Joint association of Insulin resistance and frailty index with incident ASCVD and MACE in individuals with cardiovascular-kidney-metabolic syndrome stages 0–3: a multi-cohort study

Abstract

Background

While insulin resistance (IR) and frailty are mechanistically intertwined via shared pathways of chronic inflammation and metabolic dysregulation, their synergistic impact on cardiovascular risk has yet to be fully elucidated. We aimed to evaluate the joint association of IR and frailty index (FI) with the risk of incident atherosclerotic cardiovascular disease (ASCVD) and major adverse cardiovascular events (MACE) in individuals with Cardiovascular-Kidney-Metabolic (CKM) syndrome stages 0–3.

Methods

This study enrolled 259,714 and 2632 CKM syndrome stages 0–3 from the UK Biobank (UKB), and China Health and Retirement Longitudinal Study (CHARLS). Three IR-related indices, namely the triglyceride-glucose index (TyG), metabolic score for IR (METS-IR), and estimated glucose disposal rate (eGDR), were each combined with two frailty assessments, the FI and physical FI (PFI), yielding six composite IR-FI indices for analysis. In the UKB, MACE and ASCVD were the co-primary outcomes, with heart disease and stroke as secondary outcomes. In the CHARLS, the outcomes included CVD, heart disease, and stroke. Multivariate Cox proportional hazards models, restricted cubic spline analyses, Kaplan–Meier analyses and time-dependent receiver operating characteristic curves were employed to assess associations. Mediation analysis and functional proteomic enrichment provided a framework for exploring the potential biological mechanisms.

Results

In the UKB, elevated levels of the IR-FI were associated with increased risks of ASCVD and MACE, with consistent findings observed for CVD, heart disease, and stroke in the CHARLS. Significant positive nonlinear relationships were established for the METS-IR-FI and eGDR-FI regarding cardiovascular outcomes. Mediation analysis revealed that Creactive protein, neutrophils, and leucocyte partially mediated these associations (ratio range: 1.5–12%). Proteomic analyses identified Cytokine-cytokine receptor interaction as a primary regulatory hub, while leukocyte migration, chemotaxis, and neutrophil degranulation were characterized as the central pathological execution mechanisms in this process.

Conclusions

The IR-FI are significantly associated with the incidence of cardiovascular events, an association that appears to be mediated through systemic inflammatory pathways. These novel integrated indices represent promising biomarkers for early screening and targeted clinical intervention among individuals with CKM syndrome stages 0–3.

Supplementary Information

The online version contains supplementary material available at https://doi.org/10.1186/s12933-026-03275-0.

Introduction

The Cardiovascular-Kidney-Metabolic (CKM) syndrome established by the American Heart Association (AHA) in 2023 represents a transformative systemic framework that acknowledges the profound interconnectedness among metabolic dysfunction, renal impairment, and cardiovascular pathology [1, 2]. This consensus-driven staging system provides a rigorous paradigm for risk stratification by documenting the progressive escalation of cardiovascular vulnerability from stage 0 to stage 3 [3]. Within this continuum, cardiovascular disease (CVD) remains the principal driver of both clinical morbidity and the associated economic burden [2, 4–6]. Atherosclerotic cardiovascular disease (ASCVD), encompassing ischemic heart disease (IHD), peripheral artery disease (PAD), and stroke, serves as the primary clinical manifestation and a robust predictor of catastrophic major adverse cardiovascular events (MACE) [7–9]. As these clinical events represent the culmination of terminal multi-organ deterioration, the initial stages of the CKM syndrome provide a critical window for therapeutic interception. Therefore, the precise identification of high-risk individuals within CKM syndrome stages 0–3 is essential for implementing targeted interventions to arrest disease progression and alleviate the escalating global burden of this multi-systemic disorder [2].

Insulin resistance (IR) serves as a primary pathogenic driver of CVD and metabolic syndrome [10, 11]. By inducing impaired glucose uptake, dyslipidemia, and endothelial dysfunction, IR precipitates a cascade of pathological shifts that accelerate atherosclerosis and predispose individuals to adverse cardiovascular events [12–15]. Although the hyperinsulinemic euglycemic clamp technique remains the gold standard for assessment, its technical complexity and prohibitive costs have necessitated the use of surrogate markers such as the triglyceride-glucose (TyG) index, estimated glucose disposal rate (eGDR) and metabolic score for insulin resistance (METS-IR) in large-scale epidemiological research [16–18]. While these indices are effective for identifying high-risk metabolic profiles, they offer a restricted perspective by focusing solely on biochemical derangements while neglecting the broader physiological resilience and cumulative vulnerability of the individual. This limitation is addressed by the Frailty Index (FI), which quantifies cumulative deficits across physical, cognitive, and psychological domains to characterize an individual’s global health status [10, 14, 19]. Notably, frailty is characterized by chronic low-grade inflammation, metabolic dysregulation, and sarcopenia, conditions that engage in a bidirectional pathological synergy with CVD to exacerbate vascular injury [20–22]. Within this framework, constructing composite IR-FI indices may provide superior predictive utility for cardiovascular outcomes, yet the clinical performance of such a combined approach remains insufficiently explored.

Building on previous studies of metabolic and frailty-related indices, we hypothesized that composite IR-FI indices would be associated with increased risks of ASCVD and MACE. Utilizing longitudinal data from two representative cohorts including the UK Biobank (UKB) and the China Health and Retirement Longitudinal Study (CHARLS), this research aimed to rigorously quantify these associations and evaluate the predictive performance of the IR-FI for these cardiovascular outcomes.

Methods

Study design and population

This study analyzed data from the UKB and the CHARLS cohort. In the UKB, a total of 502,248 participants were enrolled between 2006 and 2010. We excluded participants with missing data required for CKM syndrome stages definition, those not classified into CKM syndrome stages 0–3, and those with prevalent ASCVD and MACE at baseline. Individuals with missing outcome variables or covariates were also removed. Finally, 259,714 participants remained for the analysis (Additional file 2: Fig. S1). The CHARLS recruited a total of 17,596 participants between 2011 and 2020. After excluding participants with missing data on related variables, 2,632 participants were included in the FI study and 567 in the PFI study (Additional file 2: Fig. S2).

Definition of CKM syndrome

CKM syndrome stage 0 included participants with normal BMI and waist circumference. CKM syndrome stage 1 identified individuals with elevated BMI or waist circumference (WC) or prediabetes. Stage 2 identified participants with metabolic risk factors (elevated triglycerides, hypertension, diabetes, or metabolic syndrome) or moderate-to-high-risk chronic kidney disease (CKD). Stage 3 was identified based on the presence of very-high-risk KDIGO CKD stages or a high-predicted 10-year CVD risk. Stage 4 was identified based on self-reported CVD (coronary heart disease, angina, myocardial infarction, heart failure, stroke) [1, 2, 23]. Detailed definitional specifications are provided in the Additional file 2.

Definition of IR-related indices, physical frailty (PFI) and frailty index (FI)

We calculated three IR-related indices: TyG, METS-IR, and eGDR. The TyG index was defined as Ln [fasting triglycerides (mg/dL) × fasting glucose (mg/dL) / 2]. The METS-IR was calculated using the formula: Ln [2 × fasting glucose (mg/dL) + fasting triglycerides (mg/dL)] >  BMI (kg/m²) / Ln [HDL-C (mg/dL)]. The eGDR (mg/kg/min) was computed as 21.158 − (0.09 × WC) − (3.407 × HT) − (0.551 × HbA1c), where WC represents waist circumference (cm), HT indicates hypertension (coded as 1 for yes and 0 for no),  and HbA1c denotes glycosylated hemoglobin (%). In the UKB, HbA1c levels were measured using high-performance liquid chromatography (HPLC). The values were converted to percentages using the formula: (0.09148 × HbA1c mmol/mol) + 2.152) [24].

