Joint association of body mass index, abdominal obesity, and triglyceride-glucose index with stroke risk: insights from the China health and retirement longitudinal study

Abstract

Objective

Individuals who are obesity or abdominal obesity (AO) often develop insulin resistance. This study aims to explore the separate and combined effects of body mass index (BMI), AO [assessed by waist circumference (WC)], and triglyceride-glucose (TyG) index on the incidence risk of stroke.

Methods

We analyzed 8746 participants aged 45 years or older from the China Health and Retirement Longitudinal Study (CHARLS). The primary study outcome was stroke incidence. Cox proportional hazard models were applied to explore the association between the combination of BMI, AO, and TyG index with the risk of stroke.

Results

During a median follow-up of 9 years, 812 (9.28%) participants had occurred strokes. Restricted cubic splines curves revealed a significant linear relationship between BMI, WC, and TyG index with the risk of stroke, the HRs were 1.24 (95% CI: 1.02, 1.51) for BMI, 1.38 (95% CI: 1.16, 1.64) for AO, and 1.25 (95% CI: 1.04, 1.49) for TyG index. Cox regression analysis revealed that participants with BMI ≥ 28 or AO and higher levels of TyG have higher risk of stroke, the HRs were 1.48 (95% CI: 1.13, 1.93) for BMI ≥ 28 and TyG ≥ median group, 1.75 (95% CI: 1.36, 2.25) for AO and TyG ≥ median group. Compared with individuals with no biomarker elevations, the HRs for incident stroke were 1.51 (95% CI: 1.21, 1.89), 1.72 (95% CI: 1.34, 2.21), and 1.99 (95% CI: 1.45, 2.72) for those with one, two, or three biomarkers in the high-value group, respectively. Kaplan Meier curve demonstrates that individuals with three risk factors have the highest risk of stroke. The TyG index partially mediated the effects of BMI and AO on stroke events by 6.6% and 6.9%, respectively.

Conclusions

The separate and combined effects of BMI, AO, and TyG index were significantly associated with the risk of stroke, with TyG index partially mediating the impact of BMI and AO on stroke events. The findings highlighted the importance of joint evaluation of BMI, AO, and TyG index for primary prevention of stroke.

Supplementary Information

The online version contains supplementary material available at 10.1186/s12883-025-04558-x.

Introduction

The global healthcare landscape faces a mounting challenge from stroke-related mortality and disability [1, 2]. China exemplifies this trend, with stroke incidence escalating by 86% between 1990 and 2019, reaching 3.94 million new cases [2]. Despite significant advancements in prevention and treatment strategies, the incidence of stroke persists in increasing, which emphasizing the critical need for enhanced preventive strategies through modifiable risk factor identification [3, 4].

Metabolic dysfunction, particularly obesity, significantly influences cerebrovascular outcomes [5–7]. While body mass index (BMI) traditionally gauges obesity, it inadequately captures fat distribution patterns [8]. Therefore, the Lancet recommends not only evaluating obesity through BMI but also incorporating at least one additional body size measurement to better capture abdominal obesity (AO, defined by waist circumference) and fat distribution [9]. Waist circumference (WC) is a straightforward and clinically applicable method for evaluating AO. The American Heart Association proposes that the AO assessed by WC is an independent cardiovascular disease risk marker independent of BMI [10]. Evidence suggests AO independently predicts cardiovascular risk, even in individuals with normal BMI [11–13]. These metabolic alterations frequently coincide with insulin resistance, efficiently measured through the triglyceride-glucose (TyG) index [14, 15]. Recent research indicates that BMI, AO, and TyG are all independent predictors of stroke occurrence [16–22]. However, the association between the combination of BMI, AO, and TyG index with the risk of stroke also remains unclear.

To address these critical research gaps, we analyzed the data of the China Health and Retirement Longitudinal Study (CHARLS), aiming to explore the complex the association between the combination of BMI, AO, and TyG index with stroke risk. This study may provide valuable insights into improving stroke risk stratification and identifying new therapeutic targets for prevention strategies.

Methods

Study design

This investigation leverages data from CHARLS, a large-scale longitudinal health surveillance program monitoring middle-aged and elderly Chinese residents. The study framework encompasses five waves of data collection (2011–2020), covering 10,000 households and 17,708 participants across 450 communities nationwide. Standard questionnaires are implemented every 2 to 3 years to collect essential health information from participants. Following established research protocols and ethical guidelines approved by Peking University, participants provided informed consent for comprehensive health assessments [23].

Study population

From the initial cohort of 17,708 individuals, we applied specific inclusion criteria to establish our study population. After excluding cases with incomplete metabolic parameters (n = 7,956), age-related exclusions (n = 289), and those with pre-existing stroke, insufficient stroke data, or follow-up loss (n = 717). Consequently, 8962 participants were excluded, resulting in a final cohort of 8746 participants for this study (Fig. 1).

Fig. 1.

Flow chart of the study population

Exposure

The calculations for BMI and TyG index were performed as follows: (1) BMI = weight_(kg)/height_(m)²; (2) TyG index = ln[TG_(mg/dl) × FPG_(mg/dl)/2]. TG represents triglycerides, and FPG represents fasting plasma glucose. According to the latest definition of AO in China, the WC thresholds for diagnosing AO in men and women are ≥ 85 cm and ≥ 80 cm, respectively [24]. According to the criteria set by the Working Group on Obesity in China (WGOC), obesity is defined as BMI ≥ 28.0 kg/m² [25].

