The New England Journal of Medicine 
Owned, published, and © copyrighted, 2000, by the MASSACHUSETTS MEDICAL
SOCIETY

----------------------------------------------
Volume 343(24)             14 December 2000             pp 1742-1749
----------------------------------------------

Fine Particulate Air Pollution and Mortality in 20 U.S. Cities, 1987-1994
[Original Articles]
Samet, Jonathan M.;  Dominici, Francesca;  Curriero, Frank C.;  Coursac, Ivan; 
Zeger, Scott L.
From the Departments of Epidemiology (J.M.S.) and Biostatistics (F.D., F.C.C.,
I.C., S.L.Z.), School of Hygiene and Public Health, Johns Hopkins University,
Baltimore. Address reprint requests to Dr. Samet at Johns Hopkins University,
School of Public Health, 615 N. Wolfe St., Suite 6041, Baltimore, MD 21205, or
at jsamet@jhsph.edu.

----------------------------------------------

Outline

Abstract
Methods

Data Collection
Statistical Analysis

Results
Discussion
REFERENCES

Graphics

Table 1
Table 2
Table 3
Figure 1
Figure 2

----------------------------------------------

Abstract

Background: Air pollution in cities has been linked to increased rates of
mortality and morbidity in developed and developing countries. Although these
findings have helped lead to a tightening of air-quality standards, their
validity with respect to public health has been questioned.

: We assessed the effects of five major outdoor-air pollutants on daily
mortality rates in 20 of the largest cities and metropolitan areas in the United
States from 1987 to 1994. The pollutants were particulate matter that is less
than 10 microm in aerodynamic diameter (PM(10))), ozone, carbon monoxide, sulfur
dioxide, and nitrogen dioxide. We used a two-stage analytic approach that pooled
data from multiple locations.

: After taking into account potential confounding by other pollutants, we found
consistent evidence that the level of PM(10)) is associated with the rate of
death from all causes and from cardiovascular and respiratory illnesses. The
estimated increase in the relative rate of death from all causes was 0.51
percent (95 percent posterior interval, 0.07 to 0.93 percent) for each increase
in the PM(10)) level of 10 microg per cubic meter. The estimated increase in the
relative rate of death from cardiovascular and respiratory causes was 0.68
percent (95 percent posterior interval, 0.20 to 1.16 percent) for each increase
in the PM(10)) level of 10 microg per cubic meter. There was weaker evidence
that increases in ozone levels increased the relative rates of death during the
summer, when ozone levels are highest, but not during the winter. Levels of the
other pollutants were not significantly related to the mortality rate.

Conclusions: There is consistent evidence that the levels of fine particulate
matter in the air are associated with the risk of death from all causes and from
cardiovascular and respiratory illnesses. These findings strengthen the
rationale for controlling the levels of respirable particles in outdoor air. (N
Engl J Med 2000;343:1742-9.)

----------------------------------------------

Studies showing that current levels of air pollution in the cities of many
developed and developing countries are associated with increased rates of
mortality and morbidity have heightened concern that air pollution continues to
pose a threat to public health. [1-3] The evidence suggests that small airborne
particles are a toxic component of urban air pollution. Using this interpretation
of the evidence as a rationale, the Environmental Protection Agency implemented
a new standard for fine particulate matter. [4] The existing standard,
promulgated in 1987, specified the maximal levels allowable in a 24-hour period
and on an annual basis for particulate matter with an aerodynamic diameter (the
diameter of a unit-density sphere that has the same settling velocity in gas as
the particle of interest) that was less than 10 microm (PM(10))). In 1997, the
agency added standards for particulate matter that is less than 2.5 microm in
aerodynamic diameter (PM(2).5)), since the size of such particles better
corresponds to the size of particles that can penetrate to the airways and
alveoli of the lung. This decision has been controversial; critics question
whether the scientific evidence is strong enough to take regulatory action.
[5-8] A more detailed version of our methods and findings is available
elsewhere. [9]

Key findings on particulate air pollution have come from time-series analyses of
the association of air-pollution levels with the number of deaths per day. [3]
With the exception of a few studies, such as the multi-city Air Pollution and
Health: a European Approach (APHEA) project [10] and an analysis of data from
six U.S. cities, [11] most of these studies have been based on single locations
selected without a defined sampling plan. Consequently, the generalizability of
the findings is uncertain, and analytic strategies have differed among studies.
Citing these limitations, critics have questioned whether the findings indicate
an effect of air pollution generally or of particles specifically. [7,12,13]

To address these limitations, we combined information on the associations of
levels of the five major outdoor-air pollutants - PM(10)), ozone, sulfur
dioxide, carbon monoxide, and nitrogen dioxide - with daily mortality rates from
20 of the largest U.S. cities. [14] Our estimates are based on a defined sample
of the cities; statistical precision was enhanced by combining information from
multiple locations.

