Cognitive decline is often associated with aging and various pathological conditions, and represents a significant challenge for middle-aged and older adults. The impact of cognitive decline on daily life, decision-making, and independent functioning in older adults can be profound (Pan et al., 2015), which is necessitating a multidisciplinary approach to its prevention and management. Among the various strategies explored, exercise has emerged as a promising non-pharmacological intervention. However, the exact role of exercise in maintaining or improving cognitive function among middle-aged and older adults remains controversial (Johansson et al., 2022; Lavie et al., 2015).

Numerous studies have reported the beneficial effects of exercise on cognitive function, suggesting that it can delay cognitive decline and even enhance cognitive performance. Zhang et al. emphasized potential cognitive functional benefits of physical exercise for the brain health of older adults in a systematic review (Zhang et al., 2024). Similarly, Xu et al. found that moderate-intensity exercise improves cognitive function (Xu et al., 2023). The beneficial effects of exercise on cognitive function are mediated through multiple mechanisms, including enhanced cardiovascular health, weight management, and reduced inflammation. These factors are hypothesized to enhance cognitive performance, potentially mediated by elevated brain-derived neurotrophic factor levels (Norman et al., 2018). Emerging evidence further suggests that exercise may improve cognitive function through dual mechanisms: enhanced cardiovascular efficiency and strengthened functional connectivity between the anterior putamen and sensorimotor cortex (Johansson et al., 2022; Lavie et al., 2015).

However, some studies have found negative or non-significant effects of exercise on cognitive function in middle-aged or older adults. In a randomized controlled trial, the researchers found that exercise did not improve the cognitive performance in older adults over 50 years (Snowden et al., 2011). Some epidemiological studies also suggested that high-intensity or more frequent physical activity was not associated with better cognition (Wu et al., 2021; Zhou et al., 2022). This could be due to various factors, including the type, intensity, and duration of exercise (Prakash et al., 2015), as well as the health conditions of the participants (Washio et al., 2021). Besides, due to the lack of research into the mediating factors of blood pressure (Peri-Okonny et al., 2015), BMI(Xu et al., 2024), sleep (Sewell et al., 2021), blood lipid (Horowitz et al., 2020), the underlying mechanisms are not well known (Erickson et al., 2019). Notably, in older adults, research findings on BMI and cognitive function are inconsistent - some studies report a positive association, whereas others suggest a negative relationship(Monda et al., 2017).

In the present study, we analyzed the potential mediating factors between physical exercise and cognitive decline in middle-aged and older adults from the China Health and Retirement Longitudinal Study (CHARLS) cohort. The physical exercise was evaluated by intensity, frequency, and a physical activity score [PAC] calculated using metabolic equivalents [METs]. Considering that exercise may influence physiological indicators such as blood pressure, glucose, lipids, inflammation, and BMI, etc, we selected several related physiological indicators as mediators to elucidate the potential mechanisms underlying the relationship between exercise and cognitive decline.

This is a prospective cohort study. The CHARLS is a nationally representative survey initiated in 2011 that aimed to collect health-related information among Chinese who are over 45 years old and their spouses. This survey was approved by the institutional review board of Peking University, China (IRB00001052-11015). All subjects provided written informed consent at the baseline and follow up. A total of 17,714 participants were initially recruited, and biennial follow-up was carried out. This study used dataset of baseline (2011) and the wave 4 (2015). The exclusion standards were as follows: 1) participants whose cognitive function was not tested in 2011 or 2015; 2) participants with cognitive decline at baseline; 3) participants without a recording of exercise status at baseline; 4) participants with Alzheimer’s disease, brain atrophy, or Parkinson disease at baseline. According to the above exclusion standards, a total of 3,153 participants were included into the analysis.

According to the previous research of this database (Liu et al., 2021; Zhou et al., 2020), cognitive function of participants was assessed through episodic memory and mental intactness. Short-term episodic memory was evaluated by immediately recalling ten words read by the investigators, and long-term episodic memory was evaluated by recalling these ten words after 5 min. Episodic memory score was 20 points in total. Mental intactness is comprised of computing ability (subtract 7 from 100 serially for five times), time orientation (month, day, year, week and season) and picture drawing (draw two overlapping pentagons). Mental intactness score ranged from 0 to 11 points. Global cognitive score is the sum of episodic memory and mental intactness, and ranged from 0 to 31. A global cognitive score of less than 11 points was defined as cognitive decline (Liu et al., 2021).

