Interaction effects of exposure to air pollution and social activities on cognitive function in middle-aged and older Chinese adults based on a nationwide cohort study

Abstract

Background

Although there have been many studies on the relationship between ambient air pollution and cognitive functioning in developed countries, there are no studies focusing on the interaction between ambient air pollution and social activities. This study aims to examine interactive effects of ambient air pollution and social activities on cognitive function in Chinese middle-aged and older.

Methods

This study used nationally representative longitudinal survey data of China Health and Retirement Longitudinal Study (CHARLS) 2013, 2015 and 2018. The study explored the additive interaction effects of air pollutants and social activities on cognitive function in middle-aged and older adults by constructing mixed linear regression analyses containing interaction terms, as well as constructing additive interaction analyses with dummy variables containing four unordered categories that were partitioned according to median. In addition, the study further explored the interaction between air pollution and different types of social activities through an interaction term between air pollution and different types of social activities.

Results

In the model fully adjusted for covariates such as age, sex, region, we found significant coefficients on the interaction term between PM_2.5, O₃ and social activities on cognitive function (PM_2.5, β = -0.018, 95%CI: -0.029, -0.006; O₃, β = 0.017, 95%CI: 0.007, 0.027). In the interaction analysis by constructing dummy variables, we found a significant antagonistic effect between PM_2.5 and social activities (SI = 0.730, 95%CI: 0.674, 0.785), a possible antagonistic effect between NO₂ and social activities (SI = 0.697, 95%CI: 0.648, 0.747), and a possible synergistic effect between O₃ and social activities (SI = 1.769, 95%CI: 0.648, 0.747). In addition, the study found significant interactions between simple interaction, leisure and recreational, and intellectual participation social activities and air pollution.

Conclusion

Our study demonstrated an antagonistic effect of PM_2.5 and social activities on cognitive function in middle-aged and older Chinese adults.

Supplementary Information

The online version contains supplementary material available at 10.1186/s12889-024-21107-2.

Introduction

Cognitive function is the advanced psychological activity process of the human brain to receive and process external information, including aspects such as perception, memory, attention, language, and cognitive understanding [1, 2]. Numerous studies have shown that cognitive function declines with age, and cognitive decline in some older adults may further progress into mild cognitive impairment (MCI) and Alzheimer disease (AD) [2, 3]. China is one of the world's fastest aging nations, China's population over 65 is predicted to reach 365 million by 2050, with 115 million of them being over 80 years old [4, 5]. It is reported that the prevalence (924.1/100 000) and mortality rates (22.5/100 000) of AD and other dementias in China are higher than the global average [6], and the issue of cognitive impairment among middle-aged and older individuals has attracted global public attention.

Recently, the World Health Organization's newly released WHO Global Air Quality Guidelines state that the health risks associated with PM_2.5 (2.5-µm Particulate Matter, PM_2.5) and PM₁₀ (10-µm Particulate Matter, PM₁₀) are of particular importance to public health [7]. Recently, several studies have indicated that air pollution is an important influence on cognitive decline [8, 9]. A prospective cohort study from French Three-City demonstrated that higher levels of exposure to PM_2.5 accelerate the decline in cognitive function among participants [10]. A study in India showed a significant decline in cognitive function in participants living in homes with indoor air pollution compared to homes without indoor air pollution [11]. However, there are also studies that have not found any association between cognitive function and air pollution [12]. Despite the fact that scholars in developed countries have studied the connection between environmental pollution and cognitive function extensively, some studies have suggested an association between air pollutants and cognitive function, some have not found an association, and there is still heterogeneity in the available research [13–15]. Therefore, it is more meaningful to further explore the relationship between environmental pollution and cognitive function of middle-aged and older people in China, which is the world's first populous country and the world's second largest economy.

In contrast to air pollution, appropriate socialization has significant health benefits for cognitive function in middle-aged and older adults. Previous studies have found that active participation in social activities is effective in improving cognitive function and reducing the risk of cognitive deterioration in middle-aged and older adults [16, 17]. A UK study found social isolation increases risk of dementia [18]. In China, a study also showed that high levels of social contact can prevent the occurrence of dementia in the older people [19]. A previous literature points to social cohesion as a moderator of air pollution exposure and dementia [20], ignoring the interaction between social activities and air pollution may make the results less accurate, so it is important to focus on the interaction between air pollution and social activities on cognitive function. Some studies have examined the separate effects of ambient air pollution and social activities on cognitive function in middle-aged and older people, but no study in China has focused on the interactive effects of the two [21–23]. Consequently, the objective of this study is to assess the interactive effects of ambient air pollution and social activities on cognitive function in Chinese middle-aged and older adults using nationally representative cohort data.

Methods

Study population

Data from the Waves 2 (2013), 3 (2015) and 4 (2018) of the China Health and Retirement Longitudinal Study (CHARLS) were used in our study. CHARLS is a representative, nationwide longitudinal survey conducted by Peking University, targeting people over the age of 45 and covering a wide range of health, social and economic aspects of the population. The survey used multistage sampling to select respondents from 450 communities in 150 counties across 28 provinces in China [24]. After excluding respondents with missing data of the dependent variables, major covariates, and lack of exposure indicators for air pollution, we retained a total of 33355 participants. The Biomedical Ethics Review Committee of Peking University approved CHARLS, and all respondents were required to sign an informed consent form.

