Estimated glucose disposal rate and the dynamic transitions of cardiometabolic multimorbidity and all-cause mortality in prediabetes: a multistate trajectory analysis

Abstract

Aim

To examine the association between insulin sensitivity, measured by estimated glucose disposal rate (eGDR), and the onset, progression, and prognosis of cardiometabolic multimorbidity (CMM) and all-cause mortality.

Methods

This prospective study included 50,342 UK Biobank participants with prediabetes and free of baseline cardiometabolic disease (CMD). CMM was defined as the coexistence of at least two of the following: coronary artery disease, type 2 diabetes, and stroke. eGDR was calculated according to a previously published formula based on waist circumference, HbA1c, and hypertension status.A multistate Markov model assessed associations between eGDR and sequential transitions from CMD onset to CMM and death, accounting for competing risks and nonlinearity. Nonlinear dose–response relationships were assessed using penalized splines with three degrees of freedom.

Results

Over a median follow-up of 13.6 years, 12,641 participants developed a first CMD (FCMD), 2,081 progressed to CMM, and 4,847 died. Higher eGDR was associated with lower risks of FCMD (HR for per unit increase: 0.88; 95% CI: 0.87–0.89), CMM (HR: 0.93; 95% CI: 0.91–0.96), and death before FCMD (HR: 0.95; 95% CI: 0.93–0.98). No significant association was observed with mortality after CMM onset. Nonlinear associations were observed between eGDR and CMM progression across multiple transition states.

Conclusions

In individuals with prediabetes, higher insulin sensitivity is associated with a reduced risk of CMD, CMM, and all-cause mortality. Once the first CMD occurs, the association between eGDR and mortality becomes less pronounced, highlighting the critical importance of early intervention to improve insulin sensitivity.

Graphical Abstract

Supplementary Information

The online version contains supplementary material available at 10.1186/s12944-025-02699-z.

Introduction

Cardiometabolic diseases (CMDs), including coronary artery disease (CAD), type 2 diabetes mellitus (T2DM), and stroke, represent a significant global health challenge [1]. According to the World Health Organization (WHO), cardiovascular diseases (CVDs) are responsible for approximately 17.9 million deaths annually, constituting 32% of global mortality, with 80% of these deaths attributable to CAD and stroke [2]. Concurrently, the global prevalence of T2DM has reached 537 million cases, with projections estimating over 700 million by 2045 [3]. These conditions frequently coexist, and factors such as population aging, sedentary lifestyles, and rising obesity rates are accelerating the development of cardiometabolic multimorbidity (CMM) [4]. Epidemiological evidence shows that individuals with CMM face a 3- to 7-fold increased risk of all-cause mortality compared to those with single CMDs, alongside significantly elevated healthcare costs as comorbidities accumulate [5]. Specifically, having two cardiometabolic conditions reduces life expectancy by approximately 12 years at age 60, increasing to about 15 years with three conditions [5].

The central pathophysiological mechanism underlying CMM is insulin resistance (IR), a systemic process that connects glucose-lipid dysregulation to accelerated atherosclerosis, endothelial dysfunction, and chronic inflammation [6–8]. Recently, clinical assessment tools such as the estimated glucose disposal rate (eGDR) have emerged, providing practical and non-invasive methods for evaluating insulin sensitivity [9, 10]. Higher eGDR values indicate better insulin sensitivity and lower insulin resistance. Emerging evidence suggests that the eGDR index serves as a superior surrogate marker of insulin sensitivity compared with homeostasis model assessment of insulin resistance, quantitative insulin sensitivity check index, fasting insulin resistance index [9]. By incorporating measures of abdominal obesity, hypertension, and glycemic control, eGDR demonstrates superior predictive capability for cardiovascular mortality compared to traditional indices such as the triglyceride-glucose index, with reported areas under the curve ranging from 0.78 to 0.82 versus 0.60–0.68, respectively [11]. The robust predictive power of eGDR likely arises from its simultaneous assessment of vascular damage and glucotoxicity, highlighting its role as a comprehensive indicator of insulin resistance [12, 13].

Prediabetes, affecting approximately 352 million individuals globally, presents a strategic window for intervention [14]. Early preventive measures at this stage are essential, as timely interventions can significantly reduce progression to overt T2DM and lower the associated cardiovascular burden. Individuals with prediabetes, although not meeting the diagnostic thresholds for T2DM, exhibit early insulin resistance and metabolic dysfunction, conferring an approximately 17% increased risk for CVDs [15]. Lifestyle interventions have proven highly effective, preventing progression to diabetes in up to 93% of individuals, with 43% achieving a return to normoglycemia [16]. Nonetheless, current studies on CMM primarily focus on static clinical endpoints, typically evaluating the presence or absence of CMM at a single time point or assessing time-to-first-event without considering disease sequencing or dynamic progression [17, 18]. For instance, while individuals may ultimately develop T2DM, the sequence of preceding or subsequent events—such as stroke or CAD—can vary substantially. These divergent trajectories suggest that the role of insulin sensitivity may differ across distinct stages and pathways of cardiometabolic disease progression, yet conventional analytical methods often fail to capture such complexity over time. To address these gaps, our study employs multistate modeling to: (1) investigate the role of insulin sensitivity in altering disease trajectories; (2) quantify transition probabilities among distinct metabolic states;3) assess the influence of the type of first cardiometabolic disease (FCMD) on the progression to CMM and incident mortality. Our findings will provide novel insights into the dynamic progression of CMM, informing targeted and timely intervention strategies.

