Introduction

Air pollution is a major threat to health worldwide, especially in urban areas (Brunekreef and Holgate 2002; Katsouyanni et al. 2001; Samet et al. 2000; Schwela 2000; Seaton et al. 1995). For instance, more than two million premature deaths each year can be attributed to the effects of air pollution, according to the World Health Organization’s (WHO) air quality guidelines. One source of air pollution, particulate matter, is a complex mixture of microscopic particles and liquid droplets that contain acids, organic chemicals, metals, and soil or dust particles. The WHO estimates that PM2.5 concentrations in the environment contribute to approximately 800,000 premature deaths per year, and are the 13th highest cause of mortality worldwide.

The most direct impact of air pollution on human health is the induction of respiratory disease. Many epidemiological studies have observed associations between air pollutants and negative health effects, such as increased hospitalizations, greater emergency hospital visits, and higher mortality levels (DeVries et al. 2017; Dominici et al. 2006; Hansel et al. 2016; Qiu et al. 2012; Song et al. 2014; Stieb et al. 2009). The relationship between hospital admissions for respiratory disease and air pollution has been well assessed in Europe and North America, but less so in Asia, where the components of pollution, meteorological conditions, and demographic characteristics may vary according to geographical region. In China, the largest developing country in the world, rapid urbanization and industrialization not only bring economic prosperity, but also bring environmental contamination; China is experiencing deteriorating air quality due to automobile exhaust emissions and energy consumption. Air pollution caused by particulate matter has not been well controlled; thus the frequency of hazy weather is increasing. Because data on fine particulate matter (PM2.5) in China were first released in 2013, relevant studies are still comparatively rare in the literature.

To close this gap, we assessed the short-term effects of air pollutants on respiratory disease (i.e., pneumonia, chronic obstructive pulmonary disease [COPD], and asthma) hospitalizations in Hefei, China. Our dataset includes information of daily hospital admissions, six principal air pollutants, and daily meteorological data. The paper proceeds as follows. We describe our data and methods in the “Materials and methods” section. We report our main results in the “Results” section and discuss our findings in the “Discussion” section. In the “Conclusions” section, we draw the study conclusions.

Materials and methods

Study area

This study was conducted in Hefei (Geographical Coordinates: 32° 52′ N, 117° 17′ E), which is the capital city of Anhui Province and located between the Yangtze River and the Huaihe River, and beside the Chaohu Lake. It is a mid-China city nearest to the coastal cities in eastern China (Fig. 1). Hefei enjoys a subtropical humid monsoon climate featuring mild climate with distinct seasons and appropriate precipitation. Hefei was ranked the 17th most polluted city in China on the basis of its PM2.5 concentration (Greenpeace 2014).

Fig. 1
figure 1

Location of Hefei in China

Full size image

Daily hospital admissions data

Daily hospital admissions data between January 1, 2014 and March 31, 2016 were obtained from the New Rural Cooperative Medical System (NRCMS), which is a heavily subsidized voluntary health insurance program designed to reduce the risk of catastrophic health spending for rural residents in China. Patient data included gender, age, residential address, date of admission, hospital level, and principal diagnosis. Data were coded according to the International Statistical Classification of Diseases and Related Health Problems, 10th Revision (ICD-10) and Diagnosis-related Groups (DRGs) in Anhui Province, China. The following disease categories were analyzed separately: pneumonia (ICD-10 code: J12-J18; DRG code: ZDZ347); COPD (ICD-10 code: J44.901; DRG code: ZDZ348); and asthma (ICD-10 code: J45; DRG code: ZDZ267).

Air pollutants and meteorological data

Air pollutants and meteorological data were collected for the same period. Data on air pollution were obtained from an online platform (https://www.aqistudy.cn) that monitors and analyzes six air pollutants: PM2.5 (particles with diameters ≤ 2.5 μm), PM10 (particles with diameters ≤ 10 μm), SO2 (sulfur dioxide), NO2 (nitrogen dioxide), CO (carbon monoxide), and O3 (ozone). Daily concentrations for each pollutant were measured at ten fixed-site stations in Hefei (Pearl Plaza, Sanli Street, Amber Villa, Dongpu Reservoir, Changjiang Middle Road, Luyang District, Yaohai District, Baohe District, New Binhu District, and High-tech District) under the Ministry of Environmental Protection of the People’s Republic of China (Fig. 2). Meteorological data on daily average temperature (T), atmospheric pressure at sea level (SLP), average relative humidity (H), average visibility (VV), and average wind speed (V) were obtained from Weather Underground (https://www.wunderground.com).