Measurement of cognitive function

The cognitive function assessment of this study is derived from the cognitive function test section of the CHARLS questionnaire, which includes four parts: time orientation check, attention and calculation check, executive function check, and language and immediate and delayed memory capacity check [25, 26]. The TICS (Telephone Interview for Cognitive Status, TICS) was used for the time orientation check and attention and calculation check, which have good reliability in the Chinese population, with reliabilities above 0.98 for both scales [27]. In the time orientation check, respondents were asked to answer today's date (containing the day, month, year, day of the week and season) for the assessment. The attention and calculation check were evaluated by asking the respondents to subtract 7 from 100 five times in a row. In terms of executive function check, it was tested by having respondents draw graphics in the pictures shown. And the immediate and delayed memory tests are conducted by recalling 10 words. In the test, each correct answer scores 1 point, with a total score of 31, with higher scores on the test representing better cognitive functioning of the respondent. The cognitive function test portion of the CHARLS was assessed by face-to-face interviews with respondents conducted by trained interviewers.

Social activities

In this study, social activeness was constructed to quantify the social activities in which the respondents were involved. Social activeness was obtained by summing the CHARLS questionnaire on whether or not participants engage in social activities (including 11 social activities in five categories: Simple Interaction, Intellectual Participation, Fitness and Exercise, Leisure and Recreational, Organizational Groups and Helping Others) and the frequency of activities in the past month [28–30]. Among these, the simple interaction type of social activities includes hanging out and socializing with friends; Intellectual Participation includes activities such as attending school or training programs, speculating on stocks, and surfing the Internet; Fitness exercise type includes dancing, fitness, practicing qigong, etc.; Leisure and Recreational types include playing mahjong, chess, cards, and going to community rooms; Organizational Groups and Helping Others types include caring for the sick or disabled in different residences, helping family members or neighbors in different residences, and participating in voluntary or charitable activities. Therefore, we build a comprehensive index of social activeness by combining both the number of social items and social frequency [31]. The formula is as follows:

$$S={\sum }_{i=1}^{N=11}(A_i\times F_i)$$

In the formula, S stands for social activity (quantified using the indicator “social activeness”) and $A_i$ indicates whether or not socialization took place (The question in the "Health Status and Functioning" section of the CHARLS questionnaire asks "Have you done any of these activities in the last month?” (Multiple choice); Yes = 1, No = 0). Frequency of participation of $F_i$ representatives in conducting each social activity (Respondents in the "Health Status and Functioning" section of the CHARLS questionnaire were asked how often they socialize; never = 0, infrequently = 1, almost weekly = 2, almost daily = 3). The values of the quantitative indicators for social activities ranges from 0 to 33, the higher the score, the more socially active you are.

Pollution exposure

According to the 2010 Sixth Population Census of China, the urban population in mainland China accounted for 49.68% of the total population, and the urban population accounted for 47.33% of the total population in the CHARLS data, so we chose to use population-weighted pollution exposure data in order to more accurately respond to the impact of pollution exposure. We utilize training dataset and random forest algorithm after resampling to construct two-stage machine learning model to integrate information from satellites, emission inventories, model simulations, and ground observations to estimate the annual PM_2.5 concentration [32, 33]. The spatial resolution of our analysis is 0.1° × 0.1° (approximately 10 km × 10 km) [34]. The results of out-of-bag cross-validation showed high accuracy of PM_2.5 estimates, with R² values of 0.80–0.88 and RMSE values of 13.9–22.1 μg/m³ during the years 2013–2020 [32, 35]. We obtained population densities from the world gridded population at a spatial resolution of 0.1° × 0.1° [36]. The PM_2.5 concentration for each city is derived by calculating the weighted average of the population density of all grids within the city boundaries [37]. We averaged the daily PM_2.5 data from 2013, 2015, 2016 and 2018 to obtain annual data.

We also established a three-stage random forest model that integrates multiple sources of data, including Ozone Monitoring Instrument (OMI), CMAQ simulation, and satellite remote sensing O₃ (Ozone, O₃) vertical profiles, to predict O₃ concentration [33, 38]. For the period 2013–2020, the O₃ forecast model is highly accurate, with an out-of-bag cross-validation R² value of 0.84 [39]. In addition, considering the spatiotemporal heterogeneity of air pollution, we also estimated the NO₂ (Nitrogen dioxide, NO₂) concentration by using a spatiotemporal extratree machine learning model that combines data from ground observation, satellite remote sensing, and other data [40, 41], the spatial resolution of our analysis is 0.1° × 0.1°. The cross-validated coefficient of determination R² value for this dataset over the period 2013–2018 is 0.84, which is highly accurate [42]. In the same way as for PM_2.5, we obtained O₃ concentrations and NO₂ concentrations for each city after population density weighted averaging. We then averaged the daily O₃ and NO₂ data for 2013, 2015, 2016 and 2018 into annual data. In addition, the ambient temperature and humidity data were obtained from the China Meteorological Forcing Dataset (CMFD) with a spatial resolution of 0.1° × 0.1° [43]. Similarly, the ambient temperature and humidity of each city are represented by the population density-weighted average of all grids within the city boundary. Participants from the same city were assigned the same pollutant concentrations.