Method and study design

The UK Biobank is a population-based prospective cohort study designed to evaluate the associations between lifestyle, genetic factors, and health outcomes, as previously described [19]. Between 2006 and 2010, more than 500,000 participants aged 40 to 69 years were enrolled from 22 assessment centers across the United Kingdom. Baseline data collection included electronic informed consent, self-administered touchscreen questionnaires, computer-assisted interviews, physical measurements, and biological sample collections. Mortality data were linked using National Health Service identifiers.

Baseline CMDs was determined from self-reported, medical history and hospital records. Initially, 415,112 participants without known CAD, T2DM, or stroke, and with complete eGDR data were identified. The diagnostic details are provided in Table S1. Pregnant women (n = 111) and individuals with cancer (n = 39,068) were excluded to minimize potential confounding effects from acute metabolic changes. The final analytic cohort comprised 50,342 individuals with prediabetes, defined according to the 2022 American Diabetes Association criteria (HbA1c 5.7–6.4% [39–47 mmol/mol]) [20].

The UK Biobank study was approved by the North West Multi-Centre Research Ethics Committee (Reference: 11/NW/0382). This research was conducted under UK Biobank application number 107,335. The study adhered to the principles outlined in the Declaration of Helsinki and complied with the Strengthening the Reporting of Observational Studies in Epidemiology guidelines.

Assessment of eGDR

eGDR (mg/kg/min) was calculated using a previously validated equation: [21]

where waist circumference is measured in centimeters, hypertension status is coded as 1 (present) or 0 (absent), and HbA1c is expressed as a percentage.

Hypertension was ascertained based on baseline questionnaire responses, medical records, and the use of antihypertensive medications (Table S1). The blood pressure measurement method for the UK Biobank was that each participant underwent two blood pressure measurements during the assessment visit, with the measurements taken a few minutes apart. Therefore, baseline blood pressure measurements were not included, as the clinical diagnosis of hypertension requires elevated readings on multiple non-consecutive days to rule out transient elevations due to stress or white-coat effects. HbA1c levels were measured using high-performance liquid chromatography on Bio-Rad Variant II Turbo analyzers, with a reportable range of 15–184 mmol/mol (4.0–17.0%) according to the manufacturer’s specifications.

Definition of outcome

CMM is defined as the coexistence of two or more of the following CMDs: CAD, T2DM, and stroke, consistent with many previous studies [4, 22]. The incidence of cardiometabolic events in our study was ascertained using the UK Biobank First Occurrences dataset (Category 1712), which integrates information from primary care, hospital inpatient, death registry, and self-reported data, all mapped to 3-character International Classification of Diseases, Tenth Revision (ICD-10) codes. ICD codes for CAD were I20–25, for T2DM were E11, and for stroke were I60–64. All-cause mortality was determined through linkage with national death registries, including NHS Digital (England and Wales) and the NHS Central Register (Scotland). Dates of death were ascertained from official death certificates. The follow-up data for stroke were updated to April 1, 2023, and for T2DM and CAD to May 1, 2023. However, mortality data, critical for censoring in our multistate model, were only available up to December 19, 2022, due to UK Biobank’s batch data release process. Consequently, we defined December 19, 2022, as the end of the observation period. This approach is consistent with previous studies using UK Biobank data [23, 24]. Participants were followed from the date of enrollment until the occurrence of death, censoring, or December 19, 2022, whichever occurred first.

The primary outcome was the progression of CMM, evaluated through five distinct transition states: (1) baseline (free of CMDs) to FCMD; (2) baseline to death; (3) FCMD to CMM; (4) FCMD to death; and (5) CMM to death. The secondary outcome involved stratifying the FCMD into specific subtypes, including CAD, T2DM, and stroke (Fig. 1A).

Fig. 1.

Different trajectories of CMM progression.A.Numbers (percentages) of participants across the five transition stages in Transition Pattern A. B. Numbers (percentages) of participants across the eleven transition stages in Transition Pattern B. Abbreviations: FCMD, first cardiometabolic disease; CMM, cardiometabolic multimorbidity. Note: Cardiometabolic diseases included type 2 diabetes, coronary artery disease, and stroke. CMM was defined as the presence of at least two of these conditions. Note: In the analysis of specific CMDs, 440 participants diagnosed with multiple new CMDs on the same day were further excluded, as the sequence of events could not be determined

Assessments of covariates

Baseline characteristics were assessed using standardized touchscreen questionnaires and included sex, self-reported race (White vs. non-White), age, Townsend Deprivation Index (TDI; a composite indicator of socioeconomic deprivation incorporating unemployment, car ownership, home ownership, and household overcrowding, with higher scores indicating greater deprivation), education level (high school or below vs. college or above), household income (≤£30,999 vs. >£30,999), alcohol consumption status (current drinker vs. non-drinker), and medication use. Biochemical analyses were performed using the Beckman Coulter AU5800 platform. All lipid measurements and random blood glucose were performed using enzymatic methods. Missing data for continuous variables were imputed using mean values, whereas categorical variables with missing responses (e.g., ‘prefer not to answer’) were assigned to a separate category.

