A lag effect exists in the association between PM10 and mortality (Braga et al., 2001b and Schwartz, 2000). As part of the sensitivity analysis, we also examined the PM10 effects from different lag structures, including both single-day lag (from lag 0 to lag 5) and multiday lag (lag 01 to lag 05). The largest effects were observed at lag 03; thus, we focused on the results of the lag 03 model and developed a second model for the study. Additional details SJB3-019A reported in the Appendix.
Second, we investigated the potential interactive patterns of PM10 and mean temperature on daily mortality while adjusting for other covariates. We stratified the effect of PM10 on mortality by the percentile of the daily mean temperature for each city. This approach allowed us to examine the heterogeneity of the effects of PM10 across the temperature strata, and inversion provided a simple, quantitative comparison of the effect of mortality on PM10 in the different temperature strata (Morris and Naumova, 1998 and Roberts, 2004). To analyze the effects of PM10 in different temperature strata, we started with a Poisson log-linear model and included the temperature strata, for which it was assumed that the effect of PM10 on mortality was purely additive. The model (2) is described as follows:equation(2)lnEyt,c=αc+factortemp_ranget−i,c+PM10t−i,c:factortemp_rangest−i,c+stempt−i,c,df+stimet,df+factorDOWt+shumidt−i,c+spresst−i,c.