Fig. 2
figure 2

Locations of ten air pollution monitoring stations in Hefei

Full size image

Statistical analysis

For descriptive statistics, we analyzed respiratory disease hospitalizations and daily data on air pollutants and meteorological conditions using stratification factors. We used Spearman’s rank correlation test to evaluate the association between air pollutants and meteorological variables.

The daily number of hospital admissions was the outcome variable. We first calculated the mean and variance of the number of daily admissions and found the variance is slightly greater than the mean (e.g., for asthma: mean = 1.17, variance = 1.27). A common way to deal with overdispersion for counts is to use a generalized linear model framework (McCullagh and Nelder 1989). Therefore, a generalized Poisson regression model and a distributed lag nonlinear model (DLNM) were combined to evaluate short-term effects of air pollution on hospitalizations. The equations were as follows:

$$ {Y}_t=\alpha +\sum \limits_{i=1}^6{\beta}_i{X}_{i,t,l}+ ns\left(T, df\right)+ ns\left(H, df\right)+ ns\left( SLP, df\right)+ ns\left( VV, df\right)+ ns\left(V, df\right)+ ns\left( time, df\right)+\delta {DOW}_t, $$

where t is the day of the observation; Yt is the number of observed hospital admissions on day t; α is the intercept; Xi, t, l refers to the cross-basis objects obtained by applying the DLNM to six criteria air pollutants (PM2.5, PM10, SO2, NO2, CO, O3); i is the sequence number of these air pollutants; l is the lag days and βi is the vector of coefficients for Xi, t, l; T, H, SLP, VV, and V represent the aforementioned meteorological variables; ns( ) represents a natural cubic spline; df is the degree of freedom; DOWt is a categorical variable for the day of week on day t; and δ is the vector of coefficient.

In the base model, air pollutant concentration was added as a linear term. To control for spatial variation, daily pollutant concentrations were averaged from the available monitoring results of ten fixed-site stations (Cao et al. 2009). Meteorological factors, however, were considered as nonlinear terms. We controlled for them by using a natural cubic spline with the degree of freedom. Further, we employed the Akaike information criterion (AIC) to determine the df for meteorological factors (Armstrong 2006; Gasparrini et al. 2010). Meanwhile, to control for the long-term trends and seasonality, we used a natural cubic spline with 7 df per year of data for the variable time. Moreover, we used a categorical variable DOW in research model to control for the day of week effect (Lokys et al. 2018). Based on the vector of estimated coefficients in the equation, the DLNM was used to predict the effects and standard errors for combinations of air pollutants and time lags (Armstrong 2006). Fourth-degree polynomial-distributed lag models for air pollutants were adopted to estimate lag structures. To completely capture the overall effect and adjust for any potential harvesting, we used lags up to 21 days, in line with a previous study (Barnett et al. 2012; Sun et al. 2016). The cumulative effects of air pollutants on hospital admission over 0–21 days were examined and plotted.

We also explored the potential effect modification of respiratory disease risk by age (< 65 years and ≥ 65 years); gender; and season (warm season: May to October; cold season: November to April). Sensitivity analyses were carried out to examine the impact of the degree of freedom selection on the effect estimates of pollutants. We changed the degree of freedom for time, temperature, and relative humidity. We also varied the maximum lag days for the DLNM. All statistical analyses were conducted using R software, version 3.5.0, and the dlnm package (Gasparrini 2011).

Results

Table 1 presents descriptive statistics for daily hospital admissions from January 1, 2014 to March 31, 2016, in Hefei, China. Of the total of 19,465 hospital admissions for respiratory disease, 14,929 were pneumonia admissions; 3579 COPD admissions; and 957 asthma admissions. More males were hospitalized for these diseases than females. For example, males with pneumonia had 8692 hospitalizations, accounting for 58.2%. Except for COPD, there were more admissions for patient younger than 65.

Table 1 Demographic characteristics of daily hospital admission in Hefei, China, January 1, 2014 to March 31, 2016
Full size table

Table 2 presents summary statistics for daily hospital admissions, air pollution concentrations, and meteorological conditions during the study period. Mean daily counts for pneumonia, COPD, and asthma admissions were around 24, 4, and 1, respectively. The daily mean concentration of air pollutants was 73.9 μg/m3, 101.9 μg/m3, 18.2 μg/m3, 1.0 mg/m3, 30.9 μg/m3, and 66.9 μg/m3 for PM2.5, PM10, SO2, CO, NO2, and O3, respectively. Average daily temperature, atmospheric pressure at sea level, relative humidity, visibility, and wind speed were 15.4 °C, 1017.9 hPa, 73.0%, 6.0 Km, and 9.9 Km/h, respectively.

