Clean Air Act mitigate the cognitive deterioration in older adults

Air pollution may accelerate cognitive aging, it is unclear whether large-scale interventions by Clean Air Act can mitigate the cognitive deterioration. Here, we conducted a difference-in-differences analysis based on Chinese Longitudinal Healthy Longevity Survey during 2014 and 2018. Intervention group came from where the government set a strict target of reducing air pollution, whereas control group lived in areas without reduction target. Global cognitive functions were measured using the Mini-Mental State Examination (MMSE). We found the intervention group with implementing Clean Air Act had a signicantly smaller decline in MMSE score compared to the control group. Interquartile increases in PM2.5 and SO2 concentrations were signicantly associated with a faster decline of MMSE score by 1.78 and 0.92 points, respectively. Implementing stringent clean air policies, especially in low- and middle-income countries may mitigate the risk of cognitive aging in older people.

Health Longevity Survey in 2014 and 2018 were aligned with the CCAA implementation plan (from 2014 to 2017). This provides a unique quasi-experiment opportunity to assess the cognitive impact of PM 2.5 and co-pollutants. We carried out the present analysis to test the hypothesis that rapid reductions in ambient concentrations of PM 2.5 and other pollutants brought by the CCAA implementation would reduce the risk of cognitive deterioration in older adults.

Descriptive statistics
The 2812 participants were 81.0 ± 9.3 years, comprised of 50.1% women and 57.6% rural residents ( Table  1). Of all participants, 2251 (79.8%) were in the intervention group, and the remaining 561 adults were in the control group. Participants in the intervention and control groups were comparable by sex and age. We observed signi cant between-group differences for rural/urban residence (P = 0.001), ethnicity (P < 0.001), smoking status (P = 0.02) and drinking status (P = 0.04). Therefore, we applied inverse probability weighting to control for these differences in our statistical models.
Across all the participants, average MMSE scores declined from 25.8± 5.2 in 2014 to 24.0 ± 7.1 in 2018 (Table 1). The average reductions of MMSE score were 2.1 ± 5.8 and 1.7 ± 6.5 among the participants in the control group and the intervention group, respectively. Long-term exposures to (annual concentrations of) PM 2.5 , PM 10

