Effect of Combination Therapy with SGLT2 Inhibitors and GLP-1 Receptor Agonists on Myocardial Infarction and Stroke in Type 2 Diabetes: A Systematic Review and Meta-Analysis

Abstract

Introduction

Sodium-glucose cotransporter-2 inhibitors (SGLT2is) and glucagon-like peptide-1 receptor agonists (GLP-1RAs) offer cardioprotection in type 2 diabetes mellitus (T2DM). We evaluated their monotherapy versus their combined use for prevention of myocardial infarction (MI) and stroke.

Methods

Conducted according to preferred reporting items for systematic reviews and meta-analyses (PRISMA) 2020 guidelines, five databases were systematically searched through November 2025 for studies evaluating use of SGLT2i and GLP-1RA in combination versus their monotherapy in adults with T2DM. Hazard ratios (HRs) for MI and stroke were pooled utilizing random-effects models. Subgroup interaction testing and grading of recommendations assessment, development, and evaluation (GRADE) certainty assessments were performed.

Results

Of the studies, eight comprising ten comparisons were included. Combination therapy was associated with a lower overall risk of MI (HR 0.79, 95% CI 0.70–0.88; I² = 72.8%) and stroke (HR 0.85, 95% CI 0.77–0.93; I² = 63.5%) compared with their monotherapy. For MI, combination therapy significantly outperformed SGLT2i monotherapy (HR 0.74, 95% CI 0.60–0.90) but not GLP-1RA monotherapy (HR 0.80, 95% CI 0.55–1.16). For stroke reduction, combination therapy outperformed both SGLT2i (HR 0.73, 95% CI 0.54–0.99) and GLP-1RA (HR 0.92, 95% CI 0.88–0.96) monotherapy. Cerebrovascular benefit appeared stronger in adults > 65 years (HR 0.50, 95% CI 0.37–0.68). Overall certainty of evidence was very low.

Conclusions

In adults with T2DM, SGLT2i and GLP-1RA combination therapy may be associated with lower risks of MI and stroke compared with their monotherapy. While the MI benefit appears primarily to be driven by GLP-1RA, combination therapy yields an additive reduction in stroke risk, particularly in older adults. Given the very low certainty of evidence, these hypothesis-generating findings require confirmation through dedicated prospective trials.

Systematic Review Registration

https://www.crd.york.ac.uk/PROSPERO/view/CRD420261320374.

Supplementary Information

The online version contains supplementary material available at https://doi.org/10.1007/s13300-026-01911-2.

Key Summary Points

Why carry out this study?

While SGLT2 inhibitors (SGLT2i) and glucagon-like peptide-1 receptor agonists (GLP-1RA) each confer cardiovascular protection in type 2 diabetes, their additive value for specific vascular events remains unclear.

This systematic review and meta-analysis examined whether combining these agents is associated with lower risks of myocardial infarction (MI) and stroke compared with monotherapy.

What was learned from the study?

For MI, combination therapy was associated with lower risk compared with SGLT2i monotherapy but not GLP-1RA monotherapy, suggesting the observed association may primarily reflect GLP-1RA effects.

For stroke, combination therapy was associated with lower risk compared with both individual drug classes, a potential additive benefit that appeared stronger in adults aged 65 and older.

Given the very low certainty of the current evidence, these hypothesis-generating findings highlight a potential vascular-specific rationale for combination therapy that requires confirmation through dedicated prospective trials.

Introduction

Cardiovascular complications remain the primary drivers of morbidity and mortality in people with type 2 diabetes mellitus (T2DM), with myocardial infarction (MI) and stroke representing the most disabling and prognostically significant outcomes [1]. While these conditions frequently coexist, their underlying pathophysiologies diverge across distinct vascular beds. MI is fundamentally a coronary atherothrombotic event, whereas stroke is a heterogeneous condition involving complex cerebrovascular and cardioembolic pathways [2, 3]. Hence, contemporary management has transitioned from a strictly glucocentric focus toward an integrated cardiometabolic strategy designed to reduce specific cardiovascular events through pleiotropic mechanisms [4]. Sodium-glucose cotransporter-2 inhibitors (SGLT2i) and glucagon-like peptide-1 receptor agonists (GLP-1RA) serve as the foundation for this approach, providing cardio-protection via complementary pathways. SGLT2is predominantly optimize hemodynamic and cardiorenal stability, while GLP-1RAs exhibit robust antiatherosclerotic and antiinflammatory effects [5]. This mechanistic complementarity provides a compelling rationale for their combination therapy in patients who continue to carry high residual cardiovascular risk [6].

Despite growing clinical adoption, the incremental benefit of combined SGLT2i and GLP-1RA therapy over active monotherapy remains uncertain, particularly regarding individual vascular end points. A major limitation of the existing evidence is its frequent reliance on major adverse cardiovascular events (MACE) as a composite primary outcome [7]. Because MI and stroke exhibit varying sensitivities to clinical risk factors, relying on composite outcomes can easily obscure divergent treatment effects across biologically distinct vascular beds [8, 9]. Consequently, evaluating combination therapy solely through MACE may obscure whether dual therapy provides meaningful additive protection against specific individual events.

To address this gap, we conducted a systematic review and meta-analysis of both randomized controlled trials (RCTs) and observational cohorts to evaluate the comparative cardiovascular effectiveness of SGLT2i and GLP-1RA combination therapy versus their monotherapy in adults with T2DM. By disaggregating MACE and analyzing MI and stroke as distinct primary end points, we sought to determine if their divergent pathophysiologies respond uniquely to dual therapy. Furthermore, we employed rigorous subgroup interaction testing to evaluate how specific clinical determinants, including age, baseline insulin use, and cardiovascular history, modify the therapeutic response within these separate vascular beds. This high-resolution approach allows for the identification of potential clinical signals and provides an evidence-based framework to determine which patient profiles derive the greatest utility from combination therapy.

