Real-World Evidence of Multiple Air Pollutants and Mortality: A Prospective Cohort Study in an Oldest-Old Population

We aimed to report real-world longitudinal ambient air pollutants levels compared to WHO Air Quality Guidelines (AQG) and analyze multiple air pollutants’ joint effect on longevity, and the modification and confounding from the climate and urbanization with a focus on the oldest-old. This study included 13,207 old participants with 73.3% aged 80 and beyond, followed up from 2008 to 2018 in 23 Chinese provinces. We used the Cox-proportional hazards model and quantile-based g-computation model to measure separate and joint effects of the multiple pollutants. We adjusted for climate and area economic factors based on a directed acyclic graph. In 2018, no participants met the WHO AQG for PM2.5 and O3, and about one-third met the AQG for NO2. The hazard ratio (HR) for mortality was 1.07 (95% confidence interval-CI: 1.05, 1.09) per decile increase in all three pollutants, with PM2.5 being the dominant contributor according to the quantile-based g-computation model. In the three-pollutant model, the HRs (95% CI) for PM2.5 and NO2 were 1.27 (1.25, 1.3) and 1.08 (1.05, 1.12) per 10 μg/m3 increase, respectively. The oldest-old experienced a much lower mortality risk from air pollution compared to the young-old. The mortality risk of PM2.5 was higher in areas with higher annual average temperatures. The adjustment of road density considerably intensified the association between NO2 and mortality. The ambient PM2.5 and O3 levels in China exceeded the WHO AQG target substantially. Multiple pollutants coexposure, confounding, and modification of the district economic and climate factors should not be ignored in the association between air pollution and mortality.


■ INTRODUCTION
According to the World Health Organization (WHO), 99% of the world's population lives in places where air pollution levels exceed WHO guideline limits.The WHO introduced an updated air quality guideline (AQG) in 2021 and established new interim targets for common air pollutants, such as fine particulate matter (PM 2.5 ), nitrogen dioxide (NO 2 ), ozone (O 3 ), sulfur dioxide, and carbon monoxide, 1 based on systematic reviews of mortality risks in the target pollutants. 2,3t lowered the recommended annual average concentration of PM 2.5 from 10 to 5 μg/m 3.1 Since most studies included in the guideline meta-analyses and systematic reviews are from North America and Europe, where air pollution mixtures and exposure ranges differ from other regions, this will result in uncertainty in concentration−response curves. 4The annual mean levels of PM 2.5 and the mortality rate attributed to ambient air pollution in China were much higher than Northern America and Europe. 5The United Nations' Sustainable Development Goals (SDGs) report in 2022 evaluated China's progress toward SDGs 3 and 11 regarding air pollution and related health as "significant challenges remain". 6r pollution and mortality longitudinal studies have emerged in China and mainly focused on the separate association of a single pollutant and health outcome or only adopted multiple-pollutant model for mutual adjustment. 7,8ew air pollutants occur in isolation, so a better understanding of air pollution on premature mortality must take the synergistic effects of pollutant mixtures and coexposures of correlated multiple pollutants into account to establish causal relationships.While some large cohort studies in developed countries suggest PM 2.5 , O 3 , and NO 2 were causally associated with a greater mortality risk and that PM 2.5 posed the greatest risk, 9 systematic reviews indicate the quality of evidence for the long-term NO 2 and O 3 with all-cause mortality risk is considered moderate. 2,3Furthermore, our previous findings showed that NO 2 's association with mortality was explained by PM 2.5. 10 It is worth noting that NO 2 exhibits a positive correlation with road traffic, which serves as an economic indicator in developing countries like China.However, there is a notable absence of adjustments for these potential districtlevel confounders in studies investigating the relationship between air pollutants and health.
There have been limited cohorts specifically targeting the oldest-old (aged 80 years and above) population, particularly centenarians.The vast geographic and demographic diversity in China also provide opportunities to investigate various vulnerable populations and district-level disparities related to environmental health.A previous study found per capita NOx emission inequality increased from 2006 to 2015 in China. 11he author then conducted an ecological study that demonstrated that this air pollution inequality impacted health inequality based on 31 Chinese provinces. 12Despite these data at the macro level, we need real-world evidence at the individual level for better understanding and improvement in air pollution health disparity.
First, our study utilized a longitudinal cohort study of China's oldest-old population to investigate the temporal and spatial variations of three air pollutants: PM 2.5 , NO 2 , and O 3 .Second, we hypothesized that these three air pollutants would exhibit positive associations with mortality risk and used multiple air pollutants models to calculate the separate and joint health effects of air pollutants on mortality.Third, we aimed to identify vulnerable populations to air pollution among those with different demographic characteristics, lifestyles, and comorbidities, with an emphasis on the oldestold group.Furthermore, we examined the possible confounding and modification of the area-level factors.

