Association of SGLT2 Inhibitors With Suicidality, All-Cause Mortality, and Hospitalization in Bipolar Disorder: A Cohort Study of 1.23 Million Adults

Abstract

Background:

Sodium-glucose cotransporter-2 inhibitors (SGLT2is) have demonstrated promising neuropsychiatric properties in animal research.

Aims:

We aimed to examine whether SGLT2i exposure is associated with reduced incidence of suicidality, all- cause mortality, and hospitalization in bipolar disorder (BD).

Method:

This cohort study utilized the TriNetX network (2008–2025) to identify 1,230,821 adults with BD, including 36,092 SGLT2i users and 1,194,729 non-users. The outcomes were incident suicidality, all-cause mortality, and hospitalization over 5-years of follow-up. Adjusted hazard ratios (aHRs) were estimated using Cox proportional hazards models, with additional 1:1 propensity score matching (PSM) for survival analysis conducted to control for confounding. Subgroup analyses stratified results by sex, race/ethnicity, concurrent mood stabilizer use, and somatic comorbidities.

Results:

SGLT2i exposure was associated with significantly reduced risk of suicidality (aHR 0.75, 95% CI 0.71–0.79), all-cause mortality (aHR 0.55, 95% CI 0.52–0.57), and hospitalization (aHR 0.71, 95% CI 0.69–0.72). Protective associations remained consistent across subgroups, including patients receiving lithium, lamotrigine, or valproate. After PSM (31,001 matched pairs), five-year outcomes favored SGLT2i users: suicidality-free survival 94.51% versus 93.56%, overall survival 88.99% versus 79.57%, and hospitalization-free survival 72.39% versus 66.72% (all p <0.001).

Conclusions:

In this large cohort of over 1.23 million adults with BD, SGLT2i therapy was associated with substantially lower risks of suicidality, hospitalization, and all-cause mortality, with consistent benefits across demographic and clinical subgroups. These novel findings suggest SGLT2is may represent a promising therapeutic strategy to improve psychiatric and survival outcomes in BD. Prospective RCTs are warranted to evaluate long- term safety and efficacy.

INTRODUCTION

Bipolar disorder (BD) is characterized by recurrent episodes of mania/hypomania, mixed states, and depression, with significant morbidity and mortality (Singh et al., 2025b; Vieta et al., 2025). The disease burden is substantial: affected individuals predominantly experience depressive symptoms throughout their illness trajectory and face markedly increased risks of suicide, psychiatric hospitalization, and premature death relative to the general population (Elsayed et al., 2022; Vieta et al., 2025). Contemporary pharmacological management centers on mood-stabilizing medications including lithium, anticonvulsants (valproate, carbamazepine, lamotrigine), and second-generation antipsychotics (Cuellar-Barboza et al., 2020; Elmosalamy et al., 2025; Elsayed et al., 2022; Ho et al., 2020; Singh et al., 2024; Yocum and Singh, 2025). Despite these available treatments, effective therapeutic options for BD remain inadequate, underscoring an urgent need for innovative interventions to enhance clinical outcomes and reduce disease-related morbidity (Singh et al., 2022).

Sodium-glucose cotransporter-2 inhibitors (SGLT2is), originally developed as oral antihyperglycemic agents, have since demonstrated significant cardiovascular and renoprotective benefits, with several agents receiving FDA approval for chronic kidney disease (CKD) and heart failure, independent of their glucose-lowering properties (Wu et al., 2016; Yankah et al., 2024). Accumulating preclinical evidence suggests these agents may possess neuroprotective and neuropsychiatric properties, including anti-inflammatory and antioxidant effects, with demonstrated attenuation of depressive-like behaviors in animal models (Muhammad et al., 2024). Recently, a randomized controlled trial (RCT) demonstrated the efficacy of empagliflozin (an SGLT2i) as augmentation therapy for major depressive disorder, providing preliminary clinical evidence for the potential psychiatric utility of this drug class (Zandifar et al., 2024). However, there is no evidence of their efficacy in BD.

Individuals with BD exhibit disproportionately high prevalence of comorbid medical conditions, including diabetes mellitus (DM), CKD, and cardiovascular disease—conditions in which SGLT2is have demonstrated therapeutic benefit (Ercis et al., 2025, 2026; Wu et al., 2016; Yankah et al., 2024). Recent data suggests that SGLT2i use is associated with reduced rates of progression to kidney replacement therapy among individuals with comorbid BD and CKD (Singh et al., 2026). However, the impact of SGLT2i exposure on psychiatric outcomes in BD has not been systematically investigated. The objective of this hypothesis-generating study was to examine the association between SGLT2i exposure and psychiatric outcomes in adults with BD, specifically suicidality, all-cause mortality and hospitalization.

Methods

Data Source and Study Design

This retrospective cohort study leveraged a large, real-world dataset from the TriNetX network (TriNetX, LLC), which aggregates de- identified electronic health records (EHRs) from multiple healthcare organizations across the United States (Baweja et al., 2025; Palchuk et al., 2023). Adult patients (≥18 years) with a documented diagnosis of BD were identified using International Classification of Diseases, Tenth Revision (ICD-10) codes F30–F31 during the study period (2008–May 2025) (Singh et al., 2025a; Singh et al., 2026). Patients with a history of schizoaffective disorder or primary psychotic disorders (F20–F29) were excluded to minimize confounding. The sample was geographically diverse, with 28% from the Northeast, 21% from the Midwest, 33% from the South, 14% from the West, and 2% from Ex-US locations. Overall, 54% of patients were treated at academic centers and 46% at non- academic centers. This study utilized de-identified secondary data and was therefore exempt from Institutional Review Board oversight under the National Human Research Protections Advisory Committee guidelines, with waived patient consent. All procedures were conducted in conformity with the ethical standards outlined in the Declaration of Helsinki, and study reporting followed the Strengthening the Reporting of Observational Studies in Epidemiology guidelines (Vandenbroucke et al., 2007). Across the study cohort, mean follow-up was 4.37 years (SD =4.25).

