A body shape index trajectories and cognitive function among older hypertensive patients: a national cohort study

Abstract

Background

While several cross-sectional studies have suggested an association between a body shape index (ABSI) and cognitive decline, the relationship between longitudinal ABSI patterns and cognitive outcomes among older hypertensive patients remains unclear. This study aimed to investigate the association between ABSI trajectories and cognitive outcomes in this population using a longitudinal cohort design.

Methods

Data were obtained from the China Health and Retirement Longitudinal Study (CHARLS), with ABSI measurements collected in 2011, 2013, and 2015. Group-based trajectory modeling (GBTM) was used to identify the distinct ABSI trajectories. Cox proportional hazards models and linear mixed-effects models were applied to evaluate the association between ABSI trajectories and cognitive function among older hypertensive patients.

Results

A total of 1,065 older hypertensive participants were included in the study (572 males, 53.7%; 493 females, 46.3%). Three ABSI trajectory groups were identified: moderate–stable (n = 602, 56.5%), low–rapid-rising (n = 144, 13.5%), and high–slightly-increasing (n = 319, 30.0%). After adjusting for potential confounders, there was borderline evidence of effect modification by sex for incident low cognitive performance defined by a Z-score cutoff (P for interaction = 0.046). Compared with low–rapid-rising trajectory group, hazard ratio point estimates were higher in women (moderate–stable trajectory group: HR = 1.57, 95% CI 0.81–3.03; high–slightly-increasing trajectory group: HR = 1.83, 95% CI 0.94–3.58) but lower in men (moderate–stable trajectory group: HR = 0.78, 95% CI 0.48–1.27; high–slightly-increasing trajectory group: HR = 0.63, 95% CI 0.36–1.09), although sex-stratified estimates were imprecise, with confidence intervals crossing the null. In mixed-effects models of global cognitive Z-scores, baseline cognition was similar across groups; moderate–stable trajectory group showed a less steep decline than low–rapid-rising trajectory group (moderate–stable trajectory group×time: β = 0.05, 95% CI: 0.00–0.10), whereas high–slightly-increasing trajectory group was similar, with no evidence of sex modification of slopes.

Conclusions

Long-term ABSI trajectories showed heterogeneous associations with cognitive outcomes in older hypertensive patients, with potential sex-related heterogeneity for incident low cognitive performance. Defining incident events by first crossing a Z-score cutoff may capture earlier cognitive vulnerability before clinically diagnosed impairment, offering a prevention-relevant perspective on longitudinal changes in ABSI among older adults with hypertension. However, sex-specific estimates were imprecise, and replication in larger, independent cohorts is warranted.

Supplementary Information

The online version contains supplementary material available at 10.1186/s13690-026-01925-z.

Text box 1. Contributions to literature
• To our knowledge, this is among the first nationally representative longitudinal studies to link ABSI trajectories to cognitive outcomes in older adults with hypertension and to evaluate sex as a potential effect modifier.
• By evaluating both time to first crossing a Z-score cutoff and longitudinal change in global cognitive Z-scores, our study provides complementary evidence on early cognitive vulnerability in older adults with hypertension.
• If replicated, ABSI trajectory monitoring may provide a feasible marker of longitudinal body-shape changes relevant to cognitive health in hypertension care.

Introduction

By 2050, individuals aged ≥ 60 years are expected to account for 22% of the global population, nearly twice the proportion observed in 2015 (12%) [1]. Aging is associated with a range of adverse outcomes including age-related cognitive decline. Rapid deterioration of cognitive function is recognized as a key risk factor for dementia [2, 3]. It is projected that, by 2050, 153 million individuals worldwide will be living with dementia [4]. In response to this growing concern, the United Nations has designated 2021–2030 as the Decade of Healthy Aging, emphasizing cognitive health as a central priority [5]. Therefore, maintaining cognitive function and preventing dementia in older adults is an urgent scientific and public health objective. These efforts not only aim to improve the quality of life of older individuals, but also to alleviate the societal and economic burdens associated with dementia. The 2020 Lancet Commission on Dementia Prevention, Intervention, and Care reported that up to 40% of dementia cases may be preventable or delayable through the management of modifiable risk factors [6]. Consequently, identifying additional modifiable factors is critical for mitigating cognitive decline and reducing global dementia burden.

Hypertension is among the most prevalent chronic conditions in the older population [7, 8] and is now well established as a major risk factor for cognitive decline [9]. A substantial body of research indicates that hypertension contributes to cognitive decline through mechanisms such as cerebrovascular damage and the acceleration of neurodegenerative processes [10, 11]. A large-scale Chinese epidemiological study conducted between 2015 and 2018 reported that individuals with hypertension had a 1.62-fold higher risk of cognitive decline than normotensive individuals [12]. Additionally, a meta-analysis published in 2025 found that the prevalence of mild cognitive impairment among older hypertensive patients in China is 21.3% [13]. Together, these findings indicate that older adults with hypertension represent a clinically important high-risk population for cognitive decline.

Given the importance of hypertension management for preserving cognitive health, identifying additional modifiable factors within this high-risk group may help refine risk stratification and inform prevention strategies [11]. However, cognitive outcomes vary among older adults with hypertension, suggesting that additional modifiable factors may contribute to differences in the risk of cognitive decline within this population [6, 14].

Obesity, a common metabolic disorder, may play a particularly important role in hypertension-related cognitive impairment [9]. Prior studies have shown that the prevalence of obesity among hypertensive patients is more than twice that observed in normotensive individuals [15]. In China, nearly 40% of individuals with hypertension have obesity-related hypertension [16]. The synergistic effects of hypertension and obesity may exacerbate vascular damage, thereby accelerating the progression of vascular cognitive impairment in older adults with these comorbidities [9].

Importantly, age-related changes in fat distribution may differentially impact cognitive function [17]. Research suggests that abdominal obesity is a more accurate predictor of the risk of cognitive decline than generalized obesity, which is typically measured by body mass index (BMI) [17, 18]. Waist circumference (WC) is the most commonly used indicator for assessing abdominal obesity; however, its strong correlation with BMI limits its utility in capturing the unique aspects of fat distribution [19]. The body shape index (ABSI) was developed to overcome the limitations of traditional anthropometric measures. The ABSI is independent of the so-called “obesity paradox” and provides a more nuanced assessment of physical fitness. It has demonstrated superiority over other indices such as BMI, WC, and waist-to-height ratio (WHtR) in representing physical fitness [20, 21]. Although many studies have investigated the relationship between BMI and cognitive function, research on the association between ABSI and cognitive outcomes in older adults remains limited. In particular, the potential influence of ABSI on cognitive function in older individuals with hypertension has not been sufficiently explored. Moreover, existing studies have predominantly focused on cross-sectional assessments of ABSI without examining the implications of longitudinal ABSI trajectories on cognitive decline.

To address these gaps, we utilized data from the China Health and Retirement Longitudinal Study (CHARLS), focusing on older hypertensive individuals whose ABSI was measured on at least three occasions between 2011 and 2015. The present study aimed to identify distinct ABSI trajectories over time and to assess their association with cognitive function, thereby providing valuable insights for managing abdominal obesity and preventing cognitive decline in older hypertensive populations.

