Association between the combined cholesterol, high-density lipoprotein, glucose index and frailty index and new-onset cardiovascular disease in individuals with cardiovascular-kidney-metabolic syndrome stages 0–3: evidence from CHARLS

Abstract

Background

Early-stage cardiovascular-kidney-metabolic (CKM) syndrome, encompassing stages 0–3, represents a critical window for identifying individuals at risk before the development of clinical cardiovascular disease (CVD). Metabolic dysregulation and frailty-related physiological vulnerability may jointly contribute to cardiovascular susceptibility in middle-aged and older adults. However, conventional single-domain markers may not adequately capture this combined risk burden. This study aimed to evaluate the association and predictive utility of the combined cholesterol, high-density lipoprotein, glucose index and frailty index (CHG-FI) for new-onset CVD, heart disease, and stroke among individuals with early-stage CKM syndrome.

Methods

This cohort study used longitudinal data from the China Health and Retirement Longitudinal Study (CHARLS, 2011–2020), including 4,603 participants with CKM stages 0–3. New-onset CVD, defined as the occurrence of heart disease or stroke during follow-up, was the primary outcome; new-onset heart disease and stroke were analyzed separately as secondary outcomes. CHG-FI was calculated as the product of the cholesterol, high-density lipoprotein, glucose (CHG) index and frailty index (FI). Multivariable Cox proportional hazards models were used to assess the associations of CHG-FI with new-onset cardiovascular outcomes. Restricted cubic splines were applied to examine nonlinear dose–response relationships. Joint-exposure analyses were used to characterize CVD risk across combined CHG and FI categories, and multiplicative and additive interaction analyses were performed to assess whether CHG and FI interacted in relation to new-onset CVD. Predictive performance was compared with CHG, FI, triglyceride-glucose index (TyG), atherogenic index of plasma (AIP), and metabolic score for insulin resistance (METS-IR) using receiver operating characteristic curves, DeLong tests, net reclassification improvement (NRI), and integrated discrimination improvement (IDI). Formal incremental modeling was further performed by separately adding CHG-FI and comparator indices to the fully adjusted base model.

Results

During a median follow-up of 9 years, 1,189 participants (25.8%) developed new-onset CVD. In fully adjusted models, each one-unit increase in CHG-FI was associated with higher risks of new-onset heart disease (HR = 1.59; 95% CI 1.42–1.77), stroke (HR = 1.56; 95% CI 1.32–1.83), and composite CVD (HR = 1.56; 95% CI 1.41–1.72). Participants in the highest CHG-FI quartile had more than two-fold higher risks of all cardiovascular outcomes compared with those in the lowest quartile. Restricted cubic spline analyses showed nonlinear associations, with cardiovascular risk increasing more steeply across lower to moderate CHG-FI levels. In joint-exposure analyses, participants with both high CHG and high FI had the highest risk of new-onset CVD (HR = 2.05; 95% CI 1.71–2.47); however, no statistically significant multiplicative or additive interaction was observed. CHG-FI showed the highest AUCs for new-onset CVD (0.624) and heart disease (0.622) among the evaluated indices and statistically outperformed CHG, FI, TyG, AIP, and METS-IR for both outcomes; however, the absolute discriminative performance was modest. For new-onset CVD, CHG-FI showed statistically significant reclassification improvements compared with CHG, TyG, AIP, and METS-IR, with NRI values of 13.85%, 9.16%, 11.02%, and 20.90%, respectively, and yielded small but statistically significant IDI improvements ranging from 0.20 to 0.48% compared with all reference indices. In formal incremental models, adding CHG-FI to the fully adjusted base model provided statistically significant but limited improvement, with only small absolute gains over FI alone.

Conclusions

Higher CHG-FI was independently associated with increased risks of new-onset CVD, heart disease, and stroke among middle-aged and older adults with early-stage CKM syndrome. Participants with both high CHG and high FI had the greatest risk of new-onset CVD, although this pattern did not reflect a statistically significant multiplicative or additive interaction. CHG-FI showed statistically significant but modest discriminative performance, and the incremental improvements beyond comparator indices and clinical covariates were limited in magnitude. These findings suggest that CHG-FI may serve as a complementary marker reflecting integrated glucose–lipid metabolic dysregulation and frailty-related vulnerability, but it should not be considered a standalone clinical prediction tool and requires further validation before clinical application.

Supplementary Information

The online version contains supplementary material available at https://doi.org/10.1186/s12933-026-03293-y.

Research insights

What is currently known about this topic?

Metabolic dysregulation and frailty are both important contributors to cardiovascular risk in middle-aged and older adults. The cholesterol, high-density lipoprotein, glucose (CHG) index has been proposed as a marker of glucose–lipid metabolic disturbance, whereas the frailty index (FI) captures multisystem physiological vulnerability. However, whether integrating CHG with FI improves cardiovascular risk assessment in individuals with early-stage cardiovascular-kidney-metabolic (CKM) syndrome remains insufficiently explored.

What is the key research question?

Is the combined index derived from the CHG index and FI, termed CHG-FI, associated with new-onset cardiovascular disease (CVD), heart disease, and stroke among middle-aged and older adults with CKM stages 0–3? Does CHG-FI provide incremental predictive value beyond its individual components and established metabolic indices, and do CHG and FI interact in relation to new-onset CVD?

What is new?

Higher CHG-FI was independently associated with increased risks of new-onset CVD, heart disease, and stroke, with nonlinear dose-response relationships. Participants with both high CHG and high FI had the greatest risk of new-onset CVD, although no statistically significant multiplicative or additive interaction was observed. CHG-FI showed modest but favorable predictive performance and improved risk reclassification beyond conventional metabolic indices.

How might this study influence clinical practice?

CHG-FI offers a feasible approach to integrating metabolic risk and frailty assessment in early-stage CKM syndrome. In community and primary care settings, this composite marker may help identify individuals at higher risk of new-onset CVD, heart disease, and stroke, especially those with both high CHG and high FI. These individuals may benefit from closer cardiovascular monitoring and earlier preventive strategies targeting both metabolic abnormalities and frailty-related functional vulnerability.

Introduction

Cardiovascular disease (CVD) remains the leading cause of mortality and disability worldwide [1, 2]. Research has demonstrated that diverse cardiovascular and metabolic risk factors are interwoven, necessitating the integration of multiple dimensions—including glucose metabolism, lipid metabolism, and frailty-related physiological vulnerability—beyond traditional single-domain risk assessment [3]. The recently proposed cardiovascular-kidney-metabolic (CKM) syndrome framework captures this complexity by highlighting the close interconnection among metabolic dysfunction, kidney impairment, and cardiovascular injury, positioning these conditions along a systemic and continuous pathophysiological continuum rather than as discrete entities [4–6]. Epidemiological evidence indicates that CKM-related risk states are highly prevalent among adult populations, underscoring a substantial public health burden [4, 6–8]. The concept of panvascular disease further supports the view that CKM-related cardiovascular risk should not be viewed solely as isolated cardiac events, but rather understood within the broader framework of systemic vascular pathology [9]. It is well established that cardiovascular complications represent a major source of the clinical burden of CKM syndrome, and the risk of adverse cardiovascular outcomes increases with CKM stage progression [10–12]. Accordingly, preventing disease progression in individuals with early-stage CKM syndrome, namely CKM stages 0–3, represents an important public health priority. Although overt clinical CVD has not yet occurred by definition in this population, cardiometabolic abnormalities and subclinical cardiovascular or renal injury may already emerge or accumulate across these stages. This highlights the urgent need to identify simple and practical markers for improving cardiovascular risk stratification among individuals with CKM stages 0–3, thereby facilitating earlier prevention and intervention to mitigate the disease burden.

Insulin resistance (IR) and related metabolic disturbances play central roles in the development of CVD [13–15]. Several surrogate indices, including the triglyceride-glucose (TyG) index, atherogenic index of plasma (AIP), and metabolic score for IR (METS-IR), have been widely used to evaluate metabolic risk and predict cardiovascular outcomes [16, 17]. Evidence from populations with CKM stages 0–3 further indicates that IR-related composite indices are associated with future cardiovascular risk, highlighting the importance of metabolic risk assessment before the development of clinical CVD [18–20]. However, these indices capture partly different metabolic dimensions. TyG and AIP are closely related to triglyceride-centered insulin resistance and atherogenic lipid balance, whereas METS-IR incorporates anthropometric and metabolic variables to approximate insulin resistance. In contrast, the cholesterol, high-density lipoprotein, glucose (CHG) index integrates total cholesterol, HDL-C, and glucose, thereby reflecting a glucose–cholesterol metabolic dimension that may not be identical to triglyceride-based or anthropometry-based surrogate indices [21, 22]. Accordingly, CHG was selected not to replace TyG, AIP, or METS-IR, but to evaluate whether this complementary glucose–cholesterol marker, when combined with FI, could provide additional information for cardiovascular risk characterization in early-stage CKM syndrome. However, metabolic markers alone may not fully capture the broader physiological vulnerability that contributes to cardiovascular susceptibility, particularly in middle-aged and older adults.