Methods

This systematic review and meta-analysis was conducted in accordance with a prespecified protocol registered on PROSPERO (CRD420261320374). The study follows the preferred reporting items for systematic reviews and meta-analyses 2020 guidelines (PRISMA 2020).

Eligibility Criteria

Eligibility criteria were defined using the population, intervention, comparator, outcome, study design (PICOS) framework. Eligible studies included adults (aged > 18 years) with a confirmed diagnosis of T2DM. We included populations across the cardiovascular risk spectrum, including patients with established atherosclerotic cardiovascular disease, heart failure, chronic kidney disease, or multiple cardiovascular risk factors. Studies involving participants on background glucose-lowering medications, including insulin, were also considered eligible.

The intervention of interest was combination therapy utilizing a SGLT2i and a GLP-1RA. Comparators included monotherapy with either an SGLT2i or a GLP-1RA. The primary outcomes were MI and stroke, including fatal, nonfatal, or otherwise specified events as reported by individual studies. Outcomes were evaluated at the longest available follow up reported in each study. The preferred effect measure for synthesis was the hazard ratio (HR) with corresponding 95% confidence intervals (CI).

Both randomized controlled trials and nonrandomized study designs were eligible for inclusion. There were no restrictions on geographical location, clinical setting (including primary, secondary, or specialist centers), publication date, or language. Studies were excluded if they did not evaluate the comparative efficacy or safety of combination therapy versus the specified monotherapies or placebo. Furthermore, studies were excluded if they reported only MACE outcomes without providing individual data for MI or stroke. This criterion was based on outcome-reporting granularity rather than on the direction or significance of the reported effects; because the decision to report disaggregated component end points is generally prespecified by study design and independent of the observed results, this restriction was not expected to preferentially select studies with particular effect estimates. We also excluded studies where primary outcome data were not extractable or those reporting only all-cause mortality without the specified cardiovascular end points. Reviews, editorials, and case reports were also excluded.

Search Strategy

PubMed (MEDLINE), Embase, Scopus, Web of Science, and the Cochrane Central Register of Controlled Trials (CENTRAL) were searched from inception to 17 November 2025. Search terms included combinations of diabetes-related terms, such as “diabetes mellitus, type 2,” “type 2 diabetes,” “T2DM,” and “T2D,” with SGLT2 inhibitor-related terms, including “SGLT2 inhibitor,” “SGLT2i,” “empagliflozin,” “dapagliflozin,” and “canagliflozin.” These were combined with GLP-1 receptor agonist terms, such as “GLP-1 receptor agonist,” “GLP1RA,” “liraglutide,” “semaglutide,” and “tirzepatide.” To capture clinical end points, terms for cardiovascular outcomes were utilized, including “major adverse cardiovascular events,” “MACE,” “myocardial infarction,” “MI,” “stroke,” and “mortality,” alongside keywords for combined use such as “combination therapy,” “dual therapy,” “add-on,” and “concomitant.” Terms were combined using Boolean operators AND/OR and searched using Medical Subject Headings (MeSH), Emtree terms, and keywords in titles and abstracts. Reference lists of included studies and relevant reviews were manually screened to identify additional eligible articles.

Study Selection

All retrieved records were imported into EndNote V21, and duplicates were removed. Two reviewers independently screened titles and abstracts according to the predefined eligibility criteria. Full texts of potentially eligible studies were then assessed independently by the same reviewers. Disagreements were resolved through discussion or consultation with a third reviewer. The study selection process is summarized in the PRISMA flow diagram shown in Fig. 1.

Fig. 1.

PRISMA flow diagram of the study selection. MI, myocardial infarction; PRISMA, preferred reporting items for systematic reviews and meta-analyses

Data Extraction and Quality Assessment

Data were extracted independently by two reviewers using a standardized form. Extracted study characteristics included the first author, year of publication, study design, geographical location, and data source. We also extracted participant demographics, sample size, and clinical history, including diabetes duration, background therapies, and baseline cardiovascular and renal comorbidities. Specific details regarding the SGLT2i and GLP-1RA combination, comparator definitions, and the duration of follow up were recorded for each study. For clinical outcomes, the incidence of fatal or nonfatal MI and stroke was extracted.

The primary effect measure was the HR with its corresponding 95% CI. We prioritized data from the most fully adjusted statistical models and used estimates from the longest available follow up period reported. Authors were not contacted for additional information; if required data were unavailable, we utilized the published information and addressed the limitation during risk-of-bias and certainty assessments.

Risk of bias was assessed independently by two reviewers. Nonrandomized studies were assessed using the risk of bias in nonrandomized studies—of interventions (ROBINS-I) tool, evaluating confounding, participant selection, intervention classification, and other potential biases [10]. Randomized trials were evaluated using the Cochrane risk of bias tool (RoB 2.0), covering domains such as randomization, deviations from intended interventions, missing outcome data, measurement, and selection of reported results [11]. Disagreements were resolved through discussion or consultation with a third reviewer. The risk of bias owing to missing results (reporting bias) was assessed where sufficient studies were available. The certainty of evidence for each primary outcome was rated as high, moderate, low, or very low using the GRADE approach [12]. This assessment considered domains of risk of bias, inconsistency, indirectness, imprecision, and publication bias.

Data Synthesis and Statistical Analysis

All quantitative syntheses and statistical analyses were performed using R (version 4.3.2; R Foundation for Statistical Computing, Vienna, Austria) within the RStudio integrated development environment, primarily using the meta (version 7.0-0) and metafor (version 4.4-0) packages [13, 14]. Pooled effect sizes across the ten distinct comparisons were calculated as hazard ratios (HRs) with 95% CIs using a random-effects model via the inverse-variance method to account for expected clinical and methodological heterogeneity. Between-study heterogeneity was assessed using Cochran’s Q-test and quantified using the I² statistic. An I² > 50% was considered indicative of substantial heterogeneity. Where quantitative synthesis was not feasible, we provided a narrative summary of findings.

