This cohort study examines the association of initiated and sustained use of antidiabetic medications with risk for dementia, psychiatric emergency department use, or psychiatric hospitalization among older adults in the Veterans Affairs health care system.
Key Points
Question
Is treatment with sodium-glucose cotransporter 2 (SGLT2) inhibitors associated with reduced risk of dementia and other neuropsychiatric outcomes among people with mood and psychotic disorders?
Findings
In this cohort study of 112 725 patients using data from the US Department of Veterans Affairs health care system, exposure to an SGLT2 inhibitor was associated with a significantly lower risk of incident all-cause dementia and psychiatric emergency department visits.
Meaning
These findings support further research into the benefits of SGLT2 inhibitors in individuals with psychiatric disorders and a potential shared metabolic mechanism underlying psychiatric disorders and dementia.
Abstract
Importance
Individuals with mood and psychotic disorders are at an increased risk for dementia. Sodium-glucose cotransporter 2 (SGLT2) inhibitors, a class of antidiabetic medications with mitochondrial and metabolic properties, may offer protective benefits.
Objective
To evaluate whether treatment with SGLT2 inhibitors is associated with reduced risk of incident dementia and other neuropsychiatric outcomes in patients with psychiatric disorders.
Design, Setting, and Participants
This cohort study used a target trial emulation design and data from the US Department of Veterans Affairs databases from January 1, 2016, to June 1, 2024. Participants were 65 years or older with a diagnosis of major depressive disorder, bipolar disorder, or schizophrenia spectrum disorder but without prior dementia diagnosis at baseline or history of SGLT2 inhibitor use. Analyses used marginal structural models weighted by inverse probability of treatment and censoring.
Interventions
Initiation and noninitiation of SGLT2 inhibitor were calculated using intention-to-treat (ITT) analysis, and sustained and nonsustained use of SGLT2 inhibitor for ≥3 months were estimated using per-protocol (PP) analysis.
Main Outcomes and Measures
The primary outcome was incident all-cause dementia, as defined by International Classification of Diseases–coded diagnoses. Secondary outcomes were time to psychiatric emergency department (PED) visit and time to psychiatric hospitalizations. Covariates included demographic characteristics, comorbidities, psychiatric diagnoses, and medication use.
Results
In total, there were 112 725 individuals in the sample, of whom 7631 (6.8%) were exposed to an SGLT2 inhibitor. The sample had a median (IQR) age of 74.1 (69.7-77.6) years and predominantly consisted of males (104 818 [92.8%]); 49.3% of the patients had obesity. In the ITT analysis, SGLT2 inhibitor use was associated with reduced odds of all-cause dementia (odds ratio [OR], 0.61; 95% CI, 0.52-0.73) and PED visits (OR, 0.80; 95% CI, 0.66-0.97) but not psychiatric hospitalizations (OR, 0.68; 95% CI, 0.44-1.04). In the PP analysis, SGLT2 inhibitor use was associated with lower odds of all-cause dementia (OR, 0.54; 95% CI, 0.40-0.73) and psychiatric hospitalizations (OR, 0.56; 95% CI, 0.31-1.00) but not PED visits (OR, 0.74; 95% CI, 0.53-1.05).
Conclusions and Relevance
In this cohort study of older adults with mood and psychotic disorders, SGLT2 inhibitor use was associated with lower risk of dementia and PED visits. The results support a neuroprotective role of SGLT2 inhibitors in a high-risk psychiatric population.
Introduction
People with psychiatric conditions face a markedly higher risk of dementia, suggesting transdiagnostic vulnerabilities that could open new avenues of treatment for dementia prevention.1,2,3 Previously understood as important comorbidities and medication adverse effects,4 metabolic abnormalities, including insulin resistance and mitochondrial dysfunction, are increasingly recognized as core features of both psychiatric and neurodegenerative diseases.5,6,7 Evidence suggests that brain energetics represents a shared etiological pathway contributing to cognitive decline8,9,10,11 and poor treatment response12,13,14 across diagnoses.
Therapies targeting metabolic dysfunction, such as ketogenic diets, have shown promise in neuropsychiatric conditions.15,16 Sodium-glucose cotransporter 2 (SGLT2) inhibitors are antidiabetic agents with cardioprotective and nephroprotective effects and are unique for their ketogenic properties.17,18,19 These benefits may underlie their role in the central nervous system, as ketones represent higher efficiency fuel substrates for neurons than glucose,20 and the metabolic signature of SGLT2 inhibitors in the brain mimics that of other ketogenic interventions.21,22,23 Preclinical experiments have shown SGLT2 inhibitors to be associated with improved neuronal energetics,24,25 reduced expression of inflammatory cytokines in the microglia,26 and improved proxies of depressive symptoms.27,28,29
Retrospective studies indicate that SGLT2 inhibitor use is associated with reduced risk of incident dementia and depression compared with other antidiabetic medications30,31,32,33,34,35; a recent target trial emulation study found reduced risk of Alzheimer disease and Alzheimer disease–related dementias (AD/ADRD) in individuals with diabetes.36 However, few studies have looked at neuropsychiatric outcomes in psychiatric populations. Additionally, prior studies did not account for dynamic entry times, treatment adherence, or comprehensive covariate adjustment, limiting interpretability. Marginal structural models (MSMs) are well suited for a trial emulation and offer advantages over conventional methods, such as propensity score matching.37 The US Department of Veterans Affairs (VA) health care system houses richly detailed, longitudinal data well suited to MSMs.
We hypothesized that exposure to SGLT2 inhibitors may be associated with reduced dementia incidence and related psychiatric outcomes (emergency psychiatric evaluation and hospitalization) among individuals with major depressive disorder (MDD), bipolar disorder (BD), or schizophrenia spectrum disorder (SSD). We used a dynamic target trial emulation approach within the VA health care system to evaluate whether treatment with SGLT2 inhibitors is associated with reduced risk of incident dementia and other neuropsychiatric outcomes in patients with psychiatric disorders.
Methods
Study Design and Data Sources
We conducted a retrospective cohort study with a target trial emulation design to analyze the treatment effects of SGLT2 inhibitors in a transdiagnostic group of individuals with psychiatric disorders (MDD, BD, and SSD) associated with increased liability of dementia and impairments in brain energetics. The VA Boston Research and Development Committee deemed this study exempt from ethics review and waived informed consent due to the use of existing data, per 45 CFR § 46.104. We followed the Transparent Reporting of Observational Studies Emulating a Target Trial (TARGET) reporting guideline.38
We used electronic health record data from the US VA Corporate Data Warehouse, which contains comprehensive clinical information; demographics; International Classification of Diseases, Ninth Revision (ICD-9) and International Statistical Classification of Diseases and Related Health Problems, Tenth Revision (ICD-10) diagnosis codes; prescriptions; and laboratory data (eTable 1 in Supplement 1).39 Our observation window for medical history was from January 1, 2006, to June 1, 2024. The primary analysis was conducted in November 2025.
