Joint association of atherogenic index of plasma and estimated glucose disposal rate with new-onset cardiovascular disease risk in individuals with cardiovascular–kidney–metabolic syndrome stages 0–3: a 9-year nationwide prospective cohort study

Abstract

Background

The atherogenic index of plasma (AIP) and estimated glucose disposal rate (eGDR) are two composite indices derived from routine metabolic measurements and are associated with cardiocerebrovascular disease risk. In individuals with Cardiovascular–Kidney–Metabolic (CKM) syndrome stages 0–3, however, it remains unclear whether joint stratification by these markers helps summarize gradients of cardiovascular disease, heart disease, and stroke risk beyond single-marker assessment.

Methods

Using data from the China Health and Retirement Longitudinal Study (CHARLS), 5,925 participants without CVD at the start and in CKM stages 0–3 were analyzed. Participants were grouped by median AIP and/or eGDR values. Kaplan–Meier curves and Cox models assessed the link between these indicators and new CVD, heart disease, and stroke cases. Furthermore, both multiplicative and additive interactions between AIP and eGDR were assessed. The predictive value was assessed using the time-dependent Harrell’s C index, integrated discrimination improvement (IDI), and net reclassification improvement (NRI).

Results

A cohort of 5,925 participants aged 45 years and older (mean age: 57.92 ± 8.52 years) was analyzed, with 54.65% of the cohort being female. During the nine-year follow-up period, 1,467 (24.76%) participants developed incident CVD, including 1,106 (18.67%) with heart disease and 525 (8.86%) with stroke. The high AIP and low eGDR group had the highest risk, with CVD hazard ratios (HRs) of 1.35 (95% CI 1.14–1.59), heart disease HRs of 1.32 (95% CI 1.08–1.62), and stroke HRs of 1.59 (95% CI 1.19–2.12), using the low AIP and high eGDR group as the reference. Neither multiplicative nor additive interaction was statistically significant. The combined application of AIP and eGDR provided a modest improvement in predictive capability for cardiovascular disease, heart disease, and stroke.

Conclusion

In individuals with CKM stages 0–3, combined AIP and eGDR stratification captured gradients of cardiovascular risk. The combined application of these indicators may provide modest incremental value for risk stratification within CKM stages, thereby aiding in the identification of high-risk individuals during the early stages of CKM.

Graphical abstract

Supplementary Information

The online version contains supplementary material available at 10.1186/s12933-026-03200-5.

Research insights

What is currently known about this topic?

• Higher AIP and lower eGDR have each been associated with adverse cardiovascular outcomes in CKM or related populations, but evidence on their joint association in CKM stages 0–3 remains limited.

What is the key research question?

• In individuals with CKM syndrome stages 0–3, does the joint profile of high AIP and low eGDR identify higher risks of incident CVD, heart disease, and stroke?

• Does joint assessment of AIP and eGDR provide modest additional predictive information, and is there evidence of significant additive or multiplicative interaction between them?

What is new?

• In this nationally representative cohort of adults with CKM stages 0–3, the joint profile of high AIP and low eGDR identified the subgroup with the highest observed risks of incident CVD, heart disease, and stroke.

How might this study influence clinical practice?

• AIP and eGDR can be calculated from routine clinical measurements and may serve as low-cost summary markers for risk stratification in early CKM. These indices are better viewed as supplementary tools for refining risk assessment within CKM stages 0–3 and may help identify individuals with a heavier cardiometabolic burden.

Background

Cardiovascular disease (CVD) continues to be a predominant cause of mortality and morbidity on a global scale. Despite ongoing advancements in the management of traditional risk factors, the overall global burden of CVD has escalated considerably over the past three decades [1]. According to the most recent Global Burden of Disease (GBD) 2023 data, there were 626 million cases of CVD worldwide in 2023 (95% uncertainty interval [1]: 591–672 million), leading to 437 million disability-adjusted life years (DALYs) (95% UI: 401–465 million) [2]. In response to this situation, the American Heart Association (AHA) released a scientific statement in 2023, introducing the concept of Cardiovascular–Kidney–Metabolic (CKM) syndrome. This statement proposes a staging framework that integrates the interconnections between obesity, metabolic risk factors, chronic kidney disease (CKD), diabetes, and cardiovascular dysfunction into a comprehensive disease spectrum [3]. With the rising prevalence of obesity and metabolic disorders, evidence from various countries indicates that CKM syndrome imposes a substantial population burden and tends to manifest at an earlier age. The AHA advocates for the transition of risk assessment and intervention to stages 0–3, prior to the manifestation of clinical events, in order to facilitate more cost-effective primary prevention and stratified management [4, 5]. A defining feature of CKM stages 0–3 is the accelerated risk across multiple systems, which has not yet advanced to irreversible end-organ damage. Consequently, there is a need for accessible and reproducible indicators that can summarize overall cardiometabolic burden within early CKM and support pragmatic risk stratification.

From a pathophysiological standpoint, atherosclerosis fundamentally involves the infiltration and retention of apolipoprotein B(ApoB)-containing lipoprotein particles within the arterial wall, which initiates local inflammatory responses and facilitates the gradual development of plaque. Notably, even in the absence of significantly elevated low-density lipoprotein (LDL) levels, triglyceride-rich lipoproteins and their remnant particles can contribute cholesterol to the arterial wall, thereby promoting plaque progression and accounting for the residual risk that persists despite LDL management. Insulin resistance may precede the onset of diabetes for an extended period and, through mechanisms such as decreased endothelial nitric oxide bioavailability, impaired vasodilation, increased proliferation of vascular smooth muscle cells, sympathetic activation, and stimulation of the renin–angiotensin–aldosterone system, contributes to elevated blood pressure, microvascular damage, and the establishment of a pro-coagulant inflammatory milieu. These processes collectively heighten the risk of plaque rupture or thrombotic events. The CKM framework underscores that the kidneys play an active role in this process. Early renal dysfunction, along with sodium and water retention and inflammatory responses, can exacerbate hemodynamic stress and metabolic disorders, thereby establishing a positive feedback loop among organs prior to the manifestation of clinical events. While the hyperinsulinemic-euglycemic clamp technique remains the gold standard for assessing insulin resistance, its complexity and high cost limit its applicability in large-scale population studies and routine clinical practice [6]. Consequently, the estimated glucose disposal rate (eGDR), derived from conventional indicators such as waist circumference (WC), hypertension status, and glycated hemoglobin A1c (HbA1c), has been linked to cardiovascular outcomes in several studies [7]. Furthermore, the eGDR index has demonstrated superiority over traditional insulin resistance indices in predicting cardiovascular disease events [8, 9]. The atherogenic index of plasma (AIP), defined as the logarithm of the ratio of triglycerides to high-density lipoprotein cholesterol (log(TG/HDL-C)), was introduced by Dobiásová and Frohlich in 2001 [10]. This index provides a comprehensive measure of the burden of atherosclerotic lipid abnormalities. Numerous studies have demonstrated that AIP is more sensitive than conventional lipid indicators in identifying adverse lipoprotein profiles and is correlated with an elevated risk of cardiovascular disease and stroke [11–13]. At the same time, the components used to construct these indices overlap substantially with the metabolic features that define early CKM stages. This overlap is central to the interpretation of any association observed within CKM stages 0–3. In this setting, the key question is not whether these markers represent fully independent biological pathways, but whether they can capture gradients of residual cardiometabolic burden among individuals already classified within the CKM framework.

It is important to highlight that the current body of evidence concerning the early stages of CKM is fragmented, with the majority of studies concentrating on individual indicators or outcomes. Typically, research reports the association of the AIP with either overall CVD or stroke independently, or validates the inverse relationship between eGDR and overall CVD separately. However, fewer studies have undertaken a simultaneous comparison of overall CVD and specific outcomes, such as heart disease and stroke, within the CKM stages 0–3 framework. Moreover, there is a notable deficiency in systematic evaluations regarding whether the combined stratification of these two phenotypes can more effectively differentiate risk gradients in early CKM populations, elucidate their potential additive and multiplicative interaction characteristics, and determine whether a combined assessment can offer quantifiable predictive improvements beyond traditional risk factors. Addressing this gap in evidence is practically significant, as clinical decision-making during CKM stages 0–3 predominantly relies on low-cost, scalable stratification tools, rather than complex tests that are challenging to routinely implement in primary care settings.