In the study, we calculated two frailty indices: the PFI and the FI. The PFI was defined based on an adaptation of the Fried frailty phenotype. Five criteria were considered: unintentional weight loss, exhaustion, low physical activity, slow gait speed, and low grip strength. Participants were classified according to the number of criteria met: Non-frail (0), Pre-frail (1–2), or Frail (3 or more). The FI in the UKB was constructed from 49 self-reported items covering health conditions, illnesses, physical limitations, and mental well-being. For participants with fewer than ten missing items, the FI score was calculated as the proportion of deficits present among the total possible deficits, yielding a continuous score between 0 and 1. Participants were categorized as non-frail (score ≤ 0.12), Pre-frail (0.12–0.24), or Frail (> 0.24) [25]. In CHARLS, frailty was assessed using a 32-item frailty index (32-FI), which encompasses five domains: comorbidities (13 items), physical function (10 items), mobility (7 items), depression (1 item), and cognition (1 item) [26]. Except for item 32, each individual item was coded as 0 or 1. Item 32, measuring cognition, was a continuous variable ranging from 0 to 1, with higher values indicating poorer cognitive function. The 32-FI for each participant was calculated by summing the scores of existing health deficits and dividing by 32, resulting in a continuous variable ranging from 0 to 1. Higher values of 32-FI represent greater frailty. To optimize the sample size, missing data for participants with an item missing rate of < 10% across the 32 items were imputed using the median values of the relevant items.

The composite IR-FI indices were included in the analysis: TyG-FI, METS-IR-FI, eGDR-FI, TyG-PFI, METS-IR-PFI, and eGDR-PFI. TyG-FI and METS-IR-FI were calculated by multiplying TyG or METS-IR by the FI, respectively. The eGDR-FI was derived by dividing the FI by eGDR. The corresponding PFI indicators were constructed in the same manner, using PFI in place of the FI.

Definition of cardiovascular events

In this study, the primary outcomes of the UKB were ASCVD and MACE. ASCVD encompassed IHD, PAD, and stroke. Incident cases were identified using International Classification of Diseases-Tenth Revision (ICD-10) codes recorded during follow-up. The diagnostic criteria were based on specific code ranges: I20–I25, I47.2, I49.0, I46, I50, I51.4–I51.6, I51.9, and I70.9 indicated IHD; I60–I69 indicated stroke; and I70–I72, and I73.9 indicated PAD [27]. MACE was defined as the first occurrence of any of the following during follow-up: hospitalization for acute myocardial infarction (ICD-10 codes I21 or I23), ischemic stroke (I63 or I64), intracerebral hemorrhage (I61), subarachnoid hemorrhage (I60), unstable angina (I20), or heart failure (I50); or cardiovascular death. Hospitalizations were identified using primary or secondary ICD-10 diagnosis codes from hospital episode statistics, and fatal events were ascertained from death registry data. The first nonfatal hospitalization or cardiovascular death during follow-up was considered to have met the MACE endpoint [28]. To facilitate comparability with the CHARLS cohort, heart disease and stroke were designated as secondary outcomes in the UKB analyses. Heart disease was ascertained using the following ICD-10 codes: I01, I02.0, I11, I13, I20-I25, I26-I28, and I30-I52. Stroke was defined based on ICD-10 codes I60-I69. In the CHARLS, outcomes included CVD, heart disease and stroke, which were defined based on self-reported or physician-diagnosed conditions via questionnaire.

Inflammatory markers

In the UKB, CRP was measured in serum samples using immunoassay analyzers conforming to the ISO 17025:2005 standard, with all results corrected for aliquot dilution and date-of-assay drift as per the UKB Biomarker Assay Quality Procedures (Version 1.2, 2019). Neutrophil and total leukocyte counts were obtained from standard haematological tests performed on fresh whole blood within 24 h of venepuncture using automated hematology analyzers. All biomarkers were assessed at the baseline visit.

Covariates

Covariates included factors considered potential confounders in the relationship between the indices and outcomes [29]. Age at baseline was calculated as the interval between birth date and the assessment date. Sex was obtained from the self-reported data. Other baseline characteristics were collected via touchscreen questionnaires, including self-reported area of residence (urban or rural), marital status (married or other status), education level (less than high school or high school and above), smoking status (never or current or former) and drinking status (never or current or former). Several medical conditions and medication use were treated as binary variables. These variables included hypertension, diabetes, dyslipidemia, as well as use of antihypertensive, antidiabetic, and lipid-lowering medications. BMI, fasting blood glucose (FBG), total cholesterol (TC), low-density lipoprotein cholesterol (LDL-C), and glycated hemoglobin (HbA1c) were analyzed as continuous variables. We constructed two multivariable models. Model 1 adjusted for sex, age, residence, marital status, education level, smoking status, drinking status, hypertension, diabetes, dyslipidemia, and BMI. Model 2 included all covariates in model 1, with further adjustment for use of antihypertensive, antidiabetic, lipid-lowering medications, FBG, TC, LDL-C, and HbA1c. In the UKB, model 1 and model 2 further adjusted for ethnicity (Asian, Black, White, Mixed, Chinese, Other). All models satisfied the proportional hazards assumption as confirmed by Schoenfeld residuals testing. In the CHARLS cohort, all covariates passed formal collinearity diagnostics. To maintain consistency across cohorts, a harmonized covariate set was applied in the UKB analyses. To address the potential collinearity introduced by TC in the UKB context, a sensitivity analysis was conducted for Model 2 with TC excluded, and all remaining covariates satisfied the collinearity threshold. Furthermore, to mitigate the risk of mathematical coupling between the composite IR-FI indices and biochemical covariates, an additional sensitivity analysis was performed with all biochemical parameters and BMI removed from the adjustment set.

Statistical analyses

Independent samples t-tests and Pearson’s chi-square tests were used to compare baseline demographic characteristics across study populations. Continuous variables were reported as mean ± standard deviation (SD), while categorical variables were presented as frequency (percentage, %). For continuous variables that did not satisfy the assumption of normality, median values with interquartile ranges (IQR) are additionally reported in the Additional file 2: Table S1-S2. The baseline characteristics of the subsample used for PFI calculation are also described in the Additional file 2: Table S3.

