Short-term effect of PM 2.5 and O3 on non-accidental mortality and respiratory mortality in Lishui District , China

Using daily mortality and atmospheric monitoring data from 2015 to 2019, we applied a generalized additive model with time-series analysis to study the association of PM 2.5 and O 3 exposure with daily non-accidental mortality and respiratory mortality in Lishui district of China. Using attributable risk to estimate the death burden attributable to short-term exposure to O 3 and PM 2.5

The daily mortality records and the daily average concentration of atmospheric pollutants in Lishui district from January 1, 2015 to December 31, 2019, were derived from the Lishui Smart City Operating Command Center of Nanjin, the big data integration center of Lishui district government. The daily mortality records included the mortality data of the permanent population. Speci c information included age, gender, date of birth and death. We classi ed the causes of mortality based on the sole primary diagnosis coded by ICD-10 (International Statistical Classi cation of Diseases and Related Health Problems 10th Revision), including non-accidental mortality (A00-R99), and respiratory diseases (J00-J99). The environmental data included daily meteorological data and atmospheric pollutants data.

Statistical analysis
We used daily aggregated data from 2015 to 2019 to quantitatively assess the impact of PM 2.5 , and O 3 exposure on non-accidental mortality and respiratory mortality. Daily mortality, air pollution, and meteorological data were described with average standard deviations and quartiles. The relationship between air pollutants and meteorological conditions was evaluated using the spearman correlation. The data of mortality, air pollution levels, and meteorological factors were linked by the date. For the total population, the daily mortality of residents was a small probability event and obeyed the Poisson distribution and the correlation between explanatory variables and the number of mortalities per day was mainly non-linear. Thus, we constructed a generalized additive model (GAM) based on the Poisson distribution in which time-series analysis was used to establish the core model to estimate the association between mortality and air pollutant exposure. The model was as follows: Log[E(Y t )] =α + βX t + ns(Time,df)+ ns(Z t ,df)+DOW In this equation, t refers to the day of the observation; Y t is the number of daily mortalities observed on day t; E(Y t ) is the expected daily mortality rate on day t. α is the intercept; β represents the regression coe cient of the corresponding air pollutants; X t represents the pollutant concentration on day t; Z t represents the meteorological data on day t; DOW is a binary dummy variable; s is a non-linear variable with smoothing spline function. Previous studies [19][20][21][22] have usually set the degrees of freedom(df) of time to 5 to 7 and meteorological factors to 3 to 6. The degree of freedom was selected according to the minimum value of the Akaike information criterion (AIC) of the Poisson model, and the smaller AIC value indicates the preferred model [23]. Considering the applicability and AIC value of the model, 6 df was used to adjust the time trend, seasonality, and temperature, and 3 df was used to adjust relative humidity in the model.
The lag effect of air pollutants on non-accidental mortality and respiratory mortality was studied from the current day up to the 7th day (lag0-lag7). Previous studies have shown that cumulative effects may be underestimated by the single-day lag model [24]. Therefore, we further used the moving average of air pollutant concentrations from the 2ed day to the 8th day (lag01 to lag07) in the analysis. In addition, research has found that the decrease of PM 2.5 leads to the increase of photochemical ux and the acceleration of atmospheric oxidation, increasing of O 3 concentration [25]. As a result, we explored whether there is an interactive effect on the death of the two main pollutants in Lishui district, the twopollutant model was used to evaluate the confounding effect of pollutants. After establishing a statistical model that includes all control variables and checking the applicability, we separately included air pollutants into the model. In addition, the data were strati ed by gender (female and male), age (45-64 years, 65-84 years, 85 years or older), and seasons (spring, summer, autumn, winter).The results were expressed as excess risk (ER) and 95% con dence intervals (95% CI) of daily deaths associated with 10μg/m 3 increase in pollutants' concentration.
We further estimated the death burden attributable to short-term exposure to O 3 and PM 2.5. The counts of different death outcomes attributable to air pollutants were estimated using: AC ij = N ij *(RR ij − 1)/ RR ij . RR ij is the relative risk for disease j at lagi based on the relative risk functions. N ij is the death number of disease j at lagi. AC ij is the attributable counts of disease j at lagi. Then we calculate the total attributable counts of disease j (AC j ) by summing the AC ij of the study period. Finally, the population attributable fractions (PAF) were calculated by dividing the total AC j by the total number of deaths among middleaged and elderly people in Lishui district.
All Statistical analysis was performed using R software, version 4.0.3. The statistical signi cance of all analyses was de ned as P < 0.05. Table 1 shows the descriptive summary for daily mortality, air pollutants, and meteorological data in Lishui district, China, during 2015-2019. From 2015 to 2019, the total number of non-accidental mortality and respiratory mortality among the middle-aged and elderly (≥ 45 years) in Lishui district was 13,160 and 1,478 respectively. At the same time, a seasonal pattern of daily mortality was also observed, with higher mortality in winter (Fig. 1). The daily average temperature was 16.9℃ (Range: -6.7℃-34.7℃), the daily average relative humidity was 73.0% (Range: 28%-100%). The 24-hour average concentration of PM 2.5 was 43.57µg/m 3 (Range: 26µg/m 3 -171µg/m 3 ), and the maximum daily 8-hour average concentration of O 3 (MDA8 O3) was 100.13µg/m3 (Range: 2µg/m 3 -285µg/m 3 ). O 3 was moderately positively correlated with average temperature (r = 0.52, P < 0.05), was moderately positively correlated with the relative humidity (r = -0.38, P < 0.05), and was slightly negatively correlated with PM 2.5 concentration. PM 2.5 was moderately correlated with the temperature (r =-0.45, P < 0.05), and was slightly negatively correlated with the relative humidity (Fig. 2). Notes: The data were set by spring (Mar-May), summer (Jun-August), autumn (Sept-Nov) and winter (Dec-Feb) seasons.