Frailty, commonly quantified using the frailty index (FI), reflects the cumulative burden of health deficits and reduced physiological reserve [23]. It has increasingly been recognized as a cardiovascular risk marker that may provide prognostic information beyond traditional cardiometabolic factors. In middle-aged and older adults, frailty often coexists with metabolic dysregulation, together representing complementary dimensions of cardiovascular vulnerability that may not be fully captured by metabolic indices alone. Metabolic dysfunction and frailty, although closely related, represent distinct biological domains: the former involves derangements in glucose and lipid metabolism, whereas the latter reflects cumulative multisystem physiological decline and reduced physiological reserve. This distinction supports the development of integrated markers that combine metabolic and frailty-related information. This has motivated the development of composite measures that integrate metabolic markers with frailty; for example, indices such as TyG-FI and AIP-FI have shown potential value for cardiovascular risk characterization in aging populations [16, 24]. More recently, Fan et al. evaluated CHG-FI within the cardiometabolic multimorbidity (CMM) framework and reported that CHG-FI was associated with incident heart disease, stroke, diabetes, and CMM, providing important evidence for the combined assessment of glucose–cholesterol metabolic burden and frailty-related vulnerability [25]. Nevertheless, CMM and CKM syndrome represent related but distinct clinical frameworks. CMM emphasizes the coexistence of multiple cardiometabolic diseases within the same individual, whereas CKM syndrome highlights the systemic interconnection among metabolic dysfunction, kidney impairment, and cardiovascular disease across a broader disease continuum. Therefore, although Fan et al. provided important evidence for CHG-FI in relation to CMM, the role of CHG-FI in predicting new-onset CVD among individuals with early-stage CKM syndrome remains insufficiently clarified. In addition, the comparative performance of CHG-FI relative to its individual components and established metabolic indices in this population requires careful evaluation. Accordingly, the present study extends prior frailty–metabolic research by evaluating whether the integration of a glucose–cholesterol metabolic marker with systemic frailty assessment provides additional information for cardiovascular risk characterization in early-stage CKM syndrome. CHG-FI is therefore not proposed as a completely new or standalone prediction model, but as an integrated marker aimed at capturing the combined burden of glucose–cholesterol metabolic dysregulation and frailty-related physiological vulnerability.

In this study, we used data from the China Health and Retirement Longitudinal Study (CHARLS) to investigate the association of CHG-FI with new-onset CVD and its components, including heart disease and stroke, among middle-aged and older adults with CKM stages 0–3. Rather than proposing CHG-FI as a completely new or standalone prediction model, we aimed to clarify its specific relevance within the early-stage CKM framework by integrating glucose–cholesterol metabolic dysregulation with frailty-related physiological vulnerability. We further examined the joint association and interaction between CHG and FI, assessed potential nonlinear dose–response relationships, and compared the predictive performance of CHG-FI with its individual components and established metabolic indices, including TyG, AIP, and METS-IR.

Methods

Study design

Information for the present investigation was extracted from the CHARLS. Managed by Peking University through its National School of Development, this nationwide tracking project (accessible via http://charls.pku.edu.cn/en) specifically surveys Chinese citizens aged forty-five and above [26]. The baseline cohort consisted of 17,708 participants recruited in 2011 (Wave 1), with follow-up assessments conducted across four subsequent waves in 2013, 2015, 2018, and 2020. Peking University’s Biomedical Ethics Review Committee granted formal authorization for this research framework (under document IRB00001052-11015). Before taking part, every individual provided their written informed consent. Furthermore, all investigative procedures completely complied with the ethical standards outlined within the Declaration of Helsinki.

Study population

Participants were selected according to predefined inclusion and exclusion criteria. First, 777 participants were excluded due to age < 45 years. Subsequently, 9,896 individuals were excluded due to the presence of advanced CKM syndrome (Stage 4) or missing data required for CKM staging. Additionally, 115 participants were removed due to insufficient data for calculating the CHG index or FI. Participants with loss to follow-up or incomplete covariate data were further excluded (n = 2,317). Finally, 4,603 participants with CKM syndrome stages 0–3 were included in the present analysis (Fig. 1).

Fig. 1.

Flowchart of CHARLS participants. CHARLS, China health and retirement longitudinal study; CKM, cardiovascular-kidney-metabolic; CHG, cholesterol, high-density lipoprotein, glucose; FI, frailty index

Definition of CKM syndrome stages

CKM syndrome stages were defined at baseline according to the American Heart Association scientific statement [10, 27]. Stage 0 was defined as the absence of CKM risk factors; stage 1 as overweight/obesity, abdominal obesity, or adiposity-related dysfunction without other metabolic risk factors or chronic kidney disease (CKD); stage 2 as the presence of metabolic risk factors, including hypertriglyceridemia, hypertension, metabolic syndrome, or diabetes, and/or moderate- to high-risk CKD; and stage 3 as subclinical cardiovascular disease (CVD), operationalized as a 10-year predicted CVD risk ≥ 20% based on the PREVENT equations or very high-risk CKD. Stage 4 was defined as clinical CVD based on self-reported history and was excluded from the present analysis.

Assessment of metabolic indices and CHG-FI

Fasting blood glucose (FBG), total cholesterol (TC), triglycerides (TG), and high-density lipoprotein cholesterol (HDL-C) were measured using enzymatic colorimetric methods [28]. Insulin resistance-related indices were calculated using established formulas as follows:

The FI was constructed based on the deficit accumulation model established by Searle et al. [29]. Briefly, the FI included 32 health deficits covering multiple domains, including physician-diagnosed chronic diseases, sensory problems, self-rated general health, limitations in activities of daily living and instrumental activities of daily living, mobility limitations, depressive symptoms, and cognitive function. Most deficits were coded as binary variables, with 1 indicating the presence of a deficit and 0 indicating its absence; self-rated health, depressive symptoms, and cognitive function were coded according to predefined cut-off values or scoring rules, as detailed in Table S1. For each participant, the FI was calculated as the ratio of accumulated deficits to the total number of included deficits, yielding a score ranging from 0 to 1. The selection of deficits was theoretically motivated according to the deficit accumulation framework and prior CHARLS-based frailty studies, rather than data-driven selection based on their associations with cardiovascular outcomes in the present analysis. [30, 31]. Higher FI values indicated a greater burden of frailty-related health deficits. The combined CHG-FI was calculated as the product of CHG and FI:

CHG-FI=CHG× FI

Assessment of outcomes

The primary outcome was new-onset CVD, defined as the occurrence of heart disease or stroke during follow-up. New-onset heart disease and stroke were analyzed separately as secondary outcomes. Outcome information was obtained from standardized self-reported physician diagnoses in CHARLS. Heart disease was defined as a self-reported diagnosis of heart attack, coronary heart disease, angina, congestive heart failure, or other heart problems. Stroke was defined as a self-reported physician diagnosis of stroke.

Covariates

Covariates included sociodemographic characteristics, lifestyle factors, clinical conditions, and laboratory biomarkers. Sociodemographic variables included age, sex, marital status, educational level, and household registration. Lifestyle factors included smoking status, drinking status, sleep duration, and BMI. Clinical conditions included hypertension and diabetes. Diabetes was defined as a self-reported physician diagnosis, use of glucose-lowering medications, FBG ≥ 126 mg/dL, or HbA1c ≥ 6.5%. Hypertension was defined as a self-reported physician diagnosis, use of antihypertensive medications, systolic blood pressure ≥ 140 mmHg, or diastolic blood pressure ≥ 90 mmHg. Laboratory biomarkers included platelet count (PLT), blood urea nitrogen (BUN), uric acid (UA), serum creatinine (SCr), and C-reactive protein (CRP). These laboratory biomarkers were selected based on prior evidence linking hematological, renal, metabolic, and inflammatory status with cardiometabolic risk and cardiovascular outcomes, as well as their availability in CHARLS. PLT was included as a routinely available hematological marker related to thrombosis, vascular inflammation, and atherothrombotic cardiovascular processes, rather than as a confirmed causal mediator [20].

Statistical analysis

Baseline characteristics were summarized according to quartiles of CHG-FI. Continuous variables were expressed as medians with interquartile ranges because of their non-normal distributions and compared using the Kruskal–Wallis test. Categorical variables were expressed as frequencies and percentages and compared using the chi-square test or Fisher’s exact test, as appropriate.