Effect of the CCAA intervention on cognitive function
The intervention effects on MMSE score, estimated in models with different covariates, are shown in Figure 1. When adjusted for no covariates (Model 1), the intervention group had a smaller decline in MMSE score than the control group, showing that the CCAA implementation was associated with an incremental MMSE score of 1.36 (95% con dence interval [CI]: 0.47, 2.25) points. In the fully-adjusted model, this estimate on the score improvement was larger (2.45 with 95% CI of 1.32, 3.57). In a sensitivity analysis by excluding the IPW weights and the offset of MMSE trend, the estimates were generally consistent to the fully-adjusted model results (Supplementary Figure A3). Moreover, statistical signi cances for intervention effects were consistently robust when the ≥5% reduction target was replaced with three larger reduction targets. The sample size of the control/intervention group using the three stricter cut-off points were 1027/1785, 1223/1589, 1800/1012, respectively. The estimate effect on improvement of MMSE score was slightly lower in the high cut-off group compared with low cut-off group. This phenomenon might be explained by the reason that more and more provinces with actually moderate emission reduction target were assigned to the control group as the threshold went up, thus making the difference of CCAA policy strength between the two groups become smaller.
We also examined potential effect modi cations (Supplementary Figure A4). We found the effect estimates were potentially modi ed by age, ethnicity, education, marital status, or physical activity (P for intervention-covariate interaction < 0.05). We found that the CCAA bene cial effect was ampli ed by a higher education level, being married, being less old, or being physically more active. The bene cial effect estimate was smaller in participants of Han ethnicity. This might be due to ethnicity differences in genetics or lifestyle characteristics. However, this needs to be interpreted with caution, given the small number of non-Han participants (n=189; 6.72%) in our cohort.
Associations between changes in pollutant exposure and cognitive function decline Figure 2 shows MMSE score changes associated with per unit increase in pollutant exposure, adjusting for the selected covariates. Separate cross-sectional analyses of the 2014 wave data and the 2018 wave data generated similar results for all the pollutants except O 3 , indicating increasing exposure was associated with decreasing MMSE score. According to the DID analysis, increasing exposure to PM 2.5 or to SO 2 was signi cantly associated with faster cognitive decline. Comparing the results from the DID analysis and those from the cross-sectional analyses, we found the PM 2.5 effect was the most robust, as the separate 2014 and 2018 wave cross-sectional analyses both found a signi cant MMSE association with PM 2.5 . Not surprisingly, due to a strong inverse correlation between PM 2.5 and O 3 , we observed a positive association between O 3 and MMSE score in the DID analysis. Adjusting for ambient temperature generated did not change these the associations (Supplementary Figure A5). Additionally, we conducted DID analyses in subgroups strati ed by age, sex, ethnicity, rural/urban residence, marital status, and years of schooling. We found the effect modi cation similar to that found in the intervention analysis described above (Supplementary Figure A6). Finally, we also associated pollutant exposure changes with MMSE score changes using nonlinear models (Figure 3), which yielded results comparable to the effects observed in the linear models. The nonlinear models consistently showed an inverse association of MMSE score with PM 2.5 or SO 2 , and a positive association with O 3 . In a sensitivity analysis by linking MMSE scores with PM 2.5 and O 3 changes jointly, the estimate for O 3 was statistically non-signi cant while conditioning on the PM 2.5 level (Supplementary Figure A7). The two pollutant models also suggested S0 2 has a signi cant impact on MMSE change after controlling O 3 or PM 2.5 on a constant level. However, the 2018 wave cross-sectional analysis did not nd a signi cant MMSE association with SO 2 exposure. It is possible that the SO 2 effect observed in the DID model re ects the simultaneous reductions of both pollutants by controlling coal combustion, a major source of both PM 2.5 and SO 2 . 39

