The prognostic role of glycemic variability in predicting the risk of adverse cardiovascular events in patients with cardiovascular diseases: a meta-analysis Authors: Bahar Darouei, Reza Amani-Beni, Nasim Kakavand, Arsham Seifnezhad, Sadegh Mazaheri-Tehrani, Arman Shekarian, Reza Eshraghi, Ashkan Bahrami, Ehsan Amini-Salehi, Seyyed Mohammad Hashemi, Mohammad Reza Movahed, Pam R Taub, Guillermo Umpierrez Published: 2025 ABSTRACT AbstractBackgroundGlycemic variability (GV) reflects glucose fluctuations, and increased GV is associated with adverse cardiovascular outcomes. We systematically evaluated the association between GV, measured by the mean amplitude of glycemic excursions (MAGE) and standard deviation (SD), and cardiovascular risk in patients with pre-existing cardiovascular disease (CVD).Materials and methodsWe searched PubMed, Embase, Scopus, Web of Science, and CENTRAL through May 2025 for studies on GV and cardiovascular outcomes in patients with established cardiovascular disease. Random-effects models with Knapp-Hartung adjustment were used to pool the risk ratios (RRs) and 95% confidence intervals (CIs). Subgroup, sensitivity, and meta-regression analyses were used to explore heterogeneity. The risk of bias was assessed using the ROBINS-I tool.ResultsSeventeen studies comprising 6,096 patients were included. In the categorical analysis, higher MAGE (RR: 2.18; 95% CI: 1.70, 2.80; I² = 15.6%) and SD (RR: 1.94; 95% CI: 1.32, 2.87; I² = 55.9%) were associated with increased risk of major adverse cardiovascular events (MACE). Continuous MAGE was also associated with MACE (RR: 1.66; 95% CI: 1.28, 2.14; I² = 65.65%), while continuous SD showed a non-significant association. The adjusted analyses for MAGE reinforced these findings. The MAGE cut-off thresholds significantly moderated the effect sizes in the meta-regression analysis. Subgroup analyses showed between-group differences in SD according to diabetes status and in MAGE according to country and follow-up duration. The certainty of evidence (GRADE) was very low for all contrasts. In outcome-specific analyses, both MAGE and SD were associated with acute heart failure and myocardial infarction, but not cardiovascular mortality.ConclusionGV, particularly when assessed using MAGE, is a significant predictor of cardiovascular risk in patients with pre-existing CVD. While these findings support its prognostic utility, limitations such as the observational design and residual confounding warrant cautious interpretation. Further studies should establish standardized thresholds and dose-response relationships.RegistrationPROSPERO CRD420251002622.Supplementary InformationThe online version contains supplementary material available at 10.1186/s13098-025-02036-5.Keywords: Glycemic variability, Cardiovascular diseases, Mortality, Cardiovascular events, Prognostic factors Background Individuals with established cardiovascular disease (CVD), such as acute coronary syndrome (ACS) or acute heart failure (AHF), face a substantially elevated risk of recurrent major adverse cardiovascular events (MACE) and other long-term complications [1, 2]. Despite advances in pharmacological and interventional therapies, the residual risk remains high in these populations, warranting the identification of novel prognostic indicators that may allow for more personalized treatment interventions and improved secondary prevention strategies [1]. Disturbances in glucose regulation have been linked to higher rates of morbidity and mortality among patients with coronary artery disease (CAD), even in the absence of type II diabetes [3]. In this context, glycemic variability (GV) has gained recognition as an important marker that represents both oscillations in blood glucose levels and the efficiency of the body’s regulatory mechanisms [4]. GV refers to changes in blood glucose levels over time, including the magnitude and frequency of both hyperglycemic and hypoglycemic episodes [5, 6]. Recent studies suggest that GV may be more strongly associated with the severity of diabetic microvascular and macrovascular atherosclerosis than HbA1c [7] and may also serve as a predictor of adverse cardiovascular outcomes, even in the population without diabetes [8–11]. Moreover, relying solely on glycosylated hemoglobin (HbA1c) or fasting plasma glucose (FPG) for glycemic control may be inadequate for predicting long-term cardiovascular outcomes [12]. Therefore, growing attention has been directed toward the potentially harmful impact of GV on coronary arteries and there is emerging interest in addressing it as a therapeutic target for prevention [13]. Among short-term GV metrics, MAGE is most commonly derived from continuous glucose monitoring system (CGMS) or high-frequency self-monitoring of blood glucose (SMBG) profiles, and SD can be computed from either CGM or SMBG [14–16]. Other CGM-based indices, including the mean of daily differences (MODD) and continuous overall net glycemic action (CONGA), have also been reported, but are used less frequently than MAGE or SD [17–19]. Given the substantial residual risk that persists after acute coronary syndromes and acute heart failure despite guideline-directed secondary prevention, identifying early, actionable in-hospital risk markers is a priority [20, 21]. Short-term peri-invasive GV is frequent and clinically informative. GV measured during hospitalization or around PCI has predicted adverse outcomes independent of chronic glycemia in cardiology cohorts [9, 22, 23]. This focus is clinically actionable because inpatient glycemic targets are codified in major guidelines [24], enabling GV-informed strategies to be tested within secondary prevention pathways. In primary prevention settings, visit-to-visit variability of fasting glucose/HbA1c predicts incident cardiovascular disease; however, these studies used different metrics and longer time frames, introducing heterogeneity that would dilute cardiology-specific inferences [11, 25, 26]. Therefore, focusing on patients with established CVD maximizes pathophysiological