INTRODUCTION TO NMMAPS PART II
This report provides an integrated synthesis of the key
findings of NMMAPS on air pollution and morbidity and
mortality. The report begins by introducing the rationale
for the multicity approach that is used in NMMAPS and
briefly describing the statistical methods used to combine
evidence across locations. The findings on mortality are
then presented for the 2 databases: the 20 and 90 largest
US cities. In the analysis of the 20 cities, the primary analytic
thrust was toward estimating the overall effects of
PM10 and other criteria air pollutants. We used the previously
described Bayesian hierarchical model developed
for this purpose (Dominici et al 2000; Samet et al 2000a).
Air pollution–mortality associations are assessed within
the individual cities with previously described methods
(Kelsall et al 1997); the evidence is then combined across
the cities using the model of Dominici and coworkers. We
next provide the results of using a multistage, regional
modeling approach for exploring spatial heterogeneity in
the 90-city database. We also evaluate sociodemographic
and other characteristics of the cities as determinants of
heterogeneity in the effects of PM10.
Hospitalization data are also analyzed by combining
information across cities. For morbidity, the cities were
selected with preference given to those 14 locations having
the most abundant PM10 measurements. The within-cities
time-series analyses are accomplished with a distributed
lag approach developed by Schwartz (2000b). Evidence is
then combined across locations using hierarchical methods
common to meta-analysis. This approach also allows the
examination of sociodemographic characteristics of the
population as modifiers of the effect of PM10 on heart and
lung disease. In addition, the assessment of confounding by
other pollutants was done in the second stage of the hierarchical
model.