Table 2 Summary statistics for the daily number of respiratory disease admission, daily air pollution concentrations and meteorological conditions in Hefei, China, January 1, 2014 to March 31, 2016
Full size table

The Ambient Air Quality Standards (Ministry of Ecology and Environment of the People’s Republic of China 2012) of China specify annual average concentration limits for various pollutants. The limits for PM2.5, PM10, SO2, CO, NO2, and O3 are 35 μg/m3, 50 μg/m3, 50 μg/m3, 4 mg/m3, 80 μg/m3, and 100 μg/m3, respectively. Comparing the daily means of air pollution concentrations with the standards, PM2.5 and PM10 were far beyond the limits, which indicates that Hefei suffered from serious haze pollution during the study period. Therefore, we mainly examined the short-term effects of PM2.5 and PM10 on hospitalizations. Figure 3 presents the time-series distributions of PM2.5, PM10, and daily hospitalizations in Hefei from January 2014 to March 2016.

Fig. 3
figure 3

The time-series distribution of PM2.5, PM10, and daily hospitalization from January 2014 to March 2016 in Hefei

Full size image

Table 3 presents the Spearman correlations between air pollutants and meteorological variables. PM2.5, PM10, SO2, CO, and NO2 were correlated with each other, and their coefficients ranged from 0.46 to 0.80. They were also positively associated with atmospheric pressure at sea level. Most of the pollutants were negatively correlated with both temperature and humidity. We estimated the effects of atmospheric particulate matter individually using single pollutant models in the main analyses, because PM2.5 and PM10 were correlated with statistical significance (correlation coefficient = 0.79) and may be interactive with each other.

Table 3 Spearman’s rank correlation between air pollutants and meteorological variables in Hefei, China, January 1, 2014 to March 31, 2016
Full size table

Figure 4 shows the cumulative effects of a 10 μg/m3 increase in PM2.5 and PM10 levels along 21 lag days, together with the 95% confidence interval. Specific values are shown in Supplementary Tables S1–S3. For COPD, these two pollutants were significantly associated with hospitalizations. For example, the short-term effects of PM2.5 were statistically significant during the lag of 6–16 days, and that of PM10 were significant during the lag of 7–11 days. The cumulative effect of PM2.5 peaked on the 12-lag day (RR = 1.068, 95% CI 1.017 to 1.121). This means that for every 10 μg/m3 increase in PM2.5, the number of COPD hospital admissions increased by 6.8% at the 12-lag day. As for the cumulative effects of PM10, the number of COPD admissions increased by 3.1% at the peak 10-lag day (RR = 1.031, 95% CI 1.002 to 1.060). Asthma hospital admission was associated with PM10 at lag 7–14 days, and cumulative effects peaked on the 12-lag day (RR = 1.057, 95% CI 1.010 to 1.107). Pneumonia admissions did not exhibit any statistically significant relationships with any pollutants.

Fig. 4
figure 4

Cumulative relative risk of respiratory disease hospitalization in associations with a 10 μg/m3 increase in PM2.5 over 21 days, together with its 95% confidence interval

Full size image

The associations between each air pollutant and the daily number of hospital admissions can be stratified by season (warm and cold), gender, and age (< 65 years and ≥ 65 years). Figure 5 shows the cumulative effects of particle pollutants on pneumonia hospitalizations in each subgroup. We found that the cumulative effects were not statistically significant for pneumonia admissions for all season, gender, and age subgroups.

Fig. 5
figure 5

Cumulative relative risk of pneumonia hospitalization in associations with a 10 μg/m3 increase in PM2.5 and PM10 over 21 days strata by season (warm and cold), gender (male and female), age (< 65 years and ≥ 65 years), together with its 95% confidence interval

Full size image

Figure 6 shows the cumulative effects of particle pollutants on COPD hospitalizations in each subgroup. COPD admissions had a stronger association with the warm season than the cold season. For the short-term effects of PM2.5, significant lag effects were observed in females and older patients (aged ≥ 65 years). For every 10 μg/m3 increase in PM2.5, the short-term effects for females began from the 8-lag day to the 15-lag day, with the greatest effects on the 13-day lag (RR = 1.120, 95% CI 1.032 to 1.216). PM2.5 effects on older patients began from the 6-lag day to the 17-lag day, and cumulative relative risk peaked on the 12-lag day (RR = 1.093, 95% CI 1.035 to 1.154). For the short-term effects of PM10, COPD admissions were also more sensitive to females and older patients. For every 10 μg/m3 increase in PM10, the short-term effects on females began from the 9-lag day to the 11-lag day, and the effects among older patients began from the 7-lag day, and then gradually increased with each lag.