Variables

The study evaluated two primary outcomes: new-onset suicidality and all-cause mortality. Suicidality encompassed both suicidal ideation (SI) and suicide attempts, identified through ICD-10 diagnostic codes R45.851 (SI) and T14.91 (suicide attempt). All-cause mortality was defined as any death recorded within the EHR during the follow-up period, encompassing deaths attributable to somatic, psychiatric, or combined etiologies (Supplementary Tables 1 and 2).

The secondary outcome was inpatient hospitalization, defined by presence of Hospital Inpatient Services code. To ensure assessment of incident events, patients with documented history of suicidality or hospitalization prior to the index date were excluded from the respective outcome analyses.

Independent Variable:

The primary independent variable was SGLT2i use (including canagliflozin, dapagliflozin, empagliflozin, ertugliflozin, bexagliflozin, and sotagliflozin), classified as user versus non-user. Individuals with BD who were newly initiated on SGLT2is were classified into the SGLT2i group, with the index date defined as the first prescription date. Follow- up commenced at treatment initiation and was censored at outcome occurrence, loss to follow-up, or study conclusion. Individuals who never received SGLT2is formed the comparator group. This time- dependent approach ensures that exposure classification dynamically reflects real-world prescribing patterns, minimizes exposure misclassification, and minimizes the immortal time bias that may arise when exposure is fixed at a predetermined baseline date (Jain and Goldsweig, 2026). To further address this concern, an active-comparator analysis was conducted using dipeptidyl peptidase-4 inhibitors (DPP-4i) as the reference treatment (Chang et al., 2025) for the primary outcomes—new-onset suicidality and all-cause mortality—as detailed in the Statistical Analysis section. Because both agents are commonly prescribed for DM, the risk of confounding by indication is minimized.

Covariates:

Covariates included demographic characteristics (age at index date, sex, race, and ethnicity), psychiatric and somatic comorbidities, and relevant pharmacotherapy, as documented in the EHR. All covariates were ascertained within the period preceding and including the index date.

Statistical Analysis

All analyses were conducted in November 2025 using the secure TriNetX platform. Categorical variables were expressed as frequencies and proportions, and continuous variables as means with standard deviations (SDs). To compare demographic variables, psychiatric and somatic comorbidities, and pharmacotherapy between SGLT2i users and non-users, standardized mean differences (SMDs) and corresponding p-values were calculated.

Time-to-event analyses were performed to assess differences in outcome incidence between exposure groups. Cox proportional hazards models were used to estimate adjusted hazard ratios (aHRs) with 95% confidence intervals (CIs) over a five-year follow-up period. Outcomes included new-onset suicidality (suicidal ideation or attempts), all-cause mortality, and inpatient hospitalizations. Patients were censored at the time of the event, loss to follow-up, or end of the follow-up period, whichever occurred first. Models were adjusted for sociodemographic variables, psychiatric and somatic comorbidities, and concurrent pharmacotherapy. Covariates included age (continuous), sex, race, ethnicity, presence of comorbidities, and pharmacotherapy, with reference categories set as female/other, non-White, non-Hispanic/Latino, absence of comorbidities and no pharmacotherapy. A neutral outcome, acute appendicitis, was examined to identify any bias unrelated to SGLT2i use, following approaches used in previous studies (Pan et al., 2024; Su et al., 2025).

To assess the robustness of our findings, we conducted sensitivity analyses across clinically relevant subgroups defined by sex, race, ethnicity, psychopharmacologic treatment regimens (including antipsychotics, valproate, lamotrigine, and lithium), and concurrent somatic comorbidities (DM, CKD, and heart failure). Given the large number of comparisons, a threshold of p <0.001 was used to define statistical significance. A supplementary sensitivity analysis employed a more restrictive definition of SGLT2i exposure, requiring a minimum of two prescriptions during the study period to better capture sustained therapeutic engagement rather than sporadic or single-exposure use. An additional active-comparator sensitivity analysis, restricted to individuals prescribed either SGLT2i or DPP-4i, was conducted for the two primary outcomes to further minimize immortal time bias and confounding by indication.

Survival Analysis

Event-free survival probabilities for suicidality, all-cause mortality, and inpatient hospitalization were compared between SGLT2i users and nonusers over the five-year follow-up period using Kaplan–Meier survival analyses. To address potential baseline imbalances between groups, we implemented propensity score matching (PSM) incorporating the same covariates utilized in the Cox regression models. A greedy nearest-neighbor matching algorithm with a caliper width of 0.1 SDs was applied to achieve 1:1 matching without replacement, thereby minimizing residual confounding, enhancing covariate balance between comparison groups, and strengthening the interpretability of the survival analyses (Austin, 2014). SD values below 0.1 were considered indicative of adequate balance. Following PSM, baseline characteristics were well balanced between cohorts, with most SMDs below 0.10. Residual imbalance was noted for diabetes mellitus and related treatments (SMD ~0.13–0.14), likely reflecting substantial baseline differences and limited group overlap (Mortada et al., 2026). To account for this, a subgroup analysis stratified by DM status was conducted. Furthermore, PSM was applied to the active-comparator SGLT2i versus DPP-4i cohort using identical matching procedures and covariate sets. Following matching, cohorts were well balanced, with all SMDs below 0.1 (Supplementary Table 4). Comprehensive details regarding the matching, outcome definitions, and sample size characteristics are presented in the Supplementary Table 3 & 4.