coherence and translational relevance for secondary prevention. Numerous observational studies have investigated the relationship between GV and cardiovascular events in patients with established CVD. However, findings across these studies remain inconsistent, with some reporting strong associations and others failing to establish a significant link [23, 27–29]. Although previous meta-analyses have demonstrated a significant association between GV and adverse cardiovascular outcomes in patients with ACS or CAD [3, 30], their conclusions were constrained by methodological limitations and outdated data, highlighting the need for more robust and comprehensive evaluations across broader cardiovascular cohorts. To address these gaps, the present meta-analysis aimed to evaluate the prognostic significance of GV, specifically measured using MAGE and SD, in predicting adverse cardiovascular outcomes in patients with pre-existing CVD. Methods Protocol and registration This meta-analysis followed the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) guidelines [31]. The protocol was registered in the PROSPERO database (CRD420251002622). Search strategy We systematically searched PubMed, Scopus, Web of Science, CENTRAL, and Embase up to August 16, 2024, and updated the search until May 3, 2025. The selection of search terms was informed by a combination of Medical Subject Headings (MeSH), relevant keywords, and terms identified in previous systematic reviews on GV and cardiovascular outcomes, leading to combinations of keywords including “glycemic variability,” “mean amplitude of glycemic excursions (MAGE),” “standard deviation (SD),” “cardiovascular disease,” and “major adverse cardiac event (MACE).” Google, Google Scholar, and the reference lists of relevant articles were also searched manually to identify additional eligible studies. No language restrictions were applied. The detailed search strategy is provided in Supplementary Table S1. Eligibility criteria and study selection We included observational cohort, case-control, and post-hoc analyses of clinical trials studies that examined the association between GV and cardiovascular outcomes in adult patients with established cardiovascular disease. To be eligible, studies had to measure GV specifically using either the MAGE or SD of glucose levels at baseline and report effect sizes, such as risk ratios (RRs), odds ratios (ORs), or hazard ratios (HRs) with corresponding 95% confidence intervals (CIs). Additionally, studies that only reported prevalence rates or mean differences between GV groups without calculating effect estimates were excluded. The outcomes of interest included MACE, myocardial infarction (MI), AHF, cardiovascular mortality, and stroke. We excluded studies that assessed GV using other metrics, such as the coefficient of variation (CV), variability independent of the mean (VIM), MODD, CONGA, or similar indices, because these measures were rarely reported across eligible studies, which precluded consistent pooling. Additionally, we excluded conference abstracts, review articles, preclinical studies, editorials, and studies involving non-human subjects. Although the search was not limited by language, only studies published in English were included in the analysis. Four independent reviewers (A.S., N.K., B.D., and R.E.) independently screened the titles and abstracts of all retrieved records using EndNote software to identify potentially eligible studies. The full texts of the selected articles were then assessed for inclusion based on the predefined eligibility criteria. Any disagreements between the reviewers were resolved through discussion and consensus with the involvement of another reviewer (R.A.B.) when necessary. Data extraction and quality assessment From each included study, we extracted relevant data, including the first author’s name, year of publication, country, study design, sample size, population characteristics (age, male percentage, baseline diabetic status, baseline HbA1C), method of GV assessment (MAGE or SD), duration of GV measurement, GV cut-off values, follow-up duration, potential adjusted confounders in the association between GV and CV outcomes, and reported CV outcomes. When available, we collected both categorical effect estimates (e.g., highest vs. lowest GV categories) and continuous estimates (per-unit increase in MAGE or SD). Three reviewers independently extracted the data (E.A.S., B.D., and A.B.), and discrepancies were resolved by discussion or consultation with a third reviewer (R.A.B). The risk of bias (ROB) for non-randomized studies was assessed using the Risk of Bias in Non-randomized Studies - of Interventions (ROBINS-I) tool by two reviewers (E.A.S. and S.M.H.). Conflicts were settled by discussion or, when necessary, by a third-party reviewer (R.A.B). The ROBINS-I tool evaluates seven bias domains, including confounding, selection, classification of exposures, deviations from intended interventions, missing data, outcome measurement, and selective reporting [32]. Each study was independently assessed, and judgments were categorized as low, moderate, or serious risk of bias. Statistical analysis All statistical analyses were performed using Stata version 17.0 (StataCorp, College Station, TX, USA). We conducted separate meta-analyses for MAGE and SD. For the primary endpoint (MACE), exposures were modeled both as categorical (highest vs. lowest GV) and continuously (per-unit increase). For secondary endpoints (MI, AHF, and cardiovascular mortality), data availability permitted only categorical contrasts. We prespecified RR as the common effect measure. ORs were converted to RRs using standard formulas [33]. For continuous contrasts, when only HRs were available (continuous MAGE or SD for MACE), we transformed HRs to approximate RRs using published approaches [34]. For robustness, we also conducted an HR-only sensitivity analysis on the log-HR scale when possible. For categorical MAGE, categorical SD, and continuous SD, we pooled crude RRs with 95% CIs. For continuous MAGE (MACE), we synthesized the most comprehensively adjusted estimates. Meta-analyses were conducted using a random-effect model with a restricted maximum likelihood (REML) estimator to account for between-study variance. In particular, the Knapp-Hartung adjustment was applied in analyses including 10 or fewer studies. Heterogeneity across studies was quantified using the I² statistic, with values interpreted as low (< 25%), moderate (25–50%), severe (50–75%), and very severe (> 75%). The 95% prediction intervals (PIs) were also calculated to estimate the likely range of effect sizes in future comparable populations. To explore sources of heterogeneity, subgroup analyses were conducted based on diabetes status, follow-up duration, GV measurement duration, and geographic region. Additionally, meta-regression analyses were performed to evaluate the effect of study-level covariates, including the percentage of male participants and GV cut-off values, on pooled effect estimates. Specifically, meta-regression on GV cut-off was conducted for both MAGE and SD in relation to MACE to assess linear trends and inform potential dose-response relationships. Meta-regression outputs included regression coefficients, p-values, and the proportion of explained between-study variance (R²). To evaluate the robustness of the findings, leave-one-out sensitivity analyses were conducted by sequentially omitting each study to assess its influence on the overall pooled estimates. Additional sensitivity analyses were performed using adjusted RRs for categorical MAGE and MACE associations and adjusted HRs for continuous MAGE and MACE associations. Publication bias was assessed using Egger’s test, with a p-value < 0.10, indicating potential small-study effects. The trim-and-fill method was applied to identify and adjust for potentially missing studies, thereby generating bias-corrected pooled estimates. All statistical tests were two-sided, and results with P-values less than 0.05 were considered statistically significant. Certainty of evidence (GRADE) We assessed the certainty of evidence for the primary outcome (MACE) using the GRADE approach adapted for prognostic/association evidence [35]. Following a conservative convention for non-randomized prognostic studies, certainty ratings started at low and were downgraded for risk of bias, inconsistency, indirectness, imprecision, and small-study effects. We considered upgrading for large effects, dose-response, and where all plausible residual confounding or bias would attenuate the observed association. To avoid double-counting, we summarized one primary contrast per exposure metric and scale: MAGE (categorical highest vs. lowest; continuous per-unit) and SD (categorical highest vs. lowest; continuous per-unit). Final judgments are reported in a Summary-of-Findings table. Results Study selection A total of 2,931 records were identified through database searching, and an additional 258 records through other sources (Google Scholar, citation tracking, and reference lists). After removing duplicates, 1,999 records were screened by their titles and abstracts. Of these, 156 full-text articles were assessed for eligibility, and 17 studies met the inclusion criteria and were included in the meta-analysis [9, 10, 22, 23, 27–29, 36–45]. The selection process is illustrated in Fig. 1. PRISMA flowchart of the selection process Baseline characteristics of the included studies This meta-analysis included 17 observational studies conducted between 2013 and 2024, comprising 6,096 participants with established cardiovascular disease (Table 1). Most studies were single-center cohorts from China (n = 11), followed by Japan (n = 3), France (n = 2), and Australia (n = 1); one study was a two-center study. Most studies have focused on patients with acute coronary syndrome (ACS), including ST-elevation myocardial infarction (STEMI), non-STEMI (NSTEMI), and unstable angina, although a few have included broader cardiovascular populations, such as patients with stable angina or AHF. The majority of studies were prospective cohort studies (n = 14), while the remaining (n = 3) were retrospective. The mean age of the participants ranged from 52 to 73 years, and the proportion of males varied between 47.7% and 85.2%. While some studies included only patients with diabetes (n = 6), others enrolled mixed (n = 9) or non-diabetic populations (n = 2). GV was measured using MAGE (n = 9), SD (n = 6), or both (n = 3), with monitoring periods ranging from 1 to 3 days and up to weeks. The follow-up duration varied widely from 1 to 65 months. Of the 17 studies, 8 were judged to have a moderate risk of bias, 2 had a serious risk, and 7 were rated as low risk (Table S2). Table 1Baseline characteristics of the included studiesStudy (Year)CountryDesignN (participants)Type of ACSMean age, y (SD)Male (%)Diabetes, %GV metric (MAGE/SD)Monitoring period (days)GV cut-off (mmol/L)Follow-up, monthsYang et al. [43], 2024ChinaProspective (single center)484Acute STEMI63.662.6100MAGE (lab FPG)3Per 1 unit12Gerbaud et al. [36], 2022FranceProspective (single center)392AHF7369.1100SD (SMBG)NA (5 tests/day)TertilesPer 1 unit29 [6–51]Yi et al. [44], 2022ChinaRetrospective (single center)612STEMI56.380.6100SD (lab FPG)NAQuantiles7.2 ± 2.1 (days)Yamamoto et al. [28], 2021JapanProspective (single center)101Stable CAD70.878.253.5MAGE (CGMS)1Per 1 unit65 [43–90]Chai et al. [27], 2019AustraliaRetrospective post-hoc analysis (single center)121AMI63.079.361.1MAGE (SMBG), SD (SMBG)12.8 for MAGE1.6 for SDPer 1 unit3Gerbaud et al. [10], 2019FranceProspective (single center)327ACS6977.1100SD (SMBG)4 (5 tests/day)TertilesPer 1 unit16.9 ± 7.1Takahashi et al. [9], 2018JapanProspective (single center)417ACS6683.533.6MAGE (CGMS), SD (CGMS)14.1 for MAGEPer 1 unit39Su et al. [40], 2018ChinaRetrospective (single center)759NSTE-ACS62.861.3100MAGE (CGMS)1–23.9NAXia et al. [41], 2018ChinaProspective (single center)273ACS5961.2100SD (SMBG)4 times/week26Mi et al. [37], 2017ChinaProspective (single center)256STEMI62.560.50MAGE (CGMS)33.26 - Per 