Subgroup analyses were performed to investigate potential effect modification by comparator type (SGLT2i versus GLP-1RA), age (< 65 versus > 65 years), baseline insulin use, and history of cardiovascular comorbidities. Stratified subgroup analyses were conducted using the update.meta and subgroup utilities within the meta package, and statistical interaction between subgroups was evaluated using the Q-test, where a p < 0.05 was defined as significant. Sensitivity analyses were conducted to assess the impact of study design and bias control by comparing propensity score-matched cohorts with nonmatched studies. Finally, publication bias and small-study effects were assessed through visual inspection of funnel plots and statistically via linear regression of the standard error on the log-transformed hazard ratios (Egger’s test), implemented using the metabias function.

Ethical Approval

This article is based on previously conducted studies and does not contain any new studies with human participants or animals performed by any of the authors.

Results

Study Selection

The data base search identified 2414 records, of which 321 duplicates were removed. After screening 2,093 titles and abstracts, 2064 records were excluded. Of the reports, 29 underwent full-text assessment, and 21 were excluded because they reported composite MACE outcomes without separate data for nonfatal MI or stroke (n = 12), reported only all-cause mortality (n = 4), or had an ineligible design or insufficient extractable data (n = 5). We included eight unique studies, contributing ten distinct comparisons, in the quantitative synthesis (Fig. 1).

Study Characteristics

We included ten comparisons derived from eight unique studies in the quantitative synthesis [15–22]. Combination therapy with an SGLT2 inhibitor and a GLP-1RA was compared with their monotherapy. Of the comparisons, five used propensity score matching, four used nonmatched observational cohorts, and one was derived from randomized trial data. Included populations varied by age, insulin use, cardiovascular history, and comorbidity profile, as summarized in Table 1.

Table 1.

Summary characteristics of studies included in the meta-analysis

Study	Country	Data source	Design	Population	Comparator	Total n	Follow up (months)	Age (mean)	Female (%)	PS matched
Chaiyakunapruk 2025 [15]	USA	Komodo’s healthcare map	Retrospective observational cohort	Adults with T2D and history of ASCVD	Combo versus SGLT2i	164,910	14.5	63.70 (SD 10.45)	39.3%	No
Clegg 2019 [16]	Multinational (across 35 countries)	EXSCEL randomized trial dataset	Retrospective observational cohort	Adults with T2D	Combo versus GLP1RA	1150	13.3 (combo) versus 15.1 (control)	62 (SD 9)	31.0%	Yes
Liu 2025 [17]	China	Single-center hospital EMR	Retrospective observational cohort	Adults with T2D and acute coronary syndrome (ACS)	Combo versus SGLT2i	416	12	63	32.2%	Yes
Luo 2023 [18]	Hong Kong	CDARS (Hong Kong Hospital Authority)	Retrospective observational cohort	Adults with T2D	Combo versus GLP1RA	1674	65.8	52.5 (SD 10.7)	41.6%	Yes
Mann 2024 [19]	Multinational	FLOW trial (NCT03819153)	Posthoc subgroup analysis of RCT	T2DM and CKD	NR	3533	40.8	66.6	30.3%	No
Marx 2025 [20]	Multinational (across 33 countries)	SOUL trial (NCT03914326)	Double-blind placebo-controlled RCT	T2DM with ASCVD and/or CKD	NR	9650	47.5	65	21.5%	Randomization
Riley 2023 (versus SGLT2i) [21]	International	TriNetX	Retrospective cohort	T2DM and insulin users	Combo versus SGLT2i	252,247	60	58.7	49.0%	Yes
Riley 2023 (versus GLP1RA) [21]	International	TriNetX	Retrospective cohort	T2DM and insulin users	Combo versus GLP1RA	295,351	60	58.7	49.0%	Yes
Simms-Williams 2025 (versus GLP1RA) [22]	UK	UK Clinical Practice Research Datalink (CPRD)	Retrospective observational cohort	Adults with T2D using GLP-1RA	Combo versus GLP1RA	13,392	9	56.7 (SD 10.4)	45.5%	No
Simms-Williams 2025 (versus SGLT2i) [22]	UK	UK Clinical Practice Research Datalink (CPRD)	Retrospective observational cohort	Adults with T2D using SGLT2i	Combo versus SGLT2i	17,884	8.4	57.6 (SD 10.6)	47.7%	No

ASCVD, atherosclerotic cardiovascular disease; CKD, chronic kidney disease; EMR, electronic medical record; GLP-1RA, glucagon-like peptide-1 receptor agonist; NR, not reported; PSM, propensity score matching; SGLT2i, sodium–glucose cotransporter-2 inhibitor; T2D, type 2 diabetes

Risk of Bias

Risk of bias was assessed separately according to study design, using ROBINS-I for nonrandomized studies (Supplementary Figs. S1 and S2) and RoB 2 for randomized trial evidence (Supplementary Figs. S3 and S4). Among the nonrandomized studies, five were judged to have a moderate risk of bias [15–18, 22] and one was judged to have a serious risk of bias [21]. The moderate-risk studies generally used robust real-world methods, including propensity score matching or weighting, active-comparator designs, target-trial emulation or prevalent new-user approaches, and objective outcome definitions for MI and stroke. However, they remained vulnerable to residual confounding, treatment-selection bias, incomplete adherence information, and possible exposure misclassification. The single serious-risk study was downgraded primarily because of limited confounder adjustment and dependence on electronic health record capture for outcomes and mortality, increasing concern that the observed associations could be influenced by unmeasured differences between treatment groups. Conversely, the randomized trial evidence was judged to be at low risk of bias [19, 20], reflecting central randomization, double blinding, objective cardiovascular outcomes, and intention-to-treat analyses.