A directed acyclic graph (DAG; www.dagitty.net) was constructed to identify potential confounders and to guide covariate selection [25]. DAG–based variable selection was used to identify a minimally sufficient adjustment set, prioritizing direct causal pathways. Socioeconomic indicators (e.g., education, household income, and TDI) are conceptualized as upstream determinants of IR and CMD risk. These factors may exert their effects through multiple pathways, including social stress leading to adospity, as well as indirectly through health-related behaviors such as diet and physical activity [26]; accordingly, the latter were included as covariates in our model. Lipid profiles were excluded from adjustment, as they represent downstream effects influenced by factors such as sex, obesity, and diet. Moreover, Mendelian randomization studies indicate no direct causal relationship between triglyceride levels and insulin resistance [27]. Similarly, antihypertensive and lipid-lowering medications were not included, as they constitute downstream clinical interventions rather than upstream determinants of IR. Notably, available evidence does not support a direct association between these medications and insulin resistance status [28, 29]. Smoking and alcohol use were also excluded, as they are better considered as confounding or co-occurring lifestyle factors, and current evidence does not support a direct link between these exposures and insulin resistance; thus, they were not included in the final adjustment set [30, 31]. Therefore, the minimally sufficient adjustment set included age, sex, race, body mass index, random blood glucose, physical activity, and diet score (Figure S1). Furthermore, covariates were retained if they satisfied at least one of the following criteria for any outcome: (1) inducing a > 10% change in the β coefficient for the association between eGDR and the outcome, or (2) demonstrating a statistically significant association with the outcome (P < 0.1) [32, 33]. To ensure consistency and reduce residual confounding, all covariates meeting these criteria for any outcome were included in the final models. Details of this covariate selection process are summarized in Table S2.

Statistical analysis

Continuous variables are presented as means with standardized deviations (SDs), and categorical variables are expressed as counts and percentages. Associations between eGDR and transitions across disease states were initially evaluated using conventional Cox proportional hazards regression models. To further characterize dynamic disease trajectories, we implemented multistate Markov models using a clock-forward approach via the R package ‘mstate.’ Unlike standard Cox models, which analyze single endpoints, this framework explicitly accounts for competing risks and temporal dependencies among sequential transitions: (1) baseline to FCMD; (2) FCMD to CMM; (3) baseline to death; (4) FCMD to death; and (5) CMM to death, with death considered an absorbing state (Transition Pattern A, Fig. 1). Hazard ratios (HRs) and 95% confidence intervals (CIs) were estimated per 1-unit increase in eGDR for each transition. For participants who entered multiple disease states on the same date, the entry time of the preceding state was defined as 0.5 days prior to the subsequent state, in accordance with time-series modeling conventions. To further explore potential heterogeneity in the association between eGDR and progression to FCMD, we stratified the multistate transitions by CMD subtypes and constructed 11 transitions (Transition Pattern B, Fig. 1). In pattern B, 440 participants diagnosed with multiple new CMDs on the same day were further excluded, as the sequence of events could not be determined.

To investigate the dose–response relationship between eGDR and cardiometabolic transitions, we applied multistate Cox models across five predefined transitions. Nonlinear associations were assessed using penalized splines (df = 3), with eGDR values restricted to the 2.5th–97.5th percentiles to reduce the influence of outliers. Nonlinearity was tested via likelihood ratio tests comparing spline-based to linear models. Penalized splines were chosen for their automatic smoothing capabilities and flexibility without the need for prespecified knots, making them suitable for exploratory analysis of potential nonlinear patterns. A degree of freedom of 3 was selected to balance flexibility and model parsimony, as commonly applied in epidemiological studies. To verify the robustness of the results, we also performed sensitivity analyses using restricted cubic splines (RCS), with the number of knots (3–6) selected based on the lowest Akaike Information Criterion. For transitions showing significant nonlinearity (P < 0.05), piecewise Cox regression was conducted by evaluating breakpoints at 5% intervals between the 10th and 90th percentiles of eGDR. The optimal breakpoint was identified using the minimum Akaike Information Criterion, and HRs, 95% CIs, and P-values were estimated separately for eGDR values below and above the breakpoint. Moreover, individuals with lower eGDR levels tend to develop FCMD earlier and may benefit from timely interventions. In contrast, higher eGDR levels may delay FCMD onset to older ages, potentially increasing mortality risk. To examine whether age at FCMD onset mediates the relationship between eGDR and death, we performed a causal mediation analysisusing a nonparametric bootstrap approach with 1000 simulations.

Subgroups potentially more susceptible to the effects of eGDR were identified, and stratified analyses were conducted across prespecified subgroups: age (< 60 vs. ≥60 years), sex, body mass index (BMI < 30 vs. ≥30 kg/m²), race, socioeconomic status (high: −6.26 to −2.2 vs. low: −2.2 to 10.82), lipid-lowering or antihypertensive medication use, and smoking status. Interaction terms (eGDR × subgroup) were included in the models, and heterogeneity was assessed using Wald χ² tests.