Study Population
We used the Chinese Longitudinal Healthy Longevity Survey (CLHLS).The CLHLS project has the largest sample of centenarians in the world based on our knowledge (plus compatible groups of nonagenarians, octogenarians, and young-old aged 65−79).The detailed cohort design has been described previously. 13It is a longitudinal cohort designed to study longevity and aging.The survey was conceived initially to study the senior population's socioeconomic characteristics, family, lifestyle, and demographic profile.The CLHLS adopted a targeted random-sample design to ensure representativeness, even distribution across age and sex, and sufficient subsample size of the oldest-old aged 80+, plus compatible young-old aged 65− 79.Since the 2002 wave, the CLHLS was expanded from recruiting only the oldest-old in 1998 and 2000 waves to also interviewing approximately three randomly selected nearby elders aged 65−79 of predefined age and sex in conjunction with every two centenarians. 14he questionnaire design was based on international standards and was adapted to the Chinese cultural/social context and carefully tested by pilot studies and interviews. 14Analyses of health measures showed high reliability and validity on evaluated items and exceeded widely used criteria. 15We used the 2008/2009 cohort of CLHLS with urban and rural coverage in 23 provinces.The participants were enrolled in 2008 or 2009 and followed up to 2018 roughly every two years.We overlaid environmental exposure data based on the residential area with remote sensing.We included 13,207 participants after excluding 3,747 participants from the 2008/2009 cohort (see Supplementary Methods in the Supporting Information).The average age (87 vs 86), proportion of females (58% vs 55%), annual average O 3 exposure (92 vs 92 μg/m 3 ), and NO 2 exposure (19 vs 17 μg/m 3 ) were similar between the included and excluded groups.The included participants had more participants without schooling (64% vs 55%), and more participants living in rural areas (64% vs 50%), and higher annual average PM 2.5 exposure (53 vs 48 μg/m 3 ) than the excluded participants.
The research ethics committees of Peking University (IRB00001052-13074) and Duke University approved the study.All participants in the study have given informed consent.

Air Pollutant Exposure Assessment
We used the annual PM 2.5 data at 0.01°× 0.01°calculated by the Atmospheric Composition Analysis Group.They estimated the ground-level PM 2.5 for 1998−2020 (V5.GL.02) by combining Aerosol Optical Depth (AOD) retrievals from the NASA MODIS, MISR, and SeaWIFS instruments with the GEOS-Chem chemical transport model and subsequently calibrating to global ground-based observations using a Geographically Weighted Regression (GWR), as detailed in ref 16.Annual mean PM 2.5 estimates exhibit a general consistency with ground-based observations with the coefficient of determination (R 2 ) as 0.68−0.91 for Asia. 16The annual concentrations of nitrogen dioxide (NO 2 ) concentration levels (μg/m 3 ) were obtained at one-kilometer spatial resolution using a land-use regression model corrected for satellite pass time and cloud coverage. 17The land-use regression model was corrected for satellite pass time, cloud coverage was directly used for urban areas, and model performance differed regionally with R 2 as 0.52 in Asia.For rural areas, NO 2 concentrations were adjusted using surface NO 2 concentrations derived from the Ozone Monitoring Instrument satellite NO 2 columns, and the correlation between the estimated surface concentrations and ground measurements is improved from Pearson's correlation coefficient (r) of 0.51 in the original product to 0.58. 18,19A nationwide daily maximum 8-h average (MDA8) at a resolution of 0.1°× 0.1°prediction model based on the eXtreme Gradient Boosting (XGBoost) algorithm was established by combining the MDA8 ozone observations from 2013 to 2017 with concurrent ozone retrievals, aerosol reanalysis, meteorological parameters, and land-use data. 20External model testing R 2 ranged from 0.60 to 0.87 at the month level in different years. 20We matched the air pollutants concentrations to the residence of the participants.We calculated annual ambient PM 2.5 , NO 2 , O 3 and the last peakseason (May to September) ambient O 3 concentration the participants experienced.