Methods

Data Collection

Data were collected from 1987 through 1994. We began with the 20 counties deemed
the largest in the 1990 U.S. Census on the basis of population (or with logical
groupings of counties), and for the analysis, we used data for the counties that
included the associated cities, thus encompassing a population of more than 50
million. Analysis was carried out at the county level because the county was the
common coding unit for the various data sets. In this article, we refer to
cities and metropolitan areas rather than counties. Daily mortality rates were
obtained from the National Center for Health Statistics (Table 1). After
excluding deaths from external causes (e.g., accidents, suicide, and homicide)
and deaths of nonresidents, we classified the deaths according to age group
(<65 years, 65 to 74 years, and greater/equal 75 years) and cause (cardiovascular
and respiratory and other). [15] Data on selected demographic characteristics
were obtained from the 1990 U.S. Census. [16]

----------------------------------------------
Table 1. Rates of Death from All Causes and from Cardiovascular and Respiratory
Causes in 20 U.S. Cities and Metropolitan Areas, According to Various Socioeconomic
Characteristics, 1987-1994.
----------------------------------------------

Hourly temperature and dew-point data were available from the EarthInfo
compact-disk [17] data base of the National Climatic Data Center. For analysis
we used the 24-hour mean value for each day. The air-pollution data were
obtained from the data base of the Aerometric Information Retrieval System, [18]
which is maintained by the Environmental Protection Agency. For population-oriented
monitoring variables, we downloaded all available data for PM(10)), ozone,
carbon monoxide, sulfur dioxide, and nitrogen dioxide. For the pollutants
measured on an hourly basis, we calculated the 24-hour average. If the levels of
pollutants were monitored at multiple locations in a metropolitan area, we
averaged the data. To avoid the potential consequences of outlying values, we
excluded the highest and lowest 10 percent of values (10 percent trimmed mean)
and then averaged the values for each set of monitors, after the value for each
monitor had been corrected for its yearly average.

Statistical Analysis

We used a two-stage log-linear regression model. [19-21] In the first stage, a
separate log-linear regression of the daily mortality rate on air-pollution
measures and other confounders was fitted to obtain estimates of the relative
rate of mortality associated with the pollution variable and the degree of
statistical uncertainty for each of the 20 cities. In the second stage, the
estimates of the relative rates were combined for all cities (after adjustment
for the various levels of uncertainty) to obtain an overall estimate and to
assess whether city-specific characteristics modified the estimated effect of
air pollution on the relative rate of death.

In the first-stage log-linear regressions, we controlled for possible confounding
by longer-term trends resulting from changes in the size and characteristics of
the population, health status, and health care and from shorter-term effects of
seasonality and the presence or absence of influenza epidemics. To do this, we
used a flexible function that took into account the variation in the mortality
rate over periods of several months (a smoothing function with respect to
calendar time with 7 degrees of freedom per year per city, which was allowed to
differ in the three age groups). We also adjusted for the short-term effect of
weather on the risk of death by including similar smoothing functions with
respect to a specific day's temperature and the average temperature for the
three days preceding it (6 degrees of freedom) and to dew point (3 degrees of
freedom). Finally, we included indicator variables for the day of the week. This
model specification was based on extensive, previously reported exploratory
analyses. [15,22,23] In this article, our results do not reflect the degrees of
freedom used. We have found that the relative rates of air pollution were not
sensitive to the number of degrees of freedom selected for the smoothing
functions of time, temperature, and dew point. [14,15,22,23]

In the first-stage analysis, we analyzed the effect of the day on which the
pollution data were obtained (the current day, the day before, or two days
before) on the association with mortality rates. The overall effect did not vary
with the lag interval selected. Consequently, we report data for a one-day lag
between pollution variables and mortality.