Result
After adjusting the time, day of the week, and weather conditions, we evaluated the single-day lag effect (lag0-lag7) and multi-day moving average lag effect (lag01-lag07) on non-accidental mortality and respiratory mortality (  (Table 2). Table 2 The excess risk (95% CI) of daily mortality associated with 10 µg/m 3 increase.
The daily average concentration of PM 2.5 in Lishui district is 43.57µg/m 3 , which was higher than the National Ambient Air Quality Standard (NAAQS) rst-level standard, but lower than the second-level standard (the rst-level standard is 35µg/m 3 , the second-level standard is 75µg /m 3 ). The MDA8 O 3 was 100.13µg/m 3 , which was also higher than the NAAQS rst-level standard, but lower than the second-level standard (the rst-level standard is 100µg/m 3 and the second-level standard is 160µg/m). The seasonal uctuation of pollutant concentration was mainly manifested as follows. PM 2.5 was higher in spring and winter than in summer and autumn and reached its peak in summer. O 3 was higher in summer and autumn than in spring and winter, and peaks in winter. A seasonal pattern in the number of daily mortalities was also observed, with higher mortality in winter. This observed seasonal uctuation may be related to the increase in sources of pollutants and meteorological factors. In winter, industrial production, motor vehicle, and combustion emissions (such as coal, biofuels) are the most direct factors that produce PM 2.5 [26]. High temperature and su cient sunshine in summer are favorable conditions for photochemical reaction to produce O 3 [27]. Using chemical industrial solvents and emitting the volatile organic compounds and nitrogen oxides from automobile exhaust may cause high levels of O 3 [28].
In this study, we found that in the single pollutant model, PM 2.5 had acute effects on non-accidental mortality. Every 10µg/m 3 increase in PM 2.5 was associated with a 0.94 % (95%CI: 0.05%-1.83%) increase in non-accidental mortality at lag0. A study conducted in a highly polluted area in China found that 10µg/m 3 increase in PM 2.5 was associated with 0.36% (95% CI: 0.10%-0.63%) increase of non-accidental mortality [29]. Lin found that every 10µg/m 3 increase in PM 2.5 was associated with 1.5% (95% CI: 0.5%-2.5%) of non-accidental mortality among the elderly over 65 years old [30]. A study conducted in 75 cities in the United States showed that for every 10µg/m 3 increase in PM 2.5 , the non-accidental mortality rate increased by 1.18% (95% CI: 0.93%-1.44%) [31]. Another large-scale study involving multiple countries and regions found that for every 10µg/m 3 increase in PM 2.5 , the daily non-accidental mortality rate increased by 0.68% (95% CI: 0.59-0.77%) [32]. Although our data analysis results showed that the impact of PM 2.5 on non-accidental mortality in Lishui district was slightly higher, it was generally consistent with the results of previous research reports in China. This difference may be mainly related to the age difference of the exposed population. Moreover, the sources and chemical composition of PM 2.5 in different regions are different, which may also lead to different effects on mortality.
Besides, we also found that O 3 had acute effects on respiratory mortality. Every 10µg/m 3 increase in O 3 was associated with an increase in respiratory disease mortality by 1.35% (95%CI: 0.05%-2.66%) at lag7.
A study in Jinan showed that Every 10µg/m 3 increase in O 3 was associated with a 0.975% (95% CI: 0.463, 1.489) increase in respiratory mortality at lag3 [33]. Another study in Hefei showed that every 10µg/m 3 increase in O 3 led to a 2.22% (95%CI: 0.56%-3.90%) increase in respiratory mortality [29]. A Sichuan study found that every 10µg/m 3 increase in O 3 led to a 0.78% (95%CI: 0.12%-1.44%) increase in respiratory mortality [34]. Since the middle-aged and elderly population was the research object this time, the impact of O 3 in Lishui district on respiratory mortality would be slightly higher, but it was consistent with the results of domestic and foreign research [35][36][37]. With the rapid development of the economic level and the acceleration of the urbanization process, the output of industrial manufacturing was also increasing, which may lead to the increase of volatile organic compounds (VOCs) emissions [38]. This may be one of the reasons that O 3 in Lishui district had a greater impact on the respiratory mortality of the middle-aged and elderly. In this study, the impact of multi-day moving average lag was higher than that of single-day lag, but the effect was not statistically signi cant, which was consistent with Costa's research results [39].