To address this knowledge gap, we utilized the nationally representative CHARLS prospective cohort to evaluate the independent and combined associations of eGDR and AIP with the incidence of CVD, heart disease, and stroke in individuals with CKM stages 0–3 who had no baseline CVD. Additionally, we investigated potential interactions between these variables and compared the incremental predictive value of combined assessments against single indicators, with the objective of providing evidence to optimize early screening and stratified management strategies for CKM.

Methods

Study design and population

This research utilizes cohort data from the China Health and Retirement Longitudinal Study (CHARLS) (http://charls.pku.edu.cn/), a nationally representative, population-based prospective cohort study employing a multi-stage stratified sampling method to conduct household surveys across 150 counties/districts and 450 villages/communities in 28 provinces of China. The study leverages pre-existing data collected from five survey waves conducted in 2011, 2013, 2015, 2018, and 2020, without undertaking any additional new field visits. Blood samples were obtained from participants during the baseline survey. Comprehensive descriptions of the research methodology and data collection procedures of CHARLS have been documented in the literature [14]. The study employs the population from the 2011 baseline survey as the initial sample and constructs an analytical cohort by integrating follow-up data from subsequent waves. Initially, 17,708 participants were included, with exclusions applied based on the following criteria: (1) Age < 45 years or missing age information (n = 404); (2) Prevalent CVD at or before baseline (Wave 1), including heart disease or stroke (n = 2,501); (3) Missing CKM staging information or CKM stage 4 at baseline (n = 4,913); (4) Missing baseline AIP or eGDR data (n = 2,571); (5) Lost to follow-up (n = 1,391). Ultimately, 5,925 participants were included in the analysis (Fig. 1).

Fig. 1.

Flowchart of the study population. Abbreviations: AIP: atherogenic index of plasma; CVD: cardiovascular disease; CKM: Cardiovascular–Kidney–Metabolic; eGDR: estimated glucose disposal rate

Ethical approval

This study has been approved by the Biomedical Ethics Committee of Peking University Health Science Center (Ethics Approval Number: IRB00001052-11015), and all participants provided written informed consent. This study adheres to the principles of the Declaration of Helsinki, and the reporting of the study results follows the STROBE (Strengthening the Reporting of Observational Studies in Epidemiology) guidelines [15, 16].

Definition of CKM syndrome

The classification of CKM syndrome is delineated in accordance with the guidelines set forth by the American Heart Association (AHA) Presidential Advisory Statement [3]. The criteria for staging CKM syndrome are as follows: Stage 0 encompasses individuals with no risk factors for CKM syndrome; Stage 1 includes those with abdominal obesity and/or prediabetes; Stage 2 pertains to individuals with metabolic diseases, such as type 2 diabetes, hypertension, or hypertriglyceridemia, or kidney disease; Stage 3 involves early cardiovascular disease in individuals with obesity, metabolic diseases, or kidney disease; and Stage 4 comprises individuals with established clinical cardiovascular disease, including coronary heart disease, heart failure, stroke, peripheral artery disease, and atrial fibrillation [17, 18]. This study predominantly concentrates on individuals classified within CKM stages 0 through 3. Detailed criteria for CKM stages 0 to 3 are provided in Table S1, while Table S2 enumerates the specific definitions of various diseases [see Additional file 2].

Data collection

The selection of covariates was informed by existing literature and grounded in biological plausibility. Data on covariates were gathered by professionally trained investigators utilizing structured questionnaires [19]. The covariates incorporated in this study comprised: (1) demographic variables, including age, sex, marital status, educational attainment, and place of residence; (2) anthropometric measurements, such as systolic blood pressure (SBP), diastolic blood pressure (DBP), height, body mass index (BMI), and waist circumference (WC); (3) lifestyle factors, specifically smoking status and alcohol consumption; (4) disease status, encompassing hypertension, diabetes, and dyslipidemia; and (5) laboratory test results, which included triglycerides (TG), total cholesterol (TC), high-density lipoprotein cholesterol (HDL-C), low-density lipoprotein cholesterol (LDL-C), high-sensitivity C-reactive protein (hs-CRP), serum creatinine (Cr), estimated glomerular filtration rate (eGFR), fasting blood glucose (Fglu), and glycated hemoglobin (HbA1c). The laboratory parameters are comprehensively presented in Table S3 [see Additional file 2]. BMI was calculated using the formula: BMI (kg/m²) = weight (kg)/height (m)² [20]. Participants who had smoked ≥ 100 cigarettes in their lifetime were classified as ever smokers and further categorized as current or ever smokers. Never smokers were those who had smoked < 100 cigarettes in their lifetime [21]. Alcohol consumption was categorized into never drinkers, defined as individuals who never or rarely consume alcohol (less than once a month), and current drinkers, defined as those who consume alcohol more than once a month [22]. All biological indicators and physical measurements were conducted in accordance with standardized operating procedures (SOPs). The CHARLS team regularly conducts quality control monitoring to ensure that test results fall within ± 2 standard deviations of the average quality control concentration [14, 23].

Exposure assessment

The eGDR and AIP indices were computed utilizing methodologies delineated in prior research, employing the following equations: AIP = log10 (TG/HDL-C (mg/dL)), and eGDR = 21.158 − (0.09 × waist circumference(WC,cm)) − (3.407 × hypertension (yes = 1/no = 0)) − (0.551 × glycosylated hemoglobin A1c (HbA1c, %)) [7, 24]. In accordance with existing literature, the median values of AIP (− 0.05) and eGDR (10.15) within this study cohort were adopted as threshold values to categorize participants into four distinct groups: low AIP and low eGDR, high AIP and low eGDR, low AIP and high eGDR, and high AIP and high Egdr [25, 26]. We chose the median because no universally accepted clinical cut points for AIP or eGDR have been established for CKM stages 0–3, and the median-based approach yielded balanced groups for stable joint comparison. These categories were intended for internal risk stratification rather than to define clinically validated diagnostic thresholds. To reduce reliance on arbitrary categorization, we also examined AIP and eGDR as continuous variables using restricted cubic splines.

Ascertainment of outcomes

The main focus of this research was the emergence of new-onset CVD incidents. CVD was characterized as heart disease and stroke [27, 28]. Diagnoses were based on participants' self-reports or systematically gathered by medically trained investigators using standardized questionnaires from previous studies. New-onset CVD was defined as the initial occurrence of either heart disease or stroke, aligning with earlier research. In competing risk analysis, death was considered a competing event. Mortality data were collected from death certificates, medical records, or interviews with relatives during follow-up. Participants were followed up in 2013, 2015, 2018, and 2020. If a participant experienced a CVD event or died before 2020, they were not followed further. Otherwise, follow-up continued until 2020. Outcomes were ascertained based on self-reported physician diagnoses collected via standardized questionnaires, and new-onset events were defined as the first event during the follow-up period (from 2011 to the end of 2020). The time to event was defined as the interval between the onset date and the baseline date. Conversely, for participants who did not experience an event during the follow-up period, the follow-up duration was determined by the interval between the date of the last survey and the baseline assessment date [19].

Missing data

In addressing the issue of missing data for AIP and eGDR during the baseline phase, this study undertook a systematic comparison of baseline characteristics between participants with complete AIP and eGDR data (Complete group) and those with missing data (Missing group) to evaluate potential selection bias resulting from the absence of exposure data. The findings indicated that the distribution patterns of most baseline characteristics were similar across the two groups (Table S4 [see Additional file 2]). Regarding missing covariate data, the overall missing rate in this study was approximately 4%, with around 239 out of 5925 subjects having incomplete covariate data. Through the construction of a matrix and subsequent correlation analysis (Figure S1-S2 [see Additional file 1]), we identified a high incidence of common missing data and strong correlations among variables. The missing covariate data were addressed utilizing the multiple imputation by chained equations (MICE) method, as implemented in the "mice" package of the R software (version 4.4.2), under the assumption that the data were missing at random. Ten complete datasets were generated, and the results from these imputed datasets were aggregated and reported using the "pool" function within the "mice" package, following Rubin's rules [29]. The imputation model was tailored to the type of variable: predictive mean matching (PMM) was employed for continuous variables, while logistic regression was used for both binary and categorical variables. The missing rates and the corresponding imputation methods for each variable are detailed in Table S5 [see Additional file 2] [30].

