Associations of longitudinal trajectories of triglyceride-glucose index combined with classical and novel obesity indices and cardiovascular disease: evidence from a nationwide prospective cohort study in China

Abstract

Background

The triglyceride glucose (TyG) index and various obesity indices have been proven to be cost-effective indicators of cardiovascular disease (CVD) risk. This study aims to systematically investigate and compare the associations between longitudinal trajectories of TyG index combined with classical and novel obesity indices (TyG-BMI, TyG-WC, TyG-WHtR, TyG-WWI, TyG-ABSI, TyG-BRI, TyG-CVAI) and CVD.

Methods

The study sample comprised 3505 non-CVD participants from the CHARLS national cohort. Longitudinal data from Waves 1 and 3 of the national surveys were used to quantify cumulative exposure and trajectories of TyG and its obesity derivatives. A multi-model analytical framework (including logistic regression, spline regression, and weighted quantile sum regression models) was constructed to systematically examine the strength of associations between trajectories of TyG and its obesity derivatives and CVD, and the contribution of each component.

Results

During 8-year median follow-up, 411 CVD cases occurred. The study demonstrated that compared to static baseline values, longitudinal assessment of cumulative exposure and trajectory of TyG and its obesity derivatives enhanced predictive capacity for CVD. Notably, combinations of TyG with classical obesity index WC and novel obesity index CVAI (TyG-WC and TyG-CVAI) exhibited superior performance for CVD risk assessment. Compared to participants with well-controlled trajectories and low exposure levels, those with poorly controlled or highest cumulative exposure to TyG index, TyG-BMI, TyG-WC, TyG-WHtR, TyG-WWI, TyG-ABSI, TyG-BRI, and TyG-CVAI had odds ratios of 1.61/1.40, 2.13/1.70, 2.00/1.78, 1.77/1.59, 1.31/1.36, 1.37/1.30, 1.76/1.56, and 2.00/1.72, respectively. Finally, weighted quantile sum regression results indicated that cumulative exposure to obesity and triglycerides contributed most to CVD risk among all metabolic indices, suggesting that simultaneous regulation of triglycerides and obesity may be critical for reducing CVD risk.

Conclusion

In this cohort study, the longitudinal trajectories of TyG and its obesity derivatives were closely associated with CVD. Comparatively, the combinations of TyG with classical obesity index WC and novel obesity index CVAI (TyG-WC and TyG-CVAI) exhibited superior performance for CVD risk assessment, with this risk primarily driven by obesity and triglycerides.

Graphical abstract

Supplementary Information

The online version contains supplementary material available at 10.1186/s12933-025-02972-6.

Research insights

What is currently known about this topic?

• The triglyceride-glucose (TyG) index, a surrogate marker for insulin resistance, along with various classic and novel obesity indices, have been validated as cost-effective indicators for cardiovascular disease (CVD) risk.

• The combination of the TyG index with obesity indices provides a more comprehensive assessment of CVD risk.

What is the key research question?

• Can the TyG-novel obesity indices further enhance CVD risk assessment capability? Which TyG-obesity indices trajectory pattern (TyG index, TyG-BMI, TyG-WC, TyG-WHtR, TyG-WWI, TyG-ABSI, TyG-BRI, and TyG-CVAI) is most optimal for CVD risk assessment?

What is new?

• In this prospective cohort study, longitudinal trajectories of TyG in combination with both classical and novel obesity indices were significantly associated with CVD.

• Compared to static baseline values, longitudinal assessment of cumulative exposure and trajectory of TyG and its obesity derivatives enhanced predictive capacity for CVD.

• Combination indices of TyG with classical obesity index WC and novel obesity index CVAI (TyG-WC and TyG-CVAI) exhibit optimal performance for CVD risk assessment, regardless of whether evaluating static baseline values or longitudinal trajectories.

How might this study influence clinical practice?

• Long-term dynamic monitoring of the TyG and its obesity derivatives (particularly TyG-WC and TyG-CVAI) facilitates early CVD risk stratification and personalized prevention.

Background

Despite advances in medical technology, cardiovascular disease (CVD) remains the leading threat to global health. Data from three key time points revealed a concerning trend: CVD-related deaths increased by 53.7% in 2019 compared to 1990, and further rose by 6.5% in 2022 relative to 2019 [1–3]. In response to this challenge, the international medical community has reached a consensus: early identification of high-risk CVD populations and primary prevention of risk factors are the most critical intervention strategies [4, 5]. An epidemiological meta-analysis involving over 1.8 million participants globally demonstrated that CVD risk factors exhibited significant modifiable characteristics, with metabolic factors contributing over 40% to CVD incidence [6]. This finding provides a clear direction for developing targeted metabolic regulatory prevention strategies.

Insulin resistance (IR) and obesity are two pivotal metabolic issues threatening cardiovascular health [6–8]. These abnormalities not only disrupt metabolic factors such as glucose and lipids but also trigger chronic inflammatory responses, thereby accelerating CVD progression [7–9]. In standard clinical settings, IR is typically assessed using the hyperinsulinemic-euglycemic clamp test [10], while obesity is evaluated through body composition analyzers or imaging techniques [11]. However, it should be noted that the hyperinsulinemic-euglycemic clamp test and body composition analyzers/imaging techniques are not suitable for large-scale population-based surveys [11, 12], making them inconvenient tools for epidemiological research. To facilitate application in broad populations, researchers have developed the triglyceride glucose (TyG) index as a simple surrogate for IR [13], along with multiple obesity indices including classical measures (waist circumference [WC], body mass index [BMI], and waist-to-height ratio [WHtR]) and novel indices (weight adjusted waist index [WWI], body roundness index [BRI], a body shape index [ABSI], and Chinese visceral adiposity index [CVAI]) [14–18]. Subsequent numerous epidemiological studies have validated these simplified assessment tools of IR and obesity as cost-effective indicators of CVD risk [14, 19–22]. Notably, combinations of the TyG index with obesity indices provide more comprehensive CVD risk estimation compared to the TyG index alone [23–26]. These findings further underscore the importance of routinely monitoring IR and obesity for CVD risk assessment [6]. Several studies have established the predictive value of combining the TyG index with obesity indices for CVD [23–26]. However, these findings are primarily focused on combinations with classical obesity indices, with limited exploration of TyG-novel obesity indices associations. Furthermore, it remains unclear which trajectory patterns of TyG-novel obesity indices combinations are most suitable for CVD risk assessment. To address these gaps, this study aims to systematically investigate the associations between trajectory patterns (including cumulative exposure levels and clustered control trajectories) of TyG and its classical and novel obesity derivatives, and CVD events.

Methods

This study was conducted based on the China Health and Retirement Longitudinal Study (CHARLS), enrolling 3505 Chinese adults aged ≥ 45 years. CHARLS is a nationally representative prospective cohort study [27], with detailed design protocols provided in Additional file 1. Figure S1 illustrates the complete participant screening process. The study was initiated in 2011–2012, covered 450 village-level units across 28 provinces in China, and initially enrolled approximately 17,000 participants. Through follow-up surveys conducted every 2–3 years (five waves completed in 2011–2012, 2013, 2015, 2018, and 2020), it continuously tracked the health status in middle-aged and older Chinese populations. Notably, blood samples and anthropometric measurements were collected during the baseline (2011–2012) and third wave (2015) surveys, providing critical data for analyzing dynamic trajectory patterns of TyG and its obesity derivatives. Participants with complete data on TyG and its obesity derivatives from the first three survey waves and without pre-existing CVD were included in this study. Incident CVD cases were identified through subsequent follow-ups in 2018 and 2020 (note: participants were excluded if their actual CVD diagnosis time was identified before the Wave 3 survey during follow-up). The detailed study flow is presented in Fig. 1. The CHARLS protocol was approved by the Ethical Review Committee of Peking University (IRB: 00001052-11015). All participants provided written informed consent prior to enrollment. This study strictly adhered to the ethical principles outlined in the Declaration of Helsinki, and the reporting of results complied with the Strengthening the Reporting of Observational Studies in Epidemiology guidelines.

