Novel lipid and obesity indices with new-onset stroke among adults aged 50 years and older in England

Abstract

Few studies investigated the association of newly occurring indexes of obesity and dyslipidemia including atherogenic index of plasma (AIP), body roundness index (BRI), waist-to-height ratio (WHtR) with new-onset stroke in the British population among older adults. We utilized data from Wave 6 of the ELSA (the English Longitudinal Study of Ageing), with an initial pool of 4033 participants. New incidence of stroke was captured from self-report questionnaires at Waves 7 to 9 (2014–2019). For comparability, the continuous indices (AIP, BRI, and WHtR) were converted to z-scores. Correlation analysis and multivariate logistic regression were performed on these indices. Generalized additive models (GAMs) and subgroup analyses were used to identify linear relationships. 107 subjects had experienced stroke (2.7% of total subjects). Greater AIP-Z, BRI-Z, and WHtR-Z were each independently associated with an increased risk of stroke [OR 1.42 (95% CI 1.16 to 1.74) for AIP-Z; OR 1.32 (1.09 to 1.59) for BRI-Z; OR 1.33 (95% CI 1.09 to 1.63) for WHtR-Z]. Quartiles were shown in trends between AIP-Z (P-trend = 0.003), BRI-Z (P-trend = 0.032) and WHtR-Z (P-trend = 0.032). GAMs confirmed linear associations between these indices and stroke risk. The associations were more pronounced in participants aged < 65 years and in females, as well as in smokers, drinkers, and those without diabetes or hypertension. Results are robust under sensitivity analysis. In this observational study of English adults aged ≥ 50 years, the newly identified parameters AIP, BRI and WHtR showed positive linear associations with new-onset stroke risk, with a more prominent epidemiological association observed in younger elderly individuals and females. These findings provide preliminary epidemiological insights into the link between such anthropometric and metabolic parameters and stroke risk, which may inform future epidemiological studies.

Supplementary Information

The online version contains supplementary material available at 10.1038/s41598-026-41151-9.

Introduction

Stroke serves as a pivotal agent in a prominent public health concern across the globe. It represents a primary contributor to deaths and economic hardship, and it is especially common in populations who are getting older and have higher metabolic risks¹. Stroke is a principal cause of worldwide fatalities, leading to a worsening of cognitive decline, functional impairments, and need on long-term care, as shown in recent global assessments². Atherosclerotic processes culminate in ischemic or hemorrhagic symptoms; modifiable factors, such as central obesity and dyslipidemia, accelerate these processes³. In Western nations, despite advances in acute management such as reperfusion therapies, global disparities in access to thrombolysis and thrombectomy-present even in high-income countries-highlight that treatment improvements alone cannot curb stroke incidence amid escalating metabolic burdens, necessitating the need for innovative risk stratification to reduce costs and enhance survivorship quality. Traditional risk models may fail to capture the subtle yet clinically relevant interactions between lipid profiles and adiposity, which recent studies indicate are crucial to stroke susceptibility^4,5. Moreover, population-specific disparities driven by socioeconomic, ethnic, and geographic factors further emphasize the need for refined risk assessment³. Emerging evidence indicates that composite lipid-associated indexes, including the atherogenic index of plasma(AIP), provide enhanced predictive value for stroke beyond conventional measures^4,6. To lessen the increasing worldwide burden of stroke, integrated biomarkers could be useful in clinical practice by allowing for the early detection of individuals at high risk and the facilitation of proactive, individualized preventative efforts⁶.

Emerging lipid-derived indices, such as AIP—computed using the log of the ratio between triglycerides and HDL-cholesterol—serve as effective indicators of atherogenic dyslipidemia, reflecting small, dense low-density lipoprotein (sdLDL) subfractions that penetrate the arterial intima more readily and contribute to atherogenesis compared to isolated lipid measures^7–9. Recent prospective investigations affirm AIP’s superiority in prognosticating cerebrovascular risks, including stroke recurrence and severity, by reflecting systemic inflammation and endothelial dysfunction¹⁰. Paralleling this, anthropometric proxies like the Body Roundness Index (BRI), which incorporates height, weight, and waist circumference, quantify visceral fat accumulation¹¹. Similarly, the Waist-to-height ratio (WHtR) excels in ascertaining central adiposity across diverse ethnicities, correlating strongly with cardiometabolic risk factors and showing superior or comparable predictive ability to BMI and waist circumference in meta-analyses of cardiometabolic outcomes¹². Combining AIP with anthropometric indices may provide complementary insights into metabolic health, potentially enhancing risk prediction models beyond traditional assessments¹³.

