A Study On The Associations Between Ambient Air Pollution, Meteorological Factors And Daily Mortality In Guangzhou, China

BackgroundAmbient air pollution is a growing public health concern worldwide. It significantly influences the environment and human health. As other developing countries, China is suffering from ambient air pollution. In recent years, dust-haze espisodes in China has been increasing in the frequency and intensity. In this context, the public are aware of the adverse impact of ambient air pollution. Particularly, Under the Dome produced by Chai Jing further raised the public awareness of the detrimental effects of air pollution. At present, the Chinese government delivers the real-time information of air quality of 365 major cities.The level of air pollution has apparent spatial and temporal patterns. Understanding its spatial and temporal patterns would provide important information for air pollution prediction and taking effective measures to control the air pollution. Air pollution has a relation with human activities and natural environment; particularly, meteorological factors affect the diffusion and chemical action of air pollutants. Although there is increasing evidence of temporal correlations between air pollutant concentrations and meteorological factors, previous findings are inconsistent or even conflicting. Besides, statistical methods used in previous studies are commonly inappropriate. There is a need to use appropriate statistical methods so as to reveal the true relationships. The Air Pollution Index (API) is commonly used for reporting the level of ambient air pollution, including particulate matter with an aerodynamic diameter of 10μm or less (PM10), sulfur dioxide (SO2) and nitrogen dioxide (NO2). It is a more simple and generalized way to provide timely information about air quality to the public than the concentrations of specific air pollutants. In this case, revealing the spatial and temporal variations of API and further exploring the relationships between API and meteorological factors are more meaningful from the aspect of availability of air quality information and public communication. Likewise, despite intense studies assessing the shot-term associations of specific air pollutant concentrations with health outcomes, seldom did previous studies examine whether the API can be used to communicate health risks of ambient air pollution. Due to the limitation of methods used to examine the short-term associations between ambient air pollution and health outcomes, previous studies cannot thoroughly show the distributed lag effects of ambient air pollution. Thus, these studies cannot give precise cumulative effect estimate of ambient air pollution. Newly developed method should be used to explore the shape of air pollution effect. Ambient temperature is also associated with human health. Particularly, the extreme temperature has a significant detrimental influence on human health. A few studies indicated that ambient air pollution modified health effects of temperature. However, previous studies commonly examined the log linear interaction effects between ambient air pollution and temperature. Numerous studies showed that the temperature effects were log non-linear. Therefore, we speculate that the interaction effects may also be log non-linear. This is issue need to be further confirmed.Methods2.1 Study location and data sourcesThis study was conducted in Guangzhou. The Guangzhou Bureau of Environmental Protection provided daily data of air pollutants for each monitoring station from January 1,2001 to December 31,2011 (http://www.gzepb.gov.cn/). We selected seven fixed-site air monitoring stations, namely Guangya High School (GYHS), Lu Hu (LH), Tianhe Vocational High School Affiliated Experimental Kindergarten (THVK), NO.5 Middle School (NFMS), City Monitoring Station (CMS), NO.86 Middle School (ESMS) and Guangdong Business College (GDBC). Because there were no data from the last two stations before 2003, we used data collected at other five stations to examine the spatial and temporal patterns of air quality and the relationships between API and meteorological factors. Daily meteorological data were obtained from the China Meteorological Data Sharing Service System, including daily average/minimum/maximum temperature (℃), average/minimum relative humidity (%), precipitation (mm), maximum/extreme wind speed (m/s), average/minimum/maximum atmospheric pressure (hpa), and sunshine hours (h). We obtained individual data for all registered deaths at six urban central districts in Guangzhou between 1 January 2003 and 31 December 2011 from Guangzhou Center for Disease Control and Prevention. All-causes, non-accidental, cardiovascular and respiratory mortality were examined. Analyses were also stratified by age group, gender, educational attainment and occupation.2.2 Statistical methodsSpearman correlation