Comparative risk of the neurodegenerative outcomes between sodium-glucose co-transporter 2 (SGLT2) inhibitors and thiazolidinediones in type 2 diabetes: a multicentre cohort study using the Korean healthcare database (2014–2025)

Abstract

Objective

Type 2 diabetes mellitus has been associated with an increased risk of cognitive decline and dementia, with patients being 1.5–2 times more likely to develop these conditions. While both sodium-glucose co-transporter 2 (SGLT2) inhibitors and thiazolidinediones (TZDs) have shown potential neuroprotective effects in previous studies, their comparative effectiveness for preventing neurodegenerative outcomes has not been established. This study aimed to compare the risk of stroke, dementia and Alzheimer’s disease (AD) between patients treated with SGLT2 inhibitors and those treated with TZDs.

Design

Multicentre, retrospective, observational, new-user, active-comparator cohort study.

Setting

Electronic health record-based databases from 11 secondary and tertiary institutions in South Korea from 1 January 2014 to 31 July 2025. The study period began in 2014, following the post-marketing surveillance initiation of SGLT2 inhibitors in Korea (November 2013), to ensure adequate drug availability and clinical adoption.

Participants

Patients aged 40 years or older who were newly prescribed either SGLT2 inhibitors or TZDs without prior exposure.

Interventions

Propensity score matching (1:1) was performed using sex as the primary covariate due to data availability constraints in the Observational Medical Outcomes Partnership Common Data Model framework. The HRs with 95% CIs were measured via Cox regression analysis.

Results

The study analysed 24 172 matched pairs for stroke outcomes (40 483 person-years in the SGLT2 inhibitor group and 39 363 person-years in the TZD group), 25 111 matched pairs for dementia (41 924 person-years in the SGLT2 inhibitor group and 40 726 person-years in the TZD group) and 25 237 matched pairs for AD (42 139 person-years in the SGLT2 inhibitor group and 40 895 person-years in the TZD group) across 11 participating hospitals. After a 1:1 propensity score matching, the SGLT2 inhibitors showed no significant difference in stroke risk (HR 1.18, 95% CI 0.62 to 2.23, p=0.62), while having significant reductions in dementia risk (HR 0.66, 95% CI 0.45 to 0.98, p=0.04) and AD risk (HR 0.54, 95% CI 0.35 to 0.83, p=0.005). Moreover, these protective effects for neurodegenerative outcomes were shown to be consistent across multiple hospital sites.

Conclusions

SGLT2 inhibitors are associated with a reduced risk of dementia and AD compared with TZDs in patients aged 40 years or older with type 2 diabetes and have neutral effects on stroke risk. These findings confirm the potential selective neuroprotective benefits of SGLT2 inhibitors for neurodegenerative outcomes, which may inform therapeutic decision-making for diabetic patients at risk of cognitive decline.

STRENGTHS AND LIMITATIONS OF THIS STUDY.

• This large-scale multicentre study used standardised Observational Medical Outcomes Partnership Common Data Model (OMOP-CDM) data from 11 hospitals, allowing for robust real-world evidence generation across diverse healthcare settings.

• The new-user, active-comparator design minimises prevalent user bias and provides a clinically relevant comparison between two antidiabetic drug classes.

• The long follow-up period (2014–2025) allowed adequate time for capturing delayed neurodegenerative outcomes.

• Propensity score matching was limited to sex due to OMOP-CDM constraints, potentially leaving unmeasured confounding from age, comorbidities and concurrent medications.

• A competing risk analysis for mortality could not be performed due to limitations in individual patient tracking within the de-identified framework.

Introduction

Dementia incidence has been gradually increasing with advancing age, with Alzheimer’s disease (AD) representing the primary cause and rapidly becoming one of the most expensive, lethal and burdensome diseases of this century.¹,³ The relationship between diabetes mellitus (DM) and cognitive dysfunction has been well established, with epidemiological studies suggesting that type 2 DM patients are 1.5–2 times more likely to develop dementia compared with healthy individuals.³ Based on the Mayo Clinic Alzheimer’s Disease Patient Registry, impaired glucose tolerance or DM affects 80% of all AD patients, highlighting the critical intersection between metabolic and neurodegenerative diseases.⁴

Type 2 DM patients tend to be at an increased risk of developing cerebrovascular accidents and cardiovascular mortality.⁴ They also face doubled stroke risk compared with the general population, with stroke representing a major cause of long-term disability and premature mortality in this population.⁵,⁷ The complex interplay between cerebrovascular and neurodegenerative pathology in diabetic patients requires the careful consideration of the effects of antidiabetic medications on both outcomes.