Despite the potential of AIP, BRI and WHtR as predictive biomarkers for cardiometabolic and cerebrovascular risks, longitudinal studies examining their associations with stroke incidence in aging populations like those in the English Longitudinal Study of Ageing (ELSA) remain absent^4,7,13–17. Since there is a lack of research on how these integrative indices can improve stroke risk stratification, this research aims to address that information shortfall by exploring the connections among these indicators and stroke occurrence in ELSA participants aged 50 and older, while accounting for key confounding factors. This underscores the need to carry out large-scale cohort studies for exploring their predictive value, making use of ELSA’s strong framework to guide focused prevention efforts and tackle the growing worldwide burden of stroke.

Methods

Study population and data acquisition

ELSA data are publicly available; ethical approval was secured from the original study’s oversight body, the London Multicenter Research Ethics Committee (11/SC/ 0374).

ELSA gathers data from a variety of sources, including interviews, questionnaires, and clinical assessments, from English individuals living in the community who are 50 years old and older. The study is conducted every two years¹⁸. Using information from the UK Data Service, this study set exposures at Wave 6 (2012–2013) and used Waves 7–9 (2014–2019) to follow up on stroke outcomes prospectively. All subjects with full exposure and covariate data who were 50 years old or older were considered. The following were removed from the original pool of 10,601 Wave 6 participants and a total of 4033 participants remained ultimately. Figure 1 shows the method for selecting participants, including the criteria for inclusion and exclusion as well as patterns of attrition (Fig. 1).

Fig. 1.

Research flowchart. Note: Participant selection flowchart for the ELSA cohort. A total of 10,601 participants were initially enrolled at Wave 6 of ELSA.

Exposures

AIP was designated as the primary exposure variable for this study, while BRI and WHtR were defined as secondary exposure variables. The calculation formulas for each index were as follows^9,11:

All indices were calculated based on standardized measurements from Wave 6 (2012–2013), including fasting blood sample collection for lipid profile testing and anthropometric assessments, all performed by trained professional nurses. To comprehensively explore the associations of these indices with incident stroke, both the primary and secondary exposure variables were analyzed as continuous variables; additionally, each index was categorized into quartiles for stratified analysis to further elucidate potential dose-response relationships.

Stroke

Stroke was assessed as new-onset stroke in this study. New-onset stroke was identified through self-reported in ELSA Waves 7–9, among participants with no stroke history at Wave 6 baseline, coded as a binary outcome (yes/no) based on ELSA’s standardized health questionnaire^19,20.

Covariates

Covariates were chosen based on their well-documented links to stroke risk, gathered through standardized interviews and questionnaires from ELSA’s Wave 6. Age was recorded continuously and categorized (under 65 years and 65 years or older) for stratification. Gender was coded as male or female. Education level was dichotomized as higher (college or above) versus lower (high school diploma or below). Marital status was grouped into married and non-married. Smoking and alcohol use were each classified as current or non-current based on participants’ self-reported behavior. Diabetes and hypertension were also divided into two groups: Yes and No. All covariates were derived from ELSA’s structured assessments by trained interviewers, ensuring data reliability^18,21–23.