Covariates

According to previous studies [44, 45], we considered various potential confounders in the regression analysis. Demographic variables included age, sex, education level (“Primary school or below”, “Middle, High, Technical secondary school” and “Junior college and above”), marital status (“married and living with spouse”, “married but living apart from spouse” and “single, divorced and widowed”), region (“eastern”, “central” and “western”) and living area (“Rural” and “Urban”). The health behavior variables consisted of smoking status (“never smoked”, “ever smoked” and “current smoking”), drinking status (“never drinking”, “less than once a month” and “more than once a month”) and chronic diseases status (“yes” and “no”). Among them, the chronic disease prevalence status in the CHARLS questionnaire was measured by asking respondents whether they had 14 chronic diseases such as hypertension, diabetes, lung disease, liver disease, heart disease, stroke and cancer and so on. We designated the chronic disease status as yes for respondents who answered that they suffered from one or more of these chronic diseases, and no for respondents who did not suffer from any of the chronic diseases. We also controlled for meteorological variables such as temperature and humidity in our analysis.

Statistical analysis

In this article, statistical description of continuous variables that conform to normal distribution were expressed as mean ± standard deviation, continuous variables that do not conform to normal distribution are expressed as M (P₂₅, P₇₅) while categorical variables were presented as n (%). Spearman correlation analysis was used to assess the correlation among PM_2.5, O₃, NO₂, annual mean temperature and relative humidity. Considering the study design as a longitudinal study, we employed a mixed linear model to analyze the interactive effects of ambient air pollution and social activities on cognitive function in the study subjects [46, 47]. The model was formulated as follows:

$${\mathit{score}}_{i,j}={\beta }_0+{\beta }_1{x}_{i,j}+{\beta }_2{s}_{i,j}+{\beta }_3{x}_{i,j}{s}_{i,j}+{\beta }_4{z}_{i,j}+{\beta }_5{\mathit{Temp}}_{i,j}+{\beta }_6{\mathit{Hum}}_{i,j}+\mu (i)$$

In the model, i represents individuals and j represents the visit waves. The ${\mathit{score}}_{i,j}$ represents the cognitive test score of individual i at visit wave j, where β₀ indicates the intercept. β₁ stands for the regression coefficient of ambient air pollution and β₂ represents the regression coefficient for social activities. β₃ represents the coefficient of the interaction term between air pollution and social activities. $z_{i,j}$ denotes a set of adjusted covariates, and β₄ represents its regression coefficient. ${\mathit{Temp}}_{i,j}$ and ${\mathit{Hum}}_{i,j}$ represent the annual average temperature and humidity, and μ(i) is the random effect term.

The crude models include annual average temperature and humidity, annual average PM_2.5, NO₂ or O₃ concentration and social activities and their multiplicative interaction terms as fixed effects, and individual ID as a random effect to control for the impact of within-individual variation. Fully adjusted models further adjust for sex, age, marital status, educational level, region, living area, individual health behavior variables (smoking status and drinking status) and chronic disease conditions based on crude models.

We used two approaches to examine the interactive effects of air pollutants and social activities on cognitive function. First we initially explored the interaction between the two using a model containing a product interaction term. Exp (interaction term coefficient) = 1 indicates no interaction between the two, exp > 1 indicates synergy, and exp < 1 indicates antagonism. Second, to further explore the mode of action of the interaction between the two, we classified PM_2.5, NO₂,O₃ and social activeness as low and high using the median as the cut-off point and created dummy variables containing each of the four classifications (Low PM_2.5 or low NO₂ or low O₃ exposure and high social activities; high PM_2.5 or high NO₂ or high O₃ exposure and high social activities; Low PM_2.5 or low NO₂ or low O₃ exposure and low social activeness; high PM_2.5 or high NO₂ or high O₃ and low social activities). The Synergy Index (SI) and its 95% CI were calculated to test the additive interaction between PM_2.5 and social activities, SI = 1 means no additive interaction, SI > 1 means synergistic effect, and SI < 1 means antagonistic effect, and the 95% CI of the SI was calculated by Bootstraps method. To further explore the interaction between pollutants and different types of social activities, we analyzed whether there was an interaction between pollutants and different types of socialization using an interaction term between pollutants and different types of social activities after fully adjusting for covariates.$SI=\frac{\text{exp}\left(B_{11}\right)-1}{\left[\text{exp},\left(B_{01}\right),-,1\right]+\left[\text{exp},\left(B_{10}\right),-,1\right]}$

To check the robustness of the main model, we also performed sensitivity analyses. (1) We used PM_2.5 annual mean concentration and annual mean temperature and humidity data with a 1-year lag to replace the current year's data for our analysis. (2) Considering the nonlinear relationship between average annual temperature and relative humidity and cognitive function, we tested the robustness of the main model by incorporating average annual temperature and humidity as nonlinear terms in conjunction with a restricted cubic spline function. (3) Given the non-linear effect of age, we used the same methodology to include age as a non-linear term in the model. (4) Instead of population-weighted data, we used unpopulated-weighted annual average air pollutant concentrations for our analysis. (5) To retain a larger sample size, we retained all individuals with complete data on the dependent variable, and after fully adjusting for covariates, we performed complete-case analyses.