Covariates

Covariate selection was guided by clinical relevance and evaluated using a Spearman correlation matrix to identify potential information redundancy and multicollinearity (Figure S3 [see Additional file 1]). Recognizing that variables such as age, sex, height, marital status, education level, residence, and smoking and drinking status are significantly associated with metabolic exposure levels and cardiovascular outcomes, and may serve as key confounding factors at socio-demographic and lifestyle levels, these variables were incorporated as pre-specified core covariates in the primary model. Given that the calculation of eGDR involves WC, HbA1c, and hypertension status, and that the AIP is derived from the ratio of TG to HDL-C, WC, HbA1c, TG, HDL-C, BMI, and FBG, which are closely linked to their respective metabolic pathways, were excluded from the model. In the analysis of variables that are highly correlated or represent the same physiological construct, preference was given to indicators with superior clinical interpretability and statistical stability. Consequently, SBP was selected as the sole representative among blood pressure indicators, excluding DBP. Similarly, eGFR was chosen as the representative measure of renal function, with serum creatinine (Scr) omitted. LDL-C was retained as the representative indicator of lipid levels. Given that hypertension, diabetes, and dyslipidemia are well-established traditional risk factors for cardiovascular events and reflect an individual's baseline cardiovascular-metabolic risk, the final covariates were selected to account for the influence of baseline disease status on new outcomes. These covariates included age, sex, height, marriage, education, location, smoking status, drinking status, SBP, eGFR, LDL-C, hs-CRP, hypertension, dyslipidemia and diabetes [31]. To evaluate potential multicollinearity among the variables in the model, the variance inflation factor (VIF) was calculated, revealing that all included covariates had VIF values below 5 (Table S6 [see Additional file 2]).

Statistical analysis

Descriptive analysis

Descriptive analysis was performed on the baseline characteristics of the study subjects. Normally distributed continuous variables were expressed as mean ± standard deviation (SD), and group differences were assessed using analysis of variance (ANOVA). Non-normally distributed continuous variables were expressed as median and interquartile range, and group comparisons were performed using the Kruskal–Wallis test. Categorical variables were expressed as counts and percentages, and group differences were assessed using the chi-square test [32].

Risk correlation analysis

The Kaplan–Meier method was employed to generate cumulative incidence curves, while the log-rank test facilitated the comparison of differences among various exposure groups. Subsequently, a Cox proportional hazards regression model was developed to systematically evaluate the association between AIP and eGDR and the risk of incident CVD, heart disease, and stroke. To examine the influence of covariate adjustment on the results and to assess the robustness of the model, four progressively complex models were constructed: The initial model was unadjusted, incorporating only the exposure variables; Model I included adjustments for sociodemographic factors such as age, sex, height, marital status, place of residence, education level, alcohol consumption, and smoking status; Model II further incorporated adjustments for clinical and laboratory indicators, including SBP, eGFR, LDL-C, and hs-CRP; Model III additionally adjusted for the presence of hypertension, diabetes, and dyslipidemia33. Given the conceptual overlap between the exposure indices and the CKM construct, we additionally fitted a CKM-stage-adjusted model to examine whether the observed associations persisted after accounting for baseline stage membership. To further evaluate the additive interaction between AIP and eGDR, three quantitative indicators were employed: relative excess risk due to interaction (RERI), attributable proportion due to interaction (AP), and synergy index (SI). RERI quantifies the increase in excess risk from the combined exposure relative to the sum of the individual associations of the two exposures. AP indicates the proportion of the outcome attributable to the interaction, while SI denotes the ratio of the excess risk from the combined exposure to the sum of the excess risks from the individual exposures. The statistical significance of the interaction was assessed using the likelihood ratio test, and a trend P-value was calculated based on the median of various stratified groups. To control the false discovery rate and mitigate false positive results, the Benjamini–Hochberg method was applied for multiple comparisons.

Risk prediction

Utilizing the Cox proportional hazards model, we developed predictive models to assess the influence of incorporating the AIP, the eGDR, and their combined effect on the predictive performance for CVD, heart disease, and stroke risk. We computed the time-dependent Harrell's concordance index to evaluate model performance because it is suitable for censored time-to-event data and allows evaluation of discrimination over follow-up. We selected this metric to compare the relative ordering performance of models containing atherogenic index of plasma, estimated glucose disposal rate, and their combination over follow up rather than to establish direct clinical utility. Given that the C-index may be relatively insensitive to subtle improvements in predictive performance and is unable to assess model calibration, we additionally report the Integrated Discrimination Improvement (IDI) and continuous Net Reclassification Improvement (NRI) at Year 9, aiming to provide a more comprehensive and detailed assessment of the incremental predictive value.

Sensitivity analysis

To evaluate the robustness of the primary findings, we performed the following pre-specified sensitivity analyses: (1) Further adjusted Model III for baseline medication use, including antihypertensive, antidiabetic, and lipid-lowering medications. (2) Excluded participants with missing data. (3) Excluded participants who were already receiving antihypertensive, antidiabetic, or lipid-lowering treatment at baseline to minimize the influence of treatment intervention on the exposure-outcome relationship. (4) Repeated the primary analysis after excluding individuals in CKM stage 0 to assess consistency across populations with varying degrees of cardiovascular-renal metabolic risk burden. (5) Excluded participants with a follow-up duration of less than two years to mitigate the potential impact of reverse causality and early events on the findings. (6) Considering that the study cohort comprised middle-aged and elderly individuals, there was a potential for non-CVD/heart disease/stroke-related deaths to occur during the follow-up period, potentially obscuring the observation of outcome events. To address this, we employed the Fine-Gray subdistribution hazard model, treating non-corresponding outcome deaths as competing events, to perform a competing risk sensitivity analysis on the primary analysis results. We reported the hazard ratios along with their 95% confidence intervals. All statistical analyses were conducted using R software version 4.4.2, with all tests being two-sided and a significance level set at P < 0.05.

Results

Baseline characteristics

This study comprised 5,925 participants with a mean age of 57.92 ± 8.52 years, of whom 54.65% were female. The cohort was stratified into four groups according to the median baseline levels of AIP and eGDR: the low AIP & low eGDR group (n = 1,147); the high AIP & low eGDR group (n = 1,815); the low AIP & high eGDR group (n = 1,815); and the high AIP & high eGDR group (n = 1,148). Most baseline characteristics exhibited statistically significant differences across the groups. While hs-CRP levels did not differ significantly among the four groups (P = 0.21), all other baseline characteristics demonstrated significant intergroup differences (P ≤ 0.02). These characteristics included age, sex, education, region, smoking status, alcohol consumption, systolic and diastolic blood pressure, height, glycated hemoglobin, creatinine, triglycerides, fasting blood glucose, total cholesterol, HDL, LDL, waist circumference, BMI, eGFR, and the prevalence of hypertension, diabetes, hyperlipidemia, cardiovascular disease, heart disease, stroke, CKM stage, and treatment status for hyperglycemia, hypertension, and dyslipidemia. The two groups characterized by low eGDR predominantly fell within CKM stages 2–3. Specifically, the group with low AIP and low eGDR comprised 52.31% in CKM stage 2 and 35.92% in stage 3, while the group with high AIP and low eGDR included 45.51% in stage 2 and 51.07% in stage 3. Conversely, the groups with high eGDR were primarily concentrated in CKM stages 0–1. The group with low AIP and high eGDR consisted of 24.19% in CKM stage 0 and 42.20% in stage 1, whereas the group with high AIP and high eGDR comprised 6.27% in stage 0 and 15.16% in stage 1. In terms of clinical outcomes, the group with high AIP and low eGDR exhibited the highest incidence rates of CVD at 31.85%, heart disease at 23.09%, and stroke at 13.11%. In contrast, the group with low AIP and high eGDR demonstrated the lowest incidence rates, with CVD at 17.69%, heart disease at 13.33%, and stroke at 5.51% (Table 1). When the groups were dichotomized based on the median values of AIP or eGDR, consistent trends emerged (Table S7 [see Additional file 2]).