Kaplan–Meier curves were generated to compare the cumulative incidence of new-onset CVD, heart disease, and stroke across CHG-FI quartiles, and differences were assessed using log-rank tests. Multivariable Cox proportional hazards models were used to evaluate the associations of CHG-FI with new-onset CVD, heart disease, and stroke. CHG-FI was analyzed both as a continuous variable and by quartiles. The following sequential models were constructed: the crude model without adjustment; Model 1 adjusted for age, sex, education, household registration, and marital status; Model 2 further adjusted for smoking status, drinking status, sleep duration, and BMI; Model 3 further adjusted for hypertension and diabetes; and Model 4 further adjusted for platelet count, blood urea nitrogen, uric acid, C-reactive protein, and serum creatinine.

Restricted cubic spline analyses were performed in fully adjusted Cox models to explore potential nonlinear associations between CHG-FI and new-onset CVD, heart disease, and stroke. Nonlinearity was assessed using likelihood ratio tests comparing models with and without spline terms. To examine the joint association of CHG and FI with new-onset CVD, CHG and FI were dichotomized into low and high groups according to their respective median values and then cross-classified into four groups: low CHG plus low FI, high CHG plus low FI, low CHG plus high FI, and high CHG plus high FI. The low CHG plus low FI group served as the reference. Multiplicative interaction was evaluated by including a product term between CHG and FI in the fully adjusted Cox model. Additive interaction was assessed using the relative excess risk due to interaction (RERI), attributable proportion due to interaction (AP), and synergy index (SI). The null values were 0 for RERI and AP and 1 for SI.

The predictive performance of CHG-FI was compared with that of CHG, FI, TyG, AIP, and METS-IR using receiver operating characteristic (ROC) analyses. Differences in AUCs were assessed using DeLong tests. The incremental predictive value of CHG-FI was further evaluated using net reclassification improvement (NRI) and integrated discrimination improvement (IDI). In addition, formal incremental modeling was performed by first constructing a fully adjusted base model including all covariates in Model 4, and then separately adding CHG-FI, CHG, FI, TyG, AIP, or METS-IR to the base model. The incremental performance of each index was evaluated using AUC, IDI, and NRI. Subgroup analyses were performed according to age, sex, marital status, education level, household registration, BMI, smoking status, alcohol consumption, and sleep duration to examine potential effect modification.

Sensitivity analyses were conducted by applying multiple imputation for missing covariates and by excluding participants who developed cardiovascular events within the first 24 months of follow-up to minimize potential reverse causation. To evaluate the robustness of our findings to frailty operationalization, we further performed sensitivity analyses using four alternative FI specifications. First, a cardiovascular-free FI was reconstructed after removing heart disease- and stroke-related deficits from the original 32-item FI. Second, a hypertension/diabetes-free FI was reconstructed after removing hypertension- and diabetes-related deficits. Third, a cardiometabolic-overlap-free FI was reconstructed after simultaneously removing cardiovascular disease-, hypertension-, and diabetes-related deficits. Fourth, to assess the influence of subjective and psychosocial frailty components, an alternative FI was reconstructed after removing depressive symptoms and self-rated health-related deficits. For each modified FI, the denominator was adjusted according to the number of retained deficits, and the corresponding modified CHG-FI was recalculated as the product of CHG and the modified FI. The primary Cox regression analyses were then repeated using these modified CHG-FI measures. In addition, because rural/urban household registration may not fully capture current residence, we performed an additional sensitivity analysis by including current residence as a covariate in the Cox models. To assess potential multicollinearity, variance inflation factor (VIF) diagnostics were calculated for the adjusted Cox models across the three cardiovascular outcomes. VIFs were evaluated for both continuous CHG-FI models and quartile-based CHG-FI models. A VIF value below 5 was considered to indicate the absence of serious multicollinearity. To further address potential structural dependence among CKM staging, exposure construction, and covariate adjustment, we repeated the primary Cox analyses after excluding participants with baseline hypertension, excluding participants with baseline diabetes, and excluding participants with baseline hypertension and/or diabetes. In addition, baseline CKM stage was entered as an additional categorical covariate to examine whether the association between CHG-FI and cardiovascular outcomes was robust after accounting for CKM stage. The proportional hazards assumption for Cox regression models was assessed using Schoenfeld residuals. PH diagnostics were performed for the crude model and sequentially adjusted Models 1–4 across all three cardiovascular outcomes. The PH assumption was evaluated for both models using CHG-FI as a continuous exposure and models using CHG-FI quartiles. Variable-specific and global PH tests were examined, and a two-sided P value < 0.05 was considered indicative of potential violation of the PH assumption. For covariates showing evidence of PH violation, sensitivity analyses were conducted using time-dependent Cox regression by introducing interaction terms between log-transformed survival time and the covariates violating the PH assumption. To examine whether the main findings were dependent on platelet count adjustment, we performed an additional sensitivity analysis by excluding PLT from the fully adjusted model while retaining the other covariates.

We conducted all computational procedures utilizing the 4.2.2 edition of the R programming environment. Findings achieving a two-tailed probability value below 0.05 were deemed to possess statistical significance.

Results

Baseline characteristics

The baseline characteristics of the 4,603 participants, stratified by quartiles of the CHG-FI, are summarized in Table 1. The median age of the study population was 57 (51, 63) years, and 56.0% (n = 2,576) were female. The median duration of follow-up was 9 years. Compared with participants in the lowest quartile (Q1), those in the highest quartile (Q4) were older, more likely to be female, had lower educational attainment, were more likely to have rural household registration, had a higher BMI, and reported shorter sleep duration. Clinically, participants in Q4 exhibited a significantly higher prevalence of hypertension and diabetes mellitus. Regarding the clinical laboratory parameters, higher CHG-FI quartiles were accompanied by higher levels of LDL-C, FBG, TC, and CRP, whereas HDL-C levels tended to be lower.

Table 1.

Baseline characteristics

Variables	Total (n = 4,603)	Q1 (n = 1,159)	Q2 (n = 1,144)	Q3 (n = 1,151)	Q4 (n = 1,149)	P-value
Age, Median (Q1, Q3)	57 (51, 63)	54 (48, 60)	56 (51, 62)	57 (51, 63)	60 (55, 67)	< 0.001
Sex, n (%)						< 0.001
Female	2576 (56)	550 (47.5)	600 (52.4)	675 (58.6)	751 (65.4)
Male	2027 (44)	609 (52.5)	544 (47.6)	476 (41.4)	398 (34.6)
Marital, n (%)						< 0.001
Married	4177 (90.7)	1093 (94.3)	1037 (90.6)	1043 (90.6)	1004 (87.4)
Other	426 (9.3)	66 (5.7)	107 (9.4)	108 (9.4)	145 (12.6)
Education, n (%)						< 0.001
Primary school or below	3233 (70.2)	647 (55.8)	769 (67.2)	849 (73.8)	968 (84.2)
High school or above	1370 (29.8)	512 (44.2)	375 (32.8)	302 (26.2)	181 (15.8)
Household registration, n (%)						< 0.001
Rural	4005 (87)	959 (82.7)	977 (85.4)	1016 (88.3)	1053 (91.6)
Urban	598 (13)	200 (17.3)	167 (14.6)	135 (11.7)	96 (8.4)
BMI, Median (Q1, Q3)	23.2 (21, 25.8)	23.1 (21.1, 25.2)	23 (21, 25.6)	23.3 (20.9, 25.9)	23.4 (20.9, 26.4)	0.050
Smoking, n (%)						< 0.001
No	2919 (63.4)	695 (60)	688 (60.1)	744 (64.6)	792 (68.9)
Yes	1684 (36.6)	464 (40)	456 (39.9)	407 (35.4)	357 (31.1)
Drinking, n (%)						< 0.001
No	3213 (69.8)	772 (66.6)	762 (66.6)	820 (71.2)	859 (74.8)
Yes	1390 (30.2)	387 (33.4)	382 (33.4)	331 (28.8)	290 (25.2)
Sleep time, Median (Q1, Q3)	7 (5, 8)	7 (6, 8)	7 (6, 8)	6 (5, 8)	6 (4, 7)	< 0.001
Hypertension, n (%)						< 0.001
No	2901 (63.0)	892 (77.0)	749 (65.5)	677 (58.8)	583 (50.7)
Yes	1702 (37.0)	267 (23.0)	395 (34.5)	474 (41.2)	566 (49.3)
Diabetes, n (%)						< 0.001
No	3941 (85.6)	1067 (92.1)	1010 (88.3)	971 (84.4)	893 (77.7)
Yes	662 (14.4)	92 (7.9)	134 (11.7)	180 (15.6)	256 (22.3)
Plt, Median (Q1, Q3)	208 (162, 256)	212 (166, 260)	207 (160, 255)	203 (159, 253)	211 (165, 258)	0.062
BUN, Median (Q1, Q3)	15.1 (12.5, 18.1)	14.8 (12.4, 17.6)	15.1 (12.5, 18.2)	15.2 (12.5, 18.2)	15.2 (12.5, 18.4)	0.052
Scr, Median (Q1, Q3)	0.7 (0.6, 0.9)	0.8 (0.6, 0.9)	0.8 (0.6, 0.9)	0.7 (0.6, 0.9)	0.7 (0.6, 0.8)	< 0.001
UA, Median (Q1, Q3)	4.2 (3.5, 5)	4.2 (3.6, 5.1)	4.2 (3.6, 5.0)	4.2 (3.5, 5.0)	4.1 (3.4, 4.9)	0.004
CRP, Median (Q1, Q3)	0.9 (0.5, 1.9)	0.8 (0.5, 1.7)	1.0 (0.5, 1.9)	1.0 (0.5, 2.1)	1.1 (0.6, 2.2)	< 0.001
FBG, Median (Q1, Q3)	101.7 (94.1, 111.4)	100.6 (93.6, 108.4)	101.7 (94.5, 111.1)	102.2 (94, 112)	103.1 (94.9, 115.2)	< 0.001
TC, Median (Q1, Q3)	191.4 (168.2, 216.1)	186.7 (163.9, 209.9)	192.9 (169.6, 217.7)	191 (167.4, 216.1)	194.1 (172, 220.7)	< 0.001
HDL, Median (Q1, Q3)	49.9 (41, 60.3)	50.6 (41.4, 61.1)	49.5 (40.6, 59.9)	50.3 (41.8, 60.1)	49.1 (40.6, 59.5)	0.180
LDL, Median (Q1, Q3)	116 (95.1, 138.8)	113.7 (93.4, 133.8)	118.1 (96.3, 141.9)	115.2 (93.2, 139)	117.1 (96.7, 140.7)	0.001
CVD, n (%)						< 0.001
No	3414 (74.2)	977 (84.3)	885 (77.4)	834 (72.5)	718 (62.5)
Yes	1189 (25.8)	182 (15.7)	259 (22.6)	317 (27.5)	431 (37.5)
Heart disease, n (%)						< 0.001
No	3696 (80.3)	1019 (87.9)	958 (83.7)	909 (79.0)	810 (70.5)
Yes	907 (19.7)	140 (12.1)	186 (16.3)	242 (21.0)	339 (29.5)
Stroke, n (%)						< 0.001
No	4177 (90.7)	1107 (95.5)	1046 (91.4)	1032 (89.7)	992 (86.3)
Yes	426 (9.3)	52 (4.5)	98 (8.6)	119 (10.3)	157 (13.7)

