Sodium‐glucose cotransporter 2 inhibitors and the risk of late onset epilepsy: A real‐world cohort study

Abstract

Objective

Late onset epilepsy (LOE) is associated with substantial morbidity. Sodium‐glucose cotransporter 2 inhibitors (SGLT2i) may exert neuroprotective effects. This study evaluated the association between SGLT2i and risk of LOE among older adults with type 2 diabetes mellitus.

Methods

This retrospective cohort study was conducted between 2013 and 2025 with a 3‐year follow‐up using the TriNetX global network. Patients ≥60 years old with type 2 diabetes mellitus were classified into two cohorts—new SGLT2i users and new dipeptidyl peptidase‐4 inhibitors users—following a 6‐month washout of other antihyperglycemic agents except metformin. Patients with major neurological diseases or contraindications to SGLT2i were excluded. Propensity score matching was used to balance baseline characteristics between two cohorts. The primary outcomes were incident LOE, status epilepticus, and initiation of antiseizure medications. Unadjusted Cox proportional hazards models were applied to estimate hazard ratios (HRs) and 95% confidence interval (CIs).

Results

A total of 1 435 648 patients were identified. After matching, 60 203 patients were included in the SGLT2i cohort (mean age = 67.7 years, 42.3% female) and 60 203 in the control cohort (mean age = 67.7 years, 41.7% female). SGLT2i use was associated with lower risk of LOE (HR = .55, 95% CI = .44–.68), status epilepticus (HR = .38, 95% CI = .21–.69), and antiseizure medication initiation (HR = .63, 95% CI = .58–.69). SGLT2i reduced LOE risk in patients with stroke (HR = .69, 95% CI = .42–.88) and dementia (HR = .44, 95% CI = .25–.78) but not in those with traumatic brain injury or brain tumors.

Significance

SGLT2i use was associated with reduced risk of LOE among selected patients, supporting its role in an etiology‐specific therapeutic approach for older adults at risk of epilepsy.

Key points.

• SGLT2 inhibitors may exert neuroprotective effects, but their association with LOE in older adults remains unclear.

• In a real‐world cohort of 1.4 million patients, SGLT2 inhibitors were linked to lower risks of LOE, status epilepticus, and antiseizure drug use.

• SGLT2 inhibitors reduced LOE risk in patients with stroke and dementia, but not in those with traumatic brain injury or brain tumor.

• Further basic and prospective research is needed to confirm the effects of SGLT2 inhibitors on LOE and their underlying mechanisms.

1. INTRODUCTION

Epilepsy is one of the most common chronic neurological disorders, characterized by recurrent unprovoked seizures due to excessive neuronal activity. ¹ The global lifetime prevalence of epilepsy is estimated at 7.6 per 1000 persons, and the risk increases substantially with age, reaching a cumulative incidence of 4.4% by age 85 years. ¹ , ² Late onset epilepsy (LOE), defined as the onset of recurrent seizures at or after age 60 years, is increasingly recognized as a distinct clinical entity with growing research interest. ³ , ⁴ , ⁵ Cerebrovascular disease and neurodegenerative disorders account for up to 50% of cases, followed by traumatic brain injury (TBI) and brain tumors; however, the etiology remains unknown in 25%–50% of cases. ² , ⁶ , ⁷ LOE is associated with reduced quality of life and multiple adverse outcomes, including cognitive impairment, functional decline, psychiatric morbidity, and increased mortality, particularly among older adults with high burden of systemic and neurological comorbidities. ⁸ , ⁹ Although most patients achieve seizure control with antiseizure medications, approximately 13.7%–36.3% develop drug‐resistant epilepsy, ¹⁰ underscoring the need for novel therapeutic approaches.

