జర్నల్ ఆఫ్ పొల్యూషన్ ఎఫెక్ట్స్ & కంట్రోల్
అందరికి ప్రవేశం

నైరూప్య

Distributed Lag Models: An Analysis of Milan Mortality Data

Mieczyslaw Szyszkowicz and Wesley S Burr

The objective of this study is to present a new variation on Distributed Lag Non-Linear Models (DLNMs) for assessing associations between counts of health events and exposure to ambient air pollution. For illustrative purposes, a well-known data set for Milan, Italy was considered. Total Suspended Particulate (TSP) concentrations were used as the air pollution measure, and meteorological data were represented by daily mean temperature and relative humidity. Relative risks (RR) were estimated using Poisson Generalized Linear Models. Two controls for long time-scale variation were considered: a more traditional cubic regression spline smoother, and the more recent case-crossover (CC) control approach. The mortality displacement effect was estimated using DLNMs for a relatively high number of constructed lags. For the considered lags (0–45 days) and the CC approach, three regions were identified: region A (lags 0–7) with RR=1.021 (95 % confidence interval: 1.009, 1.043; region B (lags 8–27) with RR=0.981 (0.965, 0.997); and region C (lags 28–45) with RR=1.018 (1.003, 1.032). The total cumulative risk (regions A + B + C, lags 0–45) gave RR=1.019 (1.001, 1.037). The results were reported for an interquartile range (IQR=86.5) increase in TSP air pollution and are similar in structure to those previously reported, albeit at a significantly reduced level. We attribute the change to the considerable change in long timescale variation left in the residuals, as the clustering effect controls seasonal effects at a much stronger level.