Myocardial Infarction

Combination therapy was associated with a lower risk of MI compared with monotherapy overall (pooled HR 0.79, 95% CI 0.70–0.88; p < 0.001). Substantial between-study heterogeneity was observed (I² = 72.8%; heterogeneity p < 0.001) (Fig. 2).

Fig. 2.

Forest plot of the association between combination therapy with an SGLT2i and a GLP-1RA and the risk of MI compared with monotherapy. CI, confidence interval; GLP-1RA, glucagon-like peptide-1 receptor agonist; HR, hazard ratio; MI, myocardial infarction; SGLT2i, sodium–glucose cotransporter-2 inhibitor

Subgroup Analysis

By comparator type, combination therapy was associated with a lower risk of MI versus SGLT2i monotherapy (HR 0.74, 95% CI 0.60–0.90), whereas the association versus GLP-1RA monotherapy was not statistically significant (HR 0.80, 95% CI 0.55–1.16). The test for subgroup differences by comparator type was not statistically significant (p = 0.69) (Fig. 3).

Fig. 3.

Forest plot of the association between combination therapy and MI risk, stratified by comparator type. CI, confidence interval; GLP-1RA, glucagon-like peptide-1 receptor agonist; HR, hazard ratio; MI, myocardial infarction; SGLT2i, sodium–glucose cotransporter-2 inhibitor

In age-stratified analyses, the association was statistically significant among participants aged < 65 years (HR 0.69, 95% CI 0.60–0.79), but not among those aged > 65 years (HR 0.79, 95% CI 0.51–1.21). The test for subgroup differences by age was not statistically significant (p = 0.54) (Fig. 4).

Fig. 4.

Forest plot of the association between combination therapy and MI risk, stratified by age group. CI, confidence interval; HR, hazard ratio; MI, myocardial infarction

In analyses stratified by insulin use, neither insulin users (HR 0.74, 95% CI 0.53–1.03) nor noninsulin users (HR 0.85, 95% CI 0.67–1.07) showed a statistically significant association. The test for subgroup differences was not statistically significant (p = 0.50).

In analyses stratified by baseline ischemic heart disease (IHD), the association was not statistically significant among participants with IHD (HR 1.11, 95% CI 0.38–3.21) or without IHD (HR 0.81, 95% CI 0.61–1.08). The test for subgroup differences was not statistically significant (p = 0.57).

In addition, in analyses stratified by hyperlipidemia status, neither subgroup showed a statistically significant association, with wide confidence intervals among participants with hyperlipidemia (HR 1.29, 95% CI 0.30–5.59) and without hyperlipidemia (HR 1.44, 95% CI 0.16–12.54). The test for subgroup differences was not statistically significant (p = 0.93) (Table 2).

Table 2.

Pooled hazard ratios and subgroup analyses for myocardial infarction

Variables	No. (comparisons/studies)	Pooled HR (95% CI)	Between-studies heterogeneity		Test for subgroup differences		Small-study effects assessment
			I2 (%)	p value	Q	p value	B	SE	p value
Overall	10 (8)	0.79 (0.70–0.88)	72.75	0.00	–	–	−0.70	0.52	0.179
Comparator type
Combo versus SGLT2i	4	0.74 (0.60–0.90)	64.97	0.04	0.16	0.69
Combo versus GLP1RA	4	0.80 (0.55–1.16)	60.36	0.06
Propensity score matching
No	4 (3)	0.67 (0.58–0.78)	0.00	0.66	6.76	0.01
Yes	5 (4)	0.87 (0.77–0.98)	74.89	0.00
Age
< 65	4	0.69 (0.60–0.79)	5.07	0.37	0.38	0.54
> 65	3	0.79 (0.51–1.21)	25.31	0.26
Insulin
Insulin users	1	0.74 (0.53–1.03)	–	–	0.45	0.50
Without insulin	2	0.85 (0.67–1.07)	1.98	0.16
IHD
IHD	2	1.11 (0.38–3.21)	84.11	0.01	0.32	0.57
NonIHD	1	0.81 (0.61–1.08)	–	–
Hyperlipidemia
Hyperlipidemia	2	1.29 (0.30–5.59)lePara>	69.18	0.07	0.01	0.93
Nonhyperlipidemia	2	1.44 (0.16–12.54)	73.83	0.05

B and SE refer to the coefficient and standard error from Egger’s regression test. I² represents the percentage of variability attributable to between-study heterogeneity. HR, hazard ratio; CI, confidence interval; GLP-1RA, glucagon-like peptide-1 receptor agonist; SGLT2i, sodium–glucose cotransporter-2 inhibitor; IHD, ischemic heart disease

Stroke

Combination therapy was associated with a lower risk of stroke compared with monotherapy overall (pooled HR 0.85, 95% CI 0.77–0.93; p < 0.001). Moderate-to-substantial between-study heterogeneity was observed (I² = 63.5%; heterogeneity p < 0.001) (Fig. 5).

Fig. 5.

Forest plot of the association between combination therapy with an SGLT2i and a GLP-1RA and the risk of stroke compared with monotherapy. CI, confidence interval; GLP-1RA, glucagon-like peptide-1 receptor agonist; HR, hazard ratio; SGLT2i, sodium–glucose cotransporter-2 inhibitor

Subgroup Analysis

By comparator type, combination therapy was associated with lower risk for stroke versus either SGLT2i monotherapy (HR 0.73, 95% CI 0.54–0.99; I² = 83.8%) and GLP-1RA monotherapy (HR 0.92, 95% CI 0.88–0.96; I² = 0.0%). The test for subgroup differences by comparator type was not statistically significant (p = 0.14) (Fig. 6).

Fig. 6.

Forest plot of the association between combination therapy and stroke risk, stratified by comparator type. CI, confidence interval; GLP-1RA, glucagon-like peptide-1 receptor agonist; HR, hazard ratio; SGLT2i, sodium-glucose cotransporter-2 inhibitor

In age-stratified analyses, the association differed by age group, with a statistically significant test for subgroup differences (p = 0.02). Combination therapy was associated with lower risk of stroke among participants aged > 65 years (HR 0.50, 95% CI 0.37–0.68), whereas the association was not statistically significant among those aged < 65 years (HR 0.82, 95% CI 0.61–1.11) (Fig. 7).

