Introduction

In recent years, due to the fast development of urbanization in China, PM2.5 has rapidly become a serious problem (Wang et al. 2017). Previous epidemiological studies have pointed out that PM2.5 has a great impact on human health (Kioumourtzoglou et al. 2016; Bell et al. 2007 and Schlesinger 2007; Shi et al. 2016). Especially, PM2.5 has a strong impact on the incidence of respiratory diseases (Polezer et al. 2018; Liu et al. 2016; Hakantecer et al. 2008). From 1996 to 2016, the prevalence and severity of asthma have significantly increased around the world (Maio et al. 2016), and the affected population was up to 300 million (Torgerson et al. 2011). Generally, children are more susceptible to asthma than adults (Guyatt et al. 1997). In the USA, the under-4-year-olds account for 71.8% of the total patients with persistent asthma (Zahran et al. 2014). Asthma has been found to be associated with several environmental factors such as air temperature and vapor pressure (Ito et al. 1992), as well as fine particulate matter (Peng et al. 2008).

Worldwide, many studies have been conducted to explore the correlations between fine particulate matter and asthma incidence (Yun et al. 2002; An 2013; Han et al. 2014). As compared with coarse PM, fine PM are more closely related to the increase of asthma prevalence and morbidity (Keet et al. 2018). Several researches found that PM1.0 were closely associated with climate, air quality, and human health (Zang et al. 2018; Zang et al. 2019; Zou et al. 2017). Although people began to pay attention to and monitor PM1.0, studies on the relationship between PM1.0 and health are still rare. For example, in Canada, the relative risk of asthma incidence increased by 2.24% (95% confidence interval (95% CI) 0.93–5.38) for per 10 μg/m3 increase in concentrations of PM2.5 every 2 years from 2007 to 2014 in 117 populated areas (Requia et al. 2017). In the USA, pediatric asthma incidence would increase by 0.9% with each interquartile range (IQR) (6.7 μg/m3) of PM2.5 of the previous day in an agricultural community in Washington State; meanwhile, the forced expiratory volume during the first second of the forced breath (FEV1) for these affected children would drop by 0.9% (Loftus et al. 2015). The lag effect of PM2.5 on asthma incidence for the entire population in the USA occurred at lag 1 day, however, the lag was at 1 to 3 days for the African American population (Glad et al. 2012). The effect of (outdoor) atmospheric particulate matter (5.0 ppb per 10 μg/m3 increase in PM2.5) on asthma incidence is usually stronger than that of indoor particulate matter (3.3 ppb per 10 μg/m3 increase in PM2.5) (Koenig et al. 2005). In particular, traffic-related PM2.5 emission is a major concern, which was found to be positively correlated with asthma exacerbation for children in the USA (Spira-Cohen et al. 2009). When children are exposed to PM2.5 that is direct emission from traffic, the cases of asthma symptoms would keep increasing until reaching a peak at lag 3 days (Gent et al. 2009). For every 5 μg/m3 increase in the concentrations of PM2.5, the odds ratio (OR) of wheezing and shortness of breath would increase by 1.16 (95% CI 1.11–1.21) and 1.61 (95% CI 1.45–1.78), respectively (Dany et al. 2017).

In China, the impact of PM2.5 on asthma incidence has become a hot topic in recent years. From 1997 to 2002, PM2.5 was closely related to pediatric asthma admission rates for the under-18-year-olds in Hong Kong; when the concentrations of PM2.5 increased by 20.6 μg/m3, asthma-related admission rate for the under-18-year-olds would increase by 3.24% (95% CI 0.93–5.60) (Lee et al. 2006). In Shanghai, childhood asthma prevalence has significantly increased (Shanghai Medical Association Pediatric Professional Committee Respiratory Group 2014), which is closely related to PM2.5 (Pan and Kuai 2017). From 2007 to 2010, childhood asthma prevalence for the 6–14-year-olds increased from 5.92 to 10% in Shanghai (Yuan et al. 2007). For every 10 μg/m3 increase in PM2.5, the number of pediatric asthma patients for the 3-to-14-year-olds would increase by 0.78% (95% CI 0.16–1.4) in Shanghai (Mo et al. 2014). However, a comprehensive study on the effect of PM2.5 on childhood asthma prevalence for different age groups on different timescales in Shanghai has not yet been reported.