BMI, body mass index; PLT, platelet count; BUN, blood urea nitrogen; SCr, serum creatinine; UA, uric acid; CRP, C-reactive protein; FBG, fasting blood glucose; TC, total cholesterol; HDL-C, high-density lipoprotein cholesterol; LDL-C, low-density lipoprotein cholesterol; CVD, cardiovascular disease

Associations of CHG-FI with new-onset cardiovascular outcomes

During a median follow-up of 9 years, new-onset CVD occurred in 1,189 participants, representing 25.8% of the study cohort. Specifically, 907 participants developed heart disease (19.7%) and 426 developed stroke (9.3%). The Kaplan–Meier curves shown in Fig. 2 illustrate event-free survival probabilities for (A) CVD, (B) heart disease, and (C) stroke according to CHG-FI quartiles. Significant differences were observed across CHG-FI quartiles for all outcomes, with log-rank P values < 0.001. Event-free survival decreased progressively across increasing CHG-FI quartiles, with participants in Q4 consistently showing the lowest survival probability. A similar graded pattern was observed for heart disease (Fig. 2B), comparable to that for composite CVD (Fig. 2A). For stroke (Fig. 2C), although the separation between curves was slightly less pronounced than that for CVD and heart disease, the between-group differences remained statistically significant, with participants in Q4 consistently demonstrating the highest risk.

Fig. 2.

Kaplan–Meier survival curves for cardiovascular outcomes stratified by CHG-FI quartiles. Kaplan–Meier survival curves are shown for A cardiovascular disease (CVD), B heart disease, and C stroke according to quartiles of CHG-FI (Q1–Q4). Differences among groups were assessed using the log-rank test. CHG-FI, combined cholesterol, high-density lipoprotein, glucose index and frailty index

Multivariable Cox proportional hazards models were used to estimate the associations of CHG-FI with the three cardiovascular outcomes (Table 2). CHG-FI was analyzed both as a continuous variable and by quartiles. In the fully adjusted model, each one-unit increase in CHG-FI was associated with a 56%–59% higher hazard of new-onset cardiovascular outcomes. Specifically, the adjusted HRs were 1.59 (95% CI 1.42–1.77) for heart disease, 1.56 (95% CI 1.32–1.83) for stroke, and 1.56 (95% CI 1.41–1.72) for CVD. Consistent with these continuous analyses, the quartile-based analyses showed clear dose–response patterns across CHG-FI quartiles (all P for trend < 0.001). Compared with participants in Q1, those in Q4 had substantially higher risks of all outcomes, with HRs of 2.58 (95% CI 1.84–3.61), 2.47 (95% CI 1.99–3.05), and 2.43 (95% CI 2.01–2.93) for stroke, heart disease, and CVD, respectively.

Table 2.

Association between CHG-FI and CVD incidence in individuals with CKM syndrome stages 0–3

	HR (95%CI)
Exposure	Crude model	Model 1	Model 2	Model 3	Model 4
CVD
CHG-FI (per unit)	1.73 (1.58,1.89)	1.66 (1.51,1.83)	1.66 (1.51,1.83)	1.57 (1.42,1.73)	1.56 (1.41,1.72)
CHG-FI (quartiles)
Q1	1.00 (Reference)	1.00 (Reference)	1.00 (Reference)	1.00 (Reference)	1.00 (Reference)
Q2	1.5 (1.24,1.82)	1.49 (1.23,1.80)	1.48 (1.22,1.79)	1.42 (1.17,1.72)	1.43 (1.18,1.73)
Q3	1.89 (1.58,2.27)	1.88 (1.56,2.26)	1.88 (1.56,2.26)	1.76 (1.46,2.12)	1.77 (1.47,2.14)
Q4	2.76 (2.32,3.28)	2.68 (2.23,3.21)	2.67 (2.22,3.21)	2.43 (2.01,2.93)	2.43 (2.01,2.93)
P-trend	< 0.001	< 0.001	< 0.001	< 0.001	< 0.001
Heart disease
CHG-FI (per unit)	1.74 (1.57,1.92)	1.66 (1.49,1.85)	1.66 (1.49,1.85)	1.60 (1.43,1.78)	1.59 (1.42,1.77)
CHG-FI (quartiles)
Q1	1.00 (Reference)	1.00 (Reference)	1.00 (Reference)	1.00 (Reference)	1.00 (Reference)
Q2	1.38 (1.11,1.72)	1.37 (1.10,1.71)	1.36 (1.09,1.70)	1.33 (1.06,1.66)	1.34 (1.07,1.67)
Q3	1.84 (1.49,2.26)	1.81 (1.47,2.24)	1.82 (1.47,2.24)	1.74 (1.40,2.15)	1.76 (1.42,2.18)
Q4	2.74 (2.25,3.33)	2.62 (2.13,3.23)	2.63 (2.13,3.24)	2.46 (1.99,3.05)	2.47 (1.99,3.05)
P-trend	< 0.001	< 0.001	< 0.001	< 0.001	< 0.001
Stroke
CHG-FI (per unit)	1.77 (1.53,2.05)	1.74 (1.49,2.04)	1.74 (1.48,2.04)	1.57 (1.33,1.85)	1.56 (1.32,1.83)
CHG-FI (quartiles)
Q1	1.00 (Reference)	1.00 (Reference)	1.00 (Reference)	1.00 (Reference)	1.00 (Reference)
Q2	1.95 (1.39,2.73)	1.94 (1.38,2.72)	1.91 (1.36,2.68)	1.74 (1.24,2.45)	1.76 (1.25,2.47)
Q3	2.37 (1.71,3.29)	2.39 (1.72,3.32)	2.37 (1.70,3.30)	2.04 (1.46,2.84)	2.06 (1.47,2.87)
Q4	3.19 (2.33,4.36)	3.17 (2.29,4.40)	3.14 (2.26,4.38)	2.57 (1.84,3.59)	2.58 (1.84,3.61)
P-trend	< 0.001	< 0.001	< 0.001	< 0.001	< 0.001

Crude model: unadjusted

Model 1 adjusted for: Age, sex, education, household registration, and marital status

Model 2 adjusted for: Age, sex, education, household registration, marital status, smoking status, drinking status, sleep time, and BMI

Model 3 adjusted for: Age, sex, education, household registration, marital status, smoking status, drinking status, sleep time, BMI, hypertension, and diabetes

Model 4 adjusted for: Age, sex, education, household registration, marital status, smoking status, drinking status, sleep time, BMI, hypertension, diabetes, platelet count, blood urea nitrogen, uric acid, C-reactive protein, and serum creatinine

CHG-FI, combined cholesterol, high-density lipoprotein, glucose index and frailty index; CKM, cardiovascular-kidney-metabolic; HR, hazard ratio; CI, confidence interval; CVD, cardiovascular disease; BMI, body mass index

Dose–response associations assessed by RCS

To evaluate potential nonlinear dose–response associations between CHG-FI and new-onset cardiovascular outcomes, multivariable-adjusted RCS models were applied (Fig. 3). After adjustment for covariates, CHG-FI showed significant nonlinear associations with CVD risk (overall P < 0.001; nonlinearity P < 0.001). Specifically, the risk of CVD increased steeply until CHG-FI reached an inflection point of 0.815, beyond which the rate of increase appeared to attenuate (Fig. 3A). Similar nonlinear patterns were observed for heart disease and stroke, with identified inflection points of 0.817 (Fig. 3B) and 0.813 (Fig. 3C), respectively.

