Estimated glucose disposal rate and mortality in patients with cardiovascular disease: findings from two large cohorts

Abstract

Background

Cardiovascular disease (CVD) remains a leading cause of morbidity and mortality worldwide. Insulin resistance contributes to adverse cardiovascular outcomes but conventional assessments are complex and not routinely available. The estimated glucose disposal rate (eGDR), a simple surrogate of insulin resistance, may provide prognostic information in patients with established CVD.

Methods

Adults with CVD were identified from the National Health and Nutrition Examination Survey (NHANES) (n=4283) as the primary cohort and from a predefined UK Biobank (UKB) subcohort (n=31 460) as the replication cohort. Cox models, restricted cubic splines, Kaplan-Meier curves and subgroup analyses were used to evaluate associations between eGDR and mortality.

Results

During median follow-ups of 7.8 and 14.2 years, 1886 and 2218 deaths occurred in NHANES and the UKB, including 634 and 616 cardiovascular deaths, respectively. In fully adjusted models, higher continuous eGDR was associated with lower all-cause mortality in NHANES (HR, 0.859; 95% CI 0.800 to 0.923; p<0.001) and the UKB (HR, 0.792; 95% CI 0.743 to 0.844; p<0.001). Similar associations were observed for cardiovascular mortality in NHANES (HR, 0.844; 95% CI 0.736 to 0.969; p=0.016) and the UKB (HR, 0.737; 95% CI 0.656 to 0.828; p<0.001). Spline analyses showed non-linear inverse associations with both mortality outcomes in NHANES. In the UKB, the association with all-cause mortality became non-linear after multivariable adjustment, whereas the association with cardiovascular mortality was approximately linear.

Conclusion

Higher eGDR was independently associated with lower all-cause and cardiovascular mortality in patients with CVD, supporting its potential as a simple marker for mortality risk stratification.

WHAT IS ALREADY KNOWN ON THIS TOPIC

• Insulin resistance is a recognised driver of adverse cardiovascular outcomes, yet its routine clinical assessment is hampered by the complexity of available measures. Although the estimated glucose disposal rate (eGDR) has been proposed as a simple surrogate for insulin resistance, its prognostic value for mortality in a broad, generalisable cohort of patients with established cardiovascular disease (CVD) has not been systematically validated.

WHAT THIS STUDY ADDS?

• Using two large, independent cohorts, this study shows that higher eGDR is consistently associated with lower risks of all-cause and cardiovascular mortality among patients with CVD, with evidence of non-linear associations for selected outcomes.

HOW THIS STUDY MIGHT AFFECT RESEARCH, PRACTICE OR POLICY?

• Our findings support the adoption of eGDR as a simple, readily calculable tool for routine mortality risk stratification in patients with CVD, which could facilitate earlier identification of high-risk individuals in resource-limited settings without the need for specialised laboratory testing and may inform future interventional studies targeting insulin resistance in secondary cardiovascular prevention.

Introduction

Cardiovascular disease (CVD) remains the leading cause of mortality and a major contributor to disability worldwide, posing a substantial burden on global public health.^1–3 Stroke and ischaemic heart disease continue to be the primary contributors to the global CVD burden, as reflected by persistently high disability-adjusted life years.^{2 4} Population ageing and increasing life expectancy are expected to further increase the global burden of CVD, particularly among older adults, who bear a disproportionately high burden of cardiovascular morbidity and mortality.⁵ Therefore, better risk stratification among patients with established CVD is essential for identifying those at increased mortality risk and guiding individualised care.

In this context, insulin resistance (IR) has garnered increasing attention due to its strong association with CVD onset and adverse clinical outcomes.⁶ IR is a fundamental driver of various cardiometabolic diseases, including CVD and metabolic syndrome.^{6 7} Although the hyperinsulinemic-euglycemic clamp remains the gold standard for assessing IR,⁸ its complexity and cost limit its applicability in clinical and epidemiological settings. To address this, several surrogate indices have been developed—such as homeostasis model assessment as an index of IR (HOMA-IR).^{9 10} However, these methods often rely on fasting insulin or fasting glucose, which are not routinely measured and may be confounded by antidiabetic medications.