Inclusion Criteria
Our inclusion criteria required participants to be age 65 years or older and have a history of psychiatric disease (MDD, BD, or SSD), defined as having received at least 2 ICD codes from the same disease domain at any time during the study period. The age filter was applied to enrich the sample in people at risk of incident dementia during the relatively short observation window. For each patient, the psychiatric diagnosis date corresponded to the date of the second ICD code. Since SGLT2 inhibitors were first approved in 2013, inclusion was only implemented after the date of the first recorded SGLT2 inhibitor prescriptions in the VA, which in our sample corresponded to January 26, 2016.
Patients were excluded if they had a prior diagnosis of dementia or had received prescriptions for SGLT2 inhibitors prior to start of the study. We also excluded patients with diagnosed personality disorders.
Drug Exposure
The exposure was defined as receipt of an SGLT2 inhibitor prescription (full list in eTable 2 in Supplement 1) for at least 3 months. SGLT2 inhibitor use was considered as a single category, and switching from one to another was considered as continuous treatment. Treatment discontinuation was ascertained from the last date of SGLT2 inhibitor use.
Outcomes
The primary outcome was incident all-cause dementia. All-cause dementia was defined as receiving 2 ICD codes for AD, ADRD, or nonspecific dementias using definitions comparable to prior VA dementia studies. The date of the second diagnosis was treated as the outcome date to reduce potential false-positive diagnoses.40,41
Secondary outcomes were time to psychiatric emergency department (PED) utilization and time to psychiatric hospitalization. PED utilization was defined as an ED visit with primary psychiatric symptoms (eFigure 4 in Supplement 1) as a cause of admission in the VA Emergency Department Integration Software records. Psychiatric hospitalization outcome was defined as an individual psychiatry specialty admission or transfer to a VA hospital.42
These 3 outcomes, although drawn from the same cohort, were treated as separate studies since each required its own specifications (eg, controlling for prior ED visits for incident PED). As such, instead of a single study with differing end points, we designed separate trials starting from the same cohort with levels of adjustment customized according to the hypothesis tested. PED utilization and psychiatric hospitalization trials excluded participants who had experienced outcomes within 6 months of trial start. Occurrences before trial initiation were encoded as adjusting covariates.
Covariates
Covariates included age, sex, self-reported race and ethnicity, smoking, alcohol use, body mass index, hemoglobin A1c levels, and medical history covariates listed in Table 1. Race and ethnicity categories included Hispanic, non-Hispanic Black, non-Hispanic White, and other (American Indian and Alaska Native, Asian, Native Hawaiian and Other Pacific Islander, and unknown); these data were included in this analysis because they are known modifiers of dementia risk and are covariates. Models were additionally adjusted by psychiatric diagnosis and history of medication use, including use of major antidepressants and diabetic medications, which are sometimes used in combination with or as alternative to SGLT2 inhibitors. A complete list of medications is shown in eTable 2 in Supplement 1.
Table 1. Cohort Characteristics for All-Cause Dementia Trial .
| Variable | Patients, No. (%) | |||
|---|---|---|---|---|
| Temporality | Overall (N = 112 725) | SGLT2 inhibitor | ||
| Did not initiate (n = 105 094) | Initiated (n = 7631) | |||
| Demographic | ||||
| Sex | ||||
| Female | Static | 8180 (7.3) | 7907 (7.5) | 273 (3.6) |
| Male | Static | 104 545 (92.7) | 98 087 (92.5) | 7359 (96.4) |
| Age, median (IQR), y | Dynamic | 73.9 (69.5-77.5) | 73.9 (69.4-77.5) | 74.6 (71.1-77.2) |
| Racea | ||||
| Black | Static | 21 979 (19.5) | 20 557 (19.6) | 1422 (18.6) |
| White | Static | 72 324 (64.1) | 67 448 (64.1) | 4876 (63.9) |
| Otherb | Static | 10 847 (9.6) | 10 164 (9.7) | 683 (8.9) |
| Ethnicitya | ||||
| Hispanic | Static | 7626 (6.8) | 6973 (6.6) | 654 (8.6) |
| Vital | ||||
| Obesityc | Dynamic | 55 517 (49.3) | 50 204 (47.8) | 5313 (69.6) |
| HbA1c, median (IQR), % | Dynamic | 5.8 (5.6-5.9) | 5.8 (5.6-5.9) | 6.5 (5.8-6.9) |
| Lifestyle | ||||
| Tobacco use | Dynamic | 35 444 (31.4) | 33 209 (31.6) | 2235 (29.3) |
| Alcohol use | Dynamic | 3956 (3.5) | 3817 (3.6) | 139 (1.8) |
| Other drugs | ||||
| Benzodiazepines | Dynamic | 37 383 (33.2) | 34 188 (32.5) | 3195 (41.9) |
| Antidepressants | Dynamic | 90 055 (79.9) | 83 408 (79.4) | 6647 (87.1) |
| DPP-4i | Dynamic | 6273 (5.6) | 4230 (4.0) | 2043 (26.8) |
| GLP-1RA | Dynamic | 4767 (4.2) | 2693 (2.6) | 2074 (27.2) |
| Psychiatric diagnosis at baselined | ||||
| SSD | Static, dynamic | 6073 (5.4) | 5801 (5.5) | 272 (3.5) |
| BD | Static, dynamic | 8735 (7.7) | 8089 (7.7) | 646 (8.5) |
| MDD | Static, dynamic | 97 917 (86.9) | 91 204 (86.8) | 6713 (88.0) |
| MDD subtype | ||||
| Psychotic MDD | Static, dynamic | 2412 (2.1) | 2243 (2.1) | 169 (2.2) |
| Comorbidities | ||||
| T2D | Dynamic | 48 897 (43.4) | 41 859 (39.8) | 7038 (92.2) |
| Hypertension | Dynamic | 93 974 (83.4) | 86 594 (82.4) | 7380 (96.7) |
| PVD | Dynamic | 14 916 (13.2) | 13 254 (12.6) | 1662 (21.8) |
| CAD | Dynamic | 52 677 (46.7) | 47 416 (45.1) | 5261 (68.9) |
| KD | Dynamic | 28 153 (25.0) | 24 724 (23.5) | 3429 (44.9) |
| TBI | Dynamic | 4688 (4.2) | 4321 (4.1) | 367 (4.8) |
| PTSD | Dynamic | 46 660 (41.4) | 43 144 (41.1) | 3516 (46.1) |
| Neurodegeneration | Dynamic | 7589 (6.7) | 7050 (6.7) | 539 (7.1) |
| Follow-up | ||||
| Follow-up time, median (IQR), y | NA | 3.3 (1.5-5.3) | 3.0 (1.3-5.3) | 4.8 (3.2-6.4) |
| Lost to follow-upe | NA | 10 377 (9.2) | 10 066 (9.6) | 311 (4.1) |
| Outcome | ||||
| All-cause dementia | NA | 4616 (4.1) | 4485 (4.3) | 131 (1.7) |
| Death | NA | 17 891 (15.9) | 17 387 (16.5) | 504 (6.6) |
Abbreviations: BD, bipolar disorder; CAD, coronary artery disease; DPP-4i, dipeptidyl peptidase-4 inhibitor; GLP-1RA, glucagon-like peptide-1 receptor agonist; HbA1c, hemoglobin A1c; KD, kidney disease; MDD, major depressive disorder; NA, not applicable; PTSD, posttraumatic stress disorder; PVD, peripheral vascular disease; SGLT2, sodium-glucose cotransporter 2; SSD, schizophrenia spectrum disorders; T2D, type 2 diabetes; TBI, traumatic brain injury.