The protein sodium-glucose co-transporter 2 (SGLT2) is responsible for glucose reabsorption by the kidney. SGLT2 inhibitors, which are novel hypoglycaemic agents that reduce blood glucose by preventing renal glucose reabsorption, have shown promising neuroprotective potential.⁸ In murine models combining AD and type 2 DM, empagliflozin attenuated vascular damage and cognitive impairment, suggesting the neuroprotective potential of SGLT2 inhibitors.⁹ Recent real-world studies have reported that the patients prescribed SGLT2 inhibitors had lower dementia risk compared with those administered glucose-lowering agents, with one study reporting an 11% reduction.¹⁰ In real-world practice, patients with type 2 DM who are prescribed SGLT2 inhibitors tend to have a lower risk of incident dementia than those who are not.¹⁰ Another class of antidiabetic agents, thiazolidinediones (TZDs), which are peroxisome proliferator-activated receptor γ agonists with established antidiabetic efficacy, have also shown potential cognitive benefits. They decrease plasma free fatty acid concentration and fasting hyperglycaemia by exerting an insulin-reducing effect. A recent study on pioglitazone has shown that it may have therapeutic benefits. Pioglitazone studies have shown improved cerebral blood flow and reduced amyloid-beta and tau pathology in early-stage AD patients.¹¹ Clinical evidence has shown that TZDs affect cerebral glucose metabolism, reduce beta-amyloid accumulation and down-regulate inflammatory mediators, potentially improving neuronal function and memory formation.¹²,¹⁵

Given these advantages, it is clinically important to determine which drug class provides superior neuroprotection. Thus, this study aimed to compare the effects of SGLT2 inhibitors with those of TZDs on stroke, dementia and AD risk among type 2 DM patients using real-world data from multiple Korean hospitals.

Methods

Data sources

Electronic health record-based databases from 11 secondary and tertiary institutions in South Korea were used: (1) Soonchunhyang University Seoul Hospital (SCHSU), (2) Soonchunhyang University Bucheon Hospital (SCHBC), (3) Soonchunhyang University Cheonan Hospital (SCHCA), (4) Kangdong Sacred Heart Hospital (KDH), (5) Kangwon National University Hospital (KWMC), (6) Kyung Hee University Hospital (KHMC), (7) Kyung Hee University Hospital at Gangdong (KHNMC), (8) Bucheon Sejong Hospital (SEJONG_BCN), (9) Daegu Catholic University Hospital (DCMC), (10) Ajou University School of Medicine (AUMC) and (11) MyongJi Hospital (MJH). All databases contained de-identified, patient-level electronic health record data converted to the Observational Medical Outcomes Partnership Common Data Model (OMOP-CDM) version 5.3, allowing for distributed network research validated in previous international studies.

Study design

In this study, we conducted a multicentre, retrospective, observational, new-user, active-comparator study from 1 January 2014 to 31 July 2025. The study period began in 2014, following the post-marketing surveillance initiation of SGLT2 inhibitors in Korea (November 2013) to ensure sufficient drug availability and clinical adoption across participating institutions. This timing consideration was critical as dapagliflozin and empagliflozin received Korean insurance coverage in January 2016 and May 2016, respectively. Moreover, the patients with congestive heart failure were excluded from the entire study cohort to minimise potential bias from differential SGLT2 inhibitor prescribing patterns, as these agents have established cardiovascular benefits in heart failure populations that could influence treatment selection.

Study population

Type 2 DM patients aged 40 years or older who were previously naïve to both SGLT2 inhibitors and TZDs were included. The target cohort comprised patients first prescribed any SGLT2 inhibitor (eg, dapagliflozin, empagliflozin or ipragliflozin) for >180 consecutive days. The comparator cohort included the patients first prescribed any TZD (eg, rosiglitazone, pioglitazone or lobeglitazone) for >180 consecutive days. An exclusive user design wherein patients in each cohort had no prior exposure to the comparator drug class was applied. Specifically, the SGLT2 inhibitor cohort excluded any patients with previous TZD use, while the TZD cohort excluded any patients with previous SGLT2 inhibitor use, ensuring mutually exclusive treatment groups. Moreover, continuous drug exposure was established by allowing gaps of at least 180 days between prescriptions.