Firstly, multivariate Cox proportional hazards models were used to explore the associations of IR-FI with cardiovascular events. TyG-FI, METS-IR-FI, eGDR-FI were classified according to tertiles (T1, T2, T3). Subsequently, we utilized restricted cubic spline (RCS) curves to investigate the nonlinear associations between exposure indicators and outcomes, and the corresponding cutoff values identified from the RCS analyses were marked for both cohorts. Time-dependent receiver operating characteristic (ROC) analysis was performed to evaluate and compare the predictive performance of individual IR-related indices, the FI in isolation, and the composite IR-FI indices for incident cardiovascular outcomes. A higher Area Under the Curve (AUC) value indicated superior predictive accuracy. In addition, Harrell’s concordance index (C-index) with 95% confidence intervals (CIs) was calculated to further assess and compare the discriminative performance of individual and combined indices. We further explored the potential mediating factors (C-reactive protein, neutrophils, and leucocyte) underlying the association between the IR-FI and cardiovascular events. To validate the robustness of our findings, we further dichotomized the FI and IR-related indices (TyG, METS-IR, eGDR) into high and low groups, then combined them into four subgroups to analyze the association between this 4-category indicator and outcomes. To formally evaluate the joint effect of IR and frailty on cardiovascular outcomes, both multiplicative and additive interactions were assessed. Multiplicative interaction was examined by incorporating cross-product interaction terms (IR × FI) into the Cox proportional hazards models, while additive interaction was quantified by calculating the Relative Excess Risk due to Interaction (RERI). Finally, Functional enrichment analyses utilizing the Gene Ontology (GO), the Kyoto Encyclopedia of Genes and Genomes (KEGG), and the Reactome database were conducted to elucidate the fundamental signaling pathways and molecular mechanisms that mediate the associations between these indices and cardiovascular events [30]. The significance threshold applied was a Benjamini–Hochberg adjusted P value of less than 0.05. Analyses were conducted using the ‘clusterProfiler’ package in R. The enrichment background was defined as the set of measured proteins that passed quality control and were successfully mapped to gene symbols, rather than all annotated genes.

To examine whether the composite IR-FI indices serve as reliable prognostic markers in individuals at earlier stages of cardiometabolic risk, given that higher CKM stages are inherently associated with greater cardiovascular event risk, we repeated the Cox proportional hazards regression analyses in a subgroup restricted to participants classified as CKM syndrome stages 0–1. To account for the potential competing risk of non-cardiovascular mortality in a frail aging population, a sensitivity analysis was conducted in the UKB using the Fine-Gray subdistribution hazard model, with non-cardiovascular mortality treated as the competing event. Subdistribution hazard ratios (sHR) with 95% confidence intervals were estimated for all six composite IR-FI indices, using the same covariate adjustment structure as the primary Cox regression models. To further evaluate whether the observed associations were influenced by differences in cardiovascular risk across CKM syndrome stages, we conducted additional stage-specific analyses among participants classified as CKM syndrome stages 0, 1, 2, and 3, respectively. These analyses were performed to reduce the potential influence of CKM progression-related risk variation and to examine whether the associations between composite IR-frailty indices and cardiovascular outcomes were consistent within each CKM stage. In addition, to assess the potential impact of missing data, multiple imputation was performed separately in the UKB and CHARLS cohorts using the mice package in R. The imputation models incorporated demographic characteristics, lifestyle factors, clinical covariates, exposure variables, and outcome-related information. The associations between composite IR-frailty indices and cardiovascular outcomes were then re-estimated in the imputed datasets as sensitivity analyses and compared with the primary complete-case results.

Statistical analyses for the UKB and CHARLS were conducted using R (version 4.4.1) and SAS (version 9.4). A two-sided significance threshold of P < 0.05 was applied across all statistical tests.

Results

Baseline characteristics of the participants

The baseline characteristics of the included individuals with CKM syndrome stages 0–3 from two national cohorts (259,714 in the UKB, 2632 in the CHARLS) are summarized in Table 1 and Additional file 2: Table S1-S2. Regarding the distribution of CKM syndrome stages 0–3, the proportions in the UKB were 16.4%, 16.6%, 50.6%, and 16.4% respectively, whereas the CHARLS showed proportions of 20.8%, 21.8%, 38.7%, and 18.6%. The UKB participants had a mean age of 55.8 years and were predominantly White at 91.7%, while the CHARLS had a mean age of 59.7 years. Demographic analysis revealed that 45.6% of the UKB participants were male compared with 41.3% in the CHARLS. In the UKB, 91.9% of participants were urban residents and 81.5% had attained high school education or above. Meanwhile, 12.7% of participants were urban residents and 6.3% had achieved a comparable educational level in the CHARLS. In the UKB, 56.7% of participants were never smokers and 3.8% were never drinkers, compared with 64.2% never smokers and 63.5% never drinkers in the CHARLS.

Table 1.

