A systematic review and meta-analysis was conducted to assess the impact of semaglutide on MACE in patients with and without DM. The present study was carried out in accordance with the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) guidelines (18).

A systematic search of MEDLINE (PubMed; https://pubmed.ncbi.nlm.nih.gov/), Embase (https://www.embase.com/), Cochrane Central Register of Controlled Trials (CENTRAL; https://www.cochranelibrary.com/) and Web of Science for randomized controlled trials (RCTs) (https://www.webofscience.com) was carried out up to January 2025. Boolean operators (AND/OR) and field restrictions were used to enhance both precision and recall. The following words and their permutations were utilized: ‘semaglutide’ [Title/Abstract] OR ‘GLP-1 receptor agonist’ [Title/Abstract] AND [‘cardiovascular mortality’ (Title/Abstract) OR ‘myocardial infarction’ (Title/Abstract) OR ‘stroke’ (Title/Abstract) OR ‘major adverse cardiovascular events’ (Title/Abstract) OR ‘placebo’ (Title/Abstract) OR ‘randomized clinical trial’ (Title/Abstract)]. Title/Abstract field restrictions were applied to enhance specificity. ClinicalTrials.gov and the WHO International Clinical Trials Registry Platform were searched to find ongoing or unpublished trials. Grey literature (for example, conference abstracts, dissertations, regulatory documents) was excluded to maintain methodological quality and ensure data completeness. This decision was informed by evidence indicating that grey literature or unpublished research may be subject to suboptimal peer review and report few methodological details, thereby potentially introducing bias into effect estimates. The search was augmented by hand-checking the reference lists of relevant trials and systematic reviews to identify additional studies. Trials were considered eligible for inclusion if they met the following criteria: i) Design: randomized, placebo-controlled clinical trials. ii) Population: adults with T2D and non-diabetic adults. iii) Intervention: semaglutide (oral or injectable formulation). iv) Comparator: placebo or active comparator. v) Primary outcome: MACE, a composite of cardiovascular death, non-fatal myocardial infarction and non-fatal stroke. Two reviewers independently screened titles, abstracts, and full-text articles for eligibility. Any disagreement was resolved by discussion or through consultation with a third reviewer.

Two reviewers independently extracted data using a standardized data collection form. The extracted data included study characteristics (author, year, sample size, follow-up duration, intervention, comparator and concomitant therapies), patient demographics [age, sex, baseline BMI, baseline hemoglobin subunit alpha 1 (HbA1c), baseline eGFR, history of CVD, ethnicity, smoking status and hypertension status] and clinical outcomes. The primary endpoint was MACE, defined as a composite of cardiovascular death, non-fatal myocardial infarction, and non-fatal stroke. Secondary outcomes focused on safety endpoints that were extracted systematically and reviewed, including GI adverse events (nausea, vomiting, diarrhea, with incidence rates per 1,000 person-years), gallbladder disease (incidence rates and severity grading), renal adverse events (acute kidney injury, proteinuria and eGFR decline) and discontinuations due to adverse events, with reasons for withdrawal being noted (for example, intolerability and severe adverse reactions). Safety outcomes were categorized according to treatment group (semaglutide + standard care vs. placebo + standard care) to evaluate differential risk. Besides, relative risks (RR), incidence rates, and 95% confidence intervals (CI) were calculated for each event. Concomitant treatments (for example, standard care, antihypertensive treatment and use of lipid-lowering drugs) for both semaglutide and control groups were also ascertained to review their potential impact on outcomes. Disagreements among reviewers were resolved through consensus.

Two reviewers independently assessed the quality of the included RCTs using the JADAD scale, which scores study quality (maximum score of 5) based on randomization, blinding, and handling of withdrawals or dropouts (19). Disagreements were resolved through discussion.

The meta-analysis was conducted using Comprehensive Meta-Analysis (version 2; Biostat, Inc.). For each outcome, treatment effects were summarized as odds ratios (ORs) with 95% CIs. Given the variability across trials in terms of diabetes status, semaglutide dose and formulation, follow-up duration and background therapy, the authors assumed variation in true treatment effects and thus employed a DerSimonian-Laird random-effects model, which estimates the mean of a distribution of effects rather than a single common effect. Heterogeneity across included studies was assessed using prediction intervals instead of the traditional I² statistic. Accordingly, the authors reported the 95 % prediction interval, calculated as exp [pooled ln OR ± 1.96 x √ (within-study variance + τ²)], indicating the range within which the true effect of a future comparable trial is expected to lie. It is expressed on the same OR scale as the pooled estimate.

To test the robustness of results, sensitivity analyses were performed by excluding individual trials one at a time. DerSimonian-Laird τ² was used to examine continuous moderators through mixed-effects meta-regression (random effects for between-study variability and fixed effects for the moderator itself). For MACE outcomes, the dependent variable was the inverse variance-weighted natural-log odds ratio (ln-OR). Regression equation (β₀+β₁ x covariate), Pearson correlation coefficient (r), coefficient of determination (R²=r²), 95% CI for β₁ and two-tailed P-value were reported for each covariate. Moderator correlations (r) were calculated with inverse-variance weighting to reflect the precision of each study. Potential publication bias was assessed through visual inspection of funnel plots and statistical tests, including Egger's regression test and Begg's test. The trim-and-fill method was applied to adjust for potentially missing studies, and the classic fail-safe N was calculated to determine the robustness of the observed effect.

