Short-term joint effects of ambient PM2.5 and O3 on mortality in Beijing, China

Introduction In recent years, air pollution caused by co-occurring PM2.5 and O3, named combined air pollution (CAP), has been observed in Beijing, China, although the health effects of CAP on population mortality are unclear. Methods We employed Poisson generalized additive models (GAMs) to evaluate the individual and joint effects of PM2.5 and O3 on mortality (nonaccidental, respiratory, and cardiovascular mortality) in Beijing, China, during the whole period (2014–2016) and the CAP period. Adverse health effects were assessed for percentage increases (%) in the three mortality categories with each 10-μg/m3 increase in PM2.5 and O3. The cumulative risk index (CRI) was adopted as a novel approach to quantify the joint effects. Results The results suggested that both PM2.5 and O3 exhibited the greatest individual effects on the three mortality categories with cumulative lag day 01. Increases in the nonaccidental, cardiovascular, and respiratory mortality categories were 0.32%, 0.36%, and 0.43% for PM2.5 (lag day 01) and 0.22%, 0.37%, and 0.25% for O3 (lag day 01), respectively. There were remarkably synergistic interactions between PM2.5 and O3 on the three mortality categories. The study showed that the combined effects of PM2.5 and O3 on nonaccidental, cardiovascular, and respiratory mortality were 0.34%, 0.43%, and 0.46%, respectively, during the whole period and 0.58%, 0.79%, and 0.75%, respectively, during the CAP period. Our findings suggest that combined exposure to PM2.5 and O3, particularly during CAP periods, could further exacerbate their single-pollutant health risks. Conclusion These findings provide essential scientific evidence for the possible creation and implementation of environmental protection strategies by policymakers.


Introduction
Significant epidemiological research has shown that short-term exposure to ambient air pollution is substantially related to numerous detrimental health consequences (Fan et al., 2020;Stafoggia et al., 2022) (1). Among the various ambient air pollutants, particles with diameters ≤2.5 μm (PM 2.5 ) and ozone (O 3 ) are considered have serious dangerous to human health (2).
At one time, China, the world's largest developing country, had the worst air pollution issue than other countries, which led to almost 2 million premature deaths annually (3). The Chinese government has implemented a variety of pollution prevention and control measures since 2013 to protect public health, including policy changes in energy, industrial, and transportation infrastructure (4). According to (5), there was a significant reduction of 30-50% in PM 2.5 concentrations from 2013 to 2017. Despite this reduction, PM2.5 pollution episodes persist in China, especially in megacities (6,7). Furthermore, the decreased PM 2.5 also slow down the sink of hydroperoxy radicals and thus speeding up O 3 production, resulting in the ground-level O 3 levels in China have grown annually (8). Consequently, there was a cooccurrence of PM 2.5 and O 3 pollution (9)(10)(11). This cooccurrence is known as combined air pollution (CAP). CAP has received much interest in atmospheric environmental research (12,13). However, the health risks caused by CAP are still unclear.
Given that humans are exposed to more than one air pollutant in real life, biological responses to inhaled pollutants likely depend on the interaction between individual pollutants (14). Ground-level O 3 and PM 2.5 are closely related and interact with each other, and thus they may have a combined negative impact on human body (15). Traditional time-series studies have focused on assessing health effects using single-pollutant models. Research on the combined health effects of multiple pollutants has been inadequate. In recent years, some new models and methods have been developed to simultaneously quantify the combined health effects of multiple pollutants. One such technique is the use of the cumulative risk index (CRI), which involves the linear combination of individual coefficients. This approach enables accurate estimation of cumulative effects, even in cases where there is a high correlation among variables (14), and has been recommended for joint estimates of multipollutant exposure effects on health outcomes (16). However, CRI-related studies are quite limited; most of these studies have been conducted in developed countries, and studies in developing countries are lacking (17,18).
Beijing, as the capital of China, has a serious air pollution issue. CAP appears in Beijing from time to time, and the frequency continues to increase (19). It is still unknown how the CAP affects the health outcomes of a local population. Therefore, the goal of this manuscript was to evaluate the individual and combined effects of PM 2.5 and O 3 on nonaccidental and cause-specific mortality in Beijing, China, across the entire time period and during the CAP period, respectively. The joint health effects of PM 2.5 and O 3 were estimated by using the CRI index.