Originally developed for type 2 diabetes mellitus (T2DM), sodium‐glucose cotransporter 2 inhibitors (SGLT2i) are antihyperglycemic agents that inhibit glucose and sodium reabsorption in the proximal renal tubules. ¹¹ Their use has expanded rapidly owing to cardiovascular and renal benefits, ¹² with annual prescription volumes in the United States nearly doubling over the past decade. ¹³ Emerging evidence suggests that SGLT2i may also confer neuroprotective properties through attenuation of oxidative stress, augmentation of ketogenesis, and modulation of microglia‐mediated neuroinflammation—mechanisms that are linked to epileptogenesis. ¹⁴ , ¹⁵ , ¹⁶ A recent meta‐analysis of novel antihyperglycemic agents—including glucagonlike peptide‐1 receptor agonists (GLP‐1 RA), SGLT2i, and dipeptidyl peptidase‐4 inhibitors (DPP4i)—reported an overall reduced risk of LOE. However, the protective effect was driven primarily by GLP‐1 RA rather than by SGLT2i or DPP4i. ¹⁷ Evidence specific to SGLT2i remain sparse. Moreover, LOE was not a primary endpoint in these trials, ¹⁸ , ¹⁹ , ²⁰ , ²¹ and seizure events were identified through adverse events reporting, rendering these studies likely underpowered to detect rare neurologic outcomes with sufficient precision. ¹⁷ , ¹⁸ , ¹⁹ , ²⁰ , ²¹ Despite increasing clinical adoption of SGLT2i and potential neuroprotective properties, their association with LOE warrants further investigation. Given the growing burden of epilepsy in aging populations, evaluating the effects of SGLT2i may offer novel insights into risk modification and inform therapeutic strategies.

To date, no large‐scale population‐based study has examined the association between SGLT2i and LOE. Clinical evidence regarding how complex comorbidities in older adults affect the efficacy of SGLT2i is even more limited. To address these gaps, we used the TriNetX research network to evaluate the risk of LOE among patients treated with SGLT2i, with DPP4i used as an active comparator given their lack of established neuroprotective or epileptogenic effects in humans.

2. MATERIALS AND METHODS

2.1. Research design and data source

This retrospective cohort study was conducted using the TriNetX global collaborative network, which contains deidentified electronic medical records of more than 170 million patients across 136 health care organizations. ²² Data queries were conducted on August 6, 2025, and all statistical analyses were based on the dataset available on that date. The study adhered to the Strengthening the Reporting of Observational Studies in Epidemiology reporting guidelines. ²³ Data deidentification within the TriNetX network complies with the Health Insurance Portability and Accountability Act privacy rule. Ethical approval was waived by the institutional review board of Taipei Medical University (TMU‐JIRB No. N202509052). Due to the retrospective design and use of anonymized data, in accordance with the Declaration of Helsinki, the requirement for informed consent was waived.

2.2. Study population

The study period extended from March 1, 2013 to February 1, 2025, with the start date corresponding to the US Food and Drug Administration's approval of canagliflozin, the first SGLT2i. We included adults aged ≥60 years with a diagnosis of T2DM and at least one health care visit during the study period. Patients with documented diagnoses of brain tumors, TBI, hemorrhagic or ischemic stroke, neurodegenerative disorders, encephalitis, central nervous system demyelinating diseases, brain vascular malformations, or contraindications to SGLT2i were excluded if these conditions were identified at any time during the study period, either before the index date or during follow‐up. Eligible patients were divided into two cohorts according to the prescribed second‐line antihyperglycemic agents. To mitigate confounding by indication and enhance clinical relevance, an active‐comparator new‐user design was implemented; the SGLT2i cohort comprised new users of SGLT2i, whereas the control cohort comprised new users of DPP4i. New use was defined as the first prescription of either drug class following a 6‐month washout period without prescriptions of antihyperglycemic agents other than metformin. The index date was defined as the initiation of study drugs following fulfillment of eligibility criteria. Patients exposed to antihyperglycemic agents outside their assigned cohorts were excluded. Additionally, patients diagnosed with epilepsy or convulsions or prescribed antiseizure medications during the preindex period were also excluded. Figure 1 presents the flowchart of the study population selection. Detailed codes used to identify the study population are provided in Table S1.

FIGURE 1.

Study flowchart. DPP4i, dipeptidyl peptidase‐4 inhibitors; HbA1c, hemoglobin A1c; SGLT2i, sodium‐glucose cotransporter 2 inhibitors.

2.3. Outcomes

Information on patients' baseline demographics, comorbidities, laboratory data, and medication use was collected within 1 year preceding the index date. The primary outcome was the incidence of LOE, defined as the first occurrence of an International Classification of Diseases, 10th Revision, Clinical Modification (ICD‐10‐CM) diagnosis code for epilepsy (G40.0–G40.9) during follow‐up. This coding‐based algorithm has been validated, demonstrating a sensitivity of 84.4% and a specificity of 79.4%. ²⁴ We also assessed the risk of status epilepticus using G40 codes with the modifier “with status epilepticus” (G40.xx1), an approach validated with a sensitivity of 68.7% and a specificity of 92.6%. ²⁵ This strategy was necessary because, following the United States' adoption of ICD‐10‐CM, standalone status epilepticus code (G41) is not routinely used and is therefore unavailable within the TriNetX analytic platform. In addition, the risks of subsequent initiation of antiseizure medications were also analyzed. To evaluate potential residual confounding, we selected negative control outcomes (NCOs), including conjunctivitis, low back pain, and lymphoma, ²⁶ , ²⁷ unlikely to be causally affected by the exposure. Patients were followed for 3 years from the index date. Detailed codes used to define outcomes are provided in Table S1.