RESULTS

In a cohort of 1,230,821 individuals with BD, 36,092 (2.9%) were prescribed SGLT2is (Fig. 1). Lithium use was low (10.5%), whereas most patients received other psychotropics: anticonvulsants (53%), antipsychotics (54%), antidepressants (61%), and benzodiazepines (58%). SGLT2i users were older (mean age 59.2 vs. 48.0 years), more likely male (41.4% vs. 36.7%), and more frequently Black (15.7% vs. 12.9%) (Table 1). Psychiatric comorbidities were more common in SGLT2i users, including anxiety disorders (73.4% vs. 62.1%), substance use disorders (53.5% vs. 49.0%), and sleep disorders (14.9% vs. 7.1%), though suicide attempts (1.3% vs. 1.5%) and SIs (13.4% vs. 15.8%) were less frequent. Somatic conditions were substantially higher in the SGLT2i group: DM (86.0% vs. 15.6%), hypertension (86.9% vs. 35.0%), heart failure (37.8% vs. 5.7%), CKD (32.7% vs. 7.2%), and obesity (65.4% vs. 26.8%). SGLT2i users received more psychotropic medications—anticonvulsants (76.5% vs. 52.5%), antipsychotics (68.2% vs. 53.4%), antidepressants (85.8% vs. 60.5%), benzodiazepines (80.2% vs. 57.3%)—and cardiovascular medications—Angiotensin-converting enzyme (ACE) inhibitors (53.8% vs. 13.1%), angiotensin II blockers (46.1% vs. 8.0%), hypoglycemic agents (49.1% vs. 5.0%). All differences were statistically significant (p <0.001).

Figure 1.

Flow diagram for the study cohort

Table 1:

Characteristics of Bipolar Disorder Patients by SGLT2 Inhibitor Use^a

Variables	Full Cohort 1,23,0821 (%)	SGLT2i Users 36,092 (%)	SGLT2i Nonusers 1,194,729 (%)	SMD	p-value
Sociodemographic
Current Age in years, Mean (SD)	48.2+/−17.1	59.2 +/−12.6	48.0 +/−17.1	0.7465	<0.001
Age at Index in years, Mean (SD)	40.0+/−16.7	55.9+/−12.7	40.1+/−16.7	0.9520	<0.001
Gender
Male	447777 (36.38)	14,928 (41.36)	432,849 (36.70)	0.1054	<0.001
Female	782148 (63.55)	21,139 (58.57)	761,009 (63.70)	0.1053	<0.001
Unknown	896 (0.07)	25 (0.07)	871 (0.07)	0.0014	0.8008
Ethnicity
Not Hispanic or Latino	874802 (71.07)	27,265 (75.54)	847,537 (70.94)	0.1041	<0.001
Hispanic or Latino	78896 (6.41)	2,523 (6.99)	76,373 (6.39)	0.0239	<0.001
Unknown Ethnicity	277123 (22.52)	6,304 (17.)	270,819 (22.67)	0.1301	<0.001
Race
White	884185 (71.84)	26,041 (72.15)	858,144 (71.83)	0.0072	0.1773
Black or African America	159832 (12.99)	5,654 (15.67)	154178 (12.90)	0.0790	<0.001
American Indian or Alaska Native	7478 (0.61)	306 (0.85)	7,172 (0.60)	0.0292	<0.001
Asian	23246 (1.89)	710 (1.97)	22,536 (1.89)	0.0059	0.2659
Native Hawaiian or Pacific Islander	3387 (0.28)	149 (0.41)	3,238 (0.27)	0.0243	<0.001
Other	44399 (3.61)	1,223 (3.39)	43,176 (3.61)	0.0123	0.0237
Unknown Race	108294 (8.80)	2,009 (5.37)	106,285 (8.90)	0.1288	<0.001
Psychiatric Comorbidity
Anxiety disorders	768074 (62.40)	26497 (73.42)	741,577 (62.07)	0.2445	<0.001
Substance use disorders	604390 (49.10)	19,306 (53.49)	585,084 (48.97)	0.0905	<0.001
Sleep Disorders	90153 (7.32)	5,365 (14.86)	84,788 (7.10)	0.2504	<0.001
Eating disorders	34825 (2.83)	1,079 (2.99)	33,746 (2.82)	0.0098	0.0625
Suicide attempt	18791 (1.53)	469 (1.30)	18,322(1.53)	0.0198	0.0004
Suicidal ideations	193678 (15.74)	4,849(13.44)	188,829(15.81)	0.0671	<0.0001
Somatic Comorbidity
Hypertension	449516 (36.52)	31,371 (86.92)	418,145 (35.00)	1.2571	<0.001
Ischemic heart diseases	141781 (11.52)	16,305 (45.18)	125,476 (10.50)	0.8389	<0.001
Heart Failure	81856 (6.65)	13,628 (37.76)	68,228 (5.71)	0.8433	<0.001
Cerebrovascular diseases	99048 (8.05)	8,403 (23.28)	90,645 (7.59)	0.4451	<0.001
Chronic Kidney Disease	98081 (7.97)	11,784 (32.65)	86,297 (7.22)	0.6713	<0.001
COPD	128407 (10.43)	10,631 (29.46)	117,776 (9.86)	0.5088	<0.001
Overweight and obesity	344199 (27.96)	23,613 (65.42)	320,586 (26.83)	0.8396	<0.001
Diabetes mellitus	217133 (17.64)	31,034 (85.99)	186,099 (15.58)	1.9835	<0.001
Thyroid disorders	225206 (18.30)	12,843 (35.58)	212,363 (17.77)	0.4111	<0.001
Pharmacotherapy
Lithium	129162 (10.49)	3,877 (10.74)	125,285 (10.49)	0.0094	0.0805
Anticonvulsants	654837 (53.20)	27,608 (76.49)	627,229 (52.50)	0.5384	<0.001
Valproate	152887 (12.42)	6,582 (18.24)	146,305 (12.25)	0.1673	<0.001
Carbamazepine	36352 (2.95)	1,641 (4.55)	34,711 (2.91)	0.0867	<0.001
Lamotrigine	277835 (22.57)	10,153 (28.13)	267,682 (22.41)	0.1320	0.011
Antipsychotics	662442 (53.82)	24,605 (68.17)	637,837 (53.39)	0.3064	<0.001
Antidepressants	753392 (61.21)	30,242 (85.79)	723,150 (60.53)	0.5374	<0.001
SSRI/SNRI	736211 (59.81)	29,603 (82.02)	706,608 (59.14)	0.5187	<0.001
TCAs	127110 (10.33)	7,550 (20.92)	119,560 (10.01)	0.3053	<0.001
Benzodiazepines	712947 (57.92)	28,941 (80.19)	684,006 (57.25)	0.5105	<0.001
ACE inhibitors	175892 (14.29)	19,423 (53.82)	156,469 (13.10)	0.9566	<0.001
Angiotensin ii inhibitor	111652 (9.07)	16,637 (46.10)	95,015 (7.95)	0.9511	<0.001
Hypoglycemic agents	77712 (6.31)	17,854 (49.14)	59,858 (5.01)	1.1526	<0.001
NSAIDs	467833 (38.01)	16,653 (46.14)	451,180 (37.76)	0.1758	<0.001