Analyses followed an available-case approach. Missing outcome data from the publication were requested from the authors; if unavailable, the present study was included in qualitative synthesis only. All results were presented with 95% CIs, and a P<0.05 was considered to indicate a statistically significant difference.

A systematic search and selection process was undertaken to identify eligible studies investigating the effects of semaglutide on cardiovascular outcomes and related safety profiles. The search included RCTs, and high-quality clinical studies published in peer-reviewed journals. A total of 1,320 records were retrieved from database searches, with an additional 50 records identified through manual searches and reference list screening, yielding 1,370 records overall. Following the removal of 520 duplicates, 850 records were available for title and abstract screening. At this step, 670 records were excluded due to irrelevance (for example, not addressing semaglutide or cardiovascular outcomes). The full texts of the remaining 180 articles were reviewed for eligibility based on predefined inclusion and exclusion criteria, leading to the exclusion of 169 articles. Therefore, a total of 11 studies, published between 2016 and 2024, collectively encompassing over 25,000 participants, met the final inclusion criteria (12 separate comparisons) and were included in the present systematic review and meta-analysis (Fig. 1). The studies were conducted across diverse populations, including patients with T2DM, overweight/obesity and established cardiovascular risk factors. Follow-up durations varied from 52 weeks up to 39.8 months, offering both short- and long-term insights into semaglutide's efficacy and safety. The interventions primarily evaluated semaglutide at various doses (0.5,1 and 2.4 mg), with placebo or active comparators (for example, liraglutide 3.0 mg) in parallel arms. Racial composition data were reported in 7 of 11 trials. Among them, white participants made up a weighted average of 83% (75-96% range), Black participants comprised 5-12%, and other ethnic groups accounted for up to 8%. None of the trials were conducted exclusively in low- or lower-middle-income countries. None of them reported household income or insurance status and all were sponsored in high-income regions. Consequently, drug acquisition costs and affordability in low-income settings remain unknown (Table I).

A forest plot illustrating the OR and 95% CI for cardiovascular outcomes across multiple studies evaluating semaglutide against placebo is presented in Fig. 2. The overall pooled effect, represented by the red diamond, indicates a statistically significant reduction in cardiovascular risk associated with semaglutide treatment. The pooled OR of 0.69 (95% CI: 0.52-0.92) demonstrates that semaglutide significantly lowers the risk of cardiovascular outcomes.

Across the 11 included RCTs, safety signals were dominated by GI intolerance. Semaglutide administered weekly caused GI events (for example, nausea, vomiting, diarrhea, or constipation) in 74-83% of patients in STEP 1(15) and STEP 3(20), compared with 48-63% placebo. In STEP 8(21), 84% of patients experienced a GI event with semaglutide, vs. 55% with liraglutide or placebo. Discontinuation due to adverse events was also higher with semaglutide: 35 vs. 14 patients with STEP HFpEF (2.4 mg) and 24 vs. 6 patients with STEP 3(20) (2.4 mg), representing a relative risk increase of ~1.3 to 1.6-fold. Using a fixed-effects model (DerSimonian-Laird continuity-correction), the pooled risk ratio for discontinuation due to GI events was 2.32 (95% CI 1.54-3.49), indicating that semaglutide roughly doubles the likelihood of treatment withdrawal due to GI intolerance vs. placebo (Table II).

Biliary or gallbladder events were rare (<2% overall) but exhibited a higher incidence in the active treatment group; examples from STEP 3 include biliary colic and dyskinesia leading to permanent withdrawal (20). The events were primarily mild to moderate in severity, tended to occur early and resolved with no lasting sequelae (Table II).

The sensitivity analysis presented in Fig. 3 demonstrates the robustness of the overall findings, even when individual studies are sequentially removed. The pooled OR remained consistently below 1, with point estimates ranging from 0.630 to 0.761 and 95% CI that do not cross 1 in most cases. P-values remained statistically significant across all analyses, suggesting that the overall reduction in CVD risk associated with semaglutide treatment is robust and not driven by any single study. These results confirm that semaglutide consistently reduces the risk of cardiovascular outcomes, even when individual studies are excluded, reinforcing the reliability and stability of the overall effect.

An institution-level leave-one-out analysis was conducted by sequentially excluding all trials with the same senior author or sponsor. The exclusion of the three STEP trials led by the Wilding/Wadden group (n=2,3,5) or the two HFpEF trials led by Kosiborod et al (22,23) yielded pooled ORs of 0.70 (95% CI, 0.53-0.93) and 0.66 (0.49-0.90), respectively-both including the main estimate-indicating minimal institution-level effect.