Fig. 3.

Restricted cubic spline analyses of the associations between CHG-FI and incident cardiovascular outcomes. Multivariable-adjusted hazard ratios are shown for A cardiovascular disease (CVD), B heart disease, and C stroke. The red solid lines indicate estimated hazard ratios, and the shaded areas represent 95% confidence intervals. Grey histograms show the distribution of CHG-FI, and the horizontal dashed line indicates a hazard ratio of 1.0. P values for overall association and nonlinearity were calculated using likelihood ratio tests. CHG-FI, combined cholesterol, high-density lipoprotein, glucose index and frailty index

Joint association and interaction between CHG and FI

Compared with participants with low CHG and low FI, those with high CHG but low FI had a significantly higher risk of new-onset CVD (HR = 1.27, 95% CI 1.05–1.54; P = 0.015). A stronger association was observed among participants with low CHG but high FI (HR = 1.67, 95% CI 1.39–2.01; P < 0.001). Notably, participants with both high CHG and high FI exhibited the highest risk of new-onset CVD (HR = 2.05, 95% CI 1.71–2.47; P < 0.001), suggesting that the coexistence of elevated metabolic burden and frailty was associated with a substantially increased cardiovascular risk.

Table 3 presents the multivariable-adjusted interaction analyses between CHG and FI in relation to new-onset CVD. On the multiplicative scale, the interaction estimate was 0.97 (95% CI 0.76–1.23; P = 0.789), indicating no statistically significant multiplicative interaction between CHG and FI. On the additive scale, the RERI was 0.11 (95% CI − 0.23 to 0.45), the AP was 0.06 (95% CI − 0.11 to 0.22), and the SI was 1.12 (95% CI 0.78–1.61). Because the confidence intervals for RERI and AP included 0 and that for SI included 1, there was no evidence of statistically significant additive interaction between CHG and FI (Table 4).

Table 3.

Joint association of CHG and FI with incident cardiovascular disease

CHG and FI joint groups	N	Events	HR (95% CI)	P-value
Low CHG + Low FI	1,140	191	1.00 (Reference)
High CHG + Low FI	1,131	252	1.27 (1.05, 1.54)	0.015
Low CHG + High FI	1,161	326	1.67 (1.39, 2.01)	< 0.001
High CHG + High FI	1,171	420	2.05 (1.71, 2.47)	< 0.001

Hazard ratios (HRs) and 95% confidence intervals (CIs) were estimated using multivariable Cox proportional hazards models. The low CHG and low FI group was used as the reference category. Models were adjusted for age, sex, education level, household registration, marital status, smoking status, drinking status, sleep duration, body mass index, hypertension, diabetes, platelet count, blood urea nitrogen, uric acid, C-reactive protein, and serum creatinine

CHG, cholesterol–high-density lipoprotein–glucose index; FI, frailty index; HR, hazard ratio; CI, confidence interval; CVD, cardiovascular disease

Table 4.

Multiplicative and additive interactions between CHG and FI for incident cardiovascular disease

Measures	Estimates (95% CI)	P-value
Multiplicative interaction
Multiplicative Scale	0.97 (0.76, 1.23)	0.789
Additive interaction
RERI (95% CI)	0.11 (− 0.23, 0.45)	0.257
AP (95% CI)	0.06 (− 0.11, 0.22)	0.256
SI (95% CI)	1.12 (0.78, 1.61)	0.268

Multiplicative interaction was assessed by including a product term between CHG and FI in the multivariable Cox proportional hazards model. Additive interaction was evaluated using the relative excess risk due to interaction (RERI), attributable proportion due to interaction (AP), and synergy index (SI). The null values were 0 for RERI and AP and 1 for SI. Models were adjusted for age, sex, education level, household registration, marital status, smoking status, drinking status, sleep duration, body mass index, hypertension, diabetes, platelet count, blood urea nitrogen, uric acid, C-reactive protein, and serum creatinine

CHG, cholesterol–high-density lipoprotein–glucose index; FI, frailty index; CI, confidence interval; RERI, relative excess risk due to interaction; AP, attributable proportion due to interaction; SI, synergy index; CVD, cardiovascular disease

Comparative predictive performance of CHG-FI and other risk indices

To evaluate the predictive performance of CHG-FI relative to its individual components and established metabolic indices, ROC analyses were performed for new-onset CVD, heart disease, and stroke (Fig. 4 and Table 5). For composite CVD and heart disease, CHG-FI showed the highest AUCs among all evaluated indices, with AUCs of 0.624 and 0.622, respectively. DeLong tests further indicated that CHG-FI had statistically higher AUCs than CHG, FI, TyG, AIP, and METS-IR for both outcomes (all DeLong P < 0.001). However, the absolute AUC values remained modest, indicating limited standalone discriminative performance.

Fig. 4.

Receiver operating characteristic curve analyses comparing CHG-FI with other risk indices for incident cardiovascular outcomes. Receiver operating characteristic curves are shown for A cardiovascular disease (CVD), B heart disease, and C stroke. The CHG-FI was compared with its individual components, CHG and FI, and established metabolic indices, including TyG, AIP, and METS-IR. The dashed diagonal line indicates the reference line of no discrimination. CHG-FI, combined cholesterol, high-density lipoprotein, glucose index and frailty index; CHG, cholesterol, high-density lipoprotein, glucose index; FI, frailty index; TyG, triglyceride-glucose index; AIP, atherogenic index of plasma; METS-IR, metabolic score for insulin resistance

Table 5.

Receiver operating characteristic analysis comparing CHG-FI, its components, and established metabolic indices for incident cardiovascular outcomes

Test	AUC (95%CI)	Best threshold	Specificity	Sensitivity	Delong-P value
CVD
CHG-FI	0.624 (0.605–0.642)	0.420	0.494	0.686	–
CHG	0.558 (0.539–0.577)	5.218	0.466	0.634	< 0.001
FI	0.618 (0.600–0.636)	0.076	0.459	0.712	< 0.001
TyG	0.546 (0.527–0.564)	8.254	0.306	0.775	< 0.001
AIP	0.544 (0.525–0.562)	0.243	0.434	0.643	< 0.001
METS-IR	0.572 (0.553–0.591)	37.734	0.698	0.417	< 0.001
Heart disease
CHG-FI	0.622 (0.601–0.642)	0.475	0.547	0.635	–
CHG	0.540 (0.519–0.561)	5.218	0.455	0.621	< 0.001
FI	0.617 (0.597–0.637)	0.097	0.566	0.612	< 0.001
TyG	0.532 (0.512–0.553)	8.252	0.296	0.765	< 0.001
AIP	0.527 (0.507–0.548)	0.280	0.476	0.584	< 0.001
METS-IR	0.562 (0.540–0.583)	37.717	0.689	0.419	< 0.001
Stroke
CHG-FI	0.618 (0.591–0.645)	0.398	0.422	0.756	–
CHG	0.597 (0.569–0.624)	5.295	0.543	0.615	0.264
FI	0.610 (0.583–0.636)	0.075	0.418	0.754	< 0.001
TyG	0.575 (0.548–0.603)	8.285	0.315	0.808	0.024
AIP	0.580 (0.553–0.607)	0.341	0.554	0.575	0.043
METS-IR	0.598 (0.570–0.625)	31.678	0.375	0.777	0.286

CHG-FI, combined cholesterol, high-density lipoprotein, glucose index and frailty index; CHG, cholesterol, high-density lipoprotein, glucose index; FI, frailty index; TyG, triglyceride-glucose index; AIP, atherogenic index of plasma; METS-IR, metabolic score for insulin resistance; ROC, receiver operating characteristic; AUC, area under the receiver operating characteristic curve; CI, confidence interval; CVD, cardiovascular disease

For stroke, CHG-FI also showed the highest numerical AUC (0.618). Its AUC was statistically higher than that of FI (DeLong P < 0.001), TyG (DeLong P = 0.024), and AIP (DeLong P = 0.043), whereas the differences compared with CHG and METS-IR were not statistically significant despite numerically higher AUCs. These findings indicate that CHG-FI showed statistically better discrimination than several comparator indices, but the overall discriminative ability was modest across all cardiovascular outcomes.