Baseline characteristics of participants from two cohorts

Characteristics	UK Biobank					P value	CHINA CHARLS					P value
CKM stage	Total	0	1	2	3		Total	0	1	2	3
Participants, N(%)	259,714	42,674 (16.4)	42,955 (16.6)	131,413 (50.6)	42,672 (16.4)		2632	548 (20.8)	575 (21.8)	1019 (38.7)	490 (18.6)
Age, years, mean (s.d.)	55.8 (8.1)	53.0 (8.0)	54.7 (8.0)	58.0 (7.2)	53.1 (8.7)	< 0.001	59.7 (9.6)	59.3 (8.8)	55.8 (8.1)	58.1 (8.0)	67.7 (10.8)	< 0.001
Sex, N(%)						< 0.001						< 0.001
Male	118,475 (45.6)	11,612 (27.2)	14,005 (32.6)	51,730 (39.4)	41,128 (96.4)		1086 (41.3)	247 (45.1)	179 (31.1)	376 (36.9)	284 (58.0)
Female	141,239 (54.4)	31,062 (72.8)	28,950 (67.4)	79,683 (60.6)	1544 (3.6)		1546 (58.7)	301 (54.9)	396 (68.9)	643 (63.1)	206 (42.0)
Ethnicity, N(%)						< 0.001						< 0.001
Asian or Asian British	8943 (3.4)	1934 (4.5)	1549 (3.6)	3996 (3.0)	1464 (3.4)		0 (0.0)	0 (0.0)	0 (0.0)	0 (0.0)	0 (0.0)
Black or Black British	1314 (0.6)	159 (0.4)	240 (0.6)	554 (0.4)	361 (0.8)		0 (0.0)	0 (0.0)	0 (0.0)	0 (0.0)	0 (0.0)
Chinese	826 (0.3)	145 (0.3)	143 (0.3)	391 (0.3)	147 (0.3)		2632 (100.0)	548 (20.8)	575 (21.9)	1019 (38.7)	490 (18.6)
Mixed	8404 (3.2)	1305 (3.1)	1687 (3.9)	3753 (2.9)	1659 (3.9)		0 (0.0)	0 (0.0)	0 (0.0)	0 (0.0)	0 (0.0)
Other ethnic group	1985 (0.8)	310 (0.7)	342 (0.8)	864 (0.7)	469 (1.1)		0 (0.0)	0 (0.0)	0 (0.0)	0 (0.0)	0 (0.0)
White	238,242 (91.7)	38,821 (91.0)	38,994 (90.8)	121,855 (92.7)	38,572 (90.4)		0 (0.0)	0 (0.0)	0 (0.0)	0 (0.0)	0 (0.0)
Residence, N(%)						< 0.001						< 0.001
Urban	238,707 (91.9)	38,830 (91.0)	39,565 (92.1)	120,601 (91.8)	39,711 (93.1)		333 (12.7)	41 (7.5)	95 (16.5)	144 (14.1)	53 (10.8)
Rural	21,007 (8.1)	3844 (9.0)	3390 (7.9)	10,812 (8.2)	2961 (6.9)		2299 (87.3)	507 (92.5)	480 (83.5)	875 (85.9)	437 (89.2)
Marriage status, N(%)						< 0.001						< 0.001
Married or partnered	233,084 (89.7)	37,617 (88.1)	37,711 (87.8)	118,692 (90.3)	39,064 (91.5)		2130 (80.9)	454 (82.8)	499 (86.8)	839 (82.3)	338 (69.0)
Other marital status	26,630 (10.3)	5057 (11.9)	5244 (12.2)	12,721 (9.7)	3608 (8.5)		502 (19.1)	94 (17.2)	76 (13.2)	180 (17.7)	152 (31.0)
Educational level, N(%)						< 0.001						0.027
Below high school	47,982 (18.5)	4857 (11.4)	7484 (17.4)	27,887 (21.2)	7754 (18.2)		2467 (93.7)	524 (95.6)	526 (91.5)	953 (93.5)	464 (94.7)
High school or above	211,732 (81.5)	37,817 (88.6)	35,471 (82.6)	103,526 (78.8)	34,918 (81.8)		165 (6.3)	24 (4.4)	49 (8.5)	66 (6.5)	26 (5.3)
Smoking status, N(%)						< 0.001						< 0.001
Never	147,157 (56.7)	26,834 (62.9)	25,047 (58.3)	73,908 (56.2)	21,368 (50.1)		1689 (64.2)	331 (60.4)	422 (73.4)	703 (69.0)	233 (47.6)
Former	88,969 (34.3)	12,438 (29.1)	14,454 (33.6)	46,542 (35.4)	15,535 (36.4)		209 (7.9)	46 (8.4)	43 (7.5)	85 (8.3)	35 (7.1)
Current	23,588 (9.0)	3402 (8.0)	3454 (8.0)	10,963 (8.3)	5769 (13.5)		734 (27.9)	171 (31.2)	110 (19.1)	231 (22.7)	222 (45.3)
Drinking status, N(%)						< 0.001						< 0.001
Never	9796 (3.8)	1255 (2.9)	1570 (3.7)	5568 (4.2)	1403 (3.3)		1671 (63.5)	341 (62.2)	402 (69.9)	668 (65.6)	260 (53.1)
Former	7526 (2.9)	1120 (2.6)	1113 (2.6)	3890 (3.0)	1403 (3.3)		370 (14.1)	78 (14.2)	61 (10.6)	138 (13.5)	93 (19.0)
Current	242,392 (93.3)	40,299 (94.4)	40,272 (93.8)	121,955 (92.8)	39,866 (93.4)		591 (22.5)	129 (23.5)	112 (19.5)	213 (20.9)	137 (28.0)
Hypertension, N(%)						< 0.001						< 0.001
No	193,798 (74.6)	39,813 (93.3)	36,757 (85.6)	91,389 (69.5)	25,839 (60.6)		1603 (60.9)	516 (94.2)	544 (94.6)	325 (31.9)	218 (44.5)
Yes	65,916 (25.4)	2861 (6.7)	6198 (14.4)	40,024 (30.5)	16,833 (39.4)		1029 (39.1)	32 (5.8)	31 (5.4)	694 (68.1)	272 (55.5)
Diabetes, N(%)						< 0.001						< 0.001
No	239,085 (92.1)	42,438 (99.4)	41,678 (97.0)	120,024 (91.3)	34,945 (81.9)		2198 (83.5)	519 (94.7)	512 (89.0)	824 (80.9)	343 (70.0)
Yes	20,629 (7.9)	236 (0.6)	1277 (3.0)	11,389 (8.7)	7727 (18.1)		434 (16.5)	29 (5.3)	63 (11.0)	195 (19.1)	147 (30.0)
Dyslipidemia, N(%)						< 0.001						< 0.001
No	224,694 (86.5)	40,328 (94.5)	39,025 (90.9)	113,844 (86.6)	31,497 (73.8)		1200 (45.6)	358 (65.3)	243 (42.3)	377 (37.0)	222 (45.3)
Yes	35,020 (13.5)	2346 (5.5)	3930 (9.1)	17,569 (13.4)	11,175 (26.2)		1432 (54.4)	190 (34.7)	332 (57.7)	642 (63.0)	268 (54.7)
Hypertension medications, N(%)						< 0.001						< 0.001
No	216,347 (83.3)	41,245 (96.7)	41,129 (95.7)	120,663 (91.8)	23,845 (55.9)		2172 (82.5)	531 (96.9)	557 (96.9)	740 (72.6)	344 (70.2)
Yes	43,367 (16.7)	1429 (3.3)	1826 (4.3)	10,750 (8.2)	18,827 (44.1)		460 (17.5)	17 (3.1)	18 (3.1)	279 (27.4)	146 (29.8)
Diabetes medications, N(%)						< 0.001						< 0.001
No	257,705 (99.2)	42,665 (100.0)	42,945 (100.0)	130,305 (99.2)	41,790 (97.9)		2537 (96.4)	547 (99.8)	560 (97.4)	971 (95.3)	459 (93.7)
Yes	2009 (0.8)	9 (< 0.1)	10 (< 0.1)	1108 (0.8)	882 (2.1)		95 (3.6)	1 (0.2)	15 (2.6)	48 (4.7)	31 (6.3)
Dyslipidemia medications, N(%)						< 0.001						< 0.001
No	226,882 (87.4)	41,479 (97.2)	40,020 (93.2)	104,646 (79.6)	30,202 (70.8)		2516 (95.6)	543 (99.1)	555 (96.5)	960 (94.2)	458 (93.5)
Yes	32,832 (12.6)	1195 (2.8)	2935 (6.8)	26,767 (20.4)	12,470 (29.2)		116 (4.4)	5 (0.9)	20 (3.5)	59 (5.8)	32 (6.5)
BMI, kg/m2, mean (s.d.)	27.2 (4.6)	22.5 (1.7)	27.9 (3.5)	27.8 (4.7)	29.1 (4.1)	< 0.001	23.7 (4.2)	20.2 (1.9)	25.2 (3.7)	25.0 (4.1)	23.0 (4.2)	< 0.001
FBG, mmol/l, mean (s.d.)	5.1 (1.1)	4.8 (0.6)	4.9 (0.6)	5.1 (1.1)	5.3 (1.7)	< 0.001	6.1 (1.8)	5.6 (0.9)	5.8 (1.4)	6.2 (1.7)	6.5 (2.7)	< 0.001
TC, mmol/l, mean (s.d.)	5.8 (1.1)	5.5 (1.0)	5.6 (1.0)	6.0 (1.1)	5.6 (1.2)	< 0.001	5.0 (1.0)	4.9 (0.9)	5.0 (0.9)	5.2 (1.0)	4.9 (0.9)	< 0.001
LDL-C, mmol/l, mean (s.d.)	3.6 (0.8)	3.3 (0.7)	3.5 (0.7)	3.8 (0.8)	3.6 (1.0)	< 0.001	3.0 (0.9)	2.9 (0.8)	3.0 (0.9)	3.1 (0.9)	2.9 (0.8)	< 0.001
HbA1c,mmol/mol, mean (s.d.)	35.6 (6.0)	33.4 (3.1)	34.7 (4.0)	36.0 (6.0)	37.5 (8.5)	< 0.001	33.9 (7.8)	32.3 (4.5)	33.3 (6.5)	34.5 (8.3)	34.8 (10.3)	< 0.001
TyG, mean (s.d.)	8.7 (0.6)	8.2 (0.3)	8.3 (0.3)	8.9 (0.5)	9.1 (0.5)	< 0.001	8.7 (0.6)	8.4 (0.5)	8.6 (0.6)	8.9 (0.6)	8.7 (0.7)	< 0.001
METS-IR, mean (s.d.)	39.7 (9.0)	30.1 (3.1)	38.4 (5.8)	41.1 (8.6)	46.4 (8.5)	< 0.001	35.6 (8.6)	28.5 (3.8)	37.6 (7.7)	38.6 (8.6)	35.1 (9.1)	< 0.001
eGDR, mean (s.d.)	9.2 (2.2)	11.2 (1.2)	9.7 (1.7)	8.9 (2.1)	7.8 (2.2)	< 0.001	9.3 (2.3)	11.5 (1.2)	10.3 (1.4)	8.0 (2.0)	8.7 (2.4)	< 0.001
FI, mean (s.d.)	0.110 (0.065)	0.093 (0.057)	0.106 (0.063)	0.113 (0.067)	0.120 (0.068)	< 0.001	0.113 (0.066)	0.107 (0.063)	0.101 (0.058)	0.113 (0.063)	0.137 (0.076)	< 0.001
PFI, mean (s.d.)	0.743 (0.802)	0.636 (0.718)	0.772 (0.800)	0.760 (0.823)	0.768 (0.809)	< 0.001	1.146 (0.946)	1.229 (0.949)	0.975 (1.000)	0.953 (0.853)	1.455 (0.962)	< 0.001
TyG-FI, mean (s.d.)	1.0 (0.6)	0.8 (0.5)	0.9 (0.5)	1.0 (0.6)	1.1 (0.6)	< 0.001	1.0 (0.6)	0.9 (0.5)	0.9 (0.5)	1.0 (0.6)	1.2 (0.7)	< 0.001
METS-IR-FI, mean (s.d.)	4.5 (3.2)	2.8 (1.8)	4.1 (2.7)	4.8 (3.3)	5.7 (3.7)	< 0.001	4.0 (2.6)	3.0 (1.8)	3.8 (2.4)	4.3 (2.7)	4.8 (3.1)	< 0.001
eGDR-FI, mean (s.d.)	0.014 (0.038)	0.008 (0.006)	0.012 (0.009)	0.015 (0.020)	0.020 (0.085)	< 0.001	0.013 (0.010)	0.010 (0.006)	0.010 (0.007)	0.015 (0.010)	0.018 (0.013)	< 0.001
TyG-PFI, mean (s.d.)	6.5 (7.1)	5.2 (5.9)	6.4 (6.7)	6.8 (7.4)	7.0 (7.4)	< 0.001	9.9 (8.3)	10.2 (8.0)	8.6 (8.9)	8.4 (7.6)	12.6 (8.4)	< 0.001
METS-IR-PFI, mean (s.d.)	30.6 (36.2)	19.2 (21.9)	30.3 (33.0)	32.5 (38.4)	36.4 (40.8)	< 0.001	38.5 (33.7)	34.2 (27.3)	34.6 (35.6)	34.9 (33.5)	50.1 (35.7)	< 0.001
eGDR-PFI, mean (s.d.)	0.094 (0.341)	0.058 (0.066)	0.084 (0.096)	0.098 (0.168)	0.128 (0.777)	< 0.001	0.131 (0.116)	0.110 (0.090)	0.098 (0.103)	0.126 (0.119)	0.176 (0.126)	< 0.001