In this study, we assumed that PM2.5 has a hysteresis effect on childhood asthma prevalence in Shanghai, China. Based on the daily data (including hospital visit records, weather data, and the air quality data) from January 1, 2009, to December 31, 2010, we analyzed the short-term effect of PM2.5 on pediatric asthma hospital visits in Shanghai, China. We divided the entire study group into different age-subgroups and ran the analysis for a total of five timescales, including annual, Spring (from March to May), Summer (from June to August), Autumn (from September to November), and Winter (from December to the next February). As for the hysteresis effect, we used both single- and multi-pollutant models to estimate the relative risk of PM2.5 on pediatric asthma hospital visits at lag of 0–6 days (i.e., from the current day to the previous 6 days). By providing a comprehensive report on the short-term effect of PM2.5 on childhood asthma prevalence in Shanghai, China, our goal is to assist local health and human service authorities to reduce asthma prevalence in children.

Data and methods

Study area and data

Shanghai (121° 48′ E, 31° 22′ N) is located in eastern China, and the total resident population was about 24.2 million in 2017. Shanghai has a subtropical humid maritime monsoon climate, with an annual average air temperature of 17.7 °C and humidity of 69.4%. Based on the data availability, our study focuses on the period of January 1, 2009, to December 31, 2010. We obtained the daily average concentrations of air pollutants (i.e., PM2.5, PM10, SO2, and NO2) from the Shanghai Environmental Protection Bureau. We collected the daily weather data from the Shanghai Meteorological Bureau, including average air temperature (°C), wind speed (m/s), air pressure (hPa), and relative humidity (%). We acquired daily records of asthma-related hospital visits from the Shanghai Medical Insurance Affairs Management Center, including date of visit, age, sex, and diagnosis code. These daily medical records cover all the patients that have social medical insurance. It is required by the law for every resident to have social medical insurance, whether as a primary or secondary health insurance. Therefore, the acquired daily hospital visit records represent the entire resident population in Shanghai, China. In this study, we divided the entire study group into three age-subgroups, including the under-2-year-olds, the 3-to-5-year-olds, and the 6-to-18-year-olds.

Methods

We used a generalized additive model (GAM, Eq. (1)) (Samet et al. 2000) in the time series analysis to evaluate the short-term effect of PM2.5 on pediatric asthma hospital visits in Shanghai, China. We chose the relative risk (RR) as the index, which is the ratio of morbidity or mortality in the exposed and non-exposed groups. RR is an indicator of the extent to which the exposed research object is susceptible to a certain disease (asthma in this study). We presented the RR values as well as the 95% CI for per IQR increase in PM2.5 (Ma et al. 2018; Wang et al. 2012; Kan et al. 2008).

$$ \mathrm{Log}\left(E\left({Y}_i\right)\right)=\alpha +{\beta X}_i+\mathrm{DOW}+\mathrm{holiday}+s\left({\mathrm{PM}}_{10}, df\right)+s\left(\mathrm{time}, df\right)+s\left(N, df\right)+s\left(T, df\right)+s\left(P, df\right)+s\left(H, df\right)+s\left(W, df\right) $$
(1)

where Yi is the number of respiratory disease emergency department visits on day i, E(Yi) is the expected emergency department visits on day i; α is the intercept; β is the coefficient of exposure response; Xi is the explanatory variable that produces a linear effect; DOW is day of the week; holiday includes all statutory holidays; s is a nonparametric smoothing function; df is degree of freedom; PM10 indicates the average concentrations of PM10 at a df of 4; time is the calendar date variable, ranging from 1 to 730, with a df of 3; N is the child population from 2009 to 2010 in the study area, with a df of 3; T is daily average air temperature, with a df of 4; P is daily average air pressure, with a df of 3; H is daily relative humidity, with a df of 4; and W is daily average wind speed, with a df of 5.