We further assessed the incremental discrimination and reclassification performance of CHG-FI using NRI and IDI analyses (Table S2). For composite CVD, CHG-FI showed small but statistically significant IDI improvements compared with all comparator indices, with IDI values ranging from 0.20% to 0.48% (all P < 0.05). In terms of reclassification, CHG-FI showed statistically significant NRI improvements compared with CHG (NRI = 13.85%), TyG (NRI = 9.16%), AIP (NRI = 11.02%), and METS-IR (NRI = 20.90%), whereas the improvement relative to FI was positive but not statistically significant (NRI = 5.43%). For heart disease, CHG-FI provided significant improvements in both IDI and NRI compared with CHG and METS-IR, but not compared with FI, TyG, or AIP. For stroke, CHG-FI yielded significant improvements in reclassification and discrimination compared with all five reference indices. Collectively, these results suggest that CHG-FI provided statistically significant but limited incremental predictive information beyond conventional metabolic indices, and the magnitude of improvement should be interpreted cautiously.

Incremental predictive value beyond the fully adjusted base model

To further evaluate whether CHG-FI provided incremental predictive information beyond clinical covariates, we added CHG-FI and each comparator index separately to the fully adjusted base model (Table S3). For CVD, the base model plus CHG-FI yielded an AUC of 0.659 (95% CI 0.641–0.677), which was numerically higher than the AUCs obtained after adding CHG (0.634; 95% CI 0.615–0.652), FI (0.657; 95% CI 0.640–0.675), TyG (0.632; 95% CI 0.613–0.650), AIP (0.632; 95% CI 0.613–0.650), or METS-IR (0.631; 95% CI 0.613–0.650). Similar patterns were observed for heart disease and stroke. However, the absolute improvement of CHG-FI over FI alone was small, particularly for CVD and heart disease.

IDI and NRI analyses using the base model as the reference also showed that CHG-FI provided statistically significant but modest incremental information (Table S4). For CVD, adding CHG-FI to the base model yielded an IDI of 1.50% (95% CI 1.09–1.92%) and an NRI of 27.28% (95% CI 20.77–33.80%). For heart disease, the corresponding IDI and NRI were 1.32% (95% CI 0.93–1.72%) and 28.65% (95% CI 21.46–35.85%), respectively. For stroke, the corresponding IDI and NRI were 0.57% (95% CI 0.22–0.91%) and 22.69% (95% CI 12.77–32.61%), respectively. These results indicate that CHG-FI provided incremental information beyond clinical covariates, but the magnitude of improvement was limited and should not be interpreted as evidence of strong standalone predictive performance.

Subgroup analyses

To verify the robustness of the primary findings, subgroup analyses stratified by sociodemographic and lifestyle characteristics—including age, sex, marital status, educational attainment, place of household registration, BMI, smoking status, alcohol consumption, and sleep duration—were performed (Fig. 5). Overall, the positive association between elevated CHG-FI levels (comparing Q4 vs. Q1) and the risks of composite CVD, heart disease, and stroke remained highly consistent across the majority of stratified subgroups. However, tests for interaction revealed specific sources of heterogeneity. When evaluating the combined CVD endpoint, smoking habits demonstrated a notable interactive effect with the CHG-FI (P for interaction = 0.024). Concerning cardiac illness specifically, we identified meaningful effect modifications driven by tobacco use (P for interaction = 0.003) alongside participant gender (P for interaction = 0.025).

Fig. 5.

Subgroup analyses of the associations between CHG-FI quartiles and incident cardiovascular outcomes. Forest plots show the hazard ratios (HRs) and 95% confidence intervals (CIs) for A cardiovascular disease (CVD), B heart disease, and C stroke across CHG-FI quartiles, with Q1 as the reference group

Sensitivity analyses

To evaluate the robustness of our findings, a series of sensitivity analyses were conducted. First, the primary analyses were repeated after performing multiple imputation to account for missing covariate data, and the results remained materially unchanged (Table S5). Second, to mitigate the risk of reverse causation, researchers executed a delayed-entry evaluation. This approach involved omitting subjects who encountered new CVD events during the initial 24 months of tracking; ultimately, the resulting correlations aligned perfectly with our primary conclusions (Table S6). Third, to assess whether the results were sensitive to the operationalization of frailty, we reconstructed CHG-FI using four alternative FI specifications: a cardiovascular-free FI, a hypertension/diabetes-free FI, a cardiometabolic-overlap-free FI, and an FI excluding depressive symptoms and self-rated health-related deficits. Across all four alternative FI specifications, the modified CHG-FI measures remained positively associated with new-onset CVD, heart disease, and stroke in fully adjusted Cox models (Table S7–S10). Fourth, because household registration may not fully reflect current residence, we further performed a sensitivity analysis by additionally adjusting for current residence in the Cox models (Table S11). The association between CHG-FI and new-onset CVD remained materially unchanged. In the fully adjusted model including current residence, each one-unit increase in CHG-FI was associated with a higher risk of new-onset CVD (HR = 1.56; 95% CI 1.41–1.72), and participants in the highest CHG-FI quartile remained at substantially higher risk than those in the lowest quartile (Q4 vs. Q1: HR = 2.42; 95% CI 2.01–2.93; P for trend < 0.001; Table S11). In addition, VIF diagnostics were performed to evaluate potential multicollinearity in the adjusted Cox models. Across all adjusted models for CVD, heart disease, and stroke, the VIF values for all variables were below 3 in both continuous CHG-FI models and quartile-based CHG-FI models (Tables S12 and S13). These results were below the commonly used threshold of 5, suggesting that serious multicollinearity was unlikely to have materially influenced the effect estimates. To further address potential structural dependence related to hypertension and diabetes, we repeated the primary Cox analyses after excluding participants with baseline hypertension, excluding participants with baseline diabetes, and excluding participants with baseline hypertension and/or diabetes. The associations between CHG-FI and new-onset CVD, heart disease, and stroke remained materially consistent across these restricted populations (Tables S14–S16). We also performed an additional sensitivity analysis by further adjusting for baseline CKM stage in the Cox models (Tables S17). The association between CHG-FI and new-onset CVD remained materially unchanged. In the fully adjusted model additionally including CKM stage, each one-unit increase in CHG-FI was associated with a higher risk of new-onset CVD (HR = 1.55; 95% CI 1.41–1.71), and participants in the highest CHG-FI quartile remained at substantially higher risk than those in the lowest quartile (Q4 vs. Q1: HR = 2.42; 95% CI 2.01–2.93; P for trend < 0.001; Table S17). Proportional hazards diagnostics were performed using Schoenfeld residuals. Across the crude and sequentially adjusted Cox models for CVD, heart disease, and stroke, the primary exposure variable CHG-FI did not violate the PH assumption in either continuous CHG-FI models or quartile-based CHG-FI models. Global PH tests were also not statistically significant across the evaluated models, suggesting that the overall Cox models satisfied the proportional hazards assumption (Tables S18 and S19). For individual covariates, most variables satisfied the PH assumption. Mild PH violations were observed only for platelet count and blood urea nitrogen in the fully adjusted CVD model and for uric acid in the fully adjusted stroke model. No PH violation was observed for covariates in the heart disease models. To evaluate whether these covariate-level PH violations influenced the main findings, time-dependent Cox regression models were fitted by introducing interaction terms between log-transformed survival time and the covariates violating the PH assumption. The associations of CHG-FI with new-onset CVD and stroke remained materially unchanged. In the time-dependent Cox models, each one-unit increase in CHG-FI was associated with higher risks of new-onset CVD (HR = 1.560; 95% CI 1.412–1.722) and stroke (HR = 1.555; 95% CI 1.319–1.834). Quartile-based analyses also showed consistent associations for CVD and stroke (Table S20).

To examine whether the findings depended on the multiplicative construction of CHG-FI, we repeated the primary analyses using a standardized additive CHG-FI score calculated as (zCHG + zFI)/2. In fully adjusted Cox models, the additive CHG-FI score remained significantly associated with higher risks of new-onset CVD (HR = 1.45; 95% CI 1.34–1.58), heart disease (HR = 1.40; 95% CI 1.28–1.54), and stroke (HR = 1.62; 95% CI 1.42–1.85). Quartile-based analyses also showed graded associations, with participants in Q4 having higher risks than those in Q1 for CVD (HR = 2.33; 95% CI 1.93–2.81), heart disease (HR = 2.18; 95% CI 1.76–2.70), and stroke (HR = 3.13; 95% CI 2.22–4.42; Table S21). These findings suggest that the observed associations reflected the combined metabolic–frailty construct rather than the specific multiplication operation alone. Because PLT was included as a baseline hematological and thrombo-inflammatory marker in the fully adjusted model, we further repeated the primary analyses after excluding PLT from Model 4. The associations between CHG-FI and new-onset CVD, heart disease, and stroke remained materially unchanged (Table S22). These findings suggest that the main results were not driven by PLT adjustment.

