Smog in Beijing reduces life expectancy by 15 years

The effects of air pollution on human health have recently attracted increasing concern in China, in part due to the increasing number of days with very high levels of air pollution.1 2 In most Chinese cities, concentrations of PM2.5 (particulate matter with aerodynamic diameter <2.5 µm) are still far above the level recommended by the World Health Organization’s guidelines on air quality (interim target 2 level) of 10 μg/m3 (annual average) and 25 μg/m3 (24 h average).3 For example, in 2004-08, mean daily PM2.5 concentration was 105 μg/m3 in Beijing. Beijing is experiencing increasing population density, car use, and expanded construction. It is surrounded by a heavy industrial region, which provides additional sources of air pollutants carried via air flow. Consequently, the ambient pollutant mixture is complex, with the potential for combined toxic effects from many constituents. Reliable estimation of the burden of air pollution on health is essential to support evidence based government policy in this important public health area.4 5 Previous studies have examined the effects of air pollution on daily excess deaths or mortality risks using time series methods.6 7 Those studies focused on the number of deaths, but did not account for age at death, apart from broad age stratification. We argue that using the number of years of life lost (YLL) provides a complementary indicator to that of excess deaths, because it takes into account the life expectancy at death.8 Methods Data collection YLL data This study was conducted in eight districts within the urban area of Beijing. Mortality data on non-accidental causes were obtained from the death classification system at the Beijing Public Security Bureau, between 1 January 2004 and 31 December 2008. These data comprised date of death, sex, and age. All deaths were registered residents of urban areas of Beijing city. Chinese national life tables were obtained from WHO for the years 2000 and 2009 (web table S1).9 Life expectancies for 2004-08 were averaged from the years 2000 and 2009, as data were unavailable for 2004-08. We calculated YLL for each death by matching age and sex to the life tables. Daily YLL were calculated by summing the YLL for all deaths on that day. We stratified the sums by sex and age group (≤65 and >65 years). The web appendix shows an example of this calculation.

Data on air pollution and weather conditions
PM2.5 was monitored at the main campus of Peking University, located in the urban centre.10 11 Details of the monitoring station are described elsewhere.12 The monitoring station is a few hundred metres away from major roads and about 20 m above ground level. The campus is primarily residential and commercial without industrial sources or agricultural activities. Spatial variability of PM2.5 mass and chemical composition is low across the urban area of Beijing (difference <10%). Additionally, average particle number and size distributions at this monitoring site and another regional site (50 km south of Peking University) were similar in the summer.13 Therefore, the monitoring site provided reliable estimates of pollutant levels for the urban area.12 We computed the daily average concentration from 24 h values. We obtained daily data on particulate matter less than 10 μm in aerodynamic diameter (PM10), sulphur dioxide (SO2), and nitrogen dioxide (NO2) from the Beijing municipal environmental monitoring centre, which had eight fixed monitoring sites distributed in different part of the urban area.14 For each monitoring site, we calculated 24 h mean concentrations from non-missing data if at least 18 of 24 hourly concentrations of PM10, SO2, and NO2 were available.15 The daily mean concentrations of each air pollutant were calculated by averaging daily data over all monitoring sites. We obtained meteorological data on daily mean temperature, relative humidity, and air pressure from the China meteorological system’s data sharing service. The monitoring station is located at Daxing district in southeast Beijing. Data analysis The daily YLL follows a normal distribution (web fig S1). We used daily YLL as dependent variable in a five year time series model, to examine its association with air pollutants. To control for long term trend and seasonality, we used a natural cubic spline with seven degrees of freedom per year for time. The day of week was controlled for as a categorical variable. To most effectively control for the potential effects of weather conditions on mortality, we used distributed lag non-linear models for temperature, relative humidity, and air pressure. A natural cubic spline with five degrees of freedom was used for temperature, relative humidity, and air pressure, and a natural cubic spline with four degrees of freedom for lag days (≤27 days). We used previous experience of similar analyses in selecting the above parameters.16 We validated the model fit by checking the residuals to ensure that seasonality and autocorrelation had been successfully removed. Studies have shown that models of single day lags might underestimate the effect of air pollution on mortality,17 thus we used a moving average concentration over two days (lag 0-1) for our main analyses.18 We also examined the associations using a single day lag (from lag 0 to lag 3). For each pollutant, we fitted models of single pollutants and multiple pollutants models to assess the stability of the associations. In addition, we stratified analyses by sex and age (≤65 years and >65 years).

To examine the linearity of the associations between air pollutants and YLL, we used a natural cubic spline with four degrees of freedom for each air pollutant (lag 0-1 day) in single pollutant models. If the relations tended to be linear, we used a linear function; if not, we used a non-linear function with a natural cubic spline for air pollutants.

To compare the standard analysis of mortality and the analysis of YLL, we estimated the percentage change in daily mortality associated with changes in air pollutants. We used the same independent variables as the YLL model, but with daily count of deaths as the dependent variable following a Poisson distribution.

To check the adequacy of all models, we used an autocorrelation function to examine if the residuals were independent over time. R software was used to conduct statistical analyses.19 The dlnm package was used to perform distributed lag non-linear models.20 21

The mean concentrations of daily PM2.5, PM10, SO2, and NO2 were 105.1 μg/m3, 144.6 μg/m3, 48.6 μg/m3, and 64.2 μg/m3, respectively (table 1⇓). Generally, PM2.5 and PM10 had positive correlations with all other pollutants and weather conditions, while mean temperature was negatively correlated with SO2 and NO2 (table 2⇓).

The following two tabs change content below.