This study was conducted according to the Meta-analysis of Observational Studies in Epidemiology (MOOSE) Statement (10) and PRISMA 2009 statement (11). The study selection process is shown in the PRISMA (Preferred Reporting Items for Systematic Reviews and Meta-Analyses) flow diagram (Figure 1). Electronic databases such as PubMed, Web of Science and Embase were searched in accordance with combined terms: (1) “estimated glucose disposal rate” OR “eGDR” and (2) “cardiovascular” OR “peripheral arterial disease” OR “coronary artery disease” OR “stroke” OR “cardiovascular disease” OR “coronary heart disease” OR” ischemic stroke” OR “CHD” OR “CVD” OR “CAD” OR “PAD” OR “IS”. The search strategy used filters to select studies involving humans and available in English. The search for final literature concluded on March 28, 2025, and the specific strategy is detailed in the Supplementary material Table 1. The protocol for this study has been registered in the International Prospective Register of Systematic Reviews (PROSPERO 2025 CRD 420251147324. Available from https://www.crd.york.ac.uk/PROSPERO/view/CRD420251147324).

The meta-analysis included studies based on these criteria: (1) The studies were cohort studies published in full. (2) eGDR was assessed at the start. (3) Participants were adults without baseline CVD. (4) Outcomes included new cases of CVD, CHD, or stroke. (5) hazard ratio (HR) [95% confidence interval (CI)] values were given. (6) The studies adjusted risk factors for possible confounders. Studies were excluded if: (1) Participants were under 18. (2) The population had baseline CVD. (3) eGDR wasn’t measured. (4) The study wasn’t a cohort design. (5) HR (95%CI) values weren’t reported.

The eGDR was calculated as (mg/kg/min) 21.158-(0.09*WC) - (3.407*HT) - (0.551*HbA1c) [WC = waist circumference (cm), HT = hypertension (yes = 1/no = 0), and HbA1c = HbA1c (% DCCT)] (12). Hypertension was defined as systolic blood pressure  ≥ 140mmHg and/or diastolic blood pressure  ≥ 90mmHg, self-reported history of hypertension, or current use of prescribed medicine for HT. HbA1c was measured by the high-performance liquid chromatography method. Furthermore, the unit in mmol/mol was transformed to a percentage (%) using the equation: (0.09148 × HbA1c mmol/mol) + 2.152 (13). CVD was identified based on self-reported physician’s diagnosis or the International Classification of Diseases (ICD-10). The main outcome was CVD, either alone or within a composite. Secondary outcomes were CHD and stroke. CAD involved chronic ischemic heart disease, angina and acute myocardial infarction. Stroke cases included both ischemic and hemorrhagic types.

Articles from Embase, Web of Science and PubMed were transferred to EndNote X9. Duplicates were identified and removed using the “duplicate identifier” function. Titles and abstracts were initially screened and categorized as potentially eligible, uncertain eligibility, or clearly ineligible. For potentially eligible or uncertain articles, full - text reviews were conducted to assess their final eligibility against the inclusion and exclusion criteria.

Two researchers (Zhijun Zhang and Beibei Wang) separately performed information extraction from the articles. The data extracted included the following: (1) publication year, the first author’s name, and country; (2) the study design/mean follow-up time; (3) participant characteristics, such as the mean age, sample size, proportion of participants with prediabetes and diabetes, and proportion of male participants; (4) the analysis model used for the eGDR index; (5) the reported endpoint outcomes; and (6) the covariates controlled for within the multivariate analysis. Following the extraction process, the investigators cross-checked the data to verify its accuracy. Any discrepancies were resolved by consulting a third researcher (Lijuan Qu), whose judgment was accepted as the final decision.

The quality of each study was assessed using the Newcastle–Ottawa Scale (14). This scale evaluates the quality of cohort studies in three aspects: study selection, comparability between groups, and outcome assessment, with scores ranging from 1 to 9 points. Studies with a score of over 6 points were regarded as high-quality.

The hazard ratio (HR) and 95% confidence interval (CI) were utilized as a general measure to indicate the association between the eGDR and CVD risk in individuals without baseline CVD. For studies with the eGDR analyzed as a continuous variable, the HR (95% CI) of CVD risk per 1-unit increment of the eGDR was extracted. For studies that categorized the eGDR, the HR (95% CI) for CVD risk comparing individuals with the highest levels to those with the lowest levels of the eGDR was extracted. The heterogeneity among the included cohort studies was evaluated using the I² statistic (15). If the I² value exceeded 50%, it indicated significant heterogeneity and a random effects model was employed to pool the HR (95% CI) data; otherwise, a fixed-effects model was used for analysis. Moreover, the robustness of the results was assessed through sensitivity analyses executed by excluding each study once at a time (16). Predefined subgroup analyses were conducted to assess how study characteristics such as males sex (%), prediabetes/diabetes state (%), sample size, and mean follow-up time might influence the correlation between the eGDR and CVD risk. The potential for publication bias was initially evaluated by visually examining the symmetry of funnel plots (17). Subsequently, the trim-and-fill method, along with Egger’s (18) and Begg’s (19) tests, was employed as quantitative methods to further assess publication bias.