All statistical analyses were performed using R (version 4.2.3) software. The mixed linear model was constructed using the "lme4" and "lmerTest" packages, and P < 0.05 indicating statistical significance. The interaction effects are plotted using the "interactions" package. The "forestplot" packages were used for forest plot. The spatial distribution maps of annual mean PM_2.5 air pollution concentrations and cognitive function scores were plotted using ArcMap software (version 10.8).

Results

Table1 shows the basic characters of 33,355 participants from CHARLS 2013、2015 and 2018. The respondents’ average age was 60.15 (SD = 9.16), with 47.30% of respondents being male and 52.70% being female, and 82.64% of the participants were married and living with their spouses. 57.96% of the respondents live in rural areas, with the largest number of respondents (33.87%) coming from the central region. More than 60% of the respondents had a low level of education, with only primary school education or below.

Table 1.

Statistical description of participants' basic characters

Variable	Total	2013CHARLS	2015CHARLS	2018CHARLS
Number of visits	33355	10371	9616	13368
Age (years)	60.15 ± 9.16	58.59 ± 9.11	61.30 ± 8.95	60.52 ± 9.19
Sex
Male	15777(47.30%)	4118(39.71%)	4834(50.27%)	6825(51.05%)
Female	17578(52.70%)	6253(60.29%)	4782(49.73%)	6543(48.95%)
Living area
Rural	19332(57.96%)	5998(57.83%)	5651(58.77%)	7683(57.47%)
Urban	14023(42.04%)	4373(42.17%)	3965(41.23%)	5685(42.53%)
Region
Eastern	11037(33.09%)	3430(33.07%)	3151(32.77%)	4456(33.33%)
Central	11297(33.87%)	3478(33.54%)	3203(33.31%)	4616(34.53%)
Western	11021(33.04%)	3463(33.39%)	3262(33.92%)	4296(32.14%)
Marital status
Married and living with spouse	27564(82.64%)	8679(83.69%)	7955(82.73%)	10930(81.76%)
Married but living apart from spouse	2001(6.00%)	558(5.38%)	504(5.24%)	939(7.02%)
Single, divorced and widowed	3790(11.36%)	1134(10.93%)	1157(12.03%)	1499(11.21%)
Educational level
Primary school or below	20272(60.78%)	6481(62.49%)	6098(63.42%)	7693(57.55%)
Middle, High, Technical secondary school	12261(36.76%)	3658(35.27%)	3270(30.01%)	5333(39.89%)
Junior college and above	822(2.46%)	232(2.24%)	248(2.58%)	342(2.56%)
Smoking status
Never smoked	20292(60.84%)	7331(70.69%)	5442(56.60%)	7519(56.25%)
Ever smoked	4436(13.30%)	963(9.29%)	1460(15.18%)	2013(15.06%)
Current smoking	8627(25.86%)	2077(20.02%)	2714(28.22%)	3836(28.70%)
Drinking status
Never drinking	21547(64.60%)	7020(67.69%)	6190(64.37%)	8337(62.37%)
Less than once a month	2918(8.75%)	930(8.97%)	853(8.87%)	1135(8.49%)
More than once a month	8890(26.65%)	2421(23.34%)	2573(26.76%)	3896(29.14%)
Chronic diseases
Yes	22267(66.76%)	6934(66.86%)	9469(98.47%)	5864(43.87%)
No	11088(33.24%)	3437(33.14%)	147(1.53%)	7504(56.13%)

SD standard deviation

Table 2 shows the results of the descriptive analyses of participants' social activities, cognitive functioning scores, pollution exposure, and meteorological data. In the three waves of data collection in 2013, 2015, and 2018, the average scores of participants' cognitive function was 15.40 (SD = 5.10), 14.51 (SD = 5.23), and 17.94 (SD = 6.74), respectively, and the median social activities score is 1.00, and the level of social activities among the respondents was mostly at a low level. The annual mean concentration of PM_2.5 decreased from 48.42 µg/m³ in 2013 to 40.38 µg/m³ in 2018, showing a downward trend year by year. And annual mean concentration of NO₂ decreased from 32.71 µg/m³ in 2013 to 28.58 µg/m³ in 2018. But the annual average concentration of O₃ increased from 120.20 µg/m³ in 2013 to 136.78 µg/m³ in 2018, showing an increasing trend from year to year.

Table 2.

