptor agonistsself-harmsodium-glucose cotransporter 2 inhibitorstarget trial emulationtype 2 diabetes mellitus

Information

Type
Original Article
Information
Psychological Medicine , Volume 56 , 2026, e360
Check for updates
Creative Commons
Creative Common License - CCCreative Common License - BYCreative Common License - NCCreative Common License - ND
This is an Open Access article, distributed under the terms of the Creative Commons Attribution-NonCommercial-NoDerivatives licence (http://creativecommons.org/licenses/by-nc-nd/4.0), which permits non-commercial re-use, distribution, and reproduction in any medium, provided that no alterations are made and the original article is properly cited. The written permission of Cambridge University Press or the rights holder(s) must be obtained prior to any commercial use and/or adaptation of the article.
Copyright
© The Author(s), 2026. Published by Cambridge University Press

Introduction

Self-harm is defined as any act of intentional self-poisoning or self-injury, irrespective of apparent intent, encompassing behaviors such as drug overdose, ingestion of non-ingestible substances, cutting, burning, or mechanical injury (Moran et al., Reference Moran, Chandler, Dudgeon, Kirtley, Knipe, Pirkis and Christensen2024). Globally, an estimated 14 million self-harm events occur annually, although this figure is likely a substantial underestimate, especially in low- and middle-income countries (LMICs), where systematic surveillance is often lacking (Knipe et al., Reference Knipe, Padmanathan, Newton-Howes, Chan and Kapur2022; Moran et al., Reference Moran, Chandler, Dudgeon, Kirtley, Knipe, Pirkis and Christensen2024). The health and economic burden of self-harm is considerable. In 2019, 34 million disability-adjusted life years (DALYs) lost due to self-harm were associated with an estimated global cost of US$639 billion, 81% of which was borne by LMICs (Moran et al., Reference Moran, Chandler, Dudgeon, Kirtley, Knipe, Pirkis and Christensen2024).

Diabetes mellitus, a metabolic disorder primarily characterized by elevated fasting glucose, affects approximately 529 million individuals globally, 96% of whom have type 2 diabetes mellitus (T2DM) (GBD 2021 Diabetes Collaborators, 2023). T2DM patients have approximately twice the risk of self-harm compared to the general population (Conti et al., Reference Conti, Mennitto, Di Francesco, Fraticelli, Vitacolonna and Fulcheri2017; Myers, Grannemann, Lingvay, & Trivedi, Reference Myers, Grannemann, Lingvay and Trivedi2013). Sodium-glucose cotransporter 2 inhibitors (SGLT-2i) represent a novel class of oral antihyperglycemic agents that reduce blood glucose by inhibiting SGLT-2 and have demonstrated significant cardiorenal benefits in diabetes patients (American Diabetes Association Professional Practice Committee 2024b). Both SGLT-2i and glucagon-like peptide-1 receptor agonists (GLP-1RA) are commonly used at similar stages of disease progression (Shapiro et al., Reference Shapiro, Yin, Yu, Rej, Suissa and Azoulay2025). Since the Icelandic Medicines Agency issued a safety alert regarding a potential link between GLP-1RA and self-harm, numerous studies have explored this association (Gamble et al., Reference Gamble, Chibrikov, Midodzi, Twells and Majumdar2018; Hurtado et al., Reference Hurtado, Robles, Peiró, García-Sempere and Sanfélix-Gimeno2024; Shapiro et al., Reference Shapiro, Yin, Yu, Rej, Suissa and Azoulay2025; Tang et al., Reference Tang, Lu, Donahoo, Shao, Shi, Fonseca, Guo, Bian and Guo2024; Ueda et al., Reference Ueda, Söderling, Wintzell, Svanström, Pazzagli, Eliasson and Pasternak2024). Although randomized controlled trials have preliminarily examined the association between SGLT-2i and self-harm (Lu et al., Reference Lu, Tang, Kotecha, Cho, Donahoo, Bian and Guo2024; Wadden et al., Reference Wadden, Brown, Egebjerg, Frenkel, Goldman, Kushner and Fink-Jensen2024), they were constrained by limited sample sizes, few outcome events, and short follow-up periods, underscoring the need for large-scale observational studies to validate this association in real-world populations. Moreover, evidence on this topic remains scarce, particularly in LMICs.

Leveraging a target trial emulation framework with an active comparator, new user design, the aims of this investigation are to evaluate the association between SGLT-2i use and the incidence of self-harm among T2DM patients and explore potential effect modification by age, sex, and diabetes severity.

Methods

Data sources

Data for this study were extracted from the Shenzhen Public Health Data Platform (Huang et al., Reference Huang, Ke, Liu, Liu, Li, Feng and Di2025) developed by the Shenzhen Health Development Research and Data Management Center. This platform integrates multidimensional health-related resources – including electronic health records (EHRs), electronic medical records (EMRs), comprehensive demographic data, and healthcare service resources – for more than 40 million residents of Shenzhen. It interconnects 75 public hospitals and public health institutions, 106 private hospitals, all community health service centers, and over 1,600 clinics across Shenzhen. Currently, the platform hosts more than 40 million patient EHRs and over 50 billion operational data entries.

The study was approved by the Ethics Review Board. The requirement for informed consent was waived because all data were de-identified.

Study design and population

Applying a target-trial emulation framework with an active comparator, new user design (Lund, Richardson, & Stürmer, Reference Lund, Richardson and Stürmer2015) (Supplementary Figure S1 and Table S1), we assembled two independent cohorts. In Cohort 1, new users of SGLT-2i (including dapagliflozin, empagliflozin, canagliflozin, ertugliflozin, and henagliflozin) constituted the exposure group and new users of dipeptidyl peptidase-4 inhibitors (DPP-4i, including sitagliptin, saxagliptin, vildagliptin, alogliptin, linagliptin, teneligliptin, trelagliptin, and retagliptin) the comparator; in Cohort 2, new users of GLP-1RA (including semaglutide, exenatide, liraglutide, beinaglutide, lixisenatide, and dulaglutide) served as the comparator. Active comparators included GLP-1RA and DPP-4i, which, like SGLT-2i, are commonly employed as second-line options following metformin failure and are administered to patients with comparable disease severity (Chinese Diabetes Society, 2025). The index date was defined as the date of the first prescription for either SGLT-2i or DPP-4i/GLP-1RA, following a 12-month washout period after the diagnosis of T2DM. Among 36,247,311 residents with at least one diagnostic record in the platform between 1 January 2019 and 1 October 2024, all diabetic patients were considered eligible for inclusion. We excluded patients who (1) did not fulfill the 12-month washout criterion, (2) had missing age information or were aged < 18 years at the index date, (3) initiated the comparator drug within 60 days after first dispensing the exposure drug, or (4) had a history of self-harm before the index date. After applying these criteria, 139,599 eligible participants were included in Cohort 1 and 135,456 in Cohort 2 (Supplementary Figure S2).

The reporting of this study aligns with the TARGET statement (Cashin et al., Reference Cashin, Hansford, Hernán, Swanson, Lee, Jones and McAuley2025).

Outcome and follow-up

The primary outcome was self-harm, ascertained using International Classification of Diseases (ICD)-10 codes X60–X84 and Y87.0 (Shapiro et al., Reference Shapiro, Yin, Yu, Rej, Suissa and Azoulay2025). ICD-10 codes were sourced from inpatient, emergency department, and outpatient settings, capturing both primary and secondary diagnoses. Follow-up commenced on the index date and continued until the earliest of self-harm, death, or 1 October 2024.