Fig. 1.

CONSORT flow diagram. TyG: Triglyceride glucose; WC: waist circumference; BMI: body mass index; WHtR: waist-to-height ratio; WWI: weight adjusted waist index; BRI: body roundness index; ABSI: a body shape index; CVAI: Chinese visceral adiposity index; CVD: cardiovascular disease

Data collection

The study utilized multidimensional data collected through the CHARLS questionnaire, including: basic demographic characteristics (age, sex, living place, education, marital status, smoking status, and drinking status), standardized measurements of physical parameters (height, weight, and WC), comorbid chronic diseases (hypertension, diabetes, and kidney disease), and hematological indices. For chronic disease determination, we adopted the following criteria: Hypertension was defined as: (1) Average systolic blood pressure ≥ 140 mmHg; (2) Average diastolic blood pressure ≥ 90 mmHg; or (3) Physician-diagnosed history of hypertension [28]. Diabetes was determined based on the following criteria: (1) self-reported physician diagnosis; (2) fasting plasma glucose ≥ 126 mg/dL; or (3) baseline haemoglobin A1c (HbA1c) ≥ 6.5% [29]. Kidney disease was ascertained through self-reported medical history. For anthropometric measurements, the CHARLS team adhered to standardized protocols: height and weight were measured with participants in light clothing, and WC was measured at the umbilical level at the end of normal expiration. Using these anthropometric data, we calculated obesity indices including BMI, WHtR, WWI, ABSI, BRI, and CVAI. For a small proportion (4.44%) of baseline and Wave 3 follow-up measurements showing implausible physiological patterns, corrections were made using longitudinal data from the Wave 2 (2013) national survey. For anthropometric data that did not conform to the laws of physiological changes, we performed imputation with the mean of two valid follow-up measurements. The main procedure is as follows: first, we subtracted the anthropometric data from Wave 3 from that from Wave 1, and defined the data where the difference between the two exceeded 2 standard deviations as such data. Then, we further incorporated the anthropometric measurements from Wave 2 to compare with those from Waves 1 and 3, identifying the survey wave(s) where erroneous data might have occurred. Data where correctness cannot be confirmed were deleted. Finally, the erroneous data were imputed using the mean of the valid data from two survey waves (including valid data from either Wave 1 or Wave 3, along with Wave 2 data). The calculations for obesity indices were performed as follows [14–18]:

Venous blood samples were collected from participants after overnight fasting by a professional team from the Chinese Center for Disease Control and Prevention and stored at − 80 °C. Biochemical parameters, including HbA1c, fasting plasma glucose, high-density lipoprotein-cholesterol, low-density lipoprotein-cholesterol (LDL-C), total cholesterol, triglycerides (TG), high-sensitivity C-reactive protein, uric acid, and creatinine (Cr), were measured following standardized protocols. It should be noted that the blood samples from the CHARLS national surveys were analyzed at different laboratories, with Wave 1 processed at the You'anmen Center for Clinical Laboratory of Capital Medical University and Wave 3 at the KingMed Central Laboratory, both in Beijing. For data quality control, test results from both survey waves were monitored weekly to ensure they remained within quality control ranges [27, 30]. Detailed laboratory methodology parameters are provided in Tables S1 and S2 of Additional file 2.

Exposure assessment

Exposure factors included both TyG and its obesity derivatives measured at baseline, as well as cumulative exposure levels and control status trajectories of TyG and its obesity derivatives assessed by integrating data from baseline and Wave 3 survey [31–33]. The following methods were employed to evaluate the exposure factors [13]:

Cumulative TyG and its obesity derivatives = (TyG and its obesity derivatives2012 + TyG and its obesity derivatives2015)/2 ∗ time (2015–2012) [31, 32]. See Additional file 1 for details.

This study employed the k-means clustering algorithm (Euclidean distance) to analyze control trajectories of TyG and its obesity derivatives from 2012 to 2015. This method offers computational efficiency and visualization advantages [34]. The optimal number of clusters was determined using the elbow method (Additional file 1: Fig. S2), where an inflection point in the improvement of within-cluster variance was observed at k = 3, indicating limited gains in model interpretability from further increasing cluster numbers. Three distinct trajectory patterns were ultimately identified: Class 1 (well-controlled): The TyG and its obesity derivatives were at a low level at baseline, followed by a slight increase, indicating a well-controlled metabolic state; Class 2 (moderately controlled): The TyG and its obesity derivatives were at a moderate level at baseline, followed by a slight increase, indicating a moderately controlled metabolic state; Class 3 (poorly controlled): The TyG and its obesity derivatives were at a high level at baseline, followed by a slight increase or decrease, indicating a poorly controlled metabolic state (see Fig. 2 for details). We employed k-means clustering to identify distinct trajectories of TyG and its obesity derivatives over time. This data-driven method was chosen because it objectively identifies natural subgroups based on patterns of change, rather than imposing arbitrary boundaries. Compared with traditional predefined groups (e.g., low-low, low–high), k-means clustering offers several advantages: it reduces researcher bias in category definition, better captures the continuous nature of changes in TyG and its obesity derivatives, and adapts to the actual distribution of our data [35–38]. This method also enhances statistical power by aligning subgroup formation with observed patterns, thereby improving the biological interpretability of our study results [35–37, 39].

Fig. 2.

Clustering of control trajectories in TyG index, TyG-BMI, TyG-WC, TyG-WHtR, TyG-WWI, TyG-ABSI, TyG-BRI, and TyG-CVAI from 2012 to 2015. Three clusters (Class 1, Class 2, and Class 3) were identified using the k-means method with Euclidean distance. Each cluster is represented by a unique color and shape across all figures to distinguish between groups and highlight variations over time. A Scatter plots comparing standardized values of TyG index, TyG-BMI, TyG-WC, TyG-WHtR, TyG-WWI, TyG-ABSI, TyG-BRI, and TyG-CVAI between 2012 (X-axis) and 2015 (Y-axis). Standardization adjusts the data to a mean of 0 and a standard deviation of 1, facilitating comparison across years. B Line plots showing the mean values of TyG index, TyG-BMI, TyG-WC, TyG-WHtR, TyG-WWI, TyG-ABSI, TyG-BRI, and TyG-CVAI for the three clusters from 2012 to 2015. Each indicator is presented at two time points (2012 and 2015), illustrating temporal trends and highlighting differences in mean values across clusters. C, D Density plots illustrating the distributions of TyG index, TyG-BMI, TyG-WC, TyG-WHtR, TyG-WWI, TyG-ABSI, TyG-BRI, and TyG-CVAI within the three clusters in 2012 and 2015. These plots reveal significant differences in mean levels and variability among clusters, as well as shifts in distributions of indicators over time, allowing for visual comparison across the 2 years. TyG: Triglyceride glucose; WC: waist circumference; BMI: body mass index; WHtR: waist-to-height ratio; WWI: weight adjusted waist index; BRI: body roundness index; ABSI: a body shape index; CVAI: Chinese visceral adiposity index; CVD: cardiovascular disease

Ascertainment of CVD

The primary outcome for this study was incident CVD events identified in Waves 4 and 5 of national follow-up surveys [40], encompassing heart disease and stroke. Diagnosis was based on responses to the following standardized questionnaire items: “Have you ever been diagnosed with a heart attack, coronary heart disease, angina, congestive heart failure, or other heart problems by a physician?” or “Have you ever been diagnosed with a stroke by a physician?”.

Statistical analysis

The data analysis platforms included R 4.5 and Empower(R) 4.0. We first assessed the dataset missingness; given the high missingness rates in TyG and its obesity derivatives within the CHARLS dataset, this study systematically compared baseline characteristics between participants with complete data and those with missing data at baseline and Wave 3 follow-up. Analysis results indicated that the two groups had similar distributions across most baseline characteristics (Additional file 2: Tables S3–S4), suggesting that missing data might follow a missing at random mechanism, with no significant association with observed or unobserved variables [41]. Among the final 3,505 participants included in this study, the maximum missingness rate for any variable was only 0.79% (Additional file 2: Table S5, and Additional file 1: Fig. S3). Given the extremely low missingness proportion, analyses were conducted directly on the original dataset to fully preserve data information.