In this study, exercise status was categorized based on multiple methods. First, it categorized into non-exercise, light exercise, moderate exercise and vigorous exercise according to exercise intensity. Vigorous exercise makes one’s breathing much harder than normal, such as heavy lifting, digging, plowing, aerobics, fast bicycling, and cycling with a heavy load. Moderate exercise makes one’s breathing somewhat harder than normal, such as carrying light loads, bicycling at a regular pace, or mopping the floor. Light exercise includes at work and at home, walking to travel from place to place, and any other walking that for recreation, sport, exercise, or leisure. Each type of exercise that exceeds 10 minutes is regarded as effective exercise. Second, participants were classified as light exercise 1-5 days, light exercise 6-7 days, moderate exercise 1-5 days, moderate exercise 6-7 days, vigorous exercise 1-5 days, and vigorous exercise 6-7 days, based on exercise intensity and frequency. Third, we generated a variable of physical activity score (PAC) that calculated by metabolic equivalents (MET). One MET refers to the amount of oxygen consumed at rest. Light exercise consumes 3.3 MET, moderate exercise consumes 4.0 MET, and vigorous exercise consumes 8.0 MET. PAC = 3.3 × light exercise weekly duration (min) + 4.0 × moderate exercise weekly duration (min) + 8.0 × vigorous exercise weekly duration (min). In this study, non-exercise was regarded as basic PAC. PAC increased due to exercise was divided into quartiles.

Summarizing previous studies and based on our experience, we selected age, gender, marital status, education level, body mass index (BMI), systolic pressure (SP), diastolic pressure (DP), smoking, drinking, sleep duration, blood glucose, blood lipids, C-reactive protein (CRP), uric acid, and chronic diseases (hypertension, diabetes, heart problems and depression) as covariates. BMI was calculated based on the standard definition (weight/height², kg/m²). Marriage was categorized into single, married cohabitation, married but separated. Education level was divided into illiteracy, did not complete primary school, primary school, middle school, high school or above. Smoking and drinking were arranged as all the time, former, and never. Sleep duration was divided into 7-9 hours, < 7 hours, and ≥ 9 hours. Chronic diseases were coded as Yes or No. Other variables, including age, BMI, SP, DP, glucose, blood lipids, uric acid, CRP, glycosylated hemoglobin (HbA1c) and score of center for epidemiological survey-depression scale (CESD) were analyzed in the form of continuous variables.

Given the continuous variables are non-normally distributed in this study, continuous variables were presented as the median and interquartile range (IQR). Gender, marriage status, smoking, drinking, sleep duration, education levels and chronic diseases were presented as categorical variables. The Kruskal-Wallis test was employed for continuous variables test between cognitive decline and cognitive normal groups, and the Chi-square test was used for categorical variables. A binary logistic regression model was applied to identify the associations of exercise with cognitive decline from 2011 to 2015. Confounding factors were selected based on a causal path diagram created using the directed acyclic graph (DAG) approach (www.dagitty.net). The DAG method derived the most parsimonious model with theoretically and priori-experience-based adjustment, and without the risk of overadjustment and associated reduction of statistical power. According to DAG diagram, minimal sufficient adjustment sets for estimating the total effect of exercise with cognitive decline included age, gender, marriage, education, smoking, and drinking (Erickson et al., 2019; Horowitz et al., 2020; Monda et al., 2017; Peri-Okonny et al., 2015; Sewell et al., 2021; Xu et al., 2024) (Supplementary Figure 1 ). In addition to univariate regression, Model 1 adjusted for age and gender; Model 2 adjusted for age, gender, marriage status, educational levels, smoking, and drinking. The results of logistic regression were shown as odds ratios (ORs) and 95% confidence intervals (CIs).

We conducted relative mediation effect analyses to explore potential paths of the association between exercise and cognitive function. The exercise intensity, frequency and PAC were set as independent variables, respectively. The cognitive decline was set as dependent variable. Physiological factors, including BMI, SP, DP, sleep duration, blood glucose, triglyceride, NHDLC, CRP, uric acid, HbA1c and chronic diseases were included as potential mediating factors. Age and sex subgroup analysis were performed in order to observe the age and sex-specific association.

The data were processed by R version 4.1.3 software (R Foundation for Statistical Computing, Vienna, Austria). All statistical tests were two-sided, and the significance level was P < 0.05.

The baseline demographic information of the included participants was shown in Table 1. We recruited 3,151 participants cognitively normal at baseline. After 4 years of follow-up, 888 (28.2%) participants experienced cognitive declined. The participants with incident cognitive decline were older, had lower educational level, lower BMI and higher SP. The proportions of individuals who were female, single, with longer sleep duration, without drink history and hypertension were higher in the participants with incident cognitive decline. Smoking status, blood glucose, blood lipids, CRP and DP were not significantly different between the groups with and without cognitive decline.