P-values less than 0.05 (P < 0.05) were considered significant

CHARLS = China Health and Retirement Longitudinal Study; TyG = triglyceride-glucose, METS-IR = Metabolic score for insulin resistance, eGDR = estimated Glucose Disposal Rate, FI = frailty index, PFI = physical frailty index, s.d. = standard deviation; N = Number; CKM = cardiovascular-kidney-metabolic, BMI = body mass index, FBG = Fasting blood glucose, TC = Total cholesterol, LDL-C = Low density lipoprotein cholesterol, HbA1c = Glycated hemoglobin A1c

Across both cohorts, all three IR-related indices, both frailty indices, and the composite IR-FI indices differed significantly across CKM syndrome stages 0–3. Additional file 2: Tables S1-2 present the median values with interquartile ranges for continuous variables that did not satisfy normality testing; however, given the large sample sizes employed in this study, t-tests remain statistically robust and are considered appropriate for group comparisons. In addition, Additional file 2: Table S3 presents the baseline characteristics of the subsample used for PFI calculation. No significant differences were observed between this subsample and the FI analytical sample with respect to residential area, education level, disease history, medication use, FBG, TC, LDL-C, TyG, or glycated hemoglobin, supporting the representativeness of the PFI subsample.

Association of composite insulin resistance-frailty indices (IR-FI) with cardiovascular events

Figure 1 demonstrates that composite IR-FI and IR-PFI indices are significantly associated with cardiovascular events based on model 2 in CKM syndrome stages 0–3 across two cohorts. Within the UKB, the eGDR demonstrated a significant protective effect against ASCVD and MACE, whereas the FI and PFI were identified as significant risk factors. For the METS-IR, individuals in the T3 showed significantly increased risks for ASCVD and MACE by 6% and 10% (HR, 95%CI: 1.06, 1.01–1.11; 1.10, 1.04–1.16), respectively (Fig. 1 and Additional file 2: Tables S4-5). The IR-FI at the T2 and T3 was associated with elevated risks for ASCVD ranging from 14 to 18% (TyG-FI: 1.14, 1.10–1.18; METS-IR-FI: 1.17, 1.13–1.21; eGDR-FI: 1.18, 1.14–1.22) and 42% to 47% (TyG-FI: 1.42, 1.38–1.47; METS-IR-FI: 1.45, 1.40–1.50; eGDR-FI: 1.47, 1.42–1.53), respectively. Similarly, the risks for MACE in these groups increased by 13% to 16% (TyG-FI: 1.13, 1.09–1.17; METS-IR-FI: 1.16, 1.12–1.21; eGDR-FI: 1.15, 1.11–1.20) and 39% to 43% (TyG-FI: 1.39, 1.34–1.44; METS-IR-FI: 1.43, 1.37–1.48; eGDR-FI: 1.43, 1.38–1.49). Consistent with this trend, the T3 of the IR-PFI was associated with risk increases of 16% to 18% for ASCVD and 18% to 19% for MACE. When TC was excluded from Model 2, the results are presented in Additional file 2: Table S6 and remained largely consistent with those of the original Model 2. Additional file 2: Table S7 demonstrates that higher tertiles of all six composite IR-FI indices were significantly associated with increased risks of heart disease, and that the IR-PFI indices at T3 were additionally associated with a significantly elevated risk of stroke.

Fig. 1.

Associations of insulin resistance-frailty index with cardiovascular events in individuals with Cardiovascular-Kidney-Metabolic syndrome stages 0–3. Note: The model was adjusted for sex, age, race, residence, marital status, education level, smoking status, drinking status, hypertension, diabetes, dyslipidemia, BMI, antihypertensive, antidiabetic, lipid-lowering medications, FBG, TC, LDL-C, HbA1c P-values less than 0.05 (p < 0.05) were considered significant. UKB = UK Biobank, CHARLS = China Health and Retirement Longitudinal Study, TyG = triglyceride-glucose, METS-IR = Metabolic score for insulin resistance, eGDR = estimated Glucose Disposal Rate, FI = frailty index, PFI = physical frailty, ASCVD = Atherosclerotic Cardiovascular Disease, MACE = major adverse cardiovascular events, HR = hazard ratio, BMI = body mass index, FBG = Fasting blood glucose, TC = Total cholesterol, LDL-C = Low density lipoprotein cholesterol, HbA1c = Glycated hemoglobin A1c.

