Ambient Ozone, PM1 and Female Lung Cancer Incidence in 436 Chinese Counties

Ozone air pollution has been increasingly severe and has become another major air pollutant in Chinese cities, while PM1 is more harmful to human health than coarser PMs. However, nationwide studies estimating the effects of ozone and PM1 are quite limited in China. This study aims to assess the spatial associations between ozone (and PM1) and the incidence rate of female lung cancer in 436 Chinese cancer registries (counties/districts). The effects of ozone and PM1 were estimated, respectively, using statistical models controlling for time, location and socioeconomic covariates. Then, three sensitivity analyses including the adjustments of smoking covariates and co-pollutant (SO2) and the estimates of ozone, PM1 and SO2 effects in the same model, were conducted to test the robustness of the effects of the two air pollutants. Further still, we investigated the modifying role of urban–rural division on the effects of ozone and PM1. According to the results, a 10 μg/m3 increase in ozone and PM1 was associated with a 4.57% (95% CI: 4.32%, 16.16%) and 4.89% (95% CI: 4.37%, 17.56%) increase in the incidence rate of female lung cancer relative to its mean, respectively. Such ozone and PM1 effects were still significant in three sensitivity analyses. Regarding the modifying role of urban–rural division, the effect of PM1 was greater by 2.98% (95% CI: 1.01%, 4.96%) in urban than in rural areas when PM1 changed by 10 μg/m3. However, there was no modification effect of urban–rural division for ozone. In conclusion, there were positive associations between ozone (and PM1) and the incidence rate of female lung cancer in China. Urban-rural division may modify the effect of PM1 on the incidence rate of female lung cancer, which is seldom reported. Continuous and further prevention and control measures should be developed to alleviate the situation of the two air pollutants.


Introduction
There are great public health concerns due to the severe situation of air pollution in Chinese cities. Air pollution, already identified as the Group I carcinogenic factor to lung cancer diseases, can adversely affect human health through the increase in genetic damage [1]. According to the report of the State of Global Air 2019, fine particulate matter (i.e., PM 2.5 ) was responsible for approximately three million mortalities in 2017 all over the world [2]. Despite previous efforts to tackle air pollution, especially for PM 2.5 and PM 10 [3][4][5], however, the associations of human health with ambient ozone and PM 1 air pollution have not been fully understood in China.
Recently, there is increasing interest in ozone research all over China. After the implementation of clean air actions, mainly to alleviate PM 2.5 air pollution, there has been great reductions in PM 2.5 concentrations already, which have resulted primarily from the recent reduction in anthropogenic emissions [6,7]. Nonetheless, ozone concentrations have been increasing during the same period [8][9][10]. In parallel with the greatest public concern on PM 2.5 air pollution, China has been a hotspot of ozone air pollution in the world. Most