Abbreviations: ACE, angiotensin-converting enzyme; Angiotensin II inhibitor, angiotensin receptor blocker; COPD: Chronic obstructive pulmonary disease; NSAID, nonsteroidal anti-inflammatory drug; SGLT2i, sodium–glucose cotransporter-2 inhibitor; SMD, standardized mean difference; SNRI, serotonin–norepinephrine reuptake inhibitor; SSRI, selective serotonin reuptake inhibitor; TCA, tricyclic antidepressant.

^aSGLT2 inhibitor use includes canagliflozin, dapagliflozin, empagliflozin, ertugliflozin, bexagliflozin, or sotagliflozin.

After adjusting for covariates in the Cox proportional hazards models, SGLT2i use was associated with significantly lower risks of suicidality (aHR 0.75, 95% CI 0.71–0.79; p <0.001), mortality (aHR 0.55, 95% CI 0.52–0.57; p <0.001), and inpatient hospitalization (aHR 0.71, 95% CI 0.69–0.72; p <0.001) compared with non-SGLT2i use (Table 2, Fig. 2). Sensitivity analysis using a more stringent definition of SGLT2i exposure (requiring at least two prescriptions during the study period) demonstrated consistent with main findings. Overall, 71% (n =25,466) of patients met this threshold. After adjusting for covariates, sustained SGLT2i use was associated with significantly lower risks of suicidality (aHR 0.81, 95% CI 0.76–0.86; p <0.001), mortality (aHR 0.53, 95% CI 0.50–0.56; p <0.001), and inpatient hospitalization (aHR 0.59, 95% CI 0.55–0.64; p <0.001). There was no significant difference in the risk of incident acute appendicitis between the groups (aHR 0.91, 95% CI 0.71–1.78; p =0.474).

Table 2:

Adjusted Hazard Ratios^a for Clinical Outcomes with SGLT2i Use^b in Bipolar Disorder

Outcomes			aHR (95% CI)	p-value
Suicidality			0.75 (0.71, 0.79)	<0.001
Mortality			0.55 (0.52, 0.57)	<0.001
Inpatient hospitalization			0.71 (0.69, 0.72)	<0.001
Acute appendicitis			0.91 (0.71, 1.78)	0.474
Post-Hoc Comparator Analysis (SGLT2i vs DPP-4i)
Suicidality			0.78 (0.71, 0.86)	<0.001
Mortality			0.65 (0.60, 0.71)	<0.001
SustainedSGLT2i use (at least two prescriptions during the study period)
Suicidality			0.81 (0.76, 0.86)	<0.001
Mortality			0.53 (0.50, 0.56)	<0.001
Inpatient hospitalization			0.59 (0.55, 0.64)	<0.001
Subgroup Analysisa of Suicidality in Bipolar Disorder by SGLT2i Useb
Subgroup			aHRa (95% CI)	p-value
Sex	Male		0.70 (0.64, 0.76)	<0.001
	Female		0.79 (0.73, 0.85)	<0.001
Race	White		0.77 (0.72, 0.82)	<0.001
	Black		0.75 (0.64, 0.87)	<0.001
Ethnicity	Hispanic		0.70 (0.56, 0.87)	0.001
	Non-Hispanic		0.75 (0.70, 0.80)	<0.001
Medication	Antipsychotics		0.98 (0.94, 1.03)	0.392
	Lithium		0.69 (0.61, 0.79)	0.001
	Valproate		0.71 (0.64, 0.80)	<0.001
	Lamotrigine		0.76 (0.69, 0.84)	<0.001
Somatic Comorbidity	Diabetes mellitus	Present	0.97 (0.93, 1.02)	0.229
		Absent	0.55 (0.45, 0.66)	0.001
	Chronic kidney disease	Present	0.87 (0.78, 0.97)	0.011
		Absent	0.71 (0.66, 0.76)	<0.001
	Heart Failure	Present	0.79 (0.72, 0.87)	<0.001
		Absent	0.73 (0.68, 0.79)	<0.001
Subgroup Analysisa of Mortality in Bipolar Disorder by SGLT2i Useb
Subgroup			aHRa (95% CI)	p-value
Sex	Male		0.58 (0.54, 0.62)	<0.001
	Female		0.60 (0.57, 0.65)	<0.001
Race	White		0.56 (0.55, 0.61)	<0.001
	Black		0.29 (0.19, 0.45)	<0.001
Ethnicity	HispaSnic		0.47 (0.37, 0.59)	0.001
	Non-Hispanic		0.61 (0.58, 0.65)	<0.001
Medication	Antipsychotics		0.41 (0.39, 0.44)	x<0.001
	Lithium		0.60 (0.51, 0.71)	<0.001
	Valproate		0.60 (0.54, 0.64)	<0.001
	Lamotrigine		0.44 (0.35, 0.56)	<0.001
Somatic Comorbidity	Diabetes mellitus	Present	0.38 (0.36, 0.41)	<0.001
		Absent	0.92 (0.82, 1.04)	0.200
	Chronic kidney disease	Present	0.63 (0.59, 0.68)	0.011
		Absent	0.59 (0.55, 0.63)	<0.001
	Heart Failure	Present	0.67 (0.63, 0.71)	<0.001
		Absent	0.49 (0.45, 0.53)	<0.001