coefficients of de-trended APIs in 5 monitoring stations (i.e., GYHS, LH, THVK, NFMS and CMS) were calculated to assess the short-term relationships between five APIs. Then, the Seasonal-Trend Decomposition Procedure Based on Loess (STL) was used to analyze the temporal variation of API. Next, wavelet analyses were employed to explore the time scale-dependent relationships between API and meteorological factors. Subsequently, weighted API (APIw) was constructed by weighting index values for different ambient air pollutants (i.e., PM10, SO2 and NO2). Then, a Distributed Lag Non-linear Model (DLNM) was used to assess the associations of API, index values for three pollutants, APIw with mortality. The present study evaluated whether API can be used to communicate health risks by comparing the associations of API, index values for three pollutants, APIw with non-accidental mortality. In addition, a model with the interaction of type of "Main Pollutant" and API was built. This study used F test to examine whether the associations of API with mortality were the same for different "Main Pollutant" Further, stratified analyses were used to identify subpopulations vulnerable to ambient air pollution effects. Non-linear interaction terms of concentration of PM10 and mean temperature were incorporated into a Generalized Linear Model (GLM). Then, F test was again employed to examine whether the interaction effects were statistically significant. Temperature effects were calculated, given that PMio equaled 3, the 25th (48.6μg/m3),50th (71.4μg/m3) and 75th (102.6μg/m3) percentile of o. Then the distributed lag effects of temperature under different levels of PM10 (PM10 was categorized into two levels (i.e., high and low) using a cutoff value of 50μg/m3, the standard of WHO Air Quality Guidelines) were examined. All analyses were performed using R 3.1.2 and MATLAB 7.6.0. The results of statistical tests were considered statistically significant if the two-tailed P value was less than 0.05.ResultsThe days in which air quality were good accounted for 24.6% of the whole study period. Among five monitoring stations, API was higher in GYHS and NFMS, but lower in THVK and LH. APIs were highly correlated among five monitoring stations, with the Spearman rank coefficients ranging from 0.84 to 0.89 (P<0.001), but there were substantial temporal variations in five APIs. During 2001-2011, API increased before 2005, peaking in 2004, and decreased since 2005. The API was relatively high in winter and low in summer.Timescale-dependent relationships were found between API and a variety of meteorological factors. Temperature, relative humidity, precipitation and wind speed were negatively correlated with API, while diurnal temperature range and atmospheric pressure were positively correlated with API in the annual cycle. Basically, at shorter scales relative humidity, precipitation and wind speed were negatively correlated with API, while temperature, diurnal temperature range and atmospheric pressure were positively correlated with API. The relationship between sunshine hours and API was not clear.The Spearman rank coefficients of API and index values for PM10, SO2, NO2 were 0.97,0.74 and 0.87, respectively. We observed significant negative associations between API and non-accidental mortality with the highest effect occurring on the current day. The harvesting effect appeared after 2days. An increase of 10 in API was associated with a 0.88%(95%confidence interval (CI):0.50,1.27%) increase of non-accidental mortality at lag 0-2 days. The corresponding estimate of cumulative effects at lag 0-15 days was 1.03%(95%CI:0.26,1.82%). The associations of API with non-accidental mortality were similar with those of pollutant-specific index value. Specifically, an increase of 10 in index value for PM10、SO2、NO2 was associated with a 0.82%(95%CI:-0.01,1.65%)、1.09%(95%CI:0.11,2.07%) and 1.17%(95%CI:0.42,1.93%) increase of non-accidental mortality at lag 0-15 days. Compared to the association of API with non-accidental mortality at lag 0-15 days, the change of effect estimates for pollutant-specific indices was within 20%. Even though the associations of APIw with non-accidental mortality decreased with weights for "Main Pollutant", the associations of APIw with non-accidental mortality were similar with those of API. Compared to the association of API with non-accidental mortality at lag 0-15 days, the change of effect estimates for APIw was also within 20%. The interaction of type of "Main Pollutant" and API was not statistically significant (F=1.651, P=0.117). Greater associations of API with mortality were observed in the elderly (aged 65 years or above) and females. Residents with low education attainment (illiterate or primary school) were more susceptible to ambient air pollution effects compared to residents with high education attainment (middle school or above). The harvesting effect