Table 1.

Baseline characteristics according to eGDR and AIP in CKM syndrome stage 0–3

Variable	Total (n = 5925)	Low AIP & Low eGDR (n = 1147)	High AIP & Low eGDR (n = 1815)	Low AIP & High eGDR (n = 1815)	High AIP & High eGDR (n = 1148)	P value
Age, years	57.92 ± 8.52	60.04 ± 9.10	58.54 ± 8.44	57.04 ± 8.30	56.21 ± 7.83	< 0.0001
Sex						< 0.001
Female	3238(54.65)	645(56.23)	1050(57.85)	923(50.85)	620(54.01)
Male	2687(45.35)	502(43.77)	765(42.15)	892(49.15)	528(45.99)
Marriage						< 0.0001
Married	5096(86.01)	933(81.34)	1582(87.16)	1574(86.72)	1007(87.72)
Separated	829(13.99)	214(18.66)	233(12.84)	241(13.28)	141(12.28)
Education						0.02
No formal education	2783(46.98)	587(51.18)	830(45.76)	853(47.00)	513(44.69)
Primary	1313(22.16)	247(21.53)	408(22.49)	412(22.70)	246(21.43)
Second/high school/higher	1828(30.86)	313(27.29)	576(31.75)	550(30.30)	389(33.89)
Location						< 0.0001
Rural	3987(67.29)	778(67.83)	1107(60.99)	1343(73.99)	759(66.11)
Urban	1938(32.71)	369(32.17)	708(39.01)	472(26.01)	389(33.89)
Drink						< 0.0001
No	3891(65.67)	730(63.64)	1266(69.75)	1126(62.04)	769(66.99)
Yes	2034(34.33)	417(36.36)	549(30.25)	689(37.96)	379(33.01)
Smoke						< 0.0001
Current	1788(30.18)	342(29.82)	453(24.96)	601(33.11)	392(34.15)
Ever	436(7.36)	78(6.80)	171(9.42)	119(6.56)	68(5.92)
Never	3701(62.46)	727(63.38)	1191(65.62)	1095(60.33)	688(59.93)
SBP, mmHg	128.94 ± 20.44	141.28 ± 21.02	139.59 ± 20.78	117.06 ± 12.10	118.61 ± 11.89	< 0.0001
DBP, mmHg	75.45 ± 11.82	80.55 ± 12.05	81.01 ± 11.88	69.49 ± 8.79	70.98 ± 8.98	< 0.0001
Height, m	157.98 ± 8.47	157.18 ± 8.83	158.59 ± 8.59	157.79 ± 8.34	158.11 ± 8.06	< 0.001
hs-CRP, mg/L	2.39 ± 6.54	2.47 ± 6.57	2.62 ± 4.82	2.23 ± 7.46	2.20 ± 7.27	0.21
HbA1c, %	5.26 ± 0.76	5.31 ± 0.85	5.52 ± 1.03	5.09 ± 0.38	5.08 ± 0.44	< 0.0001
Cr, mg/dL	0.77 ± 0.18	0.76 ± 0.19	0.79 ± 0.19	0.76 ± 0.16	0.77 ± 0.18	< 0.0001
TG, mg/dL	130.28 ± 109.45	76.96 ± 20.95	196.20 ± 147.68	73.74 ± 21.13	168.70 ± 99.88	< 0.0001
Glucose, mmol/L	108.60 ± 32.20	107.55 ± 30.64	120.05 ± 45.97	100.42 ± 15.14	104.49 ± 19.52	< 0.0001
TC, mg/dL	194.94 ± 38.27	193.40 ± 35.18	202.43 ± 40.53	188.82 ± 35.03	194.28 ± 40.45	< 0.0001
HDL-C, mg/dL	51.76 ± 15.16	60.69 ± 13.75	41.69 ± 9.78	61.56 ± 13.49	43.26 ± 9.77	< 0.0001
LDL-C, mg/dL	117.77 ± 34.46	119.35 ± 31.97	120.35 ± 37.74	115.03 ± 31.17	116.48 ± 36.03	< 0.0001
WC, cm	84.08 ± 12.19	87.31 ± 9.61	92.09 ± 9.08	77.36 ± 10.24	78.80 ± 12.94	< 0.0001
BMI kg/m2	23.55 ± 3.80	24.03 ± 3.90	25.87 ± 3.78	21.48 ± 2.67	22.70 ± 3.03	< 0.0001
eGFR	97.09 ± 13.22	95.89 ± 13.67	95.19 ± 13.96	99.14 ± 11.68	98.03 ± 13.32	< 0.0001
Hypertension						< 0.0001
No	4644(78.78)	705(61.63)	1019(56.36)	1793(99.50)	1127(98.77)
Yes	1251(21.22)	439(38.37)	789(43.64)	9(0.50)	14(1.23)
Dyslipidemia						< 0.0001
No	5371(92.59)	1055(93.45)	1524(86.10)	1727(97.13)	1065(94.75)
Yes	430(7.41)	74(6.55)	246(13.90)	51(2.87)	59(5.25)
Glycemic_statuses						< 0.0001
DM	818(13.89)	151(13.22)	439(24.25)	102(5.66)	126(11.09)
pre_DM	2675(45.42)	533(46.67)	895(49.45)	727(40.37)	520(45.77)
NGR	2396(40.69)	458(40.11)	476(26.30)	972(53.97)	490(43.13)
CKM						< 0.0001
0	511(8.62)	0(0.00)	0(0.00)	439(24.19)	72(6.27)
1	1137(19.19)	135(11.77)	62(3.42)	766(42.20)	174(15.16)
2	2319(39.14)	600(52.31)	826(45.51)	345(19.01)	548(47.74)
3	1958(33.05)	412(35.92)	927(51.07)	265(14.60)	354(30.84)
Antihypertensive treatment						< 0.0001
No	4837(84.02)	800(73.53)	1130(65.39)	1785(99.06)	1122(98.51)
Yes	920(15.98)	288(26.47)	598(34.61)	17(0.94)	17(1.49)
Antihyperglycemic treatment						< 0.0001
No	5764(97.28)	1108(96.60)	1714(94.44)	1802(99.28)	1140(99.30)
Yes	161(2.72)	39(3.40)	101(5.56)	13(0.72)	8(0.70)
Lipid-lowering treatment						< 0.0001
No	5322(95.72)	1043(96.31)	1501(91.36)	1718(98.51)	1060(97.25)
Yes	238(4.28)	40(3.69)	142(8.64)	26(1.49)	30(2.75)
CVD						< 0.0001
No	4458(75.24)	821(71.58)	1237(68.15)	1494(82.31)	906(78.92)
Yes	1467(24.76)	326(28.42)	578(31.85)	321(17.69)	242(21.08)
Heart disease						< 0.0001
No	4819(81.33)	896(78.12)	1396(76.91)	1573(86.67)	954(83.10)
Yes	1106(18.67)	251(21.88)	419(23.09)	242(13.33)	194(16.90)
Stroke						< 0.0001
No	5400(91.14)	1030(89.80)	1577(86.89)	1715(94.49)	1078(93.90)
Yes	525(8.86)	117(10.20)	238(13.11)	100(5.51)	70(6.10)

Bold values indicate statistically significant differences

Association of AIP and eGDR with the risk of CVD, heart disease and stroke

In the CKM 0–3 stage cohort, the cumulative incidence of new-onset CVD, heart disease, and stroke demonstrated a progressive increase with extended follow-up durations, reaching up to approximately nine years. The group characterized by high AIP and low eGDR consistently exhibited elevated cumulative risk curves for all three outcomes compared to the alternative group. This disparity was evident early in the follow-up period and progressively widened over time, as indicated by the log-rank test (all p < 0.0001) (Figure S4 [see Additional file 1]). Further stratification based on combined AIP and eGDR values revealed a more pronounced separation among the four risk curves. The group with high AIP and low eGDR consistently showed the highest cumulative risk for CVD, heart disease, and stroke, whereas the group with low AIP and high eGDR demonstrated the lowest risk (all p < 0.0001) (Fig. 2).