Covariates

Covariates were selected on the basis of clinical plausibility and prior evidence(Hurtado et al., Reference Hurtado, Robles, Peiró, García-Sempere and Sanfélix-Gimeno2024; Shapiro et al., Reference Shapiro, Yin, Yu, Rej, Suissa and Azoulay2025; Tang et al., Reference Tang, Lu, Donahoo, Shao, Shi, Fonseca, Guo, Bian and Guo2024; Ueda et al., Reference Ueda, Söderling, Wintzell, Svanström, Pazzagli, Eliasson and Pasternak2024). Baseline covariates included demographic characteristics (age, sex, marriage, etc.), comorbidities (diabetes-related complications, depression, cardiovascular disease, etc.), and concomitant medications (other glucose-lowering agents, antihypertensive drugs, lipid-lowering agents, etc.), as detailed in Table 1. Covariates were measured within the 12 months prior to the index date, and the most recent available measurement was used for each individual. The ICD-10 codes corresponding to comorbidities are presented in Supplementary Table S2.

Table 1.

Baseline characteristics of cohort SGLT-2i versus DPP-4i

A detailed table comparing baseline characteristics of patient cohorts treated with S G L T-2i versus D P P-4i, showing unweighted and weighted data with standardized mean differences. See long description.
Table 1. Long description

The table is organized into seven columns: Variable, Unweighted patients (subdivided into D P P-4i and S G L T-2i), S M D, and Weighted patients (subdivided into D P P-4i and S G L T-2i), followed by a final S M D column.

Key data points include:

* Age (years): Unweighted D P P-4i is 59.74 plus or minus 12.77; S G L T-2i is 56.81 plus or minus 12.29. Weighted values for both are 58.5 plus or minus approximately 12.5.

* Males: Unweighted D P P-4i is 29,670 (55.14%); S G L T-2i is 55,790 (65.03%). Weighted values for both are 17,927.57 (59.04%).

* Body mass index (kg/m super 2): Unweighted D P P-4i is 24.55 plus or minus 2.43; S G L T-2i is 24.98 plus or minus 2.67. Weighted values for both are 24.69 plus or minus approximately 2.5.

* Duration of diabetes (years): Unweighted D P P-4i is 2.60 plus or minus 1.15; S G L T-2i is 2.74 plus or minus 1.16. Weighted values for both are 2.65 plus or minus approximately 1.15.

* Comorbidities: Includes detailed counts and percentages for Diabetic nephropathy, retinopathy, peripheral neuropathy, mental disorders (depression, Alzheimer’s, anxiety, schizophrenia), cardiovascular diseases (myocardial infarction, heart failure, stroke), and liver diseases.

* Medications: Lists usage for Biguanides, Sulfonylureas, alpha-Glucosidase inhibitors, Insulin, and various antihypertensive and lipid-lowering drugs.

In the weighted patient columns, the S M D for almost all variables is less than 0.001, indicating successful balancing of the cohorts after weighting.

Note: DPP-4i, dipeptidyl peptidase-4 inhibitors; GLP-1RA, glucagon-like peptide-1 receptor agonist; SGLT-2i, sodium-glucose cotransporter 2 inhibitors; SMD, standardized mean difference.

Statistical analysis

Continuous variables were presented as mean ± standard deviation (SD), and categorical variables were reported as frequencies and percentages. Covariate balance between exposure and comparator groups was assessed using the standardized mean difference (SMD), with an absolute SMD < 0.10 indicating adequate balance (Austin & Stuart, Reference Austin and Stuart2015). The incidence rate of self-harm was calculated as the number of events per 1,000 person-years. Cumulative incidence was estimated using the weighted Kaplan–Meier curve. Under the intention-to-treat framework, overlapping weighting Cox proportional hazards models were fitted to quantify the association between SGLT-2i initiation and self-harm occurrence. The proportional hazards assumption was evaluated by including an interaction term between treatment and time in the model, and the assumption was satisfied for all exposure-outcome comparisons. Propensity scores were derived via logistic regression to estimate the probability of assignment to each group. The overlapping weights were calculated for every patient using the standard formula (Xie et al., Reference Xie, Bowe, Xian, Loux, McGill and Al-Aly2023):

$$ W\left(t,X\right)=\frac{1/P\left(A=t/X\right)}{\sum_{g=1}^21/P\left(A=g/X\right)} $$ WtX equals 1 slash PA equals t slash upper X sigma summation g equals 121 slash PA equals g slash upper X

where W denotes the overlap weight, t indicates the treatment actually received, X is the vector of baseline covariates, P is the propensity score estimated from the multinomial logistic regression model, and g indexes one of the treatment groups. Estimation of confidence intervals was performed using robust standard errors.

Sensitivity analyses were conducted to assess result stability. (a) Self-harm was redefined using ICD-10 codes X60–X84 and Y10–Y34 (Ueda et al., Reference Ueda, Söderling, Wintzell, Svanström, Pazzagli, Eliasson and Pasternak2024). (b) Three sensitivity analyses were conducted to evaluate the impact of combination therapy by excluding patients with overlapping prescriptions during defined grace periods: (i) same-day initiation; (ii) within 30 days; and (iii) within 90 days of the first SGLT-2i and the first DPP-4i/GLP-1RA prescription. (c) Extreme weights were trimmed by omitting participants below the 1st or above the 99th percentile of the overlapping weighting distribution. (d) We repeated the primary analysis using inverse-probability-of-treatment weighting to explore the impact of different weighting methods on the results. (e) Death was treated as a competing event, and the association was analyzed using the Fine–Gray subdistribution hazard model. (f) We addressed the potential confounding effect of varying medication duration of SGLT-2i and DPP-4i/GLP-1RA by including it as a covariate in the calculation of the propensity score. (g) Laboratory parameters, including glycated hemoglobin, creatinine, lipid profile (total cholesterol, triglycerides, high-density lipoprotein cholesterol, and low-density lipoprotein cholesterol), liver function tests (alanine aminotransferase, aspartate aminotransferase, and gamma-glutamyl transferase), uric acid, and direct bilirubin, were incorporated into weight calculation. These measurements were obtained on the index date or within the preceding year, though approximately 80% of values were missing. To address this, we performed multiple imputation under the missing-at-random assumption using the chained equations approach. The imputation model included all covariates from the final multivariable analyses to ensure valid inference. Five independent imputed datasets were generated, and the main analyses were conducted separately on each. Results were pooled via Rubin’s rules to yield final estimates and 95% confidence intervals, accounting for within- and between-imputation variance. All imputation procedures were performed using the ‘mice’ package (Version 3.16.0). (h) To reduce the risk of reverse causation, we conducted sensitivity analyses by (1) performing lag analyses with exposure delays of 0.5 and 1 year, and (2) excluding those with documented cancer or psychiatric histories (such as depression, bipolar disorder, etc.). (i) Patients with a history of self-harm were retained in the analysis, and this variable was included in the calculation of the weights. (j) The exposure group was restricted to new users of dapagliflozin, which accounted for approximately 90% of SGLT-2i in Shenzhen. (k) To assess the potential impact of residual confounding, we calculated the E value and performed a positive control analysis using genital tract infection (GTI) as the outcome.