Statistical analysis

Baseline characteristics of 4033 participants were examined based on their stroke status (new onset stroke versus no stroke). To compare continuous variables, the t-test was used, while chi square test was used for categorical variable comparison. To eliminate collinearity and scale differences, each index was included in an independent model and standardized via the z-score^24,25. We implemented a multivariate logistic regression model with three adjustment levels to evaluate the relationship between exposure (AIP-Z, BRI-Z, and WHtR-Z) and the occurrence of new stroke. The initial adjustment level was not adjusted, while the moderate level considered age, gender, education level, marital status, smoking and alcohol consumption²⁴. Odds ratio (OR) and 95% confidence interval (CI) were subsequently calculated. Dose-response relationships were evaluated through quartile analyses of the standardized indices, with P for trend calculated accordingly. For the multiple parallel association analyses of AIP‑Z, BRI‑Z, and WHtR‑Z scores with the study outcome, the Benjamini–Hochberg (BH) method was used for false discovery rate (FDR) correction to control the inflation of Type I error caused by multiple testing. The potential effect changes were evaluated by age, gender, educational achievement, marital status, smoking, drinking, diabetes and hypertension. In order to capture potential nonlinear correlations, a generalized additive model (GAM) using binomial distribution and logit link function was utilized, providing a complex approach to analyze relationships that may not be purely linear. Sensitivity analysis evaluated robustness by excluding participants lacking covariates, ensuring consistency in the results of different models. There were two ways to fill in missing data: mode filling and category filling. Two sets of repeated regression models were performed to further evaluate robustness. All statistical analyses were conducted using R software (version 4.5.1) and EmpowerStats (version 4.2), following a two tailed P-value threshold of less than 0.05 to determine statistical significance²⁶. To assess the multicollinearity among variables, the Variance Inflation Factor (VIF) analysis was performed for the three core exposure variables (AIP-Z score, BRI-Z score, WHtR-Z score) and their covariates in different regression models. Model 1 was excluded as it contained only one independent variable, failing to meet multicollinearity analysis conditions. VIF < 2 was adopted as the strict judgment criterion to determine the absence of severe multicollinearity, ensuring the reliability of regression coefficient estimation^27,28. Pearson correlation coefficients were calculated to assess the strength of association among AIP-Z, BRI-Z, and WHtR-Z scores. In accordance with established literature, correlation coefficients between 0.3 and 0.5 were defined as moderate, while values exceeding 0.5 were considered high²⁹.

Results

Baseline characteristics

Descriptive analysis included 3,926 non-stroke participants and 107 stroke patients (Table 1). Stroke patients demonstrated significantly elevated metabolic scores: AIP-Z (0.34 ± 1.06 vs. -0.01 ± 1.00), BRI-Z (0.37 ± 1.06 vs. -0.01 ± 1.00), and WHtR-Z (0.37 ± 1.03 vs. -0.01 ± 1.00; all P < 0.001). They were also older (71 ± 9 vs. 65 ± 8 years, P < 0.001), had lower education levels (67.3% vs. 54.1% with high school or below, P = 0.007), and were less likely to consume alcohol (85.0% vs. 91.0%, P = 0.035). Hypertension prevalence was marginally higher in stroke patients (45.8% vs. 37.4%, P = 0.077). No significant differences were found for gender, marital status, smoking, or diabetes (all P > 0.05).

Table 1.

Characteristics of the participants at baseline.

Characteristic	Stroke		p-value
	No N = 3,926	Yes N = 107
AIP-Z score, Mean ± SD	-0.01 ± 1.00	0.34 ± 1.06	< 0.001
BRI-Z score, Mean ± SD	-0.01 ± 1.00	0.37 ± 1.06	< 0.001
WHtR-Z score, Mean ± SD	-0.01 ± 1.00	0.37 ± 1.03	< 0.001
Gender, n (%)			0.143
Man	1,738 (44.3%)	55 (51.4%)
Woman	2,188 (55.7%)	52 (48.6%)
Education, n (%)			0.007
High school and below	2,123 (54.1%)	72 (67.3%)
Above high school	1,803 (45.9%)	35 (32.7%)
Age, Mean ± SD	65 ± 8	71 ± 9	< 0.001
Marital status, n (%)			0.185
Married	2,730 (69.5%)	68 (63.6%)
Non-married	1,196 (30.5%)	39 (36.4%)
Smoke, n (%)			0.113
No	1,583 (40.3%)	35 (32.7%)
Yes	2,343 (59.7%)	72 (67.3%)
Drink, n (%)			0.035
No	353 (9.0%)	16 (15.0%)
Yes	3,573 (91.0%)	91 (85.0%)
Diabetes, n (%)			0.307
No	3,627 (92.4%)	96 (89.7%)
Yes	299 (7.6%)	11 (10.3%)
Hypertension, n (%)			0.077
No	2,458 (62.6%)	58 (54.2%)
Yes	1,468 (37.4%)	49 (45.8%)

Note: Data are presented as mean ± standard deviation (continuous variables) or n (percentage) (categorical variables). Continuous variables were compared using independent samples t-test, and categorical variables using χ²test. P < 0.05 was considered statistically significant.

