Comparative cardiovascular efficacy of antidiabetic therapies in type 2 diabetes: a systematic review and network meta-analysis of randomized trials Authors: Donya Mohammadi, Behnam Sadeghirad, Mohammad Aziz Rasouli, Lotfollah Saed, Hamid Reza Baradaran, Yousef Moradi Published: 2026 ABSTRACT BackgroundCardiovascular diseases (CVDs) remain the leading cause of morbidity and mortality among individuals with type 2 diabetes mellitus (T2DM), imposing a substantial burden on healthcare systems. This systematic review and network meta-analysis evaluated the comparative efficacy of antidiabetic medications and combination therapies in reducing cardiovascular outcomes in patients with T2DM.MethodsPubMed (Medline), Embase, Web of Science, Scopus, and the Cochrane Central Register of Controlled Trials (CENTRAL) were searched from inception to December 30, 2024, for randomized controlled trials (RCTs) assessing cardiovascular outcomes in adults with T2DM treated with metformin, sulfonylureas, thiazolidinediones (TZDs), dipeptidyl peptidase‑4 (DPP‑4) inhibitors, glucagon‑like peptide‑1 (GLP‑1) receptor agonists, sodium–glucose cotransporter‑2 (SGLT2) inhibitors, or insulin. Outcomes included cardiovascular mortality, myocardial infarction (MI), stroke, heart failure, hospitalization for cardiovascular events, and unstable angina. Risk of bias was assessed using the Cochrane RoB 2.0 tool, and analyses were conducted with a random-effects model in Stata (version 18).ResultsFrom 10,514 records, 133 RCTs involving 289,558 participants (mean age: 64.7 years) were included. Glucagon-like peptide-1 receptor agonists (GLP-1 RAs) significantly reduced cardiovascular mortality (RR = 0.85, 95% CI: 0.75–0.98; high-certainty evidence) and stroke (RR = 0.83, 95% CI: 0.74–0.93; high-certainty evidence). SGLT2 inhibitors significantly reduced heart failure (RR = 0.64, 95% CI: 0.53–0.77; high-certainty evidence) and hospitalization for cardiovascular events (RR = 0.72, 95% CI: 0.68–0.77; high-certainty evidence). An indirect comparison suggested lower cardiovascular mortality with DPP-4 inhibitors versus GLP-1 receptor agonists, although this finding should be interpreted cautiously due to network imbalance.ConclusionsGLP-1 RAs and SGLT2 inhibitors are associated with significant reductions in specific cardiovascular outcomes in patients with T2DM, particularly in reducing cardiovascular mortality, stroke, heart failure, and hospitalization. Treatment decisions should integrate patient-specific risk profiles and cost-effectiveness to optimize outcomes.Supplementary InformationThe online version contains supplementary material available at https://doi.org/10.1186/s13098-026-02226-9. Background Type 2 diabetes mellitus (T2DM) is a chronic metabolic disorder characterized by insulin resistance and/or insufficient insulin production due to a combination of genetic and environmental factors. It is a leading cause of mortality worldwide and significantly increases the risk of cardiovascular diseases (CVDs), including myocardial infarction (MI) and stroke [1, 2]. Therefore, prevention of cardiovascular complications is considered important for glycemic control in the management of T2DM [3]. A wide range of pharmacological interventions have been developed and evaluated for their efficacy in controlling hyperglycemia and reducing diabetes-related complications. Current clinical guidelines recommend glucose-lowering therapies not only for glycemic management but also for reducing cardiovascular risk in individuals with T2DM, particularly those with established atherosclerotic cardiovascular disease (ASCVD) or those at high cardiovascular risk [3–7]. Metformin remains the first-line therapy due to its effectiveness, low cost, low risk of hypoglycemia, and potential cardiovascular benefits [8, 9]. Sulfonylureas are commonly used as second-line agents after metformin failure; however, they have been associated with an increased risk of MI and hypoglycemia [10–12]. Thiazolidinediones (TZDs) improve insulin sensitivity and β-cell function and may confer cardiovascular protection [13–15]. Dipeptidyl peptidase-4 (DPP-4) inhibitors increase insulin secretion and exert anti-inflammatory effects that may contribute to cardiovascular safety [16, 17]. More recently, glucagon-like peptide-1 receptor agonists (GLP-1) and sodium‒glucose cotransporter-2 (SGLT2) inhibitors have been shown to significantly reduce both primary and secondary cardiovascular events, particularly in high-risk patients [18, 19]. Despite its potential for hypoglycemia, insulin remains a vital option for achieving glycemic targets in certain populations [20]. Despite the growing body of evidence on the cardiovascular benefits of certain antidiabetic agents, the lack of direct head‒to-head comparisons between all available therapies complicates clinical decision-making [21, 22]. Previous studies share limitations that complicate direct clinical translation and necessitate updated syntheses. A limitation of previous network