Abbreviations: aHR, adjusted hazard ratio; CI, confidence interval; SGLT2i, sodium–glucose cotransporter-2 inhibitor.

^aCox regression models were adjusted for demographic factors, psychiatric comorbidities, and treatment variables. Covariates include age at index; sex; race; ethnicity; anxiety disorders; substance use disorders; sleep disorders; hypertension; diabetes mellitus; thyroid disorders; ischemic heart disease; cerebrovascular disease; heart failure; COPD; overweight and obesity; CKD; and use of antidepressants, antipsychotics, lithium, anticonvulsants, sedative-hypnotics, ACE inhibitors, angiotensin II inhibitors, and NSAIDs

^bSGLT2 inhibitor use includes canagliflozin, dapagliflozin, empagliflozin, ertugliflozin, bexagliflozin, or sotagliflozin.

Figure 2.

Adjusted Hazard Ratios for Clinical Outcomes with SGLT2i Use in Bipolar Disorder

In the active-comparator analysis restricted to individuals prescribed either SGLT2i or DPP-4i, SGLT2i use remained associated with a significantly lower risk of suicidality (aHR 0.78, 95% CI 0.71–0.86) and all-cause mortality (aHR 0.65, 95% CI 0.60–0.71) compared with DPP-4i use (Table 2).

Subgroup analyses for suicidality showed consistent protective effects across demographic strata, including males (aHR 0.70) and females (aHR 0.79), and racial/ethnic subgroups: White (aHR 0.77), Black (aHR 0.75), Hispanic (aHR 0.70), and non-Hispanic (aHR 0.75), all p ≤0.001. Benefits were also observed across most mood stabilizers—lithium (aHR 0.69), lamotrigine (aHR 0.76), and valproate (aHR 0.71), all p ≤0.001—though not with antipsychotics (aHR 0.98, p =0.392). Among patients without DM, SGLT2i use showed stronger protective effects (aHR 0.55, p =0.001) compared to those with DM (aHR 0.97, p =0.229). Protective effects were evident in patients both with (aHR 0.87, p =0.011) and without CKD (aHR 0.71, p <0.001), as well as in those with (aHR 0.79, p <0.001) and without heart failure (aHR 0.73, p <0.001).

For mortality outcomes, subgroup analyses demonstrated robust protective effects of SGLT2is across all examined strata. Benefits appear consistent across sexes, with comparable effects in males (aHR 0.58) and females (aHR 0.60). The reduction in mortality is especially notable among Black patients (aHR 0.29) compared with White patients (aHR 0.56), and among Hispanic individuals (aHR 0.47) compared with non- Hispanic individuals (aHR 0.61), all p ≤0.001. Protective effects were consistent across all mood stabilizers—antipsychotics (aHR 0.41), lamotrigine (aHR 0.44), lithium (aHR 0.60), and valproate (aHR 0.60), all p <0.001. Among individuals with DM, SGLT2i use is associated with a markedly stronger reduction in mortality (aHR 0.38, p <0.001) compared with those not on SGLT2is, whereas the effect is not significant in individuals without DM (aHR 0.92, p =0.200). Benefits were observed in patients both with (aHR 0.63, p =0.011) and without CKD (aHR 0.59, p <0.001), as well as in those with (aHR 0.67, p <0.001) and without heart failure (aHR 0.49, p <0.001). These results indicate robust benefits of SGLT2i use across clinical and demographic subgroups for both psychiatric and mortality outcomes.