1 unit3Akasaka et al. [23], 2017JapanProspective (single center)65Stable angina71.247.70MAGE (CGMS)23.612Xia et al. [42], 2017ChinaProspective (single center)864ACS6763.753.1SD (SMBG)1 (7 times/day)21Xia et al. [29], 2017ChinaProspective (single center)746Stable angina pectoris6466.660.1SD (SMBG)3 (7 times/day)26Zhang et al. [45], 2014ChinaProspective (two center)73STEMI5568.5100MAGE (CGMS)1Tertiles - Per 1 unit11645270.10Wang et al. [22], 2014ChinaProspective (single center)34AMI62.885.282.3MAGE (CGMS), SD (CGMS)26.05 for MAGE - Per 1 unit17.03 ± 1.66Su et al. [39], 2013ChinaProspective (single center)222AMI6262.653.6MAGE (CGMS)23.9 - Per 1 unit12Su et al. [38], 2013ChinaProspective (single center)186Elderly AMI67.060.454.3MAGE (CGMS)32.55 - Per 1 unit12ACS = acute coronary syndrome; AHF = acute heart failure; AMI = acute myocardial infarction; CAD = coronary artery disease; DM = diabetes mellitus; GV = glycemic variability; CGMS = continuous glucose monitoring system; SMBG = self-monitoring of blood glucose; MAGE = mean amplitude of glycemic excursions; NSTE-ACS = non–ST-segment elevation acute coronary syndrome; SD = standard deviation Baseline characteristics of the included studies Tertiles Per 1 unit 2.8 for MAGE 1.6 for SD Per 1 unit Tertiles Per 1 unit 4.1 for MAGE Per 1 unit ACS = acute coronary syndrome; AHF = acute heart failure; AMI = acute myocardial infarction; CAD = coronary artery disease; DM = diabetes mellitus; GV = glycemic variability; CGMS = continuous glucose monitoring system; SMBG = self-monitoring of blood glucose; MAGE = mean amplitude of glycemic excursions; NSTE-ACS = non–ST-segment elevation acute coronary syndrome; SD = standard deviation Association between MAGE-Based GV and MACE (categorical) Categorical analyses of the association between MAGE-based GV and MACE risk were conducted across nine studies, demonstrating a significant relationship. A random-effects REML model with Knapp-Hartung adjustment yielded a pooled RR of 2.18 (95% CI: 1.70, 2.80, P < 0.001), indicating a clear association with MACE, and a 95% PI of 1.45 to 3.28 (Fig. 2A). A separate sensitivity analysis of adjusted RRs from four of these studies supported this finding, showing an adjusted RR of 2.43 (95% CI: 1.29, 4.61, P = 0.02) with a 95% PI of 0.60 to 9.80. I2 was 15.58% for the overall RR and 34.77% for the adjusted RR (Figure S1). Forest plots of the association between MAGE-based glycemic variability and major adverse cardiovascular events based on (A) Categorical analysis (highest vs. lowest MAGE); and (B) Continuous analysis (per unit increase in MAGE) Subgroup analyses provided further insights. Stratification by follow-up duration showed that studies with follow-up ≤ 3 months had an RR of 1.78 (95% CI: 1.11, 2.84, I² = 13.45%), while those with follow-up ≥ 12 months had a higher RR of 2.71 (95% CI: 2.13, 3.45, I² = 0.00%), with a borderline significant between-group difference (P = 0.05). Stratification by country revealed regional variations: studies from China reported an RR of 2.29 (I² = 0%), the study from Japan showed an RR of 2.68 (I² = 0%), and the one from Australia did not show a significant association. The between-group differences were also significant (P = 0.04). In the ACS-only sensitivity analysis, pooling of seven studies yielded a similar association (RR = 2.06, 95% CI: 1.57, 2.71), and in CGMS-only studies, the association remained strong and homogeneous (pooled RR = 2.41 [95% CI 2.04, 2.84]; I² = 0%). Additional subgroup analyses are presented in Table 2. When restricted to studies with low or moderate risk of bias, the association between MAGE and MACE remained significant (RR = 2.00, 95% CI: 1.53, 2.62). Table 2Subgroup and sensitivity analyses of the association between MAGE and MACESubgroupNumber of studiesMeta-analysisHeterogeneityPooled RR (95%CI)I2 (%)P between group Categorical analysis (highest vs. lowest MAGE) All 92.18 (1.70, 2.80)15.58% Region and Country 0.04 China52.29 (1.90, 2.76)0% Japan32.68 (1.34, 5.32)0% Australia11.18 (0.70, 1.99)- ACS-only population 72.06 (1.57, 2.71)10.07% GV monitoring method CGMS-only82.41 (2.04, 2.84)0% Diabetes Status 0.86 Non-DM31.96 (1.17, 3.29)0% Mix-DM52.31 (1.33, 4.02)53.62% All-DM22.15 (0.83, 5.58)0% GV measurement Duration 0.70 1 day42.10 (0.98, 4.47)59.15% 2–3 days52.33 (1.92, 2.82)0% Follow-up Duration 0.05 ≤ 3 months41.78 (1.11, 2.84)13.45% ≥ 12 months52.71 (2.13, 3.45)0% Continuous analysis (per-unit increase in MAGE) All 81.66 (1.28, 2.14)65.65% Region and Country China61.91 (1.47, 2.48)20.19% ACS-only population 71.76 (1.30, 2.38)65.67% GV monitoring method CGMS-only61.66 (1.32, 2.09)24.93% Diabetes Status 0.17 Non-DM12.02 (1.14, 3.58)- Mix-DM51.52 (1.15, 2.02)62.82% All-DM12.85 (1.46, 5.57)- GV measurement Duration 0.06 1 day31.41 (0.77, 2.58)75.27% 2 days21.55 (0.21, 11.64)4.70% 3 days32.40 (1.53, 3.77)0% Follow-up Duration 0.62 ≤ 6 months31.56 (0.71, 3.45)74.80% ≥ 12 months51.76 (1.18, 2.62)51.32%ACS = acute coronary syndrome; CI = confidence interval; DM = diabetes mellitus; GV = glycemic variability; CGMS = continuous glucose monitoring system; SMBG = self-monitoring of blood glucose; MACE = major adverse cardiovascular events; MAGE = mean amplitude of glycemic excursions; RR = risk ratio Subgroup and sensitivity analyses of the association between MAGE and MACE ACS = acute coronary syndrome; CI = confidence interval; DM = diabetes mellitus; GV = glycemic variability; CGMS = continuous glucose monitoring system; SMBG = self-monitoring of blood glucose; MACE = major adverse cardiovascular events; MAGE = mean amplitude of glycemic excursions; RR = risk ratio Meta-regression analyses illustrated that the MAGE cut-off explained between-study heterogeneity (R² = 100%), while the male proportion accounted for none (R² = 0%). Each 1-mmol/L increase in the cut-off threshold was associated with a 1.77-fold higher pooled RR for MACE (95% CI 1.21–2.59). Bubble plots are shown in Fig. 4A and Figure S2. Sensitivity analysis using the leave-one-out approach revealed that omitting individual studies