The data for this study were obtained from the CHARLS, an ongoing longitudinal survey that aims to capture the health conditions and related factors among community-dwelling adults aged 45 years and older across 28 provinces in China. CHARLS provides comprehensive information on personal characteristics, family dynamics, health status, cognitive function, retirement, and personal property status. It is a nationally representative survey that offers reliable data on middle-aged and older individuals.

The data used in our study were freely accessible to the general public and were collected through a series of exams conducted every 2 to 3 years. We utilized data from four waves: wave 1 (collected in 2011), wave 2 (collected in 2013), wave 3 (collected in 2015), and wave 4 (collected in 2018) of the CHARLS survey. These waves allowed for a longitudinal analysis of the study variables.

Out of the initial 25,586 participants enrolled, a total of 22,590 participants were excluded for reasons including missing essential information such as age, gender, region, etc. Additionally, 110 participants below the age of 45 were excluded. To ensure data quality, any participants with missing information or abnormal values for the VAI, as determined by the Tukey method based on VAI quartiles, were also excluded. After applying these exclusion criteria, a total of 2,677 eligible individuals remained in the final analysis sample. The selection process for participant inclusion is illustrated in Figure 1, providing a visual representation of the steps taken to arrive at the study sample used for analysis.

VAI was calculated using gender-specific formulas proposed by Amato et al. (2010) to account for physiological differences between males and females in terms of visceral adiposity. VAI serves as an indicator of visceral adiposity and is derived from anthropometric and metabolic parameters. The formulas used for calculating the VAI were as follows: For males: VAI = (WC/(39.68 + 1.88 × BMI)) × (TG/1.03) × (1.31/HDL), For females: VAI = (WC/(36.58 + 1.89 × BMI)) × (TG/0.81) × (1.52/HDL). Here, WC represents waist circumference, BMI is body mass index, TG stands for triglycerides, and HDLC represents high-density lipoprotein cholesterol. These formulas allowed us to calculate gender-specific VAI values, providing a more accurate assessment of visceral adiposity for males and females. The calculated VAI values were subsequently divided into tertiles, yielding three levels: Tertile 1 (T1), Tertile 2 (T2), and Tertile 3 (T3). The corresponding ranges for these levels in our study were “< 2.36,” “2.36–4.54,” and “> 4.54.”

Cognitive function assessment in the CHARLS database included measures of global cognitive score, episodic memory, and mental status. Episodic memory assessment involved the presentation of a list of Chinese words to participants, who were then asked to immediately recall as many nouns as possible. After a 5-min delay, participants were prompted to recall the nouns again (delayed recall). Episodic memory scores were computed as the sum of the immediate and delayed recall scores, ranging from 0 to 20. A higher episodic memory score indicated better performance in remembering and recalling the presented nouns. Mental status was evaluated using the Telephone Interview of Cognitive Status (TICS) questionnaire. Participants were asked to perform several tasks, including subtracting 7 from 100 multiple times, naming the current date, and redrawing a previously shown picture. The correct responses were tallied to calculate a single TICS score ranging from 0 to 11. A higher TICS score reflected better mental status and cognitive functioning. The global cognitive score was determined by summing the episodic memory score and the TICS score. This composite score provided an overall measure of cognitive function, ranging from 0 to 31. Higher global cognitive scores indicated better overall cognitive performance, considering both episodic memory and mental status. Cognitive assessments, including episodic memory and mental status, were conducted by trained investigators following standardized protocols. These assessments, along with the calculation of the global cognitive score, enabled a comprehensive evaluation of cognitive function within the CHARLS study population.

Data on various demographic and clinical characteristics were obtained for the participants from the CHARLS public database. These included age, gender, education, marital status, region (urban or rural), drinking and smoking status, as well as self-reported or physician-diagnosed chronic diseases such as cancer, hypertension, diabetes, heart disease, dyslipidemia, stroke, liver disease, kidney disease, and digestive disease. Education was categorized into three levels based on the official coding file of the CHARLS. Level 1 represented individuals with no formal education (illiterate), those who did not complete primary school but could read, those who received private tutoring (Sishu), or those who completed elementary or middle school. Level 2 included individuals who completed high school or vocational school. Level 3 encompassed respondents with two/three-year college education, a college degree, or a postgraduate degree.