Abbreviations: AIP-Z = standardized atherogenic index of plasma; BRI-Z = standardized body roundness index; WHtR-Z = standardized waist-to-height ratio; SD = standard deviation.

Models associations between obesity-related indices and stroke risk

The associations between AIP-Z, BRI-Z and WHtR-Z with stroke risk were assessed using logistic regression models (Table 2). And the unadjusted model included no covariates. Adjust I controlled for age and gender, while Adjust II further adjusted for age, sex, educational background, marital status, smoking, drinking, diabetes and hypertension.

Table 2.

Association between AIP-Z/BRI-Z/WHTR-Z score and Stroke.

Exposure	Non-adjusted	Adjust I	Adjust II
AIP-Z score (continuous)	1.40 (1.16, 1.68) < 0.001	1.45 (1.19, 1.77) <0.001	1.42 (1.16, 1.74) < 0.001
AIP-Z score
Q1	1	1	1
Q2	0.97 (0.49, 1.89) 0.923	0.89 (0.45, 1.74) 0.725	0.87 (0.44, 1.70) 0.676
Q3	1.82 (0.99, 3.32) 0.053	1.66 (0.90, 3.05) 0.104	1.60 (0.87, 2.96) 0.131
Q4	2.21 (1.23, 3.98) 0.008	2.17 (1.19, 3.95) 0.012	2.02 (1.10, 3.74) 0.024
P for trend	< 0.001	0.001	0.003
BRI-Z score (continuous)	1.39 (1.17, 1.64) <0.001	1.35 (1.12 1.62) 0.001	1.32 (1.09, 1.59) 0.005
BRI-Z score
Q1	1	1	1
Q2	1.48 (0.80, 2.76) 0.215	1.21 (0.64, 2.28) 0.553	1.22 (0.64, 2.30) 0.545
Q3	1.18 (0.61, 2.27) 0.619	0.82 (0.42, 1.61) 0.572	0.81 (0.41, 1.58) 0.533
Q4	2.72 (1.55, 4.79) <0.001	2.08 (1.17, 3.71) 0.012	1.98 (1.09, 3.59) 0.025
P for trend	<0.001	0.013	0.032
WHtR-Z score (continuous)	1.42 (1.19, 1.70) < 0.001	1.37 (1.12, 1.66) 0.002	1.33 (1.09, 1.63) 0.006
WHtR-Z score
Q1	1	1	1
Q2	1.48 (0.80, 2.76) 0.215	1.21 (0.64, 2.28) 0.553	1.22 (0.64, 2.30) 0.545
Q3	1.18 (0.61, 2.27) 0.619	0.82 (0.42, 1.61) 0.572	0.81 (0.41, 1.58) 0.533
Q4	2.72 (1.55, 4.79) 0.001	2.08 (1.17, 3.71) 0.012	1.98 (1.09, 3.59) 0.025
P for trend	<0.001	0.013	0.032

Note: Data are presented as odds ratio (OR) (95% CI) and P value. Non-adjusted model: no covariates adjusted. Adjust I: adjusted for age and gender. Adjust II: adjusted for age, gender, education level, marital status, smoking, drinking, diabetes, and hypertension. Quartiles (Q1-Q4) were divided equally by sample size. P for trend indicates the significance of dose-response relationships.

Logistic regression analysis revealed significant positive associations between all three metabolic indices and stroke risk. Clear dose-response relationships were observed, with the highest quartile (Q4) showing approximately 2-fold increased stroke risk compared to the lowest quartile (Q1) across all indices (AIP-Z: OR = 2.02; BRI-Z: OR = 1.98; WHtR-Z: OR = 1.98, all P < 0.05). These associations remained robust after progressive adjustment for demographic and lifestyle factors. The E-value results are 2.19, 1.97, and 1.99, respectively. Sensitivity analysis (Table S1) and E-value results show that the results are robust. After FDR correction using the BH method for multiple testing, the associations of continuous AIP-Z, BRI-Z, and WHtR-Z scores with the outcome remained statistically significant. The P-values for the corresponding trend tests also remained significant after correction, indicating that the core findings of the study remained robust after accounting for multiple testing (Table S5).

Threshold effect analyses and smooth curve fitting

GAMs shows that there is a linear association between AIP-Z, BRI-Z, WHtR-Z and the risk of stroke (Fig. 2 and Table 3).

Fig. 2.

