Exposure to air pollution and cognitive impairment risk: a meta-analysis of longitudinal cohort studies with dose-response analysis

Background We conducted a meta-analysis to explore the relationship between exposure to air pollution and the risk of cognitive impairment of longitudinal cohort studies. Methods PubMed, Web of Science and Wan Fang databases were searched for relevant articles of longitudinal cohort studies published between January 1950 and September 2019. The pooled relative ratio (RR) and 95% confidence interval (CI) were calculated using the random effect model. Results Ten articles involving 519 247 cases among 12 523 553 participants were included in this meta-analysis. The pooled RR of cognitive impairment per 5 μg/m3 increments in exposure to particulate matter ≤2.5 μm (PM2.5) was 1.08 (95% CI = 1.03, 1.13; I2 = 82.2%; Pheterogeneity<0.001). No association was found between nitrogen dioxide/nitrogen oxide (NO2/NOx) and ozone (O3) and cognitive impairment. For PM2.5 exposure, in subgroup analysis, the above-mentioned significant positive association was found among studies conducted in population (RRper 5 μg/m3 = 1.05; 95% CI = 1.01,1.09; I2 = 57.4%; Pheterogeneity = 0.016), in North America (RRper 5 μg/m3 = 1.13; 95% CI = 1.01,1.26; I2 = 86.7%; Pheterogeneity<0.001) and with follow-up duration >10 years (RRper 5 μg/m3 = 1.10; 95% CI = 1.03,1.17; I2 = 86.3%; Pheterogeneity<0.001). Conclusions This meta-analysis suggests that exposure to PM2.5 might increase the risk of cognitive impairment.


VIEWPOINTS PAPERS
other studies [20,21,25,26]. For O3 exposure, one study [17] revealed an inverse relationship with cognitive impairment, while another study [20] found a positive relationship with cognitive impairment. In light of the inconsistencies among the above epidemiological studies, we conducted a meta-analysis of longitudinal cohort studies to synthesize the results of existing studies to evaluate the relationship of air pollution with cognitive impairment.

METHODS
This meta-analysis was conducted according to the Preferred Reporting Items for Systematic reviews and Meta-Analyses (PRISMA) guidelines [27].

Search strategy
We searched all relevant articles in English or Chinese from PubMed, Web of Science and Wan Fang databases published between January 1950 and September 2019. The following search strategy was used: (air pollution or particulate matter or carbon monoxide or nitrogen dioxide or nitrogen oxide or sulfur dioxide or ozone) And (cognitive impairment or dementia). The exhaustive search process in PubMed is shown in Table S1 in the Online Supplementary Document. Furthermore, we reviewed the reference lists of retrieved articles to identify additional relevant articles.

Inclusion criteria
The included studies ought to meet the following criteria: (1) longitudinal cohort studies published in English or Chinese; (2) the exposure of interest was particulate (PM2.5, particles with an aerodynamic diameter ≤2.5 μm; PM10, particles with an aerodynamic diameter ≤10 μm or PM2.5-10, particles with an aerodynamic diameter 2.5-10) or gaseous (NO2, nitrogen dioxide; NOx, nitrogen oxide; O3, Ozone; CO, carbon monoxide; SO2, sulfur dioxide) air pollutants; (3) the outcome of interest was cognitive impairment mainly including dementia and AD; (4) Multivariate adjusted odds ratio (OR), relative risk (RR) or hazard ratio (HR) and their 95% confidence interval (CI) of cognitive impairment with exposures to air pollution with per unit increase in pollutant concentrations (μg/m 3 , mg/m 3 , ppb, or ppm) were available, or sufficient data could be used to convert these results.
Two investigators independently performed the literature search. If there was disagreement on an article, the consensus was reached through discussion. If data appeared in more than one study, we would select the most comprehensive data.

Data extraction
Data extracted from each identified study by two investigators independently as follows: the first author' s name, publication year, study areas and country, age range or mean age of participants, sex, follow-up duration, sample size, number of cases, measurement of exposure, the source of outcome, RRs (we presented all results with RR for simplicity) and their 95% CI with per unit increase in particulate (PM2.5, PM10, or PM2.5-10) or gaseous (NO2, NOx, O3, CO, SO2) air pollutants concentrations (μg/m 3 , mg/m 3 , ppb, or ppm), adjusted covariates.
The included cohort literature was assessed by the Newcastle-Ottawa Scale (NOS) [28].