In different intensity of exercise group, participants with light, moderate and vigorous intensity exercise all had a significant decreased incidence of cognitive decline compared with non-exercise subjects (Figure 1 A). Any frequency of moderate or vigorous exercise was associated with lower incidence of cognitive decline compared with non-exercise subjects (Figure 1 B). Any increase of PAC due to exercise were also associated with lower incidence of cognitive decline compared with non-exercise (Figure 1 C). But there was not significant difference for the incidence of cognitive decline among quartiles of PAC that increased due to exercise (Figure 1 C).

As shown in Table 2, cognitive decline was negatively associated with light, moderate and vigorous exercises with respect to no exercise (age and gender adjusted OR [95% CI]: 0.62 [0.47, 0.83], 0.45 [0.37, 0.64], 0.54 [0.41, 0.71], respectively). After adjusting the age, gender, marriage, educational level, smoking, and drinking, the significance was still present (OR [95% CI]: 0.70 [0.51, 0.95], 0.56 [0.41, 0.75], 0.55 [0.41, 0.74], respectively).

All exercise frequencies were negatively associated with cognitive decline, even adjusted for age, gender, marriage, educational level, smoking, and drinking. Besides, we discovered that the strongest correlation with cognitive decline was moderate exercise performed 1-5 days a week (adjusted OR [95% CI]: 0.51 [0.35, 0.74]) (Table 2).

Compared with basic PAC, any increased PAC due to exercise were significantly associated with stable cognitive function (adjusted OR [95% CI]: 0.69 [0.50, 0.94], 0.69 [0.50, 0.94], 0.58 [0.43, 0.80] and 0.65 [0.48, 0.89] for quartiles of PAC that increased due to exercise, respectively). The strongest correlation with cognitive decline was the third quartile of PAC that increased due to exercise (adjusted OR [95% CI]: 0.58 [0.43, 0.80]).

In adults ≥ 60 years, almost all exercise types were significantly associated with a decreased risk of cognitive decline. However, in adults < 60 years, only vigorous exercise 1-5 days (adjusted OR [95% CI]: 0.62 [0.40, 0.98]), or the second quartile of PAC that increased due to exercise (adjusted OR [95% CI]: 0.56 [0.36, 0.87]) was associated with a decreased risk of cognitive decline compared with no exercise (Supplementary Table 1 ).

In males, all exercise types were associated with decreased risk of cognitive decline. However, in females, moderate intensity (adjusted OR [95% CI]: 0.64 [0.42, 0.97]), vigorous intensity (adjusted OR [95% CI]: 0.58 [0.38, 0.89]), moderate exercise 6-7 days (adjusted OR [95% CI]: 0.63 [0.41, 0.99]), vigorous exercise 1-5 days (adjusted OR [95% CI]: 0.56 [0.34, 0.91]), or the second quartile of PAC that increased due to exercise (adjusted OR [95% CI]: 0.51 [0.33, 0.80]) was associated with a decreased risk of cognitive decline compared with no exercise (Supplementary Table 2).

We further explored whether exercise affected cognitive function through physiological indicators using a mediating effect model. SP, DP, BMI, sleep duration, HbA1c, score of CESD, uric acid, CRP, blood glucose, NHDL-C and triglyceride were included as mediating factors in the analysis. In the overall population, vigorous exercise (regardless of frequence) and the fourth quartile of PAC might increase the risk of cognitive decline by decreasing BMI (Table 3). Other physiological indicators had no role of mediating effect (Supplementary Table 3).

In age subgroups, vigorous exercise might increase the risk of cognitive decline by decreasing BMI (Supplementary Table 4 ). In sex subgroups, among women who engaged in vigorous physical activity 6-7 times per week and the fourth of PAC, the indirect effect of physical activity level on cognitive function through BMI was significant, while the direct effect of physical activity level on cognitive function was not significant, suggesting complete mediation (Supplementary Table 5). Other assumed mediating factors have no significant mediating effects in males and in females (Supplementary Table 5 ).

In this study, we investigated potential mediating factors between physical exercise and cognitive decline. The results showed that exercise was associated with decreased risk of cognitive decline, and decreased BMI may mediate the positive association between exercise and cognitive decline in females.