Methods

Study design and participants

CHARLS is a nationally representative biennial cohort study launched in 2011 targeting individuals aged 45 years and older in China, along with their spouses. Detailed descriptions of the study design have been published [22]. The Peking University Institutional Review Board approved the study, with all participants providing written informed consent.

For the present analysis, data were obtained from three CHARLS waves: 2011, 2013, and 2015. Exclusion criteria were: (1) age < 60 years, (2) missing data on ABSI and cognitive scores at baseline or during any follow-up period, (3) missing information on blood pressure measurements, depressive symptoms, and other covariates, (4) self-reported diagnosis of dementia and/or Parkinson’s disease at baseline, and (5) absence of hypertension at baseline. The sample selection process is shown in Fig. 1.

Fig. 1.

Flowchart of participant selection for the cohort study of older adults with hypertension in China, 2011–2015

Assessment of hypertension and ABSI

Blood pressure was measured using the Omron HEM-7200 sphygmomanometer. In accordance with previous studies [23, 24], hypertension was defined as meeting any of the following criteria at baseline: (1) self-reported physician-diagnosed hypertension; (2) an average systolic blood pressure ≥ 140 mmHg or diastolic blood pressure of ≥ 90 mmHg from three measurements, or (3) current use of antihypertensive medications.

Anthropometric data in CHARLS were collected following a standardized protocol. Height (Seca™ 213; 0.1 cm) and weight (Omron™ HN-286; 0.1 kg) were measured with participants barefoot and in light clothing. Waist circumference was measured in the standing position with a non-elastic tape placed horizontally at the umbilical level, directly against the skin, and recorded to the nearest 0.1 cm at the end of a normal expiration. Detailed CHARLS measurement procedures have been described elsewhere [22, 25]. ABSI was calculated using the formula established in the literature [26]:

where WC is the waist circumference (in meters), BMI is the body mass index (kg/m²), and height is measured in meters.

Cognitive function

Cognitive function was assessed across two key domains, episodic memory and executive function. The total cognitive function score ranged from 0 to 21, with higher scores indicating better cognitive performance. The detailed methodology for the cognitive assessment is provided in Supplementary Material 1. Low cognitive performance was evaluated in two steps. Step 1 involved collecting raw cognitive scores from all the available follow-up waves. These scores were standardized using the baseline mean and standard deviation to calculate Z-score [27]. For example, the mean (standard deviation) of global cognitive scores at baseline was 14.91 (1.35). Step 2 involved computing the Z-score using the following formula: . Negative Z-scores indicated low cognitive performance, i.e., performance below the baseline study-sample mean (Z < 0) [28]. This Z-score approach and the operational definition of low cognitive performance have been used in previous studies [28, 29]. Incident low cognitive performance was defined as a follow-up Z-score < 0 among participants with baseline Z ≥ 0. This operational definition is intended as a relative-to-baseline distribution outcome and does not represent a clinical diagnosis nor the clinical incidence of mild cognitive impairment or dementia. As a sensitivity analysis aligned with prior operationalizations of objective subtle cognitive decline, we evaluated an alternative cutoff (Z < − 0.5) and applied an incident design restricting to baseline Z ≥ − 0.5, with incident events defined as the first follow-up wave with Z < − 0.5 [30].

Covariates

Based on previous studies [31–33], a range of potential covariates that could confound the association between ABSI and cognitive function were included in the analysis. These covariates were categorized into two broad domains: two main categories: demographic characteristics and health-related factors. Demographic variables included age, gender (male or female), educational attainment (illiterate, primary school, middle school, high school or above), marital status (married or unmarried), and place of residence (rural or urban). Health-related factors encompass health behaviors and health conditions. Health behaviors included smoking status (yes or no) and alcohol consumption (yes or no). The health condition assessed was depression (yes or no), evaluated using the Chinese version of the 10-item Center for Epidemiologic Studies Depression Scale (CESD-10). This instrument has demonstrated good reliability and validity and is widely employed to screen for depressive symptoms [34]. In accordance with previous studies [35, 36], a CESD-10 score ≥ 10 was used as the cutoff for identifying the presence of depressive symptoms. Beyond that, the presence of chronic diseases (diabetes, stroke, dyslipidemia), medication use (anti-hypertension, anti-diabetes, anti-dyslipidemia) and baseline BP level were also involved. Diabetes mellitus was defined as a fasting blood glucose ≥ 126 mg/dL, HbA1c ≥ 6.5%, or current use of anti-diabetic therapy, or self-reported history of diabetes mellitus. Dyslipidemia was defined as a total cholesterol ≥ 240 mg/dL, current use of lipid-lowering therapy, or self-reported history of dyslipidemia. Stroke was defined as self-reported stroke diagnosed by a physician.

Statistical analysis

Baseline demographic and clinical characteristics of older hypertensive participants were summarized according to the ABSI trajectory groups. Continuous variables are presented as mean ± standard deviation (SD), while categorical variables are expressed as frequencies and percentages. Differences in baseline characteristics across ABSI trajectories were assessed using one-way analysis of variance for continuous variables and chi-square tests for categorical variables, as appropriate.

To explore the impact of ABSI trajectories on cognitive function, we identified trajectories of change in ABSI using group-based trajectory modeling (GBTM). GBTM was implemented using the “gbmt” package in R software [37]. Models specifying one to four trajectory groups were fitted using age (in years) as the timescale. Because each participant contributed three measurement occasions, the maximum identifiable polynomial degree was quadratic (d = 2) (i.e., number of time points minus one) [38]; accordingly, following standard practices, we modeled group trajectories using second-order polynomials. To avoid over-parameterization, we applied term-pruning by removing non-significant highest-order terms within each trajectory group, while also considering overall model fit and substantive plausibility [39]. The individuals were assigned to a trajectory group based on the highest posterior probability. The optimal number of trajectory groups was determined by evaluating multiple criteria: the lowest Bayesian Information Criterion (BIC), the lowest Akaike Information Criterion (AIC), an average posterior probability (AvePP) of ≥ 0.70 for each group, high entropy (≥ 0.80), odds of correct classification (OCC) > 5 and a minimum group size of at least 5% of the sample. As an additional posterior-probability diagnostic, we summarized classification uncertainty as the percentage of participants whose maximum posterior probability of trajectory-group membership (maxPP) was < 0.70 (Pct_maxpost_lt70), computed across the full sample for each fitted model; lower values indicate less between-class overlap and more confident classification [38, 40, 41]. For interpretability, trajectory labels were assigned post hoc based on relative baseline level and the direction of change over follow-up. These labels are intended as descriptive summaries rather than implying “true” immutable subtypes. Model-estimated trajectories are presented with 95% pointwise uncertainty bands. For subsequent analyses, trajectory-group membership was treated as a nominal categorical variable. To assess numerical stability, we refitted the selected three-class, pruned quadratic model using 10 sets of random starting values and confirmed identical log-likelihood, BIC, and class proportions across runs. In the pooled sample, sex differences in trajectory-group membership were evaluated using multinomial logistic regression, with classification uncertainty incorporated via posterior-probability pseudo-class draws (M = 200) and estimates pooled using Rubin’s rules.