Results of descriptive analyses of participants' social activities, cognitive function score, pollution exposure and meteorological data

Variable	2013CHARLS	2015CHARLS	2018CHARLS
Social activities (social activeness score)	1.00(0.00, 3.00)	1.00(0.00, 3.00)	1.00(0.00, 3.00)
PM2.5 concentration (µg/m3)	48.42 ± 20.63	45.77 ± 19.75	40.38 ± 15.35
NO2 concentration (µg/m3)	32.71 ± 11.76	28.78 ± 9.84	28.58 ± 9.34
O3 concentration (µg/m3)	120.20 ± 16.40	121.41 ± 18.59	136.78 ± 21.18
Temp (℃)	14.39 ± 5.07	14.35 ± 4.87	14.38 ± 4.92
Hum (%)	65.59 ± 8.41	68.82 ± 9.71	66.75 ± 10.40
Cognitive function score	15.40 ± 5.10	14.51 ± 5.23	17.94 ± 6.74

SD standard deviation, PM_2.5 fine particles less than or equal to 2.5 µm in diameter, NO₂ nitrogen dioxide, O₃ Ozone, Temp average annual temperature, Hum relative humidity

Figure 1. (A-C) shows that the spatial distribution of annual average air pollutant concentrations in provinces and cities shows significant differences. Highly polluted areas are concentrated in the Beijing-Tianjin-Hebei region and in populous provinces such as Henan and Shandong. According to Fig. 1. (D), there are distinct differences in cognitive function scores of participants from various regions. The average cognitive function scores of participants from the eastern region are slightly higher than those from the central and western regions, especially in eastern cities such as Beijing and Shanghai. (see Additional file 1, Figure A.1.a-e).

Fig. 1.

Spatial distribution of average cognitive function scores and average annual air pollutant levels in 2013, 2015 and 2018. Notes: White portion represents no data

The Spearman correlation analysis results showed that PM_2.5 was positively correlated with O₃ and NO₂ (P < 0.05), and negatively correlated with temperature and relative humidity (P < 0.05). O₃ was positively correlated with NO₂ (P < 0.05), and negatively correlated with relative humidity (P < 0.05). NO₂ was negatively correlated with temperature and relative humidity (P < 0.05). Temperature and relative humidity were positively correlated (P < 0.05). (see Additional file 1, Table A.3, Figure A.2).

Figure 2 summarizes the effects of annual mean PM_2.5, NO₂, and O₃ exposure and social activities and their interaction terms on participants' cognitive function scores. In Model 1, we found that study participants' cognitive function scores decreased with increasing annual average PM_2.5 concentration and increased with increasing social activities scores (P < 0.001). And the average decline in the study subjects' cognitive function scores was 0.075 points for every 10 µg/m³ increase in the annual average PM_2.5 concentration. (β = -0.075, 95%CI: -0.122, -0.028), and every 1-point increase in social activities scores was associated with a significant increase in cognitive functioning scores (β = 0.467, 95%CI: 0.408, 0.526). And we found a significant interaction effect of PM_2.5 and social activities on cognitive function (β = -0.023, 95%CI: -0.035, -0.011). In the fully adjusted model 2 for annual average PM_2.5 concentrations, the PM_2.5-social activities interaction term remained significant (β = -0.018, 95%CI: -0.029, -0.006). In the model where NO₂ is fully adjusted, we found that each 10 µg/m³ rise in NO₂ was significantly associated with a 0.372-point decrease in cognitive function score (β = -0.372, 95%CI: -0.459, -0.286). However, we did not find a significant interaction effect between NO₂ and social activities. In the fully adjusted model for ozone, we find significant coefficients on the interaction term between O₃ and social activities (β = 0.017, 95%CI: 0.007, 0.027).

Fig. 2.

Model estimates of the relationship between PM_2.5, NO₂, O₃ and social activities and their interaction term and participants' cognitive function scores. Notes: Model 1, 3, 5, crude model; Model 2, 4, 6, adjusted for sex, age, marital status, educational level, region, living area, individual health behavior variables (smoking status and drinking status) and chronic disease conditions based on crude models; β, Beta; CI, confidence interval

Interaction effect plots showed that higher level of social activities is associated with higher cognitive function scores at equivalent PM_2.5 levels (Fig. 3). The results of exploring the interaction between PM_2.5 and social activities using the interaction term showed a significant additive interaction between the two(P < 0.001), with a coefficient of the product term of -0.018 and exp (-0.018) < 1, suggesting that there may be a significant antagonistic effect between the two. In addition, this study found a possible synergistic effect between O₃ and social activities, with an interaction term coefficient of 0.017 and exp (0.017) > 1. Interaction effect plots showed that higher level of social activities is associated with higher cognitive function scores at equivalent PM_2.5 or O₃ levels (Fig. 3).

Fig. 3.