Following PSM, 31,001 individuals with BD prescribed SGLT2i were matched to 31,001 nonusers. The matched cohorts were well balanced in terms of demographic characteristics, psychiatric and somatic comorbidities, and concurrent medication use (Supplementary Table 3). Kaplan–Meier survival analysis demonstrated significantly higher survival probabilities among SGLT2i users compared to nonusers across all outcomes (Tables 3a, 3b, and 3c; Supplementary Fig. 1). For suicidality, survival probabilities at each time point were: 1 year (98.55% vs. 97.69%), 2 years (97.37% vs. 96.44%), 3 years (96.42% vs. 95.33%), 4 years (95.56% vs. 94.54%), and 5 years (94.51% vs. 93.56%) for SGLT2i users versus nonusers, respectively (all p <0.001). For mortality, survival probabilities were: 1 year (96.79% vs. 93.15%), 2 years (94.51% vs. 89.25%), 3 years (92.45% vs. 85.53%), 4 years (90.76% vs. 82.41%), and 5 years (88.99% vs. 79.57%) for SGLT2i users versus nonusers, respectively (all p <0.001). For inpatient hospitalization, survival without hospitalization was: 1 year (90.69% vs. 82.41%), 2 years (85.45% vs. 77.73%), 3 years (80.87% vs. 73.61%), 4 years (76.35% vs. 70.00%), and 5 years (72.39% vs. 66.72%) for SGLT2i users versus nonusers, respectively (all p <0.001). These findings suggest a robust association between SGLT2i use and improved long-term psychiatric and overall survival outcomes in individuals with BD.

Table 3a:

Kaplan–Meier Survival Analysis of Suicidality in Bipolar Disorder by SGLT2i Use^a

Cohort	Patients with Outcome (n = 31,001)b	Survival Probability at End of Time Window	χ2	df	p-value
1 Year
SGLT2i Users	335	98.55%	49.092	1	<0.001
SGLT2i Nonusers	529	97.69%
2 Years
SGLT2i Users	521	97.37%	38.695	1	<0.001
SGLT2i Nonusers	738	96.44%
3 Years
SGLT2i Users	625	96.42%	38.595	1	<0.001
SGLT2i Nonusers	884	95.33%
4 Years
SGLT2i Users	687	95.56%	34.044	1	<0.001
SGLT2i Nonusers	964	94.54%
5 Years					<0.001
SGLT2i Users	737	94.51%	30.913	1
SGLT2i Nonusers	1,042	93.56%

Abbreviations: SGLT2i, sodium–glucose cotransporter-2 inhibitor; χ², chi-square; df, degrees of freedom.

^aSGLT2 inhibitor use includes canagliflozin, dapagliflozin, empagliflozin, ertugliflozin, bexagliflozin, or sotagliflozin.

^bPropensity score matching (n = 31,001) was used to adjust for demographic factors, psychiatric comorbidities, and treatment variables. Patients with suicidality recorded prior to the index event (n = 3,301 in the SGLT2i user cohort and n = 3,308 in the nonuser cohort) were excluded to minimize bias from pre-existing outcomes and potential EHR data irregularities, such as delayed reporting or inaccurate event dates.

Table 3b:

Kaplan–Meier Survival Analysis of Mortality in Bipolar Disorder by SGLT2i Use^a

Cohort	Patients with Outcome (n = 31,001)b	Survival Probability at End of Time Window	χ2	df	p-value
1 Year
SGLT2i Users	838	96.79%	380.693	1	<0.001
SGLT2i Nonusers	1,806	93.15%
2 Years
SGLT2i Users	1,253	94.51%	471.227	1	<0.001
SGLT2i Nonusers	2,590	89.25%
3 Years
SGLT2i Users	1,515	92.45%	550.031	1	<0.001
SGLT2i Nonusers	3,191	85.53%
4 Years
SGLT2i Users	1,657	90.76%	604.057	1	<0.001
SGLT2i Nonusers	3,594	82.41%
5 Years					<0.001
SGLT2i Users	1,756	88.99%	628.331	1
SGLT2i Nonusers	3,892	79.57%

Abbreviations: SGLT2i, sodium–glucose cotransporter-2 inhibitor; χ², chi-square; df, degrees of freedom.

^aSGLT2 inhibitor use includes canagliflozin, dapagliflozin, empagliflozin, ertugliflozin, bexagliflozin, or sotagliflozin.

^propensity score matching (n = 31,001) was used to adjust for demographic factors, psychiatric comorbidities, and treatment variables. Patients with mortality recorded on or before the index date (n = 75 in the SGLT2i user cohort and n = 127 in the nonuser cohort) were excluded to minimize bias from potential EHR data irregularities, such as delayed reporting or inaccurate death dates).

Table 3c:

Kaplan–Meier Survival Analysis of Inpatient hospitalization in Bipolar Disorder by SGLT2i Use^a

Cohort	Patients with Outcome (n = 31,001)b	Survival Probability at End of Time Window	χ2	df	p-value
1 Year
SGLT2i Users	1,567	90.69%	611.336	1	<0.001
SGLT2i Nonusers	3,170	82.41%
2 Years
SGLT2i Users	2,186	85.45%	479.063	1	<0.001
SGLT2i Nonusers	3,759	77.73%
3 Years
SGLT2i Users	2,576	80.87%	418.242	1	<0.001
SGLT2i Nonusers	4,172	73.61%
4 Years
SGLT2i Users	2,835	76.35%	371.452	1	<0.001
SGLT2i Nonusers	4,454	70.00%
5 Years					<0.001
SGLT2i Users	2,990	72.39%	347.270	1
SGLT2i Nonusers	4,663	66.72%

Abbreviations: SGLT2i, sodium–glucose cotransporter-2 inhibitor; χ², chi-square; df, degrees of freedom.

^aSGLT2 inhibitor use includes canagliflozin, dapagliflozin, empagliflozin, ertugliflozin, bexagliflozin, or sotagliflozin.

^pPropensity score matching (n = 31,001) was used to adjust for demographic factors, psychiatric comorbidities, and treatment variables. Patients with inpatient hospitalization recorded prior to the index event (n = 3,301 in the SGLT2i user cohort and n = 3,308 in the nonuser cohort) were excluded to minimize bias from pre-existing outcomes and potential EHR data irregularities, such as delayed reporting or inaccurate event dates.