Cox proportional hazards models were used to assess the association between ABSI trajectories and incident low cognitive performance. Hazard ratios (HRs) and corresponding 95% confidence intervals (CIs) were reported. Event time was approximated by the first follow-up wave at which the threshold was crossed (coded as 2 or 4 years since baseline). Participants who did not cross the threshold were right-censored at their last observed wave (2013 or 2015). Proportional hazards (PH) assumptions were assessed using Schoenfeld residuals. To aid interpretation of effect sizes, we additionally reported unadjusted Kaplan–Meier cumulative risk (1 − S(t)) at 2 and 4 years by ABSI trajectory group stratified by sex, together with stratum-specific sample sizes and numbers of events. Sex effect modification was formally tested by adding an ABSI trajectory group × sex interaction term and comparing nested models using likelihood-ratio tests. Sex-specific HRs were derived as simple effects from the trajectory group × sex interaction model using linear contrasts, rather than fitting separate sex-specific Cox models, to improve precision when events were sparse [42]. Three Cox regression models were fitted, using the low–rapid-rising trajectory of ABSI as the reference group. This reference was chosen a priori because it captures the dynamic pattern of primary interest (a rapid increase from a low baseline), enabling direct comparisons of more stable trajectories against this pattern. Model 1: Adjusted for demographic covariates, including age, education level, marital status, and place of residence. Model 2: Further adjusted for health-related factors, including smoking status, drinking status, and depression. Model 3: Further adjusted for vascular/metabolic factors, including diabetes, stroke, dyslipidemia, medication use (anti-hypertension, anti-diabetes, anti-dyslipidemia), and baseline systolic/diastolic blood pressure.

To capture within-person cognitive change over time, we modeled longitudinal change in global cognition as a continuous outcome using linear mixed-effects models (LMMs) with random intercepts and random slopes, as commonly applied in large-cohort longitudinal analyses [43–46]. Time was modeled as years since baseline (0, 2, and 4 for the 2011, 2013, and 2015 waves). Fixed-effects terms included ABSI trajectory group, time, the interaction between ABSI trajectory group and time, and others covariates. Sex modification was assessed using 2-way (trajectory group × sex) and 3-way (time × trajectory group × sex) interaction tests. We additionally derived adjusted 2-year and 4-year changes from baseline (Δ) from model-based marginal estimates to align with the biennial follow-up. We fitted parallel LMMs for episodic memory and executive function to examine domain-specific cognitive change.

Subgroup analyses and multiplicity. Additional subgroup analyses (age category, marital status, residence, smoking, alcohol consumption, and depressive symptoms) were conducted within each sex by adding trajectory group × subgroup interaction terms and evaluating them using LRTs. These analyses were considered exploratory; therefore, interaction P values were additionally adjusted for multiplicity using the Benjamini–Hochberg false discovery rate (FDR) procedure within each sex, and q values were reported, as commonly implemented in large-sample cohort/database studies [43, 44, 47, 48].

Incremental predictive value beyond BMI and WC. We compared ABSI trajectory groups with conventional anthropometric indices in the same analytic cohort using an identical fully adjusted covariate set (Model 3). Incremental value was assessed by likelihood-ratio tests and changes in AIC when adding ABSI trajectory groups to models including BMI or WC. Discrimination accounting for censoring was evaluated using IPCW time-dependent AUC assessed just before the end of follow-up, with between-model differences in AUC reported as ΔAUC.

Sensitivity analyses. To assess robustness, we conducted several sensitivity analyses. First, we repeated the Cox models under alternative operational choices for the threshold-based outcome: (i) using a stricter incident threshold (global cognition z-score < − 0.5) restricted to participants with baseline z-score ≥ − 0.5, and (ii) using an expanded risk set that included participants with baseline z-score < 0. Second, sex-stratified GBTM models were fitted separately in men and women to explore potential sex-related heterogeneity in trajectory structure. Third, we refit the GBTM in the full cohort (hypertensive + normotensive) using the same specification and evaluated agreement of trajectory assignments among hypertensive participants (Cohen’s κ, adjusted Rand index [ARI], AvePP, and entropy). For cross-stratum comparisons, trajectory groups were derived in the overall cohort (overall-derived classification), and the resulting group assignments were applied to both hypertensive and normotensive participants. We further examined effect modification by hypertension using ABSI trajectory group × hypertension interactions in Cox models for incident low cognitive performance and time × trajectory group × hypertension interactions in LMMs for longitudinal outcomes. Within-stratum slope contrasts (Δβ) and difference-in-differences (ΔΔβ) were derived from the fitted time × trajectory group × hypertension LMM. We also compared the distribution of overall-derived trajectory group membership between hypertensive and normotensive participants using a Pearson chi-square test, reporting Cramér’s V as an effect size. Fourth, to address potential selection bias due to complete-case analysis arising from missing exposure, outcome, or covariate data, we applied inverse-probability weighting to reweight complete cases toward the baseline eligible cohort using stabilized, truncated weights and fitted weighted Cox models with robust standard errors. Fifth, because outcome ascertainment occurred at discrete survey waves (interval-censored onset), we assessed robustness using a discrete-time hazard model with a complementary log–log link (person–period data) and an interval-censored Weibull model. Interaction terms in sensitivity models were evaluated using likelihood-ratio tests. Sixth, using the fully adjusted (Model 3) covariate set, we assessed potential age- and sex-related influences by testing trajectory group × age-group interactions, modeling continuous age with restricted cubic splines, repeating Cox models using age-/sex-standardized ABSI z scores (including winsorization and a time-varying exposure specification), and fitting LMMs with participant-specific random intercepts and slopes for time. Seventh, to mitigate sparse data in sex-by-trajectory subgroups when evaluating sex-related heterogeneity, we assessed robustness using a sensitivity analysis based on ABSI change rates (random slopes) derived from LMMs. Using the fully adjusted (Model 3) covariate set, we fitted Cox models including an ABSI slope × sex interaction term and additionally controlled baseline/mean ABSI using two specifications (model-based ABSI level vs. observed baseline ABSI2011); interaction terms were evaluated using likelihood-ratio tests. In addition, to mitigate potential reverse causality, primary survival analyses used an incident design restricting to participants with baseline global cognition Z ≥ 0; further sensitivity analyses additionally adjusted for baseline cognition and excluded early events first detected at the 2013 wave. Detailed methods are provided in the Supplement.

All analyses were performed using R version 4.4.1. Statistical significance was set at P < 0.05.

Results

Among the 1,065 participants included in the final analysis, the median age was 66.91 years, and 572 (53.7) were male. Most participants had an educational attainment of primary school or below, were married, resided in rural areas, did not drink or smoke, and reported lower levels of depressive symptoms (Table 1). A comparison of baseline characteristics between participants included and excluded is shown in Table S1 in the Supplement.

Table 1.