Health data
In this study, we collected data on the daily death counts in Beijing from January 1, 2014, to December 31, 2016, from the Chinese Center for Disease Control and Prevention (CDC). To classify the causes of death, we used the International Classification of Diseases, Tenth Revision (ICD-10). Nonaccidental causes, cardiovascular diseases, and respiratory diseases were categorized as A00-R99, I00-I99, and J00-J99, respectively.

Statistical methods
We employed four parallel time-series Poisson generalized additive models (GAMs) to evaluate the individual and joint effects of O 3 and PM 2.5 on nonaccidental, cardiovascular, and respiratory mortality during the whole period and the CAP period. These models include a single-pollutant model, multipollutant model, nonparametric bivariate response surface model, and stratification model. First, we utilized the single-pollutant model as the basis to assess the individual effects of a single pollutant on health outcomes at different lag days, including single (lag days 0 and 1) and cumulative (lag days 01 and 04) effects. The following is an expression for Model 1: where Y t and E(Y t ) signify the daily death counts and predicted death counts on day t, respectively. α refers to the intercept. NS ( ) is the natural cubic spline function. According to the minimum Akaike information criterion (AIC), Time with the degrees of freedom (df) 6/ year was selected to control for secular trends, and the df of the daily mean temperature (Temp) and RH are both 3. DOW and Holiday are two dummy variables that indicate weekday and public holidays, Frontiers in Public Health 03 frontiersin.org respectively (23). x kt and β kt denote the specific air pollutant concentrations and the corresponding coefficient on day t, respectively. Additionally, COVs represent all covariates including time, mean temperature, relative humidity, weekday, public holidays, and the intercept, respectively. On this basis, we utilized a multipollutant model to evaluate the joint effects of PM 2.5 and O 3 on health outcomes at different lag days. The CRI, which was developed using estimates from multipollutant models, was used to assess the joint effects of multipollutant exposures (24). The multipollutant model and the formula for the CRI can be expressed as follows: where x kt and β kt denote the specific air pollutant concentrations and the corresponding coefficient on day t, respectively. The COVs are identical to those in Model (1). p indicates the type of air pollutant. CRI t denotes the joint effects of p air pollutant mixtures on day t.
The CRIs obtained from the multipollutant models were compared with the effect estimates of the single-pollutant models. If the effect estimate from the single-pollutant model was as high as the CRI from the multipollutant model, it indicated that the influence of only one pollutant was adequate to reflect the total pollutant mixture and that there were no synergistic effects.
Third, we also used a nonparametric bivariate response surface model to intuitively analyze the combined effects of PM 2.5 and O 3 on health outcomes. The model can be expressed as follows: ST () denotes the cubic regression splines. The COVs are identical to those in Model (1).
Fourth, the pollutant stratification model was employed to quantitatively assess the joint effects of PM 2.5 and O 3 on health outcomes during the CAP period. The model can be expressed as follows: where m is an indicator variable that is used to represent the CAP days. m = 1 represents co-occurring air pollution of PM 2.5 and O 3 ; otherwise, m = 0. β it and β jt represent the coefficients of O 3 and PM 2.5 on day t, respectively. The COVs are the same as those in Model (1).
To evaluate the models' robustness, several sensitivity studies were carried out. We changed the df of Time from 7 to 10 per year and the df of mean temperature and RH from 3 to 5 for the singlepollutant model. R 4.2.3 software with the "mgcv" package was used for all analyzes. For each 10-μg/m 3 increase in PM 2.5 and O 3 , the estimated individual and joint effects are shown as percentage changes (%) along with 95% confidence intervals (95% CIs). The Spearman correlation coefficients of the three mortality categories and different environmental factors are shown in Figure 1. The three mortality categories were all significantly negatively correlated with the mean temperature, RH, and O 3 concentration and significantly positively correlated with the PM 2.5 concentration. The Spearman correlation between PM 2.5 and O 3 was low even though it was statistically significant (r = −0.07, p < 0.001), indicating the possibility of interaction effects on three mortality categories. Figure 2 illustrates the individual effects of PM 2.5 and O 3 on health outcomes at different lags. The individual effects of PM 2.5 and O 3 on the three mortality categories all peaked at lag day 01. Specifically, the increase in the nonaccidental, cardiovascular, and respiratory mortality categories was 0.32% (95% CI: 0.21, 0.43%), 0.36% (95% CI: 0.21, 0.50%), and 0.43% (95% CI: 0.28, 0.58%) for each 10-μm −3 increase in the PM 2.5 concentration (lag day 01), and 0.22% (95% CI: 0.08, 0.36%), 0.37% (95% CI: 0.21, 0.53%), and 0.25% (95% CI: 0.12, 0.37%) for each 10-μg/m 3 increase in the O 3 concentration (lag day 01), respectively. Figure 3 depicts the joint effects of PM 2.5 and O 3 on health outcomes at different lags. As with the individual effects of PM 2.5 and O 3 , the joint effects of PM 2.5 and O 3 on the three mortality categories all peaked at lag day 01. The corresponding CRIs for nonaccidental, respiratory and cardiovascular mortality were 0.34% (95% CI: 0.16, 0.52%), 0.43% (95% CI: 0.21, 0.65%), and 0.46% (95% CI: 0.23, 0.70%), respectively. Importantly, for the same category of diseases, the joint effect represented by CRI was higher than for any single pollutant effect estimate at lag day 01. Overall, the CRIs implied that a single-pollutant effect did not accurately represent the whole health effects of the mixture. In the subsequent analysis, both PM 2.5 and O 3 at lag day 01 were used as the research objects. Figure 4 illustrates the combined effects of PM 2.5 and O 3 on the three mortality categories using three-dimensional visualization graphs. The response surfaces show that the combined effects of PM 2.5 (lag day 01) and O 3 (lag day 01) on nonaccidental, cardiovascular, and respiratory deaths were complicated. Notably, when high concentrations of PM 2.5 and O 3 coexisted, all three categories (nonaccidental, cardiovascular, and respiratory fatalities) reached their maximums, showing that the interaction effects could be synergistic. Table 2 depicts the individual and joint effects of PM 2.5 (lag day 01) and O 3 (lag day 01) on health outcomes during the whole period and the CAP period. For the same kind of illness, the CRIs of the joint effects during both the whole period and the CAP period were higher than any single-pollutant effect estimates. In addition, the joint effects during the CAP period were remarkably larger than those during the whole period, indicating that the CAP period further exacerbated the combined effects of PM 2.5 and O 3 on the three mortality categories.