appeared after 2 or 3 days. An increase of 10 in API was associated with a 1.54%(95%CI:0.67,2.41%),1.37% (95%CI:0.27,2.48%),1.41%(95%CI:0.43,2.41%) increase of non-accidental mortality among the elderly, females, residents with low education attainment at lag 0-15 days.There were non-linear relationships between mean temperature and mortality categories. Effects of mean temperature at high levels of PM10 were obviously greater than those for low PM10. Compared with other mortality categories, changes in temperature effects on respiratory mortality across PM10 levels were negligible. When the level of PM10 equals 0μg/m3,48.6μg/m3 (the 25th percentile of PM10),71.4μg/m3 (the 50th percentile of PM10) and 102.6μg/m3 (the 75th percentile of PM10), the overall increase in all-causes mortality risk at 0-20 days comparing 7.5℃ (the 1st percentile of temperature) and 13.3℃ (the 10th percentile of temperature) was 35.49%(95%CI:22.59,49.75%),40.69%(95%CI:28.97,53.47%),43.20%(95% CI:30.97,56.57%) and 46.70%(95%CI:32.66,62.22%), respectively. When the level of PM10 equals Opg/m3, the 25th,50th,75th percentiles, the overall increase in all-causes mortality risk at 0-20 days comparing 31.9℃ (the 99th percentile of temperature) and 29.9℃ (the 90th percentile of temperature) was 10.76%(95%CI: 0.33,22.28%),13.97%(95%CI:4.53,24.25%),15.51%(95%CI:6.21,25.63%) and 17.65%(95%CI:8.09,28.05%), respectively. That is, cold effects were stronger than hot effects for all mortality categories; with regard to the cause-specific difference, cold effects on cardiovascular mortality were greater than on all-causes and non-accidental mortality; both cold and hot effects increased with the quartiles of PMio.Men suffered more from cold-related mortality than women, with the gender difference enlarged with the quartiles of PMio. When the level of PMio equals the 25th,50th,75th percentiles, the overall increase in all-causes mortality risk among males at 0-20 days comparing 7.5℃ and 13.3℃ was 38.92%(95%CI:22.49, 57.56%),46.97%(95%CI:31.72,63.99%),50.91%(95%CI:34.88,68.85%) and 56.46%(95%CI:37.90,77.52%). The elderly were more vulnerable to cold and hot effects. When the level of PM10 equals 0μg/m3, the 25th,50th,75th percentiles, the overall increase in all-causes mortality risk among the elderly at 0-20 days comparing 7.5℃ and 13.3℃ was 41.38%(95%CI:26.59,57.91%),45.02%(95%CI:31.75, 59.64%),46.77%(95%CI:32.99,61.97%) and 49.18%(95%CI:33.48,66.72%), respectively. When the level of PM10 equals 0μig/m3, the 25th,50th,75th percentiles, the overall increase in all-causes mortality risk among the elderly at 0-20 days comparing 31.9℃ and 29.9℃ was 16.42%(95%CI:4.14,30.15%),19.82%(95%CI: 8.70,32.08%),21.45%(95%CI:10.48,33.52%) and 23.71%(95%CI:12.45, 36.11%), respectively.We identified statistically significant interaction effects between PMio and mean temperature on all-causes (F=3.028, P=0.006), non-accidental (F=3.177, P=0.005) and cardiovascular mortality (F=3.807, P=0.001), but not on respiratory mortality (F=0.745, P=0.614). Hot effects basically appeared acutely on highly polluted days, while effects were delayed for 1 day on lowly polluted days with harvesting. Cumulative effects at 0-20 days were higher on highly polluted days for non-accidental and cardiovascular mortality. But for respiratory mortality, the cumulative effects were almost the same on highly and lowly polluted days.Conclusions(1) During the study period of 2001-2011, air quality in Guangzhou was worst in 2004. A slight decrease was shown since 2005. However, air pollution there remains serious. There was an apparent seasonal pattern in air quality, with the peak of API in winter and spring. Weather conditions are important driving forces of intra-annual variation in API. There are significant coherent annual cycles between API and many meteorological parameters. The daily API is negatively correlated with daily temperature, relative humidity, precipitation and wind speed, but it is positively correlated with the daily DTR and atmospheric pressure.(2) This study confirmed that API can be used to communicate health risks of exposure to ambient air pollution. The elderly, females, residents with low educational attainment were vulnerable populations associated with the exposure to ambient air pollution. Subpopulations that are at high health risks related to ambient air pollution should be given more attention.(3) We identified the effect modification by PM10 in the non-linear temperature-mortality association. There is clear evidence that both cold and hot effects of mean temperature increased with the quartiles of PM10. The time courses of temperature effects differed on highly and lowly polluted days.(4) These findings guide how to build models to forecast air quality. In addition, our findings have important implication for how to alleviate the loss of public health due to ambient air pollution and extreme temperature.