Fig. 2.

Kaplan–Meier plot of CVD, heart disease and stroke by different AIP-eGDR groups. AIP: atherogenic index of plasma; CVD: cardiovascular disease; eGDR: estimated glucose disposal rate

To further investigate the associations, we conducted both independent and joint stratification analyses of AIP and eGDR, employing Cox proportional hazards regression models with varying levels of adjustment. In the fully adjusted Model 3 of the independent stratification analysis, the high AIP group exhibited a 13% increased risk of CVD (95% CI 1.02–1.26), a 9% increased risk of heart disease (95% CI 1.02–1.24), and a 24% increased risk of stroke (95% CI 1.03–1.49) compared to the low AIP group. Conversely, the low eGDR group demonstrated a 22% increased risk of CVD (95% CI 1.06–1.41), an 18% increased risk of heart disease (95% CI 1.00–1.38), and a 39% increased risk of stroke (95% CI 1.09–1.77) relative to the high eGDR group. Following joint stratification by AIP and eGDR, with the "low AIP & high eGDR" category serving as the reference, the "high AIP & high eGDR" group showed an 18% increased risk of CVD (95% CI 1.11–1.40), the "low AIP & low eGDR" group exhibited a 26% increased risk (95% CI 1.06–1.51), and the "high AIP & low eGDR" group demonstrated a 35% increased risk (95% CI 1.14–1.59) (all p < 0.001). In relation to heart disease outcomes, the three groups exhibited increased risk levels of 25% (95% CI 1.03–1.52), 28% (95% CI 1.05–1.55), and 32% (95% CI 1.08–1.62), respectively, with a statistically significant trend observed (p = 0.02). Regarding stroke outcomes, the "high AIP & high eGDR" group demonstrated a 10% increased risk (95% CI 1.01–1.50), while the "low AIP & low eGDR" group exhibited a 25% increased risk (95% CI 1.12–1.71). Notably, the "high AIP & low eGDR" group presented the most pronounced risk elevation at 59% (95% CI 1.19–2.12), indicating a significant overall increasing trend (all p < 0.001) (Table 2). We further adjusted for baseline CKM stage in an additional model. After this adjustment, the effect estimates were attenuated. The high AIP and low eGDR group remained associated with CVD HR 1.27 95% CI 1.05–1.53 and stroke HR 1.46 95% CI 1.06–2.00, whereas the association with heart disease was directionally similar but borderline HR 1.25 95% CI 1.00–1.55. The same pattern of attenuation was also observed in the separate analyses of AIP and eGDR (Table S8 [see Additional file 2]).

Table 2.

Multivariable cox regression analysis of new-onset CVD, heart disease, and stroke by eGDR and AIP among individuals with CKM syndrome stages 0–3

	Case	Incidence rate (95% CI)	Crude model		Model 1		Model 2		Model 3
CVD
AIP
Low AIP	647	26.65 (24.64–28.79)	ref		ref		ref		ref
High AIP	820	34.73 (32.39–37.19)	1.31(1.19,1.46)	< 0.0001	1.31(1.18,1.46)	< 0.0001	1.24(1.11,1.38)	< 0.0001	1.13(1.02,1.26)	0.03
eGDR
High eGDR	563	22.84 (21.00–24.81)	ref		ref		ref		ref
Low eGDR	904	38.89 (36.40–41.51)	1.73(1.56,1.92)	< 0.0001	1.63(1.46,1.81)	< 0.0001	1.46(1.29,1.66)	< 0.0001	1.22(1.06,1.41)	0.01
eGDR-AIP
Low AIP & High eGDR	321	21.09 (18.85–23.53)	ref		ref		ref		ref
High AIP & High eGDR	242	25.67 (22.54–29.11)	1.22(1.03,1.44)	0.02	1.24(1.05,1.46)	0.01	1.22(1.03,1.45)	0.02	1.18(1.11,1.40)	0.04
Low AIP & Low eGDR	326	35.98 (32.18–40.11)	1.73(1.48,2.02)	< 0.0001	1.62(1.39,1.89)	< 0.0001	1.46(1.23,1.74)	< 0.0001	1.26(1.06,1.51)	0.01
High AIP & Low eGDR	578	40.75 (37.50–44.22)	1.97(1.72,2.26)	< 0.0001	1.87(1.63,2.15)	< 0.0001	1.67(1.43,1.96)	< 0.0001	1.35(1.14,1.59)	< 0.001
p for trend				< 0.0001		< 0.0001		< 0.0001		< 0.001
Heart disease
AIP
Low AIP	493	19.84 (18.12–21.67)	ref		ref		ref		ref
High AIP	613	25.14 (23.19–27.21)	1.27(1.13,1.43)	< 0.0001	1.25(1.11,1.41)	< 0.001	1.2(1.06,1.35)	0.004	1.09(1.02,1.24)	0.03
eGDR
High eGDR	436	17.37 (15.78–19.08)	ref		ref		ref		ref
Low eGDR	670	27.75 (25.69–29.94)	1.61(1.43,1.82)	< 0.0001	1.51(1.34,1.71)	< 0.0001	1.4(1.21,1.62)	< 0.0001	1.18(1.00,1.38)	0.05
eGDR-AIP
Low AIP & High eGDR	242	15.63 (13.72–17.72)	ref		ref		ref		ref
High AIP & High eGDR	194	20.18 (17.44–23.23)	1.3(1.07,1.57)	0.01	1.3(1.08,1.57)	0.01	1.3(1.08,1.58)	0.01	1.25(1.03,1.52)	0.02
Low AIP & Low eGDR	251	26.79 (23.58–30.32)	1.73(1.45,2.07)	< 0.0001	1.63(1.36,1.94)	< 0.0001	1.53(1.26,1.87)	< 0.0001	1.28(1.05,1.55)	0.01
High AIP & Low eGDR	419	28.36 (25.71–31.21)	1.84(1.57,2.15)	< 0.0001	1.72(1.46,2.02)	< 0.0001	1.6(1.33,1.91)	< 0.0001	1.32(1.08,1.62)	0.01
p for trend				< 0.0001		< 0.0001		< 0.0001		0.02
Stroke
AIP
Low AIP	217	8.35 (7.28–9.54)	ref		ref		ref		ref
High AIP	308	11.99 (10.69–13.41)	1.44(1.21,1.72)	< 0.0001	1.49(1.25,1.78)	< 0.0001	1.36(1.14,1.63)	< 0.001	1.24(1.03,1.49)	0.02
eGDR
High eGDR	170	6.50 (5.56–7.55)	ref		ref		ref		ref
Low eGDR	355	13.92 (12.51–15.45)	2.16(1.80,2.60)	< 0.0001	2.07(1.72,2.50)	< 0.0001	1.72(1.38,2.15)	< 0.0001	1.39(1.09,1.77)	0.01
eGDR-AIP
Low AIP & High eGDR	100	6.23 (5.07–7.58)	ref		ref		ref		ref
High AIP & High eGDR	70	6.92 (5.40–8.74)	1.11(1.02,1.51)	0.005	1.15(1.05,1.57)	0.006	1.11(1.01,1.51)	0.01	1.1(1.01,1.50)	0.04
Low AIP & Low eGDR	117	11.78 (9.74–14.12)	1.91(1.46,2.49)	< 0.0001	1.79(1.37,2.34)	< 0.0001	1.47(1.09,1.98)	0.01	1.25(1.12,1.71)	0.03
High AIP & Low eGDR	238	15.29 (13.41–17.36)	2.48(1.97,3.14)	< 0.0001	2.47(1.95,3.12)	< 0.0001	2.01(1.54,2.63)	< 0.0001	1.59(1.19,2.12)	0.002
p for trend				< 0.0001		< 0.0001		< 0.0001		< 0.001