Fig. 7.

Forest plot of the association between combination therapy and stroke risk, stratified by age group. CI, confidence interval; HR, hazard ratio

In analyses stratified by insulin use, the test for subgroup differences was statistically significant (p < 0.001). The estimate among insulin users was HR 2.33 (95% CI 1.11–4.90), whereas the estimate among noninsulin users was HR 0.61 (95% CI 0.36–1.01). Because the insulin-user subgroup was based on a single comparison, this finding should be interpreted cautiously and is regarded as exploratory and hypothesis-generating rather than confirmatory.

When stratified by stroke definition, the association was statistically significant for ischemic stroke (HR 0.67, 95% CI 0.50–0.91) and for unspecified stroke (HR 0.91, 95% CI 0.88–0.94), but not for nonfatal stroke (HR 1.00, 95% CI 0.73–1.37) or stroke/transient ischemic attack (TIA) (HR 0.58, 95% CI 0.24–1.42). The test for subgroup differences was not statistically significant (p = 0.16) (Table 3).

Table 3.

Pooled hazard ratios and subgroup analyses for stroke

Variables		No. (studies/comparisons)	Pooled HR (95% CI)	Between studies heterogeneity		Test for subgroup differences		Small-study effects assessment
				I2 (%)	p value	Q	p value	B	SE	p value
Overall		10 (8)	0.85 (0.77–0.93)	63.48	0.00			−0.55	0.49	0.266
Comparator type	Combo versus SGLT2i	4	0.73 (0.54–0.99)	83.82	0.00	2.17	0.14
	Combo versus GLP1RA	4	0.92 (0.88–0.96)	0.00	0.73
Propensity score matching	No	4 (3)	0.84 (0.55–1.29)	76.41	0.01	0.12	0.73
	Yes	5 (4)	0.91 (0.87–0.94)	6.25	0.37
Definition of stroke	Ischemic	3 (2)	0.67 (0.50–0.91)	29.48	0.24	5.22	0.16
	Nonfatal stroke	2	1.00 (0.73–1.37)	53.16	0.14
	Unspecified	4 (3)	0.91 (0.88–0.94)	8.90	0.35
	Stroke and TIA	1	0.58 (0.24–1.42)	–	–
Age	< 65	4	0.82 (0.61–1.11)	66.14	0.03	5.41	0.02
	> 65	3	0.50 (0.37–0.68)	0.00	0.84
Insulin	Insulin users	1	2.33 (1.11–4.90)	–	–	8.55	0.00
	Without insulin	2	0.61 (0.36–1.01)	0.72	0.40

B and SE refer to the coefficient and standard error from Egger’s regression test. I² represents the percentage of variability attributable to between-study heterogeneity. HR, hazard ratio; CI, confidence interval; GLP-1RA, glucagon-like peptide-1 receptor agonist; SGLT2i, sodium-glucose cotransporter-2 inhibitor; transient ischemic attack (TIA), transient ischemic attack

Sensitivity Analysis and Small-Study Effects

In analyses stratified by propensity score matching, the pooled estimate for MI differed by matching status. Nonmatched comparisons showed a larger association (HR 0.67, 95% CI 0.58–0.78) than propensity-matched comparisons (HR 0.87, 95% CI 0.77–0.98), with a statistically significant test for subgroup differences (p = 0.01) (Fig. 8). For stroke, estimates were directionally consistent across nonmatched comparisons (HR 0.84, 95% CI 0.55–1.29) and propensity-matched comparisons (HR 0.91, 95% CI 0.87–0.94), with no statistically significant subgroup difference (p = 0.73) (Fig. 9).

Fig. 8.

Sensitivity analysis of the association between combination therapy and MI risk, stratified by propensity score matching status. CI, confidence interval; HR, hazard ratio; MI, myocardial infarction

Fig. 9.

Sensitivity analysis of the association between combination therapy and stroke risk, stratified by propensity score matching status. CI, confidence interval; HR, hazard ratio

Egger’s regression test did not suggest evidence of small-study effects for MI (p = 0.179) or stroke (p = 0.266). These findings should be interpreted cautiously given the limited number of included comparisons (Fig. 10).

Fig. 10.

Funnel plots assessing small-study effects for MI and stroke outcomes. Funnel plots display the logarithm of the hazard ratio against the standard error for MI (A) and stroke (B). Each point represents an included comparison. The vertical line indicates the pooled effect estimate, and dashed lines represent the 95% pseudo-confidence limits. HR, hazard ratio; MI, myocardial infarction

Certainty of Evidence

The overall certainty of evidence was rated as very low for both primary end points (Supplementary Table S1). For MI, the evidence was initially rated as low owing to the predominance of observational data. This was subsequently downgraded one level for serious inconsistency, driven by substantial between-study heterogeneity (I² = 72.8%, p < 0.001). No further downgrades were applied for imprecision, as the pooled confidence interval excluded the null (HR 0.79, 95% CI 0.70–0.88), nor for publication bias, as Egger’s regression was nonsignificant (p = 0.179), although the limited number of comparisons warrants cautious interpretation. Similarly, the certainty of evidence for stroke was rated as very low. The predominantly observational baseline was downgraded one level for serious inconsistency owing to moderate-to-substantial heterogeneity (I² = 63.5%, p < 0.001). The pooled estimate remained precise (HR 0.85, 95% CI 0.77–0.93), and Egger’s test did not indicate significant small-study effects (p = 0.266), precluding further downgrades.