To clearly present the cumulative exposure levels and control trajectories of TyG and its obesity derivatives, the study population was stratified by primary outcome. Characteristics of the study population were presented using differentiated approaches based on data types (see footnotes of Table 1). Between-group comparisons were conducted using the standardized difference method, with a clinical significance threshold set at 0.1 [42, 43]. In addition, we also applied the t-test or Mann–Whitney U test and chi-square test for intergroup comparisons, with a significance threshold of P < 0.05.

Table 1.

Baseline characteristics and cumulative metabolic changes stratified by incident CVD status

	CVD		Standardized difference, % (95% CI)	P value
	No	Yes
No. of subjects	3064	441
Age, years	57.00 (51.00–64.00)	59.00 (53.00–64.00)	0.10 (− 0.00, 0.20)	0.07
Sex, n (%)			0.13 (0.03, 0.23)	0.01
Male	1461 (47.68%)	182 (41.27%)
Female	1603 (52.32%)	259 (58.73%)
Marital status, n (%)			0.05 (-0.05, 0.15)	0.32
Married	2744 (89.56%)	388 (87.98%)
Other	320 (10.44%)	53 (12.02%)
Living place, n (%)			0.06 (-0.04, 0.16)	0.26
Village	2037 (66.48%)	305 (69.16%)
City	1027 (33.52%)	136 (30.84%)
Education, n (%)			0.18 (0.08, 0.28)	< 0.01
Below primary	1439 (47.00%)	221 (50.11%)
Primary schools	687 (22.44%)	118 (26.76%)
Middle school	657 (21.46%)	67 (15.19%)
High school and above	279 (9.11%)	35 (7.94%)
Kidney disease, n (%)			0.05 (− 0.05, 0.15)	0.35
No	2917 (95.92%)	414 (94.95%)
Yes	124 (4.08%)	22 (5.05%)
Hypertension, n (%)			0.28 (0.18, 0.38)	< 0.01
No	1808 (59.01%)	199 (45.12%)
Yes	1256 (40.99%)	242 (54.88%)
Diabetes, n (%)			0.09 (-0.01, 0.19)	0.07
No	2647 (86.39%)	367 (83.22%)
Yes	417 (13.61%)	74 (16.78%)
Drinking status, n (%)			0.04 (-0.06, 0.14)	0.41
No	1966 (64.21%)	292 (66.21%)
Yes	1096 (35.79%)	149 (33.79%)
Smoking status, n (%)			0.07 (-0.03, 0.17)	0.17
No	2088 (68.28%)	314 (71.53%)
Yes	970 (31.72%)	125 (28.47%)
FPG, mg/dL	101.88 (94.50–111.06)	103.86 (95.22–114.48)	0.16 (0.06, 0.26)	< 0.01
Cr, mg/dL	0.75 (0.64–0.87)	0.75 (0.63–0.85)	0.04 (-0.06, 0.14)	0.35
TC, mg/dL	192.89 (38.85)	200.67 (38.54)	0.20 (0.10, 0.30)	< 0.01
TG, mg/dL	100.89 (71.69–146.91)	108.86 (81.42–153.99)	0.10 (0.00, 0.20)	< 0.01
HDL-C, mg/dL	49.87 (40.98–60.70)	48.33 (40.21–58.76)	0.08 (-0.02, 0.18)	0.06
LDL-C, mg/dL	113.66 (92.78–135.70)	119.65 (98.10–143.62)	0.19 (0.09, 0.29)	< 0.01
hs-CRP, mg/L	0.92 (0.53–1.93)	1.21 (0.58–2.21)	0.03 (-0.07, 0.13)	0.48
HbA1c, %	5.24 (0.76)	5.41 (1.11)	0.18 (0.08, 0.28)	< 0.01
UA, mg/dL	4.22 (3.52–5.05)	4.29 (3.53–5.11)	0.02 (-0.08, 0.12)	0.70
TyG index	8.63 (0.65)	8.76 (0.71)	0.19 (0.09, 0.29)	< 0.01
Height, m	1.58 (0.08)	1.58 (0.08)	0.04 (-0.06, 0.14)	0.29
Weight, kg	58.38 (10.79)	60.31 (11.24)	0.18 (0.08, 0.28)	< 0.01
BMI, kg/m2	23.31 (3.56)	24.17 (3.65)	0.24 (0.14, 0.34)	< 0.01
WC, cm	84.48 (9.52)	87.08 (9.73)	0.27 (0.17, 0.37)	< 0.01
WHtR	0.54 (0.06)	0.55 (0.06)	0.28 (0.18, 0.38)	< 0.01
WWI	11.11 (0.84)	11.27 (0.89)	0.19 (0.09, 0.29)	< 0.01
ABSI	0.08 (0.01)	0.08 (0.01)	0.11 (0.01, 0.21)	0.03
BRI	4.12 (1.28)	4.49 (1.37)	0.28 (0.18, 0.38)	< 0.01
CVAI	89.75 (64.91–117.64)	101.35 (78.08–132.43)	0.29 (0.19, 0.39)	< 0.01
Cumulative TyG index	25.90 (1.69)	26.31 (1.86)	0.23 (0.13, 0.33)	< 0.01
Cumulative TyG-BMI	608.21 (109.62)	639.56 (114.02)	0.28 (0.18, 0.38)	< 0.01
Cumulative TyG-WC	2205.40 (326.92)	2308.60 (350.55)	0.31 (0.20, 0.40)	< 0.01
Cumulative TyG-WHtR	14.00 (2.12)	14.70 (2.33)	0.31 (0.21, 0.41)	< 0.01
Cumulative TyG-WWI	289.37 (30.78)	298.40 (34.77)	0.28 (0.18, 0.38)	< 0.01
Cumulative TyG-ABSI	2.15 (0.20)	2.20 (0.22)	0.25 (0.15, 0.35)	< 0.01
Cumulative TyG-BRI	109.58 (36.26)	121.32 (40.51)	0.31 (0.21, 0.41)	< 0.01
Cumulative TyG-CVAI	2508.64 (1062.01)	2846.95 (1111.35)	0.31 (0.21, 0.41)	< 0.01
TyG index control trajectory			0.24 (0.14, 0.34)	< 0.01
Class 1	1360 (44.39%)	149 (33.79%)
Class 2	1305 (42.59%)	207 (46.94%)
Class 3	399 (13.02%)	85 (19.27%)
TyG-BMI control trajectory			0.32 (0.22, 0.42)	< 0.01
Class 1	1165 (38.02%)	115 (26.08%)
Class 2	1306 (42.62%)	190 (43.08%)
Class 3	593 (19.35%)	136 (30.84%)
TyG-WC control trajectory			0.33 (0.23, 0.43)	< 0.01
Class 1	1194 (38.97%)	109 (24.72%)
Class 2	1246 (40.67%)	199 (45.12%)
Class 3	624 (20.37%)	133 (30.16%)
TyG-WHtR control trajectory			0.30 (0.20, 0.40)	< 0.01
Class 1	1143 (37.30%)	111 (25.17%)
Class 2	1271 (41.48%)	191 (43.31%)
Class 3	650 (21.21%)	139 (31.52%)
TyG-WWI control trajectory			0.22 (0.12, 0.32)	< 0.01
Class 1	1114 (36.36%)	122 (27.66%)
Class 2	1356 (44.26%)	201 (45.58%)
Class 3	594 (19.39%)	118 (26.76%)
TyG-ABSI control trajectory			0.21 (0.11, 0.31)	< 0.01
Class 1	1179 (38.48%)	136 (30.84%)
Class 2	1383 (45.14%)	200 (45.35%)
Class 3	502 (16.38%)	105 (23.81%)
TyG-BRI control trajectory			0.30 (0.20, 0.40)	< 0.01
Class 1	1228 (40.08%)	124 (28.12%)
Class 2	1252 (40.86%)	187 (42.40%)
Class 3	584 (19.06%)	130 (29.48%)
TyG-CVAI control trajectory			0.33 (0.23, 0.43)	< 0.01
Class 1	1149 (37.50%)	107 (24.26%)
Class 2	1264 (41.25%)	188 (42.63%)
Class 3	651 (21.25%)	146 (33.11%)