In recent years, the estimated glucose disposal rate (eGDR) has emerged as a practical and reliable surrogate for IR.¹¹ Unlike traditional indices, eGDR does not require insulin measurements and is derived from routine clinical variables—waist circumference, hypertension status and glycated haemoglobin A1c (HbA1c).¹¹ eGDR has demonstrated strong prognostic value for both microvascular complications^{12 13} and macrovascular outcomes.¹⁴ Despite these findings, evidence remains limited regarding the prognostic utility of eGDR in patients with CVD. To address this gap, we utilised data from the National Health and Nutrition Examination Survey (NHANES) and UK Biobank (UKB) to evaluate the predictive value of eGDR for mortality outcomes in CVD populations. Given its simplicity, accessibility and cost-effectiveness, eGDR holds strong potential as a clinical tool for risk stratification and long-term prognosis assessment in patients with CVD.

Methods

Data sources

Data for the present analysis were obtained from NHANES, conducted by the National Center for Health Statistics.¹⁵ All survey protocols were approved by the NCHS Research Ethics Review Board, and written informed consent was obtained from all participants. UKB is a large-scale, population-based cohort study across England, Scotland and Wales.¹⁶

Study population

From the NHANES 2001–2016 cycles, 65 474 participants were initially identified. Individuals younger than 18 years were excluded (n=9677), as were pregnant (n=1651) and lactating women (n=350). Among the remaining participants, 5017 individuals with CVD were identified. Participants with missing eGDR (n=731) or with missing follow-up data (n=3) were excluded, yielding 4283 individuals for final analyses. The UKB recruited approximately 500 000 participants overall. For the present replication analysis, we used a predefined UKB subcohort with available Patient Health Questionnaire-9. Within this subcohort, 33 677 participants had documented CVD. After excluding participants without valid eGDR measurements (n=2217), 31 460 participants with CVD were included in the final analysis.

Laboratory tests and clinical data

In NHANES, venous blood samples were collected from all participants, with a subset randomly selected to provide fasting samples. Complete blood counts were performed within 24 hours of collection to ensure analytical accuracy. Serum specimens were stored under appropriate frozen conditions and transported to certified contract laboratories for biochemical analysis.¹⁷ In UKB, blood samples were obtained at baseline at each assessment centre and transported in temperature-controlled containers to a central laboratory, where all haematological and biochemical assays were conducted.¹⁸

We included covariates across four domains: demographic and socioeconomic factors (age, gender, race, education level, marital status and poverty–income ratio (PIR) in NHANES or household income in UKB); lifestyle behaviours (body mass index (BMI), physical activity, smoking status and alcohol intake); medical history (diabetes, hypertension and cancer or malignancy) and laboratory parameters (white blood count, haemoglobin, platelet count, serum creatinine, blood urea nitrogen, uric acid, triglycerides, total cholesterol, high-density lipoprotein (HDL) cholesterol and low-density lipoprotein (LDL) cholesterol).

Definitions

The eGDR was calculated as previously reported using the following formula: eGDR (mg/kg/min)=21.158−(0.09×waist circumference (cm))−(3.407×hypertension)−0.551×HbA1 c (%), where hypertension was either 1 (presence of hypertension) or 0 (absence of hypertension).¹⁴ In NHANES, CVD was determined by a medical condition questionnaire, mainly by the affirmative answers to any of the following five questions: ‘Ever told had congestive heart failure?’, ‘Ever told you had coronary heart disease?’, ‘Ever told you had angina/angina pectoris?’, ‘Ever told you had heart attack?’ and ‘Ever told you had a stroke?’. If a positive answer was given to any of these questions, we considered the participant to have CVD.^{19 20} In UKB, CVD was determined by physician diagnoses or procedural codes (online supplemental table S1).²¹ Diabetes and hypertension were defined as the guidelines^{22 23} and the relevant International Classification of Diseases (ICD) codes in hospital records. Sociodemographic variables, including race, education level and marital status, were obtained through self-reported questionnaires in both NHANES and UKB. Lifestyle behaviours were defined based on self-reports in NHANES and on structured questionnaires or linked medical records in UKB, as detailed in online supplemental table S2. In NHANES, PIR was categorised as low (<1.3), moderate (≥1.3 to <3.5) and high (≥3.5), following established cut-offs.²⁴ In UKB, annual household income was categorised into three levels: low (<£18 000), medium (£18 000–£51 999) and high (≥£52 000).²⁵ Diagnoses of cancer or malignancy in NHANES were based on self-reported physician diagnoses. In the UKB, diagnoses of cancer were ascertained through linkage with hospital records using ICD codes. BMI was calculated using the formula: weight (kg)/ height (m)².