Mortality Outcome Assessment
The immediate family members of the subjects reported the mortality information in the follow-up surveys.The date of death would be validated by death certificates when available.We used all-cause mortality.We measured the survival time in months from the first interview to the recorded death date or last interview date.

Climate-Related Factors Measurement
Daily meteorological data of the weather monitoring stations across China between 2008 to 2018 was obtained from the China Meteorological Administration.Each study participant was matched with meteorological data collected from a monitoring station closest to their residence.We used the annual average and standard deviation of the daily mean temperature within the year as the two variables in our analyses.The season of the death or last survey month was spring (March−May), summer (June−August), fall (September−November), and winter (December−February).The elevation data are from SRTM 90m DEM Digital Elevation Database. 21e divided the geographical region based on residential location to account for climate and dietary differences: central China (Henan, Anhui, Jiangxi, Hubei, Hunan), eastern China (Shandong, Shanghai, Jiangsu, Zhejiang, Fujian), northeastern China (Heilongjiang, Jilin, and Liaoning), northern China (Beijing, Tianjin, Hebei, Shanxi, Shaanxi), southern China (Guangdong, Guangxi, and Hainan), and southwestern China (Chongqing and Sichuan).

Area Economic Measurement
We followed the CLHLS residence categories: "Urban" (including "City" and "Town") and "Rural".We defined the road density as the sum of road length within 5 km buffer around the residence.We filtered highway categories "motorway", "trunk", "primary", "secon-dary", "tertiary", and "residential" in the China street map in 2022 from OpenStreetMap (OSM) to calculate the weighted road length sum:

Other Covariates and Modifiers Measurements
We considered baseline characteristics, including age, sex, marital status, education, smoking status, drinking status, physical activity, household income, coresidence, housework frequency, social activity frequency, Mini-Mental State Examination (MMSE), Activities of Daily Living (ADL) disability and Body Mass Index (BMI).The detailed definition is shown in the Supplementary Methods (see the Supporting Information).

Statistical Analysis
Given the open cohort nature of our cohort with various subjects contributing different person-times to the analysis, we decided to use the Cox proportional hazards model to examine the association between long-term PM 2.5 , NO 2 , and O 3 exposure and all-cause mortality.We ran a series of models to test the robustness of the association.The first model was the unadjusted model (Model A), and then it was gradually adjusted for potential confounders based on the Directed Acyclic Graph (DAG) (Figure S1) or predictors of the outcome: age, sex, education, household income, marriage, coresidence, exercise, smoking, alcohol drinking, housework frequency, social activity frequency, residence (Model B), region, temperature annual mean, temperature annual SD, elevation, season of death (Model C), road density, city GDP per capita, city population density (Model D), BMI, MMSE score, and ADL score (Model E).The air pollutants were further mutually adjusted in the multiple pollutant model (Model F to K). Considering the possible multicollinearity of the three pollutants and to assess the joint effect of the pollution mixture on mortality, we used the quantile-based gcomputation model. 22Quantile-based g-computation estimates the overall mixture effect with the same procedure used by a weighted quantile sum (same formula as below), but estimates the parameters of a marginal structural model, rather than a standard regression.
ψ is equivalent to the g-computation estimator 23 of a joint marginal structural model for quantized exposures, which estimates the effect of increasing every exposure simultaneously by one quantile.w j are the weights for each exposure (here we have d exposures).ε is the error term.X j q is the quantized version of the jth exposure.β j is the effect size for exposure j.
To assess for nonlinearity, we used the restricted cubic spline to describe the concentration−response relationship.Susceptible population and area disparity were tested via stratified analyses.We further examined the association between the last two years and three years air pollution exposure and mortality for sensitivity analysis.We used R version 4.2.1 to do all the analysis.

■ RESULTS
We included 13,207 participants with a mean age of 87 (SD: 11), and 7634 (57.8%) were females.The centenarian group consisted of 81.3% females (n = 2119).A total of 9245 deaths occurred during the 61,082 person-year follow-up period.Most participants did not receive a formal education (64.1%), did