Outcomes and follow-up

The primary outcomes were incident stroke, dementia and AD occurring 180 days after treatment initiation and were identified using Systematized Nomenclature of Medicine Clinical Terms (SNOMED-CT) concepts within the OMOP-CDM framework. The index date was the first prescription date, while the cohort end date was defined as a continuous drug exposure cessation or outcome diagnosis. The time at risk spanned from 180 days post-index date to 180 days following the cohort end date. This 180-day induction period was implemented to minimise the risk of reverse causality by excluding the outcomes occurring immediately after treatment initiation that may be related to pre-existing subclinical conditions rather than drug effects, while allowing for delayed outcome manifestation that is characteristic of neurodegenerative diseases.

Statistical analysis

Large-scale propensity score matching was performed using the Observational Health Data Sciences and Informatics Cohort Method R. The propensity score model included sex as a variable due to data availability constraints in the OMOP-CDM framework. This was considered a limitation, as additional covariates such as age, comorbidities and concurrent medications would have strengthened the matching process. The ATLAS software V.2.7.6 was used, and the analysis was performed using FEEDER-NET, a Korean health data platform based on the OMOP-CDM. Greedy 1:1 matching was performed using a calliper of 0.2 times the SD. The HRs with 95% CIs were measured via Cox regression analysis. The cumulative incidence of the diseases was estimated via Kaplan-Meier analysis. The results from the individual databases were aggregated via random-effects meta-analysis. However, a competing risk analysis for mortality was not performed due to limitations in individual patient tracking within the de-identified framework. The assessment for the statistical tests of heterogeneity was performed using the χ² and I² statistics. The mean follow-up duration was measured as total person-years divided by the number of subjects; the median values could not be determined because the distributed research network architecture permits only aggregate summary statistics sharing across sites without individual patient-level data. Furthermore, all analyses were performed using the R statistical software (V.3.6.1; R Foundation for Statistical Computing).

Patient and public involvement

The patients or the public were not involved in the design, conduct, reporting or dissemination plans of this study.

Results

Baseline characteristics

Table 1 shows the baseline characteristics of the patients from all 11 participating hospitals before propensity score matching. This study included a total of 36 544 SGLT2 inhibitor users and 39 902 TZD users. Moreover, this table shows the medications that were prescribed across multiple hospitals; the less frequently used agents within each drug class that were limited to specific institutions are not shown, which explains why the percentages within drug classes do not sum to 100%. Of the SGLT2 inhibitor users, the age distribution showed the highest proportion in the 60–64 years group (17.77%), followed by the 55–59 years group (16.01%). For the TZD users, the 65–69 years age group showed the highest proportion (15.63%), followed by the 60–64 years (15.12%) and 70–74 years (14.10%) groups. The cohorts included younger patients comprising 6.95% of the SGLT2 inhibitor users and 4.55% of the TZD users in the 40–44 years group. Women comprised 37.71% of the SGLT2 inhibitor users, compared with 43.20% of the TZD users. Of the SGLT2 inhibitors, dapagliflozin was the most frequently prescribed (59.65%), followed by empagliflozin (36.56%) and ipragliflozin (3.81%). For TZDs, pioglitazone was considered the predominant choice (59.86%), followed by rosiglitazone (21.69%) and lobeglitazone (18.52%). The comorbidity patterns differed between the groups, with hyperlipidaemia present in 26.81% of the SGLT2 inhibitor users versus 14.00% of the TZD users (standardised difference=0.32). Essential hypertension was recorded in 43.53% of the SGLT2 inhibitor users and 37.71% of the TZD users. The metformin co-prescription rates were similar between the groups (57.73% vs 52.18%). Statin use showed some variation, with atorvastatin prescribed to 24.29% of the SGLT2 inhibitor users compared with 22.73% of the TZD users, while rosuvastatin use was higher in the SGLT2 inhibitor group (22.68% vs 12.56%).

Table 1. The baseline characteristics of the patients with SGLT2 inhibitors versus thiazolidinedione across 11 participating hospitals before propensity score matching.