Ambient Ozone and PM1
Data on annual surface ozone concentrations at 10 km × 10 km grids between 2014 and 2016 were acquired from the ChinaHighO3 dataset (http://doi.org/10.5281/zenodo.4400043 (accessed on 30 December 2020)). This dataset is one of the full-coverage and high-quality datasets of surface air pollution in China. Technologically, the production of surface ozone data was similar to that of the ChinaHighPM2.5 dataset [31,32]. In brief, the surface ozone concentrations were estimated using the proposed Space-Time Extra-Trees (STET) model (i.e., one of the most advanced machine learning methods in the field of remote sensing estimation of air pollution concentrations). Data as model inputs mainly included the MAIAC (Multi-Angle Implementation of Atmospheric Correction) AOD (aerosol optical depth) product, meteorological factors, surface conditions (e.g., Normalized Difference Vegetation Index (NDVI) and land use cover), pollutant emissions and population distributions, as well as the terms capturing the spatial and temporal autocorrelations of surface ozone concentrations. As reported (http://doi.org/10.5281/zenodo.4400043 (accessed on 30 December 2020)), there is high consistency between the estimated ozone data and ground-based ozone measurements (R 2 = 0.87, root-mean-square error (RMSE) = 17.10 μg/m 3 ). Figure 2A presents the spatial distributions of ozone concentrations across China in 2016.
Data on annual PM1 concentrations at 1 km 2 spatial resolution between 2014 and 2016 were obtained from our previous study. More details on PM1 estimates can be found in Wei et al. [33]. Briefly, a machine learning approach, i.e., the space-time extremely randomized trees model (STET), was developed to estimate daily PM1 concentrations in terms of the geospatial data of MAIAC AOD, MEIC pollution emissions, meteorological factors, land use, road and population, as well as the spatiotemporal information which captures the spatial and temporal autocorrelations of PM1 concentrations. According to the results  Data on annual surface ozone concentrations at 10 km × 10 km grids between 2014 and 2016 were acquired from the ChinaHighO 3 dataset (http://doi.org/10.5281/zenodo. 4400043 (accessed on 30 December 2020)). This dataset is one of the full-coverage and high-quality datasets of surface air pollution in China. Technologically, the production of surface ozone data was similar to that of the ChinaHighPM 2.5 dataset [31,32]. In brief, the surface ozone concentrations were estimated using the proposed Space-Time Extra-Trees (STET) model (i.e., one of the most advanced machine learning methods in the field of remote sensing estimation of air pollution concentrations). Data as model inputs mainly included the MAIAC (Multi-Angle Implementation of Atmospheric Correction) AOD (aerosol optical depth) product, meteorological factors, surface conditions (e.g., Normalized Difference Vegetation Index (NDVI) and land use cover), pollutant emissions and population distributions, as well as the terms capturing the spatial and temporal autocorrelations of surface ozone concentrations. As reported (http://doi.org/10.528 1/zenodo.4400043 (accessed on 30 December 2020)), there is high consistency between the estimated ozone data and ground-based ozone measurements (R 2 = 0.87, root-meansquare error (RMSE) = 17.10 µg/m 3 ). Figure 2A presents the spatial distributions of ozone concentrations across China in 2016.
Data on annual PM 1 concentrations at 1 km 2 spatial resolution between 2014 and 2016 were obtained from our previous study. More details on PM 1 estimates can be found in Wei et al. [33]. Briefly, a machine learning approach, i.e., the space-time extremely randomized trees model (STET), was developed to estimate daily PM 1 concentrations in terms of the geospatial data of MAIAC AOD, MEIC pollution emissions, meteorological factors, land use, road and population, as well as the spatiotemporal information which captures the spatial and temporal autocorrelations of PM 1 concentrations. According to the results from model validation, there is high agreement between the estimated daily PM 1 concentrations and ground-level PM 1 measures, with R 2 and RMSE equal to 0.77 and 14.6 µg/m 3 , respectively [33]. Such high agreement could also be seen when PM 1 was estimated at the seasonal and annual scales, with R 2 equal to 0.97 and RMSE lower than 4.1 µg/m 3 . To date, this dataset has been increasingly used to estimate health burden [34] Int. J. Environ. Res. Public Health 2021, 18, 10386 4 of 14 as well as the effects of PM 1 on human health in Chinese cities [21,35,36]. The spatial distributions of PM 1 concentrations in 2016 all over China are shown in Figure 2B. from model validation, there is high agreement between the estimated daily PM1 concentrations and ground-level PM1 measures, with R 2 and RMSE equal to 0.77 and 14.6 μg/m 3 , respectively [33]. Such high agreement could also be seen when PM1 was estimated at the seasonal and annual scales, with R 2 equal to 0.97 and RMSE lower than 4.1 μg/m 3 . To date, this dataset has been increasingly used to estimate health burden [34] as well as the effects of PM1 on human health in Chinese cities [21,35,36]. The spatial distributions of PM1 concentrations in 2016 all over China are shown in Figure 2B.