Descriptive statistics were used to summarize the baseline characteristics of the study population. Continuous variables were assessed for normality using the Shapiro–Wilk test. For variables that were found to be non-normally distributed, we chose to report the median and interquartile range (IQR), whereas normally distributed variables were represented using the mean and standard deviation. Categorical variables were presented as counts and percentages. One-way analysis of variance (ANOVA) was used to compare the differences in continuous variables, while the Chi-square test was employed to compare the differences in categorical variables among the three groups. The relationship between the VAI and cognitive scores was assessed using generalized estimating equation (GEE) models. Four different models were employed to examine this association, with each model incorporating specific adjustments: Model 1: Adjusted for the cognitive score in 2011. Model 2: Further adjusted for gender and age. Model 3: In addition to the adjustments in Model 2, further adjustments were made for education, marital status, smoking, and drinking. Model 4: In addition to the adjustments in Model 3, further adjustments were made for hypertension, diabetes, heart disease, stroke, dyslipidemia, liver disease, kidney disease, and digestive system disease. Furthermore, four sensitivity analyses were performed to evaluate the robustness of the findings: 1. Exclusion of individuals with cognitive decline in 2011. 2. Exclusion of individuals with brain injury, intellectual retardation, or memory-related diseases, along with those having cognitive decline in 2011. 3. Combination of exclusion of individuals with cognitive decline in 2011 and individuals with brain injury, intellectual retardation, or memory-related diseases. 4. Addition of follow-up time as a confounding factor correction variable. Model 5 is reported for the fourth sensitivity analysis. In the sensitivity analysis, individuals with low cognitive function in 2011 were excluded from the study population. Low cognitive function was defined as having a cognitive total score below two standard deviations from the mean baseline cognitive score. Participants whose cognitive total score in 2011 fell below the threshold of two standard deviations below the mean baseline cognitive score were classified as having low cognitive function. All statistical analyses were conducted using R version 4.1.2. A significance level of p < 0.05 was considered statistically significant, indicating evidence of a meaningful association.

In this research, we scrutinized data from 2,677 subjects, segregating them into three categories based on the tertiles of their VAI: T1, T2, and T3. As shown in Table 1, the average age in the T1 group was 57.3 years (Standard deviations (SD) = 8.10), in the T2 group it was 56.3 years (SD = 7.65), and in the T3 group it was 56.0 years (SD = 7.48). We observed a significant age difference among these groups (p = 0.001), with an evident declining pattern (P for trend = 0.001) as we moved from lower to higher VAI groups. When considering gender, the percentage of females escalated from T1 (28.6%) to T2 (46.9%) and T3 (57.5%), whereas the male percentage displayed a converse trend. The gender differences were statistically significant (p < 0.001), and a distinct trend was observed (P for trend<0.001). With respect to variables such as education, marital status, and geographic region, no significant discrepancies or patterns were discernible among the VAI groups. Nevertheless, lifestyle habits, specifically drinking and smoking, presented notable differences and patterns across the groups (p < 0.001 for both). The data indicated a declining ratio of drinkers and smokers in the higher VAI groups. Finally, as for health conditions including hypertension, diabetes, heart disease, and dyslipidemia, we found significant differences and patterns among the VAI groups (all p < 0.001), suggesting a greater incidence of these conditions in the higher VAI groups. Other health conditions like cancer, stroke, liver disease, kidney disease, and digestive disease did not exhibit significant differences or patterns across the groups, except for digestive disease which did show a significant pattern (P for trend = 0.04).

Our observations across each wave indicated a slight increase in cognitive scores as we transitioned from the lower VAI group (T1) to the higher VAI groups (T2, T3). As shown in Table 2, in the first wave, for instance, the global cognitive score saw a slight increase from T1 (mean = 16.3, SD = 4.04) to T3 (mean = 16.7, SD = 3.94). This trend was consistent in the second wave as well, with scores marginally rising from T1 (mean = 16.3, SD = 3.95) to T3 (mean = 16.8, SD = 3.87). Similar patterns emerged across all waves and all three cognitive measures (As shown in Figure 2). Overall, these results suggest a potential association between VAI and cognitive function, particularly in terms of global cognition and episodic memory.