We considered the effects of multiple pollutants on the relative rate of
mortality. We initially conducted univariate analyses that included PM(10))
alone and ozone alone. We then considered the effects of these two pollutants in
a bivariate model and developed trivariate models that also included sulfur
dioxide, nitrogen dioxide, or carbon monoxide. The trivariate models provided
estimates of the individual effects of carbon monoxide, sulfur dioxide, and
nitrogen dioxide on the risk of death after adjustment for PM(10)) and ozone
levels.

The second stage of the analysis provided pooled estimates of the relative rates
of mortality associated with specific pollutants and a characterization of the
effects of air pollutants among the cities. We also examined factors determining
heterogeneity in the effect of air pollution on mortality. With respect to
determinants of heterogeneity in the second stage of the analysis, we assumed
that first-stage estimates of the relative mortality rates associated with
specific pollutants followed a linear regression with the selected city-specific
demographic characteristics (Table 1) as predictor variables. The second-stage
analysis provided an estimate of the effect of each predictor variable on the
relative rate of mortality associated with PM(10)).

Model fitting was performed with use of a Bayesian statistical approach, [24]
which provides an estimate of the posterior distribution of the variable of
interest. We carried out this analysis without making a strong prior assumption
as to the value of the relative rate. The posterior distribution is used to
determine the probability that the relative rate of mortality associated with
PM(10)) has a particular value - that is, it is a measure of the strength of the
evidence. One important calculation is the posterior probability that the
relative rate of mortality associated with PM(10)) is greater than zero. The
posterior distribution can also be used to determine the 95 percent posterior
intervals. The 95 percent posterior interval encompasses 95 percent of the
posterior distribution, a Bayesian formulation similar to the 95 percent
confidence interval. All analyses were performed with use of S-Plus statistical
software. [25]

Results

The 20 cities and metropolitan areas broadly represented the United States. The
number of days for which pollution data were available varied (Table 2). Since
the Environmental Protection Agency requires levels of PM(10)) to be measured
only every six days, data for ozone and other pollutants were generally
available on more days. The mean daily values for PM(10)) ranged from about 20
microg per cubic meter to nearly 50 microg per cubic meter; the present maximal
allowable level of PM(10)) in a 24-hour period is 150 microg per cubic meter.
The average numbers of deaths per day were substantial, ranging from less than
20 to nearly 200 (Table 1). The correlation coefficients of all correlations
between pollutants for all 20 cities and metropolitan areas are provided in
Table 3. The correlation structure generally reflects the common sources of the
primary combustion-related gases (sulfur dioxide, nitrogen dioxide, and carbon
monoxide) and of PM(10)). The level of ozone was only slightly correlated with
that of PM(10)) and was not correlated with the levels of other gaseous
pollutants.

----------------------------------------------
Table 2. Mean Levels of Pollutants in 20 U.S. Cities and Metropolitan Areas.
----------------------------------------------

----------------------------------------------
Table 3. Correlation Coefficients of All Pairwise Correlations between
Pollutants for the 20 Cities and Metropolitan Areas.
----------------------------------------------

In initial univariate analyses, the level of PM(10)) was positively associated
with the rate of death from all causes in most of the 20 cities and metropolitan
areas (Figure 1). Adjustment for the effect of ozone levels had little effect on
the association, whereas the effects of the ozone level, before and after
adjustment for PM(10)) levels, tended to be more variable. The analysis of each
pollutant was also stratified according to the cause of death. The city-specific
associations between PM(10)) levels and the rate of death from cardiovascular
and respiratory causes were similar to those for the rate of death from all
causes. A previous univariate analysis stratified according to age showed no
age-associated trend. [14]

----------------------------------------------
Figure 1. Regression Coefficients for the Changes in the Rate of Death from All
Causes for Each Increase in the PM(10)) Level of 10 microg per Cubic Meter,
before and after Adjustment for Ozone Levels, and for Each Increase in the Ozone
Level of 10 ppb, before and after Adjustment for PM(10)) Levels in 20 Cities and
Metropolitan Areas. PM(10)) denotes particulate matter that is less than 10
microm in aerodynamic diameter. Bars indicate 95 percent confidence intervals.
No data on ozone were available for Minneapolis.
----------------------------------------------