	SGLT2 inhibitor		Thiazolidinedione		Std. diff
	Count, n	Percentage	Count, n	Percentage
Total number	36 544		39 902
Age group
40–44	2541	6.95	1814	4.55	0.1035
45–49	3369	9.22	2905	7.28	0.0705
50–54	4601	12.59	4181	10.48	0.0662
55–59	5850	16.01	5299	13.28	0.0772
60–64	6495	17.77	6035	15.12	0.0715
65–69	5037	13.95	6237	15.63	0.0474
70–74	3638	9.96	5628	14.10	0.1278
75–79	2620	7.17	4364	10.94	0.1316
80–84	1632	4.47	2416	6.05	0.0712
85–89	556	1.52	843	2.11	0.0442
Female	13 781	37.71	17 237	43.20	0.1120
Hyperlipidaemia	9798	26.81	5585	14.00	0.3221
Essential hypertension	15 906	43.53	15 049	37.71	0.1185
Medication use
Metformin	21 096	57.73	20 821	52.18	0.1117
Empagliflozin	31 360	36.56
Dapagliflozin	21 798	59.65
Ipragliflozin	1392	3.81
Pioglitazone			23 885	59.86
Rosiglitazone			8656	21.69
Lobeglitazone			7388	18.52
Linagliptin	2128	5.82	3375	8.46	0.1025
Sitagliptin	2295	6.28	3277	8.21	0.0746
Gemigliptin	1930	5.28	2160	5.41	0.0059
Teneligliptin	809	2.21	1019	2.55	0.0223
Saxagliptin	630	1.72	573	1.44	0.0231
Vildagliptin	915	2.50	1600	4.01	0.0849
Alogliptin	415	1.14	2538	6.36	0.2778
Gliclazide	4242	11.61	6238	15.63	0.1176
Glimepiride	5698	15.59	12 375	31.01	0.3710
Atorvastatin	8878	24.29	9068	22.73	0.0370
Rosuvastatin	8288	22.68	5011	12.56	0.2680
Pitavastatin	1489	4.07	1181	2.96	0.0605
Simvastatin	607	1.66	1882	4.72	0.1746
Pravastatin	474	1.30	919	2.30	0.0757

SGLT2 inhibitor, sodium-glucose co-transporter 2 inhibitor.

Primary outcomes

Table 2 summarises the distribution of the outcomes across 11 participating hospitals after propensity score matching. The analysis included 24 172 matched pairs for stroke outcomes (40 483 person-years in the SGLT2 inhibitor group and 39 363 person-years in the TZD group); 25 111 matched pairs for dementia (41 924 person-years in the SGLT2 inhibitor group and 40 726 person-years in the TZD group); and 25 237 matched pairs for AD (42 139 person-years in the SGLT2 inhibitor group and 40 895 person-years in the TZD group). The largest contributing site was AUMC, with over 3500 matched pairs per outcome, while the smallest contributing site was KWMC, with approximately 1000 pairs. The follow-up duration varied across institutions, with the total person-years ranging from approximately 3500 to over 10 700 per site. The mean follow-up duration was approximately 1.67 years for the SGLT2 inhibitor group and 1.62–1.63 years for the TZD group across all three outcomes.

Table 2. The number of subjects and follow-up duration in the target and comparator groups after propensity score matching (SGLT2 inhibitor vs TZD).

Hospital	Stroke		Dementia		Alzheimer’s disease
	SGLT inhibitor	TZD	SGLT inhibitor	TZD	SGLT inhibitor	TZD
	Count, n (target-years, PY)	Count, n (target-years, PY)	Count, n (target-years, PY)	Count, n (target-years, PY)	Count, n (target-years, PY)	Count, n (target-years, PY)
SCHSU	1527 (1992)	1527 (2167)	1613 (2090)	1613 (2251)	1619 (2101)	1619 (2272)
SCHBC	3219 (5402)	3219 (5378)	3289 (5508)	3289 (5492)	3307 (5542)	3307 (5520)
SCHCA	2546 (3539)	2546 (3438)	2580 (3564)	2580 (3480)	2589 (3580)	2589 (3494)
KDH	1397 (2148)	1397 (2894)	1459 (2223)	1459 (3012)	1465 (2238)	1465 (2994)
KWMC	971 (1955)	971 (1519)	974 (1933)	974 (1504)	981 (1928)	981 (1512)
KHMC	2991 (4786)	2991 (5304)	3299 (5236)	3299 (5814)	3320 (5267)	3320 (5845)
KHNMC	2573 (4332)	2573 (3741)	2714 (4550)	2714 (3968)	2729 (4580)	2729 (4001)
SEJONG_BCN	1662 (3972)	1662 (3820)	1668 (3985)	1668 (3738)	1683 (4023)	1683 (3761)
DCMC	1936 (3139)	1936 (2867)	2036 (3382)	2036 (3021)	2047 (3399)	2047 (3041)
AUMC	3564 (3139)	3564 (5647)	3681 (6378)	3681 (5856)	3687 (6389)	3687 (5835)
MJH	1786 (3029)	1786 (2588)	1798 (3075)	1798 (2590)	1810 (3092)	1810 (2620)
Total subjects (total target years)	24 172 (40 483)	24 172 (39 363)	25 111 (41 924)	25 111 (40 726)	25 237 (42 139)	25 237 (40 895)