2.4. Statistical analysis

To address baseline differences between cohorts, one‐to‐one propensity score matching (PSM) was performed using greedy nearest neighbor matching with a caliper of .1. Patients were matched for age, sex, race, comorbidities, T2DM complications, medication use (including antipsychotics, antidepressants, and benzodiazepine‐related drugs), glycated hemoglobin A1c (HbA1c), health care utilization, and socioeconomic status. Socioeconomic status was ascertained using ICD‐10‐CM codes Z55–Z65, which have been previously validated and applied in real‐world studies. ²⁸ , ²⁹ , ³⁰ , ³¹ Covariate balance was assessed using standardized mean differences (SMDs), with values ≤ .1 considered well balanced. Unadjusted cox proportional hazards models were used to estimate hazard ratios (HRs) and 95% confidence intervals (CIs) for epilepsy‐related outcomes. Cumulative incidence was assessed using time‐to‐event analysis, and group differences were compared using the log‐rank test. A two‐sided p < .05 was considered statistically significant. Moreover, E‐values were calculated to assess the potential impact of unmeasured confounding. An E‐value > 2 indicates that unmeasured confounders would require a strong association with both the exposure and outcomes to negate the observed effect, supporting the robustness against residual confounding. ³²

To evaluate whether baseline characteristics modified the effect of SGLT2i, we conducted subgroup analyses stratified by age (60–80 vs. ≥80 years), sex (male vs. female), systemic diseases (overweight, chronic kidney disease, hypertension, and hyperlipidemia), and HbA1c (<7% vs. ≥7%).

We also conducted analyses in separate high‐risk groups by independently identifying additional patient populations with a history of stroke, dementia, TBI, or brain tumor prior to the index date—conditions associated with increased susceptibility to LOE. ⁶ Within each high‐risk groups, patients were stratified by SGLT2i use to assess its association with epilepsy risk. The patient selection process is illustrated in Figure S1. This approach ensured that the observed associations reflected the effect of SGLT2i within patients with preexisting neurological conditions, while minimizing potential mediation by these comorbidities.

A prespecified exploratory secondary analysis was conducted among patients with preexisting epilepsy or convulsions to investigate whether SGLT2i use was associated with differences in subsequent epilepsy‐related clinical outcomes. After PSM using the same covariates as in the primary analysis, the risks of epilepsy‐related neurological and functional outcomes were compared between SGLT2i users and DPP4i users. Outcomes were ascertained using incident ICD‐10‐CM codes. Dementia was identified using G30, F01, G31.83, G31.0, and F03, and psychiatric disorders using F30–F39 and F41. Falls were identified using W00–W19, and decline in activities of daily living (ADL) was identified using Z74 as a proxy for functional decline. Mortality included all‐cause deaths during follow‐up.

Sensitivity analyses were performed to assess the robustness of the findings, including alternative follow‐up periods (1 year, 2 years, and follow‐up until the earliest occurrence of the studied outcomes or the last available medical record), prescription counts (≥5 and ≥10 prescriptions), treatment durations (≥1 and ≥2 years), and alternative comparators (patients receiving other commonly used second‐line agents, including GLP‐1 RA, DPP4i, sulfonylureas, thiazolidinediones, or α‐glucosidase inhibitors). To account for potential time‐varying effects, time‐split analyses were performed by dividing follow‐up into four intervals (0–6 months, 6 months–1 year, 1–2 years, and 2–3 years). We also assessed the incidence of electroencephalography utilization, defined by the presence of relevant procedure codes (CPT 1013322) during follow‐up. To address potential nonepilepsy indications for antiseizure medications, a sensitivity analysis was conducted restricting the outcome to highly specific antiseizure medications rarely prescribed for nonepilepsy diseases (e.g., vigabatrin, lamotrigine, levetiracetam, zonisamide, lacosamide, and perampanel). ³³ , ³⁴ Comorbidities related to alternative indications, including neuropathic pain and mood disorders, were incorporated into the PSM model to further balance comorbidity burden. Statistical analyses were conducted using the TriNetX built‐in analytics platform, and the forest plots and cumulative incidence curves were generated in R software (version 4.4.3).