SI conversion factor: To convert HbA1c to the proportion of total hemoglobin, multiply by 0.01.
Race and ethnicity were self-reported. Non-Hispanic Black race and Hispanic ethnicity were used as static covariates. Non-Hispanic White was the majority racial group and not included as a covariate for adjustment.
Other race included American Indian and Alaska Native, Asian, Native Hawaiian and Other Pacific Islander, and unknown. The category was not used as a covariate.
Body mass index (calculated as weight in kilograms divided by height in meters squared) greater than or equal to 30.
New diagnostic codes were dynamically time-updated.
More than 1 period from end of study window if no outcome or censor.
In addition to static variables (demographic and baseline psychiatric diagnosis), the emulation model included time-varying covariates (comorbidities, medications, and laboratories) that were updated each period using the most recent measurement within 2 years prior to trial entry and then the latest available data or, if unavailable, the last observed value (last observation carried forward). Time since psychiatric disease diagnosis was also modeled as a dynamic covariate. Additional time-varying variables in the analysis of secondary outcomes (PED utilization and psychiatric hospitalization) included the cumulative sum of ED visits, the total number of prior hospitalizations, and time elapsed since the most recent PED visit or hospitalization.
Trial Emulation Design
We implemented a dynamic trial emulation framework using MSMs in the TrialEmulation R package, version 4.0.343,44 (R Project for Statistical Computing) to estimate the treatment effect of SGLT2 inhibitor treatment. Our target estimands were (1) intention-to-treat (ITT, primary analysis) treatment effect estimate of initiating an SGLT2 inhibitor at baseline, regardless of subsequent adherence, vs not initiating1 and (2) per-protocol (PP, secondary analysis) treatment effect estimate of sustained SGLT2 inhibitor use vs sustained nonuse, reflecting biological effectiveness under continuous treatment.2 Both estimands were defined over a maximum follow-up of 7.5 years (30 periods).
The trial window of January 1, 2016, to June 1, 2024, was divided into discrete 3-month periods. Eligibility was dynamic: patients who were ineligible at and earlier period (eg, had not yet reached 65 years of age or had not yet received a second psychiatric diagnosis) could enter later trials once they satisfied all criteria. To minimize bias from reverse causal association, patients were additionally required to remain in the cohort for at least 6 months after meeting initial eligibility criteria (age ≥65 years and ≥2 psychiatric ICD codes). Time 0 for each emulated trial was therefore defined as the first period following this 6-month run-in period.
At each trial baseline (time 0), treatment assignment was determined by SGLT2 inhibitor status. Patients were classified as initiators only if they had sustained SGLT2 inhibitor use for at least one 3-month period; all other participants were assigned to the nontreatment arm. For ITT analyses, patients were followed up under their baseline assignment regardless of subsequent treatment changes. For PP analyses, patients were censored on deviation from their assigned strategy (discontinuation in the treatment arm or initiation in the control arm). Patients were followed up from their trial-specific time 0 until dementia diagnosis, death, loss to follow-up, treatment deviation (PP analysis only), or June 1, 2024, whichever occurred first. Death was treated as a censoring event.45,46 A schematic is presented in eFigure 1 in Supplement 1.
Statistical Analysis
MSMs were used to estimate ITT and PP treatment effects via weighted pooled logistic regression. For the PP analysis, inverse probability of treatment weights were applied to correct for selection bias introduced by artificial censoring at treatment deviation; inverse probability of censoring weights were additionally applied to address informative censoring due to natural loss to follow-up. For the ITT analysis, only inverse probability of censoring weights were applied to address natural censoring. Nonrandom treatment assignment at baseline was accounted for by including baseline covariates in the pooled logistic outcome model. Cause-specific hazards were estimated for both ITT and PP analyses; death was operationalized via censoring but treated as a competing event in the interpretation of risk.47 Full model specification is provided in the eMethods in Supplement 1.
Statistical significance was defined using 2-sided α = .05. Robust SEs were calculated using the sandwich variance estimator; P values for MSM coefficients were obtained via nonparametric bootstrapping.43,44 Odds ratios (ORs) from the pooled logistic regression were interpreted as approximations to hazard ratios for time-to-event outcomes. Treatment coefficients and corresponding significance at 90% and 95% were estimated.
Since our primary sample was mainly composed of patients with MDD, which could mask results specific to bipolar and psychotic disorders, we conducted a sensitivity analysis restricted to individuals with severe psychiatric conditions (psychotic MDD, BD, SSD). This analysis aimed to explore the direction of associations between SGLT2 inhibitor exposure and relevant outcomes in these high-risk groups, who reportedly had elevated dementia risk and impaired brain metabolism. To mitigate low power, we expanded the sample by lowering the eligibility criterion for trial entry to age 60 years.
To assess the association of potential differential SGLT2 inhibitor exposure with diabetes indication, we stratified the trial emulation in the primary dementia cohort by baseline diabetes status. Given its high prevalence in the VA population and its overlapping neuroinflammatory profiles with BD, MDD, or SSD, we also performed subgroup analyses stratified by baseline posttraumatic stress disorder (PTSD) status.
Death was censored in the primary analyses. Because death precludes the subsequent diagnosis of dementia and may differentially affect treatment groups, we computed competing-risk cumulative incidence functions (CIFs) (eMethods in Supplement 1) as well as performed a sensitivity analysis in which death and dementia were modeled as a composite outcome.