did not substantially alter the magnitude, direction, or statistical significance of the pooled estimates (Figure S3). The trim-and-fill analysis imputed three studies; however, Egger’s test (P = 0.13) suggested no significant small-study effects. Bubble plots showing the association between cut-off thresholds and effect sizes in (A) MAGE (categorical) and (B) SD (categorical) analyses Association between MAGE-Based GV and MACE (continuous) Analysis of eight studies evaluating the continuous association between MAGE and MACE yielded a significant pooled adjusted RR of 1.66 (95% CI: 1.28, 2.14), based on a random-effects REML model with Knapp-Hartung adjustment (Fig. 2B). This indicates an increased risk of MACE per unit increase in MAGE (mmol/L). Severe heterogeneity was observed (I² = 65.65%), and the 95% PI ranged from 0.88 to 3.12. A sensitivity analysis of five studies reporting adjusted HRs showed a similarly elevated pooled HR of 2.66 (95% CI: 1.84, 3.86), accompanied by mild heterogeneity (I² = 0%) and a narrower 95% PI (1.61, 4.40), indicating more consistent effects across studies (Figure S4). Subgroup analyses explored heterogeneity (Table 2). By region, studies from China showed a consistent association (RR = 1.91; 95% CI, 1.47, 2.48; I² = 20.19%). In the ACS-only analysis, seven studies showed a significant association (RR = 1.76, 95% CI: 1.30, 2.38). In CGMS-only studies, each unit increase in MAGE was associated with a higher MACE risk (RR = 1.66 [1.32, 2.09]), with mild-to-moderate heterogeneity (I² = 24.9%). MAGE monitoring duration suggested stronger effects with longer capture (P = 0.06): 1-day RR = 1.41 (0.77, 2.58; I² = 75.27%); 2-days RR = 1.55 (0.21, 11.64; I² = 4.70%); 3-days RR = 2.40 (1.53, 3.77; I² = 0%). Results were similar by follow-up duration (P = 0.62): ≤6 months RR = 1.56 (0.71, 3.45; I² = 74.80%) vs. ≥ 12 months RR = 1.76 (1.18, 2.62; I² = 51.32%). In the continuous analysis, the pooled effect remained robust across low- or moderate-ROB studies (RR = 1.68, 95% CI: 1.20, 2.35). Meta-regression revealed a significant inverse association between male proportion and the MAGE-MACE effect size (P = 0.009), accounting for 79% of the between-study heterogeneity (Figure S5). Robustness checks through leave-one-out sensitivity analysis confirmed that no individual study disproportionately influenced the overall estimate. Small-study effects were suggested by Egger’s test (P < 0.001); However, trim-and-fill showed no small-study. Outcome-specific analyses of MAGE-based GV In a separate analysis focused on AHF, three studies examined the prognostic role of MAGE. The pooled estimate from a random-effects model with Knapp-Hartung adjustment showed a significant association (RR = 3.19; 95% CI: 1.11, 9.12) with mild heterogeneity (I² = 0) (Figure S7). Egger’s test (P = 0.60) and trim-and-fill analysis suggested no small-study effects. Four studies on MAGE and cardiovascular mortality yielded a non-significant pooled RR of 2.15 (95% CI: 0.16, 29.33), with severe heterogeneity (I² = 65.2%) (Figure S8). Leave-one-out analysis showed consistent results (Figure S9). Egger’s regression test results (P = 0.34) and trim-and-fill analysis did not indicate small-study effects. Five studies evaluated the association between MAGE and MI. The pooled risk ratio was 2.52 (95% CI: 1.01, 6.57), which reached statistical significance (P = 0.05) with mild heterogeneity (I² = 2.04%) (Figure S10). Leave-one-out analysis revealed that omitting any study, except Su et al. (2018), led to notable changes in the magnitude and direction of the pooled effect estimate (Figure S11). Egger’s test (P = 0.0574) and the trim-and-fill method suggested small-study effects. Evidence for stroke was limited (two studies). In the diabetic NSTE-ACS PCI cohort, the high-MAGE group (≥ 3.9 mmol/L) showed a numerically higher stroke incidence than the low-MAGE group (1.8% vs. 0.9%), but the difference was not statistically significant [40]. In the STEMI CGM cohort, the “intermittent hyperglycemia” pattern, characterized by higher MAGE, had two strokes within 30 days, while the other patterns had none; no further strokes occurred through 6 months [46]. Given the limited number of studies, stroke was summarized narratively rather than pooled. Association between SD-based GV and MACE (categorical) Seven studies were included in the meta-analysis assessing the association between GV (as measured by the SD of glucose levels) and the risk of MACE. Using a random-effects model with Knapp-Hartung adjustments, the pooled RR for the highest vs. lowest SD category was 1.94 (95% CI: 1.32, 2.87), indicating a significantly elevated risk of MACE associated with higher GV (Fig. 3A). There was severe heterogeneity among the included studies (I² = 55.9%), and the 95% PI ranged from 0.85 to 4.44. Sensitivity analysis using the leave-one-out method showed no considerable change in magnitude, direction, and statistical significance (Figure S12). Forest plots of the association between SD-based glycemic variability and major adverse cardiovascular events based on (A) Categorical analysis (highest vs. lowest SD); and (B) Continuous analysis (per unit increase in SD) In the subgroup analysis, studies including only populations with diabetes showed stronger associations (RR = 2.40; 95% CI: 1.72, 3.34) than mixed populations (RR = 1.48; 95% CI: 0.63, 3.51), with a statistically significant between-group difference (P = 0.03) (Table 3). In ACS-only studies, five cohorts demonstrated a significant association between high SD and MACE (RR = 1.78, 95% CI: 1.12, 2.84 (Table 3). In SMBG-only studies, the association persisted (RR = 1.90 [1.20, 3.00]). A meta-regression showed that the SD cut-off value significantly influenced the results (R² = 89.34%, P = 0.036), but not the male sex (R2 = 0, P = 0.76) (Fig. 4B and Figure S13). Each 1-mmol/L increase in the SD threshold was associated with a 1.94-fold higher pooled RR for MACE (95% CI 1.07–3.51). Neither the trim-and-fill analysis nor Egger’s test (P = 0.25) indicated small-study