Incidence rate: Per 1000 person-years

Crude model: unadjusted

Model 1: adjusted for Age, Sex, Marriage, Education, Location, Smoke, Drink

Model 2: further adjusted for SBP, Height, eGFR, LDL-C, hs-CRP

Model 3: further adjusted for Hypertension, Dyslipidemia, Diabetes

Restricted cubic spline analysis

In subsequent dose–response analyses, we employed restricted cubic splines based on the fully adjusted Model 3 to evaluate potential non-linear associations between AIP and eGDR with the risk of three distinct outcomes, using the median of each indicator as a reference point. The findings indicated a non-linear, increasing association between AIP and the risk of developing new-onset CVD, heart disease, and stroke. Conversely, eGDR demonstrated a monotonically decreasing relationship with the risk of these outcomes. Specifically, the association between eGDR and both CVD and heart disease exhibited a non-linear pattern characterized by an initial decrease followed by a plateau, whereas the relationship with stroke was approximately linear. When the median was used as a reference, AIP values above the median were linked to an elevated risk, whereas eGDR values above the median were associated with a reduced risk. These continuous dose–response results were consistent with the findings from the four-quadrant stratification, which identified the highest risk in the group characterized by high AIP and low eGDR (Fig. 3). The continuous analyses were consistent with the four-group results and indicate that the observed pattern was not solely driven by the use of median-based cut points.

Fig. 3.

Restricted cubic spline analyses demonstrated the dose–response relationships of AIP and eGDR with the risks of new-onset CVD, heart disease, and stroke in individuals with CKM Syndrome at Stages 0–3. A–C respectively presented the associations of AIP with the risks of new-onset CVD, heart disease, and stroke, while D–F respectively illustrated the associations of eGDR with the risks of the three aforementioned outcomes. The curves represented hazard ratio estimates based on the Cox proportional hazards model, and the shaded areas indicated 95% confidence intervals. The median values of AIP and eGDR were adopted as the reference levels. P for overall represented the P-value for the overall association test, and P for nonlinearity denoted the P-value for the nonlinear component test. AIP atherogenic index of plasma eGDR estimated glucose disposal rate CVD cardiovascular disease CKM cardiovascular–kidney–metabolic. AIP: atherogenic index of plasma; CVD: cardiovascular disease; CKM: Cardiovascular–Kidney–Metabolic; eGDR: estimated glucose disposal rate

Prediction

We employed time-dependent Harrell's C-index, derived from Model 3, to assess the predictive efficacy of AIP, eGDR, and their combined application in forecasting the incidence of CVD, heart disease, and stroke. The findings indicated that within the CKM 0–3 stage cohort, the discriminative capabilities of AIP and eGDR individually were generally comparable. However, the combined use of AIP and eGDR demonstrated superior predictive performance compared to each marker alone (Fig. 4). Specifically, the Harrell's C-indices for CVD prediction were 0.643 (95% CI 0.625–0.652) for AIP, 0.642 (95% CI 0.624–0.655) for eGDR, and 0.657 (95% CI 0.646–0.665) for the combination of AIP and eGDR (Fig. 4A). For heart disease prediction, the indices were 0.628 (95% CI 0.608–0.650), 0.629 (95% CI 0.608–0.650), and 0.642 (95% CI 0.631–0.663) (Fig. 4B), and for stroke prediction, they were 0.674 (95% CI 0.647–0.700), 0.675 (95% CI 0.650–0.702), and 0.688 (95% CI 0.672–0.725) (Fig. 4C). The discrimination of the baseline model was moderate. Adding both AIP and eGDR yielded statistically significant but numerically modest improvements in Harrell’s C index (ΔHarrell’s C: 0.036 for CVD, 0.028 for heart disease, and 0.037 for stroke; all P ≤ 0.008). Similar patterns were observed for the IDI and continuous NRI (Fig. 5).

Fig. 4.

The predictive performance of AIP and eGDR for the risks of CVD, heart disease, and stroke in individuals with CKM Syndrome at stages 0–3. Predictive capacity of the AIP and eGDR on the CVD (A), heart disease (B) and stroke (C) risk. AIP: atherogenic index of plasma; CVD: cardiovascular disease; CKM: Cardiovascular–Kidney–Metabolic; eGDR: estimated glucose disposal rate

Fig. 5.

Discrimination and incremental predictive value of AIP, eGDR, and their combination for CVD, heart disease, and stroke (C-index, ΔC-index, IDI, and continuous NRI). AIP: atherogenic index of plasma; CVD: cardiovascular disease; eGDR: estimated glucose disposal rate

Interaction between AIP and eGDR

In the CKM 0–3 stage cohort, we evaluated the interaction between AIP and eGDR concerning the risk of CVD, heart disease, and stroke, employing both additive and multiplicative scales. In the fully adjusted Model 3, the additive interaction metrics for CVD outcomes were as follows: RERI =  − 0.09 (95% CI − 0.44 to 0.25), AP =  − 0.07 (95% CI − 0.36 to 0.17), and SI = 0.80 (95% CI 0.29 to 2.06), with a multiplicative interaction effect of 0.91 (95% CI 0.69 to 1.19). For heart disease outcomes, the additive interaction metrics were RERI =  − 0.21 (95% CI − 0.66 to 0.22), AP =  − 0.16 (95% CI − 0.57 to 0.15), and SI = 0.60 (95% CI 0.13 to 1.84), with a multiplicative interaction effect of 0.82 (95% CI 0.59 to 1.16). Regarding stroke outcomes, the additive interaction metrics were RERI = 0.24 (95% CI − 0.31 to 0.85), AP = 0.15 (95% CI − 0.24 to 0.42), and SI = 1.69 (95% CI 0.32 to 9.70), with a multiplicative interaction effect of 1.16 (95% CI 0.77 to 1.74). The confidence intervals for all interaction metrics encompassed the reference value indicating no interaction (RERI and AP included 0, while SI and the multiplicative interaction effect included 1), suggesting that neither additive nor multiplicative interactions between AIP and eGDR were statistically significant for CVD, heart disease, and stroke in the CKM 0–3 stage cohort (Table 3) (Figure S5 [see Additional file 1]).

Table 3.

Interaction between the AIP and eGDR on CVD, Heart disease and stroke risk in individuals at CKM stages 0–3

	Crude model Interactive effects (95% CI)	Model 1	Model 2	Model 3
CVD
Additive effect
RERI	0.02 (− 0.42–0.44)	0.01 (− 0.41–0.42)	− 0.01 (− 0.44–0.41)	− 0.09 (− 0.44–0.25)
AP	0.01 (− 0.23–0.21)	0.01 (− 0.24–0.20)	− 0.01 (− 0.29–0.22)	− 0.07 (− 0.36–0.17)
SI	1.02 (0.66–1.64)	1.01 (0.63–1.68)	0.99 (0.53–1.96)	0.80 (0.29–2.06)
Multiplicative effect	0.93 (0.72–1.22)	0.93 (0.71–1.21)	0.94 (0.70–1.25)	0.91 (0.69–1.19)
Heart Disease
Additive effect
RERI	− 0.19 (− 0.70–0.29)	− 0.21 (− 0.69–0.25)	− 0.23 (− 0.73–0.25)	− 0.21 (− 0.66–0.22)
AP	− 0.10 (− 0.42–0.14)	− 0.12 (− 0.44–0.13)	− 0.14 (− 0.51–0.14)	− 0.16 (− 0.57–0.15)
SI	0.82 (0.48–1.40)	0.77 (0.44–1.39)	0.72 (0.35–1.46)	0.60 (0.13–1.84)
Multiplicative effect	0.82 (0.60–1.11)	0.81 (0.60–1.10)	0.80 (0.58–1.12)	0.82 (0.59–1.16)
Stroke
Additive effect
RERI	0.46 (− 0.36–1.27)	0.53 (− 0.27–1.33)	0.43 (− 0.31–1.18)	0.24 (− 0.31–0.85)
AP	0.19 (− 0.17–0.43)	0.21 (− 0.13–0.45)	0.21 (− 0.18–0.48)	0.15 (− 0.24–0.42)
SI	1.45 (0.76–3.07)	1.56 (0.80–3.43)	1.74 (0.67–6.86)	1.69 (0.32–9.70)
Multiplicative effect	1.17 (0.78–1.75)	1.20 (0.80–1.81)	1.23 (0.79–1.93)	1.16 (0.77–1.74)

Subgroup analyses

We performed subgroup analyses and interaction tests, designating the Low AIP & High eGDR group as the reference category, to evaluate the robustness of the association between the four AIP/eGDR combinations and CVD, heart disease, and stroke outcomes. The stratification factors included age, sex, marital status, education level, urban versus rural residence, smoking and alcohol consumption status, history of hypertension and dyslipidemia, glucose metabolic status, and CKM staging. The overall findings indicated that the effect estimates across each subgroup were generally consistent with those of the primary analysis. Furthermore, the interaction test P-values were all greater than 0.05, suggesting no significant effect modification (Table S9-11 [see Additional file 2]).