In the active-comparator PSM analysis, SGLT2i and DPP-4i cohorts were well balanced following matching, with all SMDs below 0.10. Kaplan–Meier survival analysis demonstrated higher event-free survival probabilities for SGLT2i users compared with DPP-4i users for both primary outcomes: five-year survival for suicidality was 94.94% versus 94.15% (log-rank χ2 =5.146, p =0.023) and for mortality was 90.25% versus 83.72% (log-rank χ2 =117.483, p <0.001). These findings replicate the primary analysis results in a more rigorously controlled comparator framework, further supporting a robust association between SGLT2i use and reduced suicidality and mortality risk in individuals with BD (Supplementary Table 4).

DISCUSSION

This is the largest study to investigate the role of SGLT2is, a novel class of hypoglycemic agents, in psychiatric outcomes among individuals with BD. By systematically evaluating suicidality, mortality, and inpatient hospitalization, this comprehensive study of 1.23 million individuals with BD aimed to generate evidence at the intersection of psychiatry and metabolic medicine to inform clinical practice and guide future research. The findings demonstrate that individuals with BD prescribed SGLT2is had substantially lower risk across all measured outcomes, with a 25% reduction in suicidality, 29% reduction in inpatient hospitalization, and 45% reduction in mortality. The robustness of these findings is underscored by their consistency across diverse demographic groups—males and females, Blacks and Whites, Hispanic and non-Hispanic populations—and across various clinical contexts, including concurrent use of mood stabilizers (lithium, valproate, and lamotrigine) and presence or absence of cardiovascular comorbidity.

Notably, differential patterns emerged across outcomes that provide important mechanistic insights: for suicidality, the protective effect was present in individuals without DM but not significant in those with DM, suggesting a possible mechanism independent of glycemic control. Conversely, the mortality benefit was significant only in individuals with DM and not in those without DM, which likely reflects the established cardiovascular and renal protective effects of SGLT2is in individuals with DM who have higher baseline mortality risk. The mortality benefits in individuals with DM, while consistent across demographic groups, showed particularly pronounced effects in Black individuals, raising critical questions about differential baseline risks, disparities in standard care, or potential biological variations in drug response that merit urgent investigation. This is consistent with recent data, indicating dapagliflozin provides the greatest absolute benefit in Black patients, reflecting their higher baseline cardiovascular risk (Butt et al., 2023). SGLT2i use was associated with a greater mortality risk reduction among Hispanic individuals than non-Hispanic individuals (HR 0.47 versus 0.61), suggesting potential effect modification by ethnicity. The more pronounced benefit among Hispanic individuals may speculatively reflect their higher burden of comorbid metabolic conditions (Joseph et al., 2021; O’Hearn et al., 2022), rendering them more likely to benefit from the cardioprotective and renoprotective properties of SGLT2is. These findings should be considered hypothesis-generating, warranting dedicated studies in ethnically diverse BD populations.

These divergent patterns—suicidality benefits in individuals without DM and mortality benefits in individuals with DM—can be understood through the lens of BD as an inflammatory disease characterized by high levels of insulin resistance, mitochondrial dysfunction, systemic inflammation, and altered metabolic signaling (Bavaresco et al., 2020; Singh et al., 2025b; Solmi et al., 2021). SGLT2is may exert dual therapeutic effects through distinct but complementary pathways. First, SGLT2is induce mild ketosis, which may enhance mitochondrial bioenergetics, stabilize excitatory-inhibitory neurotransmitter imbalance, and confer neuroprotection by reducing reactive oxygen species, upregulating antioxidant enzymes, and reducing glutamatergic excitotoxicity (Asadinejad et al., 2025; Avgerinos et al., 2022; Ekanayake et al., 2020; Lin et al., 2014; Lupsa et al., 2023; Sa-Nguanmoo et al., 2017). Recent studies have demonstrated that SGLT2is enhance neurotrophic signaling, reduce inflammation, and may exert antidepressant effects, potentially through this ketogenic mechanism, which has been associated with improved mood and reduced depressive symptoms (Ali et al., 2025; Brietzke et al., 2018; Jensen et al., 2020; Qiu et al., 2017). Multiple ongoing ketogenic diet trials for BD are aimed at understanding these mechanistic pathways, and SGLT2i-induced ketosis could offer a novel, pharmacologically-induced treatment approach that serves as a critical metabolic pathway to BD treatment (Campbell et al., 2025a; Campbell et al., 2025b; Longhitano et al., 2024; Rigby et al., 2025). The ketosis-mediated neuroprotective effects may be most apparent in metabolically healthier individuals without DM, explaining the suicidality reduction observed in this subgroup. Second, in individuals with established DM and higher baseline cardiovascular risk, the predominant benefit may be through improvements in hyperglycemia and insulin resistance, potentially reducing cardiovascular mortality, which is a leading cause of death in this population (Hanke et al., 2025; Odutayo et al., 2021; Staplin et al., 2025).

To address concerns regarding residual imbalance after PSM and potential immortal time bias, we conducted a post hoc active- comparator sensitivity analysis restricted to SGLT2i or DPP-4i users. SGLT2i use remained associated with lower risks of all-cause mortality and suicidality compared with DPP-4i use. These findings are consistent with prior work by Chang et al. (2025), which reported reduced suicide- related events and mortality with SGLT2i use in BD and type 2 DM compared to DPP-4i use, and suggest that immortal time bias is unlikely to fully account for the observed associations.