Values were expressed as mean (standard deviation) or medians (interquartile range) or n (%)

HbA1c haemoglobin A1c, HDL-C high-density lipoprotein-cholesterol, hs-CRP high-sensitivity C-reactive protein, LDL-C low-density lipoprotein-cholesterol, TC total cholesterol, TG triglycerides, UA uric acid, Cr creatinine, CI confidence interval, TyG triglyceride glucose, CVD cardiovascular disease, BMI body mass index, WC waist circumference, WHtR waist-to-height ratio, ABSI a body shape index, WWI weight adjusted waist index, CVAI Chinese Visceral Adiposity Index; BRI, Body Roundness Index

This study employed multivariable Logistic regression models to evaluate the associations of TyG and its obesity derivatives (including both baseline measurements and longitudinal trajectories) with the outcome. To ensure comparability of effect estimates across variables, Z-score standardization was applied to normally distributed independent variables (Fig. 2C, D), and standardized odds ratios (ORs) with 95% confidence intervals were calculated. Additionally, the associations between tertile groups of TyG and its obesity derivatives (including baseline values and cumulative exposure) and the outcome were analyzed to provide supplementary evidence for comparing association strengths. Covariate selection was based on clinical significance and inter-variable correlations. To avoid multicollinearity, components of independent variables were excluded from adjusted models. By analyzing the covariate correlation matrix, the following highly correlated variable pairs were excluded from multivariable models (Additional file 1: Fig. S4): HbA1c (with diabetes), uric acid (with Cr), and LDL-C (with total cholesterol). Restricted cubic spline regression was used to model dose–response relationships between TyG and its obesity derivatives (including baseline measurements and cumulative exposure values) and the outcome; likelihood ratio tests were performed to evaluate potential nonlinear effects.

To evaluate the predictive performance of the TyG and its obesity derivatives (including baseline values and cumulative exposure) for CVD events, this study conducted receiver operating characteristic curve analysis and calculated the area under the curve, best threshold, sensitivity, and specificity. To further confirm the advantage of the combination of the TyG index and obesity in predicting CVD events, we compared the TyG-obesity derivatives with the TyG index and glucose alone using the Delong test [44]. Additionally, we investigated the incremental predictive performance of adding the TyG and its obesity derivatives to the Framingham Risk Score (FRS) model, and calculated the C-index and net reclassification improvement to quantify the improvement in predictive ability.

Given that TyG and its obesity derivatives are derived from routinely measured parameters, we employed weighted quantile sum (WQS) regression based on 1000 bootstrap resamples to systematically evaluate the differential contributions of individual metabolic components to CVD incidence during long-term exposure. This model utilized a standardized weight allocation algorithm (within the 0–1 range), where the weight coefficients of each component directly quantified their relative contribution to CVD risk [45].

To ensure the robustness of our findings, sensitivity analyses were conducted under different prior hypothesis frameworks to evaluate associations between longitudinal trajectories of TyG and its obesity derivatives and CVD: (1) We evaluated the associations between trajectories of TyG and its obesity derivatives and CVD across subgroups stratified by age, sex, smoking/drinking status, and hypertension/diabetes/kidney disease status, and examined potential population heterogeneity using likelihood ratio tests. (2) To mitigate potential reverse causation bias, we repeated the association analysis after excluding CVD patients diagnosed in the Wave 4 survey. (3) Missing factors were addressed by either setting dummy variables or performing mean/median imputation to generate a complete dataset; subsequent re-analyses evaluated associations between longitudinal trajectories of TyG and its obesity derivatives and CVD in this complete dataset.

Results

Study population characteristics

This study analyzed data from 3505 CHARLS participants, comprising 1643 males and 1,862 females. During the median 8-year follow-up period, there were 323 incident heart disease cases, 148 strokes, and a total of 411 CVD events. Table 1 presents the baseline characteristics and dynamic trajectories of TyG and its obesity derivatives in the study population, stratified by future CVD occurrence. Compared to participants who did not develop CVD, those who were later diagnosed with CVD were more likely to be female, had lower educational attainment, had a higher prevalence of hypertension, were more obese, and exhibited significantly worse glucose-lipid metabolic profiles.

In terms of cumulative exposure profiles, participants who experienced CVD events during follow-up exhibited significantly higher cumulative levels of TyG and its obesity derivatives compared to those without CVD events, with the most pronounced between-group differences observed in cumulative TyG-WC, cumulative TyG-WHtR, cumulative TyG-BRI, and cumulative TyG-CVAI (standardized difference = 0.31).

In terms of trajectory control performance, participants who developed CVD during follow-up demonstrated generally poorer control of TyG and its obesity derivatives compared to non-CVD participants. Notably, the control levels of TyG-WC and TyG-CVAI exhibited the largest between-group differences (standardized difference = 0.33).

Associations of baseline TyG and its obesity derivatives and CVD

In the age- and sex-adjusted logistic regression analyses (Model I), we observed significant positive associations between baseline TyG and its obesity derivatives and CVD. After further adjusting for education, living place, kidney disease, hypertension, diabetes, smoking status, and drinking status (Model II), the associations between TyG and its obesity derivatives (as continuous variables) and CVD slightly attenuated but remained statistically significant. Notably, in Model II, the associations between the highest tertile of TyG index and TyG-ABSI and CVD became non-significant. In the final model (Model III), all TyG and its obesity derivatives, except TyG index and TyG-ABSI, maintained significant positive associations with CVD [OR per standard deviation increase: TyG-BMI 1.22, TyG-WC 1.24, TyG-WHtR 1.22, TyG-WWI 1.13, TyG-BRI 1.22, TyG-CVAI 1.23]. Based on standardized analyses with TyG and its obesity derivatives treated as both continuous and categorical variables, TyG-WC and TyG-CVAI demonstrated the strongest associations with CVD (Table 2).

Table 2.