Baseline characteristics of older adults with hypertension by A Body Shape Index trajectory group in the China Health and Retirement Longitudinal Study, 2011–2015

Characteristics	Overall	ABSI Trajectories			P Value
		Moderate–Stable	Low–Rapid-rising	High–Slightly-increasing
Subjects, n	1065	602	144	319
ABSI	0.08 ± 0.01	0.08 ± 0.00	0.08 ± 0.02	0.09 ± 0.00	< 0.001
Age, years	66.91 ± 5.54	65.75 ± 5.00	68.33 ± 6.14	68.45 ± 5.70	< 0.001
Sex					< 0.001
Male	572 (53.7)	364 (60.5)	82 (56.9)	126 (39.5)
Female	493 (46.3)	238 (39.5)	62 (43.1)	193 (60.5)
Education					0.049
Illiterate	527 (49.5)	272 (45.2)	75 (52.1)	180 (56.4)
Primary school	325 (30.5)	193 (32.1)	45 (31.2)	87 (27.3)
Middle school	151 (14.2)	97 (16.1)	18 (12.5)	36 (11.3)
High school and above	62 (5.8)	40 (6.6)	6 (4.2)	16 (5.0)
Marital status					0.027
Married	854 (80.2)	500 (83.1)	111 (77.1)	243 (76.2)
Single	211 (19.8)	102 (16.9)	33 (22.9)	76 (23.8)
Residence					0.420
Rural	675 (63.4)	380 (63.1)	98 (68.1)	197 (61.8)
Urban	390 (36.6)	222 (36.9)	46 (31.9)	122 (38.2)
Smoking status					0.003
No	600 (56.3)	321 (53.3)	74 (51.4)	205 (64.3)
Yes	465 (43.7)	281 (46.7)	70 (48.6)	114 (35.7)
Drinking status					0.026
No	608 (57.1)	329 (54.7)	77 (53.5)	202 (63.3)
Yes	457 (42.9)	273 (45.3)	67 (46.5)	117 (36.7)
Depression					0.012
No	690 (64.8)	411 (68.3)	81 (56.2)	198 (62.1)
Yes	375 (35.2)	191 (31.7)	63 (43.8)	121 (37.9)
Diabetes					0.613
No	824 (77.4)	472 (78.4)	111 (77.1)	241 (75.5)
Yes	241 (22.6)	130 (21.6)	33 (22.9)	78 (24.5)
Dyslipidemia					0.409
No	804 (75.5)	452 (75.1)	115 (79.9)	237 (74.3)
Yes	261 (24.5)	150 (24.9)	29 (20.1)	82 (25.7)
Stroke					0.059
No	1032 (96.9)	579 (96.2)	144 (100.0)	309 (96.9)
Yes	33 (3.1)	23 (3.8)	0 (0.0)	10 (3.1)
Medication use
For hypertension	508 (47.7)	283 (47.0)	55 (38.2)	170 (53.3)	0.009
For diabetes	72 (6.8)	40 (6.6)	8 (5.6)	24 (7.5)	0.726
For dyslipidemia	90 (8.5)	45 (7.5)	10 (6.9)	35 (11.0)	0.151
SBP (mmHg)	147.24 (19.66)	146.42 (19.40)	147.40 (18.86)	148.70 (20.48)	0.243
DBP (mmHg)	79.86 (11.51)	80.02 (11.33)	78.70 (13.19)	80.08 (11.03)	0.430

Data are presented as mean ± SD or n (%), as appropriate. ABSI, A Body Shape Index; SBP, systolic blood pressure; DBP, diastolic blood pressure; SD, standard deviation

Based on GBTM, AIC and BIC decreased as the number of trajectory groups increased from one to three; the four-class model produced an inadmissible (boundary) solution with an empty class (0% membership) and non-estimable classification indices (NaN), and was excluded. The three-class model showed excellent classification quality (entropy = 0.985; AvePP = 0.993/0.994/0.993), with only 0.939% of participants having maxPP < 0.70, indicating minimal overlap between trajectory groups (Table S2; Figure S1), and OCC values were high across classes, supporting good separation. The three-class solution was reproducible across 10 refits using different random seeds, with identical log-likelihood, BIC, and class proportions (Tables S3). Within the three-class solution, a quadratic specification fit better than a linear model (ΔBIC = − 1248.84; Table S4). In the unpruned quadratic model, the quadratic term was supported for the moderate–stable trajectory group (P = 0.004) but not for the low–rapid-rising or high–slightly-increasing trajectory groups (P = 0.190 and 0.842, respectively), and non-significant quadratic terms were pruned in the final model (Table S4). Parameter estimates for the final three-class model are reported in Table S5.

The morphology of the trajectories is shown in Fig. 2. We identified three trajectories of ABSI as the best fit: (1) Moderate–Stable (n = 602, 56.5%), indicating a moderate baseline with minimal change over follow-up; (2) Low–Rapid-rising (n = 144, 13.5%), indicating a low baseline level with a steep positive slope; and (3) High–Slightly-increasing (n = 319, 30.0%), indicating a high baseline with a small upward trend. Participants classified in the low–rapid-rising ABSI trajectory group were more likely to engage in smoking and alcohol consumption and to exhibit depressive symptoms than those in the other trajectory groups (Table 1).

Fig. 2.

Mean trajectories of A Body Shape Index by increasing age among older adults with hypertension in the China Health and Retirement Longitudinal Study, 2011–2015. ABSI, A Body Shape Index. Solid lines show model-estimated mean trajectories; shaded areas indicate 95% pointwise prediction bands generated by the gbmt plotting routine (bands = TRUE, conf = 0.95). Trajectory labels were assigned post hoc based on the relative baseline level and the direction/magnitude of change over follow-up for interpretability and are descriptive rather than definitive subtypes. Trajectory groups should be interpreted as descriptive summaries of heterogeneity rather than immutable “true” subtypes

ABSI trajectory groups and incident low cognitive performance among older hypertensive participants

During follow-up, incident low cognitive performance (follow-up Z < 0 among baseline Z ≥ 0) occurred in 255/598 (42.6%) hypertensive participants and 297/625 (47.5%) normotensive participants. In fully adjusted Cox models, there was borderline evidence of an interaction between sex and ABSI trajectory group (P for interaction = 0.046; Table 2). Point estimates suggested differences in the direction of association in females versus males for moderate–stable trajectory group and high–slightly-increasing trajectory group relative to low–rapid-rising trajectory group, although most sex-specific estimates were imprecise, with confidence intervals crossing the null, and these findings should be interpreted cautiously. Within–trajectory-group sex contrasts (male vs. female) derived from the same interaction model are presented in Table S6, with the strongest contrast observed in high–slightly-increasing trajectory group.

Table 2.