Smooth curve fitting between AIP-Z/BRI-Z/WHTR-Z score and Stroke. Note: Smooth curve fitting of the association between standardized indices and new-onset stroke risk (generalized additive models, GAMs). (A) AIP-Z (standardized atherogenic index of plasma); (B) BRI-Z (standardized body roundness index); (C) WHtR-Z (standardized waist-to-height ratio). The x-axis represents the standardized z-score of each index; the y-axis represents the log-odds of new-onset stroke. Shaded areas indicate 95% CIs. GAMs confirmed no threshold effect, supporting linear associations between each index and new-onset stroke risk.

Table 3.

Threshold effects between AIP-Z/BRI-Z/WHtR-Z score and stroke.

Exposure:	AIP-Z score	BRI-Z score	WHtR-Z score
Model I
Fitting by the standard linear model	1.40 (1.16, 1.68) 0.0003	1.39 (1.17, 1.64) 0.0001	1.42 (1.19, 1.70) 0.0001
Model II
Inflection point (K)	-1.16	1.51	1.51
Segment effect 1 for < K	0.89 (0.21, 3.68) 0.8701	1.60 (1.25, 2.04) 0.0002	1.57 (1.24, 1.99) 0.0002
Segment effect 2 for > K	1.43 (1.17, 1.75) 0.0004	0.89 (0.47, 1.68) 0.7211	0.90 (0.40, 2.02) 0.7909
Log likelihood ratio	0.550	0.108	0.210

Note: Data are presented as OR (95% CI) and P value. Model I: standard linear model; Model II: piecewise linear model. Inflection point (K): threshold value; Segment effect 1/2: effect sizes below/above K. Log likelihood ratio compares the fit of piecewise and linear models.

Sensitivity analyses

An exploratory sensitivity analysis was performed. The results showed that excluding missing covariates yielded consistent odds ratios (AIP-Z: OR 1.42; BRI-Z: 1.31; WHtR-Z: 1.32; all P < 0.05) (Table S1) The VIF analysis results (Tables S2-S4) showed that the VIF values of all core indicators and covariates in Model 2 and Model 3 were all less than 2, ranging from 1.0004 to 1.1560.No serious multicollinearity was observed among variables, which met the requirements of subsequent regression analysis and ensured the interpretability of the effect estimation results of each indicator. AIP-Z scores were moderately correlated with BRI-Z (r = 0.432, P < 0.01) and WHtR-Z (r = 0.471, P < 0.01) (Table S6). These findings confirmed the absence of severe multicollinearity.

Subgroup analyses

Exploratory subgroup analyses stratified by age, sex, educational background, marital status, smoking, drinking, diabetes and hypertension were performed to assess associations of AIP-Z, BRI-Z, and WHtR-Z with stroke risk (Fig. 3). Interaction tests were conducted to evaluate potential effect modification across strata.

Fig. 3.

Subgroup analysis. Note: Subgroup analyses of the association between standardized indices and new-onset stroke risk. (A) AIP-Z; (B) BRI-Z; (C) WHtR-Z. Subgroups were stratified by age (< 65 years vs. ≥ 65 years), gender (male vs. female), education level (high school or below vs. above high school), marital status (married vs. non-married), smoking status (current vs. non-current), alcohol consumption (current vs. non-current), diabetes (present vs. absent), and HBP(hypertension, present vs. absent). Data are presented as ORs and 95% CIs. P for interaction indicates the statistical significance of heterogeneity in associations across subgroups.