Interaction effect plot. Notes: PM_2.5, annual mean PM_2.5 concentration; Level of social activities social activeness scores; cognition, cognitive function scores

Table 3 demonstrates the additive interaction of PM_2.5, NO₂, and O₃ annual average exposure and social activities on participants' cognitive function. The results showed that the low PM_2.5-high level of social activities group was used as the control group, the remaining three groups (high PM_2.5-high level of social activities; low PM_2.5-low level of social activities; high PM_2.5-low level of social activities) all had B values less than 0, meaning that study participants were all at significantly higher risk of declining cognitive function scores than the control group. The decline in cognitive function scores was greatest in the high PM_2.5-low level of social activities group (B = -1.482, 95%CI: -1.661, -1.303). The SI was 0.730 (95% CI = 0.674, 0.785), indicating antagonism between the two. Using the low NO₂-high level of social activities group as the control group, the beta values of the remaining three groups (high NO₂-high level of social activities; low NO₂-low level of social activities; high NO₂-low level of social activities) were all less than 0, the risk of decline in cognitive function scores of the study participants was significantly higher than that of the control group (low NO₂-high level of social activities). The decline in cognitive function scores was greatest in the high NO₂-low level of social activities group (B = -1.574, 95%CI: -1.762, -1.386). The SI was 0.697 (95% CI = 0.648, 0.747), indicating additive interaction between the two. Therefore, individuals with high levels of social activities had significantly better cognitive functioning at equivalent levels of PM_2.5 or NO₂. Using the low O3-low level of social activities group as a control group, the high O3-high level of social activities group had the best cognitive function (B = 1.863, 95%CI: 1.693,2.033). The interaction index SI was 1.769 (95%CI = 1.519, 2.019), indicating a synergistic effect.

Table 3.

The additive interaction effects of PM_2.5, NO₂, and O₃ annual average exposure and social activities on participants' cognitive function

Category	N	$B($ 95%CI)	P-value
PM2.5-Social activities	/	/
Low–High	9409	Reference
High-High	10169	-0.546(-0.708, -0.383)	< 0.001
Low-Low	7255	-1.018(-1.173, -0.862)	< 0.001
High-Low	6522	-1.482(-1.661, -1.303)	< 0.001
SI	/	0.730(0.674, 0.785)
NO2-Social activities	/	/
Low–High	8478	Reference
High-High	9163	-0.688(-0.858, -0.517)	< 0.001
Low-Low	6457	-1.021(-1.185, -0.856)	< 0.001
High-Low	5825	-1.574(-1.762, -1.386)	< 0.001
SI	/	0.697(0.648, 0.747)
O3-Social activities	/	/
Low-Low	7171	Reference
High-Low	6606	0.899(0.720, 1.079)	< 0.001
Low–High	9403	0.962(0.807, 1.117)	< 0.001
High-High	10175	1.863(1.693,2.033)	< 0.001
SI	/	1.769(1.519, 2.019)

SI synergy index, CI confidence interval

Table 4 demonstrates the interaction of air pollutants with different types of social activities. In the interaction model between PM_2.5 and five types of social activities, we found that the coefficients of the interaction terms between PM_2.5 and simple interaction (β = -0.026, 95%CI: -0.052, -0.00004) and leisure and recreation (β = -0.040, 95%CI: -0.069, -0.010) types of activities were significant. In the modeling of the interaction of NO₂ with the five types of activities, we similarly found significant coefficients on the interaction term between NO₂ and leisure and recreational social activities (β = -0.084, 95%CI: -0.140, -0.028). And in the model of O₃'s interaction with five types of social activities, we found that it interacted with intellectual participation social activities (β = -0.034, 95%CI: -0.064, -0.004).

Table 4.

Interaction between air pollutants and different types of social activities

Air pollutants	Type of social activities	$\beta ($ 95%CI)	P-value
PM2.5	Simple Interaction	-0.026(-0.052, -0.00004)	0.049
	Intellectual Participation	-0.033(-0.075, 0.009)	> 0.05
	Fitness and Exercise	-0.015(-0.065, 0.035)	> 0.05
	Leisure and Recreational	-0.040(-0.069, -0.010)	< 0.05
	Organizational Groups and Helping Others	0.008(-0.016, 0.033)	> 0.05
NO2	Simple Interaction	-0.008(-0.057, 0.041)	> 0.05
	Intellectual Participation	-0.050(-0.119, 0.019)	> 0.05
	Fitness and Exercise	0.008(-0.078, 0.094)	> 0.05
	Leisure and Recreational	-0.084(-0.140, -0.028)	< 0.05
	Organizational Groups and Helping Others	-0.003(-0.051, 0.044)	> 0.05
O3	Simple Interaction	0.004(-0.019, 0.026)	> 0.05
	Intellectual Participation	-0.034(-0.064, -0.004)	< 0.05
	Fitness and Exercise	0.015(-0.027, 0.056)	> 0.05
	Leisure and Recreational	-0.006(-0.033, 0.021)	> 0.05
	Organizational Groups and Helping Others	0.008(-0.016, 0.032)	> 0.05

Β Beta, CI confidence interval

The results of the sensitivity analysis proved that the results of our main model are robust. In our model, where we replaced the current year data with lagged data by 1 year, there was a significance of the interaction term between PM_2.5 and social activities on cognitive function (β = -0.016, 95%CI: -0.030, -0.002). In the model that incorporated annual mean temperature and relative humidity as nonlinear terms using a restricted cubic spline function, the results were similar to those of the main model. In the model incorporating age as a non-linear term, the results were similar to the main model. In addition, the interaction term between PM_2.5 and social activities remained significant when analyzed using unpopulated-weighted pollutant data (β = -0.024, 95%CI: -0.034, -0.015). The interaction remained significant in the complete-case analyses were carried out in the models that included covariates (β = -0.015, 95%CI: -0.025, -0.005). Since the sensitivity analysis results are approximate to those of the main model, the main model was stable and well-fitted. (see Additional file 1, Table A.4, Table A.5, Table A.6, Table A.7, Table A.8).