Results
According to the results of the sensitivity analyzes, the effects of O 3 (or PM 2.5 ) remained robust regardless of the change in the df of the time (see Supplementary Figure S1), the df of the mean temperature, and the df of the RH (see Supplementary Table S1).

Discussion
The CAP of PM 2.5 and O 3 has become a major environmental and health concern worldwide (7). Evaluating the short-term individual and joint effects of PM 2.5 and O 3 on health outcomes Spearman correlation matrix between the three mortality categories and different environmental factors in Beijing, China. *p < 0.05, **p < 0.01, ***p < 0.001. Non, Nonaccidental mortality; Car, Cardiovascular mortality; Resp, Respiratory mortality; Temp, mean temperature; RH, relative humidity.  (1,25). For example, a meta-analysis conducted in 272 Chinese cities by Chen et al. (26) showed that a 10-μg/m 3 increase in the PM 2.5 concentration was associated with an increase in nonaccidental, cardiovascular, and respiratory mortality of 0.27, 0.39, and 0.29%, respectively. Another meta-analysis in China (27) revealed that an increase of 10-μg/m 3 in the O 3 concentration caused increases of 0.24 and 0.27% in nonaccidental and cardiovascular mortality, respectively. In this study, the results from the single-pollutant models revealed that each 10-μg/m 3 increase in the PM 2.5 concentration caused increases of 0.32, 0.36, and 0.43% in nonaccidental, cardiovascular, and respiratory mortality, respectively, and each 10-μg/m 3 increase in the O 3 concentration caused increases of 0.22, 0.37, and 0.25% in nonaccidental, cardiovascular, and respiratory mortality, respectively, in Beijing, China. Our estimates of the PM 2.5mortality and O 3 -mortality relationships were generally consistent with those of previous studies. Percentage changes (%) in the three mortality categories associated with each 10-μg/m 3 increase in PM 2.5 and O 3 concentrations at different lag days in the single-pollutant models. Percentage changes (%) in the three mortality categories associated with each 10-μg/m 3 increase in the PM 2.5 and O 3 concentrations at different lag days in the multipollutant models.