To evaluate potential heterogeneity in the association between SGLT-2i initiation and the incidence of self-harm, we performed pre-specified subgroup analyses stratified by age (< 65 years, ≥ 65 years), sex (male, female), diabetes duration (< median duration of 2.5 years, ≥ median duration of 2.5 years), presence of diabetes complications (yes, no), and concomitant use of other antidiabetic medications (yes, no). The analytical approach involved re-estimating weights within each subgroup. Interaction P values were derived from weighted Cox proportional hazards models with interaction terms.

Statistical analysis was performed using R (Version 4.3.3). All statistical tests were two-sided, with P < 0.05 considered to indicate statistical significance.

Role of the funding source

The funding agency had no role in the design of the study; in the collection, analysis, or interpretation of data; in the writing of the manuscript; or in the decision to publish the results.

Results

Baseline character

In the cohort comparing SGLT-2i with DPP-4i, a total of 139,599 T2DM patients were enrolled, with a mean age of 57.94 ± 12.56 years, and 85,460 (61.22%) were male. The cohort comprised 85,789 (61.45%) users of SGLT-2i and 53,810 (38.55%) users of DPP-4i. In the raw cohort, users of DPP-4i were older, had a lower rate of metformin use, exhibited lower body mass index and diastolic blood pressure levels, and had a shorter duration of diabetes (Table 1).

The analysis of the SGLT-2i versus GLP-1RA cohort included 135,456 patients, with an average age of 57.39 ± 12.27 years, and was divided into 130,330 (96.22%) patients receiving SGLT-2i and 5,126 (3.78%) receiving GLP-1RA. In the unweighted analysis, patients treated with GLP-1RA were younger and had a lower frequency of sulfonylurea and meglitinide use, yet a higher prevalence of insulin therapy. Additionally, GLP-1RA users presented with higher prevalence of diabetic peripheral neuropathy, peripheral vascular disease, and fatty liver disease (Table 2).

Table 2.

Baseline characteristics of cohort SGLT-2i versus GLP-1RA

A detailed table comparing baseline characteristics of G L P-1 R A and S G L T-2 i cohorts before and after weighting. See long description.
Table 2. Long description

The table presents baseline characteristics for two cohorts: G L P-1 R A and S G L T-2 i. It is divided into Unweighted and Weighted patient sections, each with columns for both drug types and the Standardized Mean Difference (S M D).

Key variables include:

* Age (years): Unweighted G L P-1 R A is 53.59 plus or minus 14.35; S G L T-2 i is 57.54 plus or minus 12.15 (S M D 0.294). Weighted values for both are 53.95 (S M D less than 0.001).

* Males: Unweighted both at 55.68 percent. Weighted both at 56.58 percent.

* Body mass index (kg/m super 2): Unweighted G L P-1 R A is 25.29 plus or minus 2.45; S G L T-2 i is 24.92 plus or minus 2.58. Weighted both are 25.26.

* Diabetic complications: Includes nephropathy, retinopathy, and peripheral neuropathy. In the weighted cohort, all complications show an S M D less than 0.001, indicating balanced groups.

* Mental disorders: Includes depression (2.63 percent weighted), anxiety (3.71 percent weighted), and schizophrenia (2.25 percent weighted).

* Cardiovascular disease: Includes myocardial infarction, heart failure (7.46 percent weighted), and ischemic stroke (8.85 percent weighted).

* Antidiabetic medications: Lists Biguanides, Sulfonylureas, D P P-4 i, and Insulin. Weighted cohorts are balanced at an S M D less than 0.001 for most, including Insulin at 28.67 percent.

* Other medications: Includes antihypertensives (A C E inhibitors, Beta-blockers, Calcium channel blockers), lipid-lowering medications (Statins at 26.86 percent weighted), and antiplatelets (18.71 percent weighted).

Note: SGLT-2i, sodium-glucose cotransporter 2 inhibitors; DPP-4i, dipeptidyl peptidase-4 inhibitors; GLP-1RA, glucagon-like peptide-1 receptor agonist; SMD, standardized mean difference.

After applying overlapping weighting, all baseline covariates were well balanced between the two cohorts (Tables 1 and 2).

Comparison of self-harm occurrence in T2DM patients treated with SGLT-2i versus DPP-4i

After a median follow-up of 1.85 years (interquartile range [IQR]: 0.91–2.70), in Cohort 1, self-harm occurred in 567 (1.05%) and 567 (0.66%) users of DPP-4i and SGLT-2i. The weighted incidence rates were 4.62 (95% confidence interval [CI]: 4.09–5.19) and 4.04 (95% CI: 3.52–4.60) per 1,000 person-years for DPP-4i and SGLT-2i users, respectively. Cumulative incidence of self-harm during follow-up is depicted in Supplementary Figure S3. Adjustment for confounders using overlapping weighting revealed that SGLT-2i was associated with a lower risk of self-harm compared to DPP-4i users (hazard ratio [HR]: 0.86, 95% CI: 0.76–0.98), a finding that was consistently stable across a range of sensitivity analyses (Figure 1). Additionally, analysis specific to dapagliflozin showed a 13% reduction in the risk of self-harm compared to DPP-4i (HR: 0.87, 95% CI: 0.77–0.99). The primary analysis corresponded to an E value of 1.60, and the positive control analysis indicated that SGLT-2i was associated with a higher risk of GTI (HR: 1.31, 95% CI: 1.21–1.42).

Figure 1.

Primary and sensitivity analysis of the association between SGLT-2i and the incidence of self-harm. Note: A, SGLT-2i versus DPP-4i; B, SGLT-2i versus GLP-1RA. The number of outcome events, incidence rate, HR, and its 95%CI represented the results after overlapping weighting. The primary analysis defines self-harm according to ICD-10 as X60–X84 and Y87.0, while the redefined self-harm is defined as X60–X84 and Y10–Y34. Combination therapy was defined as the scenario where patients in the exposure group received the comparator drug, and those in the comparator group received the exposure drug within a specified time window. Extreme weights are defined as overlapping weights less than the 1st percentile or greater than the 99th percentile. Laboratory tests included glycated hemoglobin, triglycerides, total cholesterol, high-density lipoprotein cholesterol, low-density lipoprotein cholesterol, alanine aminotransferase, aspartate aminotransferase, γ-glutamyl transpeptidase, direct bilirubin, serum uric acid, and serum creatinine. Mental illness included depression, Alzheimer’s disease, other dementias, delirium, anxiety disorders, substance use disorders, bipolar disorder, schizophrenia, obsessive-compulsive disorder, post-traumatic stress disorder, attention-deficit/hyperactivity disorder, borderline personality disorder, eating disorders, sleep disorders, and chronic pain. Abbreviations: CI, confidence interval; DPP-4i, dipeptidyl peptidase-4 inhibitors; HR, hazard ratio; SGLT-2i, sodium-glucose cotransporter 2 inhibitors; T2DM, type 2 diabetes mellitus.

Two forest plots comparing S G L T-2 i with D P P-4 i and G L P-1 R A regarding self-harm incidence across primary and sensitivity analyses. See long description.
Figure 1. Long description

A two-panel forest plot.