Discussion

In this systematic review and meta-analysis, we find that combination therapy with an SGLT2i and a GLP-1RA is associated with a lower overall risk of MI and stroke compared with their monotherapy in adults with T2DM. Importantly, by evaluating these events as distinct end points rather than traditional composite outcomes, our analysis reveals critical nuances in the efficacy of dual-pathway intensification across different vascular beds. For MI, the observed association appears more closely aligned with the GLP-1RA component; i.e., combination therapy was associated with a lower risk relative to SGLT2i monotherapy but showed no statistically significant additional benefit over GLP-1RA monotherapy alone. In contrast, the reduction in stroke suggested additive benefit by significantly outperforming both individual drug classes—a cerebrovascular benefit that appeared stronger in older adults. However, because the included studies are predominantly observational and exhibit substantial heterogeneity, the overall certainty of evidence is very low. Therefore, while these findings suggest that the clinical utility of dual therapy depends heavily on the target vascular bed as well as baseline patient demographics, they must be interpreted cautiously until they are confirmed by dedicated randomized trials.

Our finding that combination therapy is associated with significantly lower MI risk compared with SGLT2i monotherapy, but not GLP-1RA monotherapy, is biologically consistent with the divergent event profiles observed in landmark cardiovascular outcome trials (CVOTs) [23]. MI is fundamentally an atherothrombotic event, a pathology that responds to GLP-1RAs [24]. Clinical trial data corroborate this pattern; in harmony outcomes, the GLP-1RA albiglutide significantly reduced three-point MACE in patients with T2DM and established cardiovascular disease, and among the individual components, only MI showed a statistically significant reduction (fatal or nonfatal MI: HR 0.75, 95% CI 0.61–0.90) [25]. In contrast, comprehensive CVOT meta-analyses demonstrate that while the SGLT2i class confers broad cardiovascular protection, its most profound and reproducible effects remain concentrated in mitigating heart failure and decline in renal function, whereas its effects on atherosclerotic cardiovascular events are more modest [26]. Consequently, initiating a GLP-1RA in a patient already taking an SGLT2i may address residual atherosclerotic risk, yielding the significant MI reduction seen in our pooled data. Notably, SGLT2 inhibition also exerts direct effects within the ischemic myocardium and its surrounding milieu, with reported reductions in pericoronary adipose inflammation, infarct size and inflammatory burden, and subsequent intracoronary restenosis and MACE following acute MI [27–29]. Consistent with an additive coronary benefit, combined SGLT2i/GLP-1RA therapy has been associated with fewer cardiovascular events after acute MI [30]. However, in patients being treated with a GLP-1RA, atherothrombotic pathways may already be substantially modulated. This may limit the incremental MI benefit of subsequently adding an SGLT2i; this may provide a plausible explanation for the absence of statistically significant added protection against coronary ischemia in this subgroup [23, 26]. Importantly, our subgroup analyses underscore the consistency of this relationship. We observed no significant effect modification by age, baseline insulin use, prior ischemic heart disease, or hyperlipidemia. Clinically, the lack of significant modification by prior ischemic heart disease suggests that baseline coronary disease status alone may not identify patients most likely to derive incremental MI benefit from dual therapy, although this finding is limited by imprecise subgroup estimates and few comparisons. This uniformity suggests that the observed MI association was broadly consistent across the examined patient phenotypes, regardless of their baseline metabolic or cardiovascular risk tier.

In contrast to the MI outcomes, our analysis suggests that combination therapy is associated with additional protection against stroke, with lower risk relative to either SGLT2i or GLP-1RA monotherapy. While CVOTs such as semaglutide in subjects with type 2 diabetes (SUSTAIN-6) have shown that GLP-1RAs can reduce stroke risk, the incremental benefit observed with the addition of an SGLT2i suggests that comprehensive cerebrovascular protection may require broader physiological modulation than coronary atherothrombosis alone [23, 31]. This concept is supported by our subgroup analysis demonstrating that the risk reduction is distinctly concentrated within ischemic stroke events. This specific reduction in ischemia may reflect complementary effects between the vascular and antiatherosclerotic actions of GLP-1RAs and the hemodynamic, blood-pressure-lowering, cardiorenal, and possible antiarrhythmic effects of SGLT2is [32–34]. The most compelling nuance in our cerebrovascular findings, however, is the apparent modification of this treatment affected by age. The stroke risk reduction appeared stronger in adults aged 65 years and older, yielding a 50% relative risk reduction that was absent in younger cohorts. Older adults inherently carry a higher baseline burden of advanced arterial stiffness, chronic kidney disease, and occult atrial arrhythmias, pathophysiological domains that may increase the absolute potential for benefit from combined vascular and cardiorenal risk reduction [35, 36]. Consequently, in this older demographic, the dual-pathway approach may address the broader spectrum of cerebrovascular risk, offering a biologically plausible explanation for the stronger ischemic stroke association observed in our analysis. However, this age-related signal should be interpreted with caution. Although the test for subgroup difference was statistically significant, it was based on a limited number of comparisons, and a chance finding cannot be excluded until these results are confirmed in dedicated prospective trials.