In NHANES, mortality status was determined using the publicly available Linked Mortality Files, updated through 31 December 2019. These files were created by probabilistically matching NHANES participants to death records in the National Death Index, based on a standardised algorithm developed by the National Center for Health Statistics. Causes of death were classified according to the ICD-10. Follow-up time was calculated from the date of the household interview to the date of death, loss to follow-up or the censoring date (31 December 2019).²⁰ In UKB, follow-up was defined as the time from baseline assessment to the occurrence of all-cause or cardiovascular death, loss to follow-up or end of follow-up. Dates and causes of death were obtained from national death registries. Cardiovascular mortality was defined using ICD-10 codes for diseases of the heart (I00–I09, I11, I13, I20–I51), essential hypertension and hypertensive renal disease (I10, I12, I15), cerebrovascular diseases (I60–I69) and peripheral vascular disease (I73).²⁶

Statistical analysis

For NHANES, all analyses incorporated sample weights and accounted for the complex survey design per official guidelines. For UKB, standard unweighted methods were used, as it does not employ a complex survey design. Continuous variable distributions were assessed via histograms and Q–Q plots. Missing covariates were handled using multiple imputation (m=5) and pooled using Rubin’s rules. Participants were categorised by eGDR quartiles (Q1–Q4). In NHANES, normally distributed continuous variables were summarised as mean±SD and compared using design-adjusted analysis of variance; non-normally distributed variables were presented as median (IQR) and compared using the design-adjusted Kruskal–Wallis test; categorical variables were compared using the Rao-Scott χ² test. Analogous unweighted tests were applied to UKB data.

The association between eGDR and all-cause and cardiovascular mortality was assessed separately in NHANES and UKB using three hierarchical models: Model 1 was unadjusted; model 2 was adjusted for age, gender, race, income level (PIR in NHANES; household income in UKB), education level, alcohol intake, smoking and marital status; model 3 further adjusted for diabetes, hypertension, cancer/malignancy, physical activity, BMI, white blood count, haemoglobin, platelet count, triglycerides, total cholesterol, HDL and LDL cholesterol, serum creatinine, blood urea nitrogen and uric acid. HRs with 95% CIs and p values were reported. Restricted cubic spline (RCS) models examined potential non-linear associations, with non-linearity tested by Wald tests. Subgroup analyses assessed effect modification by gender, age (≤60 vs >60 years), diabetes, hypertension, race, education, income, marital status, alcohol intake, smoking and cancer/malignancy, with interaction terms tested via Wald tests. A two-sided p<0.05 was considered significant. All analyses were performed using R V.4.3.2.

Results

Characteristics of research population

In the NHANES cohort, the analytical sample included 4283 participants, with a mean age of 65 years, and 54.7% were male. The eGDR was categorised into four quartiles: Q1 (≤4.2), Q2 (4.2–5.53), Q3 (5.53–7.20) and Q4 (7.20–12.5). Online supplemental table S3 presents the baseline characteristics of participants across these quartiles in NHANES. In the UKB cohort, 31 460 participants were included, with a mean age of 60 years, and 60.7% were male. The eGDR index was similarly categorised into four quartiles: Q1 (≤6.91), Q2 (6.91–7.73), Q3 (7.73–8.49) and Q4 (8.49–11.45). Online supplemental table S4 presents the baseline characteristics of participants across these quartiles in UKB.

Non-linear association between eGDR and mortality

In NHANES, over a median follow-up of 7.8 years (IQR 4.7–11.8), 1886 participants died, and 634 died of cardiovascular reasons. RCS analyses demonstrated a clear non-linear inverse association between eGDR and all-cause mortality across all models (p for non-linearity <0.001 in models 1–3) (figure 1A–C). A similar non-linear pattern was observed for cardiovascular mortality, with p values for the non-linear spline terms of 0.027, <0.001 and 0.006 in models 1, 2 and 3, respectively (figure 1D–F). In UKB, over a median follow-up of 14.2 years (IQR 13.5–15.0), 2218 participants died, including 616 cardiovascular deaths. The association between eGDR and all-cause mortality was approximately linear in the minimally adjusted model 1 (p for non-linearity=0.841), but became significantly non-linear in model 2 (p for non-linearity<0.001) and remained so in the fully adjusted model 3 (p for non-linearity=0.014) (figure 2A–C). By contrast, the association between eGDR and cardiovascular mortality in the UKB appeared predominantly linear, with non-significant p values for non-linearity in models 1–3 (0.615, 0.402 and 0.695, respectively) (figure 2D–F).