Panel A compares S G L T-2 i versus D P P-4 i. The table columns include Case percentage, Incidence rate per 1,000 person-years, H R with 95% C I, and P value. The primary analysis shows an H R of 0.86 with a 95% C I of 0.76 to 0.98 and a P value of 0.019. Sensitivity analyses include redefining self-harm, excluding combination therapy at 1, 30, and 90 days, excluding extreme weights, switching weighting to I P T W, and lag analyses. Most sensitivity analyses show H R values between 0.85 and 0.93, favoring S G L T-2 i, with the exception of the 1-year lag analysis which shows an H R of 0.99. The forest plot visualizes these ratios relative to a vertical null line at 1.0, with most points and error bars shifted to the left toward S G L T-2 i better.

Panel B compares S G L T-2 i versus G L P-1 R A. The primary analysis shows an H R of 0.76 with a 95% C I of 0.59 to 0.99 and a P value of 0.045. Sensitivity analyses follow the same structure as Panel A. Notable results include the exclusion of patients with history of cancer or mental illness showing an H R of 0.63 and the 0.5-years lag analysis showing an H R of 0.72. The forest plot for Panel B shows higher variability in error bar width compared to Panel A, but the majority of point estimates remain to the left of the null line, indicating a lower risk for the S G L T-2 i group.

The association between SGLT-2i and self-harm varied across different age groups (P for interaction = 0.002). Among T2DM patients aged < 65 years, SGLT-2i was not associated with self-harm occurrence (HR: 1.05, 95% CI: 0.88–1.26). However, in patients aged 65 and above, the risk of self-harm was reduced by approximately 28% among SGLT-2i users (HR: 0.71, 95% CI: 0.60–0.85) compared to DPP-4i. Heterogeneity in the study outcomes was not observed across different sexes, diabetes duration, presence of diabetes complications, or use of other antihyperglycemic medications (all P for interaction > 0.05, Figure 2).

Figure 2.

Subgroup analysis of the association between SGLT-2i and the incidence of self-harm. Note: A, SGLT-2i versus DPP-4i; B, SGLT-2i versus GLP-1RA. The number of outcome events, incidence rate, HR, and its 95%CI represented the results after overlapping weighting. Abbreviations: CI, confidence interval; GLP-1RA, glucagon-like peptide-1 receptor agonist; HR, hazard ratio; SGLT-2i, sodium-glucose cotransporter 2 inhibitors; T2DM, type 2 diabetes mellitus.

Two forest plots showing subgroup analyses of self-harm risk comparing S G L T-2 i with D P P-4 i and G L P-1 R A. Most subgroups favor S G L T-2 i with H R values below 1. See long description.
Figure 2. Long description

The image contains two forest plots, labeled A and B.

Panel A compares S G L T-2 i versus D P P-4 i. The columns from left to right are Subgroups, S G L T-2 i Case percentage and Incidence Rate, D P P-4 i Case percentage and Incidence Rate, a forest plot of H R with 95% C I, P value, and P for interaction. Subgroups include Age, Sex, Diabetes duration, Diabetes complications, and Other antidiabetic medications. In the Age subgroup, those under 65 have an H R of 1.05 (0.88 to 1.26), while those 65 and older have an H R of 0.71 (0.60 to 0.85), with a significant P for interaction of 0.002. Most other subgroups show H R values less than 1.0, favoring S G L T-2 i, with points plotted to the left of the vertical null line at 1.0.

Panel B compares S G L T-2 i versus G L P-1 R A. It follows the same column structure. For all subgroups including Age, Sex, and Diabetes duration, the H R values are consistently below 1.0, ranging from 0.69 to 0.97. The forest plot shows all data points and their horizontal confidence interval bars situated to the left of the null line, indicating a lower incidence of self-harm for the S G L T-2 i group compared to the G L P-1 R A group across all categories. P for interaction values in Panel B are all non-significant, ranging from 0.191 to 0.812.

Comparison of self-harm occurrence in T2DM patients treated with SGLT-2i versus GLP-1RA

In the SGLT-2i versus GLP-1RA cohort, users of SGLT-2i experienced a median follow-up of 1.91 years (IQR: 0.97–2.73), during which 992 (0.76%) patients experienced self-harm. In contrast, users of GLP-1RA had a median follow-up of 1.65 years (IQR: 0.88–1.69), with 69 (1.34%) cases of self-harm. In the primary analysis, the use of SGLT-2i was associated with a lower risk of self-harm compared to GLP-1RA (HR: 0.76, 95% CI: 0.59–0.99) and the corresponding E value was 1.96. However, this association was not robust across all sensitivity analyses. Notably, when alternative outcome definitions were applied (HR: 0.95; 95% CI: 0.79–1.15) or when medication duration of SGLT-2i and GLP-1RA was incorporated into the weighting algorithm (HR: 0.91; 95% CI: 0.70–1.19), no significant difference in self-harm risk was observed between SGLT-2i and GLP-1RA users (Figure 1).

In the subgroup analyses, no significant differences in the association between SGLT-2i and GLP-1RA with self-harm occurrence were observed (all P for interaction > 0.05, Figure 2).

Discussion

In this active comparator, new user, target trial emulation study, we observed that among T2DM patients, SGLT-2i use was associated with a lower risk of self-harm compared with DPP-4i use, with a more pronounced protective association observed in older patients. Additionally, SGLT-2i use was also associated with a lower risk of self-harm compared with GLP-1RA use, although this result was not significant in some sensitivity analyses.

Currently, there is only one study that has explored the risk of self-harm associated with SGLT-2i compared to DPP-4i in T2DM patients with comorbid bipolar affective disorder (Chang et al., Reference Chang, Huang, Lo, Huang, Yang and Kornelius2025). Despite observing a lower risk of self-harm occurrence associated with SGLT-2i, the sample size and number of outcome events in that study were relatively limited. Moreover, our study extended the population to general T2DM patients. The mechanisms underlying this potential protective effect remain unclear but may involve several pathways. First, SGLT-2i promote glycosuria and shift metabolism toward increased lipolysis, fatty acid oxidation, and ketogenesis (Ferrannini et al., Reference Ferrannini, Baldi, Frascerra, Astiarraga, Heise, Bizzotto and Muscelli2016). A ketogenic diet, which induces ketone body production as an alternative brain fuel, has been linked to improved mood and cognition (Brietzke et al., Reference Brietzke, Mansur, Subramaniapillai, Balanzá-Martínez, Vinberg, González-Pinto and McIntyre2018). Thus, SGLT-2i-induced ketogenesis might confer benefits against mood disorders and self-harm (Chang et al., Reference Chang, Huang, Lo, Huang, Yang and Kornelius2025). Second, SGLT-2i improve insulin resistance, glycemic control, and weight loss while reducing hypoglycemic events (American Diabetes Association Professional Practice Committee 2024a). Given that poor glycemic control (Renaud-Charest et al., Reference Renaud-Charest, Stoljar Gold, Mok, Kichler, Nakhla and Li2024) and recurrent hypoglycemia (Majumdar et al., Reference Majumdar, Sinha, Dastidar, Gangopadhyay, Ghoshal, Mukherjee and Bhattacharjee2021) are associated with elevated self-harm risk, SGLT-2i may indirectly improve mental health by mitigating these metabolic disturbances. Third, SGLT-2i exhibit anti-inflammatory and antioxidant properties (Chen et al., Reference Chen, Coronel, Hollmann, Weber and Zuurbier2022), which may alleviate neuroinflammatory responses and potentially slow neurodegenerative processes (Heimke et al., Reference Heimke, Lenz, Rickert, Lucius and Cossais2022). A meta-analysis has also linked inflammation and oxidative stress to self-harm behaviors (Vasupanrajit et al., Reference Vasupanrajit, Jirakran, Tunvirachaisakul, Solmi and Maes2022), further supporting a role for SGLT-2i in reducing self-harm risk (Chang et al., Reference Chang, Huang, Lo, Huang, Yang and Kornelius2025). In addition to their glucose-lowering and anti-inflammatory effects, SGLT-2i have demonstrated cardiovascular and renal benefits (American Diabetes Association Professional Practice Committee 2024b; Usman et al., Reference Usman, Bhatt, Hameed, Anker, Cheng, Hernandez and Butler2024), improving overall quality of life (Ding et al., Reference Ding, Spatz, Bena, Morrison, Levay, Lin and Albert2023). Given the significant association between physical and mental health (Fiorillo et al., Reference Fiorillo, de Girolamo, Simunovic, Gureje, Isaac, Lloyd and Sartorius2023), it is plausible that SGLT-2i use is associated with a lower risk of self-harm.