Exploratory stratified analysis suggested potential differences in the associations of AIP-Z, BRI-Z, and WHtR-Z with stroke risk across subgroups. For age stratification, the associations of all three indices were statistically significant in participants aged < 65 years (AIP-Z: OR = 1.51, P < 0.05; BRI-Z: OR = 1.67, P < 0.05; WHtR-Z: OR = 1.77, P < 0.05), whereas the associations were relatively weaker in those aged ≥ 65 years. In terms of gender, the associations were statistically significant in females (AIP-Z: OR = 1.61, P < 0.05; BRI-Z: OR = 1.42, P < 0.05; WHtR-Z: OR = 1.45, P < 0.05); in males, only BRI-Z and WHtR-Z were significantly associated with stroke risk (both P = 0.026). Regarding education level, AIP-Z was significantly associated with stroke risk in participants with high school education or below (OR = 1.40, P = 0.004), while WHtR-Z showed a significant association in those with college education or above (OR = 1.56, P = 0.007). For marital status, married participants had significant associations of BRI-Z (OR = 1.39, P = 0.003) and WHtR-Z (OR = 1.43, P = 0.003) with stroke risk. In terms of lifestyle habits, smokers and drinkers showed significant associations of all three indices with stroke risk (OR range: 1.34–1.47, all P < 0.05). For comorbidities, significant positive associations of all three indices with stroke risk were observed in participants without diabetes or hypertension (OR range: 1.35–1.49, all P < 0.05), while no significant associations were detected in those with these comorbidities. These subgroup findings are exploratory and should be interpreted with caution due to limited event counts in some strata. Among the subgroup interaction tests for AIP, BRI, and WHtR, only the age‑stratified interaction for WHtR reached statistical significance (P for interaction = 0.044), while all other interaction P‑values were > 0.05.

Discussion

The present study, based on data from ELSA—a nationally representative cohort of English adults aged 50 years and older—identifies significant positive epidemiological associations between standardized AIP-Z, BRI-Z, and WHtR-Z and the risk of new-onset stroke. These associations exhibited clear dose-response relationships, with participants in the highest quartile having an approximately twofold higher risk than those in the lowest quartiles (ORs: 1.98–2.02). GAMs further confirmed linear relationships for all three indices, and all study findings remained consistent following multivariable adjustment and sensitivity analyses.

The positive association with AIP extends prior ELSA-based research by Li et al. (2025)³⁰, which linked AIP to cardiovascular disease, to stroke outcomes, emphasizing AIP’s role in atherogenesis. However, while some studies in Chinese populations report nonlinear or threshold effects for AIP-stroke associations (Liu et al., 2025), our GAM analyses revealed strictly linear relationships³¹. This linearity may reflect ethnic differences in lipid metabolism between European-descended older adults and Asian populations. To further validate this linear trend, we performed threshold (piecewise) regression and conducted log-likelihood ratio tests. The results (P > 0.05) aligning with our findings from the GAM. This methodological approach is further supported by established principles of statistical power analysis, ensuring that our model complexity remains appropriate for the number of observed events³². While potential inflection points were identified, they should be interpreted as exploratory descriptive indicators rather than definitive clinical diagnostic cut-points. As demonstrated in previous epidemiological research, retaining such analyses serves as a rigorous sensitivity test to exclude the potential influence of non-linear confounding³³.

Earlier studies yielded inconsistent evidence linking BRI specifically to stroke, with some reporting null or limited associations within composite CVD outcomes³⁴. Recent longitudinal studies, including analysis of NHANES data by Gan et al. (2025) and CHARLS data by Peng et al. (2025), have demonstrated positive associations^35,36. Our findings add supportive evidence of a linear positive relationship, likely enhanced by Z-score standardization, comprehensive confounding adjustment, and dose-response detection. This supports consideration of BRI in future research on anthropometric markers related to stroke risk in aging European populations. Furthermore, these results align with Li et al. (2023), who reported superior predictive performance of WHtR compared to BMI for outcomes in ischemic stroke patients from the CNSR-III registry, reflecting the contribution of central obesity to vascular risk³⁷.

Mechanistically, AIP surrogates small dense LDL and triglyceride-rich lipoprotein impairment, accelerating atherogenesis through arterial lipid deposition and pro-thrombotic effects^38,39. BRI and WHtR effectively capture central/visceral fat accumulation, which exacerbates chronic inflammation and systemic insulin resistance through the release of proinflammatory cytokines such as IL-6 and TNF-α^40,41. These interrelated processes, in turn, promote endothelial dysfunction and dyslipidemia—key drivers of atherosclerotic plaque progression⁴². The resulting cascade ultimately leads to plaque instability and thrombosis, contributing to the incidence of ischemic stroke.