Statistical analysis
We use the inverse variance and the number of participants to weight the study-specific log RRs that can calculate the pooled RR with corresponding 95% CI to evaluate the relationship between air pollution and cognitive impairment. For consistency, all RRs were standardized to an increment of 5 μg/m 3 of particular matter (PM2.5, PM2.5-10, and PM10) concentration and 5 ppb of gaseous (NO2, NOx, O3, CO, VIEWPOINTS PAPERS SO2) concentration. It is hypothesized that there is a linear relationship of exposure to air pollution and cognitive impairment. If studies reported RRu per u units not in per 5 μg/m 3 of particular matter (PM2.5, PM2.5-10, and PM10) or 5 ppb of gaseous (NO2, NOx, O3, CO, SO2), the RRstandardized was computed by the following formula [29]: RRstandardized = RRu increment unit (eg, 5)/u Where u represents the increment utilized in the original study to assess the effects. We used a unit conversion factor: 1 ppb = 1.88 μg/m 3 for NO2/NOx and 1 ppb = 1.96 μg/m 3 for O3, and standard condition is ambient pressure of 1 atmosphere and a temperature of 25°C [30]. We adopted to I 2 statistic to evaluate the heterogeneity, and the I 2 values of 0%, 25%, 50%, and 75% represent no, low, moderate and high heterogeneity, respectively [31]. The random effect model (REM) was utilized as the pooling method. Meta-regression with restricted maximum likelihood estimation was used to explore the potential covariates that may exert substantial impacts on between-study heterogeneity [32]. Subgroup analysis was conducted by the source of outcome, study areas, follow-up duration, and adjustment status for comorbidity and smoking. We carried out the leave-one-out sensitivity analysis to examine whether individual study influences between-study heterogeneity [33]. Influence analysis was performed excluding one study at a time to ascertain whether the aggregate results could be significantly affected by a single study [34]. Additionally, the funnel plot and Egger regression asymmetry test were used to estimate the publication bias [35]. All statistical analyses were done using STATA Version 15.0 (STATA Corporation, College Station, TX, USA). A two-sided P ≤ 0.05 was equated statistically significant.

Search results and study characteristics
On the basis of search strategy, a total of 11 090 articles were identified, involving 1085 articles from PubMed, 382 articles from Web of Science, 9622 articles from Wan Fang and one article from reference lists. After removing 207 articles because of duplicates, there are 10 883 articles left. Reviewing the title and abstract later, forty-one articles were remaining. We further excluded thirty-one articles after reviewing the full-text. The flow diagram of the literature search is displayed in Figure 1. The detailed reason for full-text reviewed articles exclusion is provided in Table S2 in the Online Supplementary Document.
The quality assessment indicated that the Newcastle-Ottawa score of cohort studies was not less than 7. Specific quality assessments are presented in Table S3 in the Online Supplementary Document. The characteristics of the studies are shown in Table 1. Table 2 shows the summary risk estimates of cognitive impairment for air pollution according to study characteristics.
We carried out subgroup analysis by the source of outcome, study areas, follow-up duration, and adjustment status for comorbidity and smoking. NO2/NOx revealed a statistically significant positive association with the risk of cognitive impairment in population-based studies (RRper 5 ppb = 1.048; 95% CI = 1.004, 1.094; I 2 = 68.5%; Pheterogeneity = 0.013), whereas no association was found in hospital-based studies (

Meta-regression
As displayed in Figure 2, high heterogeneity (I 2 = 82.8%, Pheterogeneity = 0.030) was found in the analysis of PM2.5 and the risk of cognitive impairment. Hence, we performed meta-regression with the covariates of  the source of outcome (P = 0.663), study areas (P = 0.403), follow-up duration (P = 0.993), whether adjusted comorbidity (P = 0.679) and whether adjusted smoking (P = 0.475) to explore potential sources of the heterogeneity. However, no covariates contributed to the heterogeneity of between-study.
In the analysis of NO2/NOx and the risk of cognitive impairment, as displayed in Figure 3 high heterogeneity (I 2 = 93.4%, Pheterogeneity<0.001) was found. The meta-regression with the covariates of the source of outcome (P = 0.219), study areas (P = 0.065), follow-up duration (P = 0.053), whether adjusted comorbidity (P = 0.979) and whether adjusted smoking (P = 0.540) to explore potential sources of the heterogeneity, and no covariates contributed to the heterogeneity.
In the influence analysis, no individual study had an overmuch impact on the pooled effect for PM2.5 and NO2/NOx with the risk of cognitive impairment ( Figure S5 in the Online Supplementary Document and Figure S6 in the Online Supplementary Document).