Discussion
In this quasi-experimental study in a nationwide sample of older adults in China, we observed a bene cial effect of a national clean air policy (the CCAA implementation) on cognitive function. Speci cally, the covariates-adjusted mean MMSE score decline from 2014 to 2018 was 2.45 points (95% CI: 1.32, 3.57) less in the policy intervention group than in control group of participants whose annual PM 2.5 exposure reduction was <5%. Using annual average ambient pollutant concentrations at the residence level as exposure estimates, we found that increased PM 2.5 exposure was associated with decreased MMSE score across all participants in both the 2014 and 2018 waves of the survey. More importantly, we found that smaller MMSE score reductions from 2014 to 2018 were associated with greater PM 2.5 reductions over the same period. Taken together, these ndings support a potential causal role of PM 2.5 in accelerating cognitive decline in older adults, contributing to the emerging evidence, reported in previous observational studies, that long-term PM 2.5 exposure is a risk factor of cognitive function. [6][7][8][9][10][11][12][13][14][15] This study is the rst to assess the effectiveness of implementing a national clean air policy in attenuating aging-driven cognitive. By integrating the CCAA implementation targets for PM 2.5 between 2014 to 2017 with the CLHLS surveys in 2014 (before the implementation) and 2018 (after the implementation), we had the unique opportunity to design this quasi-experimental study in which we evaluated the effect of the clean air policy targeted on PM 2.5 reduction on cognitive function. In our difference-in-differences (DID) analysis approach, we further leveraged an experimental design consisting of an intervention group and control group based on differences in PM 2.5 reduction targets. This allowed us to obtain an appropriate counterfactual framework to estimate a causal effect. A strength of the DID approach is that through comparing same individual to itself at different time points, thus potential confounders such as genetic factors and lifestyles are controlled. Similar to previous DID designs, we controlled for certain slow-changing individual-level variables such as socio-demographics, lifestyles, and socioeconomic status. 17,19 We also controlled for annual temperature that may have varied with time.
These adjustments further reduced potential confounding to the effects of the CCAA intervention and PM 2.5 observed in our study.
The DID analysis approach has been used in environmental interventional studies of other health outcomes. 18 Previous studies have focused primarily on PM-related mortality. 16,17,19,40,41 In addition, our previously published works found that Clean Air Actions in China was associated with decreased household medical expenditures 42 and reduced prevalence of depressive symptoms. 43 To date, however, only one study in the United State has assessed the impact of the air pollution regulations on dementia diagnosis in people older than 75 years. 21 However, that study was a population-level analysis of the association between reductions in annual average PM 2.5 concentration from 2004 to 2013 and reductions in dementia incidence rate. Unlike the present study, that US based study did not control individual-level data such as lifestyle, socioeconomic status, health conditions, and other individual-level statuses. These individual-level variables may have changed substantially during the long evaluation period (2004-2013), which may confound the effect of the policy. In the present study, we conducted a series of sensitivity analyses considering individual-level variables and their changes, con rming the robustness of the signi cant bene cial effect on cognitive function of PM 2.5 exposure reduction brought by the CCAA implementation.
Our nding from the present study is consistent with previous epidemiologic ndings on the detrimental effects of air pollutants on brain health. [6][7][8][9][10][11][12][13][14][15] A meta-analysis of 31 studies, including both cross-sectional and longitudinal analyses, provided compelling evidence that air pollution exposure is associated with accelerated cognitive decline across the life course. 29 The present study, through a study design mimicking a randomized clinical trial, further strengthens the con dence on putative causality of similar conclusions from existing observational studies. 20 These epidemiologic ndings are supported by biological mechanisms by which ne particle matter impairs cognitive function. Experimental research has shown that ne and especially ultra ne particles (both are components of PM 2.5 ) can enter the brain either via circulation or intra-nasally by direct translocation through the olfactory bulb, leading to detrimental and toxic effects in the brain. [44][45][46] Several animal studies have shown increased neural in ammation in response to PM 2.5 exposures. [47][48][49] In a post-mortem study of 19 humans aged 34 to 83 years, who had died of non-neurologic causes, expression of cyclooxygenase-2 (COX-2), an in ammatory mediator, and accumulation of the 42-amino acid form of β-amyloid (Aβ42), a cause of neuronal dysfunction, were higher among those who had lived in severely polluted cities in Mexico than among those who had lived in less polluted cities. 50 A recent study in humans showed an association between air pollution and local gyri cation index, a marker of local brain atrophy in the aging brain, suggesting that chronic exposure to PM 2.5 may in uence the physiological aging process of the brain. 51 A major strength of this study is the use of the quasi-experimental study design to examine the potential causal link between CCAA policy and cognitive decline, by taking advantage of the rapid reduction of air pollution over a relatively short time window (4 years). The causal inference is strengthened by including a counterfactual analysis feature, i.e., the intervention group of participants who have experienced a PM 2.5 reduction exceeding a target (e.g., 5% annually) in comparison with the control group of participants had not experienced a target reduction in PM 2.5 exposure. A previous study has shown that about 90% of PM 2.5 variations during 2014 and 2017 in China were attributable to the CCAA implementation. 23 Therefore, PM 2.5 reduction could serve as a discriminative indicator for the CCAA effectiveness in our DID models. Indeed, the signi cant intervention effect based on group comparison is further strengthened by signi cant "exposure-response" relationship between MMSE score change with PM 2.5 change from 2014 to 2018. Another strength is our ability to use residence-level air pollution estimates (as opposed to city-level or ZIP-code-level widely used in previous studies) and individual-level socio-demographic and behavioral data. Moreover, we conducted a nationwide, prospective cohort study with neuropsychological data and individual characteristics in a diverse population in China, increasing the generalizability of our study ndings. As China is one of the largest LMICs in the world, our ndings may have policy implications for dementia prevention among LMICs.
This study has some limitations. First, its quasi-experimental design does not eliminate the possibility of residual confounding. Although we found improvement in cognitive decline after implementing the CCAA policy, we cannot rule out the possibility that there may be some other changes during this period of time or as a result of the policy (e.g., more physical activity or social participation after the air quality improves) which caused the slowing down of cognitive function decline. However, our methodology, including use of time and policy xed effects as well a time varying variable, helped mitigate this limitation. Second, model-based residence-level exposure estimates cannot be veri ed with measured data for all areas, although the models have been well validated. 52 Finally, the participants in our control group included those living in the areas that had been in poorer provinces with a range of other risk factors, such as cigarette smoking, and poverty. We used sensitivity analyses to evaluate the impact of some of these potential confounding factors.
Findings from this study have notable public health policy implications in the aging society. Dementia burden is projected to rapidly increase globally, especially in LMICs. Controlling any preventable risk factors is of paramount importance for a healthy aging society. The evidence generated in the presented study, with a unique strength of causal inference, support the need to implement well-designed clean air policies to reduce population and individual exposure to air pollutants especially PM 2.5 . The need is particularly urgent considering the already-known large cardio-respiratory disease burden of air pollution. 53 The policy-oriented interventions offer widespread protection and enhanced health with less reliance on conventional individual-centered lifestyle modi cations. 54 In conclusion, this quasi-experimental study adds new evidence of supporting causal inference that PM 2.5 exposure accelerates cognitive aging in older people. Considering the rapid population aging and increasing global disease burden from dementia, implementing policy-oriented interventions such as clean air policies, especially in LMICs, are of paramount importance for heathy aging.