AUMC, Ajou University Hospital; DCMC, Daegu Catholic University Hospital; KDH, Kangdong Sacred Heart Hospital; KHMC, Kyung Hee University Medical Center; KHNMC, Kyung Hee University Hospital at Gangdong; KWMC, Kangwon National University Hospital; MJH, MyongJi Hospital; SCHBC, Soonchunhyang University Bucheon Hospital; SCHCA, Soonchunhyang University Cheonan Hospital; SCHSU, Soonchunhyang University Seoul Hospital; SEJONG_BCN, Bucheon Sejong Hospital; SGLT2 inhibitor, sodium-glucose co-transporter 2 inhibitor; TZD, thiazolidinedione.

Comparative effectiveness analysis

When the data from 11 hospitals were combined, the differential effects emerged for the neurovascular outcomes versus the neurodegenerative outcomes: for stroke (figure 1), the SGLT2 inhibitors showed no significant difference compared with TZDs (HR 1.18, 95% CI 0.62 to 2.23, p=0.62). The individual hospital results varied, with MJH showing the lowest HR (0.29) and AUMC showing the highest HR (4.00), although most CIs were wide. A substantial heterogeneity was observed (I²=48%, p=0.04). Online supplemental figure 1 shows the individual hospital survival curves.

Figure 1. Forest plot for the stroke outcomes comparing sodium-glucose co-transporter 2 (SGLT2) inhibitors with thiazolidinediones. The forest plot showing HR and 95% CI for stroke outcomes across 11 participating hospitals. The SGLT2 inhibitors showed no significant difference compared with thiazolidinediones (pooled HR 1.18, 95% CI 0.62 to 2.23, p=0.62). Substantial heterogeneity was observed (I²=48%, p=0.04). AUMC, Ajou University Hospital; DCMC, Daegu Catholic University Hospital; KDH, Kangdong Sacred Heart Hospital; KHMC, Kyung Hee University Medical Center; KHNMC, Kyung Hee University Hospital at Gangdong; KWMC, Kangwon National University Hospital; MJH, MyongJi Hospital; SCHBC, Soonchunhyang University Bucheon Hospital; SCHCA, Soonchunhyang University Cheonan Hospital; SCHSU, Soonchunhyang University Seoul Hospital; SEJONG_BCN, Bucheon Sejong Hospital.

For dementia (figure 2), SGLT2 inhibitor use was associated with a significant 34% risk reduction (HR 0.66, 95% CI 0.45 to 0.98, p=0.04). This protective effect was consistent across most sites, with several hospitals showing individually favourable trends. Heterogeneity was moderate (I²=24%, p=0.22). Online supplemental figure 2 shows the individual hospital survival curves.

Figure 2. Forest plot for dementia outcomes comparing sodium-glucose co-transporter 2 (SGLT2) inhibitors with thiazolidinediones. The forest plot showing the HR and 95% CI for dementia outcomes across 11 participating hospitals. The SGLT2 inhibitors showed significant protective effects compared with thiazolidinediones (pooled HR 0.66, 95% CI 0.45 to 0.98, p=0.04), representing a 34% risk reduction. Moderate heterogeneity was observed (I²=24%, p=0.22). AUMC, Ajou University Hospital; DCMC, Daegu Catholic University Hospital; KDH, Kangdong Sacred Heart Hospital; KHMC, Kyung Hee University Medical Center; KHNMC, Kyung Hee University Hospital at Gangdong; KWMC, Kangwon National University Hospital; MJH, MyongJi Hospital; SCHBC, Soonchunhyang University Bucheon Hospital; SCHCA, Soonchunhyang University Cheonan Hospital; SCHSU, Soonchunhyang University Seoul Hospital; SEJONG_BCN, Bucheon Sejong Hospital.