Fig. 6
figure 6

Cumulative relative risk of COPD hospitalization in associations with a 10 μg/m3 increase in PM2.5 and PM10 over 21 days strata by season (warm and cold), gender (male and female), age (< 65 years and ≥ 65 years), together with its 95% confidence interval

Full size image

Figure 7 shows the cumulative effects of particle pollutants on asthma hospitalizations in each subgroup. Older individuals were more likely to be affected by PM10. The cumulative relative risk of admission with a 10 μg/m3 increase in PM10 was statistically significant for older patients at lag 1–15 days, and peaked on the 12-lag day (RR = 1.066, 95% CI 1.015 to 1.119).

Fig. 7
figure 7

Cumulative relative risk of asthma hospitalization in associations with a 10 μg/m3 increase in PM2.5 and PM10 over 21 days strata by season (warm and cold), gender (male and female), age (< 65 years and ≥ 65 years), together with its 95% confidence interval

Full size image

Discussion

In this study, we used DLNM to assess the short-term effects of atmospheric particulate matter on hospitalizations for three kinds of respiratory disease in Hefei, China. For pneumonia admissions, these estimates were not statistically significant. For COPD admissions, however, both PM2.5 and PM10 had a positive association. Also, the effects of air pollutants were more pronounced in the warm season than in the cold season, and females and individuals older than 65 years suffered a higher risk from PM2.5 exposure. For asthma, PM10 was significantly associated with admissions. Compared with younger patients, estimates were stronger for older individuals with statistical significance.

Numerous short-term effect studies have found that transient pulmonary function decline and exacerbation of respiratory symptoms are associated with particulate matter exposure. Since the associations between air pollutant concentrations and respiratory hospitalization vary by region and population, we compared our results with studies conducted in other cities. A study conducted in ten US cities (Zanobetti et al. 2000), using a Poisson regression model that simultaneously considered the effects of PM10 up to a lag of 5 days, found that a 10-μg/m3 increase in PM10 resulted in a 2.5% (95% CI 1.8 to 3.3) increase in COPD admissions and a 1.95% (95% CI 1.5 to 2.4) increase in pneumonia. Another US study, based on the National Morbidity, Mortality and Air Pollution Study statistical model, observed that the largest effects of PM2.5 on hospital admissions occurred at lags of 0 and 1 for COPD (Dominici et al. 2006). A study conducted in Bordeaux, France, assessed the associations between daily levels of PM10, NO2, and O3 and medical home visits. The risk of visits for upper and lower respiratory diseases was significantly increased by 1.5% (95% CI 0.3 to 2.7) and 2.5% (95% CI 0.5 to 4.4), respectively, during the 3 days following a 10-μg/m3 increase in PM10 levels (Larrieu et al. 2009). In Istanbul, Turkey, the highest association of each pollutant with respiratory hospital admissions was observed with PM2.5 at lag 4 and PM10 at lag 0 for an increase of 10 μg/m3 in concentrations of the pollutants (Capraz et al. 2017). Similarly, a study in Hong Kong found that a 10-μg/m3 increase in PM2.5 and PM10 concentrations was associated with a 3.1% (95% CI 1.026 to 1.036) and 2.4% (95% CI 1.021 to 1.028) increase, respectively, in COPD admissions with a cumulative lag of 0–5 (Ko et al. 2007). Although the relative risks for respiratory diseases in Hefei were similar to those of previously reported work, we found differences in the maximum effect of the air pollutants on hospitalizations regarding the number of cumulative lag days. One reason may be that we used DLNM to simultaneously estimate both the nonlinear and delayed effects of air pollution on hospitalization (Gasparrini et al. 2010). Among the various approaches that have been proposed to examine the impact of air pollution on hospitalization, the generalized additive model (GAM) has emerged as a primary method. However, these models would result in high collinearity and deviation in the analysis. Moreover, we investigated lag effects over 21 days, which is longer than most previous studies. Another reason may be that the local population has been exposed to high levels of air pollution for a long time, which may reduce its susceptibility to pollution. Therefore, the short-term effects in Hefei lasted longer than those observed in other cities.