Several sensitivity analyses were performed to assess the robustness of the findings: (1) To address uncertainty in the timing of prior CMD events, we reassigned time intervals between overlapping diagnoses using five different assumptions (30 days, 180 days, 1 year, 2 years, and 5 years); (2) Participants with baseline cancer were included in the analysis; (3) To minimize reverse causality, we excluded participants who developed FCMD within the first two years after enrollment; (4) Missing covariate data were imputed via multiple imputation, with results pooled using Rubin’s rules; (5) To further Limit the impact of outliers, analyses were repeated restricting eGDR values to the 2.5th to 97.5th percentiles of the distribution; (6) Analyses were repeated after excluding participants with missing baseline variables; (7) Additional covariates—including TDI, household income, education level, lipid levels, smoking and drinking status, and medication use (lipid-lowering and antihypertensive medications)—were included in fully adjusted models to confirm the robustness of results; (8) To address potential residual confounding arising from imbalances in baseline characteristics, we performed inverse probability weighting (IPW) based on a propensity score model. For this purpose, the exposure variable eGDR was dichotomized at its median value, classifying participants into low and high eGDR groups. In addition to the covariates adjusted for in the main models, the propensity score model further included educational level household income, alcohol consumption, and smoking status as additional variables to better account for socioeconomic and lifestyle differences. Covariate balance before and after weighting was evaluated using standardized mean differences (SMDs), with an absolute SMD < 0.1 considered indicative of adequate balance; and (9) To evaluate the robustness of our findings against potential unmeasured confounding, we computed E-values for all statistically significant transitions involving eGDR. The E-value represents the minimum strength of association—on the risk ratio scale—that an unmeasured confounder would need to have with both eGDR and the outcome, conditional on the measured covariates, in order to completely explain away the observed association. All statistical analyses were performed using R software (version 4.2.3). All tests were two-sided, and p-values < 0.05 were considered statistically significant.

Results

This study included 50,342 participants with a mean baseline age of 58.88 years (SD: 7.17), of whom 56.25% were female. The average eGDR at baseline was 8.32 (SD: 2.25) mg/kg/min. Over a median follow-up of 13.6 years (interquartile range [IQR]: 12.9–14.5; total person-years: 664,679), 12,641 participants (25.11%) developed FCMD, corresponding to an incidence rate of 210.52 per 10,000 person-years. Among those with FCMD, 2,081 (16.46%) progressed to CMM, with an incidence rate of 131.24 per 10,000 person-years. A total of 4,847 deaths occurred during follow-up, of which 1,721 (35.55%) occurred after FCMD and 485 (10.00%) after CMM (Table 1; Fig. 1A).

Table 1.

Baseline characteristics of the 50,342 individuals with prediabetes grouped by five transition stages

Variable	Total	Baseline to FCMD	FCMD to CMM	Baseline to Death	FCMD to Death	CMM to Death
No. of subjects	50,342	12,641	2081	2641	1721	485
eGDR (mg/kg/min)	8.32 (2.25)	7.41 (2.25)	6.89 (2.18)	8.11 (2.34)	7.54 (2.32)	6.82 (2.23)
Age (years)	58.88 (7.17)	59.77 (6.98)	60.36 (6.65)	62.22 (5.91)	62.31 (5.84)	62.63 (5.64)
TDI	−1.06 (3.21)	−0.77 (3.34)	−0.42 (3.46)	−0.74 (3.36)	−0.54 (3.42)	−0.44 (3.43)
BMI (kg/m2)	28.78 (5.23)	30.4 (5.46)	31.05 (5.48)	28.48 (5.66)	29.4 (5.58)	30.61 (5.77)
Physical activity (MET-min/week)	2699.05 (2368.46)	2646.43 (2367.16)	2598.07 (2375.75)	2675.79 (2329.52)	2802.36 (2555.41)	2630.41 (2507.49)
HDL (mmol/L)	1.4 (0.34)	1.3 (0.3)	1.26 (0.3)	1.39 (0.34)	1.33 (0.33)	1.28 (0.33)
LDL (mmol/L)	3.74 (0.86)	3.71 (0.89)	3.7 (0.89)	3.64 (0.88)	3.66 (0.92)	3.58 (0.87)
Random glucose (mmol/L)	5.21 (0.74)	5.33 (0.88)	5.39 (0.94)	5.18 (0.69)	5.3 (0.85)	5.41 (1)
TG (mmol/L)	1.99 (1.07)	2.25 (1.19)	2.35 (1.19)	1.92 (1)	2.13 (1.14)	2.23 (1.1)
Diet score	5.17 (1.51)	5.32 (1.5)	5.39 (1.46)	5.36 (1.53)	5.45 (1.51)	5.35 (1.5)
TC (mmol/L)	5.9 (1.13)	5.82 (1.17)	5.77 (1.17)	5.76 (1.14)	5.76 (1.2)	5.63 (1.16)
HbA1c (%)	5.88 (0.16)	5.94 (0.18)	5.97 (0.19)	5.87 (0.15)	5.92 (0.18)	5.96 (0.18)
Sex (n, %)
Men	22,026 (43.75%)	6800 (53.79%)	1284 (61.7%)	1333 (50.47%)	1041 (60.49%)	299 (61.65%)
Women	28,316 (56.25%)	5841 (46.21%)	797 (38.3%)	1308 (49.53%)	680 (39.51%)	186 (38.35%)
Race (n, %)
White	44,963 (89.32%)	11,321 (89.56%)	1859 (89.33%)	2516 (95.27%)	1630 (94.71%)	449 (92.58%)
Others	5132 (10.19%)	1252 (9.9%)	208 (10%)	120 (4.54%)	81 (4.71%)	33 (6.8%)
Missing	247 (0.49%)	68 (0.54%)	14 (0.67%)	5 (0.19%)	10 (0.58%)	3 (0.62%)
Educational levels (n, %)
High school and below	18,059 (35.87%)	4172 (33%)	657 (31.57%)	853 (32.3%)	501 (29.11%)	129 (26.6%)
College or above	20,403 (40.53%)	4602 (36.41%)	684 (32.87%)	959 (36.31%)	565 (32.83%)	139 (28.66%)
Unknown or missing	11,880 (23.6%)	3867 (30.59%)	740 (35.56%)	829 (31.39%)	655 (38.06%)	217 (44.74%)
Average total household income, £ (n, %)
< 30,999	23,713 (47.1%)	6547 (51.79%)	1142 (54.88%)	1519 (57.52%)	1028 (59.73%)	296 (61.03%)
≥ 30,999	17,843 (35.44%)	3833 (30.32%)	554 (26.62%)	627 (23.74%)	369 (21.44%)	88 (18.14%)
Missing	8786 (17.45%)	2261 (17.89%)	385 (18.5%)	495 (18.74%)	324 (18.83%)	101 (20.82%)
Smoking status (n, %)
Current smoker	42,364 (84.15%)	10,288 (81.39%)	1652 (79.38%)	1929 (73.04%)	1217 (70.71%)	359 (74.02%)
Non-Current smoker	7632 (15.16%)	2249 (17.79%)	412 (19.8%)	692 (26.2%)	485 (28.18%)	122 (25.15%)
Missing	346 (0.69%)	104 (0.82%)	17 (0.82%)	20 (0.76%)	19 (1.1%)	4 (0.82%)
Alcohol status (n, %)
Current drinker	5389 (10.7%)	1594 (12.61%)	275 (13.21%)	297 (11.25%)	223 (12.96%)	68 (14.02%)
Non-Current drinker	44,744 (88.88%)	10,980 (86.86%)	1794 (86.21%)	2333 (88.34%)	1487 (86.4%)	414 (85.36%)
Missing	209 (0.42%)	67 (0.53%)	12 (0.58%)	11 (0.42%)	11 (0.64%)	3 (0.62%)
Hypertension (n, %)
No	33,338 (66.22%)	6873 (54.37%)	955 (45.89%)	1648 (62.4%)	950 (55.2%)	208 (42.89%)
Yes	17,004 (33.78%)	5768 (45.63%)	1126 (54.11%)	993 (37.6%)	771 (44.8%)	277 (57.11%)
Lowering lipid use (n, %)
No	45,306 (90%)	10,724 (84.84%)	1661 (79.82%)	2311 (87.5%)	1420 (82.51%)	396 (81.65%)
Yes	5036 (10%)	1917 (15.16%)	420 (20.18%)	330 (12.5%)	301 (17.49%)	89 (18.35%)
Antihypertensives (n, %)
No	44,677 (88.75%)	10,338 (81.78%)	1574 (75.64%)	2240 (84.82%)	1352 (78.56%)	354 (72.99%)
Yes	5665 (11.25%)	2303 (18.22%)	507 (24.36%)	401 (15.18%)	369 (21.44%)	131 (27.01%)