Discussion

To the best of our knowledge, this is currently the first longitudinal study in developing countries on the interaction of air pollution and social activities on cognitive function in middle-aged and older adults. Using a nationally representative dataset, this study investigated the interactive effects of ambient air pollution and social activities on cognitive function in middle-aged and older Chinese adults. In this study, significant antagonistic effects were found between PM_2.5 and social activities, possible antagonistic sway between NO₂ and social activities, and synergistic effects between social activities and O₃ at high O₃ concentrations. Our results were proved to be reliable by adjusting for covariates and performing sensitivity analyses.

With the deepening of aging, more researchers are paying attention to mental health issues among middle-aged and older populations. Although a large amount of research evidence, recently, suggests a link between environmental pollution and cognitive functioning inconsistencies in the findings of existing studies remain [23, 48]. The results of a cohort study, which conducted in the United States, indicated that long-term exposure to PM_2.5 is linked to participants’ cognitive function scores decline (β = -0.22, 95% CI: -0.44, -0.01) [49]. A study using the French CONSTANCES cohort data suggests that exposure to PM_2.5 is significantly associated with a decline in cognitive function, particularly in semantic fluency (β = -4.6%, 95% CI: -2.1,–6.9) [50]. Our study is consistent with these findings. A study conducted in Los Angeles found ambient NO₂ exposure > 20 ppb is associated with reduced logical memory [51]. Research on the relationship between O₃ and cognitive function has yielded mixed results, with one study based in Taiwan, China, found that long-term exposure to O₃ jeopardizes cognitive function in older adults (OR = 1.878, 95% CI: 1.363, 2.560) [52]. Results from a Korean study showed that exposure to O₃ was associated with higher MMSE scores (β = 0.045, 95% CI: 0.027, 0.062) [53]. The study by Gatto et al. also found a positive correlation between exposure to high levels of O₃ and the logical memory scores of the study participants [51]. But in contrast to our study, a study based on the CHARLS database did not determine a causal relationship between cognitive impairment and O₃ exposure [54]. A previous study on ozone therapy noted that it reacts with the blood and positively affects metabolism, immune system regulation, antioxidant defence systems and microcirculation, and it has been more reliably studied in the treatment of neurodegenerative diseases [55]. It has been suggested that the use of personal or biomonitoring to measure individual O₃ exposure may be more appropriate for studies on O₃ and human health than for other pollutants, and that the relationship needs to be further explored in greater depth in the future [56]. Although there are differences in the risk effect values between this study and the above studies, the direction of association of the results is consistent. The minor differences in the resultant effect values may be due to differences in the sources and composition of pollutants in different countries, the way pollutant data are processed, the indicators chosen, and differences in population and culture.

Studies of biological mechanisms of air pollution-induced cognitive decline have found that there are two main pathways through which inhalable environmental particles directly damage the central nervous system. One is through the entry into the bloodstream and crossing the blood–brain barrier to reach the brain, and the other is through the olfactory nerves to directly transfer to the brain via the olfactory bulb [57]. Ambient particulate matter can also damage the central nervous system by attacking it through a variety of cellular and molecular pathways causing neuroinflammation, oxidative stress, neurological damage and cerebrovascular damage [58–60]. Damage to the central nervous system can lead to central nervous system lesions [61, 62]. In recent years, it has also been found that air pollution also causes cognitive decline due to impaired lung function [63].

Consistent with previous national and international studies, our study also found that active participation in social activities by middle-aged and older adults was beneficial to the improvement of cognitive function status [64–66]. A longitudinal study in South Korea showed that middle-aged and older adults who participated in social activities had higher cognitive function than those who did not participate [67]. There is a lack of clarity regarding the mechanisms by which socialization improves cognitive function in middle-aged and older adults, which may be related to several reasons: first, research has found that an active social life is one of the keys to maintaining brain health, and that interacting with others stimulates neural circuits in the brain that improve memory and cognitive function [68]. Secondly, a British study suggests that people who are socially active help develop cognitive reserves to better cope with age-related cognitive decline and any symptoms of dementia [69]. A study in China also suggests that intellectually engaging activities help increase cognitive reserve, thereby slowing neurobiological aging and preventing dementia [70].