meta-analyses is the scarcity of direct head‒to‒head trial evidence for certain drug classes for which a small randomized control trials (RCTs) was available [23]. Despite the growing evidence base, important gaps remain. Most studies have focused on single agents, with limited exploration of dual SGLT2 inhibitor/GLP‑1 receptor agonist therapy, leaving questions about combined benefits unanswered [24]. Moreover, the reliance on observational data and relatively short follow‑up periods (< 5 years) means that long‑term safety and outcomes beyond cardiovascular events are still poorly understood [25]. Guidelines continue to evolve, yet current analyses often fail to capture the full spectrum of available therapies. The tendency for publication bias, particularly favoring positive results with SGLT2 inhibitors and GLP‑1, may further inflate perceived benefits, underscoring the need for more balanced and comprehensive evaluations [26, 27]. This study aims to apply this advanced approach to comprehensively assess and compare the cardiovascular outcomes of various antidiabetic drug classes in patients with type 2 diabetes, thereby generating evidence-based insights to inform optimal therapeutic strategies. Methods Study design and registration This study was conducted as a systematic review and network meta-analysis of RCTs to assess the comparative cardiovascular efficacy of antidiabetic medications in patients with T2DM. The protocol was registered a priori in the PROSPERO international prospective register of systematic reviews (CRD420251061892). The review adhered to the Preferred Reporting Items for Systematic Reviews and Meta-Analyses for Network Meta-Analyses (PRISMA-NMA) guidelines [28, 29]. Eligibility criteria Studies were considered eligible for inclusion if they met the following criteria: RCTs enrolling adult participants (aged ≥ 18 years) with a confirmed diagnosis of T2DM. Eligible studies were required to investigate the effects of one or more pharmacological antidiabetic interventions, specifically metformin, sulfonylureas, TZDs, DPP-4 inhibitors, glucagon-like peptide-1 receptor agonists (GLP-1 Ras), SGLT2 inhibitors, or insulin, on cardiovascular outcomes. Acceptable comparator groups included placebo, standard care, or other active antidiabetic drug classes. To ensure adequate assessment of cardiovascular endpoints, studies were required to have a minimum follow-up duration of six months and to report at least one prespecified cardiovascular outcome, including all-cause mortality, cardiovascular mortality, heart failure, nonfatal MI, nonfatal stroke, hospitalization, and unstable angina. “Hospitalization for cardiovascular events” was defined as a composite outcome that included hospitalization due to heart failure, acute coronary syndromes, or other cardiovascular causes, as reported in the original trials. Given variability in definitions across studies, this outcome was not fully standardized and may reflect some degree of clinical heterogeneity. Only full-text articles published in peer-reviewed journals and in English were included (Supplementary Tables, Table S1). The exclusion criteria included quasi-randomized trials, conference abstracts, review articles, case reports, and studies that did not report cardiovascular outcome data or effect estimates. Duplicate publications and those with overlapping populations were carefully reviewed, and the most comprehensive or recent dataset was retained (Supplementary Tables, Table S1). Search strategy and data sources A comprehensive literature search was systematically conducted to identify relevant RCTs published from inception to December 30, 2024. The following electronic databases were searched: PubMed (Medline), Embase, Web of Science, Scopus, and the Cochrane Central Register of Controlled Trials (CENTRAL). The search strategy was developed in collaboration with an experienced medical librarian and included a combination of MeSH terms and free-text keywords such as: “Type 2 Diabetes Mellitus”, “cardiovascular diseases”, “glucagon-like peptide-1 receptor agonists”, “SGLT2 Inhibitors”, “DPP-4 Inhibitors”, “Metformin”, “Sulfonylureas”, “Thiazolidinediones”, and “Insulin”. Search filters were applied to limit the results to RCTs involving adult human participants. Only studies published in English were included in the final analysis. The reference lists of the included articles and relevant systematic reviews were manually searched to identify additional eligible studies (Supplementary Tables, Table S1). Data extraction and management Using a standardized data extraction form, two independent reviewers (DM and MAR) extracted the following information from each eligible study: first author, publication year, geographic region, study design, sample size, mean age of participants, sex distribution, intervention and comparator arms, length of follow-up, cardiovascular outcomes assessed, effect estimates (risk ratios [RRs] with 95% confidence intervals [CIs]), and funding sources. When