In the CHARLS, the T3 of the IR-FI and IR-PFI were associated with CVD risk elevations of 37% to 49% (TyG-FI: 1.44, 1.18–1.76; METS-IR-FI: 1.37, 1.12–1.68; eGDR-FI: 1.49, 1.19–1.87) and 59% to 66% (TyG-PFI: 1.66, 1.12–2.47; METS-IR-PFI: 1.59, 1.06–2.37), respectively. Further analysis in the Chinese cohort indicated that the T3 of the IR-FI compared with T1 was linked to significant risk increases of 36% to 39% for heart disease (TyG-FI: 1.36, 1.09–1.70; METS-IR-FI: 1.36, 1.08–1.70; eGDR-FI: 1.39, 1.09–1.79) and 56% to 76% for stroke (TyG-FI: 1.76, 1.22–2.54; METS-IR-FI: 1.56, 1.09–2.24; eGDR-FI: 1.74, 1.17–2.59) (Fig. 1 and Additional file 2: Tables S8-10). This trend was more attenuated in the UKB where the T3 of the IR-PFI showed a stroke risk increase of only 9% to 11%. These findings remained fundamentally consistent across the primary and sensitivity analyses conducted in model 1. Although several effect estimates were modest in magnitude, the associations were directionally consistent across different composite indices, cohorts, and cardiovascular outcomes, supporting the robustness of the overall association pattern.

Association of continuous composite insulin resistance-frailty indices with cardiovascular events

Figure 2 and Additional file 2: S3-5 evaluate the nonlinear relationship between IR-FI and cardiovascular events. In the CHARLS, the association of METS-IR-FI with CVD and heart disease showed a significant positive nonlinear relationship (CVD: P value = 0.031; heart disease: P value = 0.042), while the association between eGDR-FI and CVD exhibited a significant positive nonlinear relationship (P value = 0.017). Conversely, within the UKB, significant nonlinear relationships were observed for both METS-IR-FI and eGDR-FI in their associations with ASCVD and MACE (P value < 0.001). Furthermore, TyG-PFI and eGDR-PFI showed significant nonlinear associations with ASCVD (P value = 0.038 and < 0.001). Regarding MACE in the UKB, only eGDR-PFI displayed a significant nonlinear relationship (P value < 0.001) (Additional file 2: Figure S3).

Fig. 2.

Restricted cubic spline curve illustrating the associations of insulin resistance-frailty index with cardiovascular events in individuals with Cardiovascular-Kidney-Metabolic syndrome Stages 0–3. Note: The model was adjusted by for sex, age, race, residence, marital status, education level, smoking status, drinking status, hypertension, diabetes, dyslipidemia, BMI, antihypertensive, antidiabetic, lipid-lowering medications, FBG, TC, LDL-C, HbA1c. UKB = UK Biobank, CHARLS = China Health and Retirement Longitudinal Study, TyG = triglyceride-glucose, METS-IR = Metabolic score for insulin resistance, eGDR = estimated Glucose Disposal Rate, FI = frailty index ASCVD = Atherosclerotic Cardiovascular Disease, MACE = major adverse cardiovascular events, HR = hazard ratio, CI = confidence interval, BMI = body mass index, FBG = Fasting blood glucose, TC = Total cholesterol, LDL-C = Low density lipoprotein cholesterol, HbA1c = Glycated hemoglobin A1c

Cumulative incidence of cardiovascular events stratified by insulin resistance-frailty index

Additional file 2: Figures S6-7 illustrate the incidence rates of cardiovascular events stratified by tertiles of the IR-FI. Within both the UKB and CHARLS, significant variability was observed among the three index groups (P value < 0.001). Specifically, a clear dose-response relationship was evident where higher tertiles of the IR-FI were associated with a progressively increased incidence of cardiovascular events.

Time-dependent ROC curves of combined insulin resistance and frailty indices for predicting 5-year cardiovascular events

Given that the TyG and METS-IR serve as risk associated indicators of IR while the eGDR represents a protective factor, we prioritized the former two indices for predictive evaluation to ensure clinical comparability (Fig. 3). Time-dependent ROC analysis indicated that among individuals with CKM syndrome stages 0–3, eGDR demonstrated modestly superior predictive performance for five-year cardiovascular events compared with TyG and METS-IR, with AUC values ranging from 0.715 to 0.723. Similarly, the FI achieved slightly better discriminative performance than the PFI, with AUC values ranging from 0.712 to 0.727. Notably, the integration of IR-related indices with FI or PFI yielded a marginal improvement in predictive performance over either component used in isolation. Consistent with these findings, evaluation of all six composite IR-FI indices across both cohorts similarly demonstrated modestly improved predictive performance relative to individual IR-related indices alone (Additional file 2: Fig. S8). The C-index analysis yielded findings consistent with the time-dependent ROC results. Combined models incorporating both IR-related indices and frailty indices generally demonstrated modestly higher discriminative performance than individual indices alone (Additional file 2: Table S11).

Fig. 3.

Time-dependent ROC curves of combined insulin resistance and frailty index for predicting 5-year cardiovascular events in the UK Biobank. Note: The model was adjusted by for sex, age, ethnicity. ROC = Receiver Operating Characteristic, AUC = Area Under the Curve, TyG = triglyceride-glucose, METS-IR = Metabolic score for insulin resistance, eGDR = estimated Glucose Disposal Rate, FI = frailty index ASCVD = Atherosclerotic Cardiovascular Disease, MACE = major adverse cardiovascular events

Mediating effects of inflammatory markers

Mediation analyses were performed to quantify the mediating role of inflammatory markers in the associations between the IR-FI or IR-PFI and incident cardiovascular events (Fig. 4 and Additional file 2: S9). CRP, neutrophils, and leucocyte mediated 1.5% to 8.8% of the association between the IR-PFI and ASCVD (TyG-PFI: CRP: 3.5%; neutrophils: 2.2%; leucocyte: 1.5%; METS-IR-PFI: CRP: 4.3%; neutrophils: 2.4%; leucocyte: 1.6%; eGDR-PFI: CRP: 8.8%; neutrophils: 7.7%), and 1.6% to 12.0% of the association with MACE (TyG-PFI: CRP: 3.9%; neutrophils: 2.7%; leucocyte: 1.6%; METS-IR-PFI: CRP: 5.0%; neutrophils: 3.1%; leucocyte: 1.8%; eGDR-PFI: CRP: 12.0%; neutrophils: 10.9%). Furthermore, regarding the IR-FI, these inflammatory mediators were found to mediate 1.2% to 3.5% of the effect on ASCVD (TyG-FI: CRP: 2.8%; neutrophils: 1.8%; leucocyte: 1.2%; METS-IR-FI: CRP: 3.5%; neutrophils: 2.0%; leucocyte: 1.5%) and 1.5% to 4.6% of the effect on MACE (TyG-FI: CRP: 3.7%; neutrophils: 2.7%; leucocyte: 1.5%; METS-IR-FI: CRP: 4.6%; neutrophils: 2.8%; leucocyte: 1.8%). These results suggest that systemic inflammation serves as a modest but significant biological intermediary in the pathological pathways linking IR-FI to cardiovascular events.

Fig. 4.