3. RESULTS

3.1. Baseline characteristics of the study population

The study included 83 670 patients aged ≥60 years with T2DM who received SGLT2i (mean age = 68.0 ± 7.9 years, 36.4% female), and 80 239 patients who received DPP4i (mean age = 68.1 ± 8.9 years, 49.4% female; Figure 1, Table 1). At baseline, the SGLT2i cohort had a higher prevalence of cardiovascular disease, cerebrovascular disease, chronic kidney disease, overweight, obstructive sleep apnea, and T2DM complications. SGLT2i users also had higher prescription rates of antipsychotics or antidepressants, as well as higher rates of health care utilization. After PSM, a total of 60 203 patients were included in both the SGLT2i cohort (mean age = 67.7 ± 7.8 years, 42.3% female) and the control cohort (mean age = 67.7 ± 8.8 years, 41.7% female). All covariates, including demographics, comorbidities, and medication use, were balanced between the matched cohorts, as indicated by SMDs ≤ .1.

TABLE 1.

Baseline characteristics before and after propensity score matching.

Characteristic	Before matching a			After matching a
	SGLT2i cohort, n = 83 670	Control cohort, n = 80 239	SMD	SGLT2i cohort, n = 60 203	Control cohort, n = 60 203	SMD
Demographics
Age, years	68.0 ± 7.9	68.1 ± 8.9	.011	67.7 ± 7.8	67.7 ± 8.8	.005
Male	50 851 (60.8)	39 504 (49.2)	.234	33 605 (55.8)	33 952 (56.4)	.012
Female	30 495 (36.4)	39 602 (49.4)	.263	25 459 (42.3)	25 133 (41.7)	.011
Race and ethnicity
White	45 715 (54.6)	33 313 (41.5)	.265	29 895 (49.7)	30 053 (49.9)	.005
Black or African American	10 713 (12.8)	8867 (11.1)	.054	7513 (12.5)	7599 (12.6)	.004
Asian	6430 (7.7)	7408 (9.2)	.056	5205 (8.6)	5320 (8.8)	.007
Comorbidities
Hypertensive diseases	41 498 (49.6)	26 850 (33.5)	.332	23 008 (38.2)	22 937 (38.1)	.002
Ischemic heart diseases	18 787 (22.5)	8046 (10.0)	.342	7668 (12.7)	7565 (12.6)	.005
Heart failure	10 968 (13.1)	2381 (3.0)	.380	2486 (4.1)	2372 (3.9)	.010
Cerebrovascular diseases	2142 (2.6)	1444 (1.8)	.052	1116 (1.9)	1122 (1.9)	.001
Overweight and obesity	11 588 (13.8)	5251 (6.5)	.243	5022 (8.3)	5015 (8.3)	<.001
Hyperlipidemia	36 786 (44.0)	22 147 (27.6)	.346	19 696 (32.7)	19 557 (32.5)	.005
Alcohol‐related disorders	958 (1.1)	443 (.6)	.065	407 (.7)	412 (.7)	.001
Hepatic failure	128 (.2)	109 (.1)	.005	76 (.1)	73 (.1)	.001
Chronic kidney disease	6986 (8.3)	2813 (3.5)	.206	2714 (4.5)	2634 (4.4)	.006
Obstructive sleep apnea	6697 (8.0)	2219 (2.8)	.234	2269 (3.8)	2196 (3.6)	.006
Complications of diabetes
Ketoacidosis	222 (.3)	174 (.2)	.010	135 (.2)	142 (.2)	.002
Hyperosmolarity	543 (.6)	312 (.4)	.036	247 (.4)	252 (.4)	.001
Neurological complications	5197 (6.2)	2770 (3.5)	.129	2487 (4.1)	2458 (4.1)	.002
Circulatory complications	3655 (4.4)	1368 (1.7)	.156	1333 (2.2)	1305 (2.2)	.003
Medications
Antipsychotics	3465 (4.1)	2822 (3.5)	.033	2038 (3.4)	2037 (3.4)	<.001
Antidepressants	9081 (10.9)	5652 (7.0)	.134	5018 (8.3)	4996 (8.3)	.001
Benzodiazepine‐related drugs	1412 (1.7)	1458 (1.8)	.010	995 (1.7)	1043 (1.7)	.006
Others
Electroencephalogram	48 (.1)	49 (.1)	.002	34 (.1)	36 (.1)	.001
Nicotine dependence	4619 (5.5)	2237 (2.8)	.137	2139 (3.6)	2103 (3.5)	.003
Tobacco use	1320 (1.6)	632 (.8)	.073	595 (1.0)	581 (1.0)	.002
HbA1c 0%–5.7%	1321 (1.6)	841 (1.0)	.047	696 (1.2)	683 (1.1)	.002
HbA1c 5.7%–6.5%	9848 (11.8)	6625 (8.3)	.117	5374 (8.9)	5424 (9.0)	.003
HbA1c 6.5%–7.9%	22 195 (26.5)	15 955 (19.9)	.158	13 035 (21.7)	13 004 (21.6)	.001
HbA1c ≥ 8%	13 429 (16.0)	10 089 (12.6)	.099	8310 (13.8)	8269 (13.7)	.002
Inpatient services	6408 (7.7)	3012 (3.8)	.169	2574 (4.3)	2627 (4.4)	.004
Outpatient services	31 534 (37.7)	16 148 (20.1)	.395	16 097 (26.7)	15 828 (26.3)	.010
Patients at risk due to socioeconomic and psychosocial factors	913 (1.1)	416 (.5)	.064	382 (.6)	375 (.6)	.001