Assessment of incident stroke

The primary study outcome was stroke incidence, which was assessed by the following questions:” Have you ever been diagnosed with stroke by a physician?“. Participants were monitored across five successive survey waves beginning in 2011, with follow-up continuing until either a stroke was documented or the end of the 2020 survey cycle, depending on which occurred first [26].

Assessments of covariates

Trained research personnel systematically gathered initial participant data through standardized assessment protocols. The collected data includes gender, age, residence, education level, marital status, smoking and drinking status, height, weight, WC, systolic blood pressure (SBP), diastolic blood pressure (DBP), hypertension, diabetes, dyslipidemia, heart disease, hypertension medication, dyslipidemia medication, and diabetes medication. In addition, the laboratory test data includes FPG, total cholesterol (TC), TG, high-density lipoprotein cholesterol (HDL-C), low-density lipoprotein cholesterol (LDL-C), and hemoglobin A1c (HbA1c).

Statistical analysis

Table S1 provides details of the proportion of missing data. To mitigate the influence of missing values and reduce potential bias, multiple imputation was applied. Continuous variables with a normal distribution are expressed as means with standard errors, and intergroup differences were assessed using analysis of variance (ANOVA). For non-normally distributed data, medians and interquartile ranges are reported, with comparisons performed using the Kruskal-Wallis test. Categorical variables are described as frequencies and percentages, and differences were examined with the Chi-square test.

We constructed three analytical models with progressive covariate adjustment: unadjusted (Model 1); demographic and clinical factors (Model 2: gender, age, smoking status, drinking status, hypertension, diabetes, dyslipidemia, and heart disease); and comprehensive adjustment (Model 3: adding residence, marital status, and education level, hypertension medications, diabetes medications, dyslipidemia medications, TC, HDL-C, and LDL-C). Since there is no established clinical cut-off for the TyG index, we used its median (8.6) as the threshold, with values above the median considered elevated. According to BMI, WC, and TyG index, participants were divided into low-value (BMI < 28, Non-AO, and TyG index < median) and high-value groups (BMI ≥ 28, AO, and TyG index ≥ median). For the primary analysis, HRs for stroke events were calculated by comparing high-value groups against low-value groups as the reference. Restricted cubic spline (RCS) models were employed to characterize potential non-linear associations of BMI, WC, and TyG index with stroke risk. Subsequently, we explored the combined effects of BMI and TyG index, AO and TyG index on stroke risk, respectively. We additionally examined joint effects of BMI and TyG index, AO and TyG index, and the combined influence of all three markers by classifying participants into groups with zero, one, two, or three elevated indicators. Kaplan-Meier survival curves and log-rank tests were applied to evaluate stroke incidence.

TyG index’s potential mediating role in the relationship between BMI or AO with stroke risk was further analyzed. We employed a quasi-Bayesian mediation analysis with 100 simulations using the ‘mediation’ package. The survival outcome was modeled using the survreg method under a Weibull distribution assumption. The C-statistics, net reclassification improvement (NRI) and integrated discrimination improvement (IDI) were calculated to gauge the enhanced predictive accuracy beyond the basic model [27–29].

Subgroup analyses were conducted according to sex (men vs. women), age (45–60 vs. ≥60 years), and diabetes status (yes vs. no). For exploratory post hoc analyses, the Bonferroni correction was applied to account for multiple testing, and a P value < 0.017 (0.05/3 subgroups) was considered statistically significant. To test the robustness of our findings, several sensitivity analyses were performed. We excluded participants with missing data, those who were non-fasting, and individuals using antidiabetic or lipid-lowering drugs. In addition, E-values were calculated from Model 3 to assess the possible impact of unmeasured confounders [30]. All statistical analyses were conducted using R version 4.2.2, with P-values < 0.05, which was considered statistically significant.

Results

Baseline characteristics of participants

A total of 8,746 participants from CHARLS were included, among whom 812 (9.28%) developed stroke during a median follow-up of 9 years. The mean age was 59.33 ± 9.22 years, including 4693 (53.7%) men. Compared with those without stroke, affected individuals were typically older, more often current or former smokers and drinkers, and more frequently presented with hypertension, diabetes, dyslipidemia, or heart disease. They also reported greater use of medications for these conditions. Moreover, the stroke group demonstrated higher WC, BMI, SBP, DBP, FPG, TC, TG, LDL-C, HbA1c, and TyG index, while HDL-C levels were lower. A detailed summary of demographic and clinical features is presented in Table 1..

Table 1.