Figure 1.

RCS models illustrating the association between eGDR and all-cause mortality (A–C) and cardiovascular mortality (D–F) in NHANES.(A, D) Model 1: unadjusted. (B, E) Model 2: adjusted for age, gender, race, education level, income level, alcohol intake, smoking status and marital status. (C, F) Model 3: further adjusted for diabetes, hypertension, cancer or malignancy, physical activity, body mass index, white blood cell count, haemoglobin, platelet count, triglycerides, total cholesterol, high-density lipoprotein cholesterol, low-density lipoprotein cholesterol, serum creatinine, blood urea nitrogen and uric acid. eGDR, estimated glucose disposal rate; NHANES, National Health and Nutrition Examination Survey; RCS, restricted cubic splines.

Figure 2.

RCS models illustrating the association between eGDR and all-cause mortality (A–C) and cardiovascular mortality (D–F) in UKB. (A, D) Model 1: unadjusted. (B, E) Model 2: adjusted for age, gender, race, education level, income level, alcohol intake, smoking status and marital status. (C, F) Model 3: further adjusted for diabetes, hypertension, cancer or malignancy, physical activity, body mass index, white blood cell count, haemoglobin, platelet count, triglycerides, total cholesterol, high-density lipoprotein cholesterol, low-density lipoprotein cholesterol, serum creatinine, blood urea nitrogen and uric acid. eGDR, estimated glucose disposal rate; RCS, restricted cubic splines; UKB, UK Biobank.

Correlation between eGDR and mortality

As shown in table 1, higher eGDR was generally associated with lower all-cause and cardiovascular mortality in both NHANES and UKB, particularly when analysed as a continuous variable. In the fully adjusted model, continuous eGDR remained significantly inversely associated with all-cause mortality in NHANES (HR, 0.859; 95% CI 0.800 to 0.923; p<0.001) and UKB (HR, 0.792; 95% CI 0.743 to 0.844; p<0.001), with similar associations for cardiovascular mortality in NHANES (HR, 0.844; 95% CI 0.736 to 0.969; p=0.016) and UKB (HR, 0.737; 95% CI 0.656 to 0.828; p<0.001). In UKB quartile-based analyses, inverse associations were more evident for all-cause mortality, whereas associations with cardiovascular mortality were attenuated after full adjustment. Kaplan-Meier curves confirmed similar inverse associations in both NHANES (online supplemental figure S1) and UKB (online supplemental figure S2).

Table 1.

Associations between eGDR and mortality among participants with CVD in NHANES and the UK Biobank