Sex-specific associations of body shape index trajectory groups with incident low cognitive performance among older adults with hypertension in the China Health and Retirement Longitudinal Study, 2011–2015

Sex	Contrast	Model 1	Model 2	Model 3
		HR (95%CI)	HR (95%CI)	HR (95%CI)
Female	Moderate–stable	1.53 (0.80–2.92)	1.54 (0.80–2.96)	1.57 (0.81–3.03)
Male	Moderate–stable	0.72 (0.45–1.16)	0.73 (0.46–1.18)	0.78 (0.48–1.27)
Female	High–slightly-increasing	1.84 (0.95–3.57)	1.84 (0.95–3.58)	1.83 (0.94–3.58)
Male	High–slightly-increasing	0.58 (0.34–1.01)	0.59 (0.34–1.02)	0.63 (0.36–1.09)
	P for interaction (LRT)	0.028	0.029	0.046

Hazard ratios (HRs) and 95% confidence intervals (CIs) were estimated using Cox proportional hazards models. Low–rapid-rising trajectory group was the reference category. Model 1: adjusted for demographic factors including age, education level, marital status, and residence. Model 2: further adjusted for health-related factors including smoking status, drinking status, and depression. Model 3: further adjusted for vascular/metabolic factors including diabetes, dyslipidemia, stroke, medication use (anti-hypertension, anti-diabetes, anti-dyslipidemia), and baseline systolic/diastolic blood pressure. P for interaction was obtained from a likelihood ratio test comparing models with and without the interaction term between A Body Shape Index trajectory group and sex

In addition to relative associations, unadjusted Kaplan–Meier cumulative risk (1 − S(t)) of incident low cognitive performance at 2 and 4 years was calculated by ABSI trajectory group and sex, together with stratum-specific sample sizes and numbers of events (Table 3). Among females, the 4-year cumulative risk was 0.48 in moderate–stable trajectory group and 0.54 in high–slightly-increasing trajectory group, compared with 0.42 in low–rapid-rising trajectory group; among males, the corresponding estimates were 0.36 and 0.40, compared with 0.49 in low–rapid-rising trajectory group. Confidence intervals were wider in strata with limited sample size, particularly the female low–rapid-rising group, and these absolute-risk estimates should therefore be interpreted cautiously. These absolute-risk estimates provide descriptive context for the sex-by-trajectory interaction observed in the Cox models (Table 2). In head-to-head prediction models, adding ABSI trajectory group did not improve discrimination or reclassification beyond BMI or WC (Tables S7–S9; Figure S2).

Table 3.

Kaplan–Meier estimated cumulative risk of incident low cognitive performance at 2 and 4 years, by A Body Shape Index trajectory group and sex among older adults with hypertension in the China Health and Retirement Longitudinal Study

ABSI trajectory group	Sex	N	events	2-year cumulative risk (95% CI)	4-year cumulative risk (95% CI)
Moderate–stable	female	126	60	0.34 (0.25–0.42)	0.48 (0.38–0.56)
Moderate–stable	male	234	83	0.20 (0.15–0.25)	0.36 (0.29–0.41)
Low–rapid-rising	female	26	11	0.27 (0.08–0.42)	0.42 (0.20–0.59)
Low–rapid-rising	male	45	22	0.27 (0.13–0.39)	0.49 (0.32–0.62)
High–slightly-increasing	female	87	47	0.37 (0.26–0.46)	0.54 (0.42–0.63)
High–slightly-increasing	male	80	32	0.18 (0.09–0.25)	0.40 (0.28–0.50)

ABSI A Body Shape Index; Kaplan–Meier cumulative risk was calculated as 1 − S(t) (unadjusted). CI denotes 95% confidence interval. Estimates for the female low–rapid-rising trajectory group should be interpreted cautiously due to small sample size (N = 26). Adjusted associations are reported as hazard ratios from Cox models (Table 2)

ABSI trajectory groups and cognitive decline among older hypertensive participants

In longitudinal analyses of continuous cognitive z-scores, the associations between ABSI trajectories and cognitive decline are summarized in Table 4. After adjusting for potential confounding factors in model 3, global cognitive Z-scores declined over time in the reference group (low–rapid-rising trajectory group) (β for time = − 0.13, 95% CI − 0.17 to − 0.09). Baseline cognitive levels did not differ significantly between moderate–stable trajectory group and low–rapid-rising trajectory group (β = 0.01, 95% CI − 0.15 to 0.18), nor between high–slightly-increasing trajectory group and low–rapid-rising trajectory group (β = 0.09, 95% CI − 0.08 to 0.27). However, compared with low–rapid-rising trajectory group, moderate–stable trajectory group showed a less steep decline over time (moderate–stable trajectory group × time: β = 0.05, 95% CI 0.00 to 0.10), whereas the decline rate in high–slightly-increasing trajectory group was similar to that in low–rapid-rising trajectory group (high–slightly-increasing trajectory group × time: β = 0.00, 95% CI − 0.05 to 0.05). Domain-specific LMMs for episodic memory and executive function showed consistent directions (Table S10). To facilitate interpretation and align with the biennial survey waves, we additionally report adjusted 2-year and 4-year cumulative changes (Δ) by trajectory group and sex derived from LMMs (Table S11). No evidence of sex effect modification was detected for baseline differences or longitudinal slopes (trajectory group × sex: LRT P = 0.240; time × trajectory group × sex: LRT P = 0.280) (Table S12). We externally validated the longitudinal findings in an independent cohort, the English Longitudinal Study of Ageing (ELSA), using the same analytic framework (group-based trajectory modelling followed by longitudinal and time-to-event analyses). To enhance cross-cohort comparability despite differences in cognitive measures, cognition was standardized to baseline z-scores within each cohort. In ELSA, model fit criteria supported a 5-class ABSI solution (Supplementary Results S2; Table S13; Figure S5). Although trajectory solutions differed between cohorts (CHARLS: 3 classes; ELSA: 5 classes), ELSA similarly identified a distinct rapid-rising ABSI pattern and showed directionally consistent longitudinal patterns (Tables S14–S15; Figure S5). Time-to-event analyses in ELSA are reported in Table S16.

Table 4.

Associations between body shape index trajectory groups and longitudinal change in global cognitive function among older adults with hypertension in the China Health and Retirement Longitudinal Study, 2011–2015

Variable	Model 1	Model 2	Model 3
	β(95% CI)a	β(95%CI)a	β(95%CI)a
ABSI trajectoriesb
Low–rapid-rising	Reference	Reference	Reference
Moderate–stable	0.06 (-0.10 to 0.23)	0.04 (-0.13 to 0.20)	0.01 (-0.15 to 0.18)
High–slightly-increasing	0.14 (-0.04 to 0.32)	0.12 (-0.06 to 0.30)	0.09 (-0.08 to 0.27)
Time	-0.13 (-0.17 to -0.09)	-0.13 (-0.17 to -0.09)	-0.13 (-0.17 to -0.09)
ABSI trajectories × timec
Low–rapid-rising	Reference	Reference	Reference
Moderate–stable	0.05 (0.00 to 0.10)	0.05 (0.00 to 0.10)	0.05 (0.00 to 0.10)
High–slightly-increasing	0.00 (-0.05 to 0.05)	0.00 (-0.05 to 0.05)	0.00 (-0.05 to 0.05)

Model 1 adjusted for demographic factors, including age, sex, education level, marital status, and place of residence. Model 2 further adjusted for health-related factors, including smoking status, drinking status, and depression. Model 3 further adjusted for vascular/metabolic factors, including diabetes, stroke, dyslipidemia, medication use (anti-hypertension, anti-diabetes, anti-dyslipidemia) and baseline systolic/diastolic blood pressure. ^a The β coefficients were estimated using linear mixed-effects models. ^b The β coefficient and its 95% CI are reported as SD. ^c The β coefficient and its 95% CI are reported as SD per year. Trajectory-group coefficients represent differences in baseline global cognitive Z-scores relative to the reference group, the coefficient for time represents the annual change in global cognitive Z-scores in the reference group, and trajectory group × time coefficients represent differences in annual change in global cognitive Z-scores over follow-up relative to the reference group. ABSI A Body Shape Index, CI confidence interval, SD standard deviation

Sensitivity analysis

Sensitivity analysis 1 (alternative outcome definitions). The trajectory group × sex interaction remained statistically significant when expanding the cohort to include participants with baseline global cognition z-score < 0 (P for interaction = 0.018; Table S17). Using a stricter incident threshold (z-score < − 0.5) restricted to participants with baseline z-score ≥ − 0.5 yielded wider confidence intervals and attenuated statistical evidence for interaction (P for interaction = 0.073; Table S18), consistent with the lower event proportion under this stricter definition compared with the primary risk set (35.2% vs. 42.6%; Table S19).

