Associations of the triglyceride-glucose index, lipid-derived parameters, and their joint exposure with incident cardiometabolic multimorbidity: a nationwide prospective cohort study

Abstract

Background

The triglyceride-glucose (TyG) index and non-traditional lipid parameters (nTLPs) are accessible markers of cardiometabolic risk; however, their joint association with incident cardiometabolic multimorbidity (CMM) and the incremental predictive value of adding nTLPs to the TyG index remain unclear.

Methods

This investigation utilized longitudinal datasets sourced from the China Health and Retirement Longitudinal Study (CHARLS), encompassing 7159 eligible participants. Kaplan–Meier analysis, multivariable Cox regression, and restricted cubic splines were used to evaluate associations between joint TyG–nTLP exposure and incident CMM. Subgroup and sensitivity analyses assessed robustness. Model discrimination and incremental predictive value were examined using receiver operating characteristic curves (ROC), net reclassification improvement (NRI), and integrated discrimination improvement (IDI).

Results

The median follow-up duration was 9 years, during which 758 participants developed CMM, corresponding to a cumulative incidence of 10.59%. Compared with participants in the dual-low groups for the TyG index and nTLPs, those in the dual-high groups had a higher risk of developing CMM. The corresponding fully adjusted hazard ratios (HRs) were 1.80 (95% CI 1.51–2.14) when both the TyG index and AIP were at high levels, 2.16 (95% CI 1.74–2.67) when both the TyG index and Non-HDL-C were at high levels, 2.10 (95% CI 1.72–2.57) when both the TyG index and CRI-I were at high levels, 2.05 (95% CI 1.66–2.53) when both the TyG index and CRI-II were at high levels, and 1.95 (95% CI 1.62–2.36) when both the TyG index and LCI were at high levels. Adding nTLPs to the TyG-based model resulted in only a modest incremental improvement in predictive performance.

Conclusion

Among middle-aged and older adults, jointly elevated TyG index and nTLPs were associated with a higher risk of incident CMM, with the associations being more evident for incident diabetes and stroke than for incident heart disease. The combined assessment of the TyG index and nTLPs may provide complementary information for the early identification and risk stratification of individuals at elevated risk of CMM.

Supplementary Information

The online version contains supplementary material available at https://doi.org/10.1186/s12933-026-03307-9.

Research insights

What is currently known about this topic?

• CMM is an increasing public health burden among middle-aged and older adults.

• The TyG index and nTLPs are accessible markers of glucose-lipid metabolic disturbances, but their joint associations with incident CMM remain unclear.

What is the key research question?

• Are joint exposure categories defined by the TyG index and nTLPs associated with incident CMM in middle-aged and older Chinese adults?

What is new?

• Dual-high exposure to the TyG index and nTLPs was associated with a higher risk of incident CMM.

• The observed associations appeared to be mainly driven by incident diabetes and stroke.

• Multiplicative interactions were observed for selected nTLPs, whereas additive interaction analyses did not provide evidence of a clear positive additive interaction.

How might this study influence clinical practice?

• Joint assessment of the TyG index and selected nTLPs may provide supplementary information for early CMM risk stratification and help identify individuals at higher risk of CMM.

Introduction

Against the backdrop of population aging and increasing life expectancy, multimorbidity has emerged as a major public health concern among middle-aged and older adults worldwide [1]. Cardiometabolic diseases (CMDs) represent a group of closely related metabolic and vascular disorders, such as diabetes, heart disease, and stroke [2]. Cardiometabolic multimorbidity (CMM) refers to the simultaneous presence of two or more CMDs in an individual [3–5]. Compared with a single CMD, CMM reflects the cumulative progression of cardiometabolic abnormalities from a single-disease state to a multimorbid state, and is usually accompanied by more complex clinical management needs, more pronounced functional impairment, and a higher risk of mortality [6, 7]. Previous evidence indicates that individuals with CMM at age 60 have an average reduction in life expectancy of approximately 15 years [7]. Moreover, data from the 2017–2018 National Health and Nutrition Examination Survey (NHANES) showed that CMM was highly prevalent in later life, affecting 34.8% of adults between 60 and 79 years of age and 42.9% of individuals aged 80 years and above [8]. In China, nationally representative survey data have similarly indicated an upward trend in CMM among older adults, with prevalence estimates rising from 11.6% to 16.9% [9]. These epidemiological findings indicate that CMM imposes a high burden among older adults and shows a continuing upward trend. Therefore, the development and validation of practical and easily accessible risk markers are of great value for identifying individuals at high risk of CMM and optimizing risk stratification.

Insulin resistance (IR) may contribute to the clustering and progression of CMDs by promoting a metabolically unfavorable environment characterized by persistent inflammation, oxidative injury, impaired endothelial function, and atherogenic changes [10]. The hyperinsulinemic-euglycemic clamp provides the most direct assessment of IR, but its labor-intensive procedure, high cost, and invasive requirements make it impractical for large population-based studies and everyday clinical use [11]. Among available IR surrogates, the triglyceride-glucose (TyG) index is a practical indicator derived from fasting triglyceride and glucose levels, capturing the combined burden of lipid and glucose metabolic dysregulation [12]. A prospective analysis of UK Biobank data suggested that TyG-related metabolic burden may contribute not only to the onset of individual CMDs but also to the accumulation of multiple cardiometabolic conditions [13]. Similar evidence was also reported in a Chinese cohort study: after full adjustment for covariates, each one-unit increase in TyG was associated with a 54% increase in the risk of CMM (HR = 1.54, 95% CI 1.29–1.84) [14]. Collectively, these findings support the utility of TyG as an indicator of the metabolic burden related to IR and as a potential aid in the early identification of individuals at elevated risk of CMM.

Previous studies have shown that traditional lipid indicators (triglycerides [TG] and total cholesterol [TC]) and lipoproteins (high-density lipoprotein cholesterol [HDL-C] and low-density lipoprotein cholesterol [LDL-C]) are closely associated with the occurrence and development of cardiometabolic abnormalities [15]. Non-traditional lipid parameters (nTLPs) calculated from routine lipid measures, including the atherogenic index of plasma (AIP), non-high-density lipoprotein cholesterol (Non-HDL-C), Castelli risk index I (CRI-I), Castelli risk index II (CRI-II), and the lipoprotein combined index (LCI), can provide integrated information on lipid-related risk and have been reported to outperform traditional lipid parameters in assessing CMD risk [16–18]. In a Chinese population-based cohort, higher levels of Non-HDL-C, AIP, and LCI were associated with incident type 2 diabetes. The corresponding risk increases per interquartile range increment were 29%, 41%, and 34%, respectively [19]. Consistent with these findings, a Chinese population-based study reported positive associations of AIP, CRI-I, and CRI-II with subsequent stroke risk [20]. These findings suggest that nTLPs may help assess the atherogenic burden and lipid metabolic abnormalities associated with CMD development.

Although TyG-related metabolic indicators and lipid-derived parameters have been examined in relation to cardiometabolic outcomes, their combined exposure patterns have not been well characterized for CMM onset in middle-aged and older Chinese adults [14, 16]. Existing studies have largely focused on individual cardiometabolic outcomes, individual metabolic indicators, or composite indices incorporating obesity-related measurements [21, 22]. For example, a large-scale cohort study from China evaluated the TyG-AIP composite index in relation to stroke and showed that, compared with individual metabolic indicators, TyG-AIP may offer greater value for early stroke risk evaluation [23]. Lai et al. further demonstrated that the combination of abdominal obesity and elevated nTLPs was associated with an increased risk of incident CMM [16]. Therefore, it remains necessary to determine whether the joint assessment of the TyG index and nTLPs can more comprehensively reflect the burden of glucose and lipid metabolic dysfunction than either indicator alone and provide additional information for risk stratification of incident CMM.

Based on this, the current investigation leverages the longitudinal datasets (2011–2020) provided by the China Health and Retirement Longitudinal Study (CHARLS), aiming to systematically evaluate the joint associations of the TyG index and nTLPs with incident CMM.

Methods

Study design and participants

The CHARLS is a nationally representative prospective cohort study, and its comprehensive research data are openly accessible via the CHARLS data portal (https://charls.pku.edu.cn/index.htm). The survey aims to systematically collect multidimensional information related to aging, including health status, socioeconomic status, lifestyle factors, family structure, healthcare service utilization, as well as physical measurements and biomarkers. The CHARLS baseline survey was conducted in 2011–2012, covering 28 provinces, 150 counties/districts, and 450 villages/resident committees or urban and rural communities in China, with a total of 17,708 respondents enrolled [5, 16]. Subsequently, follow-up surveys were conducted in 2013, 2015, 2018, and 2020. The study used a multistage stratified probability sampling strategy to select participants, ensuring broad representativeness of the sample in terms of geographic distribution and demographic characteristics.

