Association between ambient air pollution and daily hospital visits for cardiovascular diseases in Wuhan, China: a time-series analysis based on medical insurance data

ABSTRACT Although evidence showed the adverse effects of air pollution on cardiovascular disease (CVDs), few studies were based on medically insured populations. We applied a generalized additive Poisson model (GAM) to estimate the short-term effects of ambient air pollution on a group of medically insured population in Wuhan, China. We extracted daily air pollution data, meteorological data, and daily hospital visits for CVDs. We found that the ambient air pollutants sulfur dioxide (SO2), nitrogen dioxide (NO2), ground-level ozone (O3) particulate matter (PM) with an aerodynamic diameter ≤10 μm (PM10), and those ≤2.5 μm (PM2.5) all increased the risk of daily hospital visits for CVDs. We also found that the effect of air pollution on daily hospital visits for CVDs is greater in the cold season than in the warm season. Our findings can be used as evidence that supports the formulation of policies for air pollution and CVDs.


Introduction
Cardiovascular diseases (CVDs) are a group of disorders of the heart and blood vessels, such as coronary heart disease (heart attack) and hypertension (increased blood pressure). According to a report from the World Health Organization (WHO 2017), CVDs are major non-communicable diseases and the number one cause of death in the world. CVDs were estimated to cause 17.9 million deaths each year and are also responsible for the disability-adjusted life years at the national level. (Zhou et al. 2019;Kayikcioğlu and Oto 2020). The incidence of CVDs has doubled since 1990, reaching nearly 94 million in 2016, and it will continue to increase in the next ten years in China (Weiwei et al. 2016;Liu et al. 2019). The increase in the incidence of CVDs has had a serious adverse effect on the economy, posing a challenge to the healthcare system and the society as a whole (Wu et al. 2016). Given that 80% of cardiovascular deaths occur in low-and middle-income countries, a 10% reduction in CVD mortality from 2011 to 2025 is expected to reduce economic losses by $377 billion US dollars (Laslett et al. 2012). According to a report of the National Center for CVDs of China, the prevalence of CVDs in China is on a continuous rise. The report estimated that 290 million people are suffering from CVDs, of which 245 million are hypertension, 11 million are coronary heart disease, and 14 million are other types of heart disease such as congenital heart disease and rheumatic heart disease (Ma et al. 2020). The risk factors for CVDs include, but are not limited to, unhealthy lifestyle, elevated blood pressure, hyperlipidemia, hyperglycemia, and air pollution (Shen and Ge 2018).
In 2019, 99% of the world's population lived in places that did not meet the WHO air quality guidelines (WHO 2021). Ambient air pollution is associated with more than seven million premature deaths globally each year, most of which occur in low-and middleincome countries (Landrigan et al. 2018;Yusuf et al. 2020). Studies have found that the elevated concentration of air pollution is associated with the increased mortality rate in many cities in China (Chen et al. 2018). As one of the most polluted cities in China, Wuhan has experienced serious air pollution in the past decade (Mbululo et al. 2019). In China, exposure to air pollution has reportedly increased the number of hospitalizations and deaths from cardiovascular diseases. (such as coronary heart disease [CHD] and hypertension) Chen et al. 2019).
At present, China's epidemiological studies on air pollution and cardiovascular diseases mainly use disease data collected directly from hospitals. As a result, the impact of air pollution on the risk of CVDs in the insured population remains to be scrutinized. Since the population covered by the insurance system mostly represents the working population, research on such population has special public health significance. The working population is supposedly younger and healthier than the general public, so their health status not only affects the reasonable allocation of medical resources, but also affects the development of the national economy. The basic medical insurance database that we obtained the data from is an important part of China's social insurance system. The outpatient, emergency, and hospitalization data contained in the database provide us with an opportunity to examine the relationship between exposure to air pollution and the risk of CVDs in the working population. In this article, we assessed the association between the short-term exposure to air pollutants and the daily hospital visits for CVDs in Wuhan, China from November 1 st, 2013 to October 31 st , 2018. We hope that our research can make up for the lack of similar studies on medical insurance populations in China.