Sensitivity analysis 2 (sex-stratified GBTM models). Sex-stratified GBTM models suggested potential sex-related differences in trajectory structure (Tables S20–S21). However, stratified solutions showed reduced cross-sex comparability and evidence of degeneracy in some candidate models, most notably an empty class with undefined diagnostics in women. Therefore, the pooled three-group solution was retained for the primary analyses to ensure a unified trajectory definition for downstream association analyses, while sex differences were evaluated within the pooled framework. In multinomial logistic regression accounting for classification uncertainty using posterior-probability pseudo-class draws (M = 200) with Rubin’s rules, males had lower odds of belonging to high–slightly-increasing trajectory group (vs. low–rapid-rising trajectory group) compared with females (adjusted OR = 0.462, 95% CI 0.251–0.852; Table S22). In sex-stratified Cox models, the trajectory group × sex interaction was borderline significant (likelihood-ratio test χ²(2) = 5.97, P = 0.051), and the proportional hazards assumption was supported (global Schoenfeld test P = 0.91; Table S23).

Sensitivity analysis 3 (trajectory reclassification, trajectory distributions by hypertension status, and hypertension effect modification). Refitting the trajectory model in the full cohort (hypertensive plus normotensive participants) yielded highly consistent trajectory assignments among hypertensive participants (κ = 0.793; 87.6% concordance; Tables S24–S26 and Figure S3A). Trajectories were also re-derived separately within normotensive participants (Table S27; Figure S3B); however, because stratum-specific trajectories are derived separately and are not intended for direct cross-stratum comparison, we used the overall-derived trajectory classification to both normotensive and hypertensive participants to compare group membership directly. Under this comparable classification, trajectory group membership did not differ by hypertension status: low–rapid-rising, 166 (14.7%) among normotensive participants vs. 130 (12.2%) among hypertensive participants; moderate–stable, 557 (49.2%) vs. 515 (48.4%); and high–stable, 408 (36.1%) vs. 420 (39.4%) (Pearson’s χ²(2) = 4.22, P = 0.121; Cramér’s V = 0.044; Table S28). Among normotensive participants, global cognitive z-scores declined over time; however, trajectory-group differences in longitudinal slopes were not statistically significant, and trajectory groups were not associated with incident low cognitive performance (Tables S29 and S30; Figure S4). In pooled LMMs including a three-way interaction term, hypertension status did not significantly modify trajectory-group differences in longitudinal slopes (P for interaction = 0.411; Table S29), and contrast-specific difference-in-differences (ΔΔβ) were small and non-significant (P = 0.229 and P = 0.586; Table S31). Similarly, there was no conventional statistical evidence of effect modification by hypertension for incident low cognitive performance (P for interaction = 0.073) (Tables S30). In an additional mixed-effects model evaluating continuous within-person change independent of the Z < 0 threshold, global cognitive Z-scores declined over time (β for time = − 0.085 per year, P < 0.001), while the hypertension×time interaction was not significant (P = 0.143) (Table S32).

Sensitivity analysis 4 (missing data). In inverse-probability-weighted Cox models accounting for selection into the complete-case incident risk set, effect estimates were broadly consistent with the complete-case analysis, and conclusions regarding the trajectory group × sex interaction were unchanged (Table S33).

Sensitivity analysis 5 (wave-based outcome ascertainment and interval censoring). Discrete-time complementary log–log and interval-censored Weibull models produced estimates very similar to those from the primary Cox models, and conclusions regarding the trajectory group × sex interaction were unchanged (Table S34).

Sensitivity analysis 6 (age/sex robustness checks). There was no evidence of age-related effect modification (P for trajectory×age-group interaction = 0.563; Table S35) or nonlinearity in age modeling (P for nonlinearity = 0.267), and age-/sex-standardized ABSI sensitivity analyses (including winsorization and a time-varying exposure model) yielded consistent findings (Table S36). In fully adjusted mixed-effects models, age and sex were associated with ABSI levels; the time×sex interaction was not significant, and the time×age interaction did not reach conventional statistical significance (Table S37).

Sensitivity analysis 7 (slope-based assessment of sex modification). To address limited precision from sparse sex-by-trajectory subgroups (notably women in the low–rapid-rising group, N = 26), we derived individual-specific ABSI change rates (random slopes) from LMMs and tested an ABSI slope × sex interaction in fully adjusted Cox models. The interaction remained statistically significant after additionally controlling baseline/mean ABSI using either a model-based ABSI level (P for interaction = 0.013) or observed baseline ABSI (P for interaction = 0.012), with slope associations directionally positive in women and weaker in men (Table S38).

Sensitivity analysis 8 (alternative reference category). As a re-parameterization check, we re-estimated the Cox and linear mixed-effects models with the moderate–stable group specified as the reference category; the substantive interpretation was unchanged (Tables S39–S40).

Reverse-causality checks. Additional adjustment for baseline cognition and exclusion of early events at the first follow-up wave did not materially change the direction of associations; however, precision was reduced and evidence for interaction was attenuated in the early-event–excluded analysis, consistent with fewer events (Tables S41–S42).

Subgroup analyses

Exploratory subgroup analyses were conducted within each sex to evaluate potential effect modification by age, marital status, residence, smoking status, drinking status, and depressive symptoms using likelihood-ratio tests (Table S43). No robust evidence of effect modification was observed after false discovery rate correction. A smoking-status interaction was nominally significant among women (P for interaction = 0.045) but did not remain significant after FDR adjustment (q = 0.268); therefore, this finding should be interpreted cautiously.

Discussion

To the best of our knowledge, this is the first study to examine sex-related differences in the association between ABSI trajectories and cognitive function among older patients with hypertension using a large, nationally representative cohort. In this study, ABSI trajectories from 2011 to 2015 were categorized into three distinct groups: moderate–stable, low–rapid-rising, and high–slightly-increasing. Our findings showed heterogeneous associations of long-term ABSI trajectories with cognitive outcomes in older hypertensive patients. LMMs suggested that the moderate–stable trajectory was associated with a modestly attenuated rate of global cognitive decline. In Cox models of incident low cognitive performance, there was borderline evidence of effect modification by sex (P for interaction = 0.046), even after adjusting for multiple confounders. These findings suggest that long-term ABSI trajectories may be relevant to cognitive health and may have potential implications for risk stratification among older adults with hypertension, pending replication in larger, independent cohorts.