The combined analysis for all 20 cities and metropolitan areas confirmed the
association between PM(10)) levels and the rate of death from all causes (Figure
2) and of death from cardiovascular and respiratory causes. Figure 2 shows the
posterior distributions of the estimated increases in the relative rates of
death from all causes associated with each increase in the PM(10)) level of 10
microg per cubic meter before and after adjustment for levels of ozone, nitrogen
dioxide, sulfur dioxide, and carbon monoxide, as well as the probability that
overall effects are greater than zero for each model. With respect to death from
all causes, the distributions are shifted toward the right, with the respective
mean increases in the number of deaths per day for each increase in the PM(10))
level of 10 microg per cubic meter (i.e., estimated relative rates) ranging
between approximately 0.3 percent and 0.6 percent. An increase in the relative
rate of 0.3 percent corresponds to a relative risk of death of 1.003. In the
model that included PM(10)) alone, the estimated increase in the relative rate
of death from all causes was 0.51 percent for each increase in the PM(10)) level
of 10 microg per cubic meter (95 percent posterior interval, 0.07 to 0.93
percent). The posterior distributions of the PM(10)) levels did not change
substantially after adjustment for the other pollutants, suggesting that the
univariate findings were not affected by confounding by other pollutants (Figure
2).

----------------------------------------------
Figure 2. Posterior Distributions of the Overall Relative Rate of Increase in
Death from All Causes for Each Increase in the PM(10)) Level of 10 microg per
Cubic Meter, before and after Adjustment for the Levels of Ozone (O(3))),
Nitrogen Dioxide (NO(2))), Sulfur Dioxide (SO(2))), and Carbon Monoxide (CO).
Values in parentheses are the posterior probabilities that the overall effects
are greater than zero. PM(10)) denotes particulate matter that is less than 10
microm in aerodynamic diameter.
----------------------------------------------

The PM(10)) level had a somewhat greater effect on the rate of death from
cardiovascular and respiratory causes than on the rate of death from all causes
and was associated with a correspondingly larger probability that the effect was
greater than zero. The estimated increase in the relative rate of death from
cardiovascular and respiratory causes was 0.68 percent for each increase of 10
microg per cubic meter in the PM(10)) level (95 percent posterior interval, 0.20
to 1.16 percent).

The univariate effects of ozone levels were examined during a one-year period
and according to season. Overall, the posterior distributions of the effects of
ozone were concentrated near zero, and there was only an even chance that the
effect was larger than zero when death from all causes and death from cardiovascular
and respiratory causes were considered separately. Because ozone levels vary
strongly with the season, we compared the effects of ozone levels during the
three hottest summer months (June, July, and August), when levels are highest,
and three cold months (November, December, and January), when levels tend to be
lowest. With the use of this stratification, the estimated relative rates of
death from all causes with each increase in the ozone level of 10 ppb were 0.41
percent (95 percent posterior interval, -0.20 to 1.01 percent) during the summer
months and -1.83 percent (95 percent posterior interval, -2.69 to -0.96 percent)
during the cold months.

The differences between cities in the relative rates did not depend on average
PM(10)) or ozone levels in a city or on city-specific demographic characteristics;
for these variables, all associated 95 percent posterior intervals included
zero. Consequently, the analyses and results for PM(10)) were not adjusted for
these city-specific characteristics.

We also analyzed the effects of levels of carbon monoxide, sulfur dioxide, and
nitrogen dioxide in a fashion similar to that of the analysis of PM(10)) levels.
After adjustment for PM(10)) and ozone levels, we found little evidence that
these pollutants had a significant effect on the relative rate of death.

Discussion

We found consistent evidence that the level of PM(10)) is associated with the
rates of death from all causes and from cardiovascular and respiratory causes.
The association of PM(10)) was not affected by the inclusion of other pollutants
in the statistical model or by the time at which data were collected. Our
findings strongly support the findings of prior studies of particulate matter
and mortality. [26] These studies, which were largely based on data from single
cities, used a variety of measures of particulate matter, including levels of
total suspended particles, black smoke (a measure of soiling of a filter that
provides an index of particle levels), PM(10)), and PM(2).5). The statistical
methods used to assess the relations between levels of pollution and the risk of
death were also heterogeneous; for example, there was no uniformity in the
approaches used to control for factors that varied over time or for other
pollutants. Nonetheless, using a weight-of-evidence approach, the Environmental
Protection Agency interpreted the results of the studies as indicating a
possibly causal association between levels of particulate matter and adverse
effects on health. [3]