This study initially enrolled 17,708 baseline participants. The analysis was restricted to participants aged ≥ 45 years. Exclusion criteria were implemented as outlined below: (1) missing baseline diagnostic information for diabetes, heart disease, or stroke (N = 387); (2) confirmed CMM at baseline (N = 434); (3) missing baseline TyG index or serum parameters (N = 5,765); (4) age < 45 years (N = 289); (5) absence of CMM diagnosis information across the follow-up waves of 2013, 2015, 2018, and 2020 (N = 3,674). The detailed inclusion and exclusion flow is illustrated in Fig. 1.

Fig. 1.

Flow chart of inclusion and exclusion criteria of participants

Calculations of the TyG index and nTLPs

The TyG index and nTLPs were calculated using the following formulas [16, 17]:

1

2

3

4

5

6

.

Diagnostic criteria for CMM events

Consistent with prior epidemiological studies, participants were classified as having CMM if they had a combination of at least two of the following conditions: diabetes, heart disease, or stroke [24]. Participants with CMM at baseline were excluded. Follow-up time was calculated from baseline to the first follow-up wave at which CMM was identified. For individuals without incident CMM, observation time was ended at the last survey in which their health information was available [7]. Although hypertension, dyslipidemia, and chronic kidney disease (CKD) are important cardiometabolic conditions or related disease states, they were not included as primary CMM components in this operational definition. In some population-based studies, these diseases or conditions are commonly considered upstream risk factors, intermediate metabolic states, or related comorbidities; whereas within this operational framework, diabetes, heart disease, and stroke are considered major clinically manifest CMD endpoints [4, 5, 25]. The ascertainment of diabetes status was based on a composite diagnostic framework, comprising: an FPG ≥ 7.0 mmol/L, a glycated hemoglobin (HbA1c) level ≥ 6.5%, active pharmacological management for blood glucose control, or a self-reported physician-diagnosed diabetes. Information on heart disease and stroke was primarily obtained from participants’ reports of a physician diagnosis, with relevant medication use considered as supplementary evidence when available. Stroke status was determined according to whether participants reported having received a stroke diagnosis from a physician.

Covariates

Covariates incorporated in this study encompass demographic characteristics, lifestyle factors, clinical measurements, biochemical indicators, history of chronic diseases, and medication use. Demographic and lifestyle characteristics integrated into this analysis spanned age, gender, and residence. Participants were categorized based on marital status (grouped into married and others, including unmarried, divorced, or widowed) and educational attainment (divided into four distinct groups: less than primary, primary, middle, and high school or higher). Residence was recorded as urban versus rural, while lifestyle factors, specifically smoking and drinking status, were treated as dichotomous variables. Body mass index (BMI) was determined from weight and height as follows: BMI = weight (kg)/height² (m²) [26]. To minimize the influence of extreme outliers and potential data-entry errors on the study results, and in accordance with data-cleaning strategies used in previous large-scale cohort studies, participants with abnormal baseline BMI were treated as having missing BMI values, namely those with BMI < 10 kg/m² or BMI > 60 kg/m² [26, 27]. Clinical measurements and biochemical indicators included: blood pressure parameters, specifically systolic blood pressure (SBP) and diastolic blood pressure (DBP); and biochemical parameters, including estimated glomerular filtration rate (eGFR) and high-sensitivity C-reactive protein (hs-CRP). Information on cancer diagnosis and lipid-lowering medication use was collected through the baseline questionnaire. The ascertainment of hypertension status was based on three distinct criteria: a documented clinical history provided by the participant, the ongoing administration of blood pressure-lowering agents, or measured mean blood pressure levels ≥ 140/90 mmHg [28].

Statistical analysis

Continuous variables were summarized as the mean ± standard deviation (SD) or median and interquartile range (IQR), as appropriate, whereas categorical variables were presented as frequencies and percentages. For group comparisons, Student’s t-test or the Wilcoxon rank-sum test was applied to continuous variables, and differences in categorical variables were examined using Pearson’s chi-square test, as applicable. In the absence of established clinical cutoff values, the TyG index and each nTLP were dichotomized at their respective medians into low and high categories. Participants were then classified into four joint exposure categories for each pairwise combination of the TyG index and each nTLP: low TyG and low nTLPs, low TyG and high nTLPs, high TyG and low nTLPs, and high TyG and high nTLPs. The dual-low group was used as the reference group. This study initially included six nTLPs. Given the mathematical equivalence between the AC, defined as ([TC − HDL-C]/HDL-C), and the CRI-I, defined as TC/HDL-C (where AC = CRI-I − 1), the two parameters yielded identical results in the median-based analyses. To avoid redundant reporting, only the results related to CRI-I are presented in the main text.

Cumulative CMM incidence across joint exposure categories was illustrated using Kaplan–Meier curves, and statistical separation among groups was examined with the log-rank test. CMM incidence rates were derived from incident cases and total follow-up person-time, and were standardized to 1000 person-years. The associations between metabolic/lipid indicators, including the TyG index, nTLPs, their joint exposure categories, and subsequent CMM were analyzed using adjusted Cox regression models. The magnitude of associations was quantified by HRs and 95% CIs. Schoenfeld residuals were examined to assess the proportional hazards assumption. Potential multicollinearity among study variables was evaluated using the variance inflation factor (VIF). Three nested models were constructed: Model 1 served as the crude model without any adjustment; Model 2 accounted for a comprehensive set of baseline sociodemographic and lifestyle covariates, including age, gender, BMI, smoking status, drinking status, education level, marital status, sleep duration, and residence; Model 3 was further adjusted for systolic and diastolic blood pressure, estimated glomerular filtration rate, high-sensitivity C-reactive protein, history of cancer, and lipid-lowering medication use.

To reduce the limitations of median-based dichotomization, additional analyses were performed using quartiles and standardized continuous variables. For categorical analyses, the TyG index and each nTLP were divided into quartiles, and the first quartile was used as the reference category. Dose–response trends were examined by assigning the median value of each quartile to participants in that category and fitting this variable continuously in Cox models. In parallel, associations were also assessed for each 1-SD increase in the corresponding indicator. To evaluate whether the TyG index and each nTLP exhibited statistical interaction in relation to the risk of incident CMM, analyses were performed on both multiplicative and additive scales using exposure levels defined by the median of each indicator. For the multiplicative scale, cross-product terms between the TyG index and each nTLP were added separately to the adjusted Cox regression models. Additive interaction was assessed using the relative excess risk due to interaction (RERI) and the attributable proportion due to interaction (AP). Evidence of positive additive interaction was considered when both RERI and AP were greater than zero and their corresponding 95% confidence intervals excluded zero. These measures reflect statistical interaction on the additive scale and should not be interpreted as evidence of biological interaction or biological synergy.

Possible nonlinear dose–response relationships were evaluated by fitting RCS functions, with four prespecified knot locations corresponding to the 5th, 35th, 65th, and 95th percentile values of each indicator. The median value served as the reference. When the test for nonlinearity was statistically significant, a two-piecewise Cox regression model was subsequently fitted to estimate the threshold value from the observed data. Model fit was compared with that of the corresponding linear Cox model using a likelihood-ratio test, and the threshold was estimated using the profile likelihood method.

The consistency of the observed associations was evaluated across prespecified population strata defined by age, BMI, gender, smoking status, drinking status, marital status, and residence. Interaction across subgroups was assessed by including cross-product terms in the models. In the primary analyses, missing covariate data were handled using multiple imputation by chained equations. Five imputed datasets were generated, and the resulting estimates were pooled according to Rubin’s rules. The robustness and consistency of the primary findings were further examined through several additional analyses, including: (1) comparing baseline characteristics between included and excluded participants; (2) a complete-case sensitivity analysis was conducted to evaluate the robustness of the primary findings; (3) excluding individuals who were receiving pharmacological treatment for diabetes, hypertension, dyslipidemia, heart disease, or stroke at baseline; (4) rerunning the analyses after restricting the sample to participants with no history of diabetes, heart disease, or stroke at study entry; (5) further repeating the analyses after restricting the sample to participants without hypertension, cancer, diabetes, heart disease, or stroke at baseline; (6) separately analyzing diabetes, heart disease, and stroke as individual outcomes. Model discrimination was assessed using receiver operating characteristic (ROC) curves constructed from the linear predictors derived from the Cox proportional hazards models, and differences between AUC estimates were evaluated using DeLong’s method. The incremental predictive value of adding the TyG index or an individual nTLP to the base model was subsequently quantified using the net reclassification improvement (NRI) and integrated discrimination improvement (IDI).

Statistical analyses and graphical presentation were performed using Stata version 18.0, IBM SPSS Statistics version 27.0, and GraphPad Prism version 10.0. A two-sided P value < 0.05 was considered statistically significant.