	NHANES						UKB
	Model 1		Model 2		Model 3		Model 1		Model 2		Model 3
	HR (95% CI)	P value	HR (95% CI)	P value	HR (95% CI)	P value	HR (95% CI)	P value	HR (95% CI)	P value	HR (95% CI)	P value
All cause death
eGDR (continuous)	0.937 (0.910 to 0.965)	<0.001	0.954 (0.923 to 0.986)	0.005	0.859 (0.800 to 0.923)	<0.001	0.766 (0.744 to 0.789)	<0.001	0.792 (0.766 to 0.818)	<0.001	0.792 (0.743 to 0.844)	<0.001
Q1	Ref		Ref		Ref		Ref		Ref		Ref
Q2	0.864 (0.676 to 1.103)	0.24	0.687 (0.549 to 0.86)	0.001	0.692 (0.551 to 0.869)	0.001	0.687 (0.618 to 0.765)	<0.001	0.694 (0.623 to 0.772)	<0.001	0.851 (0.751 to 0.964)	0.011
Q3	0.886 (0.728 to 1.077)	0.224	0.724 (0.606 to 0.865)	<0.001	0.787 (0.615 to 0.907)	0.037	0.558 (0.498 to 0.625)	<0.001	0.600 (0.535 to 0.673)	<0.001	0.803 (0.691 to 0.933)	0.004
Q4	0.708 (0.564 to 0.888)	0.002	0.787 (0.639 to 0.969)	0.024	0.908 (0.636 to 1.296)	0.595	0.439 (0.389 to 0.497)	<0.001	0.569 (0.499 to 0.649)	<0.001	0.826 (0.681 to 1.002)	0.052
P for trend	0.903 (0.847 to 0.964)	0.002	0.94 (0.881 to 0.983)	0.041	0.954 (0.853 to 1.067)	0.411	0.759 (0.730 to 0.789)	<0.001	0.814 (0.781 to 0.849)	<0.001	0.932 (0.877 to 0.982)	0.028
Cardiovascular death
eGDR (continuous)	0.931 (0.894 to 0.969)	<0.001	0.940 (0.894 to 0.990)	0.019	0.844 (0.736 to 0.969)	0.016	0.679 (0.646 to 0.715)	<0.001	0.711 (0.672 to 0.752)	<0.001	0.737 (0.656 to 0.828)	<0.001
Q1	Ref		Ref		Ref		Ref		Ref		Ref
Q2	0.916 (0.65 to 1.292)	0.617	0.713 (0.509 to 0.993)	0.048	0.71 (0.488 to 1.034)	0.074	0.688 (0.568 to 0.833)	<0.001	0.703 (0.580 to 0.852)	<0.001	1.004 (0.800 to 1.260)	0.973
Q3	0.882 (0.655 to 1.189)	0.410	0.706 (0.524 to 0.95)	0.021	0.764 (0.498 to 1.174)	0.219	0.416 (0.333 to 0.521)	<0.001	0.469 (0.374 to 0.588)	<0.001	0.787 (0.590 to 1.051)	0.104
Q4	0.675 (0.488 to 0.932)	0.017	0.729 (0.522 to 0.917)	0.032	0.757 (0.395 to 1.45)	0.401	0.284 (0.219 to 0.367)	<0.001	0.436 (0.332 to 0.573)	<0.001	0.873 (0.600 to 1.269)	0.476
P for trend	0.885 (0.807 to 0.971)	0.010	0.911 (0.822 to 0.979)	0.047	0.912 (0.748 to 1.111)	0.359	0.653 (0.605 to 0.704)	<0.001	0.729 (0.672 to 0.791)	<0.001	0.924 (0.822 to 1.039)	0.190

Model 1: unadjusted.

Model 2: adjusted for age, gender, race, education level, income level, alcohol intake, smoking status and marital status.

Model 3: further adjusted for diabetes, hypertension, cancer or malignancy, physical activity, body mass index, white blood cell count, haemoglobin, platelet count, triglycerides, total cholesterol, high-density lipoprotein cholesterol, low-density lipoprotein cholesterol, serum creatinine, blood urea nitrogen and uric acid.

95%CI, 95%confidence interval; CVD, cardiovascular diseases; eGDR, estimated glucose disposal rate; HR, hazard ratio; NHANES, National Health and Nutrition Examination Survey; UKB, UK biobank.

Subgroup analysis

In NHANES, higher eGDR was associated with lower all-cause mortality across most subgroups (figure 3A). Significant interactions were found for age, diabetes, hypertension, race, smoking and cancer history (all p for interaction <0.05). For cardiovascular mortality (figure 3B), significant interactions were observed for diabetes (p<0.001), hypertension (p=0.039) and PIR (p=0.016). In UKB, the inverse association with all-cause mortality was observed in most subgroups (figure 4A), with significant interactions for gender (p=0.044), age (p=0.004) and hypertension (p=0.022). For cardiovascular mortality (figure 4B), significant interactions were found for hypertension (p=0.006) and education (p=0.040).

Figure 3.

Association between eGDR and mortality across subgroups in patients with CVD in NHANES. (A) All-cause mortality. (B) Cardiovascular mortality. CVD, cardiovascular diseases; eGDR, estimated glucose disposal rate; NHANES, National Health and Nutrition Examination Survey; PIR, poverty–income ratio.

Figure 4.

Association between eGDR and mortality across subgroups in patients with CVD in UKB. (A) All-cause mortality. (B) Cardiovascular mortality. CVD, cardiovascular diseases; eGDR, estimated glucose disposal rate; PIR, poverty–income ratio; UKB, UK Biobank.