Association between baseline TyG and its obesity derivatives with CVD events

	OR (95% CI)
	Non-adjusted	Model I	Model II	Model III
TyG index (Per SD increase)	1.20 (1.09, 1.31)	1.19 (1.08, 1.31)	1.13 (1.01, 1.26)	1.08 (0.97, 1.22)
TyG-BMI (Per SD increase)	1.29 (1.17, 1.42)	1.30 (1.18, 1.43)	1.24 (1.12, 1.38)	1.22 (1.10, 1.36)
TyG-WC (Per SD increase)	1.32 (1.20, 1.45)	1.32 (1.20, 1.45)	1.27 (1.14, 1.41)	1.24 (1.11, 1.39)
TyG-WHtR (Per SD increase)	1.33 (1.21, 1.47)	1.31 (1.18, 1.45)	1.25 (1.12, 1.40)	1.22 (1.09, 1.37)
TyG-WWI (Per SD increase)	1.27 (1.15, 1.40)	1.23 (1.11, 1.37)	1.17 (1.04, 1.31)	1.13 (1.00, 1.27)
TyG-ABSI (Per SD increase)	1.22 (1.11, 1.35)	1.19 (1.08, 1.31)	1.13 (1.01, 1.26)	1.09 (0.98, 1.22)
TyG-BRI (Per SD increase)	1.33 (1.21, 1.46)	1.31 (1.18, 1.45)	1.25 (1.12, 1.39)	1.22 (1.10, 1.37)
TyG-CVAI (Per SD increase)	1.33 (1.20, 1.46)	1.31 (1.18, 1.44)	1.25 (1.13, 1.40)	1.23 (1.10, 1.38)
TyG index tertile groups
T1 (6.14–8.32)	1.0	1.0	1.0	1.0
T2 (8.32–8.83)	1.53 (1.19, 1.97)	1.50 (1.16, 1.93)	1.41 (1.09, 1.82)	1.36 (1.05, 1.76)
T3 (8.83–12.96)	1.48 (1.14, 1.91)	1.43 (1.11, 1.85)	1.24 (0.94, 1.63)	1.14 (0.86, 1.52)
TyG-BMI tertile groups
T1 (108.91–183.20)	1.0	1.0	1.0	1.0
T2 (183.21–215.45)	1.50 (1.15, 1.96)	1.52 (1.17, 1.99)	1.47 (1.13, 1.93)	1.45 (1.10, 1.90)
T3 (215.49–406.66)	1.90 (1.48, 2.46)	1.94 (1.49, 2.52)	1.74 (1.31, 2.30)	1.66 (1.26, 2.22)
TyG-WC tertile groups
T1 (417.77–672.60)	1.0	1.0	1.0	1.0
T2 (672.76–777.68)	1.60 (1.23, 2.09)	1.58 (1.21, 2.06)	1.53 (1.17, 2.00)	1.49 (1.14, 1.96)
T3 (777.69–1265.20)	2.00 (1.55, 2.59)	1.97 (1.52, 2.55)	1.76 (1.34, 2.33)	1.68 (1.27, 2.23)
TyG-WHtR tertile groups
T1 (2.79–4.27)	1.0	1.0	1.0	1.0
T2 (4.27–4.94)	1.50 (1.15, 1.96)	1.47 (1.12, 1.92)	1.41 (1.08, 1.86)	1.38 (1.05, 1.82)
T3 (4.94–7.59)	2.01 (1.55, 2.59)	1.92 (1.47, 2.50)	1.69 (1.28, 2.25)	1.61 (1.21, 2.15)
TyG-WWI tertile groups
T1 (56.38–90.83)	1.0	1.0	1.0	1.0
T2 (90.84–100.41)	1.36 (1.05, 1.77)	1.31 (1.01, 1.71)	1.27 (0.97, 1.66)	1.23 (0.94, 1.61)
T3 (100.42–161.60)	1.82 (1.42, 2.34)	1.69 (1.29, 2.20)	1.50 (1.13, 1.99)	1.40 (1.05, 1.87)
TyG-ABSI tertile groups
T1 (0.39–0.68)	1.0	1.0	1.0	1.0
T2 (0.68–0.74)	1.31 (1.01, 1.69)	1.26 (0.98, 1.63)	1.24 (0.95, 1.60)	1.20 (0.92, 1.55)
T3 (0.74–1.21)	1.51 (1.18, 1.94)	1.40 (1.08, 1.81)	1.24 (0.95, 1.63)	1.15 (0.87, 1.52)
TyG-BRI tertile groups
T1 (6.94–29.45)	1.0	1.0	1.0	1.0
T2 (29.45–40.64)	1.46 (1.12, 1.90)	1.44 (1.10, 1.88)	1.37 (1.05, 1.80)	1.34 (1.02, 1.76)
T3 (40.65–87.46)	1.95 (1.51, 2.52)	1.86 (1.43, 2.43)	1.63 (1.23, 2.17)	1.57 (1.18, 2.09)
TyG-CVAI tertile groups
T1 (− 149.94–627.41)	1.0	1.0	1.0	1.0
T2 (627.79–955.43)	1.66 (1.27, 2.16)	1.58 (1.21, 2.07)	1.54 (1.17, 2.02)	1.52 (1.15, 1.99)
T3 (955.50–2582.82)	2.00 (1.54, 2.58)	1.90 (1.46, 2.47)	1.68 (1.27, 2.22)	1.66 (1.22, 2.15)

OR odds ratios, CI confidence interval, SD standard deviations, CVD cardiovascular disease, other abbreviations as in Table 1

Model I adjust for age, sex

Model II adjust for age, sex, education, living place, kidney disease, hypertension, diabetes, smoking status, drinking status

Model III adjust for age, sex, education, living place, kidney disease, hypertension, diabetes, smoking status, drinking status, LDL-C, Cr, CRP

Associations between longitudinal trajectories of TyG and its obesity derivatives and CVD

As shown in Table 3, the trajectories of TyG and its obesity derivatives (including cumulative exposure and control trajectories) were analyzed for their association with CVD. Multivariable Logistic Regression analyses (Model III) revealed that cumulative TyG and its obesity derivatives (as continuous variables) were all positively associated with CVD. Specifically, per standard deviation increase in cumulative TyG index, cumulative TyG-BMI, cumulative TyG-WC, cumulative TyG-WHtR, cumulative TyG-WWI, cumulative TyG-ABSI, cumulative TyG-BRI, and cumulative TyG-CVAI were associated with 16%, 25%, 28%, 26%, 19%, 25%, 27%, and 15% increases in CVD risk, respectively. Further analyses based on tertiles of cumulative TyG and its obesity derivatives revealed similar positive associations. Notably, the highest tertile of cumulative TyG-ABSI showed a borderline- significant positive association with CVD. Overall, combining standardized analyses treating Cumulative TyG and its obesity derivatives as both continuous and categorical variables, cumulative TyG-WC and cumulative TyG-CVAI demonstrated the strongest associations with CVD.

Table 3.

Association between longitudinal trajectories in TyG and its obesity derivatives and CVD events