Data collection
China's basic medical insurance system consists of three parts, namely basic medical insurance system for urban employees, a basic medical insurance system for urban residents, and a new rural cooperative medical care system. Our data comes from the basic medical insurance database of Wuhan, China from 2013 to 2017. The database covers urban employees and residents at Wuhan city. Because the database is too large, we randomly select 1% of the data for analysis by the ID number of the insured population. We classified outcome groups by selecting the key fields of 'hypertension, heart disease, coronary heart disease' in the disease diagnosis category. The data also includes the age, gender, and socioeconomic status (such as occupation) of patients with sensitive personal information removed.
Data on air pollution was collected from ten National Ambient Air Quality Monitoring Stations in Wuhan (Figure 1). The daily concentration of each pollutant represents 24-h averages from all air quality monitoring sites in this study. Five air pollutants were included in our research, namely, nitrogen dioxide (NO 2 ), sulfur dioxide (SO 2 ), ground-level ozone (O 3 ), particulate matter (PM) with an aerodynamic diameter ≤2.5 μm (PM 2.5 ), and PM with aerodynamic diameter ≤10 μm (PM 10 ). Data on meteorological factors including ambient temperature and relative humidity were obtained from the Hubei Meteorological Service Center.

Statistical analysis
We built a time-series database based on the date, air pollutant concentrations, meteorological factors, day of the week, and hospital visits for CVDs. We used descriptive analysis to show the characteristics of hospital visits for CVDs, air pollutants, and meteorological factors. The Spearman correlation was used to estimate the relationship between daily data of air pollutants and meteorological factors. A generalized additive Poisson regression model was established to explore the short-term impact of daily air pollutant levels on hospital visits for CVDs (Ravindra et al. 2019). In the GAM model, a smoothing spline function was selected to control the confounding effects of the long-term trend and meteorological factors. The Akaike Information Criterion for quasi-Poisson (Q-AIC) (Akaike 1987) was conducted to determine the degrees of freedom (df) for time trend, relative humidity, and temperature. The model is as below: Where y t is the number of hospital visits at day t; E(yt) indicates the expected number of hospital visits for CVDs on day t; Xt represents the concentrations of air pollutants on day t; β indicates the regression coefficient, ns means a natural smoothing spline function, and DOW is an indicator variable meaning 'day of the week'. According to the minimum value of Q-AIC, we selected a smooth function of 8 df to control for long-term effects, 5 df to control temperature, and 3 df to control relative humidity.
We conducted single-pollutant models to explore the short-term effects of each air pollutant on hospital visits for hypertension, CHD, and HD. We explored the effect by using different lag structures, including a single-day lag from the current day up to the previous 7 days (lag 0-lag 7) and moving averages of the current and previous days (lag 0-1 -lag0-7). Also, we performed the seasonal analysis by dividing the annual data into the warm season (April-September) and cold season (October-March). If the correlation coefficient of two pollutants is less than 0.7 (Mukaka 2012), we put them in the two-pollutant model to explore the effect of each pollutant on hospital visits for hypertension, CHD, and HD. In addition, to explore the impact of multiple pollutants on CVDs, we also conducted the multi-pollutant model (Deng et al. 2015).
We conducted four sensitivity analyses to verify the robustness of the results. Firstly, we modified the df values between 7 and 9 for calendar time to achieve the best model fit. Secondly, we conducted stratified analyses by gender (male and female), age (65 and ≥65 years old), and socioeconomic status (blue-collar worker and white-collar worker) to further test the reliability of the results.
All statistical analyses were conducted by R-software (version 4..0) using 'mgcv' and 'nlme' packages. The results we obtained were reported as the relative risk (RR) and 95% confidence intervals (CIs) of hospital visits for CVDs associated with a 10 μg/m 3 increase in air pollutant concentrations. Effects with a p-value.05 were considered statistically significant.