Methods
This study was conducted in accordance with the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) guideline. The STROBE checklist can be found as part of the Supplementary Information.

Study design and participants
We used data from the Chinese Longitudinal Healthy Longevity Survey (CLHLS), a prospective cohort study on the determinants of healthy aging and longevity among older population (≥ 60 years) in China. Brie y, the CLHLS is a nationwide survey on a randomly selected sample from half of the counties and cities in 23 of the 31 provinces, covering about 85% of the total population of China ( Figure 4). Beginning in 1998, a new wave of survey has been conducted every 2-3 years. To reduce attrition in numbers from death and loss to follow-up, new 'make-up' participants are enrolled during each follow-up wave. Of note, the CLHLS oversampled male octogenarians, nonagenarians, and centenarians, further shaping the age distribution of the analytic sample. 24 Trained interviewers conducted the surveys at participants' homes following a structured questionnaire. They collected data on sociodemographic characteristics, lifestyle, cognitive function, psychological status, and physical capacity. More details on the sampling procedure and assessment of data quality can be found in previous publications. 25,26 The intervention and exposure assessment This interventional study was based on the effectiveness in PM 2.5 reduction through the implementation of China's Clean Air Act (CCAA) by provincial governments. We designated CLHLS participants who lived in provinces where the government set a target of reducing PM 2.5 concentrations ≥ 5% annually from 2014 onward into the intervention group and the rest of the participants into the control group (Figure 4). We identi ed 18 provinces with and 5 provinces without an established target of PM 2.5 reduction in 2014 (Supplementary Table A2). Description of CCAA was presented in Supplementary Methods, China's Clean Air Act.
We generated residence-speci c ambient concentrations of six pollutants (PM 2.5, PM 10 , SO 2 , NO 2 , O 3 , and CO) using a previously published method. 27 Brie y, we coupled the original monthly-average concentration data with participants survey dates and residential addresses to calculate the average concentrations of the six air pollutants during the 12 months preceding the survey dates. These annual average concentrations were regarded as long-term exposure estimates. We also obtained participantspeci c estimates of land surface temperature from a weather-forecast research (WRF) model. 28 Detailed descriptions of air pollution and weather assessment are presented in Supplementary Methods, Environmental Assessment.