Environment & Health
not live with a married spouse (69.4%), lived with family (83.4%), reported a lack of or inability to undertake regular exercise (60.9%) or social activities (89.2%), were never smokers (66.7%), or never alcohol drinker (69.2%), resided in rural areas (63.9%), or did not have disabilities (78.7%) (Tables 1 and S1).The oldest-old had a slightly higher exposure to PM 2.5 and NO 2 than the young-old.Their residential areas were also characterized by fewer roads and lower GDP per capita.Additionally, they had decreased engagement in housework and social activities, and a higher prevalence of ADL disability and cognitive impairment than their counterparts (Table S1).Geographical variations in air pollution levels were observed in China (Table S2).The annual ambient average PM 2.5 was the highest in places with heavy industry or colder climates, such as central China, northern China, places with the highest annual temperature variation, or with the highest population density.The median of NO 2 was higher in northern and northeastern China, urban areas, among individuals with more than primary school education, around high road density, and high annual temperature variation than their counterparts.The peak season O 3 (μg/m 3 ) was highest in eastern China, places with the lowest elevation, highest GDP per capita, highest population density, lower annual average temperature, or higher annual temperature variation.
The restricted cubic spline for PM 2.5 was supralinear (concave downward), meaning there were larger changes in risk for low concentrations compared to higher concentrations.We identified a subtle curvature in the association between NO 2 and mortality, with a more pronounced marginal effect occurring at higher levels of NO 2 exposure.The HR for annual O 3 increased at O 3 levels lower than 88 μg/m 3 and subsequently decreased to less than one as O 3 levels increased beyond 110 μg/m 3 .Similarly, the HR for peak season O 3 decreased to less than one when O 3 was higher than approximately 110 μg/m 3 (Figure 2).The p value for nonlinear Wald statistics were all less than 0.0001, suggesting nonlinear associations between the three pollutants and mortality.
In the stratified analysis, PM 2.5 and NO 2 were found to have a stronger harmful association with mortality for those under 80 years old compared to those oldest-old.There was no significant difference in HRs for females and males.Those who were married and currently living with their spouse, smoked more, drank alcohol more, reported exercising currently, doing housework every day, with higher MMSE scores, and without ADL disability had a higher mortality risk related to PM 2.5 and NO 2 than their counterparts (Figure S2).
The mortality risk associated with increasing PM 2.5 was higher in southern China [HR (95% CI): 2.24, (2.07, 2.44)] and southwestern China [2.19 (2.05, 2.34)] than in other regions, and it tended to increase with the annual average temperature increasing and the temperature annual SD decreasing.The HR of PM 2.5 became not statistically significant in the northeastern region [1.00 (0.91, 1.12)].Annual O 3 only had an HR greater than one in southern China (including Guangdong Guangxi, and Hainan) [1.14 (1.07, 1.21)] and the areas with the highest annual average temperature and the lowest temperature annual SD.The HR of NO 2 was higher in eastern China [1.37 (1.30, 1.44)], and it increased as the temperature annual SD became larger.Additionally, PM 2.5 had a slightly higher HR in rural areas compared to urban areas.We did not observe linear trends for the HR of PM 2.5 or NO 2 with the increase of road density, GDP per capita, or population density (Figure 3).