Results
Cohort Characteristics
Baseline patient-level demographic and clinical characteristics for all outcomes are shown in Table 1. Person-trial level characteristics for all outcomes are provided in eTable 3 in Supplement 1. We identified a total of 112 725 patients who met our inclusion criteria and observation window for the all-cause dementia trial. Patients had a median (IQR) age of 74.1 (69.7-77.6) years and included 104 818 males (92.8%) and 8180 females (7.3%). The median (IQR) follow-up was 3.3 (1.3-5.3) years for the all-cause dementia trial. Similar characteristics were observed in other outcomes. Cohort derivation is shown in eFigure 2 in Supplement 1.
In the all-cause dementia trial, 7631 (6.8%) individuals initiated SGLT2 inhibitor treatment and 4616 (4.1%) developed dementia during follow-up (Table 1). When examining characteristics at the person-period level (expanding from the individual level), the total observations numbered 1 666 903 or a mean (SD) of 14.8 (3.8) periods per patient (3.7 years), with 43 998 SGLT2 inhibitor–treated patient-periods.
For the PED use trial, there was a total of 105 963 patients with 7044 SGLT2 inhibitor users and a total of 3276 (3.1%) outcomes, for a total of 1 574 644 patient-periods and 40 613 treated patient-periods. Patients had a median (IQR) age of 71.2 (65.8-73.9) years and included 98 223 males (92.7%) and 7742 females (7.3%).
For the psychiatric hospitalization trial, there were 106 693 patients with 7340 users and 1199 outcomes (1.1%), yielding a total of 1 575 994 patient-periods with 42 181 treated patient-periods (eTable 3 in Supplement 1). Patients had a median (IQR) age of 71.2 (65.8-73.9) years and included 98 856 males (92.6%) and 7850 females (7.4%).
Treatment Effect Estimates
In the all-cause dementia trial, the OR was 0.61 (95% CI, 0.52-0.73) in the ITT analysis and 0.54 (95% CI, 0.40-0.73) in the PP analysis. In the PED use trial, the OR was 0.80 (95% CI, 0.66-0.97) in the ITT analysis and 0.74 (95% CI, 0.53-1.05) in the PP analysis. In the psychiatric hospitalization trial, the OR was 0.68 (95% CI, 0.44-1.04) in the ITT analysis and 0.56 (95% CI, 0.31-1.00) in the PP analysis. The log-odds are provided in Table 2.
Table 2. Main Cause-Specific Treatment Effect Estimates of Sodium-Glucose Cotransporter 2 Inhibitor Use in All-Cause Dementia, PED Utilization, and Psychiatric Hospitalization.
| Outcome | Main analysis | Sensitivity analysisa | ||
|---|---|---|---|---|
| ITT treatment coefficient, −log OR (95% CI) | PP treatment coefficient, −log OR (95% CI) | ITT treatment coefficient, −log OR (95% CI) | PP treatment coefficient, −log OR (95% CI) | |
| All-cause dementia | −0.49 (−0.66 to −0.31)b | −0.61 (−0.92 to −0.31)b | −0.37 (−0.78 to 0.04)c | −0.88 (−1.60 to −0.15)b |
| PED use | −0.22 (−0.42 to −0.03)b | −0.30 (−0.64 to 0.05)c | −0.36 (−0.74 to 0.03)c | −0.28 (−1.05 to 0.47) |
| Psychiatric hospitalization | −0.38 (−0.81 to 0.04)c | −0.58 (−1.16 to 0.00)b | NA | NA |
Abbreviations: ITT, intention to treat; NA, not applicable; OR, odds ratio; PED, psychiatric emergency department; PP, per protocol.
Sensitivity analysis was not considered for psychiatric hospitalization due to lack of outcome events in the treatment group.
Treatment coefficients significant at 95% level.
Treatment coefficients significant at 90% level.
Because ITT models were anchored to well-defined treatment baseline in treated and untreated groups, cause-specific CIFs were estimated for these analyses. For all-cause dementia, fitted CIFs showed a mean reduction of 1.3 (95% CI, 1.0-1.7) percentage points in 5-year risk of dementia for patients with initiated (1.6%; 95% CI, 1.2%-2.0%) vs noninitiated SGLT2 inhibitor (2.9%; 95% CI, 2.7%-3.1%). For the PED use and psychiatric hospitalization outcomes, SGLT2 inhibitor initiation was associated with dementia risk reductions of 0.7 (95% CI, 0.0-1.2) percentage points (initiated: 3.2% [95% CI, 2.7%-3.8%]; noninitiated: 3.8% [95% CI, 3.6%-4.2%]) and 0.2 (95% CI, 0.0-0.3) percentage points (initiated: 0.5% [95% CI, 0.4%-0.6%]; noninitiated: 0.7% [95% CI, 0.6%-0.8%]), respectively (Figure). Weight distributions for all outcomes and covariate Love plots for dementia are shown in eFigure 3 and eFigure 5, respectively, in Supplement 1.
Figure. Line Graphs of Estimated Cumulative Incidence Function Curves for Initiators vs Noninitiators for All Outcomes Over 5 Years for the Intention-to-Treat Analysis .
PED indicates psychiatric emergency department; SGLT2, sodium-glucose cotransporter 2 inhibitor.
Results of Sensitivity Analyses
Baseline demographic and clinical characteristics of the separately constructed psychotic subtype sample are shown in eTable 4 in Supplement 1. Results of the sensitivity analysis showed that the ITT analysis yielded an OR toward all-cause dementia of 0.69 (95% CI, 0.46-1.04). The PP analysis yielded an OR of 0.41 (95% CI, 0.20-0.86). For PED use, the OR was 0.70 (95% CI, 0.48-1.03) for the ITT analysis and 0.82 (95% CI, 0.35-1.60) for the PP analysis (log-odds in Table 2).
Analyses stratified by diabetes and PTSD diagnoses at baseline showed consistent protective benefits associated with initiation of SGLT2 inhibitors in all subgroups (eTables 5 and 6 in Supplement 1). Competing-risk CIFs showed concordant reductions in both dementia and death associated with SGLT2 inhibitor initiation (eFigure 6 in Supplement 1). Treatment effect estimates were also consistent in composite outcome analyses (eTable 7 in Supplement 1).
Discussion
In this trial emulation study conducted with data from the VA health care system, we found that initiation of SGLT2 inhibitor treatment was associated with a significantly reduced risk of incident dementia among older individuals diagnosed with MDD, BD, or SSD. We also found that SGLT2 inhibitor initiation was associated with reduced PED visits and that SGLT2 inhibitor adherence was associated with reduced psychiatric hospitalizations. Our findings point to SGLT2 inhibitors as candidate therapeutics in mood and psychotic disorders as well as populations with high neuropsychiatric risk and limited treatment options.