Health outcomes and covariates
The primary outcome was global cognitive function assessed using the Mini-Mental State Examination (MMSE), a screening tool for dementia. The MMSE has been frequently used to track changes in cognitive function over time, as this tool can reliably detect cognitive deterioration. 29 This tool is especially useful for people unable to go through complex clinical diagnostic testing. 30 The global score of MMSE ranges from 0 to 30, with a higher score indicating a better cognitive function. 31 We calculated cognitive decline by subtracting participants' MMSE score in 2014 from their scores in 2018. Detailed description and scoring methods on MMSE were showed in Supplementary Methods, Mini-Mental State Examination.
We assessed a range of demographic, behavioral, and socioeconomic factors as covariates, including age, sex, education, marital status, ethnicity, rural/urban residence, occupation before retirement, survey month, alcohol drinking, smoking, physical activity, fruit intake, vegetable intake, water quality, living condition, and income source. Participants were encouraged to answer as many questions as possible. No proxy was used for objective questions such as assessment of cognitive function and physical performance. 24 The detailed de nition of covariates can be seen in the Supplementary Methods, Descriptions of covariates. In addition, we included model-estimated ambient temperature (see above) as a covariate.

Statistical analyses
We calculated within-individual differences between the MMSE scores assessed in 2014 and 2018. We used mixed-effect model in which within-individual MMSE difference was dependent variable, intervention group and selected covariates were xed-effect independent variables (see below), and participant ID was a random-effect variable. We used between-group difference in within-individual MMSE score difference to interpret the impact of the intervention (CCAA).
The comparability between the intervention group and the control group may be undermined due to the differences in aspects of (1) baseline demographic characteristics, (2) long-term trend of outcome variables, and (3) changes in longitudinal risk factors during the study period. To control for those differences, rstly, we applied inverse probability weighting (IPW) to match up the two groups in terms of baseline population characteristics including age, education, sex, residence, ethnicity, occupation before retirement, alcohol drinking status, smoking status, and living condition. 32,33 Secondly, we utilized withinindividual MMSE score changes from another prior wave of survey (in 2011) to the 2014 wave as a variable to approximate the longer-term individual-level MMSE score trends. We did not use the difference of MMSE score between 2018 and 2014 minus the difference between 2014 and 2011 as the dependent variable, because it may introduce some confounding factors that change with time as it covers a relatively long time-span. Thirdly, we further adjusted for temporal changes in the longitudinal risk factors for cognitive change, including survey month, marital status, alcohol drinking, smoking, physical activity, fruit intake, vegetable intake, water quality, living condition, and income source.
To assess the relationship between MMSE score and pollutant exposure, we conducted cross-sectional analyses of the 2014 and 2018 wave data, respectively, using linear regression models controlling for selected covariates. Furthermore, we used Difference-in-Differences (DID) analysis to assess the MMSE score change in association with pollutant exposure change from 2014 to 2018. The model is described as: where i denotes subject index; t denotes index of CLHLS wave and equals to 0 (7 th wave) or 1 (8 th wave); MMSE i,t denotes MMSE score for the i th subject at the wave t; x i,t denotes air pollutant exposure; β is the change in MMSE per unit change in exposure; z i,t denotes the selected covariates described above; γ is the coe cient for a speci c covariate; η i denotes the subject-speci c xed effect; and ∆t denotes the effect of temporal trend. To control for the seasonal variation in MMSE, we included an index of survey month in the main models. We also used an alternative approach by replacing the survey month with a spline term of monthly temperature (degrees of freedom = 3).