These findings are consistent with previous research linking visceral fat to poorer cognitive function [49]. Several studies have reported associations between ABSI and cognitive function. For instance, a cross-sectional study based on the database from the NHANES showed that higher ABSI was associated with lower cognitive function in older Americans [50]. Similarly, another study found a negative association between elevated levels of ABSI and cognitive decline in older Chinese adults [51]. However, these previous investigations did not examine the dynamic trajectories of ABSI over time. Given that hypertension is a chronically progressive disease and ABSI can be dynamic, trajectory-based analyses offer a more informative approach to predict long-term cognitive outcomes. Recently, trajectory analyses using repeated measurements have been increasingly applied to study the longitudinal development of anthropometric indicators [52]. For example, West et al. found that a long-term obesity trajectory was associated with reduced volumes of brain regions involved in cognitive function [53]. Similarly, Liang et al. found that persistently high or increasing trajectories of BMI and WC beyond midlife were significantly associated with a higher risk of cognitive decline and dementia in later life [49]. Despite these findings, evidence regarding the relationship between ABSI and cognitive function in individuals with hypertension remains limited. To address this gap, this study used data from a large, nationally representative cohort to examine ABSI trajectories over a four-year period and their long-term associations with the risk of cognitive decline in older hypertensive patients. By adopting a longitudinal trajectory approach, this study expands on the limited existing evidence on ABSI and cognitive outcomes. These findings emphasize the importance of evaluating ABSI trajectories rather than relying solely on single-point measurements when assessing the risk of low cognitive performance among older patients with hypertension.

The underlying mechanisms linking changes in ABSI to increased cognitive decline among older patients with hypertension are not yet fully understood; however, several plausible explanations have been proposed. First, a gradual increase in ABSI over time may contribute to a greater accumulation of chronic low-grade inflammation. In individuals with hypertension, this inflammatory burden may be exacerbated by vascular dysfunction [9]. Chronic neuroinflammation, in turn, can damage neurons through mechanisms such as microglial activation, ultimately leading to neuronal apoptosis, brain atrophy, and cognitive decline [54, 55]. Second, prolonged inflammation may impair cerebral vascular function, reduce cerebral blood flow, and damage the endothelial lining of the blood vessels. These effects, including inflammation of the vascular wall, can increase the risk of cognitive decline [56]. Furthermore, visceral adiposity is a key driver of insulin resistance [57], which has been consistently linked to elevated cognitive decline risk in recent reviews [58, 59]. Notably, hypertensive patients often exhibit a cluster of metabolic abnormalities, including insulin resistance, which may synergistically increase their risk of cognitive dysfunction [60, 61]. Third, abdominal obesity has been associated with poor blood pressure control in patients with hypertension [62, 63]. Uncontrolled hypertension is a known contributor to structural brain abnormalities and cognitive impairment [10]. For instance, Li et al. demonstrated that elevated cumulative exposure to SBP over time was associated with an accelerated decline in cognitive function [64].

Although we did not formally assess the mediating pathways, obesity-related changes in the gut microbiota may influence host mood and behavior [65]. A higher ABSI was significantly associated with increased depressive symptoms, and this relationship was moderated by hypertension status [66]. Consistent with these findings, our study found that individuals in the low–rapid-rising trajectory group were more likely to experience depression. Therefore, it is crucial to explore these potential mediators using longitudinal analyses. For example, Mirza et al. found that only increasing depressive symptoms were associated with a higher risk of dementia, compared to the low depressive symptom trajectory [67]. Similarly, Du et al. reported that an increasing trajectory of depressive symptoms posed a greater risk of cognitive impairment [68]. These findings suggest that depression may mediate the association between ABSI and cognitive decline. While current evidence on the relationship between ABSI and cognitive function remains limited, our findings highlight the need for future studies to elucidate the biological and behavioral mechanisms underlying this association.

We observed borderline evidence of sex heterogeneity in the association between ABSI trajectory group and incident low global cognitive performance (trajectory group × sex interaction: LRT P = 0.046), suggesting that the trajectory–outcome relationship may differ between men and women on the hazard ratio scale. Within-trajectory contrasts from the interaction model suggested that, in the high–slightly-increasing trajectory group, men had a lower hazard than women (HR for men vs. women, 0.51; 95% CI, 0.30–0.86), whereas sex contrasts in the other trajectory groups were less precise. However, estimates involving the low–rapid-rising trajectory group should be interpreted cautiously due to its small sample size (women: N = 26) and resulting imprecision. Accordingly, the interaction should be considered exploratory.

Sensitivity analyses provided additional context. This apparent heterogeneity should be interpreted cautiously because it was sensitive to alternative operationalizations of the dichotomized outcome. When baseline low cognition was not excluded, effect estimates remained directionally similar and evidence of a trajectory group × sex interaction persisted. In contrast, when applying an alternative cutoff intended to capture subtle cognitive decline (Z < − 0.5) with an incident design restricting the risk set to baseline Z ≥ − 0.5, evidence for interaction was attenuated (LRT P = 0.073), which may reflect reduced precision and the altered risk set/event definition under the alternative cutoff. Notably, a slope-based sensitivity analysis that reduced reliance on sparse sex-by-trajectory subgroups yielded directionally consistent evidence of sex modification, with the ABSI change rate × sex interaction remaining statistically significant after controlling baseline/mean ABSI under two specifications. Nevertheless, these findings remain exploratory and warrant replication in larger, independent cohorts.

One potential interpretation is that the same ABSI-defined body-shape trajectory may reflect different underlying adiposity phenotypes across sexes. Men, on average, accumulate a greater proportion of abdominal/visceral adipose tissue than women [69–71], whereas women tend to have relatively more subcutaneous fat deposition [72]. Accordingly, the high–slightly-increasing ABSI trajectory group may have different metabolic implications across sexes and, in women, may be more consistent with a central/visceral adiposity phenotype, given established sex differences in fat distribution [69–72]. Because visceral adiposity is more strongly linked to systemic inflammation, endothelial dysfunction, and insulin resistance—pathways implicated in cognitive decline [17]—a high–slightly-increasing ABSI pattern in women may reflect a greater underlying metabolic–inflammatory burden and may help explain the higher cognitive risk observed in women. By contrast, when women maintain proportionally more subcutaneous fat, cardiometabolic inflammatory consequences tend to be less pronounced [17, 72]. Sex differences in immune and inflammatory responses may further modulate the inflammatory and cardiometabolic responses to central/visceral adiposity, potentially contributing to the sex heterogeneity observed in our estimates [73]. Prior studies reporting sex differences in the prognostic performance of ABSI for mortality and vascular outcomes [74, 75] also support the notion that ABSI may capture risk-relevant body-shape information differently in men and women.

Notably, “low cognitive performance” defined by crossing a Z-score cutoff reflects a distribution-based threshold rather than clinical MCI/dementia incidence, and it captures time to first threshold crossing rather than within-person cognitive change. To complement the threshold-based analyses, we also modeled cognition on the continuous scale using LMMs, which directly quantify longitudinal within-person change. In these models, moderate–stable trajectory group showed a less steep decline in global cognition than low–rapid-rising trajectory group, whereas high–slightly-increasing trajectory group did not differ materially from low–rapid-rising trajectory group; formal tests provided no evidence that sex modified baseline differences or longitudinal slopes (2-way and 3-way interaction tests not supported). Taken together, these findings suggest that sex-related heterogeneity may be more apparent on the hazard ratio scale for a dichotomized endpoint than on the continuous decline scale, underscoring that effect modification can be scale-dependent and that dichotomization may alter apparent subgroup patterns by discarding information. Accordingly, we treat the sex-stratified survival results as exploratory and avoid definitive subgroup claims.