Results

Baseline characteristics of study participants

The final analytical cohort comprised 7159 participants from the 2011–2020 CHARLS follow-up data. The median baseline age was 57.0 years, and 45.0% of participants were men. During a median follow-up of 9.0 years, 758 participants developed incident CMM, corresponding to a cumulative incidence of 10.59%. Table 1 summarizes the baseline characteristics of participants according to incident CMM status during follow-up. Specifically, compared with participants who did not develop CMM, participants with incident CMM were older and had higher BMI, SBP, DBP, TG, TC, LDL-C, AIP, Non-HDL-C, CRI-I, CRI-II, LCI, TyG index, FPG, HbA1c, and hs-CRP levels, whereas HDL-C, eGFR, and sleep duration were lower (all P < 0.05). The proportions of women and rural residents were also higher among participants who developed CMM. In addition, baseline hypertension and medication use, including antihypertensive, antidiabetic, lipid-lowering, and cardiovascular medications, were more prevalent in participants who subsequently developed CMM (all P < 0.05).

Table 1.

Baseline characteristics of participants with and without CMM

Variables	All	Non-CMM	CMM	P value
	N = 7,159	N = 6,401	N = 758
Age (years)	57.00 (51.00, 64.00)	57.00 (51.00, 63.00)	59.50 (54.00, 64.00)	< 0.001
Gender (%)				< 0.001
Male	3219 (44.96)	2924 (45.68)	295 (38.92)
Female	3940 (55.04)	3477 (54.32)	463 (61.08)
Residence (%)				0.098
Urban	2294 (32.04)	2031 (31.73)	263 (34.70)
Rural	4865 (67.96)	4370 (68.27)	495 (65.30)
Education level (%)				0.646
Below primary	3372 (47.10)	3003 (46.92)	369 (48.68)
Primary school	1575 (22.00)	1405 (21.95)	170 (22.43)
Middle school	1492 (20.84)	1343 (20.98)	149 (19.66)
High school or above	720 (10.06)	650 (10.15)	70 (9.24)
Marital status (%)				0.250
Married	6495 (90.73)	5816 (90.86)	679 (89.58)
Other marital status	664 (9.28)	585 (9.14)	79 (10.42)
BMI (kg/m2)	23.22 (20.94, 25.88)	23.07 (20.81, 25.59)	25.10 (22.27, 27.59)	< 0.001
Smoking (%)	2685 (37.50)	2426 (37.90)	259 (34.17)	0.045
Drinking (%)	2727 (38.09)	2463 (38.48)	264 (34.83)	0.050
SBP, mmHg	125.00 (113.00, 139.50)	124.50 (112.50, 138.50)	131.00 (118.50, 147.50)	< 0.001
DBP, mmHg	74.00 (66.50, 82.50)	73.88 (66.33, 82.00)	77.00 (69.00, 85.50)	< 0.001
Sleep duration, hours	6.50 (5.00, 8.00)	6.57 (5.00, 8.00)	6.00 (5.00, 8.00)	0.001
TC (mmol/L)	4.93 (4.33, 5.57)	4.90 (4.32, 5.54)	5.15 (4.49, 5.81)	< 0.001
TG (mmol/L)	1.19 (0.85, 1.75)	1.12 (0.83, 1.69)	1.45 (1.02, 2.23)	< 0.001
HDL-C (mmol/L)	1.28 (1.05, 1.55)	1.30 (1.06, 1.57)	1.17 (0.95, 1.40)	< 0.001
LDL-C (mmol/L)	2.97 (2.42, 3.55)	2.95 (2.42, 3.52)	3.11 (2.44, 3.76)	< 0.001
FPG (mmol/L)	5.65 (5.23, 6.18)	5.63 (5.20, 6.16)	6.15 (5.51, 7.12)	< 0.001
HbA1c (%)	5.10 (4.90, 5.40)	5.10 (4.90, 5.40)	5.30 (5.00, 5.79)	< 0.001
eGFR (mL/min/1.73 m2)	95.63 (85.08, 102.74)	95.96 (85.60, 103.07)	92.80 (81.23, 99.63)	< 0.001
hs-CRP (mg/L)	0.97 (0.53, 2.00)	0.93 (0.52, 1.92)	1.43 (0.70, 2.86)	< 0.001
TyG	8.59 (8.22, 9.03)	8.55 (8.20, 8.98)	8.86 (8.50, 9.44)	< 0.001
AIP	− 0.04 (− 0.24, 0.19)	− 0.05 (− 0.25, 0.17)	0.11 (− 0.11, 0.32)	< 0.001
Non-HDL-C	3.60 (3.00, 4.24)	3.56 (2.98, 4.20)	3.94 (3.24, 4.65)	< 0.001
CRI-I	3.81 (3.08, 4.76)	3.74 (3.04, 4.67)	4.46 (3.54, 5.49)	< 0.001
CRI-II	2.32 (1.76, 2.97)	2.29 (1.73, 2.92)	2.70 (2.03, 3.43)	< 0.001
LCI	13.57 (7.46, 25.52)	13.04 (7.23, 23.93)	20.71 (10.73, 37.59)	< 0.001
Medical history (%)
Hypertension	1745 (24.38)	1356 (21.18)	389 (51.32)	< 0.001
Heart disease	701 (9.79)	478 (7.47)	223 (29.42)	< 0.001
Cancer	58 (0.81)	50 (0.78)	8 (1.06)	0.426
Stroke	104 (1.45)	64 (1.00)	40 (5.28)	< 0.001
Diabetes	307 (4.29)	203 (3.17)	104 (13.72)	< 0.001
Medications (%)
Antihypertensive	1262 (17.63)	965 (15.08)	297 (39.18)	< 0.001
Lipid-lowering medications	356 (4.97)	248 (3.87)	108 (14.25)	< 0.001
Cardiovascular	469 (6.55)	320 (5.00)	149 (19.66)	< 0.001
Antidiabetic	193 (2.70)	117 (1.83)	76 (10.03)	< 0.001

Data are presented as median (interquartile range) or n (%)

CMM, cardiometabolic multimorbidity; BMI, body mass index; SBP, systolic blood pressure; DBP, diastolic blood pressure; TC, total cholesterol; TG, triglyceride; HDL-C, high-density lipoprotein cholesterol; LDL-C, low-density lipoprotein cholesterol; FPG, fasting plasma glucose; HbA1c, glycated hemoglobin; eGFR, estimated glomerular filtration rate; hs-CRP, high-sensitivity C-reactive protein; TyG, triglyceride-glucose index; AIP, atherogenic index of plasma; Non-HDL-C, non-high-density lipoprotein cholesterol; CRI-I, Castelli risk index I; CRI-II, Castelli risk index II; LCI, lipoprotein combined index

Joint associations of the TyG index and nTLPs with incident CMM

Figure 2 presents Kaplan–Meier curves for the cumulative incidence of CMM across groups categorized according to the joint exposure levels of the TyG index and each nTLP. During follow-up, the curves for different joint exposure groups gradually separated, with the dual-high exposure group generally showing a higher cumulative incidence of CMM than the dual-low exposure group. The log-rank test showed that the differences in cumulative CMM incidence across the joint exposure groups were statistically significant (all P < 0.001). The joint associations were subsequently evaluated using adjusted Cox regression models. After adjustment for potential confounders, participants in each dual-high exposure category had a significantly higher risk of developing CMM than those in the dual-low reference category. The fully adjusted HRs for the dual-high exposure groups were 1.80 (95% CI 1.51–2.14) for joint high exposure to the TyG index and AIP, 2.16 (95% CI 1.74–2.67) for TyG plus Non-HDL-C, 2.10 (95% CI 1.72–2.57) for TyG plus CRI-I, 2.05 (95% CI 1.66–2.53) for TyG plus CRI-II, and 1.95 (95% CI 1.62–2.36) for TyG plus LCI (Table 2).

Fig. 2.

Kaplan-Meier curves of CMM risk by TyG index and nTLPs (A–E). Group 1: L-TyG and L-nTLPs; Group 2: L-TyG and H-nTLPs; Group 3: H-TyG and L-nTLPs; Group 4: H-TyG and H-nTLPs. TyG, triglyceride-glucose index; AIP, atherogenic index of plasma; Non-HDL-C, non-high-density lipoprotein cholesterol; CRI-I, Castelli risk index I; CRI-II, Castelli risk index II; LCI, lipoprotein combined index

Table 2.