Sensitivity analyses

To evaluate the robustness of the primary findings, we performed six sets of sensitivity analyses: 1)We further adjusted for baseline medication use in Model III, including antihypertensive, antidiabetic, and lipid-lowering medications. 2)We excluded participants with missing data. 3)We excluded participants who had already received antihypertensive, antidiabetic, or lipid-lowering treatment at baseline to minimize the influence of treatment intervention on the exposure-outcome relationship. 4)We repeated the main analysis after excluding individuals in CKM stage 0 to assess consistency across populations with varying levels of cardiovascular-renal metabolic risk burden. 5)We excluded participants with a follow-up duration of less than two years to mitigate the potential associations of reverse causality and early events on the results. 6)Given that the study participants were middle-aged and elderly individuals, there was a possibility of non-CVD/heart disease/stroke-related deaths occurring during the follow-up period, which could impede the observation of outcome events. To address this, we employed the Fine-Gray subdistribution hazard model, designating non-corresponding outcome deaths as competing events, to perform a competing risk sensitivity analysis of the primary analysis results. We reported the subdistribution hazard ratio (sHR) and its 95% confidence interval. Utilizing the Fine-Gray subdistribution model, the effect sizes of the AIP and eGDR on CVD, heart disease, and stroke were reduced but remained statistically significant. The relationship between AIP and eGDR and the risk of CVD, heart disease, and stroke remained generally stable across multiple sensitivity analyses, thereby supporting the robustness of the main conclusions of this study (Table S12-S17 [see Additional file 2]).

Discussion

In the context of CKM syndrome, which underscores the importance of proactive risk management, we conducted a systematic evaluation of the associations of AIP and eGDR on the incidence of new-onset cardiovascular events. This study involved 5,925 Chinese individuals aged 45 years and older, categorized within CKM stages 0–3, utilizing data from the nationally representative prospective cohort study, CHARLS. Over a follow-up period of up to nine years, 1,467 new cardiovascular events were documented among the participants, comprising 1,106 cases of heart disease and 525 cases of stroke. These findings reinforce the notion that CKM stages 0–3 represent a critical period for accelerated risk accumulation [3, 34]. After comprehensive adjustment for potential confounders, AIP consistently demonstrated a positive association with the risk of all three cardiovascular outcomes, whereas eGDR exhibited a stable negative association. We employed the median for stratification, based on the distribution within the study sample. The findings indicated that, in comparison to the group with low AIP and high eGDR, the group characterized by high AIP and low eGDR exhibited the highest incidence and risk ratios for CVD, heart disease, and stroke. Specifically, the risk increased by 35% (HR: 1.35; 95% CI 1.14–1.59), 32% (HR: 1.32; 95% CI 1.08–1.62), and 59% (HR: 1.59; 95% CI 1.19–2.12), respectively. Further analysis using restricted cubic splines suggested that AIP demonstrated a generally non-linear increasing association with risk. In contrast, eGDR exhibited a decreasing trend in CVD and heart disease risk, followed by a plateau, and a more linear decrease in stroke risk. These results are therefore best understood as describing gradients of cardiometabolic risk within a metabolically defined population rather than demonstrating independent etiologic pathways [35]. Given that both AIP and eGDR are derived from metabolic variables—variables that overlap with the metrics defining early CKM stages—we fitted an additional model, adjusted for baseline CKM staging, to directly address this conceptual overlap. The aim was to quantify the extent of shared information between AIP/eGDR and the CKM construct, rather than to treat this as a primary causal model. After this adjustment, the hazard ratio for the high AIP and low eGDR group decreased from 1.35 to 1.27 for CVD, from 1.32 to 1.25 for heart disease, and from 1.59 to 1.46 for stroke. The stage adjusted results do not negate the overall pattern, but they do indicate that an important part of the observed signal is shared with the metabolic structure already embedded in CKM staging. Furthermore, subgroup analyses and multiple sensitivity analyses produced consistent results, thereby reinforcing the robustness of the primary findings.

Our findings align with existing literature. Previous research documented 620 cases of CVD among 3,429 participants over a three-year follow-up period, indicating that poor long-term control of the AIP is consistently linked to an elevated risk of early CVD in CKM [36]. Furthermore, in the CHARLS cohort, also in CKM stages 0–3, 455 new-onset stroke cases were reported during a 6.8-year follow-up, showing that the positive association between AIP and stroke outcomes can also be observed in the early stages of CKM [37]. Additionally, Zhang et al.'s analysis of the UK Biobank CKM stages 0–3 cohort, comprising 325,312 participants, demonstrated that the eGDR is not only associated with the risk of CVD outcomes in the early stages of CKM but also offers incremental predictive value [38]. Our findings indicate that within the CKM 0–3 cohort, individuals exhibiting an AIP greater than -0.05 experienced a statistically significant 13% increase in the risk of CVD (95% CI 1.02–1.26), a 9% increase in the risk of heart disease (95% CI 1.02–1.24), and a 24% increase in the risk of stroke (95% CI 1.03–1.49). Furthermore, individuals with an eGDR below 10.15 demonstrated a significantly elevated risk of CVD by 22% (95% CI 1.06–1.41), heart disease by 18% (95% CI 1.00–1.38), and stroke by 39% (95% CI 1.09–1.77). Within the framework of CKM stages 0–3, this study simultaneously evaluated CVD, heart disease, and stroke in the same nationally representative cohort by examining joint stratification, dose–response relationships, interaction, and incremental predictive value. The present study extends prior single-marker analyses by showing that the combined adverse profile of high AIP and low eGDR marks a subgroup with consistently higher event rates. The joint pattern of high AIP and low eGDR is best understood as a practical way of summarizing a heavier cardiometabolic burden within CKM stages 0–3, rather than as evidence of a distinct CKM phenotype or a novel biological subtype. Likewise, the four joint categories created from cohort medians should be regarded as internal comparison strata designed to facilitate balanced group contrasts, not as clinically established thresholds.

AIP and eGDR are clinically accessible composite markers that summarize different, although closely related, domains of cardiometabolic dysfunction and have been employed in numerous studies for cardiovascular risk stratification and prediction. In the setting of CKM stages 0–3, the interpretation of AIP and eGDR should begin with the recognition that neither index is external to the syndrome’s metabolic architecture. From a biological perspective, AIP and eGDR reflect related aspects of cardiometabolic dysfunction. AIP summarizes a more atherogenic lipid profile characterized by higher triglyceride rich lipoprotein burden and lower HDL-C, a pattern linked to remnant cholesterol exposure, lipoprotein remodeling, and plaque vulnerability [39, 40]. eGDR, in contrast, serves as a composite surrogate of insulin resistance based on waist circumference, hypertension status, and HbA1c, and lower values have been associated with vascular dysfunction, inflammation, and adverse long term outcomes [41–44]. However, within early CKM these processes are closely interconnected and are not external to the syndrome itself. Obesity, dysglycemia, hypertension, and dyslipidemia cluster together biologically and clinically, which makes it difficult to isolate truly separate pathways from indices derived from the same metabolic substrate. For this reason, our data are more compatible with an interpretation of partial overlap than with a model of two biologically independent axes. In early CKM these processes are partly overlapping and clinically entangled; for that reason, combined categorization can still refine within-stage risk gradients by capturing different projections of the same shared cardiometabolic burden. The close alignment between joint AIP/eGDR categories and baseline CKM stage distribution, together with the attenuation of effect estimates after additional adjustment for CKM stage, indicates that a meaningful part of the observed signal is shared with the metabolic structure already embedded in CKM staging, even though the remaining gradient continues to discriminate risk within that structure [45, 46].