In a meta-analysis of U.S. studies of particulate air pollution published
between 1990 and 1993, Dockery and Pope [2] estimated that each increase in the
PM(10)) level of 10 microg per cubic meter increased the relative rate of death
from all causes by 1 percent. In a subsequent update that included data from
reports published through 1995, Dockery and Pope found little change in this
estimate. [27] Schwartz [28] also performed a meta-analysis of studies published
between 1990 and 1993 but included data from London and Minneapolis in addition
to the data on the eight cities considered by Dockery and Pope. The resulting
estimated increase in the relative rate of death from all causes was 0.7 percent
for each increase in the PM(10)) level of 10 microg per cubic meter. The APHEA
project analyzed data from 12 European cities and then estimated summary
measures. For the six western European cities in the study, the mortality rate
was estimated to increase by 0.4 percent for each increase in the PM(10)) level
of 10 microg per cubic meter. In our 20-city analysis, our estimate of an
increase of approximately 0.5 percent in the rate of death from all causes for
each increase in the PM(10)) level of 10 microg per cubic meter is very similar
to the estimate of the APHEA project. [10] The fact that our estimate was lower
than those of Dockery and Pope [2] and Schwartz [28] may reflect differences in
analytic techniques and the cities selected. The initial reports included in the
meta-analyses may have been biased by the fact that studies with positive
findings are more likely to be selected for publication than those with negative
findings. Our 20-city estimate is not subject to such bias and our results
should thus be more applicable to the United States in general.

We did not find an effect of ozone levels on the overall rate of death from all
causes or from cardiovascular and respiratory causes during the full year
period. Ozone levels were positively associated with mortality rates during the
summer months when ozone levels were highest, although the 95 percent posterior
interval extended into the range indicating no effect of ozone levels on
mortality. The finding of an effect of ozone levels only during the summer may
reflect the higher levels of ozone during these months or, possibly, differences
in the characteristics of photochemical pollution during the various seasons.
Other recent studies have generally found an association between ozone levels
and the risk of death. [29] In the APHEA project, the maximal ozone levels
during a one-hour period were associated with the numbers of deaths per day in
four cities (London; Athens, Greece; Barcelona, Spain; and Paris), and a
quantitatively similar effect was found with additional data from three cities
(Amsterdam; and Basel and Zurich, Switzerland) that were not part of the APHEA
project. [30] For each increase of 50 microg per cubic meter in the one-hour
maximal level, the estimated relative risk of death was 1.029 (i.e., a 1.1
percent increase in the rate of death for each increase in the ozone level of 10
ppb), with the use of a random-effects model for combining the city-specific
data. Thurston and Ito [29] pooled data from 15 studies and estimated that the
relative risk of death was 1.036 for each increase of 100 ppb in the daily
one-hour maximal level of ozone (i.e., a 0.36 percent increase in the rate of
death for each increase in the ozone level of 10 ppb). For the summer months,
our estimate (a 0.41 percent increase in the rate of death for each increase in
the ozone level of 10 ppb) was similar to those of Thurston and Ito. Taken
together, the results of these three studies provide consistent evidence that
exposure to ozone also increases the risk of death.

The limitations of our analyses should be considered. Data on levels of PM(2).5)
are not yet available nationally, since a monitoring network for particles in
this size range is currently being implemented. We used PM(10)) levels because
they have been monitored since 1987; there is variation across the United States
in the proportion of PM(10)) mass that is made up of PM(2).5), so that the
PM(10)) level is an imperfect surrogate for the PM(2).5)level. [3] In addition,
for regulatory purposes, PM(10)) levels must only be measured every six days,
limiting the extent of available data.

Our analyses also did not address the extent to which life is shortened in
association with daily exposure to the various pollutants. The finding that the
association between PM(10)) levels and the risk of death was strongest for
cardiovascular and respiratory causes of death is consistent with the hypothesis
that persons made frail by advanced heart and lung disease are more susceptible
to the adverse effects of air pollution. The findings from several epidemiologic
studies of the longer-term effects of air pollution on the risk of death suggest
that exposure to air pollution may do more than simply shorten life by a few
days. [31,32] Several analyses of daily mortality data also indicate that the
effect of air pollution may go beyond shortening life by a few days. [33,34]

We found no evidence that key socioeconomic factors such as low socioeconomic
status affect the association between PM(10)) levels and the risk of death in
linear regression models. The medical conditions and poor health that increase
the risk of death may not be adequately reflected by the socioeconomic
indicators recorded by the U.S. Census. Thus, more specific information on
health status, rather than on social factors, may be needed to explore this
issue, particularly in relation to the susceptibility of particular groups of
people. Finally, we used county-level data for these social factors because most
of our data were categorized according to county. The variation in socioeconomic
status in a typical urban county, however, is usually considerably larger than
the variation among counties. Thus, the demographic factors considered in the
second stages of the models may be too broad to be informative.