Note: Data represent n (%) unless otherwise indicated.

Abbreviations: HbA1c, glycated hemoglobin A1c; SGLT2i, sodium‐glucose cotransporter 2 inhibitors; SMD, standardized mean difference.

^a Propensity score matching for demographics, comorbidities, medications, health care use, and socioeconomic status.

3.2. Risk of late onset epilepsy and associated outcomes

After a 3‐year follow‐up, posttreatment mean HbA1c levels were comparable between the SGLT2i cohort (7.13 ± 1.29%) and control cohort (7.01 ± 1.30%; SMD = .093, absolute difference = .12 ± .01%), with no clinically meaningful difference. SGLT2i use was associated with a significantly lower incidence of LOE compared with the control cohort (122 [.20%] vs. 251 [.42%], HR = .55, 95% CI = .44–.68, E‐value = 3.03; Figure 2). Patients in the SGLT2i cohort were less likely to develop status epilepticus (14 [.02%] vs. 41 [.07%], HR = .38, 95% CI = .21–.69, E‐value = 4.76) and to require antiseizure medications (803 [1.33%] vs. 1400 [2.33%], HR = .63, 95% CI = .58–.69, E‐value = 2.56; Figure 2). The cumulative incidence curves (Figure S2) revealed early divergence between the SGLT2i and control cohorts, with significant differences observed throughout follow‐up. The risks of NCOs were comparable between cohorts (Figure S3).

FIGURE 2.

Risk of late‐onset epilepsy and associated clinical outcomes in the matched cohorts over 3 years of follow‐up. CI, confidence interval; DPP4i, dipeptidyl peptidase‐4 inhibitors; SGLT2i, sodium‐glucose cotransporter 2 inhibitors.

3.3. Subgroup analyses

In subgroup analyses (Figure 3), the SGLT2i cohort, compared with the control cohort, demonstrated lower risks of LOE and antiseizure medication initiation across various patient demographics and comorbidities. For status epilepticus, although a protective effect of SGLT2i was observed in most subpopulations, the association reached statistical significance only among patients older than 80 years, women, and those with hypertension.

FIGURE 3.

Risk of late onset epilepsy and associated clinical outcomes among patient subgroups. *For patient counts between 1 and 10, the results were reported as ≤10 in accordance with TriNetX data protection policies; however, hazard ratios were calculated using the actual event counts. CI, confidence interval; DPP4i, dipeptidyl peptidase‐4 inhibitors; HbA1c, hemoglobin A1c; SGLT2i, sodium‐glucose cotransporter 2 inhibitors.

3.4. Risk of LOE and associated outcomes among high‐risk populations

Figure 4 presents the risk of epilepsy‐related outcomes among high‐risk groups. SGLT2i use was associated with a significantly lower risk of LOE in patients with stroke (53 [1.20%] vs. 81 [1.84%], HR = .69, 95% CI = .42–.88) and dementia (16 [.84%] vs. 43 [2.26%], HR = .44, 95% CI = .25–.78), but not in patients with TBI (14 [1.44%] vs. 16 [1.65%], HR = .97, 95% CI = .47–1.98) or brain tumor (14 [3.56%] vs. 11 [2.80%], HR = 1.53, 95% CI = .69–3.40). A similar association was observed for the risk of requiring antiseizure medications.

FIGURE 4.