The divergent impact of combination therapy on MI and stroke aligns with the complementary, nonredundant pathophysiological targets of these two drug classes. MI is predominantly an atherothrombotic event, making it plausibly responsive to GLP-1RA-mediated effects on atherosclerotic cardiovascular risk pathways, including inflammation, endothelial function, body weight, blood pressure, and plaque biology [2, 24]. Because this atherosclerotic driver may be more directly targeted by GLP-1RA therapy, the subsequent addition of an SGLT2i, whose benefits are more consistently hemodynamic, heart failure-related, and cardiorenal rather than primarily antiatherosclerotic, may provide only limited additional defense against coronary ischemia [26, 37]. This interpretation should nonetheless be tempered by emerging evidence that SGLT2 inhibition exerts direct antiinflammatory and antioxidative actions on the atherosclerotic plaque, reducing plaque thickness and macrophage infiltration while enhancing stability, and by the recognition that SGLT2 is expressed at the cardiomyocyte level in human hearts, where it appears to modulate cardiac function partly through JunD-dependent pathways in individuals with and without diabetes [38–40]. Conversely, the additive protection we observed against stroke may reflect a dual-pathway synergy. Ischemic stroke etiology is heterogeneous, driven by both large-vessel atherosclerosis and cardioembolic abnormalities [3]. While the GLP-1RA continues to stabilize the cerebrovascular wall, the SGLT2i could provide critical hemodynamic offloading. Advanced echocardiographic and vascular studies demonstrate that combining these agents has been associated with improvement in endothelial glycocalyx, pulse wave velocity, central systolic blood pressure, myocardial work indices, and left atrial strain [41, 42]. By enhancing these parameters of atrial and vascular function, dual therapy may extend protection beyond plaque stabilization toward atrial–hemodynamic and vascular remodeling pathways relevant to risk of ischemic stroke [3, 41, 42]. This comprehensive coverage of both the vascular structure and the hemodynamics provides a plausible rationale for the additive protection for stroke observed in our analysis. In keeping with this atrial–hemodynamic axis, combination therapy has been linked to reduced atrial fibrillation recurrence following catheter ablation and to improved cardiac resynchronization therapy response and clinical outcomes, further supporting a broad, complementary cardiovascular reach [43, 44]. However, these physiological concepts must be interpreted cautiously pending confirmation from dedicated randomized trials.

While the cardiovascular benefits of dual therapy are compelling, they must be contextualized within the regimen’s safety profile. Current evidence does not indicate a novel or synergistic toxicity from combining SGLT2is and GLP-1RAs; instead, adverse events remain largely class-specific [45, 46]. SGLT2is carry established risks of genital mycotic infections, volume depletion, and rare euglycemic ketoacidosis, whereas GLP-1RAs predominantly cause gastrointestinal intolerance. However, the cumulative burden of these side effects, alongside an increased risk of hypoglycemia when background insulin or insulin secretagogues are not appropriately adjusted, warrants careful clinical monitoring [47, 48]. This specific risk of hypoglycemia, combined with the premise that insulin usage frequently serves as a marker of more advanced T2DM, greater treatment complexity, and heightened baseline vascular risk, provides critical context for our subgroup finding that stroke risk appeared elevated (HR 2.33) among insulin users at baseline [47, 48]. Because this estimate stems from a single comparison, we cannot definitively distinguish a harmful pharmacological interaction from confounding by indication. Beyond hemodynamic parameters, the cardiometabolic environment targeted by these agents is profoundly influenced by adipose-driven inflammation and acute metabolic stress. For instance, neurohormonal signaling associated with obesity has been implicated in the pathogenesis of diastolic dysfunction in T2DM, while acute hyperglycemia and systemic inflammation are recognized contributors to atypical cardiomyopathies, including Takotsubo syndrome [49, 50]. Furthermore, regional expression of SGLT2 within adipose tissue has been associated with inflammatory cascades and sirtuin downregulation, ultimately impacting cardiovascular performance [51]. Ultimately, the observed cerebrovascular and coronary benefits must be carefully weighed against individual patient tolerability, renal function, and overall treatment burden.

Strengths and Limitations

A major strength of our study is the disaggregation of MACE into distinct MI and stroke end points. Because composite outcomes often obscure divergent effects across biologically distinct vascular beds, analyzing these events separately provides a more precise assessment of the potential additive benefits of dual-pathway therapy [9]. We prioritized hazard ratios from the most fully adjusted models, integrated randomized and propensity-score-matched real-world evidence [52], performed rigorous subgroup interaction testing, and utilized the formal GRADE framework to quantify certainty of evidence [12, 53]. These findings warrant cautious interpretation. First, only one randomized comparison was available and the evidence base is otherwise predominantly observational, exposing the pooled estimates to residual confounding, immortal-time bias, and treatment-selection bias [10, 54]. Since patients receiving combination therapy may systematically differ from those on monotherapy regarding unmeasured factors such as disease severity, adherence, and healthcare engagement, residual confounding cannot be excluded; thus, these associations do not establish causality. Supporting this, nonmatched cohorts yielded a larger ^****MI association than propensity-matched comparisons (HR 0.67 versus 0.87; p = 0.01).

Second, substantial heterogeneity characterized both end points. For MI, study design was a primary driver, as evidenced by the discrepancy between matched and nonmatched cohorts. The evidence base spans three types of study designs: nonmatched observational cohorts, propensity-matched cohorts, and randomized trials (the FLOW and SOUL analyses). These designs differ significantly in how they select populations, track outcomes, and control for confounding variables. This methodological diversity likely explains the inconsistent results observed across the studies. For stroke, heterogeneity was confined to the SGLT2i-comparator subgroup and was absent within the GLP-1RA-comparator subgroup. Other likely contributors include variations in baseline risk, comorbidity, background therapy, follow up duration, and stroke definitions. Consequently, these pooled estimates represent approximate central tendencies, and this inconsistency necessitated a GRADE downgrade, resulting in very low certainty of evidence [12]. Third, several subgroup analyses, particularly those stratified by insulin use, ischemic heart disease, and hyperlipidemia, rested on as few as one comparison, reducing power and increasing the risk of chance findings; the insulin-use stroke signal, from a single comparison, is especially vulnerable to confounding by indication. Given the wide confidence intervals across these analyses, all subgroup results (including the significant age difference for stroke) should be considered as exploratory and hypothesis-generating rather than definitive conclusions [53]. Finally, requiring separate data for MI and stroke excluded studies that reported only composite MACE outcomes, reducing our available evidence base. This could theoretically introduce selection bias, but only if the decision to report these specific end points was driven by the observed results. This is unlikely, as component-end-point reporting is generally prespecified in the study design. Furthermore, our Egger’s regression tests were nonsignificant for both end points, offering additional reassurance. While the small number of comparisons means we cannot entirely rule out selection bias, its overall impact is likely modest. Reliance on aggregate data also precluded absolute risk estimates and evaluation of drug-, dose-, or sequencing-specific effects [55].

