Regional Differences in Acute Health Effects of PM2.5 Using Fractional Polynomial Models. -20 City Study in Japan

摘要

Background and Aims: Last decade, many time series studies have indicated that ambient fine particulate matter (PM2.5) exposure is associated with daily mortality. These results show the harmful effect of PM2.5 but their effect sizes are heterogeneous among countries and regions. Because it is well known that seasonal effects and weather conditions have great influence on the density, measurement and the effect of PM2.5, most studies use generalized additive model with spline function for adjusting these covariates. There are some studies that adjust for the regional differences using meta-analytic method or Bayesian hierarchical model. In epidemiologic studies, regional differences are often treated as random effects. But there is no study based on the mixed-effects model (generalized linear model including random and fixed effects). We apply the fractional polynomial models to our study data. Using fractional polynomial models, we can adjust many covariates like weather conditions and co-pollutants flexibly, and adjust for regional differences using random effects, simultaneously. Our purpose is to detect regional effect of PM2.5. Materials and Methods: Time series data contain daily counts of death (respiratory and cardiovascular causes of death, all cause death without accidental causes) and daily concentrations of air pollutants (PM2.5 using the TEOM, O3, NO2), weather conditions (temperature and relative humidity) in Japanese 20 cities from 2001 to 2007. We apply 1st and 2nd order fractional polynomial models, and choose one model with minimum deviance. We detect regional differences using interaction term between PM2.5 and region. We use Gibbs sampling (approximation algorithm) to solve likelihood function approximately. Results and Conclusions: There are no regional differences of the acute effect of PM2.5 between regions. But when we limited the early spring season (February to May), we detect regional differences between western and eastern side of Japan.