Mediation analysis of insulin resistance-physical frailty index and cardiovascular events in the UK Biobank. Note: The model was adjusted by for sex, age, race, residence, marital status, education level, smoking status, drinking status, hypertension, diabetes, dyslipidemia, BMI, antihypertensive, antidiabetic, lipid-lowering medications, FBG, TC, LDL-C, HbA1c. Only statistically significant mediation pathways are presented. *, P < 0.05; **, P < 0.01; ***, P < 0.001. TyG = triglyceride-glucose, METS-IR = Metabolic score for insulin resistance, eGDR = estimated Glucose Disposal Rate, PFI = physical frailty index ASCVD = Atherosclerotic Cardiovascular Disease, MACE = major adverse cardiovascular events, CRP = C-reactive protein, BMI = body mass index, FBG = Fasting blood glucose, TC = Total cholesterol, LDL-C = Low density lipoprotein cholesterol, HbA1c = Glycated hemoglobin A1c

Functional enrichment analyses of composite insulin resistance-frailty indices and cardiovascular events

To further elucidate the molecular underpinnings of the observed clinical associations, functional enrichment analyses were performed across the GO, Reactome, and KEGG databases as illustrated in Fig. 5 and Additional file 1. Within the GO biological process analysis, pathways related to leukocyte migration, chemotaxis, and taxis were consistently enriched for all combined indices (GeneRatio > 12%). The Reactome pathway analysis further emphasized the critical role of neutrophil degranulation and interleukin signaling in mediating the pathological effects of these combined indices. Concurrently, the KEGG analysis identified the Cytokine-cytokine receptor interaction as central regulatory hubs across both ASCVD and MACE (GeneRatio > 25%).

Fig. 5.

Functional enrichment analyses of insulin resistance-frailty index and cardiovascular events utilizing the Gene Ontology, the Kyoto Encyclopedia of Genes and Genomes, and the Reactome database. Note: The model was adjusted for sex, age, residence, marital status, education level, smoking status, drinking status, hypertension, diabetes, dyslipidemia, BMI, antihypertensive, antidiabetic, lipid-lowering medications, FBG, TC, LDL-C, HbA1c. GO = Gene Ontology, KEGG = Kyoto Encyclopedia of Genes and Genomes, TyG = triglyceride-glucose, METS-IR = Metabolic score for insulin resistance, eGDR = estimated Glucose Disposal Rate, FI = frailty index ASCVD = Atherosclerotic Cardiovascular Disease, MACE = major adverse cardiovascular events, CRP = C-reactive protein, BMI = body mass index, FBG = Fasting blood glucose, TC = Total cholesterol, LDL-C = Low density lipoprotein cholesterol, HbA1c = Glycated hemoglobin A1c

Joint exposure analysis

Further evaluation of the synergistic impact of IR and frailty on cardiovascular events involved stratifying participants into four distinct categories according to high or low levels of their respective indices (Additional file 2: Tables S12-15). HRs were calculated using the group characterized by low IR and low FI as the reference, although the grouping for the eGDR was reversed to account for its protective nature. Compared with the reference group, the other three categories demonstrated significantly increased risks for cardiovascular events to varying degrees. Furthermore, formal interaction analysis within the UKB revealed significant multiplicative interactions between IR and FI regarding the risk of ASCVD, MACE, and stroke. In the UKB, statistically significant additive interactions were identified between TyG and FI, and between METS-IR and FI, respectively.

Sensitivity analyses

To mitigate the potential influence of mathematical coupling between the composite indices and biochemical covariates, a sensitivity analysis was conducted with all biochemical parameters and BMI excluded from the adjustment set. The results, presented in Additional file 2: Table S16, confirmed that the associations between the composite IR-FI indices and both ASCVD and MACE remained robust. Among individuals classified as CKM syndrome stages 0–1, the T3 of all six composite IR-FI indices was associated with significantly elevated risks of ASCVD and MACE, with risk increases ranging from 12 to 47% and 11% to 43%, respectively (Additional file 2: Table S17). To evaluate the robustness of the primary findings to competing risks, Fine-Gray subdistribution hazard models were fitted in the UKB cohort with non-CVD mortality as the competing event. As presented in Additional file 2: Table S18, all six composite IR-FI indices remained significantly associated with cardiovascular outcomes under the competing risks framework, with subdistribution hazard ratios directionally and quantitatively consistent with the primary Cox regression estimates, confirming that the observed associations are not materially influenced by informative censoring from non-CVD mortality. In stage-specific analyses stratified by CKM syndrome stage, the associations between composite IR-frailty indices and cardiovascular outcomes were generally consistent across participants in CKM syndrome stages 0, 1, 2, and 3, suggesting that the main findings were not solely driven by the progressive increase in cardiovascular risk across CKM stages (Additional file 2: Table S19). In addition, the multiple imputation analyses using the mice package yielded results broadly consistent with the primary complete-case analyses. In the UKB, the highest tertiles of TyG-FI, METS-IR-FI, and eGDR-FI remained significantly associated with increased risks of ASCVD, with HRs of 1.46, 1.49, and 1.52, respectively, and MACE, with HRs of 1.43, 1.47, and 1.49, respectively. Similar patterns were observed for the PFI-based composite indices (Additional file 2: Table S20). In CHARLS, the highest tertiles of TyG-PFI, METS-IR-PFI, and eGDR-PFI also remained associated with increased cardiovascular risk in the fully adjusted model. These sensitivity analyses support the robustness of the observed associations and suggest that they were unlikely to be materially affected by CKM stage progression or missing data handling (Additional file 2: Table S21).

Discussion

This study evaluated the associations between composite IR-FI indices and cardiovascular outcomes among individuals with CKM syndrome stages 0 to 3 across two independent cohorts. Higher IR-FI/PFI levels were consistently associated with increased risks of ASCVD, MACE, CVD, heart disease, and stroke. RCS analyses suggested nonlinear associations for several composite indices, while mediation and proteomic enrichment analyses provided supportive evidence for inflammatory immune pathways. These findings suggest that composite IR-FI indices may help characterize cardiovascular risk among individuals with CKM syndrome stages 0 to 3.

The TyG, METS-IR, and eGDR serve as reliable surrogate markers for IR [17, 31, 32]. Consistent with prior investigations, elevated levels of the TyG and METS-IR are closely associated with increased cardiovascular risk among individuals in CKM syndrome stages 0–3, while the eGDR exhibits a protective effect [2, 10, 33]. Parallel to the predictive utility of the METS-IR regarding stroke, our findings confirm that the eGDR alone achieves area under the curve values between 0.71 and 0.72 for cardiovascular outcomes after adjusting for age and sex [2]. From a distinct dimension, the FI serves as an indicator of diminished physiological reserve and heightened susceptibility to stressors [34, 35]. We defined two versions including PFI based on five characteristics to facilitate rapid community or epidemiological screening and a more comprehensive FI comprising forty-nine items in the UKB and thirty-two items in the CHARLS for nuanced physiological evaluation. Our findings confirm that higher FI/PFI lead to an increased risk of ASCVD, MACE and CVD. Existing literature establishes the TyG as a causal determinant of frailty and further indicates that frailty synergizes with IR to heighten cardiovascular risk within the Chinese population [36, 37]. These premises provide a strong rationale for evaluating IR and frailty jointly. Although we observed evidence of joint associations between IR-related indices and frailty, the formal interaction analyses did not consistently support a synergistic effect. Therefore, the combined indices should be interpreted primarily as measures of cumulative or overlapping vulnerability rather than proof of biological synergy.

The IR-FI and IR-PFI integrate metabolic dysfunction and frailty into composite measures, providing a practical framework for evaluating their combined association with cardiovascular outcomes. Importantly, these product- or ratio-based indices were not intended to replace formal multiplicative or additive interaction models, but rather to provide a single pragmatic metric reflecting the combined burden of IR and frailty for risk stratification. Although previous studies proposed the atherogenic index of plasma (AIP)-FI, that marker relies solely on lipid parameters and fails to reflect glucose or HbA1c, thereby providing an incomplete characterization of IR status [14]. Furthermore, investigations involving a single metabolic index combined with a single FI in a single cohort often lack robustness and generalizability. To address these limitations, we calculated six variations of the composite IR-FI indices across multiple cardiovascular outcomes. The consistent associations observed between these indices and the risk of ASCVD and MACE ensure the robustness of our conclusions. The composite indices showed modestly improved predictive performance compared with individual IR-related indices, suggesting potential value for risk stratification, although further validation is needed before clinical application.