■ DISCUSSION
Among over ten thousand older adults in our study, the average PM 2.5 and peak season O 3 levels did not meet the WHO guideline interim target one.In contrast, the annual average NO 2 has met the AQG level since 2015.The study found PM 2.5 was the dominant mortality risk factor among the three pollutants.Interestingly, the oldest-old were less vulnerable to PM 2.5 and NO 2 than those younger than 80.The associations between air pollutants and mortality also varied in different districts of China and were influenced by temperature, road density, GDP per capita, and population density.
China has made policy efforts to reduce air pollution over the past decades, including the issuance of the Atmosphere Ten Articles in 2013 and the Blue Sky Defense War in 2018.These efforts have led to significant improvements in China's air quality, with the ambient exposure for our study participants decreasing greatly from 59.6 to 39.4 μg/m 3 during 2013−2019.According to the WHO, the data reporter of SDG target 11.6.2, the annual mean levels (μg/m 3 ) of PM  5 However, despite the decrease, the average levels of PM 2.5 in the sampled areas in 2019 were still higher than the WHO interim target one of 35 μg/m 3 .The majority of places were also above the WHO AQG second interim target of 25 μg/m 3 .
Unlike developed countries, the change of NO 2 in our study was relatively small, and the annual average level was lower than the WHO AQG level of 10 μg/m 3 since 2014.NO 2 increased from 2006 to 2011 and decreased since 2011.The change trend is similar to the findings of another study, which estimated the annual mean increase rate as 0.348 ± 0.132 μg/ m 3 between 2005 to 2011, and an annual mean rate of decrease as 0.312 ± 0.188 μg/m 3 between 2012 and 2019. 24 3 has increased while PM 2.5 decreased.O 3 is formed through complex radical chain-reaction chemistry.There was a potential trade-off between O 3 precursors emissions-reducing and particle pollution-reducing. 25The majority areas in mainland China also suffered peak season ozone concentrations that were higher than the first interim target: 100 μg/ m 3 of WHO 2021 guideline (target 2:70; AQG: 60 μg/m 3 ). 26eak season average O 3 remained around 108 μg/m 3 before 2015 and increased to around 123 μg/m 3 in 2017 in our study.According to another study, the population-weighted median concentrations of predicted MDA8 ozone was 89.34 μg/m 3 in mainland China in 2013 and reached 100.96 μg/m 3 in 2019. 26he gap can be due to the difference in the sampling district, O 3 source and estimation methods.There is no long-term annual limit for O 3 in the Environment Air Quality Standard (GB3095-2012) of China so far.NO 2 and O 3 have not been included in the WHO SDG related indicator report yet.
The range of the last year PM 2.5 was 14.8−133.0μg/m 3 , with the lowest level higher than the WHO AQG target.The concentration−response curve showed no safe threshold for increased mortality risk.The steep relationship within about 60 μg/m 3 indicated a stronger association in the relatively low exposure range than in the high range exposure.The last year's annual average NO 2 ranged from 1.2 to 109.0 μg/m 3 , with the lowest level lower than the WHO AQG level.The mortality risk also increased with the NO 2 concentration increasing, and there was also no threshold.This should be new evidence for the guideline update.Notably, the source of NO 2 was mainly the traffic.Road density was positively associated with NO 2 and also negatively associated with mortality [HR (95% CI) of the fourth quartile vs first quartile: 0.72 (0.66, 0.78)].Road density, GDP per capita, and population density were neglected confounders in most previous NO 2 studies.Peak season O 3 ranged from 20 to 182 μg/m 3 , with the lowest level lower than the WHO AQG level.Higher O 3 was associated with lower mortality risk on average [HR (95% CI): 0.87 (0.86, 0.89) for annual O 3 and 0.89 (0.87, 0.90) for peak season O 3 ].Previous evidence on long-term O 3 and mortality risk is modest.Pooled results based on nine studies showed no significant association between increased all-year O 3 exposure and all-cause mortality as HR being 0.97 (95% CI: 0.93, 1.02) per 10 μg/m 3 with a large heterogeneity. 27A more recent cohort based meta-analysis found a positive association between long-term O 3 and all-cause mortality [HR(95% CI): 1.014, 1.009−1.019for 10 nmol mol −1 incremental warm season O 3 ]. 28However, this meta-analysis used large-scale O 3 exposure health risk studies up to 2022, which did not cover Asian, African, or Latin American regions.The latest finding based on another Chinese cohort identified an HR of 1.18 (95% CI: 1.13, 1.23) per 10 μg/m 3 increase of O 3 and all-cause mortality. 29Their population had a mean age of 57.7 (SD: 10.4), while our study population was aged 80 on average.The different sources and resolution of O 3 exposure could be other reasons for the contradictory findings.Besides, higher O 3 was associated with higher mortality risk in the low range of about less than 110 μg/m 3 , and then the HR decreased to less than one in the high range of O 3 .Among those who were exposed to high ozone concentrations (exceeding 100 μg/m 3 ), a substantial majority resided in eastern China, which is the most developed district of China.We suspect that there may be residual social economic confounding.Furthermore, the associations between NO 2 and mortality became gradually weaker for longer exposure time windows of the last two or three years, while it was stable for PM 2.5 and O 3 (Table S3).More discussions on the separate association between air pollutants and mortality are in the extended discussion (see Extended discussion in Supporting Information).
O 3 and secondary PM 2.5 are both pollutants produced through chemical chain reactions of volatile organic compounds (VOCs) and oxides of nitrogen (NOx).Two-pollutant models can be difficult to interpret when the correlation between pollutants is high or exposures for pollutants are assessed with different methods or at a different spatial resolution. 2 In the review, overall RRs for PM 2.5 were much lower in studies that specified two pollutant models adjusted for NO 2 compared to the single pollutant estimates [1.02 (95% CI 1.00, 1.04) in two-pollutant models versus 1.07 (95% CI 1.05, 1.08) in single pollutant models].RRs remained stable after adjusting for O 3 : 1.08 (95% CI 1.04, 1.11) based upon seven studies, respectively. 2Associations between NO 2 or O 3 and mortality were attenuated upon adjustment for copollutants in some studies but not in others. 3In our study, the HR of PM 2.5 increased after adjusting for peak season O 3 , and the HR of NO 2 was attenuated significantly after adjusting for PM 2.5 , and the HR of O 3 did not change much after adjusting for PM 2.5 or NO 2 .In our previous study, the association between NO 2 and mortality even became insignificant after adjusting for PM 2.5 without adjustment of road density. 10The different sources of our PM 2.5 , NO 2 , and O 3 data can limit the comparability.The correlations were not very strong, with the highest Pearson coefficient as 0.37 between PM 2.5 and NO 2 .We also noticed that in another national study in China, the HR for mortality of NO 2 even doubled after adjusting for PM 2.5. 30As discussed above, the population age, air pollution source, and resolution may lead to the inconsistency.PM 2.5 was still the dominant pollutant associated with mortality in our study.A study using U.S. Medicare data also indicates that PM 2.5 posed the greatest risk among PM 2.5 , NO 2 , and O 3. 9 The LIFEWORK study in The Netherlands found PM 2.5 was the