Publication bias
The visual scrutiny of the funnel plot ( Figure S7 in the Online Supplementary Document and Figure S8 in the Online Supplementary Document) seemed to be asymmetrical for PM2.5 and NO2/NOx, while the Egger' s test displayed no evidence of significant publication bias in the analysis between PM2.5 (P = 0.155) and NO2/NOx (P = 0.792) and the risk of cognitive impairment, respectively.

DISCUSSION
This meta-analysis of longitudinal cohort studies included 10 articles to quantitatively evaluate the association between air pollution exposure and the risk of cognitive impairment. The results demonstrated that exposure to PM2.5 was significantly related to an increased risk of cognitive impairment. More specifically, for every 5 μg/m 3 increase in PM2.5 concentration, the risk of cognitive impairment increased by 8%. Nevertheless, there is no statistical association between NO2/NOx and O3 exposure and the risk of cognitive impairment. And the correlation direction did not change after sensitivity analysis. For PM2.5 exposure, subgroup analysis showed significant associations in population-based studies, studies that conducted in North America and studies with follow-up duration >10 years.
Several biological mechanisms have been put forward for the positive relationship of particulate matter (PM) with cognitive impairment. First, exposure to air pollution, particularly PM might cause neuroinflammation, oxidative stress and change brain innate immune responses in early adulthood [36]. At present, inflammation and oxidative stress have been confirmed as basic mechanisms by which air pollution may affect central nervous system disease (CNS) [9]. Second, PM can activate microglia [37], excessive and chronic activation may lead to neurotoxicity [38]. Importantly, microglial activation has been implicated in the progression of diseases such as dementia [39]. Third, the olfactory bulb is another pathway [40] through which PM enters the body, reaching the brain directly and inducing a series of changes such as increased the level of amyloid-β42, hyperphosphorylated τ, and neural degeneration [41,42].
High heterogeneity appeared in this meta-analysis of PM2.5 and NO2/NOx and the risk of cognitive impairment. To search for potential heterogeneity, we performed the following work. Meta-regression was performed to detect the potential factors that contributed to heterogeneity between studies, however, no factors were found to do with it. We also conducted influence analysis, and the results indicated that no individual study had an excessive impact on the pooled effect of PM2.5 and NO2/NOx and the risk of cognitive impairment. For PM2.5 exposure, the leave-one-out sensitivity analysis indicated that two studies [19,20] affected the heterogeneity and after further excluding two studies, the pooled RR was decreased but not altered substantially. In two studies, the source of outcome was from the hospital-based population, which was likely to not cover all patients and could introduce selection bias.
Our study presents several advantages. First, our meta-analysis included a large number of participants from longitudinal cohort studies, providing high statistical power and making it more likely to obtain a VIEWPOINTS PAPERS reasonable conclusion. Second, the pooled RR of this meta-analysis was based on data on dose-response relationships in the original studies, thus the existence of causation was further supported. Furthermore, to minimize between-study variation, we normalized the exposure levels of PM2.5, NO2/NOx and O3 across studies into uniform units, and the pooled RR was standardized per 5μg/m 3 increments for PM2.5 and 5 ppb for NO2/NOx and O3. Third, original studies were all fully taken into account potential confounders such as age. Moreover, sex, comorbidities, and smoking were adjusted for in most studies. As well, the quality assessment score of each study was higher than 7, demonstrating that the quality of the included articles was generally good.
Nevertheless, several limitations of our meta-analysis should be recognized. First, although as much as possible potential confounders were adjusted for in most studies, several studies still did not adjust for potential confounders, such as comorbidities. Comorbidities include heart disease, stroke, diabetes, hypertension and so on, which may elevate cognitive impairment risk [43][44][45]. In the analysis of PM2.5 and cognitive impairment, a significant association was found in studies adjusting for comorbidity, but no association was found in studies that did not adjust for comorbidity, therefore the combined result was underestimated. Second, varied exposure assessment methods differed in its ability to provide estimates of individual exposure levels, which might increase the instability of the results to some extent. Third, the definition of cognitive impairment was inconsistent, resulting in misclassification bias.

CONCLUSIONS
In summary, the results of this meta-analysis demonstrated that exposure to PM2.5 was associated with an increase in cognitive impairment risk. The results may have substantial public health significance for the prevention of cognitive impairment through air pollution interventions.