Sensitivity and effect modi cation analyses
First, we analyzed whether cognitive decline and PM 2.5 exposure had a different linear trend in intervention group versus the control group before the 2014 CCAA implementation using the 2011 and 2014 waves of CLHLS dataset (Supplementary Figure A1). Second, we applied different stricter cut-offs of the PM reduction targets (10%, 12%, and 15%), e.g., participants who lived in the places where the government set a target of reducing PM 2.5 concentrations of more than 10% as the intervention group and the rest of the participants were assigned to the control group Third, we repeated the analysis without IPW weights and without offset of longer-term MMSE trend. Fourth, we examined the interaction between pollutant exposure and intervention group indicator to assess whether exposure-outcome relationships were different between the two groups. Fifth, we explored the nonlinear exposure-response association through replacing the linear term of exposure by its spline expansions. Finally, we derived two-pollutant nonlinear models through parametrizing the exposure term using a two-dimensional spline function. The two-pollutant exposure-response associations were utilized to explore how temporal variation in a pollutant was linked with the MMSE score change, given the scenario where the other modelled pollutant was unchanged (Supplementary Methods, Two-pollutant exposure-response analysis).
All statistical analyses were performed using R (version 3.3.2; R Foundation for Statistical Computing, Vienna, Austria). Imputation was performed using the mice package. 34 The linear mixed-effects and xed-effects models were inferred using the lme4 package and the plm package, 35 respectively. Inverse probability weights were calculated using the ipw package. 36 Penalized spline functions were parameterized using the mgcv package, 37 and inference of the nonlinear mixed-effects models was done using the gamm4 package. 38

Declarations Data availability
The data that support the ndings of this study are available from Center for Healthy Aging and Development Studies, Peking University but restrictions apply to the availability of residential data, which were used under license for the current study, and so are not publicly available. Data are however available from the authors upon reasonable request and with permission of Prof Yi Zeng (zengyi@nsd.pku.edu.cn).

Competing interests
The authors declare that they have no competing interests. Contributions Y.Y., X.T., Z.J., W.H. and Z.Y. designed the research and directed its implication; Y.Y., and X.T. prepared and analyzed the data; Y.Y., L.X., and X.T. drafted the manuscript; all co-authors revised the manuscript.   The estimated effect of the CCAA on changes in MMSE score by difference-in-differences models. Note: CCAA, China's Clean Air Act; MMSE, Mini-Mental State Examination. Model 1: adjusted for inverse probability weighting from baseline and offset of MMSE trend; Model 2: further adjusted survey month based on Model 1; Model 3: further adjusted marital status, alcohol drinking, smoking, and physical activity based on Model 2; Model 4: further adjusted intake of fruit and vegetables, water quality, living condition, and income source. Effect estimates were also provided using different cut-offs of the annual PM reduction targets (≥ 10%, ≥ 12%, and ≥ 15%) other than ≥5%.  Map of study areas with changes in annual mean of PM2.5 from 2014 to 2017. The study participants were designated into control (black dot) and intervention (red dot) groups, according to the preestablished target of reduced concentration of ne particulate matters (PM2.5) after 2014. Intervention group: target of annual PM2.5 reduction ≥ 5% under CCAA; Control group: without reduction target under CCAA. In the present analysis, we used the data collected in the 7th (2014) and 8th (2018) waves of the CLHLS. These two waves spanned an ideal time period for exploring a potentially causal relationship between cognitive function and air pollution using a quasi-experimental framework, as PM2.5 exposure reduced substantially in a fraction of CLHLS participants while economic, socio-behavioral, and other factors remained relatively stable over these 4 years. The analytical samples comprised individuals who (1) were free of dementia, stroke, and bed-ridden status at the 2014 wave, (2) had cognitive function examinations at the both waves, (3) provided primary residential addresses, allowing for assessing air pollution exposures, (4) were non-movers during follow-up, and (5) had no missing data for any of the confounding variables. Our nal sample size consisted of 2812 participants (Supplementary Methods, Flow chart). CLHLS was approved by the Biomedical Ethics Committee, Peking University (IRB00001052-13074). All participants provided written informed consent prior to participation. Note: The designations employed and the presentation of the material on this map do not imply the expression of any opinion whatsoever on the part of Research Square concerning the legal status of any country, territory, city or