In exploratory subgroup analyses, we evaluated potential effect modification by several covariates within each sex. Although a smoking-status interaction was nominally observed among women, this signal did not persist after controlling for multiple testing and was based on relatively small strata; therefore, it should be interpreted as hypothesis-generating rather than confirmatory. Lewis et al. conducted a large-scale web-based observational cohort study involving adults aged 18 to 85 years and found that smoking and cardiovascular disease may affect verbal learning and memory across adulthood in sex-specific ways. Specifically, their findings suggest that smoking may have a more detrimental impact on cognitive health in women than in men [76]. It is possible that the adverse metabolic and vascular impacts of both smoking and central obesity may synergistically increase cognitive vulnerability. Nevertheless, given the limited precision of estimates in some strata and the exploratory nature of these analyses, future studies with larger samples and repeated measurement of lifestyle factors are needed to clarify whether smoking truly modifies the association between ABSI trajectories and incident low cognitive performance.

These findings help interpret how longitudinal body-shape patterning, as captured by ABSI trajectories, relates to cognitive performance among older adults with hypertension and highlight the importance of evaluating effect modification across outcome scales and definitions. Sex-related heterogeneity was more evident for the threshold-based endpoint at the earlier transition to below-average performance (incident Z < 0 among those with baseline Z ≥ 0) but was attenuated under the stricter cutoff intended to capture subtler cognitive decline (incident Z < − 0.5; baseline Z ≥ − 0.5), consistent with cut-point–dependent changes in the risk set and event yield and the resulting loss of precision. In complementary analyses treating cognition as a continuous outcome using LMMs, we found no evidence of sex-by-trajectory interaction, suggesting that any apparent heterogeneity may be threshold- and scale-dependent and may vary with outcome definition. Accordingly, the sex-stratified and interaction findings should be considered exploratory and warrant replication in independent cohorts to assess robustness across outcome definitions and threshold choices. From a prevention standpoint, given evidence from a recent review, public health strategies such as promoting physical activity, managing hypertension, and supporting smoking cessation remain key to dementia prevention, with physical activity and blood pressure control being especially well supported by high-level evidence and should be prioritized [77]. In this cohort, ABSI trajectory group did not improve discrimination or reclassification beyond BMI or WC in fully adjusted head-to-head prediction models over ~ 4 years of follow-up. Nonetheless, lack of incremental prediction does not preclude etiologic relevance: ABSI trajectories may provide a complementary longitudinal exposure phenotype for characterizing body-shape patterning over time and generating testable hypotheses about heterogeneity across population subgroups (e.g., by sex). These observations are hypothesis-generating and warrant replication and external validation before informing population risk assessment, screening, or targeted prevention strategies.

From a clinical and methodological perspective, we derived the primary ABSI trajectories a priori among hypertensive older adults to enhance interpretability in this high-risk population. To assess generalizability, we performed complementary analyses in the overall cohort. We re-derived ABSI trajectory groups, aligned group labels by trajectory shape/level, and found high concordance of trajectory membership among hypertensive participants compared with the hypertension-only model (κ = 0.793; ARI = 0.627). In addition, interaction tests provided no evidence of effect modification by hypertension status in either longitudinal or time-to-event analyses, indicating that the main findings were generally consistent across hypertension strata and not materially dependent on restricting analyses to hypertensive participants. The main strengths of this study lies in its large, longitudinal cohort with repeated measurements of ABSI trajectories over time. It is the first to examine both baseline and follow-up associations between dynamic ABSI and subsequent cognitive performance and examined potential sex-related heterogeneity among older hypertensive individuals within a nationally representative cohort.

However, several limitations should be acknowledged. First, the primary analysis was restricted to older Chinese adults with hypertension, which may limit the generalizability of the findings to the general older population. Further studies in population-based cohorts of older adults are still needed to better understand the role of ABSI in cognitive change. Second, we assessed robustness in an independent cohort (ELSA), but differences between cohorts and incomplete alignment of available covariates mean this should be viewed as supportive evidence rather than exact replication. In ELSA, residential area and lipid-lowering medication use were unavailable; therefore, residual confounding cannot be excluded. Future studies should validate these findings in additional populations. Third, given the observational design of this study, causal relationships between ABSI and cognitive function cannot be inferred. Future longitudinal randomized controlled trials are needed to establish causality. Fourth, evidence for sex-specific associations was imprecise because some sex-by-trajectory strata were small (e.g., women in the low–rapid-rising group, N = 26; 11 events), and the sex-related differences appeared to depend on the threshold used to define incident low cognitive performance. Stricter thresholds produced fewer incident events and greater uncertainty, and therefore the sex × trajectory interaction findings should be interpreted cautiously as exploratory and warrant replication in larger, independent cohorts. Finally, the ABSI trajectory group × sex interaction was attenuated after excluding events first detected at the 2013 wave and additionally adjusting for baseline global cognition, suggesting that effect-modification inferences may be sensitive to early events and model specification. Thus, the findings should be interpreted with caution because residual confounding and potential reverse causality cannot be excluded, and causal inference is limited by the observational design.

Conclusions

ABSI trajectories showed heterogeneous associations with cognitive outcomes in older patients with hypertension, with patterns varying by analytic scale and outcome definition. LMMs showed modest trajectory-related differences in global cognitive decline, with moderate–stable trajectory group exhibiting a less steep decline than the low–rapid-rising trajectory group. In Cox models of incident low cognitive performance, we observed borderline evidence of sex-related effect modification, with sex-stratified point estimates suggesting opposite directions in women and men; however, sex-stratified estimates were imprecise and the interaction attenuated under an alternative cutoff, supporting an exploratory, hypothesis-generating interpretation. Replication in larger, independent cohorts is warranted before ABSI-trajectory monitoring is considered for routine risk stratification of early low cognitive performance in older adults with hypertension.

Authors’ contributions

YH designed the study. YH, YMD and TTS contributed to methodology. YMD wrote the main manuscript text and analysed the data. TTS, ZXZ, KS, RM, YJS contributed to software, data curation and validation. YH contributed to manuscript supervision, funding acquisition, project administration, resources, writing-review and editing. All authors reviewed the manuscript and approved the final version.

Funding

This work was supported by the National Natural Science Foundation of China (72174092; 72574106), Pharmaceutical Research Fund Project of Jiangsu Pharmaceutical Association (JSPA-KY-202402), Postgraduate Research Innovation Program of Jiangsu Province (KYCX24_2082), the ‘’Unveiling and Commanding’’ Project of the School of Marxism, Nanjing Medical University, and the Young academic leaders of Qing Lan Project in Jiangsu province.

Data availability

The datasets used and/or analyzed during the current study are available fromd the corresponding author upon reasonable request.

Declarations

Ethics approval and consent to participate

This study was conducted in accordance with the Declaration of Helsinki. All methods were carried out in accordance with relevant guidelines and regulations. This study was approved by the Peking University Institutional Review Board (IRB00001052-11015). All participants have given their written informed consent to participate in the study.

Consent for publication

Not applicable.

Competing interests

The authors declare no competing interests.