Furthermore, our study found that the protective association between SGLT-2i and self-harm was more pronounced in older patients. Social support is a crucial factor in preventing self-harm (Hou et al., Reference Hou, Wang, Guo, Zhang, Liu, Qi and Zhou2022), and older patients, compared with younger patients, often have more difficulty accessing social support (Valtorta et al., Reference Valtorta, Moore, Barron, Stow and Hanratty2018), which may predispose them to self-harm. In younger patients, the protective effect of SGLT-2i might be masked by better social support. Higher medication adherence in older patients (Kirkman et al., Reference Kirkman, Rowan-Martin, Levin, Fonseca, Schmittdiel, Herman and Aubert2015), which enhances the efficacy of SGLT-2i (Ciardullo et al., Reference Ciardullo, Savaré, Rea, Perseghin and Corrao2024), may partly explain the observed phenomenon. Additionally, existing studies (Hanlon et al., Reference Hanlon, Butterly, Wei, Wightman, SAM, Alsallumi and DA2025; Wu et al. Reference Wu, Liu, Zheng, Zou, Gu, Zhou and Wu2022) also support the idea that antidiabetic medications improve psychological health and cardiorenal function more significantly in older T2DM patients.

The European Medicines Agency (European Medicines Agency, 2025) and the U.S. Food and Drug Administration (Chiappini et al., Reference Chiappini, Vickers-Smith, Harris, Papanti Pelletier, Corkery, Guirguis and Schifano2023) have shown concerns about the increased risk of self-harm associated with GLP-1RA based on pharmacovigilance analyses (Nakhla et al., Reference Nakhla, Nair, Balani, Ujjawal, Arun Kumar, Dasari and Dani2024; Schoretsanitis et al., Reference Schoretsanitis, Weiler, Barbui, Raschi and Gastaldon2024). This finding may be related to the effects of GLP-1RA on the hypothalamic–pituitary–adrenal axis, rapid weight loss, disturbances in brain-derived neurotrophic factor, and diminished appetite (Shapiro et al., Reference Shapiro, Yin, Yu, Rej, Suissa and Azoulay2025; Tang et al., Reference Tang, Lu, Donahoo, Shao, Shi, Fonseca, Guo, Bian and Guo2024). The existing meta-analyses(Lu et al., Reference Lu, Tang, Kotecha, Cho, Donahoo, Bian and Guo2024) and secondary analyses (Wadden et al., Reference Wadden, Brown, Egebjerg, Frenkel, Goldman, Kushner and Fink-Jensen2024) of randomized controlled trials and target trial emulations(Gamble et al., Reference Gamble, Chibrikov, Midodzi, Twells and Majumdar2018; Hurtado et al., Reference Hurtado, Robles, Peiró, García-Sempere and Sanfélix-Gimeno2024; Shapiro et al., Reference Shapiro, Yin, Yu, Rej, Suissa and Azoulay2025; Tang et al., Reference Tang, Lu, Donahoo, Shao, Shi, Fonseca, Guo, Bian and Guo2024; Ueda et al., Reference Ueda, Söderling, Wintzell, Svanström, Pazzagli, Eliasson and Pasternak2024) appear to suggest that there is no statistically significant difference in the risk of self-harm between users of GLP-1 receptor agonists and SGLT-2 inhibitors in Western populations. In the primary analysis of this study, SGLT-2i use was associated with a lower risk of self-harm compared with GLP-1RA use. The discrepancy between our findings and existing studies may be attributed to the greater accessibility and acceptance of psychological treatments in Western populations compared to Chinese populations (Lu et al., Reference Lu, Xu, Huang, Li, Ma, Xu and Zhang2021; Mental Health America, 2025), which may partially mask the psychological burden associated with GLP-1RAs. In addition, the limited number of GLP-1RA users in this study may undermine the generalizability of this subpopulation, and the absence of data on dosage or medication adherence further compromises the reliability of comparisons between SGLT-2i and GLP-1RA. Consequently, the finding of a higher self-harm risk associated with GLP-1RA relative to SGLT-2i should be interpreted with caution.

To our knowledge, this study is the first to investigate the association between SGLT-2i and self-harm in an LMIC population. We implemented a target trial emulation framework, ensuring our observational study design preserved key features of randomized clinical trials to facilitate causal inference. This was operationalized through an active comparator, new user design employing two commonly used second-line antihyperglycemic agents, DPP-4i and GLP-1RA, as reference treatments to minimize immortal time bias, time-lag bias, and indication bias. Furthermore, extensive sensitivity and subgroup analyses were conducted to enhance the robustness of our findings. The issue of self-harm has been largely overlooked by governments worldwide, particularly in LMICs (Knipe et al., Reference Knipe, Padmanathan, Newton-Howes, Chan and Kapur2022; Moran et al., Reference Moran, Chandler, Dudgeon, Kirtley, Knipe, Pirkis and Christensen2024). The findings of this study provide compelling evidence supporting the safety profile of SGLT-2i and suggest that their integration into standard therapeutic regimens for T2DM in LMICs could help prevent a substantial number of self-harm cases at the population level. Given the increasing prevalence of T2DM and the considerable and growing burden of self-harm in these regions, the potential protective role of SGLT-2i could have profound implications for public health planning and socioeconomic development. Additionally, our findings raise concerns regarding the safety of GLP-1RAs in LMICs. The potential association between GLP-1RA use and self-harm risk should be acknowledged and addressed by the medical community, regulatory bodies, and policymakers. Given the expanding use of GLP-1RAs in the management of T2DM, there is an urgent need for large-scale, well-designed studies in these regions to rigorously assess the safety profile of GLP-1RAs in LMICs.