Results
The descriptive statistics of air pollutants, meteorological factors, hypertension, CHD, HD from November 2013 to October 2018 were shown in Table 1. During the five-year study period, 1,153,045 patients with hypertension, 180,777 patients with CHD, and 202,683 patients with HD were included in our analysis. The average daily hospital visits for hypertension, CHD, HD were 631, 98, and 110, respectively. There were more hospital visits for patients with hypertension in men than women, but the opposite is true for CHD. Hospital visits for hypertension, CHD, HD were higher in the elderly (≥65 years old) and blue-collar workers. The daily mean concentrations of SO 2 , NO 2 , O 3 , PM 10 , and PM 2.5 were 17.50 μg/m 3 , 47.60 μg/m 3 , 55.47 μg/m 3 , 96.32 μg/m 3 , and 63.31 μg/ m 3 , respectively. The average ambient temperature and relative humidity were 17.28 and 78.25% in the 1,827 days of observation during 2013-2018. The distribution of air pollutant concentrations, meteorological factors, and hospital visits for hypertension, CHD, and HD is presented in Fig. S1. Table S1 shows the Spearman correlation coefficients of air pollutants, ambient temperature, and relative humidity, which range from 0.007 (ambient temperature and relative humidity) to 0.740 (PM 2.5 and PM 10 ). SO 2 , NO 2 , PM 10 , and PM 2.5 were all positively correlated with each other and were negatively correlated with ambient temperature and relative humidity. O 3 was positively correlated with ambient temperature (r = .659), while it was negatively correlated with the remaining four air pollutants and relative humidity. Figure 2 shows the RRs and 95% CIs of hospital visits for hypertension, CHD, and HD associated with a 10 μg/m 3 increase in pollutant concentrations at lag 0, 1, 2, 3, 0-3 in the single pollutant model. The complete results regarding the single-day lag model and the cumulative day lag model were shown in Fig. S2. The effects of exposure to air pollution on hypertension, CHD, and HD have similar trends. SO 2 , NO 2 , PM 10 , and PM 2.5 were significantly associated with daily hospital visits for CVDs. In the single-day lag model, the effects of the four air pollutants on daily hospital visits for CVDs is highest at lag 0 and then shows a downward trend. For each 10 μg/m 3 increase in SO 2 , NO 2 , PM 10 , and PM 2.5 concentrations, the RRs of daily hospital visits for hypertension increased by 3.8% (95%CI: 1.8%, 5.9%), 2.5% (95%CI: 1.9%, 3.2%), 0.5% (95%CI: 0.2%, 0.7%), and 0.7% (95%CI: 0.3%, 1.1%) at lag 0, respectively; the RRs of daily hospital visits for CHD increased by 3.6% (95%CI: 1.8%, 5.5%), 2.6% (95%CI: 1.9%, 3.4%), 0.4% (95%CI: 0.1%, 0.7%), and 0.5% (95%CI: 0.1%, 0.9%) at lag 0, respectively; the RRs of daily hospital visits for HD increased by 3.6% (95%CI: 1.4%, 5.8%), 2.1% (95%CI: 1.4%, 2.7%), 0.3% (95%CI: 0.1%, 0.6%), and 0.4% (95%CI: 0, 0.8%) at lag 0, respectively. In the multi-day lag model, the effects of SO 2 and NO 2 on hypertension, CHD, and HD remain significantly at lag 0-3. The seasonal analysis shows the RRs and 95% CIs of hospital visits for hypertension, CHD, and HD associated with a 10 μg/m 3 increase in pollutant concentrations at different lag days during the cold and warm seasons (Figure 3). In both the single-day lag model and the multi-day lag model, SO 2 , NO 2 , PM 10 , and PM 2.5 have a stronger effect on hypertension, CHD, and HD in the cold season than in the hot season. On the contrary, O 3 has a stronger effect in the warm season than in the cold season.