TDI Townsend Deprivation Index, LDL low-density lipoprotein cholesterol, TG triglycerides, BMI body mass index, HDL high-density lipoprotein cholesterol, HbA1c glycated hemoglobin

Compared with the overall cohort, individuals who developed FCMD or progressed to CMM tended to have higher BMI, triglyceride levels, proportion of males, alcohol consumption, and hypertension prevalence, along with lower eGDR, household income, and educational attainment. Among specific FCMD types, 6,655 participants (13.34%) developed T2DM, 1,160 (2.32%) developed stroke, and 4,386 (8.79%) developed CAD. Progression to CMM occurred in 723 (10.86%) of those with T2DM, 200 (17.24%) of those with stroke, and 718 (16.37%) of those with CAD (Fig. 1B).

Multistate model analysis

The traditional Cox proportional hazards models indicated that higher eGDR levels were significantly associated with lower risks of FCMD, CMM, and all-cause mortality (Table 2). Specifically, each 1-unit increase in eGDR was associated with a 12% reduction in the risk of FCMD (HR: 0.88; 95% CI: 0.87–0.89), a 17% reduction in the risk of CMM (HR: 0.83; 95% CI: 0.81–0.85), and a 6% reduction in the risk of all-cause mortality (HR: 0.94; 95% CI: 0.92–0.95).

Table 2.

Associations between eGDR and transitions from baseline to FCMD, CMM, and then death

	Case	HR (95% CI)	P-value
Traditional Cox model
Endpoints
FCMD	12,641	0.88 (0.87,0.89)	< 0.001
CMD	2081	0.83 (0.81, 0.85)	< 0.001
Death	4847	0.94 (0.92,0.95)	< 0.001
Multistate models
Transitions
Baseline → FCMD	12,641	0.88 (0.87, 0.89)	< 0.001
FCMD → CMM	2081	0.93 (0.91, 0.96)	< 0.001
Baseline → Death	2641	0.95 (0.93, 0.98)	< 0.001
FCMD → Death	1721	1.02 (0.99, 1.05)	0.152
CMM → Death	485	0.97 (0.92, 1.02)	0.187
Specific CMDs*
Baseline → FCMD
Baseline → Stroke	1160	0.94 (0.91, 0.98)	< 0.001
Baseline → T2DM	6655	0.86 (0.85, 0.87)	< 0.001
Baseline → CAD	4386	0.89 (0.88, 0.91)	< 0.001
FCMD → CMM
Stroke → CMM	200	0.88 (0.82, 0.95)	0.002
T2DM → CMM	723	0.92 (0.88, 0.96)	< 0.001
CAD → CMM	718	0.93 (0.89, 0.97)	0.001
Baseline → Death	2641	0.95 (0.93, 0.98)	< 0.001
FCMD → Death
Stroke → Death	319	1.0 (0.94,1.07)	0.912
T2DM → Death	576	1.02 (0.97, 1.07)	0.438
CAD → Death	826	1.0 (0.96, 1.04)	0.813
CMM → Death	407	0.94 (0.89,1.0)	0.059