We computed the linear trends and 95% CI by applying the natural logarithm of the effect sizes and the 95% CI for the eGDR categories, in accordance with the method described by Greenland and Longnecker (20). Nonlinear dose - response analyses were performed via robust error meta-regression., as per the approach described by Ma and Xu et al. (21, 22) The sample fitting process was conducted in two stages. Initially, a dose–response analysis was conducted separately for each study. Subsequently, the dose–response data from these individual studies were integrated using a random-effects model. This model necessitates information on the known levels of eGDR, the natural logarithm of the HR, the number of cases and the person-year (calculated by multiplying the average follow-up time by the number of cases) within each exposure range (22). When quantitative eGDR values were unavailable, missing values were imputed using the method detailed by Xu et al. (21). This approach allows for the use of either the exposure median or mean. In cases where neither the mean nor median is provided but a range of values is reported, the exposure level can be approximated as follows: for closed intervals, the midpoint between the upper and lower bounds is used; for open intervals, the interval length is inferred from the adjacent group, and the midpoint of this interval is taken as the average exposure level (23). The meta-analysis and statistical analysis were performed using R (4.2.2) software. A p-value less than 0.05 or 95% CI excluding 1 was regarded as statistically significant.

As depicted in Figure 1, the database search process was conducted systematically. A total of 614 articles were retrieved from the initial search of Web of Science, PubMed and Embase databases, following the removal of duplicate entries. During the preliminary screening of titles and abstracts, 571 articles were deemed irrelevant and excluded. Subsequently, 31 articles were further excluded based on the criteria outlined in Supplementary material Table 2. In the end, twelve cohort studies (24–35) were finalized for inclusion in the meta-analysis.

The characteristics of the twelve cohort studies are shown in Supplementary material Table 3. Twelve cohort studies with 547,287 subjects were included, with follow-up durations ranging from 5.6 to 14.1 years. These studies were carried out in China, Sweden, England, and the United States of America. All studies were cohort studies and published between 2022 and 2024.The average age of the participants across the twelve studies spanned from 56.3 to 62.9 years. Two studies had a male participant proportion exceeding 50%, while the remaining studies featured a male participant proportion under 50%. Additionally, three articles featured sample sizes exceeding 10,000, whereas nine articles reported sample sizes below 10,000.

This meta-analysis included twelve cohort studies. Their quality was evaluated via the Newcastle–Ottawa Scale, where the highest possible score is 8. The assessment showed that three studies achieved a score of 7, while the remaining seven studies scored 8. Thus, all included cohort studies were deemed high-quality (Supplementary material Table 4).

For CVD analysis, a total of nine cohorts (25–29, 31, 33–35)were examined, covering 73,577 participants. A random-effects model was used, with eGDR being treated as a categorical variable. The pooled results from the nine cohorts demonstrated that participants in the highest eGDR category experienced a significantly reduced CVD risk compared to those in the lowest eGDR category at baseline (HR = 0.58; 95% CI 0.53–0.63; I² = 52.4%; Figure 2A). This matched the meta-analysis of eGDR as a continuous variable, with a 12% (HR: 0.88, 95% CI 0.85–0.91, I²=77.4%) reduction in CVD risk for every 1-unit increase (Figure 2A). Dose-response curves treating eGDR as a categorical variable showed a negative linear relationship was observed between the eGDR and CVD risk (P_nonlinear = 0.120) (Figure 3A). Supplementary material Table 5 presents estimates for the linear exposure effect analysis for eGDR.

In subgroup evaluations, those in the highest eGDR group showed a significantly lower CVD risk than the lowest group, and this finding was consistent regardless of male sex (%), prediabetes/diabetes statue (%), sample size, and mean follow-up time (P > 0.05 for each subgroup; Table 1). Omitting one study at a time in sensitivity analyses generated analogous results (HR for the eGDR analyzed as a categorical variable: 0.57–0.60, all P<0.05) (Supplementary material Figure 1A). Funnel plots exhibited symmetrical features on visual inspection when eGDR was evaluated as a categorical variable, implying a low propensity for publication bias (Supplementary Figure 1B). The P values were 0.750 and 0.835 based on Begg’s and Egger’s regression, respectively, further suggesting no publication bias (Supplementary material Figures 2A, B). After incorporating one study using the trim-and-fill method, the HR (95% CI) remained largely unchanged, indicating that the combined effect size results are reliable (Supplementary material Figures 3A, B).