Risk of late onset epilepsy and associated clinical outcomes among high‐risk populations. *For patient counts between 1 and 10, the results were reported as ≤10 in accordance with TriNetX data protection policies; however, hazard ratios were calculated using the actual event counts. CI, confidence interval; DPP4i, dipeptidyl peptidase‐4 inhibitors; n/a, not applicable; SGLT2i, sodium‐glucose cotransporter 2 inhibitors.

3.5. Secondary analyses

Figure 5 presents the risks of neurological and functional outcomes among patients with preexisting epilepsy. The risks of developing dementia and psychiatric disorders were numerically lower among SGLT2i users; however, the reduction in psychiatric disorders did not reach statistical significance, possibly reflecting limited statistical power to detect modest associations given the small sample size. Regarding the functional outcomes (including decline in ADL and fall events) and all‐cause mortality, no significant differences were observed.

FIGURE 5.

Secondary analysis results in patients with baseline epilepsy or convulsions. *For patient counts between 1 and 10, the results were reported as ≤10 in accordance with TriNetX data protection policies; however, hazard ratios were calculated using the actual event counts. CI, confidence interval; DPP4i, dipeptidyl peptidase‐4 inhibitors; SGLT2i, sodium‐glucose cotransporter 2 inhibitors.

3.6. Sensitivity analyses

We performed multiple sensitivity analyses to test the stability of our findings. Consistent associations between SGLT2i and lower risks of LOE, status epilepticus, and antiseizure medication initiation were observed across varying follow‐up periods, prescription counts, treatment durations, and alternative comparator (Figure S4). In time‐split analyses, SGLT2i use consistently conferred protection against LOE across all time intervals (Figure S5). This association extended to a significantly lower risk of initiating highly specific antiseizure medications (344 [.58%] vs. 455 [.76%], HR = .81, 95% CI = .71–.93; Figure S6). Electroencephalography utilization was also reduced in the SGLT2i cohort compared with the control cohort (132 [.21%] vs. 184 [.30%], HR = .78, 95% CI = .63–.98; Figure S6). The concordant pattern across both clinical diagnostic evaluations and LOE events, together with the consistent results of multiple sensitivity analyses, underscores the robustness of the association between SGLT2i use and a lower LOE burden.

4. DISCUSSION

In this real‐world study, SGLT2i use was associated with approximately 45% lower risk of LOE, 62% lower risk of status epilepticus, and 37% lower risk of antiseizure medication initiation. The results were generally consistent across sensitivity analyses and subpopulations stratified by demographics and comorbidities; however, no protective effect against LOE was observed in patients with TBI or brain tumors. In secondary analyses of patients with preexisting epilepsy, SGLT2i use was not associated with decline in ADL or mortality but was associated with lower risks of dementia and mood disorders.

Our finding aligned with preclinical evidence showing that SGLT2i attenuated seizures and epileptogenesis in animal models. ¹⁵ , ¹⁶ , ³⁵ Abdelaziz et al. ³⁵ demonstrated that empagliflozin mitigated pentylenetetrazole (PTZ)‐induced seizures in rats model through modulation of epileptogenic pathways, including downregulation of neuronal PAS domain protein 4 (Npas4) and cAMP response element‐binidng protein (CREB), and activation of brain‐derived neutrophic factor‐tropomyosin receptor kinase B (BDNF–TrkB) signaling. Erdogan et al. ¹⁵ reported that dapagliflozin attenuated seizure activity in a PTZ‐induced murine epilepsy model by reducing glucose availability and neuronal sodium influx, thereby stabilizing membrane excitability. Liu et al. ¹⁶ demonstrated that dapagliflozin exerted neuroprotection in a pilocarpine‐induced status epilepticus model by modulating microglia‐mediated neuroinflammation and oxidative stress. Moreover, SGLT2i promote mild, sustained ketogenesis, which enhances mitochondrial bioenergetics and modulates both inhibitory (e.g., γ‐aminobutyric acidergic, adenosine triphosphate‐sensitive potassium channels) and excitatory (e.g., glutamate transporters) neurotransmission, thereby contributing to seizure control. ³⁶ , ³⁷ Although some mechanisms—such as attenuation of oxidative stress and anti‐inflammation—are shared by SGLT2i and DPP4i, ¹⁷ SGLT2i possess distinct mechanistic pathways, including ketogenic properties, direct modulation of membrane excitability, and more specific anti‐inflammatory actions, which could confer additional antiepileptic effect. This benefit may act through neuroprotective mechanisms rather than glycemic control alone, as posttreatment HbA1c levels were similar between matched cohorts.