Both the traditional Cox regression and the multistate models were adjusted for age, sex, race, body mass index, random blood glucose, physical activity, and diet score

*In the analysis of specific CMDs, 440 participants diagnosed with multiple new CMDs on the same day were further excluded, as the sequence of events could not be determined

Multistate model analyses further revealed stage-specific effects of eGDR on disease progression (Table 2). From baseline to FCMD, each 1-unit increase in eGDR was associated with a 12% lower risk (HR: 0.88; 95% CI: 0.87–0.89). From FCMD to CMM, the risk was reduced by 7% (HR: 0.93; 95% CI: 0.91–0.96). From baseline to death, a 1-unit increase in eGDR corresponded to a 5% reduction in risk (HR: 0.95; 95% CI: 0.93–0.98), similar to the FCMD–CMM transition. However, eGDR was not significantly associated with mortality risk after the onset of FCMD or CMM (all P > 0.05).

In analyses stratified by FCMD subtype, higher eGDR levels were associated with reduced risks of developing T2DM, stroke, and CAD. Specifically, each 1-unit increase in eGDR was associated with a 14% lower risk of T2DM, 11% lower risk of stroke, and 6% lower risk of CAD. However, eGDR did not significantly reduce the risk of death following any of the FCMD subtypes (all P > 0.05;Table 2).

Exposure–response curves demonstrated a significant nonlinear inverse association between eGDR and the risk of most cardiometabolic transitions (P for nonlinearity < 0.05; Fig. 2). Nonlinear inverse associations between eGDR and the risk of cardiometabolic transitions were observed in most pathways (P for nonlinearity < 0.05; Fig. 2). Transitions from baseline to FCMD and from FCMD to CMM followed similar trajectories, with inflection points identified at eGDR levels of 8.88 and 8.77, respectively. Below these thresholds, higher eGDR was associated with a slight reduction in cardiometabolic risk, while a steeper decline in risk was noted beyond the cut-offs. In the pathways from baseline to death and from FCMD to death, mortality risk decreased markedly with rising eGDR levels below thresholds of 5.09 and 4.87, respectively. However, the rate of decline attenuated at higher levels. Interestingly, in the FCMD-to-death transition, the curve is U-shaped and mortality risk increased when eGDR exceeded 4.87. A mediation analysis was performed using an eGDR threshold of 4.87 to examine whether age at FCMD onset mediated the association between eGDR and mortality (Figure S3). The analysis showed that higher eGDR was associated with a reduced risk of death. Notably, when eGDR exceeded 4.87, age at FCMD onset significantly mediated this association, with a mediation proportion of 15.12% (Figure S3). No evidence of a nonlinear association was observed in the transition from CMM to death. Similar patterns were observed across FCMD subtypes in stratified analyses (Figure S4). Additionally, the RCS analysis revealed similar dose-response patterns in the relationship between eGDR and cardiometabolic transition risks (Figure S5). This consistency between the RCS and spline results further supports the robustness of the observed nonlinear associations between eGDR and cardiometabolic outcomes.

Fig. 2.

Dose-response analysis. Exposure–response associations between eGDR levels and transitions from FCMD, subsequent CMM, and death. Models were adjusted for age, sex, race, body mass index, random blood glucose, physical activity, and diet score. FCMD, first cardiometabolic disease; CMM, cardiometabolic multimorbidity; eGDR, glucose disposal rate

Subgroup analysis

Significant effect modification was observed in the transition from baseline to FCMD by age (< 60 vs. ≥60 years), sex, BMI (< 30 vs. ≥30 kg/m²), smoking status, and use of antihypertensive or lipid-lowering medications (P for interaction < 0.05), but not by race or socioeconomic status (Fig. 3). The inverse association between eGDR and FCMD risk was stronger among participants aged < 60 years, women, non-smokers, those with BMI < 30 kg/m², and those not using antihypertensive or lipid-lowering medications. Nevertheless, higher eGDR levels were consistently associated with lower FCMD risk across all subgroups.

Fig. 3.

Subgroub analysis Associations between eGDR and morbidity transitions among participants with prediabetes, stratified by potential effect modifiers. (a) Association between eGDR and transition from baseline to FCMD; (b) Association between eGDR and transition from FCMD to death; (c) Association between eGDR and transition from FCMD to CMM; (d) Association between eGDR and transition from CMM to death. Abbreviations: HR, hazard ratio; CI, confidence interval; FCMD, first cardiometabolic disease; CMM, cardiometabolic multimorbidity. eGDR, glucose disposal rate. Multistate models were adjusted for age, sex, race, body mass index, random blood glucose, physical activity, and diet score. Associations are expressed as HRs (95% CIs) per 1-unit increase(mg/kg/min) in eGDR

For transitions from FCMD to CMM, FCMD to death, and CMM to death, the associations between eGDR and outcomes were generally consistent across most subgroups, with no significant interactions detected (Fig. 3). Similarly, no significant effect modification was observed for the baseline-to-death transition across most strata (Figure S2).