Associations of the TyG index and nTLPs with the risk of CMM

Subgroups	Incidence ratea	Model 1		Model 2		Model 3
		HR (95% CI)	P value	HR (95% CI)	P value	HR (95% CI)	P value
AIP and TyG
Group 1	6.31	Reference	—	Reference	—	Reference	—
Group 2	6.77	1.07 (0.73, 1.57)	0.718	0.99 (0.68, 1.46)	0.977	0.98 (0.67, 1.44)	0.940
Group 3	11.55	1.83 (1.35, 2.48)	< 0.001	1.61 (1.18, 2.19)	0.003	1.47 (1.08, 2.02)	0.015
Group 4	15.28	2.42 (2.05, 2.86)	< 0.001	1.99 (1.67, 2.37)	< 0.001	1.80 (1.51, 2.14)	< 0.001
P for trend			< 0.001		< 0.001		< 0.001
Non-HDL-C and TyG
Group 1	5.15	Reference	—	Reference	—	Reference	—
Group 2	8.82	1.71 (1.32, 2.22)	< 0.001	1.62 (1.25, 2.11)	< 0.001	1.63 (1.25, 2.11)	< 0.001
Group 3	12.51	2.43 (1.91, 3.08)	< 0.001	2.21 (1.74, 2.81)	< 0.001	2.10 (1.65, 2.67)	< 0.001
Group 4	15.95	3.10 (2.53, 3.80)	< 0.001	2.43 (1.97, 3.00)	< 0.001	2.16 (1.74, 2.67)	< 0.001
P for trend			< 0.001		< 0.001		< 0.001
CRI-I and TyG
Group 1	5.35	Reference	—	Reference	—	Reference	—
Group 2	9.28	1.73 (1.33, 2.27)	< 0.001	1.64 (1.25, 2.15)	< 0.001	1.60 (1.22, 2.10)	< 0.001
Group 3	11.97	2.24 (1.74, 2.87)	< 0.001	2.04 (1.59, 2.63)	< 0.001	1.95 (1.51, 2.51)	< 0.001
Group 4	15.79	2.95 (2.44, 3.57)	< 0.001	2.37 (1.95, 2.89)	< 0.001	2.10 (1.72, 2.57)	< 0.001
P for trend			< 0.001		< 0.001		< 0.001
CRI-II and TyG
Group 1	5.45	Reference	—	Reference	—	Reference	—
Group 2	8.16	1.50 (1.15, 1.94)	0.003	1.43 (1.10, 1.87)	0.008	1.41 (1.08, 1.83)	0.011
Group 3	12.70	2.32 (1.84, 2.94)	< 0.001	2.10 (1.65, 2.67)	< 0.001	1.96 (1.54, 2.49)	< 0.001
Group 4	15.89	2.91 (2.39, 3.56)	< 0.001	2.32 (1.88, 2.85)	< 0.001	2.05 (1.66, 2.53)	< 0.001
P for trend			< 0.001		< 0.001		< 0.001
LCI and TyG
Group 1	5.67	Reference	—	Reference	—	Reference	—
Group 2	9.65	1.70 (1.27, 2.28)	< 0.001	1.56 (1.16, 2.10)	0.003	1.59 (1.18, 2.14)	0.002
Group 3	12.64	2.23 (1.71, 2.91)	< 0.001	2.13 (1.63, 2.79)	< 0.001	2.01 (1.54, 2.63)	< 0.001
Group 4	15.27	2.70 (2.26, 3.22)	< 0.001	2.17 (1.80, 2.61)	< 0.001	1.95 (1.62, 2.36)	< 0.001
P for trend			< 0.001		< 0.001		< 0.001

Model 1: No covariates adjusted

Model 2: Adjusted for age, gender, BMI, smoking status, drinking status, education level, marital status, sleep duration, and residence

Model 3: Further adjusted for SBP, DBP, eGFR, hs-CRP, cancer, and lipid-lowering medications

Group 1: L-TyG and L-nTLPs; Group 2: L-TyG and H-nTLPs; Group 3: H-TyG and L-nTLPs; Group 4: H-TyG and H-nTLPs

TyG, triglyceride-glucose index; AIP, atherogenic index of plasma; Non-HDL-C, non-high-density lipoprotein cholesterol; CRI-I, Castelli risk index I; CRI-II, Castelli risk index II; LCI, lipoprotein combined index; HR, hazard ratio; CI, confidence interval

^a Incidence rates are presented per 1,000 person-years of follow-up

To further assess the robustness of these associations independent of median-based categorization, the TyG index and nTLPs were also analyzed as standardized continuous variables and quartiles (Table S1). After full covariate adjustment, each 1-SD increment in the TyG index and in each nTLP was positively associated with incident CMM. The corresponding HRs were 1.33 (95% CI 1.25–1.43) for the TyG index, 1.26 (95% CI 1.17–1.35) for AIP, 1.16 (95% CI 1.09–1.24) for Non-HDL-C, 1.19 (95% CI 1.12–1.26) for CRI-I, 1.18 (95% CI 1.12–1.26) for CRI-II, and 1.11 (95% CI 1.06–1.16) for LCI. Quartile-based analyses showed positive trends across increasing quartiles of the TyG index and nTLPs (all P for trend < 0.001). Compared with participants in the lowest quartile of the TyG index, those in the highest quartile had a higher risk of incident CMM (HR = 2.46, 95% CI 1.93–3.15).

Fully adjusted Cox regression models incorporating mutual adjustment were used to evaluate the independent associations of the TyG index and each nTLP with the risk of developing CMM (Table S2). The results showed that, across all models, the TyG index remained positively associated with incident CMM (all P < 0.001). After adjustment for the TyG index, the associations of Non-HDL-C, CRI-I, and LCI with incident CMM were attenuated and no longer statistically significant. Notably, CRI-II remained independently associated with incident CMM (HR = 1.13, 95% CI 1.06–1.20). AIP showed an inverse estimate after mutual adjustment with the TyG index (HR = 0.81, 95% CI 0.69–0.94). Assessment of multicollinearity indicated that the VIF values were low in the fitted models (1.09–1.14), with tolerance values above 0.87 (Table S3), suggesting no substantial multicollinearity.

Assessment of nonlinear associations

Multivariable-adjusted RCS models were used to characterize the dose–response relationships between nTLPs and the risk of incident CMM, stratified by TyG index status. Among participants with a high TyG index, significant overall associations with CMM risk were observed for Non-HDL-C and CRI-I (P for overall < 0.05; Fig. 3). The association between CRI-I and CMM risk appeared approximately linear, with no statistically significant evidence of nonlinearity (P for nonlinearity = 0.086). In contrast, Non-HDL-C showed a significant nonlinear association with the risk of incident CMM (P for nonlinearity < 0.05). Therefore, a two-piecewise Cox regression model was further applied to identify the threshold of this association, which was identified at 2.99 (Table 3). When Non-HDL-C was below 2.99, higher levels were associated with a lower risk of incident CMM (HR = 0.52, 95% CI 0.31–0.88; P = 0.014); when Non-HDL-C was above 2.99, the association became positive but did not reach statistical significance (HR = 1.08, 95% CI 0.99–1.19; P = 0.080). Among participants with a low TyG index, the overall associations between nTLPs and the risk of incident CMM were not statistically significant (all P for overall > 0.05; Figure S1).

Fig. 3.

Dose–response associations between nTLPs and CMM risk among participants with a high TyG index (A–E). All models were adjusted for age, gender, BMI, smoking status, drinking status, education level, marital status, sleep duration, residence, SBP, DBP, eGFR, hs-CRP, cancer, and lipid-lowering medications. TyG, triglyceride-glucose index; AIP, atherogenic index of plasma; Non-HDL-C, non-high-density lipoprotein cholesterol; CRI-I, Castelli risk index I; CRI-II, Castelli risk index II; LCI, lipoprotein combined index; HR, hazard ratio; CI, confidence interval

Table 3.

Threshold effect analysis of Non-HDL-C in relation to CMM risk among participants with a high TyG index

Variable	Adjusted HR (95% CI)	P value
Non-HDL-C
Total	1.15 (1.06, 1.25)	0.001
Knot	2.99
Non-HDL-C < 2.99	0.52 (0.31, 0.88)	0.014
Non-HDL-C > 2.99	1.08 (0.99, 1.19)	0.080
P for log-likelihood ratio		0.016

Cox proportional hazards models were used to estimate HR and 95% CI. Models were adjusted for age, gender, BMI, smoking status, drinking status, education level, marital status, sleep duration, residence, SBP, DBP, eGFR, hs-CRP, cancer, and lipid-lowering medications. Non-HDL-C, non-high-density lipoprotein cholesterol; HR, hazard ratio; CI, confidence interval

Subgroup, sensitivity, and interaction analyses

Subgroup analyses were performed according to age, BMI, gender, smoking status, drinking status, marital status, and residence. The associations between dual-high exposure and incident CMM were generally directionally consistent across subgroups (Fig. 4). No statistically significant interactions were observed across the predefined subgroups (all P for interaction > 0.05), suggesting that the association of dual-high exposure to the TyG index and nTLPs with incident CMM did not materially vary according to demographic or clinical characteristics.

Fig. 4.