Frontiers in Public
Frontiers in Public Health 06 frontiersin.org In the multipollutant models, our findings suggested that the estimates of the joint effects of the two air pollutants on mortality were higher than those for any individual effect for the same kind of illness. Consistent with our findings, a study conducted by Lei et al. (28) in Hefei, China, indicated that the effects of the health risks caused by PM 2.5 on nonaccidental mortality increased when O 3 was included, and vice versa, indicating that O 3 and PM 2.5 could aggravate each other's unfavorable health effects. A cross-sectional study conducted in six countries revealed a synergistic interaction effect of PM 2.5 and O 3 on disease deterioration (29). However, in contrast to our findings, Qu et al. (30) observed that when O 3 was included, the effect of PM 2.5 on nonaccidental mortality was reduced. Moreover, several earlier studies showed no interaction effects of PM 2.5 and O 3 (31,32). This inconsistency could be attributed to differences in the chemical FIGURE 4 Bivariate response surfaces of PM 2.5 and O 3 for nonaccidental, cardiovascular, and respiratory deaths in Beijing, China. Furthermore, the differences in study methods and individual sensitivity to pollutants can also lead to different results (33).
Notably, the patterns of the combined effects of PM 2.5 and O 3 on mortality demonstrated that coexisting high concentrations of PM 2.5 and O 3 could have synergistic effects on three mortality categories (34). Biological mechanisms have been somewhat postulated to explain the potential interaction effect of PM 2.5 and O 3 pollution on respiratory and cardiovascular mortality, despite the lack of clear evidence for a direct synergistic effect of the two pollutants on illnesses. For example, a few toxicology experiments on rats validated that the particulate matter served as a carrier for O 3 , delivering O 3 into the body (35). Inhaling particles and O 3 together had a synergistic impact on airway responsiveness and allergic inflammation in mice (36), suggesting that combined exposure to O 3 and PM 2.5 markedly increased health risks (37). Therefore, people, especially those with chronic respiratory and cardiovascular diseases, should strengthen protection measures and reduce outdoor activities, especially on CAP of PM 2.5 and O 3 days.
The key advantage of this study is as follows: Current research on CAP primarily focuses on the characteristics of changes in PM 2.5 and O 3 concentrations, meteorological causes, and their mutual influences. However, there is less emphasis on the joint health effects of PM 2.5 and O 3 during CAP periods (7,38). Furthermore, traditional multipollutant models mainly focus on describing the difference in the health effects of a single pollutant before and after the addition of other pollutants without quantifying the combined effects of multiple pollutants (6). Our study differs from traditional studies, as we utilized multiple methods to examine the harmful health effects associated with exposure to one and two pollutants. We also conducted stratification studies on pollution, with a specific focus on the combined health effects of PM 2.5 and O 3 during the CAP period. Furthermore, we used the CRI to accurately quantify the joint effects of PM 2.5 and O 3 during both the whole and CAP periods. This approach addresses the limitations of previous research to a significant extent (16).
There are several limitations of our study that should be acknowledged. First and foremost, due to the difficulty in obtaining disease data in China, the study only included a 3-year disease death time series, and the time coverage was relatively limited. The latest year's death data could not be obtained, which could reduce the statistical power. Second, in keeping with many previous studies (4, 39), we did not collect data on the real-time pollution exposure levels of individuals and only used the outdoor air pollutant concentration to represent individual PM 2.5 and O 3 exposure levels, which inevitably led to some deviation in the results (33). Third, the two most dangerous pollutants in China at this time are PM 2.5 and O 3 . This study only tentatively carried out research on the interaction effect between PM 2.5 and O 3 on public health and did not carry out in-depth research on interaction effects with other air pollutants (such as O 3 and nitrogen dioxide, sulfur dioxide and PM 2.5 ). Therefore, with the improvement of research methods at a later stage, further in-depth study of the health effects of interactions between different air pollutants on human health should be carried out.

Conclusion
Our findings showed that exposure to PM 2.5 and O 3 may be significant risk factors for nonaccidental, cardiovascular, and respiratory mortality in Beijing, China. Moreover, we found that combined exposure to PM 2.5 and O 3 could amplify their individual effects on three mortality categories, particularly during CAP of PM 2.5 and O 3 periods. Therefore, during the CAP periods, the public should take timely preventive measures and reduce outdoor activities to some extent to reduce air pollution hazards.

Data availability statement
The data analyzed in this study is subject to the following licenses/ restrictions: Authors are not allowed to disclose data. Requests to access these datasets should be directed to YZ, zhangy881208@126.com.
Frontiers in Public Health 08 frontiersin.org