Baseline characteristics of participants with and without stroke

Characteristic	Overall	Non-Stroke	Stroke	p-value
	N = 8,746	N = 7,934	N = 812
Sex				0.473
Men	4,693 (53.7%)	4,267 (53.8%)	426 (52.5%)
Women	4,053 (46.3%)	3,667 (46.2%)	386 (47.5%)
Age, year	59.33 ± 9.22	59.13 ± 9.27	61.34 ± 8.44	<0.001
Residence				0.810
Rural	3,004 (34.3%)	2,722 (34.3%)	282 (34.7%)
Urban	5,742 (65.7%)	5,212 (65.7%)	530 (65.3%)
Marital status				0.042
Marred	7,680 (87.8%)	6,985 (88.0%)	695 (85.6%)
Other	1,066 (12.2%)	949 (12.0%)	117 (14.4%)
Education level				0.283
No formal education	4,222 (48.3%)	3,830 (48.3%)	392 (48.3%)
Primary school	1,923 (22.0%)	1,726 (21.8%)	197 (24.3%)
Middle school	1,742 (19.9%)	1,590 (20.0%)	152 (18.7%)
High school or above	859 (9.8%)	788 (9.9%)	71 (8.7%)
Smoking status				0.001
Never	5,364 (61.3%)	4,899 (61.7%)	465 (57.3%)
Former	753 (8.6%)	657 (8.3%)	96 (11.8%)
Current	2,629 (30.1%)	2,378 (30.0%)	251 (30.9%)
Drinking status				0.008
Never	5,353 (61.2%)	4,879 (61.5%)	474 (58.4%)
Former	716 (8.2%)	627 (7.9%)	89 (11.0%)
Current	2,677 (30.6%)	2,428 (30.6%)	249 (30.7%)
Hypertension	3,565 (40.8%)	3,075 (38.8%)	490 (60.3%)	<0.001
Diabetes	1,484 (17.0%)	1,308 (16.5%)	176 (21.7%)	<0.001
Dyslipidemia	4,232 (48.4%)	3,755 (47.3%)	477 (58.7%)	<0.001
Heart disease	1,008 (11.5%)	848 (10.7%)	160 (19.7%)	<0.001
Hypertension medications	1,646 (18.8%)	1,365 (17.2%)	281 (34.6%)	<0.001
Diabetes medications	323 (3.7%)	266 (3.4%)	57 (7.0%)	<0.001
Dyslipidemia medications	410 (4.7%)	336 (4.2%)	74 (9.1%)	<0.001
Height, m	1.58 ± 0.09	1.58 ± 0.08	1.58 ± 0.09	0.526
Weight, kg	58.71 ± 11.31	58.44 ± 11.16	61.34 ± 12.40	<0.001
Waist circumference, cm	85.27 ± 9.97	84.95 ± 9.89	88.38 ± 10.20	<0.001
BMI, kg/m2	23.48 ± 3.75	23.38 ± 3.71	24.45 ± 4.00	<0.001
SBP, mmhg	130.54 ± 21.31	129.75 ± 20.92	138.30 ± 23.43	<0.001
DBP, mmhg	75.80 ± 12.05	75.45 ± 11.92	79.24 ± 12.76	<0.001
FPG, mg/dl	102.60 (94.50, 113.58)	102.24 (94.32, 113.04)	104.58 (96.12, 117.95)	<0.001
TC, mg/dl	194.06 ± 38.75	193.71 ± 38.78	197.53 ± 38.29	0.007
TG, mg/dL	106.20 (74.34, 153.99)	104.43 (74.34, 152.22)	117.71 (84.96, 168.37)	<0.001
HDL‑C, mg/dl	51.30 ± 15.35	51.56 ± 15.38	48.75 ± 14.82	<0.001
LDL‑C, mg/dl	116.51 ± 35.69	116.15 ± 35.65	120.03 ± 35.86	0.003
HbA1c, %	5.10 (4.90, 5.40)	5.10 (4.90, 5.40)	5.20 (4.90, 5.52)	<0.001
TyG index	8.69 ± 0.68	8.67 ± 0.67	8.84 ± 0.69	<0.001

BMI Body mass index, DBP Diastolic blood pressure, FPG Fasting plasma glucose, HbA1c Hemoglobin A1c, HDL-C High density lipoprotein cholesterol, LDL-C Low-density lipoprotein cholesterol, SBP Systolic blood pressure, TC Total cholesterol, TG Triglycerides, TyG index Triglyceride-glucose index

Associations of BMI, AO, and TyG index with stroke incidence

Table 2 summarizes the associations of BMI, AO, and TyG index with stroke events. Using Chinese population–specific definitions, after adjustments for confounding variables, participants with obesity (BMI ≥ 28.0 kg/m²) or AO (WC ≥ 85 cm for men and ≥ 80 cm for women) had significantly higher stroke risk compared with those without these conditions. Specifically, compared with the group of BMI < 28, the adjusted HRs for BMI ≥ 28 group were 1.24 (95% CI: 1.02, 1.51). Additionally, by comparing AO of with non-AO, the adjusted HRs were 1.38 (95% CI: 1.16, 1.64) for stroke events. Furthermore, compared with the group of TyG index < median, the adjusted HRs for TyG index ≥ median group were 1.25 (95% CI: 1.04, 1.49). Additionally, RCS analysis confirmed linear relationships between these metabolic indicators and stroke risk (all P-value < 0.05, and all P for nonlinear > 0.05) (Fig. 2).

Table 2.