Incidence Rate of Female Lung Cancer
Annual age-standardized incidence rates of trachea, bronchus and lung cancer for females (i.e., the incidence rate of female lung cancer hereafter) between 2014 and 2016 were acquired from the 2017-2019 China Cancer Registry Annual Reports [37][38][39]. This health outcome, defined as the incidence number of lung cancer for females per 100,000 people per year in a registry (county/district), was age-standardized in terms of Segi's world population. The annual reports used in the present study were publicly released by the Chinese Cancer Registry at the National Cancer Centre of China with the aim of providing timely, representative and comprehensive information on cancer diseases from specific causes all over China (including the incidence rate of female lung cancer in the present study). For example, data on cause-specific cancer diseases for each of the 682 Chinese cancer registries were reported in the 2019 China Cancer Registry Annual Report [39]. These registries were located in 31 of the 34 province-level administrative regions in China. Figure 2C exhibits the spatial distributions of the incidence rate of female lung cancer in 2016.

Socioeconomic Characteristics and Smoking Covariates
Six socioeconomic covariates were chosen according to their reported associations with lung cancer diseases in prior studies [24,40,41]. These covariates included finance per capita, population, average education years, proportions of construction and manufacturing workers, and urban-rural division. Data on the first two variables were derived from the 2015-2017 China Statistical Yearbook (County-Level), while other data were obtained from the tabulation of the 2010 population census of the People's Republic of China. Figure 2D-F show the spatial distributions of educational attainment, financial level and urban-rural attributes for 436 Chinese cancer registries (counties/districts), respectively.
Data on smoking covariates were acquired from the 2015 China Health and Retirement Longitudinal Study (CHARLS) wave3. This publicly accessible dataset was released by the National School of Development, Peking University (http://charls.pku.edu.cn/en/page/ data/2015-charls-wave4 (accessed on 31 May 2017)). The CHARLS survey aims to provide the timely information on socioeconomic and health conditions of Chinese residents who are 45 years and older across China, which is therefore representative and comprehensive at the national scale [42]. There are 10,257 households and 17,708 individuals recruited by the CHARLS survey with a spatial coverage of 28 Chinese province-level administrative regions [42]. To date, this dataset has been used to identify the influential factors of mental and physical health [43][44][45].

Statistical Analysis
The effects of ozone and PM 1 on the incidence rate of female lung cancer were estimated using two linear regression models. In Model 1, we solely included air pollutant (i.e., ozone and PM 1 ), time and location factors. Notably, the effect of ozone (and PM 1 ) is expressed as the change in the incidence rate of female lung cancer relative to its mean when ozone (and PM 1 ) changed by 10 µg/m 3 . In Model 2, we further adjusted for socioeconomic factors, including finance per capita, average education years, proportions of construction and manufacturing workers, population and urban-rural division (as a dummy variable). These socioeconomic covariates were selected primarily as a result of their connections with lung cancer diseases reported in previous studies [24,40,41].
Then, we performed three sensitivity analyses to test the sensitiveness of ozone and PM 1 effects. Firstly, we tested whether the effects of the two air pollutants are still significant after the adjustment of smoking covariates. Smoking prevalence (shortened to smoking_p) in combination with smoking strength (shortened to smoking_s, namely, the number of cigarettes smoked per day), was selected according to the reported connections of lung cancer diseases with the two smoking covariates in prior studies [46]. Notably, the CHARLS-derived smoking dataset is available at city level and does not cover all counties/districts of the present study. As a response, the same smoking characteristics were attributed to counties/districts located in the same city, which left around 48% of the whole sample. Based on the dataset of around 48% of the whole sample, we examined the effects of ozone and PM 1 in the two situations, i.e., with and without the controls of smoking factors.
Secondly, we tested whether the effects of ozone and PM 1 found in the present study are sensitive to the further control of additional air pollutant (i.e., SO 2 ). The additional air pollutant is measured as the annual mean SO 2 concentrations aggregated in each county/district. The SO 2 data at 0.5 • × 0.625 • spatial resolution were acquired from the time-series dataset of M2TMNXAER (V5.12.4), which is freely published by the Global Modelling and Assimilation Office of NASA in USA. Details about the production of the SO 2 dataset can be found in Randles et al. [47] and Buchard et al. [48]. Currently, the SO 2 dataset has been increasingly used in air pollution studies [49,50]. Thirdly, we examined whether the effects of ozone and PM 1 are still significant when estimating the effects of these two targeted air pollutants and additional air pollutant (i.e., SO 2 ) in the same model. One of the aims is to determine the relative importance of the three air pollutants in increasing the incidence rate of female lung cancer.
Finally, we investigated the modification effect of urban-rural division on the association between ozone (and PM 1 ) and the incidence rate of female lung cancer. Firstly, the whole dataset was stratified according to urban-rural division. Following the method of urban and rural division adopted in the cancer registry annual report as well as many previous studies conducted in China [39,51,52], the present study used counties and districts to represent rural and urban areas, respectively. We conducted the comparison of the effects of ozone (and PM 1 ) between urban and rural groups in terms of Model 2. Then, the stratified dataset was combined, and the interaction of ozone (and PM 1 ) with urban-rural dummy variable was added to Model 2; this was used to investigate whether the effect of air pollution significantly varies between urban and rural areas. Notably, the urban-rural dummy variable was not included in the model because this variable had a high collinearity with its interaction (i.e., the interaction between urban-rural dummy variable and ozone (and PM 1 )). Table 1 presents the descriptive statistics of health outcome, air pollutants and socioeconomic factors for 436 Chinese cancer registries (i.e., counties/districts). The mean incidence rate of female lung cancer for 436 Chinese counties/districts was 22.42 per 10 5 people. As shown in Table 1, a great variation in the incidence rate of female lung cancer among 436 counties/districts was also observed, with the standard deviation of 8.85. With regard to the two air pollutants, the mean values of ozone and PM 1 were 84.32 µg/m 3 and 34.67 µg/m 3 , respectively. The values of the two air pollutants also varied considerably all over the counties/districts (Table 1). Such large variations were also observed for socioeconomic covariates (Table 1).