Environment & Health
most relevant contributor to the positive association with the mortality among a mixture of five components: PM 2.5 , NO 2 , etc. 31 We also saw air pollution mortality risk was modified by individual level characteristics.It is thought there was a higher risk for individuals living in a lower socioeconomic status (SES).Those with higher SES tended to have more advantages in life habits and resources, health care access, and air pollution exposure reduction.In our study, relatively younger people, those living with a spouse or without cognitive impairment or ADL disability, were surprisingly more susceptible to PM 2.5 and NO 2 .This may be related to more exposure to ambient air pollution through more outdoor activities for those younger and healthier participants.Another possible reason may be that other health issues afflicting the elderly population masked the effects of air pollutants.A Hong Kong study also found a higher risk for the younger group and stated this might be explained by the healthy survivor effect 32 that those who survived are more resistant.The concentration−response relationship was also reported to be different in different ages.A study in Norway found that the risk of death of NO 2 from all-causes started to increase from 40 μg/m 3 in younger subjects, whereas the relationship was linear across the concentration range (2−73 μg/m 3 ) in the oldest age group. 33Males had a nonsignificant higher mortality risk of both PM 2.5 and NO 2 than females, which was consistent with a previous study in Hong Kong, 32 while some studies in the USA reported a significant gender difference. 34,35WHO reported age-standardized mortality rate (95% CI) attributed to ambient air pollution in China (per 100 000 population) was higher in males at 82.69 (64.83−102.8)compared to females at 48.41 (37.58−60.46). 5The mortality risk of those having a higher household income was impacted less by the air pollution in our study.But we also found that those with higher education did not have a lower risk for either PM 2.5 or NO 2 .The stratified HRs for PM 2.5 had a nonlinear trend with education level increasing and indicated weaker associations in the highest and lowest education group and stronger associations among the middle-level education group.We further found that this PM 2.5 risk difference for the two education groups only existed in females, and not in males.People with less education in the US were at higher risk of mortality associated with air pollution exposure. 36The associations did not differ consistently by educational level in three Latin American cities. 37 Another inconsistency was that there were different demographic and SES indexes in different countries.For example, there was a great difference in the air pollution risk among different races 34 in the US population, while race may not be a significant factor in other countries.For different lifestyles, heavy smokers suffered greater from both PM 2.5 and NO 2 , and heavy drinkers suffered greater from PM 2.5 in our study.
The areal-level SES can interact with individual SES and play a role in the association with health.The effect size of PM 2.5 for mortality was slightly larger in rural areas than in urban areas and tended to be higher in areas with lower road density and low population density.In China, a developing country, areas with higher road density and urban areas can have higher air pollution levels but also usually have richer social resources and higher economic levels.On the one hand, residents living in such areas can have a higher awareness of air pollution and adopt more methods to reduce personal exposure to air pollution.On the other hand, rich social resources like good health care in places with convenient transportation could reduce the mortality of the residents.
The effect size of PM 2.5 for mortality was larger in the South than in the North of China despite PM 2.5 being higher in the north than the south in our study, which was consistent with another study that used the same cohort for the Chinese old population aged 65 or older. 7O 3 was also only harmful in southern areas in our study.The different geographic characteristics, climates, and lifestyles between northern and southern China can play a role.When stratified by temperature, PM 2.5 and O 3 both showed higher risk in areas with higher temperature and lower temperature variations.The frequency of opening windows in southern areas was also usually higher than in the north in winter, which can impact indoor air pollution exposure, especially for the O 3 , whose concentration was very different between indoors and outdoors.Another study for Chinese men aged 40 or older found a higher risk for the north than the south. 38They sampled different provinces, included only males, and surveyed a younger population compared to our study design.
Individual-level, area-level SES, and air pollution are intertwined. 39A multilevel cohort study in Norway found that the effect of social deprivation at the neighborhood level was independent of and stronger than the individual social deprivation in the association between air pollution and mortality. 40We also found that the geographic region affected the effect of PM 2.5 and O 3 , and the road density, GDP, population density, and temperature affected the effect of PM 2.5 and NO 2 greatly.The residual area-level confounding may help explain the mixed results for the stratified analysis.These differences need further validation and call for more sensitivity analysis for subpopulations and specific contexts.It also indicated a need for specialized standards in addition to the overall target.
Our study has several strengths.We used a large cohort of oldest-old individuals, and followed up for 10 years, across multiple geographical areas.We accounted for multiple air pollutants over a long period of time.We adjusted for a variety of possible individual and area-level confounders such as age, sex, individual SES, individual smoking, other individual lifestyles, and area-level environment and economic confounders like road density, GDP per capita, climate region, and temperature.
However, there are also some limitations to our study.First, we used ambient air pollution derived from satellite data, which may not accurately represent real exposure.Second, different data sources and estimation methods were used for the air pollutants, which may limit comparability in the multipollutant model.Third, we used all-cause mortality, which limited our ability to test cause-specific mortality.Also, we did not consider comorbidities which can be a source of confounding.Lastly, our sample was mainly composed of participants aged 80 or older, and the majority lived in rural areas, which may limit the generalizability of our findings.