Incidence risk of stroke stratified by BMI, AO, and TyG index

Characteristic	Event, n		Model 1			Model 2			Model 3
		HR	95%CI	p-value	HR	95%CI	p-value	HR	95%CI	p-value
Total	812
BMI
BMI < 28	675	Ref			Ref			Ref
BMI ≥ 28	137	1.61	1.34, 1.94	< 0.001	1.31	1.08, 1.59	0.007	1.24	1.02, 1.51	0.033
AO
No	234	Ref			Ref			Ref
Yes	578	1.72	1.47, 2.00	< 0.001	1.44	1.22, 1.70	< 0.001	1.38	1.16, 1.64	< 0.001
TyG index
TyG < median	321	Ref			Ref			Ref
TyG ≥ median	491	1.58	1.37, 1.81	< 0.001	1.26	1.06, 1.48	0.007	1.25	1.04, 1.49	0.015

HR  Hazard Ratio, CI  Confidence Interval, BMI Body mass index, HDL-C High density lipoprotein cholesterol, LDL-C Low-density lipoprotein cholesterol, TC Total cholesterol, TyG index Triglyceride-glucose index

Abdominal obesity (AO), men, waist circumference ≥ 85 cm, women, waist circumference ≥ 80 cm

Model 1: unadjusted for any covariates

Model 2: adjusted for gender, age, smoking status, drinking status, hypertension, diabetes, dyslipidemia, and heart disease

Model 3: adjusted for gender, age, residence, marital status, education level, smoking status, drinking status, hypertension, diabetes, dyslipidemia, heart disease, hypertension medications, diabetes medications, dyslipidemia medications, TC, HDL-C, and LDL-C

Fig. 2.

Restricted cubic spline curves for stroke risk by BMI (A), waist circumference (B), and TyG index (C) Adjusted for gender, age, residence, marital status, education level, smoking status, drinking status, hypertension, diabetes, dyslipidemia, heart disease, hypertension medications, diabetes medications, dyslipidemia medications, TC, HDL-C, and LDL-C

Associations of BMI and TyG index, AO, and TyG index with incident stroke

As shown in Table 3, the combined influence of BMI and TyG index, as well as AO and TyG index, was evaluated with Cox regression. Participants with both elevated BMI (≥ 28) and high TyG index (≥ median) had greater risk of stroke (HR: 1.48, 95% CI: 1.13, 1.93) compared with those with low levels of both. Additionally, compared to individuals with Non-AO and TyG < median, participants with AO and TyG ≥ median had the higher risk of stroke (HR: 1.75, 95% CI: 1.36, 2.25).

Table 3.

Risk of stroke upon coexposure stratifed by the BMI, AO, and TyG index

Characteristic	Event, n			Model 1				Model 2			Model3
		HR	95%CI	p-value	HR		95%CI		p-value	HR	95%CI	p-value
BMI and TyG index
BMI < 28 and TyG < median	291	Ref				Ref				Ref
BMI < 28 and TyG ≥ median	30	1.52	1.30, 1.77	< 0.001		1.25	1.05, 1.49	0.012		1.25	1.04, 1.51	0.017
BMI ≥ 28 and TyG < median	384	1.52	1.04, 2.21	0.030		1.37	0.93, 2.00	0.109		1.27	0.87, 1.87	0.216
BMI ≥ 28 and TyG ≥ median	107	2.20	1.76, 2.75	< 0.001		1.58	1.23, 2.04	< 0.001		1.48	1.13, 1.93	0.004
AO and TyG index
Non-AO and TyG < median	130	Ref				Ref				Ref
Non-AO and TyG ≥ median	104	1.58	1.22, 2.04	< 0.001		1.42	1.08, 1.86	0.011		1.45	1.10, 1.91	0.008
AO and TyG < median	191	1.70	1.36, 2.12	< 0.001		1.59	1.26, 2.00	< 0.001		1.55	1.23, 1.96	< 0.001
AO and TyG ≥ median	387	2.28	1.87, 2.79	< 0.001		1.78	1.41, 2.24	< 0.001		1.75	1.36, 2.25	< 0.001

HR  Hazard Ratio, CI  Confidence Interval, BMI Body mass index, HDL-C High density lipoprotein cholesterol, LDL-C Low-density lipoprotein cholesterol, TC Total cholesterol, TyG index Triglyceride-glucose index

Abdominal obesity (AO): men, waist circumference ≥ 85 cm; women, waist circumference ≥ 80 cm

Model 1: unadjusted for any covariates

Model 2: adjusted for gender, age, smoking status, drinking status, hypertension, diabetes, dyslipidemia, and heart disease

Model 3: adjusted for gender, age, residence, marital status, education level, smoking status, drinking status, hypertension, diabetes, dyslipidemia, heart disease, hypertension medications, diabetes medications, dyslipidemia medications, TC, HDL-C, and LDL-C

The combined impact of BMI, AO, and TyG index on stroke risk

Figure 3 shows the combined effect of BMI, AO, and TyG index on the cumulative incidence for stroke events. The research results indicate that high levels of BMI, AO, and TyG index contributed independently to increased stroke risk, while the combination of all three conferred the greatest risk. Multivariable-adjusted HRs were 1.51 (95% CI: 1.21,1.89) for one elevated biomarker, 1.72 (95% CI: 1.34, 2.21) for two, and 1.99 (95% CI: 1.45, 2.72) for three, compared with participants with no elevated biomarkers (Table 4).

Fig. 3.

Combined effect of BMI, AO, and TyG index on the cumulative incidence for stroke events

Table 4.