	OR (95% CI)
	Non-adjusted	Model I	Model II	Model III
Cumulative TyG index (Per SD increase)	1.25 (1.14, 1.38)	1.24 (1.13, 1.37)	1.19 (1.07, 1.33)	1.16 (1.03, 1.30)
Cumulative TyG-BMI (Per SD increase)	1.31 (1.19, 1.44)	1.33 (1.21, 1.47)	1.28 (1.15, 1.42)	1.25 (1.12, 1.40)
Cumulative TyG-WC (Per SD increase)	1.35 (1.22, 1.49)	1.35 (1.22, 1.49)	1.30 (1.17, 1.45)	1.28 (1.14, 1.43)
Cumulative TyG-WHtR (Per SD increase)	1.36 (1.24, 1.50)	1.35 (1.22, 1.49)	1.29 (1.15, 1.44)	1.26 (1.12, 1.41)
Cumulative TyG-WWI (Per SD increase)	1.32 (1.20, 1.45)	1.29 (1.16, 1.43)	1.22 (1.09, 1.37)	1.19 (1.05, 1.34)
Cumulative TyG-ABSI (Per SD increase)	1.28 (1.16, 1.40)	1.24 (1.12, 1.37)	1.18 (1.06, 1.32)	1.15 (1.03, 1.29)
Cumulative TyG-BRI (Per SD increase)	1.35 (1.23, 1.48)	1.33 (1.20, 1.47)	1.27 (1.14, 1.42)	1.25 (1.12, 1.39)
Cumulative TyG-CVAI (Per SD increase)	1.36 (1.23, 1.50)	1.34 (1.21, 1.48)	1.29 (1.15, 1.43)	1.27 (1.13, 1.41)
Cumulative TyG index tertile groups
T1 (21.14–25.05)	1.0	1.0	1.0	1.0
T2 (25.05–26.44)	1.49 (1.15, 1.93)	1.45 (1.12, 1.88)	1.40 (1.07, 1.82)	1.37 (1.05, 1.79)
T3 (26.44–36.16)	1.73 (1.34, 2.23)	1.68 (1.30, 2.18)	1.50 (1.14, 1.96)	1.40 (1.05, 1.86)
Cumulative TyG-BMI tertile groups
T1 (108.91–214.14)	1.0	1.0	1.0	1.0
T2 (139.19–262.16)	1.57 (1.20, 2.04)	1.59 (1.22, 2.08)	1.52 (1.16, 2.00)	1.52 (1.16, 2.00)
T3 (157.95–406.66)	1.89 (1.46, 2.44)	1.94 (1.49, 2.52)	1.74 (1.31, 2.30)	1.70 (1.28, 2.27)
Cumulative TyG-WC tertile groups
T1 (1327.91–2042.17)	1.0	1.0	1.0	1.0
T2 (2042.38–2340.49)	1.73 (1.32, 2.26)	1.70 (1.30, 2.23)	1.62 (1.23, 2.13)	1.58 (1.20, 2.08)
T3 (2340.90–3638.68)	2.09 (1.61, 2.71)	2.07 (1.59, 2.69)	1.86 (1.40, 2.46)	1.78 (1.34, 2.36)
Cumulative TyG-WHtR tertile groups
T1 (8.97–12.94)	1.0	1.0	1.0	1.0
T2 (12.94–14.88)	1.51 (1.16, 1.97)	1.48 (1.13, 1.94)	1.40 (1.07, 1.85)	1.38 (1.05, 1.82)
T3 (14.89–23.32)	1.99 (1.54, 2.57)	1.92 (1.47, 2.51)	1.68 (1.26, 2.24)	1.59 (1.19, 2.13)
Cumulative TyG-WWI tertile groups
T1 (208.57–274.76)	1.0	1.0	1.0	1.0
T2 (274.82–301.64)	1.20 (0.92, 1.56)	1.16 (0.89, 1.51)	1.09 (0.83, 1.43)	1.06 (0.80, 1.39)
T3 (301.70–456.10)	1.79 (1.40, 2.29)	1.66 (1.27, 2.17)	1.46 (1.10, 1.93)	1.36 (1.02, 1.82)
Cumulative TyG-ABSI tertile groups
T1 (1.52–2.06)	1.0	1.0	1.0	1.0
T2 (2.06–2.23)	1.17 (0.90, 1.52)	1.13 (0.87, 1.47)	1.09 (0.83, 1.42)	1.05 (0.80, 1.37)
T3 (2.23–3.15)	1.69 (1.32, 2.17)	1.57 (1.22, 2.03)	1.39 (1.06, 1.82)	1.30 (0.98, 1.72)
Cumulative TyG-BRI tertile groups
T1 (31.53–90.66)	1.0	1.0	1.0	1.0
T2 (90.66–123.35)	1.39 (1.07, 1.82)	1.37 (1.04, 1.79)	1.29 (0.98, 1.70)	1.28 (0.97, 1.69)
T3 (123.38–283.61)	1.92 (1.50, 2.48)	1.84 (1.41, 2.40)	1.63 (1.23, 2.16)	1.56 (1.17, 2.08)
Cumulative TyG-CVAI tertile groups
T1 (− 223.80–2001.21)	1.0	1.0	1.0	1.0
T2 (2002.12–2977.61)	1.43 (1.09, 1.87)	1.37 (1.04, 1.79)	1.33 (1.01, 1.75)	1.31 (0.99, 1.73)
T3 (2977.96–7081.90)	2.08 (1.62, 2.68)	1.98 (1.53, 2.56)	1.78 (1.35, 2.34)	1.72 (1.30, 2.27)
TyG index control trajectory
Class 1	1.0	1.0	1.0	1.0
Class 2	1.45 (1.16, 1.81)	1.42 (1.13, 1.78)	1.33 (1.06, 1.68)	1.29 (1.02, 1.63)
Class 3	1.94 (1.46, 2.60)	1.91 (1.43, 2.56)	1.72 (1.25, 2.36)	1.61 (1.15, 2.24)
TyG-BMI control trajectory
Class 1	1.0	1.0	1.0	1.0
Class 2	1.47 (1.15, 1.88)	1.51 (1.18, 1.93)	1.44 (1.12, 1.87)	1.44 (1.11, 1.86)
Class 3	2.32 (1.78, 3.04)	2.41 (1.83, 3.18)	2.19 (1.63, 2.95)	2.03 (1.58, 2.88)
TyG-WC control trajectory
Class 1	1.0	1.0	1.0	1.0
Class 2	1.75 (1.37, 2.24)	1.73 (1.35, 2.22)	1.66 (1.29, 2.14)	1.61 (1.25, 2.09)
Class 3	2.33 (1.78, 3.06)	2.32 (1.76, 3.05)	2.09 (1.56, 2.81)	2.00 (1.48, 2.70)
TyG-WHtR control trajectory
Class 1	1.0	1.0	1.0	1.0
Class 2	1.55 (1.21, 1.98)	1.52 (1.18, 1.96)	1.44 (1.11, 1.87)	1.41 (1.08, 1.83)
Class 3	2.20 (1.69, 2.88)	2.12 (1.60, 2.81)	1.87 (1.38, 2.54)	1.77 (1.30, 2.41)
TyG-WWI control trajectory
Class 1	1.0	1.0	1.0	1.0
Class 2	1.35 (1.07, 1.72)	1.30 (1.01, 1.65)	1.20 (0.93, 1.54)	1.15 (0.89, 1.48)
Class 3	1.81 (1.38, 2.38)	1.65 (1.23, 2.21)	1.41 (1.03, 1.92)	1.31 (0.95, 1.80)
TyG-ABSI control trajectory
Class 1	1.0	1.0	1.0	1.0
Class 2	1.25 (0.99, 1.58)	1.20 (0.95, 1.51)	1.13 (0.89, 1.44)	1.07 (0.84, 1.37)
Class 3	1.81 (1.38, 2.39)	1.67 (1.26, 2.22)	1.47 (1.08, 1.99)	1.37 (1.00, 1.87)
TyG-BRI control trajectory
Class 1	1.0	1.0	1.0	1.0
Class 2	1.48 (1.16, 1.88)	1.46 (1.14, 1.87)	1.39 (1.08, 1.79)	1.36 (1.05, 1.75)
Class 3	2.20 (1.69, 2.87)	2.12 (1.60, 2.81)	1.85 (1.37, 2.50)	1.76 (1.30, 2.39)
TyG-CVAI control trajectory
Class 1	1.0	1.0	1.0	1.0
Class 2	1.60 (1.24, 2.05)	1.52 (1.18, 1.97)	1.48 (1.14, 1.92)	1.45 (1.12, 1.89)
Class 3	2.41 (1.84, 3.15)	2.29 (1.75, 3.01)	2.07 (1.55, 2.76)	2.00 (1.47, 2.65)

OR odds ratios, CI confidence interval, SD standard deviations, CVD cardiovascular disease; other abbreviations as in Table 1

Model I adjust for age, sex

Model II adjust for age, sex, education, living place, kidney disease, hypertension, diabetes, smoking status, drinking status

Model III adjust for age, sex, education, living place, kidney disease, hypertension, diabetes, smoking status, drinking status, LDL-C, Cr, CRP

We also evaluated the associations between control trajectories of TyG and its obesity derivatives and CVD (Table 3). In unadjusted Model to Model II analyses, we observed that compared to participants with well-controlled trajectories, those with poorly controlled trajectories exhibited significant positive associations with CVD risk. In the final model (Model III), we observed that poorly controlled TyG-BMI, TyG-WC, and TyG-CVAI exhibited the strongest associations with CVD, while poorly controlled TyG-WWI and TyG-ABSI showed borderline significant positive associations with CVD. Combining findings from cumulative exposure and control trajectory analyses, these results suggest that TyG-WWI and TyG-ABSI have poorer stability in CVD risk assessment.

Dose–response relationship

The dose–response relationships between TyG and its obesity derivatives with CVD (including baseline static levels and follow-up cumulative exposure) are presented in Fig. 3. After rigorous adjustment for confounding factors, we observed linear associations between baseline measurements of TyG and its obesity derivatives, as well as follow-up cumulative exposure, with CVD (p for nonlinearity > 0.05).

Fig. 3.

Dose–response relationship between baseline/cumulative TyG and its obesity derivatives and CVD risk using restricted cubic spline. A TyG index; B TyG-classical obesity indices; C TyG-novel obesity index. Adjust for age, sex, education, living place, kidney disease, hypertension, diabetes, smoking status, drinking status, LDL-C, Cr, CRP. TyG: Triglyceride glucose; WC: waist circumference; BMI: body mass index; WHtR: waist-to-height ratio; WWI: weight adjusted waist index; BRI: body roundness index; ABSI: a body shape index; CVAI: Chinese visceral adiposity index; CVD: cardiovascular disease

Predictive analysis

The receiver operating characteristic curve analysis demonstrated that the TyG-obesity indices were significantly superior to both glucose and the TyG index in predicting CVD events (all Delong p < 0.05), with TyG-WC and TyG-CVAI showing the highest predictive accuracy (Additional file 2: Table S6). This finding further supports the results of the association analysis.