Discussion

This study evaluated the prognostic value of eGDR in patients with CVD using NHANES and UKB data. Higher eGDR was consistently associated with lower all-cause and cardiovascular mortality across most subgroups. The consistent predictive value observed in both cohorts underscores eGDR’s generalisability and robustness as a risk stratification tool. These findings highlight eGDR’s clinical utility as a simple, accessible and cost-effective marker for long-term prognosis and treatment guidance in patients with CVD, with potential to enhance personalised care.

Although the mechanisms linking eGDR to mortality in patients with established CVD are not fully clarified, IR—reflected by lower eGDR—is widely recognised as a key contributor to adverse prognosis. IR promotes chronic inflammation, with elevated cytokines damaging the endothelium and impairing vasoregulation.^{27 28} Downstream effects of IR also include oxidative stress and endothelial dysfunction, with excessive reactive oxygen species impairing nitric oxide signalling and promoting vasoconstriction and thrombosis.^{27 29 30} Lipid metabolism disturbances may also influence outcomes. In our study, higher eGDR was associated with lower triglycerides, which was associated with previous studies,³¹ and prior studies suggest that triglyceride-rich lipoproteins may contribute to residual cardiovascular risk, particularly in older adults.^{32 33} At the myocardial level, IR impairs substrate utilisation and reduces myocardial energetic efficiency, potentially contributing to cardiac dysfunction.³⁴ Taken together, these findings support a multifaceted pathophysiological framework linking lower eGDR to elevated mortality in individuals with CVD. eGDR may, therefore, serve as a valuable prognostic marker for identifying high-risk patients in clinical practice.

Previous studies across diverse populations have demonstrated a consistent inverse association between eGDR and mortality. Evidence from cohorts of adults with cardiovascular-kidney-metabolic syndrome³⁵ has consistently shown an inverse, predominantly linear relationship between eGDR and adverse outcomes. Similar linear dose–response associations were reported in NHANES and other general population studies.^{36 37} However, some studies have reported potential non-linear associations with all-cause mortality while maintaining linear trends for cardiovascular mortality,³⁸ highlighting an ongoing debate about the exact shape of the eGDR–mortality relationship. Our study clarifies the eGDR–mortality association in two large, ethnically diverse cohorts with established CVD. In NHANES, we found a robust non-linear inverse relationship between eGDR and all-cause mortality. In UKB, the association was approximately linear in the minimally adjusted model but became significantly non-linear after further adjustment. For cardiovascular mortality, the association was non-linear across models in NHANES but remained linear in UKB. These findings suggest that while higher eGDR consistently predicts lower mortality, the shape of this relationship may vary across populations and outcomes. Our evidence suggests that non-linearity is more likely to appear in US populations and primarily for all-cause mortality, whereas cardiovascular mortality may follow a more linear pattern in certain populations, consistent with prior studies.^{36 38} These differences may reflect population heterogeneity or varying cardiometabolic burdens. By integrating two independent cohorts with detailed spline analyses, our study extends existing evidence by revealing both the generalisability and the nuanced shape of the eGDR–mortality association in patients with CVD. The consistent inverse association reinforces eGDR’s prognostic value, while the observed non-linearities underscore the need for population-specific considerations in clinical risk stratification. Notably, some quartile-based associations were attenuated after full adjustment, particularly for cardiovascular mortality, possibly because of reduced statistical efficiency after categorisation, fewer events within individual quartiles, and adjustment for cardiometabolic factors closely related to eGDR. These findings suggest that continuous eGDR may better capture the underlying risk gradient than arbitrary categorical cut-offs. Future studies are warranted to determine clinically meaningful eGDR thresholds and to evaluate its incremental value for cardiovascular risk stratification.