In order to achieve an optimal fitting in the model, we used the Akaike information criterion (AIC) (Akaike 1975) to detect and adjust the df of variable smoothing terms (Guo et al. 2011). The smaller the AIC value is, the better fitting of the model is. In order to acquire the smallest AIC value in the model, we adjusted and screened the df of each variable in Eq. (1) to select the optimal combinations. Then we derived the exposure response coefficient in the optimal model to compute RR values (as well as the 95% CI) for per IQR increase in PM2.5 (35.63 μg/m3) using Eqs. (2) and (3).

$$ RR=\exp \left(\beta \times IQR\right) $$
(2)
$$ RR\left(95\%\mathrm{CI}\right)=\exp \left[\left(\beta \pm 1.96\mathrm{SE}\right)\times \mathrm{IQR}\right]\Big] $$
(3)

RR indicates the effect of the per unit change in the concentrations of air pollutants (per IQR) on the number of respiratory disease emergency department visits. If RR is greater than 1, it indicates a positive effect; if RR is smaller than 1, it indicates a negative effect. In addition, we analyzed the lag effects of air pollutants on the number of respiratory disease emergency department visits on the current day (lag 0) and the previous 1–6 days (lag 1–6 days). We selected the lag days that show the highest RR in the core model for further analysis. We used both single- and multi-pollutant models in the analysis. Single-pollutant model (Eq. (1)) has PM2.5 as the only air pollutant, and multi-pollutant model (Eq. (4)) also has SO2, and NO2 as the other air pollutants. In the multi-pollutant model, we added the other air pollutants one at a time to run the analysis.

$$ \mathrm{Log}\left(E\left({Y}_i\right)\right)=\alpha +{\beta X}_i+{\gamma Z}_i+\mathrm{DOW}+\mathrm{holiday}+s\left(\mathrm{time}, df\right)+s\left(N, df\right)+s\left(T, df\right)+s\left(P, df\right)+s\left(H, df\right)+s\left(W, df\right) $$
(4)

where Zi is the concentrations of the added air pollutants, and other parameters are the same as in Eq. (1).

We used the MGCV package in R3.5.0 to adjust the single- and multi-pollutant models. The statistical significance level of RR is at p < 0.05.

Results

From 2009 to 2010, daily average concentrations of PM2.5, PM10, and SO2 were 44.04 μg/m3, 76.68 μg/m3, and 31.20 μg/m3, respectively, in Shanghai, China. Daily average pediatric asthma hospital visits were 169, including 79 for the under-2-year-olds, 61 for the 3-to-5-year-olds, and 29 for the 6-to-18-year-olds (Table 1). For the entire study group, the number of pediatric asthma hospital visits was positively correlated with the concentrations of air pollutants at a statistically significant level of p < 0.01. Among the three types of air pollutants, the highest correlation coefficient was between PM10 and the number of pediatric asthma hospital visits for the entire study group. In particular, the under-2-year-olds subgroup showed a relatively high positive correlation coefficient (p < 0.01) between the concentrations of PM2.5 and the number of pediatric asthma hospital visits. For the entire study group, the correlation coefficients between the number of pediatric asthma hospital visits and the four weather elements were all statistically significant at p < 0.01. For the four weather elements, the number of pediatric asthma hospital visits for the entire study group was negatively correlated with wind speed, relative humidity, and average air temperature, but positively correlated with air pressure. Among the three age-subgroups, the number of pediatric asthma hospital visits was most closely related to air pressure and average air temperature for the under-2-year-olds, to relative humidity for the 3-to-5-year-olds, and to wind speed for the 6-to-18-year-olds (Table 2).