Discussion

In this nationwide cohort of middle-aged and older adults with early-stage CKM syndrome, higher CHG-FI levels were independently associated with increased risks of new-onset CVD, heart disease, and stroke. We also observed significant nonlinear dose–response relationships between CHG-FI and these cardiovascular outcomes. In joint-exposure analyses, participants with both high CHG and high FI had the greatest risk of new-onset CVD, whereas those with only one elevated component showed intermediate risk. However, formal interaction analyses did not provide evidence of statistically significant interaction between CHG and FI on either the multiplicative or additive scale. In predictive analyses, CHG-FI showed statistically significant but modest discriminative performance, and the magnitude of incremental reclassification improvement was limited compared with established metabolic indices. These findings suggest that CHG-FI may serve as a complementary marker reflecting the combined burden of glucose–lipid metabolic dysregulation and frailty-related vulnerability, rather than a standalone clinical prediction tool for cardiovascular risk stratification.

The observed association between CHG-FI and cardiovascular outcomes is biologically plausible and consistent with previous evidence linking metabolic dysfunction and frailty to cardiovascular risk. The CHG index integrates glucose and lipid parameters and has been proposed as a marker related to insulin resistance and cardiometabolic vulnerability [21, 32, 33]. Insulin resistance may contribute to atherosclerosis, endothelial dysfunction, vascular stiffness, and subsequent cardiovascular events [11, 34, 35]. In parallel, frailty reflects the accumulation of multidimensional health deficits and reduced physiological reserve, and has been increasingly recognized as an important predictor of adverse cardiovascular outcomes. Prior studies have shown that worsening frailty is associated with higher CVD risk, stroke incidence, heart failure, and mortality [36–38]. Emerging studies that combine metabolic markers with FI, including TyG-FI and AIP-FI, also support the concept that integrating metabolic dysfunction with frailty may improve cardiovascular risk characterization in aging populations [16, 24].

In this study, CHG-FI was constructed as an integrated index combining metabolic dysregulation and frailty-related biological vulnerability. Our findings should be interpreted in the context of prior frailty–metabolic composite index studies. Previous studies have shown that TyG-FI and AIP-FI are associated with adverse cardiometabolic and cardiovascular outcomes, supporting the concept that metabolic dysfunction and frailty-related physiological vulnerability jointly contribute to cardiovascular susceptibility [16, 24, 39]. Recent CHG-FI-related evidence from Fan et al. further showed that CHG-FI was associated with incident heart disease, stroke, diabetes, and cardiometabolic multimorbidity within the CMM framework, supporting the value of integrating glucose–cholesterol metabolic burden with frailty-related vulnerability [25]. In line with these findings, our results showed that CHG-FI was positively associated with incident CVD, heart disease, and stroke among individuals with CKM stages 0–3. Unlike prior CHG-FI research focused on CMM, our study extends this evidence to the CKM syndrome framework and provides a broader head-to-head comparison with CHG, FI, TyG, AIP, and METS-IR. However, given the modest improvement in discrimination, CHG-FI should be viewed as a complementary marker of metabolic–frailty vulnerability rather than a standalone clinical prediction model. Therefore, the value of CHG-FI in the present study lies not in establishing a new superior prediction tool, but in providing an integrated representation of glucose–cholesterol metabolic dysregulation and frailty-related physiological vulnerability within the early-stage CKM framework. The consistent positive associations observed across CVD, heart disease, and stroke indicate that individuals with higher CHG-FI levels may carry a greater combined burden of cardiometabolic impairment and reduced physiological reserve. The nonlinear dose–response patterns further suggest that cardiovascular risk may increase disproportionately across the lower to moderate range of CHG-FI and then rise more gradually at higher levels. This pattern may reflect a threshold-like accumulation of metabolic and frailty-related deficits, beyond which additional increases in CHG-FI confer relatively smaller incremental risk. These findings support the epidemiological relevance of CHG-FI as a composite indicator of metabolic–frailty vulnerability in early-stage CKM syndrome, but do not establish it as a standalone clinical prediction tool.

The joint-exposure analysis provided additional clinically interpretable evidence regarding the combined roles of CHG and FI. Compared with participants with low CHG and low FI, those with either high CHG or high FI alone had elevated risks of new-onset CVD, while participants with both high CHG and high FI had the highest risk. This graded pattern suggests that the coexistence of glucose–cholesterol metabolic dysregulation and frailty-related vulnerability identifies a subgroup with particularly high cardiovascular risk within early-stage CKM syndrome. However, because formal interaction analyses did not show statistically significant multiplicative or additive interaction, this finding should be interpreted as evidence of cumulative or parallel risk burden rather than a synergistic biological interaction. However, the absence of statistically significant multiplicative or additive interaction indicates that the joint elevation in risk should not be interpreted as evidence of a supra-additive or synergistic effect. Rather, these findings suggest that CHG and FI may contribute to cardiovascular risk through partially independent or parallel pathways. Therefore, the value of CHG-FI may lie in its ability to capture the cumulative burden of two complementary risk domains, rather than in detecting a statistically significant interaction between them.

CKM staging, exposure construction, and covariate adjustment are not entirely independent because hypertension, diabetes, and glucose-related information contribute to more than one analytical component. This structural overlap should be considered when interpreting the findings. However, the purpose of this study was not to establish that CHG-FI is independent of CKM staging, but to examine whether CHG-FI can further characterize cardiovascular risk within individuals already classified as CKM stages 0–3. From this perspective, CKM syndrome provides the clinical framework for identifying an early-stage preventive population, whereas CHG-FI provides an integrated marker reflecting glucose–cholesterol metabolic dysregulation and frailty-related vulnerability within that population. Additional analyses excluding participants with baseline hypertension, diabetes, or both, as well as analyses further adjusting for baseline CKM stage, yielded materially consistent results. These findings suggest that the observed associations were not solely explained by structural overlap among CKM staging, CHG-FI construction, and covariate adjustment.

The comparative predictive analyses further support the potential utility of CHG-FI as a complementary risk characterization marker. CHG-FI showed the highest AUCs for new-onset CVD and heart disease among the evaluated indices, with statistically better discrimination than CHG, FI, TyG, AIP, and METS-IR. For stroke, CHG-FI also had the highest numerical AUC, although its advantage over CHG and METS-IR did not reach statistical significance. However, the overall discriminative performance remained modest, with AUC values of 0.618–0.624 and only small improvements over FI alone (0.610–0.618), indicating limited incremental discriminative value. This may be because CHG-FI relies on routinely available clinical indicators and cannot fully capture the complex metabolic heterogeneity underlying CVD. Recent metabolomic and lipidomic studies have suggested that specific lipid metabolite panels combined with machine-learning models may achieve high diagnostic discrimination for CAD and T2DM, although such diagnostic findings are not directly comparable with prospective prediction of incident CVD [40]. In addition, NRI and IDI analyses indicated only limited improvements in risk reclassification when CHG-FI was compared with conventional metabolic indices, and the magnitude of these changes remained modest. From a clinical perspective, such statistically significant but numerically small improvements in NRI and IDI may not necessarily translate into meaningful changes in individual-level risk classification or clinical decision-making; therefore, their practical relevance should be interpreted with caution. Importantly, these findings suggest that the primary value of CHG-FI is not in substantially improving predictive discrimination, but rather in providing an integrated representation of metabolic and frailty-related vulnerability. Future studies are warranted to define clinically meaningful thresholds for improvement in discrimination and reclassification metrics, and to determine whether such incremental gains can lead to measurable benefits in risk-guided prevention strategies. Furthermore, while CHG-FI offers additional predictive information beyond conventional metabolic indicators, its discriminative ability remains modest and insufficient for standalone application in clinical practice. As such, CHG-FI should be regarded as a complementary tool within a broader assessment framework for cardiometabolic function and frailty status.