effect sizes were not directly reported, they were calculated from available raw data. Where necessary, the corresponding authors were contacted to obtain missing or unpublished information. Disagreements were resolved through discussion or third-party arbitration (YM). Risk of bias assessment The risk of bias in the included RCTs was evaluated via the Cochrane Risk of Bias 2.0 tool, which examines five key domains: the randomization process, deviations from intended interventions, missing outcome data, outcome measurement, and selection of reported results. Each study was categorized as having a low risk of bias, some concerns, or a high risk of bias. Two reviewers (DM and MAR) assessed the risk of bias independently and in duplicate via a modified Cochrane Risk of Bias 2.0 tool, and disagreements were resolved by the adjudication of a third reviewer (YM) [30, 31]. Statistical analysis A random-effects network meta-analysis was conducted via Stata version 18.0 (StataCorp, College Station, TX, USA) [32, 33]. Dichotomous outcomes were synthesized as RRs with corresponding 95% CIs. Network geometry plots were generated to illustrate the structure and connectivity of treatment comparisons. Authors quantified between‑study heterogeneity using the I² statistic and evaluated global inconsistency with the design‑by‑treatment interaction model. Transitivity was assessed by comparing distributions of key effect modifiers (e.g., baseline cardiovascular risk, follow-up duration) across treatment comparisons. In particular, authors evaluated whether baseline cardiovascular risk, duration of diabetes, and follow-up time were similarly distributed across comparisons involving GLP-1 receptor agonists and DPP-4 inhibitors, given that GLP-1 receptor agonist trials often enrolled higher-risk populations. Global inconsistency was evaluated via the design-by-treatment interaction model, whereas local inconsistency was assessed via loop-specific and node-splitting approaches [34, 35]. Treatments were ranked using surface under the cumulative ranking curve (SUCRA) values. Heterogeneity was quantified via the I² statistic and visually inspected via forest plots; values exceeding 50% were considered indicative of substantial heterogeneity [36]. Potential publication bias was explored via funnel plots and Egger’s test for outcomes informed by ten or more studies [37, 38]. All-cause mortality and cardiovascular mortality were analyzed as distinct outcomes within separate network meta-analysis models. Treatment rankings were generated independently for each outcome and were not combined within a single network structure. To assess the robustness of mortality outcomes, we additionally performed sensitivity analyses restricted to (i) cardiovascular outcome trials (CVOTs) and (ii) trials with ≥ 12 months follow‑up. Assessing the certainty of the evidence The certainty of evidence for each network estimate was assessed using the GRADE (Grading of Recommendations Assessment, Development and Evaluation) framework. For direct comparisons, risk of bias, inconsistency, directness, and publication bias were evaluated. For indirect comparisons, certainty was determined by the lowest rating among contributing direct comparisons, with dominant first-order loops prioritized when available; otherwise, higher-order loops were considered. In cases of intransitivity, such as differences in patient characteristics, interventions, or comparators, certainty was rated down. For overall network estimates, the higher rating between direct and indirect evidence was retained unless incoherence or imprecision required further downgrading [39, 40]. To operationalize the GRADE process, the CINeMA (Confidence in Network Meta-Analysis) tool was employed. CINeMA is a web-based platform aligned with the GRADE framework that evaluates evidence across six domains: within-study bias, reporting bias, indirectness, heterogeneity, imprecision, and inconsistency. By providing a user-friendly interface and automated reporting, CINeMA facilitates transparent and standardized assessment of confidence in network meta-analysis results. Summary of results using the GRADE approach To enhance the interpretability of the NMA findings, a minimally contextualized GRADE approach was applied. Interventions were ranked from most to least effective based on effect estimates and the certainty of evidence [41–43]. For each outcome, interventions were classified into three categories: Worst [placebo and interventions showing no difference from placebo (95% CI includes the null].Intermediate [interventions superior to placebo but not superior to other active agents].Best [interventions superior to at least one intermediate intervention]. Worst [placebo and interventions showing no difference from placebo (95% CI includes the null]. Worst [placebo and interventions showing no difference from placebo (95% CI includes the null]. Intermediate [interventions superior to placebo but not superior to other active agents]. Intermediate [interventions superior to placebo but not superior to other active agents]. Best [interventions superior to at least one intermediate intervention]. Best [interventions superior to at least one intermediate intervention]. Each category was further stratified by certainty of evidence relative to placebo, distinguishing between moderate/high-certainty and low/very low-certainty groups. Results Description of the evidence The total number of articles related to antidiabetic drugs in the PubMed (Medline), Embase, Web of Science, Scopus, and CENTRAL databases was 10,514, comprising 2,795 articles in PubMed, 2,327 articles in Scopus, 4,793 articles in Web of Science, and 599 articles in CENTRAL. For each database, a separate library was created in EndNote software, and the retrieved articles from each database were transferred to their respective libraries. To identify and remove duplicate studies and prepare for initial screening, all the articles from the libraries of the four databases were subsequently consolidated into a new library in EndNote. Ultimately, after merging and removing duplicates, 6,520 articles were selected for screening on the basis of title and abstract. Following the evaluation of the articles on the basis of title and abstract, 1,388 studies were chosen for full-text review. These articles were assessed according to the inclusion and exclusion criteria, and 133 studies related to antidiabetic drugs [44–176] met the criteria for inclusion in the meta-analysis (Fig. 1). Characteristics and quality assessment of the included studies In total, 133 studies including 289,558 participants were included in the analysis [44–176]. The mean age of the participants was 64.7 years, and 47.4% were women. The mean follow-up duration of the studies was 1.64 years. Overall, 45.7% of patients had a prior history of cardiovascular disease (Supplementary Tables, Table S2). All studies were quality-assessed via the Cochrane RoB 2 tool. Except for two studies (CONFIDENCE and Abdul-Ghani, 2017), which had some risk of bias, the remaining studies were rated as having a low risk of bias (Supplementary Figures, Figure S1). The direct and indirect effects of various antidiabetic drugs on different cardiovascular outcomes are presented in the Supplementary Tables, Table S3. Network consistency Network consistency results for all outcomes are presented in Fig. 2. Analyses were based on: 105 trials (198,360 participants) for all‑cause mortality; 64 trials (240,711 participants) for cardiovascular mortality; 34 trials (63,785 participants) for heart failure; 29 trials (172,617 participants) for hospitalization due to cardiovascular events; 45 trials (208,593 participants) for MI; 45 trials (210,574 participants) for stroke; and 15 trials (23,923 participants) for unstable angina. Across all treatment networks, no evidence of inconsistency or heterogeneity was observed (Fig. 2). For major outcomes, global heterogeneity levels were low to moderate (I² range: >25% for most outcomes). Design‑by‑treatment interaction models did not indicate statistically significant inconsistency (p > 0.05 for all outcomes). Cardiovascular outcomes All-cause mortality On the basis of the GRADE framework, interventions were classified according to their relative effectiveness and certainty of evidence across key cardiovascular outcomes. For all‑cause mortality, although no statistically significant differences were detected, the ranking analyses indicated that metformin and GLP‑1 receptor agonists may represent comparatively more favorable treatment options (Table 2). In the network meta-analysis of 105 studies including 198,360 participants (127,972 receiving intervention, 5,050 experiencing the outcome), no statistically significant differences were observed in all-cause mortality across antidiabetic drug comparisons. Several treatment combinations, including metformin + insulin (RR = 0.15, 95% CI: 0.00–8.43), GLP-1 RAs + insulin (RR = 0.18, 95% CI: 0.00–9.15), and thiazolidinedione + insulin (RR = 0.19, 95% CI: 0.00–9.53), yielded extremely wide confidence intervals, reflecting sparse data and high imprecision. These estimates should be considered exploratory and unreliable, and they should not be interpreted as evidence of benefit (Table 1a, Fig. 2a). GLP-1 RAs potentially reduced mortality compared with DPP-4 inhibitors (RR = 0.84, 95% CI: 0.51–1.30) and placebo (RR = 0.87, 95% CI: 0.37–1.78), although differences were not statistically significant (p > 0.05). SGLT2 inhibitors and other combinations (e.g., DPP-4 inhibitors + TZD) showed variable effects, but most comparisons were limited by lack of direct evidence or wide confidence intervals, reinforcing the need for cautious interpretation. Table 1Network estimated RRs (95% CIs) of Pharmacological Interventions to Reduce Cardiovascular Outcomes in Patients with T2DM T2DM [T2DM [Type 2 Diabetes Mellitus], TZDs [Thiazolidinediones], DPP-4 inhibitors [Dipeptidyl Peptidase‑4 Inhibitors], GLP‑1 RA/GLP‑1A – [Glucagon‑Like Peptide‑1 