Forest plots of multivariable-adjusted subgroup analyses for dual-high exposure to the TyG index and nTLPs in relation to incident CMM (A–E). Models were adjusted for age, gender, BMI, smoking status, drinking status, education level, marital status, sleep duration, residence, SBP, DBP, eGFR, hs-CRP, cancer, and lipid-lowering medications. TyG, triglyceride-glucose index; AIP, atherogenic index of plasma; Non-HDL-C, non-high-density lipoprotein cholesterol; CRI-I, Castelli risk index I; CRI-II, Castelli risk index II; LCI, lipoprotein combined index; HR, hazard ratio; CI, confidence interval; NA, not available

Interaction effects were evaluated on the multiplicative as well as the additive scale (Table S4). Statistically significant multiplicative interactions were detected between the TyG index and Non-HDL-C, CRI-I, and LCI (all P for multiplicative interaction < 0.05). In contrast, there was no evidence of statistically significant positive additive interaction between the TyG index and any nTLP, as both the RERI and AP estimates had 95% confidence intervals that included zero.

The consistency of the primary findings was further assessed using multiple sensitivity analyses. First, the baseline characteristics of included and excluded participants were descriptively compared to assess potential selection bias arising from missing data (Table S5). The associations observed in the primary multiple-imputation analysis remained broadly consistent when the analyses were restricted to participants with complete covariate data (Table S6). Second, the results remained broadly consistent with those of the primary analysis after separately excluding participants who were receiving medications for diabetes, hypertension, dyslipidemia, heart disease, or stroke at baseline, as well as those who had diabetes, heart disease, or stroke at baseline (Tables S7 and S8). Additional analyses restricted to participants without hypertension, cancer, diabetes, heart disease, or stroke at baseline yielded similar results (Table S9). Component-specific analyses were further conducted for incident diabetes, heart disease, and stroke separately (Table 4 and Table S10). In fully adjusted analyses, joint exposure to the TyG index and nTLPs was closely associated with increased risks of developing diabetes and stroke; however, after full adjustment, the association between the dual-high exposure group and incident heart disease did not reach statistical significance.

Table 4.

Component-specific associations of the joint exposure of the TyG index and nTLPs on the risk of incident diabetes, heart disease, and stroke (N = 6,047)

Subgroups	Diabetesb		P value	Heart diseaseb		P value	Strokeb		P value
	Incidence ratea	HR (95% CI)		Incidence ratea	HR (95% CI)		Incidence ratea	HR (95% CI)
AIP and TyG
Group 1	8.80	Reference	–	18.17	Reference	–	6.71	Reference	–
Group 2	8.08	0.78 (0.53, 1.14)	0.192	22.98	1.27 (1.01, 1.60)	0.043	9.34	1.40 (0.98, 2.00)	0.066
Group 3	16.80	1.73 (1.30, 2.31)	< 0.001	17.34	0.90 (0.69, 1.17)	0.424	7.86	1.10 (0.74, 1.62)	0.649
Group 4	19.70	1.82 (1.55, 2.14)	< 0.001	21.17	1.03 (0.91, 1.18)	0.613	10.65	1.45 (1.19, 1.77)	< 0.001
P for trend			< 0.001			0.838			0.001
Non-HDL-C and TyG
Group 1	7.99	Reference	–	18.19	Reference	–	5.97	Reference	–
Group 2	10.16	1.23 (0.96, 1.57)	0.101	19.94	1.04 (0.88, 1.24)	0.621	9.20	1.57 (1.20, 2.06)	0.001
Group 3	17.81	2.00 (1.61, 2.48)	< 0.001	18.42	0.95 (0.79, 1.15)	0.611	8.60	1.39 (1.04, 1.85)	0.024
Group 4	20.12	2.03 (1.68, 2.45)	< 0.001	21.86	1.02 (0.88, 1.18)	0.828	11.17	1.69 (1.34, 2.14)	< 0.001
P for trend			< 0.001			0.984			< 0.001
CRI-I and TyG
Group 1	7.59	Reference	–	18.23	Reference	–	6.23	Reference	–
Group 2	11.96	1.43 (1.16, 1.77)	0.001	20.35	1.09 (0.90, 1.31)	0.366	9.37	1.46 (1.10, 1.93)	0.009
Group 3	15.32	2.08 (1.72, 2.52)	< 0.001	21.14	1.09 (0.90, 1.31)	0.384	9.91	1.53 (1.15, 2.04)	0.003
Group 4	20.75	2.24 (1.93, 2.61)	< 0.001	20.52	0.97 (0.84, 1.06)	0.723	10.42	1.49 (1.19, 1.86)	0.001
P for trend			< 0.001			0.721			< 0.001
CRI-II and TyG
Group 1	8.00	Reference	–	18.04	Reference	–	5.57	Reference	–
Group 2	10.10	1.12 (0.89, 1.46)	0.299	20.21	1.11 (0.93, 1.33)	0.252	9.92	1.81 (1.38, 2.37)	< 0.001
Group 3	17.07	1.85 (1.49, 2.31)	< 0.001	19.96	1.04 (0.87, 1.24)	0.671	9.08	1.58 (1.19, 2.10)	0.001
Group 4	20.53	2.02 (1.67, 2.45)	< 0.001	21.07	1.01 (0.86, 1.17)	0.941	10.93	1.79 (1.40, 2.28)	< 0.001
P for trend			< 0.001			0.933			< 0.001
LCI and TyG
Group 1	8.41	Reference	–	17.66	Reference	–	6.47	Reference	–
Group 2	10.15	1.08 (0.80, 1.46)	0.627	24.17	1.35 (1.11, 1.65)	0.003	9.78	1.51 (1.11, 2.08)	0.010
Group 3	16.24	1.75 (1.35, 2.27)	< 0.001	27.81	1.01 (0.80, 1.27)	0.950	8.61	1.29 (0.92, 1.81)	0.136
Group 4	19.99	1.93 (1.63, 2.28)	< 0.001	21.30	1.06 (0.92, 1.21)	0.442	10.65	1.49 (1.21, 1.84)	< 0.001
P for trend			< 0.001			0.714			< 0.001

Group 1: L-TyG and L-nTLPs; Group 2: L-TyG and H-nTLPs; Group 3: H-TyG and L-nTLPs; Group 4: H-TyG and H-nTLPs

TyG, triglyceride-glucose index; AIP, atherogenic index of plasma; Non-HDL-C, non-high-density lipoprotein cholesterol; CRI-I, Castelli risk index I; CRI-II, Castelli risk index II; LCI, lipoprotein combined index; HR, hazard ratio; CI, confidence interval

^a Incidence rates are presented per 1,000 person-years of follow-up

^b Multivariable-adjusted for age, gender, BMI, smoking status, drinking status, education level, marital status, sleep duration, residence, SBP, DBP, eGFR, hs-CRP, cancer, and lipid-lowering medications

Discriminative performance and incremental predictive value

ROC curve analysis based on Cox model-derived linear predictors was conducted to evaluate model discrimination for incident CMM. The discriminative performance and incremental predictive value of different models are shown in Table 5; Fig. 5. The AUC of the basic model was 0.703 (95% CI 0.688–0.717). Adding the TyG index or individual nTLPs produced statistically significant but modest improvements in model discrimination, with AUC values ranging from 0.709 to 0.725 (all P < 0.001 by DeLong’s test). Among the individual indicators, adding the TyG index to the basic model resulted in the greatest gain in predictive performance, with the AUC increasing to 0.725 and the NRI reaching 48.03% (95% CI 43.59%–53.47%) and an IDI of 1.17% (95% CI 0.77%–1.58%). Compared with the model containing the basic predictors plus the TyG index, the addition of CRI-II yielded the highest AUC among the combined models (AUC = 0.729, 95% CI 0.714–0.742), with statistically significant improvements in NRI (37.00%, 95% CI 30.76%–43.33%) and IDI (0.18%, 95% CI 0.01%–0.36%).

Table 5.

Predictive performance and incremental predictive value of the TyG index and nTLPs for incident CMM

Variables	AUC		NRI (%)		IDI (%)
	Estimate (95% CI)	P value	Estimate (95% CI)	P value	Estimate (95% CI)	P value
Basic model	0.703 (0.688, 0.717)	Reference	Reference	–	Reference	–
Basic model + TyG	0.725 (0.711, 0.739)	< 0.001	48.03 (43.59, 53.47)	< 0.001	1.17 (0.77, 1.58)	< 0.001
Basic model + AIP	0.717 (0.703, 0.731)	< 0.001	42.01 (35.83, 48.20)	< 0.001	0.53 (0.27, 0.79)	< 0.001
Basic model + TyG + AIP	0.725 (0.710, 0.740)	< 0.001	24.37 (18.97, 29.77)	< 0.001	0.31 (0.08, 0.53)	0.007
Basic model + Non-HDL-C	0.709 (0.694, 0.723)	< 0.001	36.06 (30.44, 41.68)	< 0.001	0.25 (0.05, 0.45)	0.014
Basic model + TyG + Non-HDL-C	0.725 (0.711, 0.739)	< 0.001	20.77 (15.42, 26.13)	< 0.001	0.00 (-0.03, 0.03)	0.993
Basic model + CRI-I	0.715 (0.701, 0.730)	< 0.001	42.07 (36.33, 47.81)	< 0.001	0.39 (0.16, 0.62)	0.001
Basic model + TyG + CRI-I	0.724 (0.710, 0.738)	< 0.001	27.94 (19.93, 35.96)	< 0.001	0.03 (-0.04, 0.10)	0.466
Basic model + CRI-II	0.713 (0.698, 0.728)	< 0.001	43.01 (37.06, 48.97)	< 0.001	0.33 (0.08, 0.59)	0.011
Basic model + TyG + CRI-II	0.729 (0.714, 0.742)	< 0.001	37.00 (30.76, 43.33)	< 0.001	0.18 (0.01, 0.36)	0.039
Basic model + LCI	0.711 (0.696, 0.726)	< 0.001	36.27 (30.39, 42.15)	< 0.001	0.18 (-0.02, 0.38)	0.080
Basic model + TyG + LCI	0.725 (0.711, 0.739)	< 0.001	35.57 (22.47, 48.68)	< 0.001	0.01 (-0.03, 0.05)	0.624