Insights from preclinical studies have laid the groundwork for exploring the clinical potential of SGLT2i in epilepsy management. In a retrospective study conducted within a single municipal district in China, SGLT2i use was associated with a 29% lower risk of epilepsy compared with DPP4i. ³⁴ However, the study's restricted geographic and ethnic scope may limit its generalizability. ³⁴ Furthermore, their outcomes were not age‐specific and included younger adults, among whom the pathophysiology and risk factors may differ from the elderly, thereby limiting direct relevance to LOE. ³⁴ More recently, Cheng et al. ²⁸ examined a relevant issue using the TriNetX network, reporting a 19% lower risk of epilepsy among adults with T2DM treated with GLP‐1 RA compared with DPP4i. The stronger protective association observed with SGLT2i in our study may reflect several methodological differences. Unlike Cheng et al., ²⁸ we excluded individuals with preexisting or incident major neurological comorbidities rather than adjusting for them through PSM, resulting in a cohort with a lower baseline burden of epilepsy‐related conditions and a profile closer to that of the general population. In addition, our cohort was restricted to patients receiving only the assigned study drug (SGLT2i or DPP4i), minimizing treatment effect dilution from concomitant therapies. Beyond design considerations, SGLT2i induce mild ketosis and reduce neuronal sodium influx, which may confer additional antiepileptogenic effects. ¹⁵ , ³⁶ , ³⁷ These methodological and biological factors may underlie the stronger protective effect against LOE in our study.

Our findings suggested that the protective effects of SGLT2i varied by underlying etiologies. The observed risk reduction in patients with prior stroke or dementia may be explained by multiple neuroprotective mechanisms, including improved mitochondrial function, attenuation of neuroinflammation and microglial activation, and prevention of myelin remodeling. ¹⁴ Enhanced expression of brain‐derived neurotrophic factor and increased synaptogenesis further support neuronal recovery. ³⁸ Additionally, by reducing postischemic hyperglycemia, SGLT2i may ameliorate ischemic neuronal injury and subsequent epileptogenesis. ³⁹ In contrast, epilepsy following TBI or intracranial neoplasms is mainly driven by structural damage and gliosis, ⁴⁰ , ⁴¹ forming entrenched epileptogenic networks unlikely to be reversed by systemic metabolic modulation. ⁴² These findings underscored the importance of etiology‐specific prevention, and suggested that the neuroprotective effects of SGLT2i may be more relevant to chronic neurovascular dysregulation than structural pathology. However, given the limited sample size and unclear mechanisms underlying these differences, our results should be interpreted with caution.

In secondary analyses, our findings demonstrated a lower incidence of dementia and mood disorders among SGLT2i users, consistent with prior studies. ¹⁴ , ⁴³ This effect may be mediated by enhanced hippocampal plasticity, reduced apoptosis and inflammation, and stimulation of neurotrophic factors, which may enhance cognitive and emotional functions. ⁴⁴ , ⁴⁵ Despite neurologic benefits, SGLT2i did not improve functional outcomes or all‐cause mortality in our study. This disconnect underscored the complexity of translating neurocognitive improvements into functional gains and highlighted the need for complementary strategies to preserve function in older adults with epilepsy.

Our study provides the first large‐scale, real‐world evidence that SGLT2i reduced the risk of LOE across diverse demographics and comorbidities. The new‐user design ensured comparable baseline characteristics between cohorts, with no recent exposure to other second‐line antihyperglycemic agents. This design allows any differences in epilepsy risk to become apparent shortly after SGLT2i initiation and likely explains the early divergence of cumulative incidence curves. Time‐split analyses further demonstrated that the protective association emerged early after treatment initiation and was sustained throughout follow‐up. Additionally, lower risks of status epilepticus and initiation of antiseizure medications suggested a potential role of SGLT2i in attenuating epilepsy severity.

Given that SGLT2i may also reduce the risk of stroke and dementia, ¹⁴ , ⁴⁶ both established risk factors for epilepsy, we excluded patients with baseline or incident stroke or dementia in the primary analysis. Furthermore, in analyses of high‐risk groups, consistent associations in patients with prior stroke or dementia suggested that the protective effect of SGLT2i against LOE is independent of differences in stroke or dementia occurrence. The null association of NCOs and primary outcome E‐values > 2 indicated that the reduced LOE risk with SGLT2i was unlikely to be explained solely by unmeasured confounding. Given the high prevalence of comorbidities, polypharmacy, and vulnerability to epilepsy‐related adverse outcomes in older adults, ⁴⁷ our findings support tailored strategies for selecting glucose‐lowering therapies based on individual risk profiles. For high‐risk patients with T2DM and comorbid stroke or dementia, SGLT2i may be preferentially considered as a second‐line antihyperglycemic agent to mitigate epilepsy risk, whereas no protective effect was observed in patients with epilepsy risk related to TBI or brain tumors.

