Part 2. Association of daily mortality with ambient air pollution, and effect modification by extremely high temperature in Wuhan, China
Zhengmin Qian,Qingci He,Hung-Mo Lin,Lingli Kong,Dunjin Zhou,Shengwen Liang,Zhichao Zhu,Duanping Liao,Wenshan Liu,Christy M Bentley,Jijun Dan,Beiwei Wang,Niannian Yang,Shuangqing Xu,Jie Gong,Hongming Wei,Huilin Sun,Zudian Qin,HEI Health Review Committee
Abstract:Fewer studies have been published on the association between daily mortality and ambient air pollution in Asia than in the United States and Europe. This study was undertaken in Wuhan, China, to investigate the acute effects of air pollution on mortality with an emphasis on particulate matter (PM*). There were three primary aims: (1) to examine the associations of daily mortality due to all natural causes and daily cause-specific mortality (cardiovascular [CVD], stroke, cardiac [CARD], respiratory [RD], cardiopulmonary [CP], and non-cardiopulmonary [non-CP] causes) with daily mean concentrations (microg/m3) of PM with an aerodynamic diameter--10 pm (PM10), sulfur dioxide (SO2), nitrogen dioxide (NO2), or ozone (O3); (2) to investigate the effect modification of extremely high temperature on the association between air pollution and daily mortality due to all natural causes and daily cause-specific mortality; and (3) to assess the uncertainty of effect estimates caused by the change in International Classification of Disease (ICD) coding of mortality data from Revision 9 (ICD-9) to Revision 10 (ICD-10) code. Wuhan is called an "oven city" in China because of its extremely hot summers (the average daily temperature in July is 37.2 degrees C and maximum daily temperature often exceeds 40 degrees C). Approximately 4.5 million residents live in the core city area of 201 km2, where air pollution levels are higher and ranges are wider than the levels in most cities studied in the published literature. We obtained daily mean levels of PM10, SO2, and NO2 concentrations from five fixed-site air monitoring stations operated by the Wuhan Environmental Monitoring Center (WEMC). O3 data were obtained from two stations, and 8-hour averages, from 10:00 to 18:00, were used. Daily mortality data were obtained from the Wuhan Centres for Disease Prevention and Control (WCDC) during the study period of July 1, 2000, to June 30, 2004. To achieve the first aim, we used a regression of the logarithm of daily counts of mortality due to all natural causes and cause-specific mortality on the daily mean concentrations of the four pollutants while controlling for weather, temporal factors, and other important covariates with generalized additive models (GAMs). We derived pollutant effect estimations for 0-day, 1-day, 2-day, 3-day, and 4-day lagged exposure levels, and the averages of 0-day and 1-day lags (lag 0-1 day) and of 0-day, 1-day, 2-day, and 3-day lags (lag 0-3 days) before the event of death. In addition, we used individual-level data (e.g., age and sex) to classify subgroups in stratified analyses. Furthermore, we explored the nonlinear shapes ("thresholds") of the exposure-response relations. To achieve the second aim, we tested the hypothesis that extremely high temperature modifies the associations between air pollution and daily mortality. We developed three corresponding weather indicators: "extremely hot," "extremely cold," and "normal temperatures." The estimates were obtained from the models for the main effects and for the pollutant-temperature interaction for each pollutant and each cause of mortality. To achieve the third aim, we conducted an additional analysis. We examined the concordance rates and kappa statistics between the ICD-9-coded mortality data and the ICD-10-coded mortality data for the year 2002. We also compared the magnitudes of the estimated effects resulting from the use of the two types of ICD-coded mortality data. In general, the largest pollutant effects were observed at lag 0-1 day. Therefore, for this report, we focused on the results obtained from the lag 0-1 models. We observed consistent associations between PM10 and mortality: every 10-microg/m3 increase in PM10 daily concentration at lag 0-1 day produced a statistically significant association with an increase in mortality due to all natural causes (0.43%; 95% confidence interval [CI], 0.24 to 0.62), CVD (0.57%; 95% CI, 0.31 to 0.84), stroke (0.57%; 95% CI, 0.25 to 0.88), CARD (0.49%; 95% CI, 0.04 to 0.94), RD (0.87%; 95% CI, 0.34 to 1.41), CP (0.52%; 95% CI, 0.27 to 0.77), and non-CP (0.30%; 95% CI, 0.05 to 0.54). In general, these effects were stronger in females than in males and were also stronger among the elderly (> or = 65 years) than among the young. The results of sensitivity testing over the range of exposures from 24.8 to 477.8 microg/m3 also suggest the appropriateness of assuming a linear relation between daily mortality and PM10. Among the gaseous pollutants, we also observed statistically significant associations of mortality with NO, and SO2, and that the estimated effects of these two pollutants were stronger than the PM10 effects. The patterns of NO2 and SO2 associations were similar to those of PM10 in terms of sex, age, and linearity. O3 was not associated with mortality. In the analysis of the effect modification of extremely high temperature on the association between air pollution and daily mortality, only the interaction of PM10 with temperature was statistically significant. Specifically, the interaction terms were statistically significant for mortality due to all natural (P = 0.014), CVD (P = 0.007), and CP (P = 0.014) causes. Across the three temperature groups, the strongest PM10 effects occurred mainly on days with extremely high temperatures for mortality due to all natural (2.20%; 95% CI, 0.74 to 3.68), CVD (3.28%; 95% CI, 1.24 to 5.37), and CP (3.02%; 95% CI, 1.03 to 5.04) causes. The weakest effects occurred at normal temperature days, with the effects on days with low temperatures in the middle. To assess the uncertainty of the effect estimates caused by the change from ICD-9-coded mortality data to ICD-10-coded mortality data, we compared the two sets of data and found high concordance rates (> 99.3%) and kappa statistics close to 1.0 (> 0.98). All effect estimates showed very little change. All statistically significant levels of the estimated effects remained unchanged. In conclusion, the findings for the aims from the current study are consistent with those in most previous studies of air pollution and mortality. The small differences between mortality effects for deaths coded using ICD-9 and ICD-10 show that the change in coding had a minimal impact on our study. Few published papers have reported synergistic effects of extremely high temperatures and air pollution on mortality, and further studies are needed. Establishing causal links between heat, PM10, and mortality will require further toxicologic and cohort studies.