In our study, significant interaction effects between eGDR and age were observed, particularly for all-cause mortality. In both NHANES and UKB, the inverse association between eGDR and all-cause mortality was more pronounced among younger individuals (≤60 years). These findings are consistent with the results of previous research.³⁹ Lower eGDR in younger individuals may indicate early-onset IR, which results in prolonged exposure to metabolic stress, thereby accelerating vascular injury and increasing mortality risk.³⁹ In contrast, older individuals often accumulate competing mortality risks (eg, cancer, neurodegeneration), which may dilute the relative contribution of IR and weaken eGDR’s predictive capacity. This pattern aligns with a previous large-scale study using CHARLS and NHANES, where eGDR showed a protective effect primarily in younger adults with cardiovascular–kidney–metabolic syndrome.³⁵ The authors suggested that younger individuals, with greater metabolic reserve, may benefit more from improved insulin sensitivity. Our findings support this and further demonstrate that age-related modification of the eGDR–mortality relationship is robust across ethnically and geographically distinct cohorts, underscoring the importance of considering age when using eGDR as a prognostic biomarker.

In our study, the eGDR–mortality association varied significantly by diabetes status, especially in NHANES. Among non-diabetic individuals, higher eGDR was associated with lower all-cause and cardiovascular mortality. In those with diabetes, however, the associations were null or even reversed, particularly for cardiovascular mortality in NHANES. This may reflect that higher eGDR in non-diabetic individuals typically indicates better insulin sensitivity and a healthier metabolic profile, whereas in diabetic patients, it may paradoxically signal catabolic states or intensive treatment effects that increase vulnerability to adverse outcomes. This observation aligns with prior research.⁴⁰ Li et al⁴⁰ found that the inverse association between eGDR and CVD mortality was stronger in non-diabetic individuals with atherosclerotic CVD, and interaction by diabetes status was significant. One possible explanation is that glucose-lowering medications used in diabetes, particularly insulin sensitisers, may mitigate the adverse effects of IR. A meta-analysis of nine clinical trials reported that pioglitazone reduced major adverse cardiovascular events in patients with IR or type 2 diabetes.⁴¹ Other studies have also suggested that the mechanisms of atherosclerosis progression differ between diabetic and non-diabetic patients, which may partly underlie differences in the prognostic impact of IR.⁴² Additionally, in patients with diabetes, the management of glycaemic status may play a more dominant role in determining outcomes than IR alone.⁴³ This could further attenuate the association between eGDR and mortality in diabetic populations.

Although both cohorts showed inverse associations between eGDR and mortality, the shape of the associations and some subgroup interactions differed. These discrepancies may stem from several factors⁴⁴: differences in baseline characteristics and eGDR distribution; longer follow-up in UKB, which may amplify competing risks; and unmeasured confounders such as medication use, healthcare access and lifestyle. These contextual factors highlight the need to interpret eGDR-based risk predictions with consideration of population and healthcare system characteristics.

Several limitations should be acknowledged. First, eGDR was assessed only at baseline, precluding evaluation of longitudinal changes in IR. Second, despite adjusting for multiple covariates, residual confounding (eg, diet, medication adherence, genetics) cannot be excluded. Third, mortality rates differed substantially between NHANES and UKB, likely reflecting differences in cohort characteristics, sampling frameworks and follow-up. These differences may limit direct quantitative comparisons between cohorts. Nevertheless, the consistent direction of the eGDR–mortality association supports the robustness of our findings. Fourth, the UKB replication analysis was based on a predefined subcohort with available Patient Health Questionnaire-9, which may have introduced selection bias and may limit the generalisability of the findings to the broader CVD population. Future studies in more diverse settings with longitudinal eGDR trajectories are needed to confirm and extend these findings.

Conclusions

In conclusion, eGDR is consistently negatively associated with both all-cause and cardiovascular mortality in patients with CVD across two large cohorts, NHANES and UKB. The negative relationship between eGDR and mortality remained robust in most subgroup analyses, further supporting its reliability. These findings highlight eGDR’s potential as a simple, cost-effective tool for risk stratification and personalised care in patients with CVD, with consistent predictive value across diverse populations.

Data availability statement

Data are available in a public, open access repository. This study was conducted using data from two publicly available sources: the UK Biobank (Application Number 103409; www.ukbiobank.ac.uk) and the National Health and Nutrition Examination Survey (NHANES; www.cdc.gov/nchs/nhanes/).

Ethics statements

Patient consent for publication

Consent obtained directly from patients.

Ethics approval

This study is officially registered with the UK Biobanking Resource Center under application number 103409. The studies involving human participants were reviewed and approved by the NCHS Research Ethics Review Board and the North West Multicenter Research Ethics Committee (REC reference for UK Biobank 11/NW/0382). Participants gave informed consent to participate in the study before taking part.