Sensitivity analysis

Results remained robust across multiple sensitivity analyses: (1) varying the time intervals (30 days, 180 days, 1, 2, and 5 years) for participants who entered multiple disease states on the same day (Table S3); (2) restricting eGDR to the central 95% range (3.89–11.68) (Table S4); (3) including participants with baseline cancer (Table S5); and (4) excluding individuals with incident FCMD within two years of enrollment or missing covariates (Table S7 and Table S7). Consistent results were also observed with multiple imputation (five datasets pooled using Rubin’s rules; Table S8). Further adjustment for socioeconomic factors, lipid profiles, lifestyle behaviors, and medication use did not materially alter the associations (Table S9).

In the IPW analysis, covariate balance was substantially improved, with all SMDs reduced to < 0.04 and most to < 0.01, indicating excellent covariate balance across groups (Table S10). In the IPW-adjusted Cox regression models, participants in the high eGDR group exhibited significantly lower risks for all outcomes compared to those in the low eGDR group: HR for FCMD was 0.65 (95% CI: 0.62–0.68; P < 0.001); for CMM, HR was 0.48 (95% CI: 0.43–0.55; P < 0.001); and for all-cause mortality, HR was 0.77 (95% CI: 0.71–0.84; P < 0.001). After weighting, the effective sample sizes were 10,572 in the low eGDR group and 14,120 in the high eGDR group, reduced from their original (unweighted) sizes of 25,227 and 25,115, respectively (Table S11). These results support the robustness of the inverse association between higher eGDR and reduced risks of incident cardiometabolic conditions and mortality. Furthermore, for the key transitions, E-values ranged from 1.29 to 1.53, indicating that a relatively strong unmeasured confounder would be required to negate the observed effects (Table S12). These findings suggest that the associations are reasonably robust to potential unmeasured confounding.

Discussion

Based on prospective data from 50,342 individuals with prediabetes in the UK Biobank, this study is the first to comprehensively evaluate how eGDR dynamically influences the risk of developing CMD, CMM, and mortality using a multistate modeling approach. Compared with conventional Cox regression analyses, the multistate model not only quantified the stage-specific effects of eGDR but also revealed distinct protective patterns and nonlinear threshold effects. Overall, greater insulin sensitivity (reflected by higher eGDR) was strongly associated with a reduced risk of CMD, CMM, and mortality. However, this protective association with mortality was not observed after the onset of FCMD. Dose–response analyses further revealed a nonlinear relationship between eGDR and mortality at this stage. Specifically, inflection points were identified at approximately 9 mg/kg/min for CMD/CMM onset and 5 mg/kg/min for mortality, suggesting that the prognostic relevance of eGDR varies across stages of disease progression. In contrast, no significant association was found between eGDR and mortality after CMM onset. By integrating multistate modeling with nonlinear analysis, this study addressed key limitations of traditional linear approaches in capturing disease transitions and identifying risk thresholds. These findings support eGDR as a practical and informative marker for risk stratification, with potential to guide targeted interventions and inform individualized prevention strategies across the disease trajectory in people with prediabetes.

Our findings are consistent with previous studies examining the association between eGDR and individual CMDs [34–36]. However, most prior research has focused on single disease endpoints without systematically evaluating the dynamic role of eGDR across the full progression of CMM—from a disease-free state to initial CMD onset, subsequent development of CMM, and ultimately death. Moreover, earlier studies often did not account for interactions among CMDs or competing risks of mortality, limiting their capacity to capture real-world disease trajectories. Disease progression typically follows a distinct sequence, with risk profiles emerging at various stages over time. Longitudinal analyses reveal that the relationship between improvements in eGDR and adverse outcomes is highly stage-dependent. In the early stages, prior to FCMD onset, each 1-unit increase in eGDR is associated with an approximate 18% reduction in the risk of FCMD and a 5% reduction in mortality. As the disease progresses to the FCMD stage, the protective effect of eGDR diminishes, resulting in only a modest 7% reduction in the risk of CMM. At the CMM stage, however, the association between eGDR and mortality becomes negligible. This attenuation of protective effects likely reflects the accumulation of irreversible pathological changes, including endothelial dysfunction and organ fibrosis, which progressively become the dominant determinants of mortality risk [37]. Moreover, at the CMM stage, patients often exhibit multi-system involvement and pronounced metabolic dysregulation, where improving insulin sensitivity alone may no longer suffice to mitigate the adverse prognosis associated with widespread organ damage. These findings underscore the clinical value of early interventions targeting insulin sensitivity and highlight the need for tailored therapeutic strategies at different stages of disease progression.