However, several limitations warrant consideration. First, as an observational study, our findings may be subject to residual and unmeasured confounding (e.g. acute psychosocial stress and baseline depression severity). Although the positive control analysis confirmed the expected association between SGLT-2i and GTI, the E values were relatively small; for instance, 1.60 for the comparison of SGLT-2i versus DPP-4i, suggesting that an unmeasured confounder with a hazard ratio of 1.60 for both SGLT-2i use and self-harm could fully attenuate some of our main findings. Given the inherent complexity of factors influencing SGLT-2i prescription and self-harm occurrence, the possibility of residual confounding cannot be completely ruled out. Second, given the later market entry and higher cost of GLP-1RA in China, the number of GLP-1RA users in our study was limited. Although our analysis suggested a lower risk of self-harm associated with SGLT-2i compared with GLP-1RA, this finding should be interpreted with caution. Third, although previous studies have suggested that even short-term SGLT-2i use, through elevated ketone levels and improved insulin sensitivity(Ferrannini et al., Reference Ferrannini, Baldi, Frascerra, Astiarraga, Heise, Bizzotto and Muscelli2016; Merovci et al., Reference Merovci, Solis-Herrera, Daniele, Eldor, Fiorentino, Tripathy and DeFronzo2014), may lower the risk of self-harm in patients with T2DM, the limited follow-up in our study and the non-significant findings in some lag analyses undermine the robustness of our estimates, underscoring the need for replication in larger, longer term cohorts. Fourth, this study only obtained information regarding the occurrence of self-harm, but lacked assessment of the severity of self-harm. In future studies, we will collect more detailed information to explore the association between SGLT-2i use and different severities of self-harm. Fifth, while outcome diagnoses were cross-checked against Chinese clinical terminology and physician notes, their primary ascertainment relied on ICD-10 codes, introducing potential misclassification that could affect estimate accuracy. However, this misclassification is unlikely to be differential between groups, thus biasing results toward the null rather than overestimating the effect (Jeong et al., Reference Jeong, Park, Noh, Bea, Filion, Yu and Shin2023). Finally, although this study provides the first evidence of the association between SGLT-2i and self-harm in LMICs, our findings are derived exclusively from a Chinese cohort. Therefore, caution is warranted when extrapolating these results to other populations.

In conclusion, this study provides preliminary evidence suggesting a potential beneficial association between SGLT-2i and reduced self-harm risk in T2DM patients compared with DPP-4i and GLP-1RA, supporting the safety profile of SGLT-2i. Meanwhile, these findings also highlight pertinent safety concerns regarding the use of GLP-1RA in LMICs. Future studies are needed to verify these observations via randomized controlled trials, investigate the dose–response relationship, and explore the underlying mechanisms involved.

Supplementary material

The supplementary material for this article can be found at http://doi.org/10.1017/S0033291726105984.

Funding statement

This work was supported by the National Natural Science Foundation of China (grant number: 82173616).

Competing interests

No potential conflicts of interest relevant to this article were reported.

Ethical standards

The study was approved by the Ethics Review Board of Shenzhen Jianyan Digital Management (Project Number: 2025007). The requirement for informed consent was waived because all data were de-identified.