Our study found that social activities can mitigate the hazards of pollutants to individuals with higher activity levels. In the analysis of the interaction between air pollutants and different types of social activities, we found that air pollutants mainly interacted significantly with simple interaction, intellectual participation and leisure and recreational social activities. There is limited research on the interaction between PM_2.5 and social activities on cognitive function. However, a study had shown that long-term exposure to air pollution is a risk factor for depression in older adults [71]. Nevertheless, social engagement can mitigate this risk [72]. Higher levels of psychological well-being (PWB) is associated with higher cognitive functioning [73]. Previous research has pointed to more frequent intellectual and social activities being associated with better cognitive function [74]. A study had shown that good social adjustment improves cognitive impairment caused by poor ambient air quality [75]. A study located in the United States showed that when stratified, greater combined air pollution exposure increased the risk of dementia in participants with low social cohesion (HR = 1.34, 95% CI = 1.04, 1.72), but not in those with high social cohesion (HR = 1.00, 95% CI = 0.93, 1.06) [20]. Our findings are similar to these studies [72, 76]. However, unlike us a study in the United States did not find a significant additive interaction between neighborhood social environment and air pollution [77]. In addition, high levels of social activities are synergistic with O₃ exposure. One study found an anti-inflammatory effect of exercise in mice exposed to air pollution, which may be a potential mechanism by which air pollution interacts with social activities to affect cognition [78]. A previous study pointed out that social activities can activate and strengthen different neurobiological pathways and functions, or induce more efficient use of brain networks, leading to improved cognitive function [79]. Exploring the interaction between air pollution and social activities on the cognitive function of middle-aged and older Chinese individuals can help motivate them to engage in positive social interactions, effectively preventing the occurrence of cognitive impairment.

This study has several strengths. First, there is limited evidence in developing countries such as China on the interaction of air pollution and social activities on cognitive function in middle-aged and older adults. Compared with other existing studies [8, 80–82], our study explored the interaction between air pollution and social activities, providing further public health recommendations for the prevention of cognitive function decline in middle-aged and older adults in China. Second, our study used nationally representative longitudinal follow-up data with a large sample size, and used population density-weighted pollutant data, with statistically robust and highly convincing results. Therefore, this study is conducive to enriching the research findings and content on the impact of air pollution on cognitive function among both middle-aged and older individuals nationwide, improving preventive measures for cognitive impairments among this demographic, improving health status among middle-aged and older individuals, and enhancing their quality of life.

There are several limitations to this study as well. First, the survey data used in this study did not provide the specific address information and specific dates of interviews of the respondents, therefore we were only able to measure the annual air pollution exposure levels of the respondents at a city level, which might lead to overlooking exposure differences among respondents from the same city. Second, respondent exposures in this study were matched based on the city in which the respondents lived at the time of the survey, which could lead to misclassification of exposure levels if the respondent's residence changed. Third, while our study has fully accounted for potential confounding variables, there are still some variables that may affect the results that may have been omitted. Fourth, as we did not use causal analysis for further analyses, our findings only demonstrate a correlation between participants' social activities and cognitive function, but fail to demonstrate a causal relationship between the two, which can be further explored in subsequent studies. Finally, the present study did not consider the possible nonlinear relationship between O₃ and cognitive functioning, which needs to be further explored in future studies.

This research has some policy applications. Owing to the relatively low level of development in developing countries and the inadequacy of environmental protection systems, middle-aged and older people are more vulnerable to the negative health effects of ambient air pollution as the decline in their physical functions with age. Our findings suggest that: first, the relevant policy makers should refine their policies on the management of different environmental pollutants to improve air quality. At the same time, publicity efforts should be strengthened to educate the public about the hazards of PM_2.5 on cognitive functions and to mitigate the adverse effects of PM_2.5. Thirdly, the government, the community and the older institutions should focus on enriching the forms of social activities for the middle-aged and the older adults, improving the relevant recreational and leisure facilities, organizing more rich community activities, and actively encouraging them to take the initiative to participate in rich and colorful social activities. In conclusion, this study is of great significance for developing countries to promote air quality management, improve preventive measures targeting the middle-aged and older adults' cognitive impairment, and improve the mental health of them.

Conclusion

In conclusion, this study suggests that there is a significant antagonistic effect between PM_2.5 and social activities, PM_2.5 increases the risk of cognitive decline in middle-aged and older adults, but individuals with higher levels of social activity were significantly better in terms of cognitive function. There may also be an antagonistic effect between NO₂ and social activities. In addition, this study found that participation in simple interaction, leisure and recreational, and intellectual participation social activities was more conducive to reducing the health burden of air pollution on middle-aged and older adults. This study prompts relevant policy makers to refine the prevention and control policies for different air pollutants and propose targeted measures to help Chinese middle-aged and older people alleviate the health burden of ambient air pollution. At the same time, middle-aged and older people should be encouraged to actively to engaging in social activities in weather with better air quality conditions to better protect cognitive function health. In addition, this study enriches the empirical evidence on the effects of ambient air pollution on the cognitive functions of middle-aged and older adults in developing countries.

Authors’ contributions

S .Y., Y. Z.: Conceptualization, Methodology, Data curation and analysis, Writing–original draft, Software. W. G., S. Z., R. L., B. A.: Data curation, Methodology, Software. H. L., Y. S., L. W.: Methodology, Writing–review and editing. C. H.: Conceptualization, Methodology, Writing–review and editing.

Funding

Chunlei Han was supported by National Natural Science Foundation Project (72374033).

Data availability

No datasets were generated or analysed during the current study.

Declarations

Ethics approval and consent to participate

Ethical approval for all the CHARLS waves was granted from the Biomedical Ethics Review Committee of Peking University, and all participants were required to provide written informed consent. The ethical approval number was IRB00001052-11015.

Consent for publication

Not applicable.

Competing interests

The authors declare no competing interests.