An interesting finding is that the effects of PM2.5 on daily admissions were higher than the effects of PM10. Researchers now generally agree that PM2.5 is more detrimental to human health than PM10 (Li et al. 2016), because PM2.5 can be inhaled more deeply into the small airways and alveoli of the lung. A larger surface area makes it possible for each mass to have a greater concentration of adsorption or toxic air pollutants, whereas PM10 may be deposited in the upper portion of the airway. In the lung, particulate matter can induce alveolar inflammation, which may aggravate underlying lung diseases (Dominici et al. 2006; Li et al. 2016).

In this study, we found a seasonality of air pollution effects on COPD hospital admissions, which rose in warm weather in agreement with several previous studies conducted in US and European cities. Previous work has demonstrated that the air pollution mixture may include a higher proportion of toxic components during warmer months (Peng et al. 2005). In addition, exposure to air pollutants may increase during warm months because of increased outdoor activities and open windows, leading to distinctive health outcomes (Stafoggia et al. 2016).

In the stratifying analysis according to gender, for COPD, the relative risks of exposure to fine particulate matter for males are smaller than those for females. This is consistent with the results of previous studies that suggest females are more vulnerable to outdoor air pollution (Anderson et al. 1997; Bell et al. 2015; Tao et al. 2014). Evidence from genetic and biological perspectives has shown that women have greater deposition of air pollution particles in their lungs than men, leading to a higher susceptibility to health conditions (Mitsakou et al. 2007). Moreover, differences in health by gender are also affected by genetic and biological differences, lifestyle behaviors (tobacco and alcohol consumption, exercise and diet), and psychosocial factors (critical life events, stress, psychological resources) (Denton et al. 2004).

Regarding age-specific effects, we found higher estimates of air pollutant-associated COPD hospitalizations for people older than 65 years. A European project investigated the short-term health effects of particulates in eight European cities and found that admissions for COPD plus asthma in patients older than 65 years increased by 1.0% (95% CI 0.4 to 1.5) after the concentration of PM10 increased by 10 (Anderson et al. 1997). Surprisingly, the same results were observed in the association between asthma admissions and PM10. Young children with asthma have long been considered a vulnerable group to adverse effects from air pollution (Weinmayr et al. 2009). However, a piece of research demonstrated that seniors (≥ 65 years) with asthma are more likely to face adverse effects associated with air pollution, but scant literature was available to support this point (Guarnieri and Balmes 2004). Our research provides evidence for that from an epidemiological perspective. Likewise, in light of previous research which reported that the elderly was the most vulnerable group to the effect of PM10 in South Korea, and the relative risk of asthma admission rates with 10 μg/m3 increase of PM10 significantly increased 2.1% (RR= 1.021, 95% CI 1.013 to 1.029) for the elderly (Park et al. 2013). Many physiological regulatory functions have been shown to weaken with age (Burns 2004), and elderly individuals’ bodies are therefore less able to withstand the effects of air pollutants, which may increase the risk of health events.

This study examined the associations between PM2.5 and PM10 and respiratory disease hospitalizations in Hefei, China. We used DLNM and longer lags to estimate the short-term effects, which produced a high predictive ability and overcame several limitations of previous research.

Our study has several limitations. First, there are some other risk factors (e.g., pollen levels, meteorological parameters) which may influence the admissions of respiratory disease. For example, positive associations between pollen levels and asthma morbidity (Murray et al. 2006). However, in this study, no information was used to control possible aeroallergen effects. In addition, previous studies have shown nonlinear relationship between absolute temperature and childhood asthma (Xu et al. 2018). Therefore, understanding whether air pollutants and other environment factors work independently or jointly in triggering the occurrence of respiratory disease should be studied further. Second, our data are from a city, and thus it may be challenging to extend our results to other cities due to climate and other differences. Third, air pollution data were collected from fixed-site monitoring stations rather than individual exposure, so there might be some inevitable assessment error. Furthermore, individual characteristics such as genetic factors, occupational exposure, and smoking history would affect the morbidity of respiratory diseases. Unfortunately, our database did not contain such individual information. Finally, there is no standard for selection of the maximum lag days.

Conclusions

Associations between air pollutants (e.g., PM2.5 and PM10) and hospitalizations for respiratory diseases (e.g., pneumonia, COPD, and asthma) do exist with a lag effect in Hefei, China. Furthermore, our results suggest that season, gender, and age modify the short-term effects of air pollution. Our findings strongly support local authorities in developing targeted prevention and control measures to protect vulnerable groups from being adversely affected by the serious air pollution.