effects. Table 3Subgroup and sensitivity analyses of the association between SD and MACESubgroupNumber of studiesMeta-analysisHeterogeneityPooled RR (95%CI)I2 (%)P between group Categorical analysis (highest vs. lowest SD) All 71.94 (1.32, 2.87)55.91% Region and Country 0.81 China41.88 (1.41, 2.49)0% Australia & France32.09 (0.31, 14.29)87.59% ACS-only population 51.78 (1.12, 2.84)60.11% GV monitoring method SMBG-only61.90 (1.20, 3.00)61.95% Diabetes Status 0.03 Non-DM & mix-DM31.48 (0.63, 3.51)52.23% All-DM42.40 (1.72, 3.34)0% Follow-up Duration 0.09 ≤ 6 months31.47 (0.50, 4.36)52.23% ≥ 6 months42.30 (1.65, 3.20)0% Continuous analysis (per-unit increase in SD) All 51.19 (0.88, 1.61)99.27% ACS-only population 41.30 (0.84, 2.01)74.20% GV monitoring method 0.22 SMBG-only31.10 (0.73, 1.63)99.26% CGMS-only21.58 (0.04, 62.82)61.56%CI = Confidence interval; DM = Diabetes mellitus; GV = Glycemic variability; CGMS = continuous glucose monitoring system; SMBG = self-monitoring of blood glucose; RR = risk ratio; SD = Standard deviation Subgroup and sensitivity analyses of the association between SD and MACE CI = Confidence interval; DM = Diabetes mellitus; GV = Glycemic variability; CGMS = continuous glucose monitoring system; SMBG = self-monitoring of blood glucose; RR = risk ratio; SD = Standard deviation Association between SD-Based GV and MACE (Continuous) In five studies reporting SD as a continuous variable, the pooled estimate from a random-effects model with Knapp–Hartung adjustment showed a non-significant association between GV and MACE (RR = 1.19; 95% CI: 0.88, 1.61) (Fig. 3B). The 95% PI was 0.62 to 2.28, with very severe between-study heterogeneity (I² = 99.27%). In ACS-only analyses, continuous SD showed no significant association with MACE (RR = 1.30, 95% CI: 0.88, 1.61) (Table 3). Neither SMBG-only nor CGMS-only subgroups provided reliable evidence of an association. Leave-one-out analysis confirmed that the overall results were not driven by any single study (Figure S14). Egger’s regression test (P = 0.0214) suggested significant small-study effects, whereas trim-and-fill analysis did not indicate its presence. Outcome-Specific analyses of SD-Based GV Four studies explored the association between SD-based GV and AHF. Using a random-effects model with Knapp–Hartung adjustment, the pooled RR was 2.19 (95% CI: 2.08, 2.31), revealing a strong association with AHF risk (Figure S15). There was mild heterogeneity (I² = 0%), and the 95% PI (2.05, 2.35) reinforced the consistency of this effect. Leave-one-out analysis showed that no single study overly influenced the outcome (Figure S16). Egger’s test (P = 0.616) and trim-and-fill analysis indicated no small-study effects. Three studies examining the association between GV, as measured by the SD of glucose levels, and the risk of CV mortality were included. The pooled RR was 2.10 (95% CI: 0.13, 35.07), and moderate heterogeneity was observed (I² = 29.12%) (Figure S17). Egger’s test (P = 0.31) and trim-and-fill analysis did not indicate a significant small-study effect. Moreover, based on four studies on the association between SD-based GV and the risk of MI, the pooled RR for high vs. low SD was 2.40 (95% CI: 1.82, 3.18) (Figure S18). There was mild heterogeneity across the studies (I² = 0%), and the 95% PI for the pooled effect was 1.65–3.51. Subgroup analysis by leave-one-out showed that omitting each study did not change the significance and magnitude of the pooled results (Figure S19). Egger’s test (P = 0.3148) and trim-and-fill analysis suggested no significant small-study effect. Evidence for stroke with SD-based GV was limited (n = 2). In a cohort evaluating visit-to-visit FPG variability, higher variability was not independently associated with nonfatal stroke (Q4 vs. Q1 adjusted HR 1.11; 95% CI 0.70, 1.77 and per 1-SD increase HR: 1.03; 95% CI 0.86, 1.25) [12]. In an ACS inpatient cohort classifying GV by the SD of inpatient glucose, stroke events were rare and not significantly different between groups (0.7% vs. 1.7% for high- vs. low-SD) [42]. Given the small number of studies, stroke was summarized narratively and not pooled. Certainty of evidence (GRADE) Using GRADE for prognostic associations, the certainty of evidence for MACE was very low across all contrasts (MAGE and SD, categorical and continuous). MAGE (categorical, highest vs. lowest) showed a large association, but was downgraded for ROB and indirectness. MAGE (continuous) was downgraded for ROB, indirectness, inconsistency, and small-study effects and upgraded for dose-response and confounding. SD (categorical) was downgraded for inconsistency and indirectness. SD (continuous) was downgraded for inconsistency, indirectness, and small-study effects (Table S3). Discussion This comprehensive meta-analysis examined the association between GV and cardiovascular outcomes, focusing on two commonly used metrics: MAGE and SD of glucose values. Our findings consistently support the role of increased GV as a predictor of adverse cardiovascular events, with stronger and more consistent associations observed for MAGE compared to SD of glucose values. A significant association was observed between MAGE and MACE in both categorical comparisons (highest vs. lowest GV) and continuous analyses. Interestingly, sensitivity analyses and PIs suggested that isolated studies did not drive this association. Additional analyses restricted to studies reporting adjusted risk ratios and hazard ratios further confirmed our main findings. This finding highlights the need for a consensus on clinically meaningful GV thresholds, particularly for MAGE, to enhance comparability and applicability across future research. In subgroup analyses of MAGE, studies using longer GV monitoring durations (≥ 3 days) and those conducted in China demonstrated more consistent effects. Similarly, longer follow-up durations (≥ 12 months) were associated with stronger associations, although the between-group difference was only marginally significant. The analysis of SD yielded similar trends, although the