Table 1 Statistics of daily average weather elements (including air temperature T, relative humidity RH, wind speed W, and air pressure P), concentrations of air pollutants (including SO2, PM2.5, and PM10), and pediatric asthma hospital visits (for the entire study group three age-subgroups of under-2-year-olds, 3-to-5-year-olds, and 6-to-18-year-olds) data from 2009 to 2010 in Shanghai, China. “x̅” indicates the mean value, “SD” stands for the standard deviation, “Min” is the minimum value, “Max” is the maximum value, “Med” is the median, “P25” is the 25 percentile, “P50” is the 50 percentile, and “P75” is the 75 percentile
Full size table
Table 2 Correlation coefficients between the number of pediatric asthma hospital visits (for all study objects and the three age-subgroups) and the concentrations of air pollutants (i.e., PM2.5, PM10, and SO2), as well as weather elements (i.e., wind speed W, relative humidity RH, air pressure P, and average air temperature Tavg) from 2009 to 2010 in Shanghai, China
Full size table

From 2009 to 2010 in Shanghai, China, at lag 4 days, annual pediatric asthma hospital visits for the entire study group would rapidly increase as the concentrations of PM2.5 were increasing. When the concentrations of PM2.5 were increased to 120 μg/m3, annual pediatric asthma hospital visits for the entire study group would reach a peak, then began to fall (Fig. 1). From lag 0 to lag 6 days (current day to the previous 6 days), the highest RR (1.060) of PM2.5 on annual pediatric asthma hospital visits occurred at lag 4 days for the entire study group. Among the three age-subgroups, the highest RR of PM2.5 on annual pediatric asthma hospital visits occurred at lag 5 days for the under-2-year-olds (with a RR of 1.061), at lag 4 days for the 3-to-5-year-olds (with a RR of 1.071), and at lag 2 days for the 6-to-18-year-olds (with a RR of 1.052). Overall, PM2.5 showed a relatively stronger short-term effect on annual pediatric asthma hospital visits for the 3-to-5-year-olds as compared with the other two age-subgroups (Fig. 2).

Fig. 1
figure 1

Exposure-response curve for concentrations of PM2.5 and logarithm of RR in annual pediatric asthma hospital visits for the entire study group at lag 4 days from 2009 to 2010 in Shanghai, China

Full size image
Fig. 2
figure 2

RRs of PM2.5 on annual pediatric asthma hospital visits for the entire study group as well as the three age-subgroups (under-2-year-olds, 3-to-5-year-olds, and 6-to-18-year-olds) at lag 0 (current day) and lag 1–6 days (previous 1–6 days) from 2009 to 2010 in Shanghai, China. RR values are statistically significant at p < 0.05 with the exceptions of that for the entire study group at lag 1 day and that for the 6-to-18-year-olds at lag 0. The “central dot” represents the mean of RR, and the vertical line extending from the central dot indicates the 95% CI of RR. The same definitions apply in Fig. 3

Full size image

In Spring, the highest RR (1.041) of PM2.5 on seasonal pediatric asthma hospital visits occurred at lag 0 for the entire study group. As for the three age-subgroups, the highest RR (1.045, 1.039, and 1.025 for the under-2-year-olds, 3-to-5-year-olds, and 6-to-18-year-olds, respectively) of PM2.5 on seasonal pediatric asthma hospital visits all occurred at lag 0 (Fig. 3(a)). In Summer, the highest RR of PM2.5 on seasonal pediatric asthma hospital visits occurred at lag 0 for the entire study group (with a RR of 1.192) as well as the three age-subgroups (with a RR of 1.216, 1.267, and 1.054 for the under-2-year-olds, 3-to-5-year-olds, and 6-to-18-year-olds, respectively) (Fig. 3(b)). In Autumn, the highest RR of PM2.5 on seasonal pediatric asthma hospital visits occurred at lag 3 days for the entire study group (1.073) and the under-2-year-olds (1.112). Meanwhile, the highest RR of PM2.5 on seasonal pediatric asthma hospital visits occurred at lag 0 for the 3-to-5-year-olds (1.064) and the 6-to-18-year-olds (1.081) (Fig. 3(c)). In Winter, the highest RR of PM2.5 on seasonal pediatric asthma hospital visits occurred at lag 4 days for the entire study group (1.087) and the 6-to-18-year-olds (1.089). Meanwhile, the highest RR of PM2.5 on seasonal pediatric asthma hospital visits occurred at lag 0 for the under-2-year-olds (1.089) and at lag 5 days for the 3-to-5-year-olds (1.132) (Fig. 3(d)). In general, RR of PM2.5 on seasonal pediatric asthma hospital visits was relatively high in Summer for the entire study group as well as the under-2-year-olds and 3-to-5-year-olds. As for the 6-to-18-year-olds, RR of PM2.5 on seasonal pediatric asthma hospital visits was relatively high in Winter. In Spring and Autumn, the under-2-year-olds subgroup was more susceptible to asthma caused by PM2.5 than the other two subgroups. By contrast, in Summer and Winter, the 3-to-5-year-olds subgroup was more susceptible to asthma caused by PM2.5 than the other two subgroups.