Receptor Agonists], SGLT2i [Sodium–Glucose Cotransporter‑2 Inhibitors]]. Table 1a: Estimates with extremely wide confidence intervals are based on sparse data. These are exploratory and were not used for overall treatment rankings Table 1Network estimated RRs (95% CIs) of Pharmacological Interventions to Reduce Cardiovascular Outcomes in Patients with T2DM T2DM [T2DM [Type 2 Diabetes Mellitus], TZDs [Thiazolidinediones], DPP-4 inhibitors [Dipeptidyl Peptidase‑4 Inhibitors], GLP‑1 RA/GLP‑1A – [Glucagon‑Like Peptide‑1 Receptor Agonists], SGLT2i [Sodium–Glucose Cotransporter‑2 Inhibitors]]. Table 1a: Estimates with extremely wide confidence intervals are based on sparse data. These are exploratory and were not used for overall treatment rankings Network estimated RRs (95% CIs) of Pharmacological Interventions to Reduce Cardiovascular Outcomes in Patients with T2DM T2DM [T2DM [Type 2 Diabetes Mellitus], TZDs [Thiazolidinediones], DPP-4 inhibitors [Dipeptidyl Peptidase‑4 Inhibitors], GLP‑1 RA/GLP‑1A – [Glucagon‑Like Peptide‑1 Receptor Agonists], SGLT2i [Sodium–Glucose Cotransporter‑2 Inhibitors]]. Table 1a: Estimates with extremely wide confidence intervals are based on sparse data. These are exploratory and were not used for overall treatment rankings Cardiovascular mortality On the basis of the GRADE framework, in reducing cardiovascular death, GLP-1 RAs demonstrated significant benefit and high certainty, whereas insulin again showed limited efficacy (Table 2). In the network meta-analysis for the outcome of cardiovascular mortality, which was based on 64 studies with a total population of 240,711 participants, the results indicated varying effects of antidiabetic treatments. The network geometry (Fig. 2b) illustrates the direct and indirect comparisons among all the treatments. The pairwise meta-analysis RRs and 95% CIs are detailed in Table 1b. An indirect network estimate suggested that DPP-4 inhibitors were associated with a lower risk of cardiovascular mortality compared with GLP-1 receptor agonists (RR = 0.85; 95% CI: 0.75–0.98). This finding is based solely on indirect evidence, as no head-to-head CVOTs directly compare these two drug classes. Node-splitting analysis did not demonstrate statistically significant inconsistency; however, this comparison should be interpreted with caution due to potential violations of the transitivity assumption, including differences in baseline cardiovascular risk across contributing trials. In sensitivity analyses limited to CVOTs and to trials with ≥ 12 months follow-up, effect estimates for all-cause mortality and cardiovascular mortality were directionally consistent with the main analysis, although confidence intervals were wider as expected due to fewer contributing trials. Heart failure Based on the GRADE framework, SGLT2 inhibitors emerged as the most effective intervention for reducing heart failure risk (RR = 0.64; 95% CI: 0.53–0.77), supported by high-certainty evidence (Table 2). In the network meta-analysis of heart failure, which included 34 studies and 63,785 participants, varying effects of antidiabetic treatments were observed. The network geometry (Fig. 2c) illustrates the direct and indirect comparisons among all treatments. SGLT2 inhibitors: RR = 0.64 (95% CI: 0.53–0.77), high-certainty evidenceCombination therapy with SGLT2 inhibitors + DPP-4 inhibitors: RR = 0.64 (95% CI: 0.53–0.77)Combination therapy with GLP-1 RAs + insulin: RRs ranging 0.45–0.71, consistently narrow CIsTZDs: RR = 0.41 (95% CI: 0.20–0.85)Sulfonylureas: RR = 0.40 (95% CI: 0.19–0.84) SGLT2 inhibitors: RR = 0.64 (95% CI: 0.53–0.77), high-certainty evidence SGLT2 inhibitors: RR = 0.64 (95% CI: 0.53–0.77), high-certainty evidence Combination therapy with SGLT2 inhibitors + DPP-4 inhibitors: RR = 0.64 (95% CI: 0.53–0.77) Combination therapy with SGLT2 inhibitors + DPP-4 inhibitors: RR = 0.64 (95% CI: 0.53–0.77) Combination therapy with GLP-1 RAs + insulin: RRs ranging 0.45–0.71, consistently narrow CIs Combination therapy with GLP-1 RAs + insulin: RRs ranging 0.45–0.71, consistently narrow CIs TZDs: RR = 0.41 (95% CI: 0.20–0.85) TZDs: RR = 0.41 (95% CI: 0.20–0.85) Sulfonylureas: RR = 0.40 (95% CI: 0.19–0.84) Sulfonylureas: RR = 0.40 (95% CI: 0.19–0.84) GRADE results. High certainty supports SGLT2 inhibitors as the most effective therapy for heart failure prevention (Table 2). Although TZDs (RR = 0.41; 95% CI: 0.20–0.85) and sulfonylureas (RR = 0.40; 95% CI: 0.19–0.84) appeared to show protective effects, these estimates were primarily derived from indirect comparisons with limited data and should be interpreted with caution. These findings are inconsistent with established clinical evidence and may reflect sparse data, variability in outcome definitions, or potential reporting biases. Table 2GRADE-Based Classification of Cardiovascular Outcomes Across Antidiabetic Interventions: Relative Effects, Rankings, and Certainty of EvidenceTZDs [Thiazolidinediones], DPP-4 inhibitors [Dipeptidyl Peptidase‑4 Inhibitors], GLP‑1 RA/GLP‑1A – [Glucagon‑Like Peptide‑1 Receptor Agonists], SGLT2 inhibitors [Sodium–Glucose Cotransporter‑2 Inhibitors] Table 2GRADE-Based Classification of Cardiovascular Outcomes Across Antidiabetic Interventions: Relative Effects, Rankings, and Certainty of EvidenceTZDs [Thiazolidinediones], DPP-4 inhibitors [Dipeptidyl Peptidase‑4 Inhibitors], GLP‑1 RA/GLP‑1A – [Glucagon‑Like Peptide‑1 Receptor Agonists], SGLT2 inhibitors [Sodium–Glucose Cotransporter‑2 Inhibitors] GRADE-Based Classification of Cardiovascular Outcomes Across Antidiabetic Interventions: Relative Effects, Rankings, and Certainty of Evidence TZDs [Thiazolidinediones], DPP-4 inhibitors [Dipeptidyl Peptidase‑4 Inhibitors], GLP‑1 RA/GLP‑1A – [Glucagon‑Like Peptide‑1 Receptor Agonists], SGLT2 inhibitors [Sodium–Glucose Cotransporter‑2 Inhibitors] Hospitalization for cardiovascular events Based on the GRADE framework, both SGLT2 inhibitors (alone or combined with DPP-4 inhibitors) were among the most effective interventions for reducing hospitalization due to cardiovascular events (Table 2). In the network meta-analysis of hospitalization outcomes, which included 29 studies and 172,617 participants, varying effects of antidiabetic treatments were observed. The network geometry (Fig. 2d and Table 1d) illustrates the direct and indirect comparisons among all treatments. GLP-1 RAs: RR = 0.72 (95% CI: 0.68–0.77)Combination therapy with sulfonylurea + insulin: RR = 0.68 (95% CI: 0.59–0.78) GLP-1 RAs: RR = 0.72 (95% CI: 0.68–0.77) GLP-1 RAs: RR = 0.72 (95% CI: 0.68–0.77) Combination therapy with sulfonylurea + insulin: RR = 0.68 (95% CI: 0.59–0.78) Combination therapy with sulfonylurea + insulin: RR = 0.68 (95% CI: 0.59–0.78) GRADE results: Moderate to high certainty confirms protective effects of SGLT2 inhibitors and GLP-1 RAs against CV-related hospitalizations (Table 2). Myocardial infarction On the basis of the GRADE framework, in the context of MI, GLP-1 had favorable effects with high certainty, whereas TZD was identified as potentially effective but supported by low-certainty evidence (Table 2). In the network meta-analysis for the outcome of MI, on the basis of 45 studies with a total population of 208,593 participants, the results indicated varying effects of antidiabetic treatments. For the outcome of MI, a statistically significant reduction in risk was observed with only DPP-4 inhibitors compared with sulfonylureas (RR = 0.69; 95% CI: 0.48–0.98) (Fig. 2e and Table 1e). Stroke According to the GRADE framework, stroke prevention was most strongly associated with GLP-1 RAs (RR = 0.83; 95% CI: 0.74–0.93), supported by high-certainty evidence (Table 2). In the network meta-analysis of stroke, which included 45 studies and 210,574 participants, varying effects of antidiabetic treatments were observed. The network geometry illustrates the direct and indirect comparisons among all treatments (Fig. 2f and Table 1f). GLP-1 RAs: RR = 0.83 (95% CI: 0.74–0.93), high-certainty evidenceDPP-4 inhibitors: RR = 0.83 (95% CI: 0.74–0.93), protective compared with comparatorsGLP-1 RAs + insulin: RR = 0.73 (95% CI: 0.54–0.98), significant protective effect GLP-1 RAs: RR = 0.83 (95% CI: 0.74–0.93), high-certainty evidence GLP-1 RAs: RR = 0.83 (95% CI: 0.74–0.93), high-certainty evidence DPP-4 inhibitors: RR = 0.83 (95% CI: 0.74–0.93), protective compared with comparators DPP-4 inhibitors: RR = 0.83 (95% CI: 0.74–0.93), protective compared with comparators GLP-1 RAs + insulin: RR = 0.73 (95% CI: 0.54–0.98), significant protective effect GLP-1 RAs + insulin: RR = 0.73 (95% CI: 0.54–0.98), significant protective effect GRADE results: High certainty supports GLP-1 RAs as the most effective agent for stroke prevention (Table 2). Unstable angina On the basis of the GRADE framework, for unstable angina, SGLT2 inhibitors was classified among the most effective interventions with moderate to high certainty, whereas sulfonylurea and TZD were categorized as potentially least effective, both of which are supported by low-certainty evidence (Table 2). In the network meta-analysis for the outcome of unstable angina, which was based on 15 studies with a total population of 23,923 participants, the results indicated varying effects of antidiabetic treatments. For the outcome of unstable angina, no treatment comparisons demonstrated statistically significant differences (Fig. 2g; Table 1 g). Discussion This systematic review and network meta-analysis of 133 RCTs including nearly 290,000 participants provides comprehensive comparative evidence on the cardiovascular effects of antidiabetic drug classes. These findings confirm that GLP-1 RAs were associated with significant reductions in cardiovascular mortality and stroke, while SGLT2 inhibitors demonstrated the greatest benefit for heart failure and hospitalization outcomes. DPP-4 inhibitors showed largely neutral cardiovascular effects. An unexpected signal suggesting lower cardiovascular mortality with DPP-4 inhibitors compared with GLP-1 RAs was observed; however, this