Effects of Ambient Ozone and PM 1
The results of the spatial association between ozone and the incidence rate of female lung cancer are shown in Table 2. In general, there was a positive effect of ozone. As shown in Table 2, if there was a 10 µg/m 3 increase in ozone, then the change in the incidence rate of female lung cancer relative to its mean was 4.91% (95% CI: 5.46%, 16.54%) in Model 1. With the control of socioeconomic characteristics in Model 2, there was a slight decrease in the effect of ozone. Specifically, a 10 µg/m 3 increase in ozone was positively associated with a 4.57% (95% CI: 4.32%, 16.16%) increase in the incidence rate of female lung cancer relative to its mean (Table 2). * for p < 0.1, ** for p < 0.05 and *** for p < 0.01. If ozone changes by 10 µg/m 3 , then the change in the incidence rate of female lung cancer relative to its mean = (10× ozone coefficient)/mean incidence rate. Table 3 exhibits the results of the effect of PM 1 on the incidence rate of female lung cancer. Generally, there were positive associations of the female lung cancer incidence rate with PM 1 . In Model 1 without socioeconomic controls, the change in the incidence rate of female lung cancer relative to its mean was 4.60% (95% CI: 3.91%, 16.70%), when PM 1 changed by 10 µg/m 3 . A similar pattern of results was observed in Model 2. Specifically, as shown in Table 3, if there was a 10 µg/m 3 increase in PM 1 , the change in the incidence rate of female lung cancer relative to its mean was 4.89% (95% CI: 4.37%, 17.56%). ** for p < 0.05 and *** for p < 0.01. If PM 1 changes by 10 µg/m 3 , then the change in the incidence rate of female lung cancer relative to its mean = (10× PM 1 coefficient)/mean incidence rate.

Control of Smoking Factors
Sensitivity analyses of air pollution effects to the control of smoking characteristics are shown in Figure 3. In general, ozone and PM 1 effects were not sensitive to the smoking control. As shown in Figure 3A, there was a positive association between ozone and the incidence rate of female lung cancer without the adjustment of smoking factors. When further controlling for smoking covariates, the effect of ozone was still significant ( Figure 3A); smoking prevalence was also positively correlated with the incidence rate of female lung cancer ( Figure 3A). A similar pattern of results was observed for PM 1 ( Figure 3B). In partic-ular, both PM 1 and smoking prevalence exerted their significant effects on the incidence rate of female lung cancer after the adjustment of smoking factors ( Figure 3B).