Fig. 3
figure 3

RR of PM2.5 on seasonal (Spring, Summer, Autumn, and Winter) pediatric asthma hospital visits for the entire study group as well as the three age-subgroups (under-2-year-olds, 3-to-5-year-olds, and 6-to-18-year-olds) at lag 0 (current day) and lag 1–6 days (previous 1–6 days) from 2009 to 2010 in Shanghai, China. “a” on the error bar means that RR values are statistically significant at p < 0.05

Full size image

When adding SO2 or NO2 or both SO2 and NO2 in the analysis model, RR of PM2.5 on annual pediatric asthma hospital visits showed little change for the entire study group, the under-2-year-olds, and the 3-to-5-year-olds but a gradual increase for the 6-to-18-year-olds. For the entire study group and the under-2-year-olds, RR of PM2.5 on seasonal pediatric asthma hospital visits in Spring barely changed when adding SO2 in the GAM but decreased when adding NO2 or both SO2 and NO2. For the 3-to-5-year-olds, RR of PM2.5 on seasonal pediatric asthma hospital visits in Spring showed similar changes as the annual metric when adding SO2 or NO2 or both SO2 and NO2 in the GAM. As for the 6-to-18-year-olds, RR of PM2.5 on seasonal pediatric asthma hospital visits in Spring was not statistically significant (p < 0.05) when adding SO2 or NO2 or both SO2 and NO2 in the GAM. In Summer, RR of PM2.5 on seasonal pediatric asthma hospital visits generally increased when adding NO2 in the GAM but decreased when adding SO2 or both SO2 and NO2, with the exception of the 6-to-18-year-olds. In Autumn and Winter, for the entire study group and the under-2-year-olds, RR of PM2.5 on seasonal pediatric asthma hospital visits decreased when adding SO2 or NO2 or both SO2 and NO2 in the GAM. For the 3-to-5-year-olds, RR of PM2.5 on seasonal pediatric asthma hospital visits increased in Autumn but decreased in Winter when adding SO2 or NO2 or both SO2 and NO2 in the GAM. As for the 6-to-18-year-olds, RR of PM2.5 on seasonal pediatric asthma hospital visits only decreased when adding SO2 in Autumn or when adding NO2 in Winter (Table 3).

Table 3 RR of PM2.5 on annual and seasonal (i.e., Spring, Summer, Autumn, and Winter) pediatric asthma hospital visits at different lag days for the entire study group as well as the three age-subgroups (i.e., under-2-year-olds, 3-to-5-year-olds, and 6-to-18-year-olds) in single- (PM2.5) and multi-pollutant (PM2.5 and SO2, PM2.5 and NO2, or PM2.5, SO2, and NO2) models from 2009 to 2010 in Shanghai, China. RR values that are statistically significant at p < 0.05 are set in boldface. The selected lag days represent the highest RR values among lags of 0 (current day) to 6 days (previous 1–6 days)
Full size table