In our study, we employed generalized estimating equations to investigate the influence of the VAI on three cognitive scores: global Cognitive, episodic memory, and mental status. As shown in Table 3, For global cognitive scores, all models showed a significant positive association with VAI. In Model 1, which adjusted for the cognitive score in 2011, the regression coefficient (β) for VAI was 0.074 (95% CI: 0.034 to 0.113, p < 0.001). This association remained significant after further adjusting for gender and age in Model 2 (β = 0.067, 95% CI: 0.027 to 0.106, p < 0.001), and additional adjustments for education, marital status, smoking, and drinking in Model 3 (β = 0.070, 95% CI: 0.031 to 0.108, p < 0.001). The significance persisted in Model 4, which further adjusted for hypertension, diabetes, heart disease, stroke, dyslipidemia, liver disease, kidney disease and digestive diseases (β = 0.064, 95% CI: 0.025 to 0.104, p = 0.001). For episodic memory scores, the association with VAI was also significant across all models, with the highest β in Model 1 (β = 0.071, 95% CI: 0.039 to 0.104, p < 0.001) and the lowest β in Model 4 (β = 0.051, 95% CI: 0.019 to 0.082, p = 0.002). However, for mental status scores, the association with VAI was not significant across all models.

In the stratified analysis, we further investigated the effect of the VAI on global cognitive (Table 4), episodic memory (Table 5), and mental status scores (Table 6), using T1 as the reference group. The analysis was adjusted for cognitive scores from 2011 and potential confounders such as hypertension, diabetes, tumors, heart diseases, stroke, and dyslipidemia. For global cognitive scores, the gender-stratified analysis showed a significant positive association with VAI for male participants in both the T2 (β = 0.424, p = 0.021) and T3 (β = 0.598, p = 0.003) groups. No significant association was found for female participants in either group (T2: β = −0.056, p = 0.836; T3: β = 0.053, p = 0.842). The age-stratified analysis showed a significant positive association in the T3 group for participants aged <60 years (β = 0.406, p = 0.030). In the education-stratified analysis, both education groups in the T3 group showed significant associations (Education 1: β = 0.567, p < 0.001; Education 2&3: β = 0.595, p = 0.076). In the region-stratified analysis, a significant association was observed in the T3 group for participants from rural villages (β = 0.643, p = 0.002). The marital status-stratified analysis showed a significant association in the T3 group for married participants (β = 0.589, p < 0.001). The smoking-stratified analysis showed a significant association in the T2 and T3 groups for participants who reported smoking (β = 0.419, p = 0.034 and β = 0.495, p = 0.021). For the episodic memory scores, there is a statistically significant association consistent with the global cognitive scores for T2 or T3 in the following subgroups: males (T2: β = 0.293, p = 0.035; T3: β = 0.438, p = 0.004), individuals under the age of 60 (T3: β = 0.312, p = 0.025), education level 1 (T2: β = 0.335, p = 0.005; T3: β = 0.487, p < 0.001), individuals with religious affiliation (T3: β = 0.547, p < 0.001), married individuals (T2: β = 0.370, p = 0.001; T3: β = 0.514, p < 0.001), individuals who consume alcohol (T3: β = 0.389, p = 0.006), and individuals who smoke or not. No statistically significant associations were observed for mental status scores in any of the stratified analyses.

In the first sensitivity analysis (Table 7), which involved excluding individuals with a history of brain injury, intellectual retardation, or memory-related diseases, we observed significant associations between total VAI and global cognitive function in four Models (β = 0.080, p < 0.001; β = 0.070, p < 0.001; β = 0.074, p < 0.001; β = 0.069, p < 0.001), as well as between the T3 VAI tertile and global cognitive function (β = 0.432, p = 0.002; β = 0.368, p = 0.091; β = 0.390, p = 0.005; β = 0.347, p = 0.013). Similar interpretations can be made for the remaining sensitivity analyses (Tables 8–10). Additionally, the p-value for the trend across the VAI tertiles is provided. For example, in the first sensitivity analysis for global cognitive function across four models (Table 7), we found a statistically significant trend (p < 0.001). Similar interpretations can be made for the other sensitivity analyses. Similarly, we observed a similar relationship between VAI and episodic memory. For instance, in the first sensitivity analysis across four Models, we found a positive correlation between Total VAI and episodic memory scores (β = 0.082, p < 0.001; β = 0.074, p < 0.001; β = 0.078, p < 0.001; β = 0.072, p < 0.001), as well as between the T3 VAI tertile and episodic memory scores (β = 0.464, p = 0.001; β = 0.410, p = 0.004; β = 0.434, p = 0.002; β = 0.386, p = 0.007). The value of p for the trend across the VAI tertiles was less than 0.05 in all four models. Similar conclusions can be drawn from the other sensitivity analyses (Tables 8–10). However, for mental status scores, there was no significant correlation observed with VAI in any of the models or sensitivity analyses (Tables 7–10).