When the TyG and its obesity derivatives were added to the FRS model (Additional file 2: Table S7), we observed a significant improvement in the model’s ability to predict CVD: the C-index increased significantly (p value < 0.05), and the net reclassification improvement was greater than 0. These findings suggest that adding the TyG and its obesity derivatives provides significant incremental value to the FRS risk model for CVD prediction.

WQS analysis

The WQS regression model was employed to quantify the differential contributions of cumulative exposure components in TyG and its obesity derivatives to CVD. Analysis revealed that cumulative TG and cumulative obesity had significantly higher weights than other metabolic parameters (Fig. 4), indicating that long-term lipid metabolic dysregulation and obesity status serve as core drivers of CVD pathogenesis.

Fig. 4.

Relative weights of cumulative obesity index, cumulative FPG, and cumulative TG contributing to CVD risk via WQS regression. A Contribution rates of components in the cumulative TyG index; B Contribution rates of components in the cumulative TyG-BMI; C Contribution rates of components in the cumulative TyG-WC; D Contribution rates of components in the cumulative TyG-WHtR; E Contribution rates of components in the cumulative TyG-WWI; F Contribution rates of components in the cumulative TyG-ABSI; G Contribution rates of components in the cumulative TyG-BRI; H Contribution rates of components in the cumulative TyG-CVAI. CVD: cardiovascular disease; WQS: weighted quantile sum; TG: triglycerides; FPG: fasting plasma glucose; WC: waist circumference; BMI: body mass index; WHtR: waist-to-height ratio; WWI: weight adjusted waist index; BRI: body roundness index; ABSI: a body shape index; CVAI: Chinese visceral adiposity index; Cum: cumulative

Sensitivity analysis

Sensitivity Analysis 1: We explored the associations between longitudinal trajectories of TyG and its obesity derivatives with CVD across different populations. The results showed that, whether evaluating cumulative exposure (Fig. 5) or control trajectories (Additional file 2: Tables S8–S14), the associations of TyG and its obesity derivatives with CVD remained consistent across subgroups stratified by sex, age (< 60/≥ 60 years), smoking/drinking status, or comorbidities (hypertension, diabetes, or kidney disease).

Fig. 5.

Forest plot presenting subgroup-specific associations of cumulative TyG and its obesity derivatives with CVD. TyG: Triglyceride glucose; WC: waist circumference; BMI: body mass index; WHtR: waist-to-height ratio; WWI: weight adjusted waist index; BRI: body roundness index; ABSI: a body shape index; CVAI: Chinese visceral adiposity index; CVD: cardiovascular disease

Sensitivity Analysis 2: Given that recent-onset CVD events might affect TyG and its obesity derivatives, participants diagnosed with CVD during Wave 4 were excluded. Upon repeating all trajectory-CVD analyses in the refined cohort, the findings remained consistent with the primary results (Additional file 2: Table S15).

Sensitivity Analysis 3: After replicating the association analysis between trajectories of TyG and its obesity derivatives and CVD in the complete dataset, no evidence inconsistent with the primary analysis was detected (Additional file 2: Table S16).

Discussion

In this prospective cohort study, longitudinal trajectories of TyG in combination with both classical and novel obesity indices were significantly associated with CVD. Among the combinations with various obesity indices, the integration of TyG with the classical obesity index WC and the novel obesity index CVAI (TyG-WC and TyG-CVAI) demonstrated superior utility for CVD risk assessment. These associations were independent of sex, age, smoking/drinking status, and comorbidities (hypertension, diabetes, or kidney disease). Furthermore, the findings remained robust across multiple sensitivity analyses.

Comparative analysis with previous similar studies

Comparison and analysis of baseline TyG and its obesity derivatives in association with CVD

Extensive research has been conducted on the application of IR combined with obesity for CVD risk assessment [23–26]. According to existing studies, both the TyG index and obesity indices alone are cost-effective indicators for CVD risk [14, 19–22], whereas their combination generally further enhances the predictive capability for CVD risk [23–26]. Regarding baseline measurements combining the TyG index with classical obesity indices, extensive research in recent years generally supports that BMI, WC, and WHtR can enhance the TyG index's capacity to assess CVD risk [23–26, 46–49]. Among these combinations, TyG-WC and TyG-WHtR have been identified as more closely associated with CVD in the vast majority of studies [23–26, 46–51]. Based on the findings of the current research, after standardizing the data, both continuous and categorical variable analyses demonstrated that TyG-WC could serve as a superior tool for CVD risk assessment.

Notably, recent studies have also exploratorily analyzed the capacity of combining the TyG index with novel obesity indices for CVD risk assessment. For instance, in a recent cohort study by Zhu et al., the TyG index was combined with the BRI, revealing that TyG-BRI exhibited a stronger association with CVD than the TyG index alone [52]. Regarding the combination of the TyG index with the WWI, Duan et al. demonstrated that baseline TyG-WWI appeared unsuitable for CVD risk assessment due to the susceptibility of their association to confounding factors [53]. Similar findings were also observed in the current study, where our results indicated a borderline positive association between baseline TyG-WWI and CVD. Additionally, recent stroke-related studies have evaluated TyG index combined with other novel obesity indices (ABSI, BRI, and CVAI) [54, 55], revealing that TyG-CVAI outperforms other TyG-novel obesity indices combinations in stroke risk assessment. Collectively, previous association studies on baseline TyG-classical/novel obesity indices and CVD have provided substantial reference and comparative evidence, emphasizing the importance of simultaneously evaluating IR and obesity for CVD risk assessment. However, it is important to note that previous studies have primarily focused on combining the TyG index with a single or a limited number of obesity indices, resulting in relatively fragmented evidence that makes it challenging to determine the optimal TyG-obesity indices combination for CVD risk assessment. In the current study, building on prior research, we systematically integrated multiple combinations of the TyG index with both classical and novel obesity indices. Through standardized analysis of the results, we identified baseline TyG-WC and TyG-CVAI as the most practical and effective simple tools for CVD risk assessment.

Comparative analysis of associations between longitudinal trajectories of TyG and its obesity derivatives and CVD

In the field of longitudinal data analysis, cumulative exposure models and trajectory clustering techniques have emerged as significant breakthroughs for evaluating disease progression characteristics. By integrating follow-up measurement data from two or more time points, this methodological framework systematically uncovers the temporal cumulative effects and dynamic patterns of study parameters [31–33]. To date, several studies on the associations between cumulative exposure/clustering trajectories of TyG and its obesity derivatives with CVD have been reported. The majority of these studies analyzed single or two TyG-obesity indices combinations [56–60], while only a limited number simultaneously examined multiple TyG-classical/novel obesity indices combinations [61, 62]. Based on analysis results from Zhu et al. and Chu et al. using cumulative exposure/controlled trajectories of multiple TyG-obesity indices [61, 62], repeated measurements of TyG-WC have been identified as the most significant metabolic combination for CVD risk assessment. This finding was also supported in the current study. Compared to the studies by Zhu et al. and Chu et al. [61, 62], our analysis employed a dual validation mechanism (including standardized ORs and comparative subgroup analyses) to validate that TyG-WC, among classical obesity indices combinations, is the most applicable for CVD risk assessment. In contrast to Chu et al.'s study [62], the current research further incorporated TyG-CVAI and evaluated the association between clustering trajectories of TyG and its obesity derivatives and CVD. Following more comprehensive model adjustments and integration of cumulative exposure and clustering trajectory analyses, we observed that the longitudinal trajectories of TyG-WWI and TyG-ABSI exhibited borderline positive associations with CVD. This finding suggests that TyG-WWI and TyG-ABSI exhibit relatively poorer stability in CVD risk assessment compared to other TyG-obesity derivatives. In contrast to Zhu et al.’s study [61], the current research further expanded the scope to evaluate associations between trajectory changes of TyG-novel obesity indices and CVD, and established the significant role of TyG-CVAI in CVD risk evaluation. It should also be noted that in numerous previous studies examining the associations between TyG and its obesity derivatives' change trajectories and CVD, a significant proportion of researchers may have inadvertently used independent variable measurement data from the same time points as outcome assessment as trajectory evaluation data [56–58, 60, 62]. In summary, compared to previous similar studies, our research employed more rigorous population selection criteria and covariate adjustment strategies, and comprehensively evaluated the impact of TyG index combined with both classical and novel obesity indices on CVD. Our findings not only reaffirmed the significant role of TyG-WC trajectory changes in CVD risk assessment, but also for the first time, identified the critical influence of TyG-CVAI trajectory changes on CVD.