The underlying mechanisms by which the newly proposed composite IR-FI indices influence cardiovascular risk have remained largely unaddressed in previous research. Extensive evidence suggests that IR triggers the release of pro-inflammatory cytokines, precipitates oxidative stress, and impairs endothelial function, collectively accelerating atherosclerosis and thrombotic tendencies [12, 38–40]. Simultaneously, frailty is characterized by chronic systemic inflammation, impaired immune responses, and mitochondrial dysfunction, all of which contribute to accelerated vascular aging and endothelial damage [41, 42]. Our findings suggest that inflammatory pathways may partly contribute to the observed associations between IR-FI indices and cardiovascular outcomes. Exploratory mediation analyses suggested partial statistical mediation by CRP, leukocytes, and neutrophils [43]. Although these proportions are modest, they are consistent with findings from large-scale epidemiological studies examining complex cardiometabolic pathways, in which the total effect on cardiovascular risk is characteristically driven by a convergence of biological mechanisms rather than any single mediating pathway. It is important to note that CRP, leucocytes, and neutrophils represent only a limited subset of the broader inflammatory milieu. However, the interaction analyses did not consistently support a synergistic effect. In the UKB, negative RERI estimates for TyG and METS-IR suggested sub-additive rather than synergistic interaction on the additive scale, whereas in CHARLS most multiplicative and additive interaction tests were non-significant. Therefore, the combined effect of insulin resistance and frailty should be interpreted cautiously as reflecting cumulative or overlapping vulnerability rather than definitive biological synergy.

The proteomic enrichment analyses provided additional exploratory biological context. Because proteomic data were measured only in a UKB subsample, these findings should not be interpreted as directly generalizable to the entire analytic population. Enriched pathways included cytokine-cytokine receptor interaction, leukocyte migration, chemotaxis, interleukin signaling, and neutrophil degranulation. These pathways are consistent with inflammation- and immune-related processes, but they do not establish causal mechanisms.

This study has several strengths. We used two independent longitudinal cohorts, evaluated six composite IR-FI/PFI indices, and assessed multiple cardiovascular outcomes. The inclusion of both comprehensive FI and PFI allowed us to examine frailty from different perspectives. In addition, mediation and proteomic enrichment analyses provided supportive biological context for the observed epidemiological associations. These features strengthen the robustness of the findings and provide a basis for further validation.

Several limitations of this investigation warrant consideration. Firstly, the absence of repeated measurements for the IR-FI precluded the assessment of cumulative exposure and temporal fluctuations throughout the follow-up period. All inflammatory biomarkers employed in the mediation analysis were assessed at baseline only. The absence of repeated measurements means that transient conditions such as acute infection or physical activity may have influenced these values, potentially limiting their ability to reflect chronic systemic inflammation. Additionally, because the exposure, mediators, and outcomes were not temporally separated at distinct assessment points, the causal directionality assumed in the mediation framework warrants cautious interpretation. Secondly, cardiovascular outcomes in the CHARLS were derived from self-reported data, which may introduce recall bias and lacks the precision of official diagnostic records or specific event timing. The PFI-related analyses in the CHARLS are based on a smaller subsample and should be interpreted with appropriate caution regarding statistical power and potential residual selection bias. Thirdly, while the composite IR-FI indices demonstrated modestly superior predictive performance over individual components, formal evaluation of clinical utility through net reclassification improvement, calibration assessment, and decision-curve analysis was not conducted in the present study. Future investigations incorporating these metrics would provide a more comprehensive assessment of the clinical applicability of the composite indices. Furthermore, inherent discrepancies between the UKB and CHARLS regarding outcome selection and the specific components of frailty definitions may restrict the direct comparability of results between the two cohorts. Finally, the observational nature of this research prevents the establishment of definitive causal inferences. Future investigations employing Mendelian Randomization and mechanistic studies are essential to validate these findings and elucidate the underlying biological pathways.

Conclusions

Across these two longitudinal cohorts, elevated levels of the composite IR-FI/PFI indices among individuals in CKM syndrome stages 0–3 were closely associated with a heightened risk of multiple cardiovascular events. While inflammatory markers and immune related signaling pathways offer a theoretical framework for the underlying mechanisms, future research should prioritize repeated assessments of the combined index along with further mechanistic validation. These findings establish the IR-FI as a reliable tool for optimizing risk stratification and a practical biomarker for targeted interventions in the stages 0–3 of CKM syndrome.

Abbreviations

UKB

UK Biobank

CHARLS

China Health and Retirement Longitudinal Study

TyG

Triglyceride-glucose

METS-IR

Metabolic score for insulin resistance

eGDR

Estimated Glucose Disposal Rate

FI

Frailty index

PFI

Physical frailty

s.d.

Standard deviation

N

Number

CKM

Cardiovascular-kidney-metabolic

BMI

Body mass index

WC

Waist circumference

FBG

Fasting blood glucose

TC

Total cholesterol

LDL-C

Low density lipoprotein cholesterol

HbA1c

Glycated hemoglobin A1c

GO

Gene Ontology

KEGG

Kyoto Encyclopedia of Genes and Genomes

RCS

Restricted cubic spline

ROC

Receiver operating characteristic

ICD-10

Classification of Diseases-Tenth Revision

CKD

Chronic kidney disease

IHD

Ischemic heart disease

PAD

Peripheral artery disease

AIP

Atherogenic index of plasma

Author contributions

W.Z. had full access to all the data in the study, take responsibility for the integrity of the data and the accuracy of the data analysis. Y.Q.H.,Y.D.X.,W.G.,L.J.C., and H.J.Z. contribute equally to this work. Concept and design: Y.Q.H., Y.D.X. and W.G.. Acquisition, analysis, or interpretation of data: Y.Q.H., Y.D.X., Z.Y.Z., T.W., T.Y.S., Y.L., X.S.,Y.Y., and H.W.. Drafting of the article: Y.Q.H., Y.Z., Y.D.X. and Y. Y. Critical revision of the article for important intellectual content: All authors. Statistical analysis: Y.Q.H.. Obtained funding: Y.Y. and H.W.. Administrative, technical, or material support: W.Z, X.S.. Supervision: H.W..

Funding

This study was supported by grants from the National Natural Science Foundation of China (82403381 and 82400854), the Noncommunicable Chronic Diseases-National Science and Technology Major Project (2026ZB0556800), the National Key R&D Program of China (2022YFD2101500), the Sailing Program of Shanghai Rising-Star Program (24YF2722700), Innovative research team of high-level local universities in Shanghai, Zhejiang Province Medical and Health Science and Technology Project (2025ZR046) and Medical-Engineering Cross Foundation of Shanghai Jiao Tong University (YG2025LC14).

Availability of data and materials

The datasets that were used and evaluated in this study can be obtained from the corresponding author upon making a reasonable request.

Declarations

Ethical approval and consent to participate

The UK Biobank received ethical approval from the North West Multi-Center Research Ethics Committee (Approved Research ID: 194423, Approval date: October 30, 2024). All participants gave written informed consent before enrollment in the study, which was conducted in accordance with the principles of the Declaration of Helsinki. The CHARLS datasets were available at http://charls.pku.edu.cn/en/. All participants provided written informed consent. All participants were informed and agreed to participate in this study.

Consent for publication

Not applicable.

Competing interest

The authors declare no competing interests.