The epidemiologic evidence that levels of particulate matter are associated with
the risk of mortality and morbidity has prompted the promulgation of a new
standard for PM(2).5) in the United States and a rethinking of guidelines for
particulate matter in Europe. Our analyses provide evidence that particulate air
pollution continues to have an adverse effect on the public's health and
strengthen the rationale for limiting levels of respirable particles in outdoor
air.

Supported by a contract with the Health Effects Institute, an organization
jointly funded by the Environmental Protection Agency (EPA R824835) and
automotive manufacturers. The contents of this article do not necessarily
reflect the views and policies of the Health Effects Institute, the Environmental
Protection Agency, or manufacturers of motor vehicles or engines. Also supported
by a grant from the National Institute of Environmental Health Sciences (P30 ES0
3819-12, to Johns Hopkins Center in Urban Environmental Health). Dr. Dominici is
the recipient of a Rosenblith Young Investigator Award from the Health Effects
Institute.

REFERENCES

1. Bascom R, Bromberg PA, Costa DL, et al. Health effects of outdoor air
pollution. Am J Respir Crit Care Med 1996;153:477-98. 

2. Dockery DW, Pope CA III. Acute respiratory effects of particulate air
pollution. Annu Rev Public Health 1994;15:107-32. [Medline Link] [BIOSIS
Previews Link] 

3. Environmental Protection Agency, Office of Air Quality Planning and
Standards. Review of the National Ambient Air Quality Standards for Particulate
Matter: policy assessment of scientific and technical information. OAQPS staff
paper no. EPA-452\\R-96-013. Washington, D.C.: Government Printing Office, 1996.


4. Environmental Protection Agency (EPA). National ambient air quality standards
for particulate matter. Fed Regist 1997;62:138. 

5. Gamble JF, Lewis RJ. Health and respirable particulate (PM(10))) air
pollution: a causal or statistical association? Environ Health Perspect
1996;104:838-50. [Medline Link] [BIOSIS Previews Link] 

6. Gamble JF. PM(2).5) and mortality in long-term prospective cohort studies:
cause-effect or statistical associations. Environ Health Perspect 1998;106:535-49.
[Medline Link] [BIOSIS Previews Link] 

7. McClellan RO, Miller FJ. An overview of EPA's proposed revision of the
particulate matter standard. CIIT Act 1997;17:1-24. 

8. Kaiser J. Showdown over clean air science. Science 1997;277:466-9. [Medline
Link] 

9. Samet JM, Zeger S, Dominici F, Dockery D, Schwartz J. The National Morbidity,
Mortality, and Air Pollution Study (NMMAPS). 1. Methods and methodological
issues. Cambridge, Mass.: Health Effects Institute, 2000. (See http://healtheffects.org/pubs/samet.pdf
.) 

10. Katsouyanni K, Touloumi G, Spix C, et al. Short-term effects of ambient
sulphur dioxide and particulate matter on mortality in 12 European cities:
results from times series data from the APHEA project: Air Pollution and Health:
a European Approach. BMJ 1997;314:1658-63. [Fulltext Link] [Medline Link]
[BIOSIS Previews Link] 

11. Schwartz J, Dockery DW, Neas LM. Is daily mortality associated specifically
with fine particles? J Air Waste Manag Assoc 1996;46:927-39. [Medline Link] 

12. Moolgavkar SH, Luebeck EG. A critical review of the evidence on particulate
air pollution and mortality. Epidemiology 1996;7:420-8. [Medline Link] 

13. Lipfert FW, Wyzga RE. Air pollution and mortality: the implications of
uncertainties in regression modeling and exposure measurement. J Air Waste Manag
Assoc 1997;47:517-23. [Medline Link] 

14. Dominici F, Samet J, Zeger SL. Combining evidence on air pollution and daily
mortality from the largest 20 U.S. cities: a hierarchical modeling strategy. J R
Stat Soc [A] 2000;152:397-406. 