For AD (figure 3), the protective effect was more significant, with SGLT2 inhibitors associated with a 46% risk reduction (HR 0.54, 95% CI 0.35 to 0.83, p=0.005). Multiple hospitals showed consistent protective effects, with MJH showing the strongest benefit (HR 0.29). Heterogeneity was low (I²=20%, p=0.93). Online supplemental figure 3 shows the individual hospital survival curves.

Figure 3. Forest plot for Alzheimer’s disease outcomes comparing sodium-glucose co-transporter 2 (SGLT2) inhibitors with thiazolidinediones. The forest plot showing the HR and 95% CI for Alzheimer’s disease outcomes across 11 participating hospitals. The SGLT2 inhibitors showed significant protective effects compared with thiazolidinediones (pooled HR 0.54, 95% CI 0.35 to 0.83, p=0.005), representing a 46% risk reduction. Low heterogeneity was observed (I²=20%, p=0.93). AUMC, Ajou University Hospital; DCMC, Daegu Catholic University Hospital; KDH, Kangdong Sacred Heart Hospital; KHMC, Kyung Hee University Medical Center; KHNMC, Kyung Hee University Hospital at Gangdong; KWMC, Kangwon National University Hospital; MJH, MyongJi Hospital; SCHBC, Soonchunhyang University Bucheon Hospital; SCHCA, Soonchunhyang University Cheonan Hospital; SCHSU, Soonchunhyang University Seoul Hospital; SEJONG_BCN, Bucheon Sejong Hospital.

Publication bias assessment

Table 3 summarises the meta-analysis results across all outcomes. The results of the publication bias assessment using the AS-Thompson test showed no significant asymmetry for any outcome (stroke, p=0.78; dementia, p=0.44; AD, p=0.92), supporting the robustness of our findings. The meta-analysis showed substantial heterogeneity for stroke outcomes (I²=48%, p=0.04), while the neurodegenerative outcomes showed low to moderate heterogeneity (dementia, I²=24%, p=0.22; AD, I²=20%, p=0.93). The consistency of the protective effects for the neurodegenerative outcomes across diverse hospital settings, despite varying patient populations and practice patterns, strengthens the internal validity of our results.

Table 3. The summary of the meta-analysis results.

Outcome	Studies	HR	95% CI	P value	I2 (%)	Heterogeneity p
Stroke	11	1.18	0.62 to 2.23	0.62	48	0.04
Dementia	11	0.66	0.45 to 0.98	0.04	24	0.22
Alzheimer’s disease	11	0.54	0.35 to 0.83	0.005	20	0.93

Discussion

Principal findings

This large-scale multicentre study across 11 Korean hospitals provides novel real-world evidence comparing the SGLT2 inhibitors and TZDs for neurovascular and neurodegenerative outcomes in type 2 diabetes. Our key finding—that SGLT2 inhibitors provide selective neuroprotection against dementia and AD without affecting stroke risk—has important clinical implications. The 34% reduction in dementia risk and 46% reduction in AD risk associated with SGLT2 inhibitor use suggests the clinically meaningful benefits that are consistent with the findings from recent observational studies.

Comparison with cardiovascular outcome trials

Our neutral findings for stroke are inconsistent with the mixed results from cardiovascular outcome trials. While the Effect of Sotagliflozin on Cardiovascular and Renal Events in Patients with Type 2 Diabetes and Moderate Renal Impairment Who Are at Cardiovascular Risk trial showed a 34% stroke reduction with sotagliflozin, the Canagliflozin Cardiovascular Assessment Study and Empagliflozin Cardiovascular Outcome Event Trial in Type 2 Diabetes Mellitus Patients trials reported neutral effects, and some analyses suggested a possible increased risk.¹⁶,¹⁸ Furthermore, our real-world data support the overall safety of the SGLT2 inhibitors regarding cerebrovascular outcomes while highlighting their selective benefits for neurodegenerative conditions.

Comparison with observational studies

Previous studies compared SGLT2 inhibitors with a placebo, while our study compared them with TZDs, which are commonly used in real-world clinical practice. Our findings showed no significant difference in stroke risk between the SGLT2 inhibitors and TZDs (HR 1.18, 95% CI 0.62 to 2.23), which is consistent with studies that have reported SGLT2 inhibitors as having a neutral effect on stroke risk. This suggests the comparable cerebrovascular safety profiles between the two drug classes. The magnitude of dementia risk reduction in the present study exceeds that reported in recent observational studies. A 2024 Korean National Health Insurance study (n=358 862) reported a 19% AD risk reduction when comparing SGLT2 inhibitors with all oral antidiabetic drugs,¹⁹ while an Ontario study among residents aged ≥66 years found a 20% dementia risk reduction when comparing SGLT2 inhibitors with dipeptidyl peptidase-4 inhibitors.¹⁸ Our findings suggest even greater benefits (34% dementia reduction, 46% AD reduction) when the SGLT2 inhibitors were compared specifically with TZDs. Despite the differences in the comparator groups and study populations, the convergent protective effects across these studies still provide robust evidence for SGLT2 inhibitor neuroprotection.