Clinical Implications and Future Directions

Our findings are consistent with current American Diabetes Association (ADA) guidelines, which recommend tailoring diabetes treatments to a patient’s specific cardiovascular risks rather than focusing solely on glucose control [4, 47]. While the cardioprotective benefits of individual SGLT2i and GLP-1RA therapies are well established, the very low certainty of our current findings precludes their use as clinical guidance. At most, these results generate hypotheses regarding how combination therapy might eventually be tailored to a patient’s specific vascular risk profile. For example, it remains entirely speculative whether patients primarily at risk for coronary atherothrombosis derive specific value from optimized GLP-1RA therapy, or if older adults at high risk for ischemic stroke would benefit from earlier combination therapy. These clinical concepts require rigorous confirmation in dedicated, prospective trials before they can inform routine practice. Importantly, our findings should not be interpreted as sufficient evidence to mandate dual therapy solely for the primary prevention of MI or stroke. Instead, potential reductions in distinct vascular events should be viewed as an additional rationale for combination therapy when both agents are already clinically appropriate for managing the patient’s broader cardiometabolic phenotype, including weight, glycemic control, and heart failure risk [4, 47].

However, to transition these insights into definitive prescribing guidelines, the field requires dedicated, prospectively powered randomized controlled trials. Future research must move beyond posthoc analyses of composite MACE end points [9]. Specifically, trials or rigorously designed target-trial emulations using high-quality real-world data should be explicitly designed to evaluate combination therapy against active monotherapy comparators [56, 57]. These studies should be powered for individual coronary and cerebrovascular events, capturing critical granularity such as stroke subtypes, fatal versus nonfatal events, and recurrent events, and structured to clearly compare the efficacy of simultaneous initiation versus sequential add-on therapy. Furthermore, future studies should stratify outcomes by baseline insulin use to clarify whether the adverse cerebrovascular signal observed in our isolated subgroup represents a true pharmacological interaction or mere confounding by disease severity [53]. Finally, future research must also integrate safety, tolerability, adherence, cost, and access considerations, as the ultimate clinical value of dual-pathway intensification depends heavily on long-term feasibility and careful patient selection in routine practice [4, 47]. Notably, none of the included studies enrolled heart transplant recipients, despite growing clinical interest in this population. Both SGLT2is and GLP-1RAs are increasingly used to manage posttransplant diabetes mellitus and obesity. Early evidence suggests these agents may be safe and metabolically beneficial, with potential relevance for mitigating cardiac allograft vasculopathy, an accelerated, diffuse coronary process that remains a leading cause of long-term graft failure [58, 59]. However, available data are limited. Furthermore, transplant-specific concerns, including drug–drug interactions with immunosuppressants, delicate volume status, and heightened infection risk, require dedicated clinical evaluation. Therefore, our findings cannot be extrapolated to transplant recipients, and prospective studies in this highly specific population are needed.

Conclusions

This systematic review and meta-analysis found that combination therapy with an SGLT2i and a GLP-1RA is associated with a lower overall risk of MI and stroke compared with monotherapy in adults with T2DM. Importantly, the incremental benefit of this dual-pathway intensification is not uniform across vascular beds. The protective effect against MI appears predominantly mediated by the GLP-1RA component, whereas the reduction in stroke may reflect additive vascular, hemodynamic, and cardiorenal effects that appeared stronger in older adults. While these findings offer a physiologically plausible rationale for considering distinct cardiovascular events separately in future research, the substantial heterogeneity and predominantly observational nature of the current evidence base yield a very low overall certainty of evidence. Consequently, these results should be interpreted as hypothesis-generating insights rather than definitive proof of causal benefit. Ultimately, establishing the true clinical utility and optimal sequencing of combined incretin and cardiorenal therapies will require confirmation from dedicated, rigorously designed prospective studies.

Supplementary Information

Below is the link to the electronic supplementary material.

Declarations

Author Contribution

Mohammad E. Khamseh contributed to the study conceptualization, supervision, and discrepancy resolution. Roya Ghafoury contributed to the study design and protocol development, literature search, and original draft preparation. Roya Ghafoury and Mobin Naghshbandi were both involved in study screening and selection, data extraction, risk of bias evaluation, and GRADE certainty of evidence assessment. Mobin Naghshbandi additionally performed the statistical analysis, meta-analysis, and software visualization. Faramarz Ismail-Beigi, Mojtaba Malek, Sanjay Kalra, and Mohammad E. Khamseh contributed to intellectual oversight and clinical data interpretation. All authors critically reviewed and edited the manuscript, approved the final version, and agree to be accountable for all aspects of the work.

Funding

No funding or sponsorship was received for this study or publication of this article.

Medical Writing/Editorial Assistance

ChatGPT (OpenAI, San Francisco, CA, USA) was used to assist with grammar and language editing during the preparation of this manuscript. No funding was received for this assistance. The authors reviewed and verified all content and take full responsibility for the integrity and accuracy of the work.

Data Availability

This study is based on aggregate data from previously published studies; no new primary data were generated. All data supporting the findings are contained within the article and its tables. The extracted data set and the analysis code are available from the corresponding author upon reasonable request.

Conflict of Interest

Roya Ghafoury has nothing to disclose. Mobin Naghshbandi has nothing to disclose. Faramarz Ismail-Beigi has nothing to disclose. Mojtaba Malek has nothing to disclose. Mohammad E. Khamseh and Sanjay Kalra have nothing to disclose other than the following: Mohammad E. Khamseh and Sanjay Kalra are Editorial Board members of Diabetes Therapy. Mohammad E. Khamseh and Sanjay Kalra were not involved in the selection of peer reviewers for the manuscript nor any of the subsequent editorial decisions.

Ethical Approval

This article is based on previously conducted studies and does not contain any new studies with human participants or animals performed by any of the authors.