References

American Diabetes Association Professional Practice Committee. (2024a). 9. Pharmacologic approaches to Glycemic treatment: Standards of Care in Diabetes-2024. Diabetes Care, 47(Suppl_1), S158–S178. https://doi.org/10.2337/dc24-S009CrossRefGoogle Scholar
American Diabetes Association Professional Practice Committee. (2024b). 10. Cardiovascular disease and risk management: Standards of Care in Diabetes-2024. Diabetes Care, 47(Suppl_1), S179–S218. https://doi.org/10.2337/dc24-S010CrossRefGoogle Scholar
Austin, P. C., & Stuart, E. A. (2015). Moving towards best practice when using inverse probability of treatment weighting (IPTW) using the propensity score to estimate causal treatment effects in observational studies. Statistics in Medicine, 34(28), 3661–3679. https://doi.org/10.1002/sim.6607CrossRefGoogle ScholarPubMed
Brietzke, E., Mansur, R. B., Subramaniapillai, M., Balanzá-Martínez, V., Vinberg, M., González-Pinto, A., & McIntyre, R. S. (2018). Ketogenic diet as a metabolic therapy for mood disorders: Evidence and developments. Neuroscience and Biobehavioral Reviews, 94, 11–16. https://doi.org/10.1016/j.neubiorev.2018.07.020CrossRefGoogle ScholarPubMed
Cashin, A. G., Hansford, H. J., Hernán, M. A., Swanson, S. A., Lee, H., Jones, M. D., & McAuley, J. H. (2025). Transparent reporting of observational studies emulating a target trial—The TARGET statement. JAMA, 334(12), 1084–1093. https://doi.org/10.1001/jama.2025.13350CrossRefGoogle ScholarPubMed
Chang, E. W., Huang, J. Y., Lo, S. C., Huang, C. N., Yang, Y. S., & Kornelius, E. (2025). Association between SGLT-2 inhibitors and suicide risk in type 2 diabetes and bipolar: A real-world cohort study. Frontiers in Pharmacology, 16, 1601118. https://doi.org/10.3389/fphar.2025.1601118CrossRefGoogle ScholarPubMed
Chen, S., Coronel, R., Hollmann, M. W., Weber, N. C., & Zuurbier, C. J. (2022). Direct cardiac effects of SGLT2 inhibitors. Cardiovascular Diabetology, 21(1), 45. https://doi.org/10.1186/s12933-022-01480-1CrossRefGoogle ScholarPubMed
Chiappini, S., Vickers-Smith, R., Harris, D., Papanti Pelletier, G. D., Corkery, J. M., Guirguis, A., & Schifano, F. (2023). Is there a risk for Semaglutide misuse? Focus on the Food and Drug Administration’s FDA adverse events reporting system (FAERS) pharmacovigilance dataset. Pharmaceuticals (Basel), 16(7). https://doi.org/10.3390/ph16070994CrossRefGoogle Scholar
Chinese Diabetes Society. (2025). Guideline for the prevention and treatment of diabetes mellitus in China (2024 edition). Chinese Journal of Diabetes Mellitus 17(1), 16–139Google Scholar
Ciardullo, S., Savaré, L., Rea, F., Perseghin, G., & Corrao, G. (2024). Adherence to GLP1-RA and SGLT2-I affects clinical outcomes and costs in patients with type 2 diabetes. Diabetes/Metabolism Research and Reviews, 40(4), e3791. https://doi.org/10.1002/dmrr.3791CrossRefGoogle ScholarPubMed
Conti, C., Mennitto, C., Di Francesco, G., Fraticelli, F., Vitacolonna, E., & Fulcheri, M. (2017). Clinical characteristics of diabetes mellitus and suicide risk. Frontiers in Psychiatry, 8, 40. https://doi.org/10.3389/fpsyt.2017.00040CrossRefGoogle ScholarPubMed
Ding, Q., Spatz, E. S., Bena, J. F., Morrison, S. L., Levay, M., Lin, H., & Albert, N. M. (2023). Association of SGLT-2 inhibitors with treatment satisfaction and diabetes-specific and general health status in adults with cardiovascular disease and type 2 diabetes. Journal of the American Heart Association, 12(17), e029058. https://doi.org/10.1161/jaha.122.029058CrossRefGoogle ScholarPubMed
European Medicines Agency. (2025). EMA statement on ongoing review of GLP-1 receptor agonists. https://www.ema.europa.eu/en/news/ema-statement-ongoing-review-glp-1-receptor-agonists (accessed 6 August 2025).Google Scholar
Ferrannini, E., Baldi, S., Frascerra, S., Astiarraga, B., Heise, T., Bizzotto, R., & Muscelli, E. (2016). Shift to fatty substrate utilization in response to sodium-glucose cotransporter 2 inhibition in subjects without diabetes and patients with type 2 diabetes. Diabetes, 65(5), 1190–1195. https://doi.org/10.2337/db15-1356CrossRefGoogle ScholarPubMed
Fiorillo, A., de Girolamo, G., Simunovic, I. F., Gureje, O., Isaac, M., Lloyd, C., & Sartorius, N. (2023). The relationship between physical and mental health: An update from the WPA working group on managing comorbidity of mental and physical health. World Psychiatry, 22(1), 169–170. https://doi.org/10.1002/wps.21055CrossRefGoogle Scholar
Gamble, J. M., Chibrikov, E., Midodzi, W. K., Twells, L. K., & Majumdar, S. R. (2018). Examining the risk of depression or self-harm associated with incretin-based therapies used to manage hyperglycaemia in patients with type 2 diabetes: A cohort study using the UK Clinical Practice Research Datalink. BMJ Open, 8(10), e023830. https://doi.org/10.1136/bmjopen-2018-023830CrossRefGoogle ScholarPubMed
GBD 2021 Diabetes Collaborators. (2023). Global, regional, and national burden of diabetes from 1990 to 2021, with projections of prevalence to 2050: A systematic analysis for the Global Burden of Disease Study 2021. Lancet, 402(10397), 203–234. https://doi.org/10.1016/s0140-6736(23)01301-6CrossRefGoogle Scholar
Hanlon, P., Butterly, E., Wei, L., Wightman, H., SAM, A., Alsallumi, K., & DA, M. A. (2025). Age and sex differences in efficacy of treatments for type 2 diabetes: A network meta-analysis. JAMA, 333(12), 1062–1073. https://doi.org/10.1001/jama.2024.27402CrossRefGoogle ScholarPubMed
Heimke, M., Lenz, F., Rickert, U., Lucius, R., & Cossais, F. (2022). Anti-inflammatory properties of the SGLT2 inhibitor empagliflozin in activated primary microglia. Cells, 11(19). https://doi.org/10.3390/cells11193107CrossRefGoogle ScholarPubMed
Hou, X., Wang, J., Guo, J., Zhang, X., Liu, J., Qi, L., & Zhou, L. (2022). Methods and efficacy of social support interventions in preventing suicide: A systematic review and meta-analysis. Evidence-Based Mental Health, 25(1), 29–35. https://doi.org/10.1136/ebmental-2021-300318CrossRefGoogle ScholarPubMed
Huang, Q., Ke, L., Liu, L., Liu, Y., Li, Y., Feng, G., & Di, Q. (2025). Heatwave warnings mitigate long-term cardiovascular disease risk from heat-related illness: A real-world prospective cohort study. The Lancet Regional Health. Western Pacific, 55, 101468. https://doi.org/10.1016/j.lanwpc.2025.101468CrossRefGoogle Scholar
Hurtado, I., Robles, C., Peiró, S., García-Sempere, A., & Sanfélix-Gimeno, G. (2024). Association of glucagon-like peptide-1 receptor agonists with suicidal ideation and self-injury in individuals with diabetes and obesity: A propensity-weighted, population-based cohort study. Diabetologia, 67(11), 2471–2480. https://doi.org/10.1007/s00125-024-06243-zCrossRefGoogle ScholarPubMed
Jeong, H. E., Park, S., Noh, Y., Bea, S., Filion, K. B., Yu, O. H. Y., & Shin, J. Y. (2023). Association of adverse respiratory events with sodium-glucose cotransporter 2 inhibitors versus dipeptidyl peptidase 4 inhibitors among patients with type 2 diabetes in South Korea: A nationwide cohort study. BMC Medicine, 21(1), 47. https://doi.org/10.1186/s12916-023-02765-2CrossRefGoogle ScholarPubMed
Kirkman, M. S., Rowan-Martin, M. T., Levin, R., Fonseca, V. A., Schmittdiel, J. A., Herman, W. H., & Aubert, R. E. (2015). Determinants of adherence to diabetes medications: Findings from a large pharmacy claims database. Diabetes Care, 38(4), 604–609. https://doi.org/10.2337/dc14-2098CrossRefGoogle ScholarPubMed
Knipe, D., Padmanathan, P., Newton-Howes, G., Chan, L. F., & Kapur, N. (2022). Suicide and self-harm. Lancet, 399(10338), 1903–1916. https://doi.org/10.1016/s0140-6736(22)00173-8CrossRefGoogle ScholarPubMed
Lu, J., Xu, X., Huang, Y., Li, T., Ma, C., Xu, G., & Zhang, N. (2021). Prevalence of depressive disorders and treatment in China: A cross-sectional epidemiological study. Lancet Psychiatry, 8(11), 981–990. https://doi.org/10.1016/s2215-0366(21)00251-0CrossRefGoogle Scholar
Lu, Y., Tang, H., Kotecha, P., Cho, H., Donahoo, W. T., Bian, J., & Guo, J. (2024). Newer glucose-lowering drugs and risk of suicidal ideation and behaviours: A network meta-analysis of randomized outcome trials. Diabetes, Obesity & Metabolism, 26(10), 4783–4786. https://doi.org/10.1111/dom.15822CrossRefGoogle ScholarPubMed
Lund, J. L., Richardson, D. B., & Stürmer, T. (2015). The active comparator, new user study design in pharmacoepidemiology: Historical foundations and contemporary application. Current Epidemiology Reports, 2(4), 221–228. https://doi.org/10.1007/s40471-015-0053-5CrossRefGoogle ScholarPubMed
Majumdar, S., Sinha, B., Dastidar, B. G., Gangopadhyay, K. K., Ghoshal, S., Mukherjee, J. J., & Bhattacharjee, K. (2021). Assessing prevalence and predictors of depression in type 2 diabetes mellitus (DM) patients - The DEPDIAB study. Diabetes Research and Clinical Practice, 178, 108980. https://doi.org/10.1016/j.diabres.2021.108980CrossRefGoogle Scholar
Mental Health America. (2025). The State of Mental Health in America 2025. https://mhanational.org/research/state-of-mental-health-in-america-2025/ (Accessed 6 August 2025).Google Scholar
Merovci, A., Solis-Herrera, C., Daniele, G., Eldor, R., Fiorentino, T. V., Tripathy, D., & DeFronzo, R. A. (2014). Dapagliflozin improves muscle insulin sensitivity but enhances endogenous glucose production. The Journal of Clinical Investigation, 124(2), 509–514. https://doi.org/10.1172/jci70704CrossRefGoogle ScholarPubMed
Moran, P., Chandler, A., Dudgeon, P., Kirtley, O. J., Knipe, D., Pirkis, J., & Christensen, H. (2024). The Lancet Commission on self-harm. Lancet, 404(10461), 1445–1492. https://doi.org/10.1016/s0140-6736(24)01121-8CrossRefGoogle ScholarPubMed
Myers, A. K., Grannemann, B. D., Lingvay, I., & Trivedi, M. H. (2013). Brief report: Depression and history of suicide attempts in adults with new-onset type 2 diabetes. Psychoneuroendocrinology, 38(11), 2810–2814. https://doi.org/10.1016/j.psyneuen.2013.06.013CrossRefGoogle ScholarPubMed
Nakhla, M., Nair, A., Balani, P., Ujjawal, A., Arun Kumar, P., Dasari, M., & Dani, S. S. (2024). Risk of suicide, hair loss, and aspiration with GLP1-receptor agonists and other diabetic agents: A real-world pharmacovigilance study. Cardiovascular Drugs and Therapy, 39(6), 1331–1341. https://doi.org/10.1007/s10557-024-07613-wCrossRefGoogle ScholarPubMed
Renaud-Charest, O., Stoljar Gold, A., Mok, E., Kichler, J., Nakhla, M., & Li, P. (2024). Suicidal ideation, suicide attempts, and suicide deaths in adolescents and young adults with type 1 diabetes: A systematic review and meta-analysis. Diabetes Care, 47(7), 1227–1237. https://doi.org/10.2337/dc24-0411CrossRefGoogle Scholar
Schoretsanitis, G., Weiler, S., Barbui, C., Raschi, E., & Gastaldon, C. (2024). Disproportionality analysis from World Health Organization data on Semaglutide, liraglutide, and suicidality. JAMA Network Open, 7(8), e2423385. https://doi.org/10.1001/jamanetworkopen.2024.23385CrossRefGoogle ScholarPubMed
Shapiro, S. B., Yin, H., Yu, O. H. Y., Rej, S., Suissa, S., & Azoulay, L. (2025). Glucagon-like peptide-1 receptor agonists and risk of suicidality among patients with type 2 diabetes: Active comparator, new user cohort study. BMJ, 389, r845. https://doi.org/10.1136/bmj.r845Google Scholar
Tang, H., Lu, Y., Donahoo, W. T., Shao, H., Shi, L., Fonseca, V. A., Guo, J., Bian, J., & Guo, J. (2024). Glucagon-like Peptide-1 receptor agonists and risk for suicidal ideation and Behaviors in U.S. older adults with type 2 diabetes: A target trial emulation study. Annals of Internal Medicine, 177(8), 1004–1015. https://doi.org/10.7326/m24-0329CrossRefGoogle ScholarPubMed
Ueda, P., Söderling, J., Wintzell, V., Svanström, H., Pazzagli, L., Eliasson, B., & Pasternak, B. (2024). GLP-1 receptor agonist use and risk of suicide death. JAMA Internal Medicine, 184(11), 1301–1312. https://doi.org/10.1001/jamainternmed.2024.4369CrossRefGoogle ScholarPubMed
Usman, M. S., Bhatt, D. L., Hameed, I., Anker, S. D., Cheng, A. Y. Y., Hernandez, A. F., & Butler, J. (2024). Effect of SGLT2 inhibitors on heart failure outcomes and cardiovascular death across the cardiometabolic disease spectrum: A systematic review and meta-analysis. The Lancet Diabetes and Endocrinology, 12(7), 447–461. https://doi.org/10.1016/s2213-8587(24)00102-5CrossRefGoogle ScholarPubMed
Valtorta, N. K., Moore, D. C., Barron, L., Stow, D., & Hanratty, B. (2018). Older adults’ social relationships and health care utilization: A systematic review. American Journal of Public Health, 108(4), e1–e10. https://doi.org/10.2105/ajph.2017.304256CrossRefGoogle ScholarPubMed
Vasupanrajit, A., Jirakran, K., Tunvirachaisakul, C., Solmi, M., & Maes, M. (2022). Inflammation and nitro-oxidative stress in current suicidal attempts and current suicidal ideation: A systematic review and meta-analysis. Molecular Psychiatry, 27(3), 1350–1361. https://doi.org/10.1038/s41380-021-01407-4CrossRefGoogle ScholarPubMed
Wadden, T. A., Brown, G. K., Egebjerg, C., Frenkel, O., Goldman, B., Kushner, R. F., & Fink-Jensen, A. (2024). Psychiatric safety of Semaglutide for weight Management in People without Known Major Psychopathology: Post hoc analysis of the STEP 1, 2, 3, and 5 trials. JAMA Internal Medicine, 184(11), 1290–1300. https://doi.org/10.1001/jamainternmed.2024.4346CrossRefGoogle ScholarPubMed
Wu, K., Liu, H., Zheng, J., Zou, L., Gu, S., Zhou, R., & Wu, X. (2022). Diabetes treatment is associated with better cognitive function: The age disparity. Frontiers in Aging Neuroscience, 13, 753129. https://doi.org/10.3389/fnagi.2021.753129CrossRefGoogle ScholarPubMed
Xie, Y., Bowe, B., Xian, H., Loux, T., McGill, J. B., & Al-Aly, Z. (2023). Comparative effectiveness of SGLT2 inhibitors, GLP-1 receptor agonists, DPP-4 inhibitors, and sulfonylureas on risk of major adverse cardiovascular events: Emulation of a randomised target trial using electronic health records. The Lancet Diabetes and Endocrinology, 11(9), 644–656. https://doi.org/10.1016/s2213-8587(23)00171-7CrossRefGoogle ScholarPubMed
Figure 0