Prior studies have shown SGLT2is to reduce the risk of depression in individuals with DM (Mui et al., 2023; Wium-Andersen et al., 2022), although in this study we were not able to measure new onset of depressive episodes due to the inherent limitations of ICD coding in EHRs. However, suicidality is a robust and clinically critical outcome, especially in BD where individuals have almost 20 times higher risk of suicide compared to the general population (Hu et al., 2023; Schaffer et al., 2015; Singh et al., 2025b). The fact that benefits persisted across different mood stabilizers and in individuals with or without CKD and heart failure suggests that both neuropsychiatric mechanisms mediated by ketosis and cardiometabolic mechanisms may be operative depending on the outcome and patient population. While observational studies cannot establish causation, these findings may support the hypothesis that ketosis may mediate the sustained antidepressant and anti-suicidal effects of SGLT2is, and future clinical trials are urgently required to test the mood stabilizing effects and mechanistic pathways of reducing suicidality in BD. These findings open new therapeutic avenues for a vulnerable population with substantially elevated mortality and suicide risk (Chan et al., 2022; Miller and Black, 2020). They underscore the complex interplay between metabolic and psychiatric pathophysiology in BD and highlight the potential for repurposing metabolic agents as adjunctive treatments to address both suicide risk and premature mortality. Particular attention must be given to health equity implications, reduced hospitalizations, and the optimization of benefits across all patient subgroups.

Strengths.

This study represents one of the largest investigations of BD to date, with 1.23 million adults. providing substantial statistical power to detect differences among groups and conduct comprehensive subgroup analyses. The consistency of findings across diverse demographic groups, concurrent mood stabilizer use, and various comorbidity profiles strengthens confidence in the robustness of the results. To address potential confounding, we employed PSM, and findings remained consistent for all three outcomes in the matched cohort. The use of clinically robust outcomes—suicidality and all-cause mortality—minimizes measurement bias, and the mortality data, while derived from EHRs, has been validated in prior studies (Harrison et al., 2020; Wang et al., 2022) and is unlikely to introduce differential bias between groups. The multiple subgroup analyses consistently demonstrated similar direction and magnitude of effects, further supporting the validity of our findings.

Limitations.

As an observational retrospective study, causality cannot be established between SGLT2i use and the observed outcomes. The diagnosis of BD was based on ICD codes, which carries inherent risk of misclassification, although this approach has been used in prior research (Kumar et al., 2023). The dataset lacks granular clinical information regarding BD severity, treatment adherence, medication efficacy, and mood symptoms, limiting our ability to fully evaluate the role of disease state on treatment outcomes. The cohort is predominantly from the United States with approximately 75% White participants, which may impact generalizability to other populations and healthcare systems. The particularly pronounced mortality benefit observed in Black individuals requires further validation, as prior research with other medication classes has shown differential effects by race (Temple and Stockbridge, 2007). Furthermore, as TriNetX includes individuals receiving care within participating healthcare systems, this study is not population- based and may underrepresent those without healthcare access, limiting generalizability. Furthermore, the absence of socioeconomic variables including income, education, occupation, and marital status may contribute to residual confounding. Residual imbalance was noted for diabetes mellitus and related treatments, likely reflecting substantial baseline differences and the strong indication bias inherent to SGLT2i prescribing. To address this, we conducted a pre-planned subgroup analysis stratified by diabetes status. As DM is both a primary therapeutic indication for SGLT2i and an independent risk factor for mortality in BD, residual confounding by DM may partially explain the observed associations, particularly for mortality outcomes (Mortada et al., 2026).

A further potential source of bias is immortal time bias, whereby SGLT2i users must survive to treatment initiation while follow-up in the comparator group begins at BD diagnosis, introducing an inherent asymmetry in at-risk time between groups. However, this is consistent with real-world clinical practice (Jain and Goldsweig, 2026; Tyrer et al., 2022). Additionally, an active-comparator analysis restricted to SGLT2i versus DPP-4i users yielded similar reductions in mortality and suicidality, suggesting that immortal time bias is less likely to fully explain the observed associations. A further limitation of EHR-based data is that mortality and prior clinical events occurring outside participating healthcare systems may be incompletely captured, potentially leading to under-ascertainment of outcomes (including all-cause mortality and suicidality) and misclassification of incident events. This may attenuate observed associations and bias estimates toward the null, as some prevalent cases may be inadvertently included due to incomplete historical records. Although acute appendicitis was included as a negative control outcome based on prior published literature (Pan et al., 2024; Su et al., 2025)., low event counts in the matched cohort yielded wide CIs, limiting the statistical power and interpretive value of this analysis. Despite these limitations, the large sample size, PSM approach, consistency across subgroups, and robust clinical outcomes strongly support the urgent need for RCTs to investigate the antidepressant, mood- stabilizing, and neuroprotective effects of SGLT2is in BD.

CONCLUSIONS

Bipolar disorder is associated with substantial functional impairment and elevated mortality rates. In this large-scale hypothesis-generating cohort study of 1.23 million adults with BD, we provide the first evidence that SGLT2 inhibitors may lower the risk of suicidality, overall hospitalization and all-cause mortality and across diverse demographic groups, various clinical contexts, as well as presence or absence of cardiovascular comorbidity. These novel findings require replication in independent cohorts and formal testing in clinical trials. While preliminary, they may have potential implications for clinical practice and underscore the need for prospective RCTs to assess efficacy, dose range, and long-term safety and effectiveness. If confirmed, these results could meaningfully impact treatment outcomes and survival in patients with BD.

Data availability statement:

To gain access to the data in the TriNetX research network, a request can be made to TriNetX (https://live.trinetx.com), but costs may be incurred, a data sharing agreement would be necessary, and no patient identifiable information can be obtained.