15. Kelsall JE, Samet JM, Zeger SL, Xu J. Air pollution and mortality in
Philadelphia, 1974-1988. Am J Epidemiol 1997;146:750-62. [Medline Link] [BIOSIS
Previews Link] 

16. Bureau of the Census. Statistical abstract of the United States. Washington,
D.C.: Government Printing Office, 1990. 

17. National Climatic Data Center databases TD-3280, TD-3281. Boulder, Colo.:
EarthInfo, 1994. 

18. Office of Air Quality Planning and Standards. Aerometric information
retrieval system. Research Triangle Park, N.C.: Environment Protection Agency,
1999. 

19. Hastie TJ, Tibshirani RJ. Generalized additive models. New York: Chapman
& Hall, 1990. 

20. Lindley DV, Smith AFM. Bayes estimates for the linear model. J R Stat Soc
[B] 1972;34:1-41. 

21. Morris CN, Normand S-L. Hierarchical models for combining information and
for meta-analysis. In: Bernardo JM, Berger JO, Dawid AP, Smith AFM, eds.
Bayesian statistics 4: proceedings of the Fourth Valencia International Meeting.
Oxford, England: Oxford University Press, 1992:321-44. 

22. Samet JM, Zeger SL, Berhane K. The association of mortality and particulate
air pollution. In: Particulate air pollution and daily mortality: replication
and validation of selected studies: the phase I report of the Particle
Epidemiology Evaluation Project. Cambridge, Mass.: Health Effects Institute,
1995. 

23. Samet J, Zeger S, Kelsall J, Xu J, Kalkstein L. Air pollution, weather and
mortality in Philadelphia. In: Particulate air pollution and daily mortality:
analyses of the effects of weather and multiple air pollutants: the phase IB
report of the Particle Epidemiology Evaluation Project. Cambridge, Mass.: Health
Effects Institute, 1997. 

24. Gelman A, Carlin JB, Stern HS, Rubin DB. Bayesian data analysis. London:
Chapman & Hall, 1995. 

25. S-PLUS version 3.4. Seattle: MathSoft, 1998 (software). 

26. Samet JM, Zeger S, Dominici F, et al. The National Morbidity, Mortality, and
Air Pollution Study (NMMAPS). 2. Morbidity and mortality from air pollution in
the United States. Cambridge, Mass.: Health Effects Institute, 2000. (See
http://healtheffects.org/pubs/samet2.pdf .) 

27. Dockery D, Pope A. Epidemiology of acute health effects: summary of
time-series studies. In: Wilson R, Spengler JD, eds. Particles in our air:
concentrations and health effects. Cambridge, Mass.: Harvard University Press,
1996:123-47. 

28. Schwartz J. Air pollution and daily mortality: a review and meta analysis.
Environ Res 1994;64:36-52. [Medline Link] [BIOSIS Previews Link] 

29. Thurston GD, Ito K. Epidemiological studies of ozone exposure effects. In:
Holgate ST, Samet JM, Koren HS, Maynard RL, eds. Air pollution and health. San
Diego, Calif.: Academic Press, 1999:485-510. 

30. Touloumi G, Katsouyanni K, Zmirou D, et al. Short-term effects of ambient
oxidant exposure on mortality: a combined analysis within the APHEA project. Am
J Epidemiol 1997;146:177-85. [Medline Link] [BIOSIS Previews Link] 

31. Dockery DW, Pope CA III, Xu X, et al. An association between air pollution
and mortality in six U.S. cities. N Engl J Med 1993;329:1753-9. [Fulltext Link]
[Medline Link] [BIOSIS Previews Link] 

32. Pope CA III, Thun MJ, Namboodiri MM, et al. Particulate air pollution as a
predictor of mortality in a prospective study of U.S. adults. Am J Respir Crit
Care Med 1995;151:669-74. [Medline Link] [BIOSIS Previews Link] 

33. Zeger SL, Dominici F, Samet J. Harvesting-resistant estimates of air
pollution effects on mortality. Epidemiology 1999;10:171-5. [Medline Link] 

34. Schwartz J. Harvesting and long term exposure effects in the relation
between air pollution and mortality. Am J Epidemiol 2000;151:440-8. [Medline
Link] 

----------------------------------------------