In recent studies, the relationship between exercise and cognitive function remains confused. In this study, we found that individuals who engaged in light, moderate, or vigorous levels of exercise had a significant decreased risk of cognitive decline compared to those with no exercise habits, suggesting that any levels of exercise may be beneficial for cognitive function. Consistently, previous studies have shown that exercise interventions can have multiple benefits on cognitive function (Anders et al., 2021; López-Ortiz et al., 2021; Venegas-Sanabria et al., 2022). However, some studies have conflicting views. Lamb et al. (2018) found that a moderate to high intensity aerobic and strength exercise training programme did not slow cognitive impairment in participants with mild to moderate dementia. Other studies used the same population as this study, but concluded that vigorous exercise is not associated with better cognition, either at baseline or at follow-up (Wu et al., 2021). Different approaches to measuring cognitive function may explain the heterogeneity of results across the studies. In addition, it is likely that moderate exercise had a stronger effect than vigorous exercise compared with no exercise. The reason for this phenomenon remains to be elucidated. But our results were similar to previous studies. For example, Zhou et al. (2022) used linear regression models to analyze the association between exercise and cognitive score in a cross-sectional database, and found that vigorous exercise is associated with lower cognitive score compared with no exercise. In another study, any type of exercise was associated decreased risk of cognitive decline, and moderate exercise had the strongest effect size (Ding et al., 2021).

We conducted a subgroup analysis to investigate the association between exercise and cognitive function in different age groups. The statistically significant association was mainly shown in participants aged 60 years or above. This is consistent with Rosa et al.’s study. They suggested that elderly individuals who exercise are more likely to maintain cognitive function (De la Rosa et al., 2020). These findings underscore age as a critical factor in the exercise-cognition relationship, highlighting its essential role as a moderating factor. Besides, this also might be attributed to several potential mechanisms, such as the regulation of β-amyloid conversion, the synthesis and release of neurotrophins (Pahlavani, 2023) and inflammation (Jensen et al., 2019). Older participants are more vulnerable to these conditions, and the benefit of exercise, therefore, may be more easily amplified.

In the subgroup analysis by gender, we found that exercise was associated with better cognitive function in men and women. Just like the results in the overall population, moderate exercise has the strongest effect size, regardless of men and women. Researchers think heavy exercise can also induce physical damage, such as stress and excessive physical load, which may harm cognitive function (Chang et al., 2012). Inconsistently, a study found that the tertile of PAC was significantly associated with decreased risk of cognitive decline in females, but not significant in males (Luo et al., 2022). However, we found a stronger association between exercise and cognition in males. In addition to considering evaluation methods for exercise, the link between sex hormones and cognitive performance could be one reason for the observed differences between males and females. Sex hormones, such as androgen and estrogen, have been shown to play a crucial role in the cognition (Hamson et al., 2016; Szoeke et al., 2021). Compared to androgens, estrogen may have a more favorable effect on cognitive function. Currently, studies have found that patients with estrogen therapy showed improvement in cognitive function (Szoeke et al., 2021). These results may partly explain the sex-different finding of current study.

We further explored whether exercise could affect cognitive function through several physiological indicators using a mediation effect model. We found that strong exercise may increase the risk of cognitive decline by reducing BMI. This is consistent with the finding of obesity paradox (Monda et al., 2017). Such as, in a study, among participants who were assigned to the low BMI group (≤ 24.5 kg/m²), declined BMI was associated with worse cognitive function (Li et al., 2020). Research on the obesity paradox suggests that a low BMI may be associated with an elevated risk of cognitive impairment. Consequently, high-intensity exercise could potentially associate with cognitive impairment by reducing BMI. Alternatively, the reduction in BMI might attenuate the protective association between high-intensity exercise and cognitive impairment. This indicates that BMI reduction could act as a mediator in the relationship between high-intensity exercise and cognitive impairment, potentially diminishing or reversing its beneficial effect. Our findings suggest that middle-aged and older women may need to maintain exercise but maintain an appropriate weight to reduce the risk of cognitive decline.

The strength of this study is that the exercise was evaluated using multiple methods. In this study, physical exercise was assessed across multiple dimensions: intensity, frequency and PAC. This integrated approach not only overcomes limitations in existing literature where exercise metrics are often examined in isolation, but also enables more granular analysis of exercise-cognition associations. Consequently, our findings significantly advance the evidence base by delineating how distinct exercise dimensions collectively influence cognitive outcomes through their interplay with physiological indicators. Furthermore, mediating pathways were explored, providing mechanistic insights. However, several limitations are acknowledged. First, the date of cognitive decline was not identified although this is a prospective study. We can only confirm whether cognitive decline occurred after 4 years follow-up. Second, since we only utilized data from two time points and were unable to accurately determine the timing of cognitive decline onset, calculating the rate of cognitive decline proved methodologically challenging. Third, the causal relationship cannot be directly concluded, which may be finally determined by interventional study. Forth, self-report of physical activity may introduce varying degrees of recall bias.

In this study, we found that higher intensity and frequency of exercise was associated with better cognitive function, and mainly in older adults aged 60 and above. The sexual difference was not obvious. Exercise may affect cognitive function by changing BMI.