Nine cohorts (25–30, 32–34)were included in the stroke analysis. The combined findings showed that higher eGDR corresponded to a lower stroke risk when comparing the highest and lowest categories (HR = 0.62; 95% CI 0.56–0.69; I²=24.8%) (Figure 2B). These findings were aligned with the meta-analysis of eGDR as a continuous variable, with a 16% (HR: 0.84, 95% CI 0.76–0.93, I²=83.0%) reduction in stroke risk for every 1-unit increase (Figure 2B). Dose-response analysis treating eGDR as a categorical variable showed a negative linear correlation between eGDR and stroke risk (P_nonlinear=0.084) (Figure 3B). Supplementary material Table 5 presents HR(95%CI) estimates for the linear exposure effect analysis for the eGDR. Omitting one study at a time in sensitivity analyses generated analogous results (HR for the eGDR analyzed as a categorical variable: 0.59–0.65, all P<0.05) (Supplementary material Figure 1C). When eGDR was analyzed as a categorical variable, the funnel plots appeared symmetric upon visual assessment, indicating a minimal risk of publication bias. (Supplementary material Figure 1D). The P values were 0.061 and 0.181 based on Begg’s and Egger's regression, respectively (Supplementary material Figures 2C, D). The trim-and-fill method was used to add three studies, and the HR (95% CI) did not change significantly, indicating that the combined effect size results were robust (Supplementary material Figures 3C, D).

Two cohorts (24, 30)were included in the ischemic stroke analysis. The aggregated data showed that participants in the highest eGDR group exhibited a decreased risk of ischemic stroke risk (HR = 0.64; 95% CI 0.49–0.85; I²=68%) when compared with those in the lowest eGDR group (Figure 2C). This results was in line with the Huan et al. (24) treating eGDR as a continuous variable, with a 13% (HR: 0.87, 95% CI 0.83–0.90) reduction in ischemic stroke risk for every 1-unit increase (Supplementary material Table 1).

The pooled estimates from two cohort studies (27, 28) indicated that higher eGDR was related to a lower CHD risk when treated as a categorical variable (HR = 0.46; 95% CI 0.25–0.83; I²=88.8%), and consistent results were found when eGDR was treated as a continuous variable (HR = 0.85; 95% CI 0.83–0.87; I²=49.6%) (Figure 2D). Similarly, higher eGDR was correlated with a reduced myocardial infarction risk in both categorical (HR = 0.51; 95% CI 0.33–0.78; I² = 57.5%) and continuous analyses (HR = 0.82; 95% CI 0.68–0.99; I² = 69.8%) (Figure 2E).

As far as we are aware, this is the first meta-analysis to explore the association between eGDR and CVD risk in a general population that was free from CVD at baseline, which allows for a more accurate assessment of the predictive value of eGDR in individuals without pre-existing CVD. By exclusively including cohort studies, we circumvented potential recall bias inherent in cross-sectional designs, thereby strengthening the causal inference in the observed association. Moreover, we conducted separate meta-analyses treating eGDR as both a categorical and a continuous variable, a methodological approach that provided complementary perspectives and further validated the robustness of our primary conclusions. Despite the presence of significant heterogeneity, we conducted sensitivity, dose–response, and subgroup analyses, which yielded robust and reliable results. Additionally, cohort studies released in the past three years, all of our selected studies were of high quality and they featured extensive sample sizes along with lengthy follow-up intervals.

Although this meta-analysis has notable strengths and potential clinical relevance, several limitations exist that warrant consideration when evaluating the results. First, despite the use of random-effects models to account for heterogeneity, significant heterogeneity was still observed in some analyses. This heterogeneity may stem from differences in participants’ races and comorbidities, as well as methods for measuring eGDR and CVD outcomes. Second, the potential for residual confounding, such as dietary patterns (46, 47), sleep quality and sleep duration factors (48), and liver fibrosis (49) may have an impact on eGDR, in the original cohort studies cannot be fully excluded, although most studies adjusted for multiple potential confounders. Third, most studies were from China and only three from Sweden/USA, limiting ethnic diversity and generalizability to global cardiovascular risk patterns. Future research must validate eGDR cutoffs and predictive accuracy in diverse Western and multi-ethnic cohorts. Fourth, it is important to note that the observational nature of the included studies limits our ability to establish causality between eGDR and cardiovascular outcomes. While the findings suggest a strong association, confounding factors and reverse causation cannot be entirely ruled out. Finally, eGDR formula was originally validated using clamp studies in type 1 diabetes populations, and its metabolic validity in different populations has not been directly established. However, this meta-analysis aims to evaluate predictive validity rather than metabolic validity, and the potential non-differential misclassification bias is more likely to make our risk estimation conservative than exaggerated.

Our meta-analysis reveals that higher eGDR is associated with a significantly lower risk of CVD, stroke, and CHD. This indicates that eGDR could serve as a valuable marker for predicting CVD risk in individuals without baseline CVD. Future research should focus on further exploring the underlying mechanisms and assessing the predictive power of eGDR in diverse populations.