Against this background, the interaction analyses are informative precisely because they define the boundary of interpretation. The absence of significant interaction on either scale argues against a supra-additive or synergistic effect, and instead supports the interpretation that the joint high-risk pattern reflects clustering of overlapping but non-identical adverse metabolic features within a common cardiometabolic network [23, 34, 47, 48]. Specifically, our findings support combined AIP/eGDR assessment as a tool for risk summarization and refinement of within-stage cardiovascular gradients.

This study also has several strengths. It used a large nationally representative cohort with long-term follow-up, focused on the clinically important pre-event stages of CKM, and assessed overall CVD alongside heart disease and stroke rather than relying on a single endpoint. In addition, the analytical framework incorporated spline analyses, interaction testing, multiple sensitivity analyses, competing risk analyses, and further adjustment for baseline CKM stage, allowing a more transparent assessment of both the robustness and the limits of the observed associations.

This study has several limitations. First and most importantly, the exposures were partially constructed from metabolic components that overlap with the definition of early CKM stages. Accordingly, the observed associations should not be interpreted as evidence of independent causal pathways or a new CKM subtype, but rather as a re-expression of cardiometabolic risk within a metabolically defined population. This conceptual overlap constrains etiologic inference and is central to the interpretation of the study. Second, although extensive covariate adjustment was performed, residual confounding from diet, physical activity, socioeconomic factors, medication adherence, and other unmeasured characteristics cannot be excluded in an observational cohort. Third, AIP and eGDR were assessed at baseline only, which precluded evaluation of longitudinal trajectories, cumulative exposure, and transitions across CKM stages over time. Fourth, the cardiovascular outcomes were based mainly on self-reported physician diagnoses, and the composite CVD endpoint did not allow adjudication of more specific disease phenotypes. Stroke subtype information was also unavailable. Fifth, the four joint categories were defined using cohort-specific median values and should be viewed as internal strata for comparison rather than clinically validated cut points. Sixth, the study population was restricted to Chinese adults aged 45 years and older. Differences in age structure, genetics, diet, adiposity pattern, health care access, and background cardiovascular risk may limit the transportability of the findings to younger individuals or non-Chinese populations.

Future studies should address the specific questions left unresolved by the present analysis. Longitudinal cohorts with repeated measurements are needed to determine whether trajectories or cumulative exposure to AIP and eGDR improve risk stratification beyond single baseline values. It will also be important to compare the added value of these indices directly against CKM stage itself and against their individual component variables in order to clarify whether they provide information beyond existing metabolic constructs. In addition, external validation, calibration assessment, and evaluation of net clinical benefit are needed in younger populations, non Chinese cohorts, and different care settings. Clinically meaningful thresholds and reassessment intervals should be established before these indices are considered for broader implementation.

From an implementation standpoint, the most plausible clinical role of AIP and eGDR in CKM stages 0–3 is as adjunctive, low-cost summary markers embedded within existing CKM and cardiovascular risk-assessment pathways, rather than as stand-alone screening tests or replacements for established risk equations. Because both indices are derived from variables already captured in routine care—triglycerides, HDL-C, waist circumference, blood pressure or hypertension status, and HbA1c—they can be calculated at the same encounter in which CKM stage is assigned and global cardiovascular risk is estimated using established tools, where locally appropriate, including CKM-oriented equations such as PREVENT. Their greatest practical value may lie in within-stage refinement: when stage-based assessment or global risk estimation appears heterogeneous, a concordantly adverse profile of high AIP and low eGDR should not be interpreted as defining a distinct phenotype, but it may help identify individuals with a heavier residual cardiometabolic burden who warrant closer review of central adiposity, blood pressure control, glycemic status, lipid optimization, kidney function, medication adherence, and lifestyle support. A pragmatic strategy is therefore to recalculate AIP and eGDR whenever the underlying measurements are repeated rather than to create a separate testing pathway. For clinically stable individuals in CKM stages 0–1, annual reassessment aligned with routine lipid and kidney surveillance may be sufficient; by contrast, in CKM stages 2–3, after initiation or intensification of lipid-lowering, antihypertensive, or glucose-lowering therapy, or in those with a persistently adverse joint profile, reassessment at approximately 3- to 6-month follow-up visits may be more informative. This schedule is consistent with current chronic metabolic care frameworks, in which blood pressure is measured at every routine visit or at least every 6 months, lipid profiles are obtained at initial evaluation and annually thereafter, repeated 4–12 weeks after initiation or dose change of lipid-lowering therapy, and UACR/eGFR are assessed at least annually and more frequently in established CKD49, 50. Such use may be particularly attractive in primary care and resource-limited settings because it does not require specialized assays, insulin measurements, or advanced imaging, and can instead be incorporated into standardized, risk-based chronic disease workflows and team-based care models that prioritize early identification of high-risk individuals and more efficient allocation of limited preventive resources [51, 52]. Nonetheless, clinically actionable thresholds, external validation across diverse populations, and demonstration of net clinical benefit beyond established CKM and cardiovascular risk tools remain necessary before broad routine implementation can be recommended.

Conclusion

Among adults with CKM stages 0 to 3, higher AIP and lower eGDR were associated with higher risks of incident CVD, heart disease, and stroke, and the combination of high AIP and low eGDR identified the group with the highest observed risk. Joint assessment of these indices provided a modest improvement in risk discrimination and reclassification beyond conventional covariates.

Supplementary Information

Below is the link to the electronic supplementary material.

Abbreviations

AHA

American Heart Association

AIP

Atherogenic index of plasma

ANOVA

Analysis of variance

AP

Attributable proportion due to interaction

BMI

Body mass index

CHARLS

China Health and Retirement Longitudinal Study

CI

Confidence interval

CKD

Chronic kidney disease

CKM

Cardiovascular–Kidney–Metabolic

CVD

Cardiovascular disease

DALYs

Disability-adjusted life years

DBP

Diastolic blood pressure

eGDR

Estimated glucose disposal rate

eGFR

Estimated glomerular filtration rate

FBG/Fglu

Fasting blood glucose

GBD

Global Burden of Disease

HbA1c

Glycated hemoglobin A1c

HDL-C

High-density lipoprotein cholesterol

HR

Hazard ratio

hs-CRP

High-sensitivity C-reactive protein

IDI

Integrated discrimination improvement

IQR

Interquartile range

LDL-C

Low-density lipoprotein cholesterol

MICE

Multiple imputation by chained equations

NRI

Net Reclassification Improvement

PMM

Predictive mean matching

ROC

Receiver operating characteristic

RERI

Relative excess risk due to interaction

SD

Standard deviation

SI

Synergy index

SBP

Systolic blood pressure

Scr

Serum creatinine

sHR

Subdistribution hazard ratio

STROBE

Strengthening the Reporting of Observational Studies in Epidemiology

TC

Total cholesterol

TG

Triglycerides

UI

Uncertainty interval

VIF

Variance inflation factor

WC

Waist circumference

Author contributions

Y.F. designed the study and collected the data. Z.S. analyzed the data and drafted the manuscript. M.W. and Y.G. prepared the figures and revised the manuscript. X.L. revised and reviewed the manuscript. All authors have read and approved the final manuscript.

Funding

This work was supported by the National Natural Science Foundation of China (42275197), the Key Projects of Tianjin Municipal Health Commission (TJWJ2023XK007).

Data availability

Datasets used in this study are available from the CHARLS repository: [http://charls.pku.edu.cn/en].

Declarations

Ethics approval and consent to participate

The studies involving human participants were reviewed and approved by the Ethics Review Committee of Peking University. All participants provided written informed consent to participate in this study (IRB No. 00001052–11015).

Consent for publication

Not applicable.

Competing interests

The authors declare no competing interests.