TyG, triglyceride-glucose index; AIP, atherogenic index of plasma; Non-HDL-C, non-high-density lipoprotein cholesterol; CRI-I, Castelli risk index I; CRI-II, Castelli risk index II; LCI, lipoprotein combined index; AUC, area under the receiver operating characteristic curve; NRI, net reclassification improvement; IDI, integrated discrimination improvement; CI, confidence interval

The basic model was adjusted for age, gender, BMI, smoking status, drinking status, education level, marital status, sleep duration, residence, SBP, DBP, eGFR, hs-CRP, cancer, and lipid-lowering medications. The incremental indices (NRI and IDI) were calculated based on two levels: first, by comparing “Basic model + single indicator” versus the “Basic model” to evaluate the incremental contribution of each individual marker; second, by comparing “Basic model + TyG + individual nTLP” versus “Basic model + TyG” to assess the additional discriminatory gain provided by each nTLP beyond the TyG index. All P values were two-sided, with P < 0.05 considered statistically significant

For AUC comparisons, models containing a single indicator were compared with the basic model, whereas models containing the TyG index plus an individual nTLP were compared with the basic model plus the TyG index

Fig. 5.

ROC curves for CMM prediction (A–E). The basic model was adjusted for age, gender, BMI, smoking status, drinking status, education level, marital status, sleep duration, residence, SBP, DBP, eGFR, hs-CRP, cancer, and lipid-lowering medications. TyG, triglyceride-glucose index; AIP, atherogenic index of plasma; Non-HDL-C, non-high-density lipoprotein cholesterol; CRI-I, Castelli risk index I; CRI-II, Castelli risk index II; LCI, lipoprotein combined index; AUC, area under the receiver operating characteristic curve

Table 6.

Predictive performance and incremental utility of the TyG index and nTLPs in the joint assessment of CMM risk

Variables	AUC		NRI (%)		IDI (%)
	Estimate (95% CI)	P value	Estimate (95% CI)	P value	Estimate (95% CI)	P value
Basic model	0.703 (0.688, 0.717)	Reference	Reference	–	Reference	–
Basic model + TyG	0.725 (0.711, 0.739)	< 0.001	48.03 (43.59, 53.47)	< 0.001	1.17 (0.77, 1.58)	< 0.001
Basic model + AIP	0.717 (0.703, 0.731)	< 0.001	42.01 (35.83, 48.20)	< 0.001	0.53 (0.27, 0.79)	< 0.001
Basic model + TyG + AIP	0.725 (0.710, 0.740)	< 0.001	24.37 (18.97, 29.77)	< 0.001	0.31 (0.08, 0.53)	0.007
Basic model + Non-HDL-C	0.709 (0.694, 0.723)	< 0.001	36.06 (30.44, 41.68)	< 0.001	0.25 (0.05, 0.45)	0.014
Basic model + TyG + Non-HDL-C	0.725 (0.711, 0.739)	< 0.001	20.77 (15.42, 26.13)	< 0.001	0.00 (-0.03, 0.03)	0.993
Basic model + CRI-I	0.715 (0.701, 0.730)	< 0.001	42.07 (36.33, 47.81)	< 0.001	0.39 (0.16, 0.62)	0.001
Basic model + TyG + CRI-I	0.724 (0.710, 0.738)	< 0.001	27.94 (19.93, 35.96)	< 0.001	0.03 (-0.04, 0.10)	0.466
Basic model + CRI-II	0.713 (0.698, 0.728)	< 0.001	43.01 (37.06, 48.97)	< 0.001	0.33 (0.08, 0.59)	0.011
Basic model + TyG + CRI-II	0.729 (0.714, 0.742)	< 0.001	37.00 (30.76, 43.33)	< 0.001	0.18 (0.01, 0.36)	0.039
Basic model + LCI	0.711 (0.696, 0.726)	< 0.001	36.27 (30.39, 42.15)	< 0.001	0.18 (-0.02, 0.38)	0.080
Basic model + TyG + LCI	0.725 (0.711, 0.739)	< 0.001	35.57 (22.47, 48.68)	< 0.001	0.01 (-0.03, 0.05)	0.624

TyG, triglyceride-glucose index; AIP, atherogenic index of plasma; Non-HDL-C, non-high-density lipoprotein cholesterol; CRI-I, Castelli risk index I; CRI-II, Castelli risk index II; LCI, lipoprotein combined index; AUC, area under the receiver operating characteristic curve; NRI, net reclassification improvement; IDI, integrated discrimination improvement; CI, confidence interval

The basic model was adjusted for age, gender, BMI, smoking status, drinking status, education level, marital status, sleep duration, residence, SBP, DBP, eGFR, hs-CRP, cancer, and lipid-lowering medications. The incremental indices (NRI and IDI) were calculated based on two levels: first, by comparing “Basic model + single indicator” versus the “Basic model” to evaluate the incremental contribution of each individual marker; second, by comparing “Basic model + TyG + individual nTLP” versus “Basic model + TyG” to assess the additional discriminatory gain provided by each nTLP beyond the TyG index. All P values were two-sided, with P < 0.05 considered statistically significant

For AUC comparisons, models containing a single indicator were compared with the basic model, whereas models containing the TyG index plus an individual nTLP were compared with the basic model plus the TyG index

Discussion

Using data from a large-scale prospective cohort, this study systematically assessed the association between joint exposure to the TyG index and nTLPs and incident CMM among middle-aged and older Chinese adults. The results showed that, compared with participants with dual-low exposure or single-high exposure, participants with dual-high exposure to the TyG index and nTLPs had a higher risk of developing CMM. This association remained generally consistent in subgroup and sensitivity analyses. However, when diabetes, heart disease, and stroke were evaluated as outcome events, the association was more evident for the risks of diabetes and stroke, whereas no statistically significant association with incident heart disease was observed after full covariate adjustment. In prediction analyses, the TyG index exhibited the best discriminative performance among the evaluated single indicators. When CRI-II was incorporated into the TyG-based model, statistically significant but modest improvements were observed in model discrimination and risk reclassification. Therefore, the joint assessment of the TyG index and nTLPs may provide supplementary information for identifying individuals with a higher cardiometabolic burden, but its incremental predictive value and clinical applicability still require further validation.

Against the backdrop of accelerating global population aging, the burden of CMDs continues to rise and has become a major challenge for public health systems worldwide [1, 3, 4]. CMM refers to the concurrent presence of at least two CMDs and represents the cumulative and interrelated effects of metabolic, vascular, and cardiovascular dysfunction over the course of aging [29, 30]. Zhang et al. reported that CMM, as a common comorbidity pattern, is associated with a more severe disease burden and a higher risk of mortality than any single CMD [31]. In this study, participants were considered to have CMM when any two or more of the following conditions were present: diabetes, heart disease, and stroke. This operational definition was aligned with the disease data captured in CHARLS and the criteria adopted in prior studies using the same cohort [6, 21, 32]. Therefore, the findings should be interpreted within this operational definition and should not be directly generalized to broader definitions of cardiometabolic multimorbidity that include hypertension, chronic kidney disease, or dyslipidemia [33]. IR is considered a key metabolic abnormality involved in the development of diabetes, stroke, and cardiovascular diseases [10]. By combining fasting triglyceride and fasting glucose levels, the TyG index serves as a practical proxy for IR and is commonly applied in population-based studies of metabolic risk [29, 34]. A prospective cohort study of 6,986 participants revealed that TyG index levels were significantly and positively associated with the risk of CMDs; specifically, higher TyG index levels were associated with a significant increase in the risk of heart disease and stroke by 15% and 34%, respectively [22]. Zhang et al. reported a positive linear association between the TyG index and CMM risk in RCS analyses, with each 1-unit increment in the TyG index associated with a 1.54-fold higher risk of CMM [14]. Consistent with previous evidence, the present study further indicates that higher TyG index levels are associated with an increased risk of CMM, suggesting that IR-related metabolic abnormalities may be involved in the formation and accumulation of CMM risk.