Exploratory stratified analyses suggested potential population heterogeneity, which may provide tentative insights for targeted risk profiling. Specifically, the associations of the three indices were statistically significant in relatively younger participants (< 65 years; ORs 1.51–1.77) but relatively weaker in those aged ≥ 65 years; this observation may be tentatively explained by age-related reductions in nitric oxide bioavailability and elastin degradation, which could increase susceptibility to cerebral ischemia^37,43,44. The associations were also more prominent in females (statistically significant for all three indices) compared with males (significant only for BRI-Z and WHtR-Z), which is consistent with postmenopausal central fat redistribution and hormonal effects on cerebral vasculature⁴⁵. Additionally, the associations were more pronounced in smokers and drinkers, likely reflecting synergistic oxidative stress and inflammatory effects^46,47. Notably, the associations were significant primarily in participants without diabetes or hypertension, which tentatively suggests that these indices may show stronger associations with stroke risk in metabolically healthier subgroups. It should be emphasized that all these subgroup-related interpretations are exploratory and preliminary; due to the limited number of stroke events in some subgroups, the findings require cautious interpretation and further validation in larger cohorts.

Strengths and limitations

This research exhibits various strengths. First, the use of longitudinal data from the ELSA cohort allowed for the assessment of temporal associations; GAMs and threshold effect analyses provided nuanced insights into how these indices correlate with stroke risk across age and gender subgroups. The representativeness of middle-aged and older British adults provides a valuable epidemiological basis for understanding risk patterns in aging Western populations. Finally, the incorporation of novel indices reveals significant longitudinal associations with stroke incidence. Rather than providing tools for clinical risk stratification or immediate prevention strategies, these findings suggest that AIP, BRI, and WHtR may serve as complementary epidemiological indicators. Future research is needed to determine whether these indices could supplement existing assessment approaches in broader population-based studies, while their direct clinical application should be interpreted with caution given the exploratory nature of the current findings⁴⁸.

However, certain limitations should be acknowledged. Being an observational study, the possibility of residual confounding persists, and critical variables such as diabetes, hypertension, dietary habits, and genetic influences were not included. Covariates were measured only at baseline, limiting assessment of temporal changes. Self-reported stroke outcomes may be affected by recall bias or misclassification due to the lack of objective confirmation. The relatively low event rate also reduces statistical power for subgroup analyses. Furthermore, the findings pertain specifically to British populations, necessitating validation within more ethnically diverse groups.

Conclusion

In the ELSA cohort of English adults aged ≥ 50 years, AIP, BRI, and WHtR showed independent, linear positive associations with new-onset stroke risk, with clear dose-response patterns and robustness across adjustments and sensitivity analyses. Subgroup analyses suggested potential variations in these associations, particularly among participants aged < 65 years and females. Overall, these findings highlight longitudinal epidemiological links between these indices and increased stroke risk in the study population. Future studies are needed to validate these preliminary findings across diverse ethnic groups and further explore their potential utility in long-term risk assessment.

Supplementary Information

Below is the link to the electronic supplementary material.

Abbreviations

AIP

Atherogenic index of plasma

BH

Benjamini–Hochberg

BRI

Body roundness index

CI

Confidence interval

ELSA

the English Longitudinal Study of Ageing

FDR

false discovery rate

GAM

Generalized additive model

HR

Hazard ratio

OR

Odds ratio

WHtR

Waist-to-height ratio

Author contributions

XRY conceived and designed the study. ZYX, HY, XRY conducted the research. XRY, ZYX wrote the manuscript. HY participated in funding acquisition. All authors read and approved the final manuscript.

Funding

This work was supported by the National Natural Science Foundation of China (82505302); the Natural Science Foundation of Sichuan Province (No. 2024NSFSC1861); the Technology Innovation R&D Project of Chengdu Science and Technology Bureau (No. 2024-YF05-00521-SN);Sichuan Province Cadres Health Research Project (2026-109).

Data availability

The datasets analyzed during the current study are publicly available from the UK Data Service repository under the ELSA dataset (DOI: 10.5255/UKDA-SN-5050-1). All data generated or analyzed are included in this published article and its supplementary information files. Raw data can be accessed upon registration with the UK Data Service.

Competing interests

The authors declare no competing interests.

Ethics approval and consent to participate

The English Longitudinal Study of Ageing (ELSA) was approved by the London Multicentre Research Ethics Committee (MREC/01/2/91; 11/SC/0374). All participants provided informed consent for the original ELSA study. This secondary analysis used de-identified, publicly available data from the UK Data Service and does not constitute human subjects research, thus no additional ethical approval was required by the Institutional Review Board of Chengdu University of Traditional Chinese Medicine. The study adhered to the Declaration of Helsinki and STROBE guidelines for cohort studies.