Discussion

In this study, we found that the number of pediatric asthma hospital visits was positively correlated with average air pressure but negatively correlated with average wind speed, relative humidity, and air temperature. These results are consistent with that from Chan et al. (2018), Hu et al. (2015), and Jun et al. (2009). From 2009 to 2010 in Shanghai, China, the daily number of pediatric asthma hospital visits would increase with the increase in daily concentrations of PM2.5. From 2006 to 2007 in Australia, asthma-related emergency department visits would increase by 1.96% (95% CI 0.02–3.94%) for per IQR increase in area PM2.5 (8.6 μg/m3) (Haikerwal et al. 2016). From 2010 to 2014 in the state of Florida, USA, for per 10 μg/m3 increase in PM2.5, asthma-related emergency department visits would go up by 2% for the 5-to-12-year-olds and 0.6% for the 13-to-18-year-olds; when the concentrations of PM2.5 reached 64.66 μg/m3, RR of PM2.5 on childhood asthma incidence was up to 1.08 (95% CI 1.05–1.11) (Mehra 2017).

We observed some lag effects of PM2.5 on the number of pediatric asthma hospital visits from 2009 to 2010 in Shanghai, China. Annually, PM2.5 showed the strongest effect on the number of pediatric asthma hospital visits at a lag of 4 days for the under-2-year-olds, a lag of 5 days for the 3-to-5-year-olds, and a lag of 2 days for the 6-to-18-year-olds. In Spring and Summer, the highest RR of PM2.5 on the number of pediatric asthma hospital visits occurred unanimously at a lag of 0 days. In Autumn, the highest RR of PM2.5 on the number of pediatric asthma hospital visits occurred at a lag of 3 days for the under-2-year-olds, but a lag of 0 day for both 3-to-5-year-olds and 6-to-18-year-olds. In Winter, the highest RR of PM2.5 on the number of pediatric asthma hospital visits occurred at a lag of 0 day for the under-2-year-olds, a lag of 5 days for the 3-to-5-year-olds, and a lag of 4 days for the 6-to-18-year-olds. The lag effects of PM2.5 on pediatric asthma hospital visits could be related to seasonal allergens such as pollen. Similar lag effects were also reported in other locations. For example, in Changhua (central county) and Yunlin (western county) in Taiwan, RR of PM2.5 on the number of pediatric asthma hospital visits for the under-15-year-olds was higher on lag 2 days than on the current day (the concentrations of PM2.5 increased from 36.17 to 81.26 μg/m3) (Chien et al. 2018). In the state of Massachusetts, USA, a case-crossover study on the effect of early life exposure to PM2.5 on risk of acute asthma clinical encounters among children showed that the OR value of PM2.5 exposure on children asthma clinical encounters was close to 1.0 at lags of 0, 1, and 2 days (Khalili et al. 2018). In particular, for per IQR (5.9 μg/m3) increase in PM2.5, the clinical risk of asthma for children with low birth weight (LBW < 2500 g) would increase at a lag of 1 day, with an OR value of 1.08 (95% CI 1.01–1.15) (Khalili et al. 2018). In Tacoma, WA, USA, RR of PM2.5 on asthma-related emergency department visits was 1.04 (95% CI 1.01–1.07) at lag 2 days and 1.03 (95% CI 1.0–1.06) at lag 3 days (Mar et al. 2010).

We ran seasonal analysis for the three different age-subgroups and found that RR of PM2.5 on pediatric asthma hospital visits was relatively high in Spring and Autumn for the under-2-year-olds, in Summer and Winter for the 3-to-5-year-olds, and in Winter for the 6-to-18-year-olds. A previous study reported that with per 10 μg/m3 increase in PM2.5, the number of asthma emergency visits would increase by more in warm season (with a rate of 3.7%, 95% CI 0.5–6.9%) than in cold season (with a rate of 2.6%, 95% CI 0.7–4.6%) (Fan et al. 2016). Nevertheless, another study pointed out that the effect of PM2.5 on asthma was stronger in cold season than in warm season (Jung et al. 2017), mostly due to the fact children were exposed to higher concentrations of PM2.5 in Winter than in Summer (Zhang et al. 2015). In the Northern New Jersey-New York City metro area of the USA, asthma incidence was closely correlated with air pollution in relatively warm months based on a 15-year summary of airborne allergens (Port et al. 2006). During the warm season, the amount of pollen (e.g., trees, grass, weeds, and ragweed) reaches the peak, which is positively correlated to the incidence of respiratory diseases (CarracedoMartinez et al. 2010). Therefore, some studies only focus on the effects of atmospheric particulate matter on asthma during warm season (Gleason et al. 2014). However, in some regions of the world, the concentrations of PM2.5 reach the peak in Winter mostly due to direct emissions produced by burning fuel (especially coal) for heat (Song et al. 2015).