Subgroup analysis showed that air pollutants were signi cantly related to non-accidental and respiratory mortality in different genders and seasons. Women were more sensitive to be affected by PM 2.5 on nonaccidental mortality. This was consistent with the research of Shin [40]and Hu [41]. Because women may have stronger airway responsiveness, combined with hormones or other factors, and therefore had a stronger physiological response to air pollutants [42,43]. However, there was also con icting research evidence that men were more susceptible to the impact of PM 2.5 on non-accidental mortality [44,45]. In contrast, we found that men were more susceptible to the effects of O 3 on respiratory mortality than women. A research carried out in Shenzhen also found the same result [46]. This can be explained by the fact that pneumonia and bronchitis were more common in men, and many men have a smoking history and different occupational exposures, which may exacerbate the impact of O 3 on respiratory mortality [47].
As far as the seasonal effect is concerned, the O 3 concentration in summer had a statistically signi cant effect on non-accidental mortality. This is consistent with the research of Zanobetti[48]. In summer, as the temperature rises, the ozone precursor substances in the air produce O 3 faster, which has harmed the health of the population [49]. Arch also found that although the concentration of O 3 in winter is at the lowest level throughout the year, the effect of O 3 on non-accidental mortality was also great. Wang's research results were consistent with ours. A study in Nanjing found that the concentration of indoor O 3 in winter may be greater than that of outdoor O 3 , and indoor O 3 produced by electrical equipment is harmful to people's health [50]. This may be because the middle-aged and elderly spend more time indoors in winter, and outdoor O 3 exposure cannot represent the actual O 3 exposure of them. Our research also found that ozone in spring and autumn had an effect on non-accidental mortality but was not statistically signi cant. Research conducted in many regions of East Asia found that O 3 levels in different seasons have varying degrees of impact on non-accidental mortality [51]. This may be due to geographical heterogeneity [52]. To identify susceptible groups, we also explored the potential modi cation effects of age, but in our study, we did not observe signi cant modi cation effects of age groups.
In This study has some limitations. First, we used the average concentration of air pollutants at the monitoring station as the population exposure level, without considering the indoor exposure. This would lead to exposure measurement errors and deviations in the accuracy and intensity of risk estimates.
Secondly, time-series analysis was an ecological study that requires a large sample size. The sample size of respiratory mortality in this study was relatively small, which may lead to unstable results. Finally, this study did not collect information on smoking history, body mass index, drug history, and educational level. These potential confounding factors may also have a potential impact on the association between air pollution and mortality.

Conclusion
In conclusion, this study shows that among the middle-aged and elderly in Lishui district, China, shortterm exposure to PM 2.5 and O 3 would increase the risk of non-accidental death and respiratory death, and air pollutants have a lag effect on the health of the population. This nding would help Lishui district authorities to adjust the existing air pollution management standards and implement more stringent air pollutant emission control policies, accelerating the construction of national health demonstration zones. Spearman correlation coe cients between daily air pollutants and meteorological parameters. The ER (95%CI) associated with 10 μg/m3 increase of mortality; (a) PM2.5 led to non-accidental mortality;(b) O3 led to respiratory mortality.