Additional key findings and clinical implications of the current study

Through systematic analysis of multiple TyG-obesity derivatives, this study further demonstrated that longitudinal assessment provides superior capacity for CVD risk identification compared to single baseline assessment. While these findings have also been reported in previous similar studies [53, 60], few researchers have provided explanations for this incremental risk effect. Regarding this incremental risk effect, we propose the following clinical and methodological insights for future research reference: (1) Non-static nature of IR and obesity: It is well-established that IR and obesity are not static conditions but exhibit adaptive dynamic fluctuations in response to metabolic changes [63–67]. In summary, single-timepoint measurements can only capture transient metabolic states, failing to distinguish between temporary fluctuations and persistent pathological states. Conversely, longitudinal monitoring of IR and obesity indices enables more accurate quantification of cumulative damage to the organism. Therefore, in data analysis, this time-dependent cumulative effect—where subjects' IR status or obesity status shows progressive deterioration compared to baseline—will be correspondingly reflected in the incremental CVD risk. (2) Threshold phenomenon: Compared to single-timepoint measurements, dynamic monitoring can capture decompensation thresholds in insulin sensitivity and obesity indices. According to previous research, when insulin sensitivity and obesity indices reach certain thresholds, they significantly increase CVD risk, potentially exhibiting J-shaped or U-shaped curve associations [68–71]. Consequently, compared to single measurements, threshold phenomena detected through dynamic monitoring may accumulate CVD risk through exponential growth curves. (3) Correction of measurement bias: In scientific research, measurement errors and regression dilution bias are critical factors affecting data accuracy and result reliability. Compared to single-timepoint measurements, repeated measurements can effectively mitigate these errors and biases, thereby enhancing the credibility of research findings [72–74]. Furthermore, repeated measurements improve model predictive capacity and parameter estimation accuracy [75].

This study provides practical guidance for improving public health prevention and control strategies and clinical risk assessment. It must be noted that accurately identifying high-risk CVD patients remains a major challenge in clinical medicine [76, 77], with one key limitation being that traditional risk prediction models struggle to effectively capture the dynamic interactions between IR and obesity [78, 79]. Based on the current study and previous similar research findings [23–26, 46–62], we propose that combination indices of TyG with classical obesity index WC and novel obesity index CVAI (TyG-WC and TyG-CVAI) exhibit optimal performance for CVD risk assessment, regardless of whether evaluating static baseline values or longitudinal trajectories. In other words, for middle-aged and elderly Chinese populations, focusing on TyG-WC or TyG-CVAI provides effective CVD risk assessment whether through single-timepoint health screenings or longitudinal monitoring. From the perspective of primary prevention in the general population, both the TyG index and obesity indices possess advantages of simple accessibility and excellent reproducibility. These strengths significantly enhance result replicability, population-level applicability, and widespread implementation.

Strengths and limitations

Based on nationwide prospective cohort data in China, this study employed a combined assessment approach incorporating both static baseline assessments and longitudinal trajectory analyses, marking the first comprehensive and systematic investigation of associations between TyG combined with both classical and novel obesity indices and incident CVD. These findings provide a relatively accurate strategy for TyG-combined obesity indices in CVD prevention tailored to middle-aged and elderly Chinese populations. The results underscore the clinical importance of incorporating longitudinal trajectories of TyG and its obesity derivatives (particularly TyG-WC and TyG-CVAI) into monitoring systems for CVD risk assessment and stratified management.

Several limitations should be acknowledged, with recommendations for targeted improvements in subsequent research: (1) While this study completed reasonable cumulative exposure assessment and cluster trajectory analysis under existing data conditions, further enhancing research value could involve incorporating more frequent repeated measurements in future work to systematically dissect the associations between detailed dynamic trajectory fluctuations of TyG and its obesity derivatives and CVD. (2) This study utilized CHARLS registry questionnaire data, where CVD diagnosis relied on self-reported physician-diagnosed information from participants. This data collection method may introduce recall bias and potential misclassification of cases [80]. However, Sabanayagam et al.'s recent validation study demonstrated a generally low false-positive rate for self-reported CVD, suggesting limited impact of potential misclassification bias [81]. (3) Although the CHARLS study has a prospective design strength, its observational nature [82] precludes active intervention design for the study population, which prevents this research from exploring the impact of interventions on associations between TyG and its obesity derivatives and CVD risk; Furthermore, due to the observational nature of the study, a causal relationship cannot be inferred. (4) CVD has a significant multifactorial nature; although current statistical analyses have adjusted for some confounding factors, the findings may still be influenced by residual confounding [83]. (5) Existing evidence primarily holds reference value for middle-aged and elderly populations. Given the persistently rising incidence of CVD among younger groups [84], conducting targeted studies in younger populations is particularly essential. (6) Current findings are exclusively derived from Chinese population data; it is necessary to validate the generalizability of these results across other countries in future research.

Conclusion

In this cohort study, the longitudinal trajectories of TyG and its obesity derivatives were closely associated with CVD. Comparatively, the combinations of TyG with classical obesity index WC and novel obesity index CVAI (TyG-WC and TyG-CVAI) exhibited superior performance for CVD risk assessment, with this risk primarily driven by obesity and TyG. These findings underscore the critical importance of simultaneously evaluating both TyG index and obesity for primary CVD prevention in middle-aged and elderly populations.

Abbreviations

TyG

Triglyceride glucose

IR

Insulin resistance

CVD

Cardiovascular disease

WC

Waist circumference

BMI

Body mass index

WHtR

Waist-to-height ratio

WWI

Weight adjusted waist index

BRI

Body roundness index

ABSI

A body shape index

CVAI

Chinese visceral adiposity index

HbA1c

Haemoglobin A1c

WQS

Weighted quantile sum

LDL-C

Low-density lipoprotein-cholesterol

TG

Triglycerides

Cr

Creatinine

ORs

Odds ratios

FRS

Framingham Risk Score

Author contributions

YZ, WW and XL: Conceptualization, methodology, supervision, and project administration. CW, SM-H and GB-X: writing-original draft preparation. SH-Z, ZY-X, LX, HC-L, QW, WW, YZ and XL: writing-reviewing and editing. YZ and SM-H: Software. YZ, WW and XL: formal analysis and validation. YZ and GB-X: data curation. All authors read and approved the final manuscript.

Funding

This work was supported by the National Natural Science Foundation of China [82460078, 82460091, 81670370 and 82360073]; the Natural Science Foundation of Jiangxi Province [20232BAB216004, 20224ACB206004 and 20224BAB216015]; and the Science and technology project of Health Commission of Jiangxi Province [202410005, 202410011 and 202510148].

Data availability

CHARLS datasets are available for download at the CHARLS home website (http://charls.pku.edu.cn/en).

Declarations

Ethics approval and consent to participate

The CHARLS protocol was approved by the Ethical Review Committee of Peking University (IRB: 00001052–11015). All participants provided written informed consent prior to enrollment. This study strictly adhered to the ethical principles outlined in the Declaration of Helsinki, and the reporting of results complied with the Strengthening the Reporting of Observational Studies in Epidemiology guidelines.

Consent for publication

Not applicable.

Competing interests

The authors declare no competing interests.