Exposure-response analysis revealed a significant nonlinear relationship between eGDR and the risk of CMM and mortality, with distinct patterns observed across different stages of disease progression. In the pathways leading to FCMD and CMM, the reduction in cardiometabolic risk associated with increasing eGDR gradually diminished. However, when eGDR exceeded 8.7–8.9 mg/kg/min, the risk dropped significantly. This suggests that, in the early stages of disease, moderate improvement in insulin sensitivity may lower FCMD risk by mitigating early pathological processes, such as oxidative stress, endothelial dysfunction, and inflammation [38, 39]. Conversely, during later disease progression, vascular injury and metabolic dysfunction may already be advanced, requiring greater improvements in insulin sensitivity to achieve meaningful reductions in CMM risk.Nevertheless, more intensive intervention is required to achieve substantial reductions in cardiometabolic risk. Unlike the CMD transition model, in the mortality pathway, increasing eGDR was inversely associated with mortality risk, but the reduction in risk plateaued at higher levels of eGDR, indicating that the protective effect reaches saturation. Notably, in the FCMD-to-death pathway, eGDR demonstrated a unique U-shaped relationship, with an inflection point at 4.87. Mediation analysis revealed that the negative association between eGDR and mortality was significantly mediated by the age of FCMD onset, with a mediation proportion of 15.2%. These findings suggest that while higher insulin sensitivity may delay FCMD onset, once the disease manifests, patients are typically older, and age-related physiological decline and comorbidities may increase susceptibility to mortality.

Subgroup analyses revealed significant heterogeneity in the association between eGDR and FCMD progression, highlighting the importance of individualized clinical risk assessment. First, age-stratified analyses showed a stronger protective effect of higher eGDR among individuals < 60 years, possibly reflecting greater physiological responsiveness to improved insulin sensitivity in younger adults, whereas age-related pathophysiologic changes may attenuate this effect in older populations [40]. Second, sex-stratified analyses demonstrated a more pronounced association in women, consistent with known sex-specific differences in insulin metabolism, potentially mediated by estrogen’s regulatory role in insulin signaling. Third, stratification by BMI revealed that the association between eGDR and FCMD was stronger among individuals with BMI < 30 kg/m², underscoring the relevance of insulin resistance even in non-obese populations and supporting the concept of distinct metabolic phenotypes, such as metabolically healthy obesity and metabolically unhealthy normal weight [41]. Additionally, the association between eGDR and FCMD risk was weaker in individuals on antihypertensive or lipid-lowering therapy, suggesting that pharmacologic therapies may partially mitigate the impact of insulin resistance on disease progression [42]. Collectively, these findings support the clinical utility of eGDR as a stratified risk marker, and underscore the need to tailor preventive strategies based on individual demographic and clinical characteristics.

Strengths and limitations

Methodologically, we applied a multistate Markov model to more accurately capture the dynamic progression of CMDs and account for competing risks of all-cause mortality, in contrast to traditional static endpoint analyses. Guided by causal inference principles and DAG, we systematically identified and selected potential confounders. Covariates included in the models were rigorously evaluated based on their influence on both exposure–outcome associations and outcome prediction, thereby minimizing the risk of over-adjustment bias. To ensure the robustness of our findings, we conducted extensive sensitivity analyses, including variations in follow-up definitions, extreme value exclusions, and multiple imputation approaches—all of which yielded consistent results.

Nonetheless, several limitations should be acknowledged. First, UK Biobank participants tend to be healthier and have higher educational and socioeconomic status than the general population, potentially limiting generalizability. Second, eGDR is calculated from waist circumference, HbA1c, and hypertension status—the latter derived from self-reports, medical records, and medication use—introducing potential measurement error and misclassification. Third, potential unmeasured confounders, including environmental exposures (e.g., air pollution), mental health status, and genetic susceptibility, may influence the observed associations between eGDR and cardiometabolic outcomes [43–45]. Due to data limitations, these variables could not be adjusted for in the current study; however, they may contribute to residual confounding and warrant investigation in future research. Despite adjustments for known confounders, the potential impact of these unmeasured variables remains a key limitation of our analysis. Fourth, as an observational study, causal inference remains limited despite the prospective design. Fifth, although follow-up was relatively long, metabolic diseases typically evolve over decades, and our study may not fully reflect long-term disease trajectories. Lastly, our findings are based exclusively on a UK population; given known ethnic differences in insulin sensitivity, extrapolation to other populations should be done with caution.

Conclusions

In individuals with prediabetes, higher insulin sensitivity, as assessed by the estimated glucose disposal rate eGDR, is associated with a reduced risk of CMD, CMM, and all-cause mortality. However, this protective effect of eGDR diminishes as the disease progresses. Once the first CMD event occurs, the association between eGDR and mortality becomes less pronounced. As a clinically accessible marker of insulin sensitivity, eGDR shows considerable potential for dynamic prevention, with distinct threshold effects observed across different stages of disease progression. Our findings underscore eGDR as a valuable biomarker for the prevention, risk stratification, and monitoring of transitions between comorbidities in cardiovascular metabolic diseases.

Authors’ contributions

Conceptualization: WKC; Methodology: WKC; Formal analysis and investigation: WKC and XRF; Writing—original draft preparation: WKC; Writing—review and editing: WKC, XRF, PFK, JJL and BT; Resources: WKC; Supervision: JJL and BT. All authors contributed to subsequent revisions and approved the final version. All authors read and approved the final manuscript.

Funding

This work was supported by the Regional Disease Joint Research Center Project, Institute of Health, Hefei Comprehensive National Science Center (Grant No. 2024bydjk005).

Data availability

Data can be accessed from a public and open repository. Interested researchers can apply for access to the UK Biobank data at https://www.ukbiobank.ac.uk and https://biobank.ndph.ox.ac.uk/ukb.

Declarations

Ethics approval and consent to participate

The UK Biobank was established with ethical clearance from the North West Multi-Centre Research Ethics Committee (REC reference: 11/NW/0382). Written informed consent have been provided by all participants.

Consent for publication

Not applicable.

Competing interests

The authors declare no competing interests.