This study has some limitations. First, although NCOs and E‐values suggested minimal residual confounding, we cannot exclude unmeasured factors such as genetic variation, family history, and lifestyle. Second, TriNetX lacks data on medication dose, frequency, and adherence. Although we approximated exposure using the prescription counts and treatment duration, these metrics may not precisely reflect the true cumulative exposure.

Third, accurate diagnosis of epilepsy requires detailed patient histories, neurological examinations, and electroencephalography. In this study, due to unavailable clinical data, we adopted a by‐proxy definition of LOE based on ICD‐10‐CM codes (G40.x), an approach commonly used in large administrative database studies. ²⁸ , ³⁴ However, using diagnosis codes to ascertain seizure recurrence among patients with preexisting epilepsy may be particularly susceptible to outcome misclassification. Therefore, seizure recurrence was not evaluated in the secondary analysis. In addition, although more stringent outcome definitions—such as requiring two or more diagnosis claims or combining diagnostic codes with antiseizure medication use—may improve specificity, these approaches were not feasible due to platform constraints. To address these limitations, we restricted the cohort to older adults (≥60 years), applied specific diagnostic codes, ²⁴ and excluded individuals with prior epilepsy or antiseizure medication use to better capture incident epilepsy. Notably, the LOE incidence (.2%–.4%) in our cohort was comparable to that reported in previous population‐based studies, ⁴ , ⁴⁸ supporting the accuracy of our case ascertainment.

Fourth, status epilepticus was ascertained using ICD‐10‐CM G40.xx1, rather than the specific code (G41), which may underestimate standalone status epilepticus events. Such underascertainment is likely to be nondifferential between matched cohorts and would be expected to bias the estimates toward the null. Nevertheless, a significantly lower risk of status epilepticus was still observed among SGLT2i users in the primary analysis, supporting the robustness of this association.

Fifth, we observed more antiseizure medication initiations than incident epilepsy diagnoses, likely reflecting nonepilepsy indications and preventive prescribing before formal diagnosis. ⁴⁹ However, consistent results from sensitivity analysis restricted to highly specific antiseizure medications suggested that the observed association was unlikely to be driven by nonepilepsy conditions. Finally, mechanistic interpretations of our findings remained limited, as the database lacks direct measures of brain pathology, electrophysiology, neuroimaging, or neuronal injury biomarkers.

In conclusion, our study provided real‐world evidence that SGLT2i use is associated with lower risk and severity of LOE in older adults. However, the protective effect was not observed in patients with TBI or brain tumors. These findings support an etiology‐specific approach when prescribing SGLT2i. Further basic research and large‐scale prospective studies are needed to clarify the effects of SGLT2i on LOE and their underlying mechanisms. Overall, SGLT2i may represent a promising disease‐modifying agent for adults aged ≥60 years to prevent LOE.

AUTHOR CONTRIBUTIONS

Bing‐Hua Lin: Conceptualization (lead); methodology (supporting); formal analysis (lead); writing—original draft (lead); writing—review and editing (equal). Hui‐Min Huang: Conceptualization (lead); methodology (supporting); formal analysis (lead); writing—original draft (lead); writing—review and editing (equal). Hui‐An Lin: Conceptualization (supporting); methodology (supporting); writing—review and editing (equal). Sheng‐Feng Lin: Conceptualization (lead); methodology (lead); formal analysis (supporting); resources (lead); supervision (lead); project administration (lead); writing—review and editing (equal). All authors reviewed and approved the final manuscript.

CONFLICT OF INTEREST STATEMENT

None of the authors has any conflict of interest to disclose. We confirm that we have read the Journal's position on issues involved in ethical publication and affirm that this report is consistent with those guidelines.

Supporting information

Appendix S1.

Lin B‐H, Huang H‐M, Lin H‐A, Lin S‐F. Sodium‐glucose cotransporter 2 inhibitors and the risk of late onset epilepsy: A real‐world cohort study. Epilepsia. 2026;67:3444–3456. 10.1002/epi.70239

DATA AVAILABILITY STATEMENT

The data used in this study were obtained from the TriNetX research network and are not publicly available. However, aggregate data or analytic codes are available from the corresponding author upon reasonable request.