associations were somewhat attenuated in comparison with MAGE. Categorical comparisons showed a directionally consistent relationship with cardiovascular risk, whereas continuous analyses demonstrated weaker effects. This discrepancy likely reflects the differential sensitivity of the GV metrics. MAGE focuses on major glucose excursions and, therefore, better reflects the impact of postprandial spikes, which are known to drive oxidative stress, vascular damage, and cardiovascular events [5, 47]. In contrast, SD represents global variability and does not differentiate between minor and significant fluctuations [47]. This may partially explain the stronger and more consistent predictive performance of MAGE observed in this study. These observations align with the growing recognition that not only the degree, but also the timing and acuteness of glycemic fluctuations influence vascular injury. Moreover, for SD, subgroup effects were more variable. Although the overall association remained, between-group differences were significant for diabetes status, indicating a stronger association in studies that restricted participants to those with diabetes. Although the associations, particularly for MAGE in categorical contrasts, were consistent and clinically meaningful, the certainty of evidence was very low. These ratings support a cautious interpretation and underscore the need for diverse, well-adjusted cohorts using standardized GV definitions. Our results are broadly consistent with those of the previous studies. A meta-analysis by Zhang et al. reported a significant association between GV and cardiovascular outcomes in patients with ACS (RR = 1.76; MAGE: RR = 2.14 SD) [30]. However, their approach combines categorical and continuous estimates into a single model. This limits the ability to distinguish the prognostic value of specific GV measures. In contrast, our meta-analysis employed a more rigorous, stratified methodology, conducting separate analyses for categorical versus continuous exposures. Moreover, our study included a broader spectrum of CVD, encompassing ACS, stable CAD, and AHF, thereby enhancing generalizability. Notably, our results showed stronger and more consistent associations for MAGE than for SD, reflecting differences in patient populations, exposure definitions, or analytic approaches compared to previous meta-analyses. In another study, Pu et al. showed that higher MAGE levels were significantly associated with increased MACE risk in CAD patients, with no subgroup differences according to diabetes status [3]. Consistent with their results, our MAGE subgroup analysis also showed no significant difference in the diabetes status. However, a key distinction emerged in our SD analysis, in which we observed a significant between-group difference, suggesting that the impact of SD-based GV may be more prominent in individuals with diabetes. Several mechanistic pathways link GV to cardiovascular pathologies. Glycemic fluctuations may contribute to adverse outcomes by promoting endothelial dysfunction, primarily through increased oxidative stress and the formation of advanced glycation end products [44, 48, 49]. Additionally, recurrent hypoglycemia and inconsistent insulin use are observed in patients with high GV and may further exacerbate vascular instability and greater oxidative stress than persistent hyperglycemia, primarily by enhancing mitochondrial superoxide generation [48–50]. A higher GV has also been linked to more severe coronary lesions [51] and increased plaque vulnerability on intravascular ultrasound imaging [52]. Intermittent hyperglycemia may also trigger endothelial apoptosis, accelerate atherosclerosis, impair autonomic function, and enhance platelet thrombogenicity, collectively contributing to greater cardiovascular risk [53–55]. Notably, even 48–72 h of peri-invasive glycemic swings can leave changes that persist beyond discharge. Oscillating glucose provokes disproportionate oxidative stress and endothelial dysfunction compared with sustained hyperglycemia [56]. In vivo, brief glucose fluctuations accelerate endothelial apoptosis, heighten inflammatory cytokines, and impair vasorelaxation, causing durable cardiovascular injury [57]. Acute hyperglycemia can rapidly damage the endothelial glycocalyx and activate coagulation, seeding a prothrombotic, low-NO milieu that is relevant to stented segments and the microvasculature [58, 59]. GV also enhances platelet hyperreactivity and platelet-driven inflammation, effects that may persist despite standard antiplatelet therapy [60]. Importantly, the “down” swings of GV are not benign: experimental hypoglycemia increases platelet activation, inflammation, and impairs fibrinolysis, even after glucose normalizes [61]. During ACS, glycemic swings intersect with ischemia-reperfusion biology, amplifying ROS via Txnip/p38-MAPK signaling and worsening myocardial injury and remodeling [62]. Finally, transient spikes can imprint ‘metabolic memory’ through persistent epigenetic activation of NF-κB in vascular endothelium, sustaining a pro-inflammatory phenotype days after normoglycemia [63]. Collectively, these mechanisms provide biological plausibility that GV captured over 48–72 h reflects and reinforces vascular and platelet pathobiology not fully reversed by secondary prevention, thereby linking short-term GV to higher long-term MACE risk. A particularly important observation from our meta-regression analysis showed that the cut-off thresholds used to define high GV explained a large portion of the observed heterogeneity. This finding not only underscores the need for consensus around standardized GV thresholds but also raises the possibility of a non-linear relationship between GV and cardiovascular risk. Future studies should explore this potential dose-response pattern through appropriately designed analyses, as the risk associated with GV may not increase uniformly across its spectrum. Moreover, randomized controlled trials are needed to assess whether therapeutic strategies that reduce GV can improve