Table 1.Baseline characteristics of cohort SGLT-2i versus DPP-4iTable 1. long description.

Figure 1

Table 2.Baseline characteristics of cohort SGLT-2i versus GLP-1RATable 2. long description.

Figure 2

Figure 1.Primary and sensitivity analysis of the association between SGLT-2i and the incidence of self-harm. Note: A, SGLT-2i versus DPP-4i; B, SGLT-2i versus GLP-1RA. The number of outcome events, incidence rate, HR, and its 95%CI represented the results after overlapping weighting. The primary analysis defines self-harm according to ICD-10 as X60–X84 and Y87.0, while the redefined self-harm is defined as X60–X84 and Y10–Y34. Combination therapy was defined as the scenario where patients in the exposure group received the comparator drug, and those in the comparator group received the exposure drug within a specified time window. Extreme weights are defined as overlapping weights less than the 1st percentile or greater than the 99th percentile. Laboratory tests included glycated hemoglobin, triglycerides, total cholesterol, high-density lipoprotein cholesterol, low-density lipoprotein cholesterol, alanine aminotransferase, aspartate aminotransferase, γ-glutamyl transpeptidase, direct bilirubin, serum uric acid, and serum creatinine. Mental illness included depression, Alzheimer’s disease, other dementias, delirium, anxiety disorders, substance use disorders, bipolar disorder, schizophrenia, obsessive-compulsive disorder, post-traumatic stress disorder, attention-deficit/hyperactivity disorder, borderline personality disorder, eating disorders, sleep disorders, and chronic pain. Abbreviations: CI, confidence interval; DPP-4i, dipeptidyl peptidase-4 inhibitors; HR, hazard ratio; SGLT-2i, sodium-glucose cotransporter 2 inhibitors; T2DM, type 2 diabetes mellitus.Figure 1. long description.

Figure 3

Figure 2.Subgroup analysis of the association between SGLT-2i and the incidence of self-harm. Note: A, SGLT-2i versus DPP-4i; B, SGLT-2i versus GLP-1RA. The number of outcome events, incidence rate, HR, and its 95%CI represented the results after overlapping weighting. Abbreviations: CI, confidence interval; GLP-1RA, glucagon-like peptide-1 receptor agonist; HR, hazard ratio; SGLT-2i, sodium-glucose cotransporter 2 inhibitors; T2DM, type 2 diabetes mellitus.Figure 2. long description.

Supplementary material: File

Li et al. supplementary material

Li et al. supplementary material
Download Li et al. supplementary material(File)
File 305.5 KB