Hazard ratios for stroke events for combined effects of BMI, AO, and TyG index

Characteristic	Event, n	Model 1				Model 2			Model 3
		HR	95%CI	p-value	HR	95%CI	p-value	HR	95%CI	p-value
BMI, AO, and TyG index
0 biomarkers at risk	129	Ref			Ref			Ref
1 biomarker at risk	266	1.62	1.32, 2.00	< 0.001	1.52	1.22, 1.89	< 0.001	1.51	1.21, 1.89	< 0.001
2 biomarkers at risk	311	2.13	1.73, 2.61	< 0.001	1.74	1.38, 2.20	< 0.001	1.72	1.34, 2.21	< 0.001
3 biomarkers at risk	106	2.81	2.17, 3.63	< 0.001	2.11	1.57, 2.83	< 0.001	1.99	1.45, 2.72	< 0.001

HR  Hazard Ratio, CI  Confidence Interval, BMI Body mass index, HDL-C High density lipoprotein cholesterol, LDL-C Low-density lipoprotein cholesterol, TC Total cholesterol, TyG index Triglyceride-glucose index

Abdominal obesity (AO): men, waist circumference ≥ 85 cm; women, waist circumference ≥ 80 cm

Model 1: unadjusted for any covariates

Model 2: adjusted for gender, age, smoking status, drinking status, hypertension, diabetes, dyslipidemia, and heart disease

Model 3: adjusted for gender, age, residence, marital status, education level, smoking status, drinking status, hypertension, diabetes, dyslipidemia, heart disease, hypertension medications, diabetes medications, dyslipidemia medications, TC, HDL-C, and LDL-C

Mediation analyses and incremental predictive

Mediation analyses, as presented in Table 5, indicated that the association between BMI with incident stroke was partially mediated by TyG index, with mediation proportions of 6.6% (P-value < 0.05). Moreover, TyG index accounted for 6.9% of the mediating effect between AO and stroke risk (P-value < 0.05) (Table 5). These results indicate that BMI and AO may indirectly contribute to stroke risk through their effects on TyG index. To further evaluate the added predictive value of BMI, AO, and TyG index, Harrell’s C-statistics, calibration, and reclassification metrics were assessed. The C-statistic increased from 0.641 (95% CI: 0.624,0.658) for the basic model to 0.669 (95% CI: 0.653,0.685). Compared with the basic model, the combined model improved classification with NRI = 0.185 (95% CI: 0.114,0.256) and IDI = 0.003(95% CI: 0.001,0.004) (all P < 0.001) (Table S2).

Table 5.

Mediation analyses of stroke risk factors

Exposures	Mediator	Total effect		Indirect effect		Direct effect		Proportion mediated, % (95% CI)
		Coefficient (95% CI)	P value	Coefficient (95% CI)	P value	Coefficient (95% CI)	P value
TyG	BMI	−69.36 (−226.23, −11.27)	< 0.001	−3.27(−9.64, −0.70)	< 0.001	−66.09(−215.88, −9.97)	< 0.001	4.9 (2.3, 11.9)
TyG	AO	−69.78(−242.05, −11.39)	< 0.001	−4.21 (−11.68, −0.94)	< 0.001	−65.57 (−228.00, −10.13)	< 0.001	6.5 (2.9, 14.4)
BMI	TyG	−1.34 (−2.33, −0.57)	< 0.001	−0.09 (−0.17, −0.03)	< 0.001	−1.26(−2.23, −0.52)	< 0.001	6.6 (2.4, 11.9)
AO	TyG	−0.73(−1.36, −0.29)	< 0.001	−0.05 (−0.10, −0.02)	< 0.001	−0.68 (−1.29, −0.27)	< 0.001	6.9 (2.4, 12.9)

HR  Hazard Ratio, CI  Confidence Interval, Abdominal obesity (AO): men, waist circumference ≥ 85 cm; women, waist circumference ≥ 80 cm; BMI Body mass index, HDL-C High density lipoprotein cholesterol, LDL-C Low-density lipoprotein cholesterol, TC Total cholesterol, TyG index Triglyceride-glucose index

Adjusted for gender, age, residence, marital status, education level, smoking status, drinking status, hypertension, diabetes, dyslipidemia, heart disease, hypertension medications, diabetes medications, dyslipidemia medications, TC, HDL-C, and LDL-C

Subgroup analysis

Subgroup analyses were limited to the three prespecified factors—age, sex, and diabetes status. The findings indicated that the combined impact of these three factors on stroke occurrence remained stable across various subgroups, including non-diabetes patients, regardless of sex and age (Table 6). After Bonferroni correction (P < 0.017), there was no significant interaction observed between the combination of BMI, AO, and TyG index and stroke risk in all subgroups (All P for interaction > 0.017).

Table 6.