These results align with findings of prior studies in nonpsychiatric populations36 and observational studies performed across geographies.31,33,48,49 A meta-analysis of randomized trial data on SGLT2 inhibitors and AD yielded equivocal results, reflecting potential exclusion of participants with psychiatric disorders or cognitive impairment.50 A Danish registry study30 found that individuals receiving SGLT2 inhibitors had a lower risk of depression than with any other antidiabetic medication (OR, 0.47; 95% CI, 0.38-0.58) and even lower than in a nondiabetic comparator group (OR, 0.63; 95% CI, 0.61-0.65), suggesting treatment effects for mood beyond diabetes control.30 In the same dataset, SGLT2 inhibitors were associated with the lowest nominal risk of dementia among antidiabetic drugs.31 Additional trial emulation work comparing glucagon-like peptide-1 receptor agonist, SGLT2 inhibitors, and dipeptidyl peptidase-4 inhibitors found no significant differences in hospitalization rates.51 A study in individuals with BD reported reduced suicide risk with SGLT2 inhibitor use.52
Taken together with all-cause dementia, PED use, and psychiatric hospitalization outcomes and the composite dementia-death analysis (eTable 7 and eFigure 6 in Supplement 1), our findings support prior studies and suggest an overall protective profile. Moreover, our findings confirm that this protective association is not attributed to differential mortality.
A growing body of evidence supports an overlapping pathophysiological mechanism between psychiatric disease and dementia, including impaired insulin signaling, mitochondrial dysfunction, and inflammation affecting the central nervous system.9,53,54 Our results support reconceptualizing neuropsychiatric disorders as partly metabolically mediated brain diseases55,56 and strengthening the case for SGLT2 inhibitors as candidate therapeutics.
SGLT2 inhibitors induce ketosis and replicate key metabolic features of ketogenic therapies,21,22,23,54 which have been linked to significant improvements in psychiatric symptoms in pilot studies.16,57 One randomized study found that empagliflozin improves mood in augmentation.58 Our findings provide more support for clinical studies of these medications.
Strengths and Limitations
A strength of this study is that the MSM approach is suited to large, longitudinal datasets from the VA health care system. It also accounted for time-varying confounding, immortal time bias, and informative censoring from loss to follow-up or treatment deviation.
Among the study limitations is that, despite adjustment and the trial emulation design, residual confounding may persist from variables such as severity of psychiatric illness, baseline cognitive function, and differential exposure related to diabetes indications. Additionally, the limited precision of diagnoses based on ICD codes and prescription codes and simplification of assumptions for dynamic covariates, such as the last-observation-carried-forward approach, may contribute bias. Despite the large sample size, the predominantly male population limits generalizability. While some adverse effects of SGLT2 inhibitors, such as urinary tract infections and candidal vaginitis, seemed to be higher in females,59 there was no clear evidence that SGLT2 inhibitor–induced metabolic changes differed by sex. The sample was primarily composed of individuals with MDD. Although subgroup analyses in psychotic subtypes, as well as exploratory analyses in diabetes and PTSD groups, suggested consistent directionality in treatment effects, these results were underpowered and should be interpreted carefully; future studies should examine these diagnostic groups as data become available. Future studies may explore whether SGLT2 inhibitors are associated with improved outcomes in PTSD, where emerging evidence implicates neuroinflammation and metabolic abnormalities. Sensitivity analyses suggested the associations held regardless of PTSD diagnosis (eTable 6 in Supplement 1).
Although psychiatric emergency visits and hospitalizations served as proxies for symptom severity, more granular clinical or biomarker data were not included and should be considered in the future. Studies also need to evaluate the association between SGLT2 inhibitor treatment and psychiatric outcomes in other datasets and to evaluate differential treatment effects by diagnosis, age, and metabolic factors. Given that SGLT2 inhibitors now have indications in chronic kidney disease and heart failure, it may be possible to validate whether any protective properties are independent of diabetes diagnosis (exploratorily suggested in eTable 5 in Supplement 1). Although in our study we comprehensively controlled for medication use and changes in status over time, adjusting for total prior exposure to the medications as a function of time was not included; future studies with larger samples should also consider this aspect. The diagnostic criterion of having 2 ICD codes for 1 of the conditions included could introduce diagnostic ascertainment inaccuracy and possible misclassification but aligns with best practices in the literature.
Conclusions
In this cohort study among older individuals with mood and psychotic disorders, we found that exposure to SGLT2 inhibitor was associated with reduced risk of all-cause dementia and, in a secondary analysis, reduced risk of PED use. These findings support the hypothesis of shared metabolic vulnerability across psychiatric and neurodegenerative diseases. Further investigation is warranted of SGLT2 inhibitors as potential transdiagnostic treatment options in individuals with high-risk psychiatric disorders.
eMethods. Trial Emulation Methodology Details
eTable 1. Comorbidity Definitions and ICD Codes Used
eTable 2. Complete List of Medications Considered
eTable 3. Demographic and Clinical Characteristics of Cohort for All Outcomes at Person-Period Level for the Main Analysis Cohort (BIP, SCZ, MDD). Outcome-Specific Covariates Are Shown in the Bottom Rows.
eTable 4. Demographic and Clinical Characteristics of Cohort for Outcomes I, II at Person-Period Level for the Sensitivity Analysis Cohort (BIP, SCZ, PMDD)
eTable 5. Sensitivity Analyses for Subgroups Stratified by Diabetes Status at Baseline for Emulated Trial for Dementia Outcome
eTable 6. Sensitivity Analyses for Subgroups Stratified by PTSD Status at Baseline for Emulated Trial for Dementia Outcome
eTable 7. Effects of SGLT2-i Treatment on Composite Outcome of Dementia or Death, in the Primary Dementia Cohort
eFigure 1. Schematic Diagram of Trial Emulation Design
eFigure 2. Cohort Derivation Diagram
eFigure 3. Truncated Distributions of Combined Inverse Censoring and Treatment Weights of Marginal Structural Models (Main Analysis). Weights Are Shown for ITT and PP Effects for All Outcomes. Densities Are Shown in Log Scale.
eFigure 4. Complete List of Psychiatric Criteria Keywords in Emergency Department Admission Descriptions
eFigure 5. Love Plot of Standardized Mean Differences (SMDs) for All Covariates in ITT and PP Analyses on Dementia Outcome
eFigure 6. Estimated Cumulative Incidence of Dementia, Cumulative Incidence of Death, and Event-Free Survival Over Follow-Up. Estimates Were Derived From Weighted Pooled Logistic Regression Models Accounting for Competing Risks; Event-Free Survival Denotes Remaining Alive and Free of Dementia.