Potential mechanisms

Several potential mechanisms may explain the selective neuroprotective effects of SGLT2 inhibitors observed in the present study. These agents activate the cellular energy sensors (eg, AMPK, SIRT1) and induce metabolic reprogramming that improves autophagy and reduces neuroinflammation.^{20 21} SGLT2 inhibitors suppress NLRP3 inflammasome activation and promote ketogenesis, providing alternative energy substrates for neurons while reducing oxidative stress.^{22 23} Moreover, they modulate the neurotrophic factors (eg, brain-derived neurotrophic factor, nerve growth factor, glial cell line-derived neurotrophic factor) essential for neuronal survival and synaptic plasticity.²³ This multi-pathway neuroprotection may confirm the significant reductions in dementia and AD risk observed with SGLT2 inhibitor use compared with TZDs.^{20 23}

Clinical implications

The findings of this study have substantial clinical relevance. The 34%–46% reduction in the neurodegenerative outcomes with SGLT2 inhibitors compared with TZDs represents a meaningful difference that should inform prescribing decisions for diabetic patients at cognitive risk. Considering the growing burden of dementia in ageing diabetic populations, the preferential use of SGLT2 inhibitors could have a significant public health impact.

Limitations

Our study had several important limitations requiring transparent acknowledgement. The most significant constraint was our limited propensity score matching, which included only sex due to the OMOP-CDM framework restrictions. This represents a substantial limitation, as unmeasured confounding from age distribution, diabetes duration, severity indicators, comorbidities and concurrent medications may have influenced our results. Moreover, body weight data were not consistently available across participating hospitals in the OMOP-CDM framework, representing another potential source of unmeasured confounding that could influence the treatment selection and outcomes. However, the magnitude and consistency of the observed protective effects across multiple sites, as evidenced by the low statistical heterogeneity (I²=24% for dementia and 20% for AD), suggest that residual confounding alone is unlikely to fully explain our findings.

The temporal considerations of our study design also need further discussion. Although we restricted the analysis of the electronic health record-based databases from 2014 to 2025 when both drug classes were available, differential adoption patterns may have introduced selection bias. SGLT2 inhibitors, as newer agents, may have been preferentially prescribed to different patient populations than TZDs, although our new-user design partially addresses this concern.

Our inability to perform a competing risk analysis for mortality represents another limitation, although the impact may be less significant in our relatively younger cohort (aged ≥40 years) compared with studies focusing on elderly populations. Nevertheless, for long-term outcomes like dementia, competing mortality could still influence the effect estimates, particularly in the subset of patients aged ≥65 years who comprised a substantial proportion of our cohort.

Outcome ascertainment relied entirely on SNOMED-CT diagnostic codes without validation through cognitive testing or clinical assessment. This may have resulted in misclassification bias, although such bias would likely be non-differential between the treatment groups. The requirement for 180-day continuous treatment may have resulted in the selection of more adherent patients, potentially limiting generalisability.

Conclusions

SGLT2 inhibitors have shown neutral effects on stroke risk but significant protective effects against dementia (34% risk reduction) and AD (46% risk reduction) compared with TZDs in patients aged 40 years or older with type 2 diabetes. These findings from a large, multicentre real-world cohort suggest that SGLT2 inhibitors provide selective neuroprotective benefits for neurodegenerative outcomes without additional cerebrovascular risk. While the methodological constraints of our study—particularly limited covariate adjustment and the lack of competing risk analysis—prevent definitive causal inference, the magnitude and consistency of the observed associations across diverse clinical settings confirm the effectiveness of SGLT2 inhibitors in treating diabetic patients at an increased risk for cognitive decline. Future prospective studies incorporating comprehensive baseline characteristics, cognitive assessments and the appropriate handling of competing risks are important to confirm these promising findings and determine the underlying mechanisms.

Data availability statement

Data are available upon reasonable request.