Subgroups analysis of the joint association of BMI, AO, and TyG index with stroke incidence

Characteristic	Case/total	0 biomarkers at risk	1 biomarker at risk	2 biomarkers at risk	3 biomarkers at risk	P for interaction
Sex						0.200
Men	426/4693	Ref	1.48 (1.04,2.09)	1.48 (1.04,2.09)	1.48 (1.04,2.09)
Women	386/4053	Ref	1.46 (1.09,1.96)	1.71 (1.23,2.39)	2.06 (1.3,3.25)
Age						0.500
45–60	351/4780	Ref	1.34 (0.95,1.89)	1.64 (1.13,2.39)	1.98 (1.25,3.13)
≥ 60	461/3966	Ref	1.69 (1.26,2.26)	1.82 (1.31,2.53)	1.92 (1.24,2.97)
Diabetes						0.545
No	636/7262	Ref	1.58 (1.25,1.99)	1.69 (1.29,2.22)	1.87 (1.29,2.69)
Yes	176/1484	Ref	1.23 (0.51,3.01)	2.02 (0.83,4.92)	2.46 (0.96,6.33)

HR  Hazard Ratio, CI  Confidence Interval. Abdominal obesity (AO): men, waist circumference ≥ 85 cm; women, waist circumference ≥ 80 cm, BMI Body mass index, HDL-C High density lipoprotein cholesterol, LDL-C Low-density lipoprotein cholesterol, TC Total cholesterol, TyG index Triglyceride-glucose index

Adjusted for gender, age, residence, marital status, education level, smoking status, drinking status, hypertension, diabetes, dyslipidemia, heart disease, hypertension medications, diabetes medications, dyslipidemia medications, and BMI

Sensitivity analysis

We conducted a series of sensitivity tests to confirm the robustness of the results. The exclusion of participants with incomplete information did not alter the associations (Table S3). Similarly, removing non-fasting subjects or those on antidiabetic or lipid-lowering therapy yielded comparable outcomes (Tables S4 and S5). In addition, the E-value derived for the combined indicators under Model 3 was 3.39, indicating that only a confounder of considerable magnitude could nullify the observed relationships.

Discussion

This nationwide prospective cohort study sought to elucidate the joint effects of three critical variables, the BMI, AO, and TyG index, on the stroke incidence. The results indicate that each factor was independently associated with higher stroke risk, and their joint elevation further amplified this risk, with the highest stroke risk observed when all three biomarkers were at higher levels. Additionally, mediation analysis indicates that the TyG index plays a significant mediating role in the relationship between BMI and AO and stroke risk. Taken together, these findings suggest that integrating BMI, AO, and TyG index may improve the precision of stroke risk stratification.

This study applied the Chinese population specific criteria recommended by the Chinese Obesity Working Group (WGOC) to define general obesity and AO. These thresholds can better capture the metabolic and cardiovascular risks of Asian adults Emphasizing these definitions related to culture and physiology can ensure that our research results accurately reflect the risk patterns associated with obesity in the Chinese population, and enhance the translational value of our research results for stroke prevention strategies in specific regions. Our findings support a clear association between elevated TyG index and stroke risk, extending evidence from earlier prospective investigations. Both the Kailuan study and the Rural Chinese cohort demonstrated that higher TyG index values independently predict future stroke [31, 32]. Consistently, we found that increased BMI was significantly associated with stroke risk, confirming earlier reports linking adiposity to cerebrovascular events [33, 34]. Since BMI is a limited proxy for adipose distribution, we further assessed AO through WC, which showed a significant association with stroke incidence. Furthermore, linear relationships emerged between these three markers and stroke occurrence, which indicate that effective management of BMI, AO, and TyG index is critical for reducing the risk of stroke.

Although these single biomarkers can predict the incidence of stroke, but the predictive accuracy is still limited. Recent evidence suggests that using a combination of multiple biomarkers can provide a more comprehensive evaluation compared to a single biomarker [35–38]. Our analysis demonstrated that the combination of two of the three biomarkers demonstrates a higher correlation with stroke risk than reliance on any single factor alone. Specifically, compared to individuals with BMI < 28 and TyG < median, participants with BMI ≥ 28 and TyG ≥ median had the higher risk of stroke. Additionally, compared to individuals with Non-AO and TyG < median, participants with AO and TyG ≥ median had the higher risk of stroke. To capture the combined contribution of BMI, AO, and TyG index, participants were divided according to whether zero, one, two, or three biomarkers were in the elevated category, with those without any considered the reference. After adjusting for potential confounding variables, the highest risk was observed in those with three biomarkers simultaneously elevated. This was followed by participants with 2 biomarkers at risk, and subsequently by those with 1 biomarker at risk. The research results indicate that the BMI, AO, and TyG index contributed independently to increased stroke risk, and their simultaneous elevation markedly amplifies risk. Unlike the study by Zhang et al., which examined the combined effects of the TyG index and the a body shape index (ABSI) on stroke [39], our research comprehensively assessed three essential and complementary metabolic dimensions: general obesity (measured by BMI), abdominal obesity (assessed through waist circumference), and insulin resistance (evaluated using the TyG index). This approach highlights the importance of a thorough metabolic risk analysis. Given the simplicity and practical applicability of BMI and waist circumference, our proposed model offers enhanced public health and clinical utility, making it more suitable for stroke prevention and primary healthcare settings in real-world scenarios. Our findings align in part with those reported by Liang W et al., who also used CHARLS data to examine the combined effects of the TyG index and obesity measures on stroke risk [40]. Both studies consistently demonstrated that higher TyG levels and excess adiposity increase the likelihood of stroke events, reinforcing the link between metabolic dysfunction and cerebrovascular outcomes. However, our study extends their work by simultaneously integrating three interrelated markers—BMI (general obesity), AO (reflecting central fat distribution), and TyG index (a surrogate of insulin resistance)—to capture a broader metabolic profile. We further evaluated cumulative and mediating effects and conducted extensive sensitivity and subgroup analyses, providing a more comprehensive understanding of metabolic contributions to stroke risk. Although these studies share a conceptual foundation, our multidimensional approach adds incremental evidence, while the similarity in conclusions underscores the reproducibility and robustness of the observed associations. We acknowledge that these studies analyzed overlapping populations of CHARLS and therefore have certain limitations. Future research using external or multi-ethnic cohorts is necessary to validate the generalizability of these findings beyond the Chinese population.