Control of Smoking Factors
Sensitivity analyses of air pollution effects to the control of smoking characteristics are shown in Figure 3. In general, ozone and PM1 effects were not sensitive to the smoking control. As shown in Figure 3A, there was a positive association between ozone and the incidence rate of female lung cancer without the adjustment of smoking factors. When further controlling for smoking covariates, the effect of ozone was still significant ( Figure  3A); smoking prevalence was also positively correlated with the incidence rate of female lung cancer ( Figure 3A). A similar pattern of results was observed for PM1 ( Figure 3B). In particular, both PM1 and smoking prevalence exerted their significant effects on the incidence rate of female lung cancer after the adjustment of smoking factors ( Figure 3B).  Table 4 presents the results of the sensitivity analysis when SO2 (i.e., the additional air pollutant) is controlled. In general, the effects of ozone and PM1 were robust to the adjustment of additional air pollutant (i.e., SO2). As shown in Table 4, a 10 μg/m 3 increase in ozone was positively associated with a 3.95% (95% CI: 2.89%, 14.84%) increase in the incidence rate of female lung cancer relative to its mean; notably, the effect of SO2 on the  Table 4 presents the results of the sensitivity analysis when SO 2 (i.e., the additional air pollutant) is controlled. In general, the effects of ozone and PM 1 were robust to the adjustment of additional air pollutant (i.e., SO 2 ). As shown in Table 4, a 10 µg/m 3 increase in ozone was positively associated with a 3.95% (95% CI: 2.89%, 14.84%) increase in the incidence rate of female lung cancer relative to its mean; notably, the effect of SO 2 on the incidence rate of female lung cancer was also significant (β = 3.34%, 95% CI: 2.35%, 12.63%) ( Table 4). A similar pattern of results was seen for PM 1 (Table 4). In particular, as presented in Table 4, there were positive associations of the incidence rate of female lung cancer with PM 1 and SO 2 . ** for p < 0.05 and *** for p < 0.01. When ozone, PM 1 or SO 2 changed by 10 µg/m 3 , the change in the incidence rate of female lung cancer relative to its mean = (10× coefficient for ozone, PM 1 or SO 2 )/mean incidence rate.

Ozone, PM 1 and Additional Air Pollutant (SO 2 ) Effects in the Same Model
The results of the sensitivity analysis estimating ozone, PM 1 and additional air pollutant (SO 2 ) effects in the same model are presented in Table 5. In general, the effects of the three air pollutants were still significant. Without the adjustment of socioeconomic covariates, each air pollutant, including ozone, PM 1 and SO 2, was positively associated with the incidence rate of female lung cancer; the greatest effect size was for ozone, followed by PM 1 and SO 2 . A similar pattern of results was observed when socioeconomic factors were adjusted. In particular, as shown in Table 5, the association with the incidence rate of female lung cancer was strongest for ozone (β = 3.94%, 95% CI: 1.28%, 6.60%), followed by PM 1 (β = 3.45%, 95% CI: 0.28%, 6.61%) and SO 2 (β = 2.32%, 95% CI: −0.15%, 4.79%). Table 5. Sensitivity analysis of estimating effects of ozone, PM 1 and the additional air pollutant (i.e., SO 2 ) in the same model. * for p < 0.1, ** for p < 0.05 and *** for p < 0.01. When ozone, PM 1 or SO 2 changed by 10 µg/m 3 , the change in the incidence rate of female lung cancer relative to its mean = (10× coefficient for ozone, PM 1 or SO 2 )/mean incidence rate.

Modification Effect of Urban-Rural Division
Tables 6 and 7 exhibit the results of urban-rural modification effect on the association between ozone (and PM 1 ) and the incidence rate of female lung cancer. Generally, urbanrural division (rural group as the reference group) positively modified the effect of PM 1 .
In the stratified dataset, as shown in Table 6, the effect of PM 1 was larger in urban than in rural areas with the coefficients of 0.11 and 0.08, respectively; in the combined dataset, as shown in Table 7, the change in the incidence rate of female lung cancer relative to its mean was higher by 4.14% (95% CI: 2.68%, 15.88%) in urban than in rural areas if there was a 10 µg/m 3 increase in PM 1 . With regard to ozone, however, no modifying role of urban-rural division was observed. Specifically, despite the significant effect of the interaction between ozone and urban-rural dummy variable, ozone was not positively associated with the incidence rate of female lung cancer in the urban group (Table 6). ** for p < 0.05 and *** for p < 0.01. When ozone or PM 1 changed by 10 µg/m 3 , the change in the incidence rate of female lung cancer relative to its mean = (10× coefficient for ozone or PM 1 )/mean incidence rate.  ** for p < 0.05 and *** for p < 0.01. When ozone or PM 1 changed by 10 µg/m 3 , the change in the incidence rate of female lung cancer relative to its mean = (10× coefficient for ozone or PM 1 )/mean incidence rate.