Comparably, when the authors re-ran the primary random effects model after eliminating both lower-dose trials (12,15), the pooled OR was 0.621 (95% CI, 0.424-0.909; P=0.014), which remained constant from the estimate of all studies (OR 0.68; Fig. S1). This suggests that the protective effect is not being driven by the lower dose studies. This was further confirmed by dose-stratified meta-analysis, which showed no difference between the higher dose and lower dose studies' effects [≤1 mg: OR=0.74 (0.55-1.00); 2.4 mg: OR=0.66 (0.46-0.94); Test for subgroup difference: Q_between=0.29, P=0.59].

Furthermore, four trials enrolled participants with T2D (SUSTAIN-6, PIONEER-6, STEP 2, STEP HF + T2D) and seven trials enrolled non-diabetic cohorts (SELECT, STEP 1,3,4,5,8, and HFpEF). In the diabetes subgroup, the pooled OR for MACE was 0.72 (95% CI, 0.54-0.96; I²=22%), whereas in the non-diabetes subgroup it was 0.67 (0.46-0.98; I²=18%). A between-subgroup Q-test (Q_between=0.25, df=1) yielded P=0.62, indicating no statistically significant interaction; the 32% overall risk reduction therefore appears consistent across glycemic and non-glycemic settings.

The assessment of publication bias using multiple statistical methods, including the funnel plot (Fig. 4), Begg and Mazumdar's rank correlation, Egger's regression intercept, fail-safe N, and Duval and Tweedie's trim-and-fill method, yielded the following insights: Kendall's Tau was -0.363 without continuity correction (P=0.099) and -0.348 with continuity correction (P=0.114). Although the P-values did not reach statistical significance, a weak trend toward asymmetry was observed. Egger's regression intercept was -2.185 with a standard error of 0.917, yielding P-values of 0.019 (1-tailed) and 0.038 (2-tailed). These results indicate a statistically significant asymmetry, suggesting the potential presence of small-study effects or publication bias. The classic fail-safe N analysis yielded a Z-value of -4.380, indicating that 48 missing studies would be required to nullify the observed effect (P>0.05), suggesting that the findings are robust against the possibility of missing data. Besides, Orwin's fail-safe N OR of 0.981, close to the trivial criterion value of 1.0, suggests that the overall effect was meaningful and unlikely to be overturned by additional missing studies. In Duval and Tweedie's trim-and-fill method, no studies were imputed (adjusted values remain identical to observed values), indicating no evidence of missing studies to the left or right of the mean. This further supports the robustness of the findings. Overall, the findings appear robust, with minimal potential for publication bias.

The distribution of true effects by presenting the standardized mean difference (g) is illustrated in Fig. 5. The mean effect size is 0.69 (95% CI, 0.46-0.92), indicating a moderate and statistically significant positive effect. Additionally, the prediction interval, reflecting variability across comparable populations, ranges from -0.81 to 2.18. This wider interval suggests that despite the overall positive effect, there is heterogeneity, and individual studies may report varying results. These findings confirm an average robust impact while emphasizing the importance of considering potential variability across different settings or populations.

The moderator analysis evaluated the influence of various covariates, including sex, age, systolic and diastolic blood pressure, HbA1c, body weight, LDL, and total cholesterol, on the log OR. A significant positive association was found for age (slope=0.025, P<0.001), HbA1c (slope=0.155, P=0.002), diastolic blood pressure (slope=0.100, P=0.008), and systolic blood pressure (slope=0.100, P=0.008), indicating that higher values of these covariates are linked to an increased log OR. Conversely, significant negative associations were found for body weight (slope=-0.126, P=0.005), total cholesterol (slope=-0.059, P<0.001), and LDL (slope=-0.012, P<0.001), suggesting that higher values of these variables correspond to a reduced log OR. Sex showed no significant association with the log OR (slope=-0.053, P=0.107) (Fig. 6). Besides, meta regression on weekly dose (mg) showed no linear relationship between the dose and larger cardiovascular risk reduction meaning that; higher dose does not systematically cause greater cardiovascular risk reduction [Slope=0.02 (95% CI, 0.07 to 0.03), P=0.38; Fig. S2]. The meta-regression equations (ln OR=β₀ + β₁ x covariate) along with goodness-of-fit values for each panel in Fig. 6 are presented in Table III.

The JADAD quality analysis evaluates the methodological rigor of the included clinical trials based on criteria such as randomization, blinding, and reporting of withdrawals/dropouts, yielding scores ranging from 0 to 5. Out of the 11 studies assessed, 8 (2,11,14,15,20,22-24) received the highest JADAD score of 5, indicating excellent methodological quality characterized by proper randomization, double-blinding, and clear reporting of withdrawals/dropouts. Meanwhile, 3 (21,25,26) received a JADAD score of 4, mainly due to limitations in blinding, though randomization and withdrawals/dropouts were adequately addressed. These results highlight that most studies included in the analysis are of high methodological quality, thereby supporting the reliability and robustness of their findings (Table IV).