BMI and WC are established anthropometric measures utilized to evaluate both general and AO. BMI, AO, and TyG index are all associated with metabolic abnormalities, IR, and an increased risk of stroke [41–46]. Consequently, it is reasonable to consider the potential interactions among these three factors. In our study, the TyG index explained 6.6% of the relationship between BMI and stroke risk and 6.9% of the relationship between AO and stroke risk. These findings indicate that both BMI and AO may contribute to stroke susceptibility, at least partly, through pathways involving the TyG index.

To further explore the relationship between these three biomarkers and stroke risk, subgroup analysis was carried out. Interestingly, this link was present in participants without diabetes but was not significant in those with diabetes, regardless of sex and age. The observed association was significant in participants without diabetes but not in those with diabetes, which can be attributed to several factors. Firstly, the link may not adequately capture the individual differences present among patients with diabetes, leading to a more pronounced relationship between additional risk markers (BMI, AO, and TyG index) and stroke prediction in non-diabetic individuals [47]. Secondly, patients with diabetes often experience chronic insulin resistance and metabolic disorders, which may saturate the risk pathways and make the incremental effects of these additional markers less apparent [7, 48, 49]. Moreover, individuals with diabetes frequently engage in active risk management through treatments that target diabetes, as well as lipid-lowering and antihypertensive therapies. Such interventions may dilute the observed association between diabetes and stroke risk. Lastly, it is important to note that blood glucose control varies significantly among patients with diabetes. The pathological features associated with poorly controlled diabetes can differ markedly from those seen in well-controlled diabetes, potentially obscuring any correlation with stroke risk [50, 51]. Our sensitivity analyses confirmed the combined impact of these three biomarkers with stroke incidence. This association remained significant even after excluding participants with missing data, removing all non-fasting participants, and excluding those who used antidiabetic or lipid-lowering medication. These findings emphasize the need for combined assessment of these three biomarkers as primary prevention for stroke events in clinical practice for different populations.

This study represents a pioneering effort to integrate the BMI, AO, and TyG index, and the exact mechanism is still unclear, but there are several potential explanations. Firstly, the combined index analysis of BMI, AO, and TyG index provides more comprehensive information. BMI can reveal the general obesity, TyG index can be utilized as a biological indicator for evaluating IR, while AO supplements the information on fat distribution. The integration of these three indicators provides a more comprehensive understanding of stroke risk. Additionally, relying solely on BMI, AO, and TyG index can only capture a single aspect of potential pathological changes, while combined indicator analysis can capture differences in multiple indicators and predict stroke earlier. Moreover, there are differences in levels of biomarkers among individuals, BMI, AO, and TyG index provide insights into obesity and IR from different perspectives. Furthermore, the observed association between an elevated TyG index, and increased stroke risk may be biologically plausible through several interrelated mechanisms. Elevated triglycerides are a hallmark of insulin resistance and atherogenic dyslipidemia, both of which contribute to endothelial dysfunction and reduced nitric oxide bioavailability, leading to vascular damage [52–54]. Triglyceride-rich lipoproteins can infiltrate the arterial intima, undergo oxidation and glycation, and subsequently trigger pro-inflammatory signaling cascades involving monocyte adhesion, foam cell formation, and smooth muscle cell proliferation [55–59]. These processes accelerate atherosclerotic plaque growth and destabilization, increasing the likelihood of plaque rupture and thromboembolism. Moreover, insulin resistance—reflected by a higher TyG index—is associated with oxidative stress, chronic low-grade inflammation, and a prothrombotic state, all of which amplify cerebrovascular vulnerability [60–66]. Collectively, these mechanisms provide a coherent biological framework linking elevated triglycerides and metabolic dysfunction to the heightened stroke risk observed in this study.

Our findings expand current knowledge by demonstrating how combined physiological markers contribute to cerebrovascular risk assessment. With the global surge in stroke incidence, particularly among individuals with adiposity-related metabolic alterations, this research offers crucial insights into prevention strategies [67–71]. Our comprehensive effects analysis provides evidence that the mechanism of stroke involves multiple pathways, therefore targeted prevention strategies may need to vary among individuals.