Discussion
Despite the implementation of multiple air-clearing actions, ozone concentrations have been increasing, and the magnitude and frequency of severe ozone air pollution were both higher in China than in most developed countries. Ozone has now become another air pollutant threating human health in Chinese cities on which to focus. PM 1 is likely to be more detrimental to the human body than coarser PMs (e.g., PM 2.5 and PM 10 ). Meanwhile, lung cancer has already become the second-most common type of cancer incidences for females in China, with the incidence rate of female lung cancer at 42.28 per 100,000 people in 2016 [39]. Despite such significance, however, few nationwide studies have investigated the effects of the two prominent air pollutants on the incidence rate of female lung cancer in China. As one of the earliest attempts in China, this nationwide study estimated the effects of ozone and PM 1 in 436 cancer registries (counties/districts) of China between 2014 and 2016.
We found a positive effect of ozone on the incidence rate of lung cancer. This is in line with the findings from previous Chinese and Western studies. In particular, a nationwide study acquiring data from 75 Chinese counties/districts between 1990 and 2009 indicated that ozone was positively associated with the incidence rate of lung cancer in the spatial age-period-cohort model [51]. Similarly, using data of 22.2 million US Medicare beneficiaries from 2000 to 2008, another nationwide study indicated the adverse effect of ozone exposure on lung cancer-associated mortality in the United States [53]. The positive associations of lung cancer diseases with ozone were also reported in other studies [54,55].
We found a greater effect of PM 1 on the incidence rate of female lung cancer in urban than in rural areas. This finding is consistent with those of previous studies in relation to the modifying role of socioeconomic factors on the effects of PMs. As indicated in our previous study, the increase in the incidence rate of male lung cancer relative to its mean was larger by 2.47% (95% CI: 0.38%, 4.55%) in urban areas than in rural areas when PM 1 changed by 10 µg/m 3 [50]. Such findings regarding urban-rural division's modifying role on PM effects were further enhanced by many other studies [41,51,56]. For example, when there was a 10 µg/m 3 increase in PM 2.5 in a nationwide study of China, the relative risk of lung cancer incidence for urban areas was 1.06 (95% CI: 1.04, 1.08), which is higher than 1.04 (95% CI: 1.00, 1.08) for the rural group [51]. Theoretically, three differences (i.e., material resources, biological factors and psychological stress) may be responsible for the varying effects of air pollution among different socioeconomic groups.
In the present study, as discussed in our previous studies [50,57], the difference in smoking behaviours between urban and rural areas may explain the differential effects between the two groups. Briefly, there were higher cigarette-associated indicators, including smoking strength (i.e., the number of cigarettes smoked per day) and the prevalence of cigarette smoking among all smokers, in urban than in rural areas of China [58]; at the same time, the age at which people began regularly smoking was also younger in urban than in rural groups of China [58]. The more severe smoking-associated situation in urban areas may explain the higher smoking-related hazard risks (e.g., lung cancer incidence and mortality) in urban than in rural areas [58]. This may cause people living in urban areas to be more vulnerable to exposure to air pollution (including PM 1 ), and thus more greatly affected by PM 1 , since previous studies have reported the differential effects of air pollution on the physical health of human beings among people having different smoking behaviors [59,60]. The finding in the present study supports the argument that socioeconomic factors can modify the association between air pollution and human health (including physical and mental health).
There are several strengths in the present study. Firstly, this is one of the few large-scale studies in China with data collected from 436 Chinese counties/districts. This nationwide study provides representative and comprehensive evidence of the effects of ozone and PM 1 from a developing setting (i.e., China) where the magnitude and frequency of severe ozone pollution are all larger than those of most developed countries. Secondly, this study focuses on ozone air pollution, which has become another health risk in Chinese cities on which to focus. We also pay attention to PM 1 which accounts for a large proportion of the dominant PM 2.5 air pollution in China. Thirdly, collecting data from 436 registries (counties/districts) in the present study enables us to further investigate modification effects of socioeconomic factors (i.e., urban-rural division) in the setting of China where urban-rural division is highly prominent.