The mechanisms linking CHG-FI to CVD, heart disease, and stroke are likely multifactorial. The CHG component reflects disturbances in glucose and lipid metabolism and may serve as a surrogate marker of insulin resistance, whereas FI captures cumulative multisystem physiological decline. Although our interaction analyses did not support a statistically significant departure from additivity or multiplicativity between CHG and FI, these two domains may still contribute to cardiovascular risk through complementary biological pathways. Insulin resistance may promote atherosclerosis and cardiovascular events through endothelial dysfunction, reduced nitric oxide bioavailability, oxidative stress, vascular inflammation, vascular smooth muscle cell proliferation, pro-atherogenic vascular remodeling, and lipid dysregulation [41–43]. Mechanistically, impaired IRS/PI3K/Akt/eNOS signaling and relatively preserved MAPK-related signaling may favor vasoconstriction, endothelial activation, inflammation, and adverse vascular remodeling, while hyperglycemia and mitochondrial reactive oxygen species may further aggravate nitric oxide deficiency and oxidative stress. Frailty, on the other hand, is characterized by chronic low-grade inflammation, impaired immune regulation, mitochondrial dysfunction, and reduced capacity to recover from physiological stress [44, 45]. Recent evidence further suggests that endothelial senescence may provide a potential link among metabolic dysregulation, chronic inflammation, microvascular damage, and systemic metabolic dysfunction [46]. With advancing age and metabolic stress, senescent endothelial cells may promote vascular inflammation, arterial stiffness, microvascular rarefaction, nitric oxide deficiency, and impaired vascular repair [47]. This is consistent with the concept of inflammageing, which links chronic age-related inflammation with CVD, frailty, and multimorbidity; meta-analytic evidence has also shown that frailty is associated with elevated inflammatory biomarkers, including CRP and interleukin-6 [48, 49]. Together, these pathways provide biological plausibility for the association between CHG-FI and cardiovascular outcomes, suggesting that elevated CHG-FI may capture the convergence of glucose–lipid metabolic dysregulation and frailty-related physiological vulnerability in early-stage CKM syndrome.

Our findings may have several clinical and public health implications. First, CHG-FI is derived from routinely available laboratory measurements and a deficit-based frailty assessment, making it potentially feasible for use in community-based and primary care settings. Given the high population burden of early CKM stages, such a simple composite marker may be particularly relevant for scalable screening and early risk stratification. Second, by integrating metabolic and frailty-related information, CHG-FI may help identify individuals with early-stage CKM syndrome who are at elevated risk of future cardiovascular events but may not be fully captured by metabolic indices alone. Third, the joint-exposure findings indicate that individuals with both high CHG and high FI represent a particularly vulnerable subgroup who may benefit from more intensive risk monitoring and early preventive strategies. These strategies may include cardiometabolic risk control, frailty assessment, lifestyle modification, and interventions targeting physical function and physiological reserve.

This study has several strengths. First, to our knowledge, this is among the first studies to evaluate CHG-FI as a composite marker for new-onset CVD, heart disease, and stroke among individuals with early-stage CKM syndrome. Second, the use of a nationwide longitudinal cohort with a relatively long follow-up period enhances the reliability and relevance of our findings. Third, we assessed CHG-FI not only as an exposure associated with cardiovascular outcomes but also as a potential predictive marker, using ROC, DeLong, NRI, and IDI analyses. Fourth, we performed joint-exposure and formal interaction analyses to clarify whether CHG and FI were associated with new-onset CVD jointly or interactively. Finally, subgroup and sensitivity analyses were conducted to evaluate the consistency and robustness of the main findings.

Several limitations should also be acknowledged. First, CHG and FI were assessed only at baseline, which precluded evaluation of longitudinal changes, cumulative exposure, or temporal trajectories of these indicators. Second, cardiovascular outcomes were ascertained using self-reported physician diagnoses rather than adjudicated medical records, which may have introduced recall bias or outcome misclassification. Such misclassification may be particularly relevant for heart disease, which encompasses a heterogeneous group of clinical conditions, including heart attack, coronary heart disease, angina, congestive heart failure, and other heart problems, with potentially different pathophysiological mechanisms and diagnostic certainty. If outcome misclassification was non-differential with respect to baseline CHG-FI, it may have attenuated the observed associations and reduced the apparent predictive performance of CHG-FI. However, if misclassification was related to frailty status, healthcare access, or health-seeking behavior, the direction of bias could be more difficult to predict. In addition, because CHARLS did not provide systematic medical-record verification or adjudicated cardiovascular outcomes, we were unable to perform formal internal validation of self-reported CVD, heart disease, or stroke diagnoses. Although our sensitivity analysis excluding cardiovascular events occurring within the first 24 months of follow-up partly reduced the possibility of reverse causation, it did not directly validate the reliability of self-reported outcomes. Therefore, the associations and discrimination estimates should be interpreted with caution. Third, although the 32-item FI was constructed according to the deficit accumulation framework rather than data-driven selection, different FI specifications may capture somewhat different dimensions of frailty. Nevertheless, sensitivity analyses using alternative FI specifications excluding cardiovascular-related, hypertension/diabetes-related, and subjective psychosocial deficits yielded materially consistent results. Fourth, CHG and FI were dichotomized in the joint-exposure and interaction analyses, which may have resulted in information loss and reduced statistical power to detect interaction effects. Therefore, the absence of statistically significant interaction should be interpreted cautiously. Fifth, hypertension and diabetes were included as covariates because they are established cardiovascular risk factors; however, because they were also included as deficits in the original FI, potential redundancy and overadjustment could not be fully excluded. To address this concern, we performed sensitivity analyses using a hypertension/diabetes-free FI and a cardiometabolic-overlap-free FI, and the results remained materially consistent. We also conducted VIF diagnostics across the adjusted Cox models, and all VIF values were below 3, suggesting no evidence of serious multicollinearity. Nevertheless, as with all observational analyses, residual confounding and model-dependence from covariate selection cannot be completely excluded. Sixth, the study population consisted of middle-aged and older Chinese adults, and caution is needed when generalizing these findings to younger populations or other ethnic groups. Because the cohort was rural-predominant according to both household registration and current residence, differences in metabolic risk profiles, frailty phenotypes, healthcare access, diagnostic awareness, and health-seeking behaviors compared with urban or non-Chinese populations may further limit external validity. Accordingly, the CHG-FI threshold values and nonlinear inflection points derived in this study should be interpreted cautiously when applied to other populations and require external validation in more diverse cohorts before broader clinical use. Seventh, CKM staging, CHG-FI construction, and covariate adjustment were not completely independent because hypertension, diabetes, and glucose-related information contributed to more than one analytical component. Although sensitivity analyses excluding hypertension and/or diabetes, using modified FI specifications, and additionally adjusting for baseline CKM stage yielded materially consistent results, structural dependence among these constructs cannot be fully eliminated. Therefore, our findings should be interpreted as risk characterization within the early-stage CKM framework rather than as evidence that CHG-FI represents a fully independent biological pathway. In addition, although CHG-FI was constructed using a product term to represent the combined burden of CHG and FI, this formulation was pragmatic and exploratory rather than biologically definitive. Sensitivity analyses using a standardized additive CHG-FI score yielded materially consistent results, but the optimal method for integrating CHG and FI remains uncertain. Future studies should evaluate additive, weighted, and data-driven formulations in external validation cohorts before clinical use.

Conclusions

In this nationwide cohort of middle-aged and older adults with early-stage CKM syndrome, higher CHG-FI was independently associated with increased risks of new-onset CVD, heart disease, and stroke. Nonlinear analyses suggested that cardiovascular risk increased more steeply across the lower to moderate ranges of CHG-FI. Joint-exposure analyses showed that participants with both high CHG and high FI had the greatest risk of new-onset CVD; however, no statistically significant multiplicative or additive interaction was observed, indicating that CHG-FI reflects the combined burden of metabolic dysregulation and frailty rather than a detectable synergistic effect. Compared with established metabolic indices, CHG-FI showed statistically significant but modest discriminative performance, and its incremental reclassification improvement was limited in magnitude. These findings suggest that CHG-FI may serve as a complementary marker of integrated glucose–lipid metabolic dysregulation and frailty-related vulnerability among individuals with early-stage CKM syndrome, but it should not be considered a standalone clinical prediction tool and requires further external validation before clinical application.

Author contributions

Conceptualization, ZS, KW, JQ, KK, SW and WZ; methodology and Software, ZS, CG, JQ, WW and YX; data curation, ZS and YT; writing—original draft, ZS, KW and CG; writing—review and editing, ZS and WZ; supervision and project administration, WZ. All authors have read and agreed to the published version of the manuscript.

Funding

This work was supported by the Noncommunicable Chronic Diseases-National Science and Technology Major Project (Grant No.2023ZD0503604).

Data availability

No datasets were generated or analysed during the current study.

Declarations

Ethics approval and consent to participate

The study protocol was approved by the Biomedical Ethics Review Committee of Peking University (Approval No. IRB00001052-11015). Written informed consent was obtained from all participants, and the study was conducted in strict adherence to the principles of the Declaration of Helsinki.

Consent for publication

Not applicable.

Competing interests

The authors declare no competing interests.