Our investigation offers several key strengths in advancing stroke risk assessment. Firstly, it marks the first-time combination of the BMI, AO, and TyG index for predicting stroke risk. Secondly, the prospective, nationwide longitudinal design with substantial participant enrollment strengthens result reliability. Thirdly, comprehensive subgroup analyses across diverse populations offer valuable clinical insights. Additionally, the utilization of readily accessible clinical markers enhances practical applicability. Finally, a key strength of the present study is the use of a combined index that integrates BMI, AO, and the TyG index. Each of these components reflects a different but interrelated aspect of metabolic health—BMI captures general adiposity, AO represents fat distribution and central obesity, and the TyG index serves as a biochemical surrogate of insulin resistance. By combining these measures, the index provides a more comprehensive and multidimensional picture of metabolic dysfunction than any single parameter. This holistic approach enables identification of individuals with overlapping metabolic abnormalities who might otherwise be overlooked when assessed by one indicator alone. Moreover, such integration may allow earlier detection of stroke susceptibility, as it accounts for both anthropometric and biochemical pathways contributing to cerebrovascular risk.

However, certain limitations merit acknowledgment. Firstly, our analysis is based on baseline data for the BMI, AO, and TyG index, which limits our ability to evaluate how changes in these indices over time may affect stroke risk. Given the 9-year median follow-up, these metabolic parameters may have changed substantially during the study period because of lifestyle modification, pharmacologic treatment, or aging-related body composition shifts. Such temporal changes could result in exposure misclassification and potential underestimation of the true association between long-term metabolic status and stroke risk. Future studies with repeated measurements or cumulative exposure analyses are warranted to evaluate how dynamic changes or trajectories of BMI, AO, and TyG index influence cerebrovascular outcomes more accurately. Secondly, another important limitation concerns the self-reported assessment of stroke diagnosis. In CHARLS, incident stroke was identified by asking participants whether they had ever been diagnosed with stroke by a physician. Although this approach is widely used in population-based cohorts and has been shown to have good validity in both Chinese and Korean populations, reliance on self-reported information may still lead to misclassification, particularly for mild or subclinical events [72, 73]. Such nondifferential misclassification is likely to attenuate the observed associations, potentially underestimating the true effect size between metabolic indicators and stroke risk. Therefore, while our findings provide valuable evidence for risk stratification, they should be interpreted cautiously. Future studies linking cohort data with hospital records or imaging-confirmed stroke registries would help validate and refine these associations. Thirdly, another limitation of this study is that stroke events were identified based on participants’ self-reported physician diagnoses, and detailed information regarding stroke subtypes (e.g., atherosclerotic, cardioembolic, or lacunar infarction) was not available in the CHARLS dataset. Because these ischemic subtypes have distinct pathophysiological mechanisms and metabolic risk profiles, we were unable to assess whether the associations of BMI, abdominal obesity, and TyG index differ across stroke subtypes. Future studies incorporating neuroimaging and hospital-based clinical data are warranted to examine the differential impact of these metabolic indicators on specific stroke phenotypes. Fourthly, as there is no universally accepted clinical threshold for the TyG index, we used a queue specific median (8.6) to divide participants into two groups. Although this method has statistical validity in the current population, the distribution of TyG index may vary depending on race, age structure, and underlying metabolic status. Therefore, applying this median threshold to other populations, especially non Chinese or young people, may require recalibration and external validation. Future research should aim to determine the critical values of TyG index with clinical significance and specific populations, in order to improve the generalizability and comparability of the study. Additionally, while participants with non-fasting blood samples and those using antidiabetic or lipid-lowering medications were excluded, several other potential metabolic confounders could not be entirely accounted for. These include conditions such as sarcopenic obesity that may occur in neuromuscular disorders, excessive alcohol consumption, familial hypertriglyceridemia, and other rare metabolic diseases. The presence of these unmeasured factors may have introduced residual confounding or misclassification bias. Therefore, future studies should incorporate comprehensive metabolic profiling and genetic screening to better understand these influences. Lastly, the study’s focus on Chinese middle-aged and elderly populations may restrict broader demographic application.

Conclusion

Our study observed that the BMI, AO, and TyG index were valuable tools for predicting stroke occurrence, and the combined evaluation provided more consistent risk stratification. Furthermore, the TyG index partially mediates the relationship between BMI and AO on stroke risk. This research not only emphasizes the combined predictive value of these indices but also provides new insights into the underlying mechanisms, which could be valuable for future intervention strategies.

BMI, body mass index; DBP, diastolic blood pressure; FPG, fasting plasma glucose; HbA1c, hemoglobin A1c; HDL-C, high density lipoprotein cholesterol; LDL-C, low-density lipoprotein cholesterol; SBP, systolic blood pressure; TC, total cholesterol; TG, triglycerides; TyG index, triglyceride-glucose index.

Authors’ contributions

GH conceived and designed the study and wrote the main manuscript text. JH analyzed the data. JZ conducted the literature search and prepared figures. DZ performed the manuscript review. All authors reviewed and approved the final manuscript.

Funding

This work was supported by the “National Science and Technology Major Project (No. 2023ZD0503900, 2023ZD0503902)” and the “Suzhou Multicenter Clinical Research Program for Major Disease (No.DZXYJ202408)”.

Data availability

The data supporting the findings of this study are available on the CHARLS website (http://charls.pku.edu.cn/).

Declarations

Ethics approval and consent to participate

CHARLS was approved by the Institutional Review Board of Peking University (approval number: IRB00001052-11015 for the household survey and IRB00001052-11014 for blood samples), and all participants provided written informed consent.

Consent for publication

Not applicable.

Competing interests

The authors declare no competing interests.

Clinical trial number

Not applicable.