finding is derived from indirect comparisons and network estimates, and may be influenced by differences in baseline cardiovascular risk and study population characteristics. This observation is based solely on indirect evidence, as no head-to-head CVOTs directly compare these two drug classes. Differences in baseline cardiovascular risk and trial populations across the network may have influenced this result. It should not be interpreted as evidence of clinical superiority of DPP-4 inhibitors over GLP-1 RAs. This indirect comparison between DPP‑4 inhibitors and GLP‑1 RAs is entirely hypothesis‑generating and should not be interpreted as evidence of comparative superiority, given the absence of head‑to‑head cardiovascular outcome trials, the reliance on disconnected indirect paths, and potential imbalances in baseline cardiovascular risk. These findings are consistent with current clinical guidelines, including those from the American Diabetes Association (ADA) and the European Society of Cardiology/European Association for the Study of Diabetes (ESC/EASD)], which recommend GLP-1 RAs and SGLT2 inhibitors, especially for patients at high risk of cardiovascular disease [18, 177–179]. Compared with previous meta-analyses, this study expands the evidence base by incorporating a larger number of randomized trials, applying a structured GRADE and CINeMA framework, and evaluating a broad range of cardiovascular outcomes across all major drug classes. Previous syntheses often focused narrowly on newer agents or single outcomes, whereas our analysis integrates all major drug classes and outcomes, providing a more comprehensive comparative framework. SGLT2 inhibitors provide strong protection against heart failure through plasma‑volume and renal mechanisms, whereas GLP‑1 RAs reduce stroke and cardiovascular mortality via improvements in vascular inflammation and endothelial function. These advantages contrast with sulfonylureas and insulin, which increase hypoglycemia and weight gain, and with TZDs, which elevate heart‑failure risk despite potential benefits for myocardial infarction. DPP‑4 inhibitors remain cardiovascularly neutral but are well tolerated and represent a safer alternative to sulfonylureas, particularly for older or medically complex patients. Although our network analysis suggested a numerically protective effect of TZDs for heart failure, this finding stems from sparse indirect evidence and contrasts with prior cardiovascular outcome trials (e.g., PROactive) and extensive observational data showing increased heart failure risk with pioglitazone and rosiglitazone. This discrepancy likely reflects (i) the limited number of heart‑failure‑specific events in the contributing RCTs, (ii) variability in outcome adjudication, and (iii) the indirectness of the comparisons. Therefore, TZD‑related estimates in our network should be interpreted very cautiously. Overall, combining GLP‑1 or SGLT2 inhibitors with metformin offers the most favorable cardiovascular profile, though treatment selection must consider tolerability, infection or gastrointestinal risks, and the higher cost of newer agents [9, 82, 114, 154, 176, 178, 180–184]. In a network meta-analysis, DPP-4 inhibitors appeared to reduce cardiovascular mortality compared with GLP-1 RAs (RR 0.85; 95% CI 0.75–0.98). This result is unexpected and conflicts with the findings of major CVOTs, such as LEADER, SUSTAIN-6, and REWIND, as well as current clinical guidelines, all of which consistently emphasize the cardiovascular benefits of GLP-1 RAs [82, 114, 160]. Several methodological factors may explain this discrepancy. First, the comparison relies largely on indirect rather than direct evidence because few head‑to‑head trials exist, and indirect estimates are more susceptible to bias and may not accurately capture true treatment differences. Second, baseline cardiovascular risk varied substantially across trials: participants in DPP‑4 inhibitor studies generally had lower cardiovascular risk than those enrolled in GLP‑1 RA CVOTs, influencing event rates and the resulting comparative estimates. Third, heterogeneity in trial design, follow‑up duration, and outcome definitions likely contributed to inconsistent effect sizes, adding noise to the network and distorting relative comparisons. The evidence base is also imbalanced, with large, rigorous CVOTs available for GLP‑1 RAs but fewer and smaller studies for DPP‑4 inhibitors, a structural asymmetry that can skew indirect comparisons and generate counterintuitive signals. Overall, these findings should be interpreted cautiously: they do not contradict the robust evidence supporting GLP‑1 RAs for cardiovascular risk reduction but instead underscore the need to interpret network meta‑analysis results within the context of direct clinical trial data and established guideline recommendations. Clinical decision-making should account for individual patient factors, including treatment cost, potential side effects (e.g., urinary tract infections with SGLT2 