The results of RR and 95% CIs of hospital visits for hypertension, CHD, and HD based on the single-and two-pollutant models were shown in Table 2. The results of multi-pollutant models were presented in Table S2. Considering the collinearity between pollutants, the correlation coefficient between the two pollutants has to be less than 0.7 to include them in the two-pollutant model, otherwise would be excluded from the analysis. The effect of PM 10 and PM 2.5 on CHD and HD decreased after the adjustment for SO 2 in the two-pollutant model. Meanwhile, after adjusting for NO 2 , the effects of SO 2 and PM 2.5 on hypertension, CHD, and HD become statistically insignificant. In addition, NO 2 can strengthen the effect of O 3 on CHD and HD. Table 3 shows the results of the stratified analysis by gender (male and female), age (65 and ≥65 years old), and socioeconomic status (blue-collar worker and white-collar worker) at lag 0. We have not observed considerable differences in the effects of air pollution on CVDs in terms of age, gender, and socioeconomic status. All results regarding the stratified analysis are presented in Fig. S3 (gender), Fig. S4 (age), and Fig. S5 (socioeconomic status). Since the number of CVDs in late 2017 to the most of 2018 was significantly higher than the previous year (see Fig. S1), we compared the differences in the effects of air pollution on CVD in the months when the number of visits was high and the effects in the previous year (the months when the number of visits was low). The results show that similar trends were observed in both periods (Fig. S6). Fig. S7, Fig S8, and Fig S9 show that the exposure-response (E-R) relationship curves for the associations between air pollutants and daily hospital visits for hypertension, CHD, and HD.

Discussion
We conducted a time-series analysis to explore the association between air pollution and hospital visits for CVDs from 2013 to 2018 in Wuhan, China. We found that the short-term exposures to SO 2 , NO 2 , PM 10 , and PM 2.5 were significantly associated with the risk of hypertension, CHD, and HD. The stratified analysis also showed that the significant association between O 3 and hospital visits for CVDs was only observed in the warm period. Our results show that the largest effect of air pollution on CVDs occurred at lag 0 (single-day lag) and lag 0-3 (multi-day lag). We also found that the association is stronger in the cold season than in the warm season. As far as we know, this is the first study to explore the association between daily hospital visits for CVDs and air pollution in Wuhan based on medical insurance data.
In the past few decades, the environment in China has faced great challenges due to rapid industrial development and urbanization. The increase in the number of vehicles and energy consumption and the decrease in green coverage has affected air quality to varying degrees. Continuous haze weather is commonly seen in China, causing serious environmental hazards, especially in cities with large populations (Maji et al. 2018). Wuhan, the largest city in central China, has a population of 12 million and is one of the most polluted cities in China. Our study suggested that the 24-hour average concentrations of PM 10 (96.09 µg/m 3 ) and PM 2.5 (63.03 µg/ m 3 ) far exceeded the air quality standards set by the WHO (45 µg/m 3 and 15 µg/m 3 , respectively). The impact of air pollution on the cardiovascular system is frequently reported worldwide. A study conducted in Ahvaz, Iran indicated that the risk of hospital admission for CVDs increased by 0.6% (95% CI: 0.1 to 1%) for every increase of 10 μg/m 3 of NO 2 at lag 0 (Dastoorpoor et al. 2019). A time-series study by Zhang et al. (2017) found that exposures to SO 2 , NO 2 , and PM 10 were associated with a 5.26% (95%CI: 3.31% to 7.23%), 2.71% (95%CI: 1.23% to 4.22%), and 0.68% (95%CI: 0.33% to 1.04%) increase in cardiovascular mortality at lag 0-3. Another study conducted by systematic review and meta-Analysis reported that short-term exposure to SO 2 , PM 2.5 , and PM 10 is significantly associated with a 4.6% (1.2% to 8.1%), 6.9% (0.3% to 14.1%), and 2.4% (1.6% to 3.2%) increase in the risk of hypertension (Cai et al. 2016). Our findings are generally consistent with the results of past studies in two dimensions. One is the effects of air pollution on cardiovascular disease and the second is the time or moment when the strongest effects occur. However, some studies have also reported different findings. A time-series study conducted in Guangzhou, China found that PM has no significant effect on CVDs ). In addition, we did not observe a significant association between the short-term exposure to O 3 and the increased mortality of CVDs that other study has already reported (Mazidi and Speakman 2018). This inconsistency may be due to the following reasons: 1) the statistical methods used in these studies are different from ours; 2) the different geographical location of the study may cause differences in results; 3) the demographic characteristics of the exposed population are different.