By contrast, several limitations and prospects should be well noted and discussed. Firstly, exposure to air pollution was operationalized as the registry-aggregated concentrations of ozone (and PM 1 ), and thus did not consider individual mobility. Hence, as in most prior ecological studies related to air pollution [4,24,61], there are errors in exposure measurements of the two air pollutants in the present study. Secondly, it is not feasible for us to examine the lag effects of ozone and PM 1 on the incidence rate of female lung cancer in this work. Some studies have pointed to the lag effects of air pollution, including the single-and moving-average lags [50,57,62]. However, our data with respect to ozone and PM 1 are available from 2014 and after (female lung cancer data are available from 2006 to 2016), so it is not feasible to investigate the potential long-latency of lung cancer development associated with exposure to ozone and PM 1 . If data on the two air pollutants prior to 2014 are available, this limitation should be well addressed and handled.
Thirdly, similar to those of our previous studies [50,57], the lack of smoking data at county/district level may make our findings concerning air pollution effects sensitive to the control of smoking covariates. As a response, we derived city-level smoking data from the CHARLS survey (which left around 50% of the whole registries for the sensitivity analysis) to test whether the effects of ozone and PM1 are still significant with the control of smoking factors. Such operationalization has two main limitations. On the one hand, our findings acquired from the whole registries may still not be robust to the control of smoking factors, although the sensitivity analysis using the dataset of around 50% of the whole registries has shown the robustness of air pollution effects to the control of smoking factors. One the other hand, CHARLS-derived smoking data are available at city level, so counties/districts located in the same city were attributed with the same smoking characteristics in the present study. This operationalization suffers from problems such as ecological fallacy. If data on county-level smoking covariates are available, such limitations should be well considered and thus addressed.
Fourthly, the uneven distribution of cancer registries inherited from the China Cancer Registry Annual Report may bias the estimate of air pollution effect on the incidence rate of female lung cancer in the present study. Although the cancer registries included in the annual report are dispersed over 31 of 34 Chinese province-level regions, they are not evenly distributed. That is, most registries (counties/districts) are concentrated in Southeast China, while registries are quite scarce in the west of China. Such uneven distribution inherited from original health data is likely to bias the effect estimate in the present study. Fifthly, despite some efforts [15,19,63], it is still not sufficient to conclude that finer particulate matters have greater effects on human health (physical and psychological health), especially in developing countries where data of PMs (especially for PM 1 ) are usually scarce or quite limited. This highlights the great need to examine the effects of multiple PMs (e.g., PM 1 , PM 2.5 and PM 10 ) at a nationwide scale in the future.

Conclusions
Ozone and PM 1 were positively associated with the incidence rate of female lung cancer in China. Moreover, urban-rural division may modify the association of the incidence rate of female lung cancer with PM 1 , with a higher effect of PM 1 observed in urban areas. There was no modifying role played by urban-rural division for ozone. The implications of our findings are two-fold. On the one hand, this study suggests that the continuous clear-air actions in China, especially the strict prevention and control strategies of air pollution, should be well designed to consider not only the dominant PM 2.5 air pollution in China, but also PM 1 and ozone which has become another focus of health risks in Chinese cities. On the other hand, area-specific measures, such as reducing the disparities in access to healthcare resources between urban and rural areas, should be well considered and developed to reduce urban-rural disparities in the health effects of air pollution (especially for PMs) in China.  Data Availability Statement: Publicly available datasets were analyzed in this study. This data can be found here: https://weijing-rs.github.io/product.html (accessed on 2 August 2021).