inhibitors or gastrointestinal issues with GLP-1), and patient preferences. Metformin remains recommended as a first-line therapy because of its low cost and safety [9], but adding SGLT2 inhibitors or GLP-1 to patients with high CVD risk can be an effective strategy. This approach aligns with the patient-centered care paradigm, emphasizing a balance between efficacy, safety, and economic accessibility [18]. Nonetheless, despite its comprehensiveness and inclusion of 133 RCTs, this network meta-analysis has several limitations that should be considered when interpreting the results. These include the limited number of direct comparisons for certain drug classes, clinical heterogeneity among participants (e.g., diabetes severity, baseline cardiovascular disease, and comorbidities), and a mean follow-up duration of only 1.5 years, which may be insufficient to capture long-term cardiovascular outcomes. Furthermore, some treatment comparisons, particularly those involving insulin combinations or dual GLP-1 RAs receptor agonist/SGLT2 inhibitor therapy, produced effect estimates with extremely wide or clinically implausible confidence intervals. Additionally, the definition of hospitalization for cardiovascular events varied across trials and often represented a composite endpoint, which may introduce heterogeneity and limit comparability across studies. These results are primarily driven by sparse data and high imprecision and should not be interpreted as evidence of benefit, but rather considered hypothesis-generating. This uncertainty contributes to heterogeneity and may limit the generalizability of the findings. Nevertheless, the assumptions underlying the network meta-analysis, including transitivity and comparability across treatment nodes, were carefully assessed and confirmed by a team of clinical experts and methodological specialists. Future studies with larger sample sizes, more homogeneous study designs, longer follow-up periods, and dedicated trials on combination therapies are needed to generate more reliable and clinically interpretable estimates. Conclusion This network meta-analysis of 133 randomized controlled trials involving 289,558 patients with type 2 diabetes demonstrates that GLP-1 RAs and SGLT2 inhibitors provide the most robust cardioprotective benefits. GLP-1 RAs notably reduced cardiovascular mortality and stroke, while SGLT2 inhibitors were most effective in lowering heart failure and cardiovascular hospitalization. TZDs lowered MI risk but were associated with an increased risk of heart failure. Metformin remains an appropriate first-line therapy due to its safety, affordability, and moderate cardiovascular benefits, whereas sulfonylureas are less favorable due to their association with hypoglycemia and weight gain. However, these findings should be interpreted with caution. The mean follow-up duration of included trials (~ 1.6 years) may be insufficient to fully capture long-term cardiovascular outcomes, particularly mortality. Certain treatment comparisons, especially those involving insulin combinations, were based on sparse data and yielded wide or clinically implausible confidence intervals. Clinical heterogeneity across trials, including differences in baseline cardiovascular disease, diabetes duration, and comorbidities, may also limit the generalizability of these results. Optimal treatment decisions should therefore consider patient-specific factors, including cardiovascular risk profile, side effects, costs, and preferences. Combination therapy with GLP-1 RAs or SGLT2 inhibitors alongside metformin may offer particular value for high-risk individuals. Future large, well-designed, long-term RCTs are needed to confirm these findings and provide more precise, clinically interpretable estimates across all antidiabetic therapies. Fig. 1Flow diagram of the trials’ selection process Flow diagram of the trials’ selection process Fig. 2Network profiles for the included trials comparing different antidiabetic drug classes in prevention of cardiovascular outcomes (Fig. 2a: Estimates with extremely wide confidence intervals are based on sparse data. These are exploratory and were not used for overall treatment rankings; Fig. 2b: The apparent reduction in cardiovascular mortality for DPP-4 inhibitors vs. GLP-1 RAs is based on indirect comparisons. This result should not be interpreted as evidence of clinical superiority.) Network profiles for the included trials comparing different antidiabetic drug classes in prevention of cardiovascular outcomes (Fig. 2a: Estimates with extremely wide confidence intervals are based on sparse data. These are exploratory and were not used for overall treatment rankings; Fig. 2b: The apparent reduction in cardiovascular mortality for DPP-4 inhibitors vs. GLP-1 RAs is based on indirect comparisons. This result should not be interpreted as evidence of clinical superiority.) Supplementary Information Supplementary Material 1 Supplementary Material 1 Supplementary Material 2 Supplementary Material 2