nTLPs are novel composite biomarkers derived from conventional lipid parameters and can more comprehensively reflect the state of systemic metabolic dysregulation in the body [16, 17]. A large-scale study based on the NAGALA cohort suggested that nTLPs may aid in identifying middle-aged and older individuals at risk of type 2 diabetes [18]. Another large cohort study based on a Chinese population showed that elevated nTLPs, including Non-HDL-C, CRI-I, and CRI-II, are independent predictors of ischemic stroke in patients with hypertension [35]. However, the role of nTLPs in combination with the TyG index remains unclear. Given that CMM is a complex condition involving both glucose metabolism and lipid metabolism abnormalities, the combined evaluation of IR-related and lipid-related indicators may help characterize cardiometabolic risk more comprehensively than single markers alone. Previous cohort studies have evaluated the associations of TyG-related obesity indices, such as TyG-BMI, TyG-WC, and TyG-WHtR, with CMM risk, suggesting that combining IR-related information with anthropometric indicators may improve cardiometabolic risk stratification. In contrast, the present study focused on the joint assessment of the TyG index and multiple lipid-derived parameters, thereby emphasizing the combined contribution of IR-related and lipid-related metabolic information [5, 13]. In addition, previous studies on the TyG index and lipid-related parameters have mainly focused on AIP and cardiovascular or cerebrovascular outcomes, rather than multiple nTLPs and incident CMM [36, 37]. For example, Zeng et al. evaluated the effect of joint exposure to the TyG index and AIP on cardiovascular disease, and the results showed that, compared with the low TyG and low AIP group, the high TyG and high AIP group had a 27% increased risk of cardiovascular disease (HR = 1.27, 95% CI 1.10–1.43) [36]. Therefore, the present study further evaluated whether multiple lipid-derived parameters could provide additional information for identifying the risk of incident CMM beyond the metabolic risk reflected by the TyG index, thereby extending the existing evidence. The results showed that participants with dual-high exposure to the TyG index and nTLPs generally had the highest CMM risk, whereas those with elevated nTLPs alone had a relatively lower risk. This pattern suggests that nTLPs may identify lipid-related metabolic abnormalities, but their risk implication may be more pronounced when accompanied by elevated TyG index levels. However, the magnitude of association varied across different TyG and nTLP combinations, and the risk estimates for some dual-high groups did not clearly exceed those observed for high TyG exposure alone. These findings indicate that the joint exposure categories should be interpreted primarily as a risk stratification approach rather than direct evidence of biological synergy.

To further evaluate the robustness of the findings derived from median-based dichotomization, this study categorized the TyG index and nTLPs into quartiles for analysis. Additional quartile-based and continuous-variable analyses showed broadly consistent positive associations, supporting the robustness of the main findings [38]. Nevertheless, dichotomization may still lead to information loss and should not be regarded as a clinically optimal threshold [39]. The mutually adjusted analyses further suggested that several lipid-derived indicators may share overlapping metabolic information. In particular, when multiple correlated metabolic and lipid indicators were simultaneously included in the model, the associations of some nTLPs were attenuated, and AIP showed an inverse association with CMM risk. This finding may be related to statistical suppression or competitive adjustment caused by the inclusion of highly related metabolic indicators in the same model. Although VIF diagnostics did not indicate severe multicollinearity, the shared components and biological overlap among these indicators should be considered when interpreting the results [40]. Given the shared components and overlapping metabolic information among the above indicators, the interpretation of the joint-exposure findings should not rely solely on the risk differences across the four joint-exposure categories, but should also incorporate formal interaction analyses on both the multiplicative and additive scales, while fully considering the scale dependence of statistical interaction. On the additive scale, the interaction indices for the TyG index combined with Non-HDL-C, CRI-I, CRI-II, and LCI did not support a statistically significant positive additive interaction. For the TyG and AIP combination, the positive estimates of RERI and AP should also be interpreted cautiously because they did not reach statistical significance. Statistically significant multiplicative interactions were identified between the TyG index and several nTLPs. However, additive interaction analyses based on RERI and AP did not provide evidence of statistically significant positive additive interaction. Therefore, these findings should be interpreted as evidence of scale-dependent statistical interaction rather than evidence of biological interaction or synergistic biological effects. Such inconsistency between additive and multiplicative interaction is not uncommon in epidemiological studies. Nian et al. found evidence supporting a multiplicative-scale interaction between TyG and AIP (P = 0.021), whereas the additive interaction did not reach statistical significance (RERI = 0.280, 95% CI -0.220 to 0.779) [37]. Notably, because additive interaction is generally more suitable for assessing excess risk at the population level, the absence of significant additive interaction suggests that the findings of this study are insufficient to support the existence of biological synergy. Instead, the main value of the joint exposure analysis lies in describing whether individuals with simultaneous IR-related and lipid-related metabolic abnormalities have a higher cardiometabolic risk.

From a biological perspective, the observed associations are plausible. IR may be associated with impaired glucose utilization, compensatory hyperinsulinemia, endothelial dysfunction, oxidative stress, and low-grade inflammation, all of which are potentially involved in CMD development [10, 13, 41]. Lipid-derived parameters, particularly Non-HDL-C, CRI-II, and AIP, may reflect atherogenic lipoprotein burden and triglyceride-rich lipoprotein metabolism, which are closely related to vascular injury and atherosclerotic processes [42, 43]. Under IR-related metabolic conditions, increased hepatic very-low-density lipoprotein production, altered triglyceride metabolism, and reduced lipid clearance may contribute to dyslipidemia [30, 44]. Therefore, elevated TyG index and nTLPs may jointly reflect a shared metabolic milieu characterized by glucolipid dysregulation. Although these mechanisms have been supported by previous studies, they were not directly examined in the present observational cohort. Future studies incorporating repeated biomarker measurements, inflammatory markers, vascular function indicators, and mechanistic endpoints are needed to further clarify the biological pathways linking IR, lipid abnormalities, and CMM progression.

To explore the stability of the association between the joint exposure to the TyG index and nTLPs and the risk of CMM, this study conducted systematic subgroup and interaction analyses based on key demographic characteristics and lifestyle factors. These included age (45–59, 60–79, and ≥ 80 years), residence (urban and rural), BMI (< 24 and ≥ 24 kg/m²), gender (male and female), drinking status, and smoking status [5, 13, 22, 45]. No significant interactions were observed across these stratification variables, suggesting that the observed association was generally stable among middle-aged and older Chinese adults. Although the strength of association appeared greater in some subgroups, including participants younger than 60 years, males, rural residents, and those with a history of smoking, these differences were not supported by statistically significant subgroup interactions. Previous studies have also described comparable subgroup findings, suggesting that TyG-related metabolic indicators may show stronger relevance to CMM risk among men [46]. However, because subgroup analyses may be affected by differences in sample size, event number, and residual confounding, these findings should be interpreted as exploratory rather than confirmatory. In addition, to ensure the robustness of the study findings, this study performed a series of sensitivity analyses. By comparing the baseline characteristics between the included and excluded participants, and verifying the consistency between the results of the complete-case analysis and multiple imputation, these analyses reduced concerns regarding potential selection bias due to missing data, although such bias cannot be completely excluded [16]. The relationship between joint exposure and incident CMM persisted across sensitivity analyses after participants with pre-existing chronic diseases or relevant medication use were excluded, suggesting that baseline disease burden or treatment status did not fully account for the observed findings. The component-specific analyses provided an important clarification for interpreting the overall CMM findings. In analyses of individual CMD components, joint exposure to the TyG index and nTLPs was strongly associated with diabetes and stroke but not with incident heart disease in fully adjusted models. This suggests that the overall association with CMM may be mainly driven by diabetes and stroke. This pattern is biologically plausible because the TyG index directly reflects glucose and triglyceride metabolism, and nTLPs are closely related to lipid-related vascular risk [10, 37, 42, 47]. In contrast, heart disease is a heterogeneous outcome that may include chronic ischemic heart disease, heart failure, and other cardiac conditions influenced by multiple metabolic and non-metabolic factors, such as blood pressure, cardiac remodeling, socioeconomic status, psychological stress, genetic susceptibility, and family history [48, 49]. For instance, a large-scale cohort study in Sweden found that low educational attainment, unemployment, poor economic conditions, signs of depression, and psychological stress showed significant links with subsequent progression of angina and heart disease [50]. Although a graded pattern may be observed in crude incidence, the lack of statistical significance after full adjustment indicates that the association between joint exposure to the TyG index and nTLPs and incident heart disease warrants further investigation. These observations suggest that CMM should be viewed as a composite endpoint composed of heterogeneous disease components, rather than as a condition in which all components follow the same risk pattern.