■ CONCLUSIONS
Our study provides real-world evidence comparing to WHO AQG that contributes to a better understanding and addressing the indicators outlined in SDG 3.9 and 11.6, specifically focusing on advanced-aged adults in specific populations and regions.We observed a significant gap between current levels of PM 2.5 , O 3 and the World Health Organization's Air Quality Guideline (AQG) level, highlighting the need for continued Environment & Health efforts to reduce PM 2.5 exposure.Additionally, despite the increase in road density, it is crucial to prioritize measures aimed at maintaining low levels of NO 2 .Further research is necessary to fully comprehend the relationship between individual and area-level SES and the risk posed by air pollution.

* sı Supporting Information
The Supporting Information is available free of charge at https://pubs.acs.org/doi/

Figure 1 .
Figure 1.Change trend of the air pollution around the participants' residence from 2006 to 2019.(a) Note: Abbreviation: AGQ − Air Quality Guideline.(b) In September 2013, the China State Council introduced a policy known as the "Atmosphere Ten Articles."This policy outlined ten measures aimed at preventing and controlling air pollution including industrial restructuring, clean energy promotion, etc.In June 2018, the China State Council printed and distributed the Three-Year Action Plan for Winning the Blue Sky Defense War aiming to reduce emissions of major air pollutants and greenhouse gases, and decrease the number of days with high air pollution.

Figure 2 .
Figure 2. The spline for the air pollutants and mortality.Note: All the models adjusted for age, sex, education, household income, marriage, coresidence, exercise, smoking, alcohol drinking, housework frequency, social activity frequency, residence, region, temperature annual mean, temperature annual standard deviation, elevation, season of death, road density, city Gross Domestic Product per capita, city population density, body mass index, Mini-Mental State Examination score, Activities of Daily Living core.

Figure 3 .
Figure 3. Association between air pollutants and mortality stratified by area characteristics.The unit: temperature -Celsius (°C), elevation -meter, weighted road length -meters in a 5 km radius, GDP per capita -RMB, population density -population per square kilometer.
The city-level GDP per capita (yuan) and population number per kilometer in 2008 were from the CNKI China Economic and Social Big Data Research Platform.

Table 2 .
Association between Air Pollutants and Mortality Risk a a Abbreviations: SD − standard deviation, GDP − Gross Domestic Product, BMI − body mass index, MMSE − Mini-Mental State Examination, ADL − Activities of Daily Living.