In different seasons, the maximum RR value appears in different lag days. According to the research in New York State from 2000 to 2005, PM2.5 has the greatest impact on respiratory hospitalizations at lag 4, 1, 4, 0, and 0 days around the year, Spring, Summer, Fall, and Winter, respectively. And RR values in Summer are much higher than other seasons, which is consistent with the results of this study (Jones et al. 2015). In Taipei and Kaohsiung, the maximum OR values of PM2.5 on childhood asthma hospitalization are found at lag 2, 3, 1, and 3 days for Spring, Summer, Autumn, and Winter, respectively, with the RR values of 0.998 (95% CI 0.823–1.210), 1.085 (95% CI 0.739–1.593), 0.966 (95% CI 0.787–1.186), and 1.109 (95% CI 0.885–1.388) (Kuo et al. 2018).

In the multi-pollutant models, RR of PM2.5 on pediatric asthma hospital visits was generally higher than that in the single-pollutant model from 2009 to 2010 in Shanghai, China. However, the amplitude of the increase in RR varied to different extents for the three age-subgroups during the four seasons. In Beijing, China, both the single- and multi-pollutant models showed the highest increase percentage (IP) of RR for the under-15-year-olds (Ma et al. 2018). In Atlanta, GA, USA, the multi-pollutant model usually showed smaller estimates on the number of pediatric asthma emergency department visits than the single-pollutant model (Winquist et al. 2014). In addition, the warm-season joint impact of models without nonlinear effects is usually stronger than that for the cold season (Winquist et al. 2014). O3 is a pollutant of primary concern in Hong Kong, China; the multi-pollutant model showed stronger effects of O3 on asthma-related hospitalization rate than that of NO2 and SO2 (Ko et al. 2010). In St. John’s, Canada, the multi-pollutant model showed that respiratory disease–related emergency department visits were caused more by O3 (with a rate of 3.9%, 95% CI 0.8–7.2%) than by SO2 (with a rate of 3.7%, 95% CI 1.5–6.0%) (David et al. 2000).

Inevitably, there are some limitations of the results in this study. First, we used the average daily concentrations of air pollutant that are derived from a variety of monitoring stations without eliminating the inherent measurement bias (Chen et al. 2012). Second, we assumed that all individuals in the study group were equally exposed to the same air pollutants at all times, which might not be consistent with reality. Third, due to the nature of the analysis model, multicollinearity between air pollutants and weather elements might cause some bias in the results. Therefore, we suggest that future studies should (a) homogenize air quality data from multiple monitoring stations; (b) acquire detailed information about the actual exposure in air pollutants for the study group; and (c) minimize the bias caused by multicollinearity in the analysis model.

Conclusions

Our study showed that PM2.5 has a significant short-term effect on the number of pediatric asthma hospital visits in Shanghai, China, from 2009 to 2010. We detected different seasonal lag effects for the entire study group as well as the three age-subgroups. For the entire study group, the highest RR value occurred at a lag of 4 days. Within the study group, the highest RR value occurred at a lag of 5 days for the under-2-year-olds, a lag of 4 days for the 3-to-5-year-olds, and a lag of 2 days for the 6–18-year-olds. Overall, the under-2-year-olds subgroup was the most affected by PM2.5 in pediatric asthma hospital visits, especially in Summer. When introducing other air pollutants (including SO2 and NO2) in the analysis model, the short-term effect of PM2.5 on pediatric asthma hospital visits is unchanging. The results in this study could be used by the local health and human services authority to help lowering childhood asthma prevalence in Shanghai.