cardiovascular outcomes. Early evidence suggests that GV is modifiable in clinical practice. In a randomized trial among adults with metabolic syndrome and prediabetes, time-restricted eating (TRE) reduced CGM-derived intradaily and interdaily variability over 3 months [64]. In type 2 diabetes, time-restricted eating TRE improved CGM-based control and reduced daily glucose SD and time above range, and a randomized crossover study of early TRE reduced MAGE and time in hyperglycemia independent of weight change [65, 66]. Although conducted in a primary-prevention population, these data support using GV as a treatment-responsive metric that can be trended over time to assess response to intervention and be tested in secondary-prevention settings. Future studies should also explore which GV metric is most predictive in various clinical contexts. Additionally, efforts should be made to develop and validate standardized cut-off points for MAGE and SD. Nonetheless, some limitations must be acknowledged. First, most of the included studies were observational in design, which limits causal inference. Also, these findings apply to patients with pre-existing CVD; extrapolation to primary-prevention populations should be cautious. Although adjusted estimates were used wherever available, and the risk of bias was formally assessed, residual confounding cannot be fully excluded. Second, considerable heterogeneity was present in some pooled estimates, particularly in the continuous analyses. This was partly explained by study-level factors, such as GV cut-off values, follow-up duration, and diabetes status, but unexplained heterogeneity remains a challenge. Third, variability in GV measurement methods, including differences in glucose monitoring duration, data collection frequency, and analytical tools, may have introduced nonsystematic errors into the estimates. Additionally, some outcomes, such as cardiovascular mortality and AHF, were underrepresented, which restricted the statistical power of these endpoints. Because most included cohorts were conducted in East Asia (China and Japan), external validity to non-East Asian populations is uncertain, and regional practice patterns and patient characteristics may limit generalizability. Finally, GRADE ratings were predominantly low/very low, driven by risk of bias, inconsistency (high I² and PIs crossing the null for continuous contrasts), indirectness (East Asia-dominant evidence; variable MACE definitions), and small-study effects. This does not negate the observed associations but limits confidence in the magnitude and generalizability. Conclusions This meta-analysis provides promising evidence that a higher GV, particularly as measured by MAGE, is independently associated with an increased risk of major cardiovascular events in patients with pre-existing CVD. These findings were consistent across diverse populations and analytical approaches and remained robust in multiple sensitivity and subgroup analyses. Importantly, we identified that the thresholds used to define high GV significantly influenced the strength of the observed associations, highlighting the need to standardize GV assessment. Compared to traditional markers, such as HbA1c, GV may offer additional insight into short-term glycemic instability, which appears to contribute uniquely to cardiovascular pathology. However, the certainty of evidence for these associations was very low, with important heterogeneity and limited geographic representation. Our findings supports GV as a candidate prognostic marker that warrants further prospective validation before routine clinical integration. Future studies should investigate whether interventions aimed at reducing GV can significantly improve cardiovascular outcomes and whether such approaches should be tailored to patient-specific risk profiles and GV patterns. Supplementary Information Supplementary Material 1 Acknowledgements None. Author contributions [B.D.]: Conceptualization; Methodology; Formal analysis; Data Curation; Writing – Original Draft; Visualization; [**R.A.B.]: ** Conceptualization; Methodology; Formal analysis; Validation; Supervision; Writing – Review & Editing; Software; Project administration; **[N.K.]: ** Investigation; Data Curation; Writing – Original Draft; **[A.S.]: ** Investigation; Data Curation; Writing – Original Draft; **[S.M.T.]: ** Validation; Writing – Review & Editing; **[A.Sh.]: ** Methodology; Writing – Review & Editing; **[R.E.]: ** Investigation; Data Curation; Writing – Original Draft; **[A.B.]: ** Data Curation; Visualization; Writing – Original Draft; **[E.A.S.]: ** Investigation; Validation; Writing – Review & Editing; **[S.M.H.]: ** Investigation; Validation; Writing – Review & Editing; **[M.R.M.]: ** Conceptualization; Supervision; Writing – Review & Editing; **[P.R.T.]: ** Conceptualization; Methodology; Validation; Supervision; Writing – Review & Editing; **[G.U.]: ** Supervision; Conceptualization; Writing – Review & Editing. All authors have read and approved the final version of the manuscript and agree to be accountable for all aspects of the work. Funding No funding or sponsorship was received for this study or for the publication of this article. Data availability All extracted summary data used in this meta-analysis are available in the included publications and in the Supplementary Material. Declarations Ethics approval and consent to participate This article is based on previous studies and does not contain any new studies with human participants or animals performed by any of the authors. Consent for publication Not applicable. Competing interests The authors declare no competing interests. Footnotes Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. Bahar Darouei and Reza Amani-Beni contributed equally to this work and share first authorship. References Associated Data Supplementary Materials Supplementary Material 1 Data Availability Statement All extracted summary data used in this meta-analysis are available in the included publications and in the Supplementary Material.