eReferences
Data Sharing Statement
References
- 1.Stroup TS, Olfson M, Huang C, et al. Age-specific prevalence and incidence of dementia diagnoses among older US adults with schizophrenia. JAMA Psychiatry. 2021;78(6):632-641. doi: 10.1001/jamapsychiatry.2021.0042 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 2.Ahearn EP, Szymanski BR, Chen P, Sajatovic M, Katz IR, McCarthy JF. Increased risk of dementia among veterans with bipolar disorder or schizophrenia receiving care in the VA health system. Psychiatr Serv. 2020;71(10):998-1004. doi: 10.1176/appi.ps.201900325 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 3.Elser H, Horváth-Puhó E, Gradus JL, et al. Association of early-, middle-, and late-life depression with incident dementia in a Danish Cohort. JAMA Neurol. 2023;80(9):949-958. doi: 10.1001/jamaneurol.2023.2309 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 4.Goff DC, Cather C, Evins AE, et al. Medical morbidity and mortality in schizophrenia: guidelines for psychiatrists. J Clin Psychiatry. 2005;66(2):183-194. doi: 10.4088/jcp.v66n0205 [DOI] [PubMed] [Google Scholar]
- 5.Stein A, Zhu C, Du F, Öngür D. Magnetic resonance spectroscopy studies of brain energy metabolism in schizophrenia: progression from prodrome to chronic psychosis. Curr Psychiatry Rep. 2023;25(11):659-669. doi: 10.1007/s11920-023-01457-1 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 6.Lam XJ, Xu B, Yeo PL, Cheah PS, Ling KH. Mitochondria dysfunction and bipolar disorder: from pathology to therapy. IBRO Neurosci Rep. 2023;14:407-418. doi: 10.1016/j.ibneur.2023.04.002 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 7.Xu L, Liu R, Qin Y, Wang T. Brain metabolism in Alzheimer’s disease: biological mechanisms of exercise. Transl Neurodegener. 2023;12(1):33. doi: 10.1186/s40035-023-00364-y [DOI] [PMC free article] [PubMed] [Google Scholar]
- 8.Hamer JA, Testani D, Mansur RB, Lee Y, Subramaniapillai M, McIntyre RS. Brain insulin resistance: a treatment target for cognitive impairment and anhedonia in depression. Exp Neurol. 2019;315:1-8. doi: 10.1016/j.expneurol.2019.01.016 [DOI] [PubMed] [Google Scholar]
- 9.Henkel ND, Wu X, O’Donovan SM, et al. Schizophrenia: a disorder of broken brain bioenergetics. Mol Psychiatry. 2022;27(5):2393-2404. doi: 10.1038/s41380-022-01494-x [DOI] [PubMed] [Google Scholar]
- 10.Rowland LM, Pradhan S, Korenic S, et al. Elevated brain lactate in schizophrenia: a 7 T magnetic resonance spectroscopy study. Transl Psychiatry. 2016;6(11):e967. doi: 10.1038/tp.2016.239 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 11.Yu L, Jin J, Xu Y, Zhu X. Aberrant energy metabolism in Alzheimer’s disease. J Transl Int Med. 2022;10(3):197-206. doi: 10.2478/jtim-2022-0024 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 12.Calkin C, McClelland C, Cairns K, Kamintsky L, Friedman A. Insulin resistance and blood-brain barrier dysfunction underlie neuroprogression in bipolar disorder. Front Psychiatry. 2021;12:636174. doi: 10.3389/fpsyt.2021.636174 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 13.Calkin CV, Chengappa KNR, Cairns K, et al. Treating Insulin Resistance With Metformin as a Strategy to Improve Clinical Outcomes in Treatment-Resistant Bipolar Depression (the TRIO-BD Study): a randomized, quadruple-masked, placebo-controlled clinical trial. J Clin Psychiatry. 2022;83(2):21m14022. doi: 10.4088/JCP.21m14022 [DOI] [PubMed] [Google Scholar]
- 14.Calkin CV, Ruzickova M, Uher R, et al. Insulin resistance and outcome in bipolar disorder. Br J Psychiatry. 2015;206(1):52-57. doi: 10.1192/bjp.bp.114.152850 [DOI] [PubMed] [Google Scholar]
- 15.Grammatikopoulou MG, Goulis DG, Gkiouras K, et al. To keto or not to keto? A systematic review of randomized controlled trials assessing the effects of ketogenic therapy on Alzheimer disease. Adv Nutr. 2020;11(6):1583-1602. doi: 10.1093/advances/nmaa073 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 16.Sethi S, Wakeham D, Ketter T, et al. Ketogenic diet intervention on metabolic and psychiatric health in bipolar and schizophrenia: a pilot trial. Psychiatry Res. 2024;335:115866. doi: 10.1016/j.psychres.2024.115866 [DOI] [PubMed] [Google Scholar]
- 17.Anker SD, Butler J, Filippatos G, et al. ; EMPEROR-Preserved Trial Investigators . Empagliflozin in heart failure with a preserved ejection fraction. N Engl J Med. 2021;385(16):1451-1461. doi: 10.1056/NEJMoa2107038 [DOI] [PubMed] [Google Scholar]
- 18.Leaf DE, Bansal N. Another win for SGLT2 inhibitors in chronic kidney disease. N Engl J Med. 2025;392(8):822-823. doi: 10.1056/NEJMe2500233 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 19.Lupsa BC, Kibbey RG, Inzucchi SE. Ketones: the double-edged sword of SGLT2 inhibitors? Diabetologia. 2023;66(1):23-32. doi: 10.1007/s00125-022-05815-1 [DOI] [PubMed] [Google Scholar]
- 20.Sethi S, Ford JM. The role of ketogenic metabolic therapy on the brain in serious mental illness: a review. J Psychiatr Brain Sci. 2022;7(5):e220009. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 21.Avgerinos KI, Mullins RJ, Vreones M, et al. Empagliflozin induced ketosis, upregulated IGF-1/insulin receptors and the canonical insulin signaling pathway in neurons, and decreased the excitatory neurotransmitter glutamate in the brain of non-diabetics. Cells. 2022;11(21):3372. doi: 10.3390/cells11213372 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 22.Hone-Blanchet A, Antal B, McMahon L, et al. Acute administration of ketone beta-hydroxybutyrate downregulates 7T proton magnetic resonance spectroscopy-derived levels of anterior and posterior cingulate GABA and glutamate in healthy adults. Neuropsychopharmacology. 2023;48(5):797-805. doi: 10.1038/s41386-022-01364-8 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 23.Campbell IH, Needham N, Grossi H, et al. A pilot study of a ketogenic diet in bipolar disorder: clinical, metabolic and magnetic resonance spectroscopy findings. BJPsych Open. 2025;11(2):e34. doi: 10.1192/bjo.2024.841 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 24.Pawlos A, Broncel M, Woźniak E, Gorzelak-Pabiś P. Neuroprotective effect of SGLT2 inhibitors. Molecules. 2021;26(23):7213. doi: 10.3390/molecules26237213 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 25.Gui Z, Wang J, Zhang Y, et al. Dapagliflozin improves diabetic cognitive impairment via indirectly modulating the mitochondria homeostasis of hippocampus in diabetic mice. Biofactors. 2024;50(1):145-160. doi: 10.1002/biof.1998 [DOI] [PubMed] [Google Scholar]