This study further investigated the dose–response relationships between different nTLPs and incident CMM across strata of the TyG index. In the high-TyG group, the overall association between CRI-I and incident CMM was statistically significant, and no statistically significant evidence of nonlinearity was observed. In contrast, Non-HDL-C showed a statistically significant nonlinear association with incident CMM. A two-piecewise Cox regression model was further applied to assess the threshold effect, identifying a data-driven inflection point of 2.99. When Non-HDL-C was below this inflection point, each 1-unit increase in Non-HDL-C was associated with a lower risk of incident CMM (HR = 0.52, 95% CI 0.31–0.88). When Non-HDL-C was above this inflection point, the direction of the association became positive, and each 1-unit increase in Non-HDL-C was associated with a possible 8% increase in CMM risk (HR = 1.08, 95% CI 0.99–1.19), although this association did not reach statistical significance. Notably, the inverse association observed below the inflection point should not be interpreted as indicating a protective effect of higher Non-HDL-C, and the data-driven inflection point of 2.99 should not be directly regarded as a clinically meaningful risk threshold. In the high-TyG group, the overall associations between the remaining nTLPs and incident CMM did not reach statistical significance; in the low-TyG group, the overall associations between all nTLPs and incident CMM were also not statistically significant. The nonlinear association patterns observed in this study differed from those reported in previous studies. A study conducted in a Chinese population found significant nonlinear relationships between several nTLPs and the risk of prediabetes. For example, when AIP was ≤ 0.524, increasing AIP levels were associated with an increased risk of prediabetes; whereas when CRI-II was > 2.059, the risk of prediabetes gradually decreased with increasing CRI-II levels [17]. Zou et al. also reported a nonlinear association between AIP and stroke risk: when AIP was < 0.357, increasing AIP levels were significantly associated with an increased risk of stroke, whereas when AIP exceeded 0.357, the strength of the association between AIP and stroke risk was markedly attenuated [47]. The differences among studies may be related to variations in population characteristics, baseline metabolic status, the distributions of lipid indicators, outcome definitions, covariate-adjustment strategies, and nonlinear model specifications. Therefore, the nonlinear association patterns and data-driven inflection point observed in this study require further validation in other independent cohorts.

Existing evidence suggests that when the TyG index is incorporated into conventional risk models, it may help improve CMM risk prediction, as reflected by changes in AUC, NRI, or IDI [13, 14]. A cohort study including 374,274 participants from the UK Biobank showed that TyG-WHtR and TyG-WC had significantly higher NRI and IDI, suggesting relatively better incremental predictive value for CMM risk [13]. Meanwhile, lipid-derived parameters may provide additional predictive information for cardiometabolic-related outcomes [51–53]. For example, a large cohort study from Sweden reported that CRI-I and Non-HDL-C were important predictors for early risk assessment of ischemic heart disease in middle-aged women [51]. In the present study, adding the TyG index to the base model produced the largest improvement in model discrimination among the individual indicators evaluated, with the AUC rising from 0.703 to 0.725 and significant improvements observed in both NRI and IDI. However, evidence regarding the joint assessment of the TyG index and nTLPs for CMM risk prediction remains limited. A UK cohort study suggested that integrating IR and dyslipidemia-related indicators into a composite metabolic score may improve risk prediction compared with single indicators [30]. Although the TyG and CRI-II model yielded the highest AUC among the tested combinations, its absolute improvement over the TyG-only model was small, suggesting limited incremental predictive value. Therefore, jointly assessing the TyG index and CRI-II may help refine CMM risk classification. However, given the modest improvement in predictive performance, its clinical applicability remains uncertain and requires further validation in independent cohorts.

Several strengths of this study should be noted. First, the analysis used data from CHARLS, a nationwide longitudinal cohort spanning 28 provinces in China, with follow-up extending up to 9 years. This design provided sufficient longitudinal information and enhanced the reliability of the observed associations. Second, the combined assessment of the TyG index, which provides an indirect estimate of IR, and nTLPs, which capture lipid-related metabolic disturbances, enabled a broader evaluation of how metabolic dysfunction relates to the subsequent development of CMM than approaches relying on either indicator alone. Third, multiple analytical approaches were applied, including Cox proportional hazards models, restricted cubic spline analysis, quartile-based and continuous-variable analyses, interaction analyses, subgroup analyses, sensitivity analyses, and prediction performance assessment. These analyses allowed us to examine the consistency, nonlinearity, and incremental value of the observed associations from different perspectives. Fourth, component-specific analyses helped clarify that the observed association with CMM was mainly related to diabetes and stroke rather than heart disease, thereby providing a more nuanced interpretation of the composite outcome.

The findings of this study should be interpreted in light of several limitations. First, because this analysis included only middle-aged and older adults from China, whether the findings can be extended to other ethnic populations, younger individuals, or groups with different lifestyle profiles remains to be confirmed. Second, CMM was defined as the coexistence of at least two of diabetes, heart disease, and stroke. This operational definition did not include hypertension, chronic kidney disease, or dyslipidemia and therefore may not capture the full spectrum of cardiometabolic multimorbidity. Future studies should examine whether the present findings remain robust under broader CMM definitions. Third, because there are no universally accepted clinical cutoffs for the TyG index and nTLPs, median-based stratification was used to define exposure categories. Although quartile-based and continuous-variable analyses were conducted, categorization may still result in information loss and may not correspond to clinically meaningful thresholds. Fourth, the TyG index and nTLPs were assessed at a single baseline time point, which limited the ability to evaluate longitudinal trajectories and prevented inference regarding temporal or causal relationships between IR and lipid abnormalities. Fifth, because the TyG index incorporates fasting glucose and diabetes was one component of the CMM outcome, partial component overlap may have inflated the association between TyG-related exposure and incident CMM. This issue should be considered particularly when interpreting diabetes-driven results. Sixth, several nTLPs are mathematically derived from overlapping lipid components, and some indices share similar biological information. Although no severe multicollinearity was detected, shared components among lipid-derived parameters may still influence model interpretation. Seventh, diabetes, heart disease, and stroke status was determined largely from participant-reported information, leaving the possibility of recall bias and outcome classification errors. Eighth, although the models included multiple covariates, unmeasured or inadequately captured factors, including diet, physical activity, socioeconomic status, medication adherence, inflammatory status, and genetic background, may still have contributed to residual confounding. Ninth, although the addition of the TyG index and nTLPs resulted in statistically significant improvements in discrimination and reclassification, the absolute improvement in predictive performance was modest, and the models were not externally validated in an independent cohort. Finally, due to the observational cohort design, this study can only support associations rather than establish causal relationships between the TyG index, nTLPs, and CMM risk.

Conclusion

Based on data from a large-scale prospective cohort study in China with a median follow-up of 9 years, this study systematically evaluated the association between joint exposure to the TyG index and nTLPs and the risk of incident CMM. The associations were more evident for incident diabetes and stroke than for incident heart disease. Among the individual indicators evaluated, the TyG index showed the best risk discrimination, while the further addition of CRI-II yielded a statistically significant but modest incremental improvement. Combined assessment of the TyG index and nTLPs may provide supplementary information for the early identification and risk stratification of middle-aged and older adults at high risk of CMM. Given the observational nature of this study, these findings and their clinical applicability require further validation in independent cohorts.

Supplementary Information

Below is the link to the electronic supplementary material.

Abbreviations

TyG

Triglyceride-glucose index

nTLPs

Non-traditional lipid parameters

CMM

Cardiometabolic multimorbidity

CMDs

Cardiometabolic diseases

CHARLS

China Health and Retirement Longitudinal Study

TC

Total cholesterol

TG

Triglyceride

HDL-C

High-density lipoprotein cholesterol

LDL-C

Low-density lipoprotein cholesterol

HbA1c

Glycated hemoglobin

FPG

Fasting plasma glucose

SBP

Systolic blood pressure

DBP

Diastolic blood pressure

hs-CRP

High-sensitivity C-reactive protein

eGFR

Estimated glomerular filtration rate

IR

Insulin resistance

BMI

Body mass index

HR

Hazard ratio

CI

Confidence interval

RCS

Restricted cubic splines

ROC

Receiver operating characteristic curve

AUC

Area under the ROC curve

NRI

Net reclassification improvement

IDI

Integrated discrimination improvement

RERI

Relative excess risk due to interaction

AP

Attributable proportion due to interaction

VIF

Variance inflation factor

Author contributions

GYW and PZG designed the study. GYW, PZG, and LQQ analyzed the data. TSK assisted with data collection and interpretation. LQQ supervised the study. GYW wrote the manuscript with input from all authors. All authors have read and approved the final manuscript.

Funding

The authors declared that financial support was received for this work and/or its publication. This work was supported by National Natural Science Foundation of China (NSFC, Grant No. 31300978).

Data availability

This study is based on publicly available data from the CHARLS, which can be accessed through the official website (https://charls.pku.edu.cn/index.htm) after completing registration and application.

Declarations

Ethics approval and consent to participate

The CHARLS study was approved by the Ethics Committee of Peking University (approval number for the household survey: IRB00001052-11015; approval number for blood collection: IRB00001052-11014), and informed consent was obtained from all participants. Access to the data was obtained through registration.

Competing interests

The authors declare no competing interests.