The latest scientific statement from the American Heart Association believes that there is a causal relationship between exposure to PM 2.5 and cardiovascular morbidity and mortality (Brook et al. 2010). The possible physiological and molecular mechanisms involved are still in the process of exploration. One of the possible mechanisms is that inhalation of particulate matter in the environment can cause the body to produce pro-oxidant substances (reactive oxygen species, etc.), pro-inflammatory biological mediators (interleukin 6, etc.), the acute phase reactants (C-reactive protein, etc.), and vasoactive hormones (endothelin, etc.), resulting in systemic inflammation and oxidative stress. These reactions in the lung will eventually affect the cardiovascular system through blood circulation (Gurgueira et al. 2002;). Meanwhile, evidence suggests that particulate matter can pass through the lung epithelium into the circulatory system or interact with lung receptors (direct action) to induce an acute cardiovascular response (Nemmar et al. 2002;Fiordelisi et al. 2017). In addition, air pollutants can cause the body's autonomic nervous system to malfunction and activate pathways of the central nervous system, leading to increased blood pressure and heart rate variability. Due to inhalation of air pollutants, the nose, bronchus, and lung C nerve fiber subtypes will activate many receptors that can affect sensory nerves. (Rajagopalan et al. 2018).
The seasonal analysis shows that the effect of most air pollutants on hospital visits for CVDs is stronger in the cold season than in the warm season except for O 3 . The results are consistent with previous studies (Song et al. 2019). Like our study, a previous study (Brook and Kousha 2015) also found that the effect of O 3 on CVDs is also more pronounced in the warm season rather than in the cold season. This effect may be due to a potential link between temperature and ozone. Ozone is a photochemical air pollutant. It is formed by the reaction of nitrogen oxides and volatile organic compounds (VOCs) with oxygen in the atmosphere exposed to sunlight (Li et al. 2017). Therefore, more ozone is produced at high temperatures, which may damage fibrinolysis and affect thrombus removal (Kahle et al. 2015).
In addition to the single-pollutant models, we also constructed two-pollutant models to assess the effect of air pollution on CVDs. To avoid collinearity, two pollutants included in the model have to meet the criteria that their correlation coefficients must be less than 0.7. After including the second pollutant in the model, the results are slightly different from those produced by the single-pollutant model. However, it should be noticed that after adjusting for SO 2 or NO 2 , the effects of co-pollutants on CVDs are universally changed (intensified or weakened). This may indicate that these two pollutants are the major confounders between air pollutants and cardiovascular diseases.
A strength of our study is that we use the basic medical insurance data to study the relationship between air pollutant concentrations and the hospital visits for CVDs. This study has several limitations. First, we obtained the data on pollutants from air quality monitoring stations rather than personal exposure samplers. Also, we assume that everyone is exposed to the same level of air pollution each day. This may lead to exposure misclassification, an inevitable flaw in such types of ecological studies. Second, our data comes from basic medical insurance data, so the conclusion does not apply to the general public. Third, we lack the monitoring of the concentration of indoor air pollutants. Given that people spend a large part of their time indoors, indoor air pollution may have an impact on CVDs.

Conclusions
We found that exposure to SO 2 , NO 2 , PM 10 , and PM 2.5 increased the risk of daily hospital visits for CVDs in Wuhan, China, especially during the cold season. Our findings not only can be used as a reference studying the financing and the allocation of medical resources in the future but can also be used served as a piece of evidence that supports the formulation of policies for air pollution and CVDs.