- 26.Heimke M, Lenz F, Rickert U, Lucius R, Cossais F. Anti-inflammatory properties of the SGLT2 inhibitor empagliflozin in activated primary microglia. Cells. 2022;11(19):11. doi: 10.3390/cells11193107 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 27.Borikar SP, Chitode GV, Tapre DN, Lokwani DK, Jain SP. Empagliflozin ameliorates olfactory bulbectomy-induced depression by mitigating oxidative stress and possible involvement of brain derived neurotrophic factor in diabetic rats. Int J Neurosci. 2025;135(12):1502-1518. [DOI] [PubMed] [Google Scholar]
- 28.Fathy N, Labib MA, Essam RM, El-Boghdady NA. The interplay between MiR-134/BDNF and LKB1/AMPK/SIRT1 accentuates the antidepressant efficacy of empagliflozin in ovariectomized rats. ACS Chem Neurosci. 2024. doi: 10.1021/acschemneuro.4c00313 [DOI] [PubMed] [Google Scholar]
- 29.Muhammad RN, Albahairy MA, Abd El Fattah MA, Ibrahim WW. Empagliflozin-activated AMPK elicits neuroprotective properties in reserpine-induced depression via regulating dynamics of hippocampal autophagy/inflammation and PKCζ-mediated neurogenesis. Psychopharmacology (Berl). 2024;241(12):2565-2584. doi: 10.1007/s00213-024-06663-0 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 30.Wium-Andersen IK, Osler M, Jørgensen MB, Rungby J, Wium-Andersen MK. Diabetes, antidiabetic medications and risk of depression - a population-based cohort and nested case-control study. Psychoneuroendocrinology. 2022;140:105715. doi: 10.1016/j.psyneuen.2022.105715 [DOI] [PubMed] [Google Scholar]
- 31.Wium-Andersen IK, Osler M, Jørgensen MB, Rungby J, Wium-Andersen MK. Antidiabetic medication and risk of dementia in patients with type 2 diabetes: a nested case-control study. Eur J Endocrinol. 2019;181(5):499-507. doi: 10.1530/EJE-19-0259 [DOI] [PubMed] [Google Scholar]
- 32.Mui JV, Li L, Chou OHI, et al. Comparing sodium-glucose cotransporter 2 inhibitors and dipeptidyl peptidase-4 inhibitors on new-onset depression: a propensity score-matched study in Hong Kong. Acta Diabetol. 2023;60(7):917-927. doi: 10.1007/s00592-023-02063-6 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 33.Mui JV, Zhou J, Lee S, et al. Sodium-glucose cotransporter 2 (SGLT2) inhibitors vs. dipeptidyl peptidase-4 (DPP4) inhibitors for new-onset dementia: a propensity score-matched population-based study with competing risk analysis. Front Cardiovasc Med. 2021;8:747620. doi: 10.3389/fcvm.2021.747620 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 34.Akimoto H, Tezuka K, Nishida Y, Nakayama T, Takahashi Y, Asai S. Association between use of oral hypoglycemic agents in Japanese patients with type 2 diabetes mellitus and risk of depression: a retrospective cohort study. Pharmacol Res Perspect. 2019;7(6):e00536. doi: 10.1002/prp2.536 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 35.Wu CY, Iskander C, Wang C, et al. Association of sodium-glucose cotransporter 2 inhibitors with time to dementia: a population-based cohort study. Diabetes Care. 2023;46(2):297-304. doi: 10.2337/dc22-1705 [DOI] [PubMed] [Google Scholar]
- 36.Tang H, Donahoo WT, DeKosky ST, et al. Heterogeneous treatment effects of GLP-1RAs and SGLT2is on risk of Alzheimer’s disease and related dementia in patients with type 2 diabetes: insights from a real-world target trial emulation. Alzheimers Dement. 2025;21(6):e70313. doi: 10.1002/alz.70313 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 37.Hernán MA, Robins JM. Using big data to emulate a target trial when a randomized trial is not available. Am J Epidemiol. 2016;183(8):758-764. doi: 10.1093/aje/kwv254 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 38.Cashin AG, Hansford HJ, Hernán MA, et al. Transparent Reporting of Observational Studies Emulating a Target Trial-The TARGET Statement. JAMA. 2025;334(12):1084-1093. doi: 10.1001/jama.2025.13350 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 39.Price LE, Shea K, Gephart S. The Veterans Affairs’s Corporate Data Warehouse: uses and implications for nursing research and practice. Nurs Adm Q. 2015;39(4):311-318. doi: 10.1097/NAQ.0000000000000118 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 40.Ramos-Cejudo J, Corrigan JK, Zheng C, et al. Antidepressant exposure and long-term dementia risk in a nationwide retrospective study on US veterans with midlife major depressive disorder. Alzheimers Dement. 2024;20(6):4106-4114. doi: 10.1002/alz.13853 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 41.Logue MW, Miller MW, Sherva R, et al. ; Million Veteran Program . Alzheimer’s disease and related dementias among aging veterans: examining gene-by-environment interactions with post-traumatic stress disorder and traumatic brain injury. Alzheimers Dement. 2023;19(6):2549-2559. doi: 10.1002/alz.12870 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 42.Vincent BM, Wiitala WL, Burns JA, Iwashyna TJ, Prescott HC. Using Veterans Affairs Corporate Data Warehouse to identify 30-day hospital readmissions. Health Serv Outcomes Res Methodol. 2018;18(3):143-154. doi: 10.1007/s10742-018-0178-3 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 43.Su L, Rezvani R, Seaman SR, Starr C, Gravestock I. TrialEmulation: an R package to emulate target trials for causal analysis of observational time-to-event data. arXiv. Preprint posted online February 19, 2024. doi: 10.48550/arXiv.2402.12083 [DOI]
- 44.Danaei G, García Rodríguez LA, Cantero OF, Logan RW, Hernán MA. Electronic medical records can be used to emulate target trials of sustained treatment strategies. J Clin Epidemiol. 2018;96:12-22. doi: 10.1016/j.jclinepi.2017.11.021 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 45.Caniglia EC, Rojas-Saunero LP, Hilal S, et al. Emulating a target trial of statin use and risk of dementia using cohort data. Neurology. 2020;95(10):e1322-e1332. doi: 10.1212/WNL.0000000000010433 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 46.Zimmerman SC, Choi M, Jiang C, et al. Statin initiation and dementia incidence in a large health care system from 1997 to 2020: a target trial emulation study. Neurology. 2025;105(2):e213855. doi: 10.1212/WNL.0000000000213855 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 47.Lau B, Cole SR, Gange SJ. Competing risk regression models for epidemiologic data. Am J Epidemiol. 2009;170(2):244-256. doi: 10.1093/aje/kwp107 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 48.Kim HK, Biessels GJ, Yu MH, et al. SGLT2 inhibitor use and risk of dementia and Parkinson disease among patients with type 2 diabetes. Neurology. 2024;103(8):e209805. doi: 10.1212/WNL.0000000000209805 [DOI] [PubMed] [Google Scholar]
- 49.Chuang WL, Tsai SC, Yang YC, et al. Correlation between the utilization of sodium-glucose cotransporter-2 (SGLT2) inhibitors and the risk of dementia: a nationwide population-based study in Taiwan. Naunyn Schmiedebergs Arch Pharmacol. 2025;398(5):5835-5843. doi: 10.1007/s00210-024-03660-0 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 50.Jaiswal V, Mashkoor Y, Raj N, Rajak K, Jaiswal A, Fonarow GC. Association between SGLT2 inhibitors and risk of dementia and Parkinson’s disease: a meta-analysis of 12 randomized controlled trials. Am J Med. 2024;137(11):1136-1141. doi: 10.1016/j.amjmed.2024.06.030 [DOI] [PubMed] [Google Scholar]
- 51.Jacobsen SL, Köhler-Forsberg O, Danielsen AA, Rohde C. Efficacy and safety of GLP1-RAs compared to SGLT2is and DPP-4is in individuals with schizophrenia and diabetes: a Danish nationwide target-trial emulation study. Psychiatry Res. 2025;351:116592. doi: 10.1016/j.psychres.2025.116592 [DOI] [PubMed] [Google Scholar]
- 52.Chang EW, Huang JY, Lo SC, Huang CN, Yang YS, Kornelius E. Association between SGLT-2 inhibitors and suicide risk in type 2 diabetes and bipolar: a real-world cohort study. Front Pharmacol. 2025;16:1601118. doi: 10.3389/fphar.2025.1601118 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 53.Bhatia S, Rawal R, Sharma P, Singh T, Singh M, Singh V. Mitochondrial dysfunction in Alzheimer’s disease: opportunities for drug development. Curr Neuropharmacol. 2022;20(4):675-692. doi: 10.2174/1570159X19666210517114016 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 54.Liebers DT, Ebina W, Iosifescu DV. Sodium-glucose cotransporter-2 inhibitors in depression. Harv Rev Psychiatry. 2023;31(4):214-221. doi: 10.1097/HRP.0000000000000374 [DOI] [PubMed] [Google Scholar]
- 55.Hasan-Olive MM, Lauritzen KH, Ali M, Rasmussen LJ, Storm-Mathisen J, Bergersen LH. A ketogenic diet improves mitochondrial biogenesis and bioenergetics via the PGC1α-SIRT3-UCP2 axis. Neurochem Res. 2019;44(1):22-37. doi: 10.1007/s11064-018-2588-6 [DOI] [PubMed] [Google Scholar]
- 56.Mujica-Parodi LR, Amgalan A, Sultan SF, et al. Diet modulates brain network stability, a biomarker for brain aging, in young adults. Proc Natl Acad Sci U S A. 2020;117(11):6170-6177. doi: 10.1073/pnas.1913042117 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 57.Danan A, Westman EC, Saslow LR, Ede G. The ketogenic diet for refractory mental illness: a retrospective analysis of 31 inpatients. Front Psychiatry. 2022;13:951376. doi: 10.3389/fpsyt.2022.951376 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 58.Zandifar A, Panahi M, Badrfam R, Qorbani M. Efficacy of empagliflozin as adjunctive therapy to citalopram in major depressive disorder: a randomized double-blind, placebo-controlled clinical trial. BMC Psychiatry. 2024;24(1):163. doi: 10.1186/s12888-024-05627-0 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 59.Dave CV, Schneeweiss S, Patorno E. Comparative risk of genital infections associated with sodium-glucose co-transporter-2 inhibitors. Diabetes Obes Metab. 2019;21(2):434-438. doi: 10.1111/dom.13531 [DOI] [PMC free article] [PubMed] [Google Scholar]
Associated Data
This section collects any data citations, data availability statements, or supplementary materials included in this article.
Supplementary Materials
eMethods. Trial Emulation Methodology Details
eTable 1. Comorbidity Definitions and ICD Codes Used
eTable 2. Complete List of Medications Considered
eTable 3. Demographic and Clinical Characteristics of Cohort for All Outcomes at Person-Period Level for the Main Analysis Cohort (BIP, SCZ, MDD). Outcome-Specific Covariates Are Shown in the Bottom Rows.
eTable 4. Demographic and Clinical Characteristics of Cohort for Outcomes I, II at Person-Period Level for the Sensitivity Analysis Cohort (BIP, SCZ, PMDD)
eTable 5. Sensitivity Analyses for Subgroups Stratified by Diabetes Status at Baseline for Emulated Trial for Dementia Outcome
eTable 6. Sensitivity Analyses for Subgroups Stratified by PTSD Status at Baseline for Emulated Trial for Dementia Outcome
eTable 7. Effects of SGLT2-i Treatment on Composite Outcome of Dementia or Death, in the Primary Dementia Cohort
eFigure 1. Schematic Diagram of Trial Emulation Design
eFigure 2. Cohort Derivation Diagram
eFigure 3. Truncated Distributions of Combined Inverse Censoring and Treatment Weights of Marginal Structural Models (Main Analysis). Weights Are Shown for ITT and PP Effects for All Outcomes. Densities Are Shown in Log Scale.
eFigure 4. Complete List of Psychiatric Criteria Keywords in Emergency Department Admission Descriptions
eFigure 5. Love Plot of Standardized Mean Differences (SMDs) for All Covariates in ITT and PP Analyses on Dementia Outcome
eFigure 6. Estimated Cumulative Incidence of Dementia, Cumulative Incidence of Death, and Event-Free Survival Over Follow-Up. Estimates Were Derived From Weighted Pooled Logistic Regression Models Accounting for Competing Risks; Event-Free Survival Denotes Remaining Alive and Free of Dementia.
eReferences
Data Sharing Statement
