电话：010-85195300
E-mail：cassoa@cass.org.cn
地址：中国北京建国门内大街5号中国社会科学院1#楼
扫一扫关注微信
国家哲学社会科学文献中心版权所有
Associations between Air Pollution and Mortality in Phoenix, 1995-1997
Therese F. MarWe evaluated the association between mortality outcomes in elderly individuals and particulate matter (PM) of varying aerodynamic diameters (in micrometers) [[PM.sub.10], [PM.sub.2.5], and [PM.sub.CF] ([PM.sub.10] minus [PM.sub.2.5])], and selected particulate and gaseous phase pollutants in Phoenix, Arizona, using 3 years of daily data (1995-1997). Although source apportionment and epidemiologic methods have been previously combined to investigate the effects of air pollution on mortality, this is the first study to use detailed PM composition data in a time--series analysis of mortality. Phoenix is in the arid Southwest and has approximately 1 million residents (9.7% of the residents are [is greater than] 65 years of age). PM data were obtained from the U.S. Environmental Protection Agency (EPA) National Exposure Research Laboratory Platform in central Phoenix. We obtained gaseous pollutant data, specifically carbon monoxide, nitrogen dioxide, ozone, and sulfur dioxide data, from the EPA Aerometric Information Retrieval System Database. We used Poisson regression analysis to evaluate the associations between air pollution and nonaccidental mortality and cardiovascular mortality. Total mortality was significantly associated with CO and [NO.sub.2] (p [is less than] 0.05) and weakly associated with [SO.sub.2], [PM.sub.10], and [PM.sub.CF] (p [is less than] 0.10). Cardiovascular mortality was significantly associated with CO, [NO.sub.2], [SO.sub.2], [PM.sub.2.5], [PM.sub.10], [PM.sub.CF] (p [is less than] 0.05), and elemental carbon. Factor analysis revealed that both combustion-related pollutants and secondary aerosols (sulfates) were associated with cardiovascular mortality. Key words: cardiovascular, composition, factor analysis, particulate matter, [PM.sub.2.5], [PM.sub.10], sources. Environ Health Perspect 108:347-353 (2000). [Online 25 February 2000]
http://ehpnet1.niehs.nih.gov/docs/2000/108p347-353mar/abstract.html
The associations between air pollution, especially particulate matter (PM), and adverse human health effects have been well documented (1-10). PM is associated with decreased respiratory function, aggravation of existing respiratory and cardiovascular conditions, altered defense mechanisms, and even premature death. The most susceptible populations include those with preexisting respiratory or cardiovascular conditions, asthmatics, children, and the elderly (8,11).
To date, few epidemiology studies have used PM measures other than size-segregated mass as the exposure metric. Schwartz et al. (12) looked at episodes of high coarse particle concentration in Spokane, Washington, and found that windblown dust episodes were not associated with increased mortality. In the Harvard Six Cities Study, Schwartz et al. (3) found a significant association between nonaccidental mortality and particulate matter [is less than or equal to] 2.5 [micro]m in aerodynamic diameter ([PM.sub.2.5]) and sulfur. They did not find a significant association with particulate matter [is less than or equal to] 10 [micro]m in aerodynamic diameter ([PM.sub.10]) or the coarse fraction of PM [[PM.sub.CF] ([PM.sub.10] minus [PM.sub.2.5])]. In contrast, Ostro et al. (13) found that [PM.sub.10] dominated by coarse particles was associated with an increase in mortality in the Coachella Valley in California. The differences in the results from these two studies may be due to the particulate composition as well as the difference in the amount of [PM.sub.CF]. In the eastern United States, [PM.sub.2.5] is dominated by sulfates (34%), whereas in the western and central United States it is dominated by organic carbon (OC) from motor vehicles and vegetative burning (39%) (14). The average [PM.sub.2.5]/[PM.sub.10] ratio for the Six Cities Study (3) was 0.6 (based on the 50th percentiles) as compared to a ratio of 0.3 for Phoenix, Arizona (15).
The goal of the present study was to evaluate the associations between daily air pollution and total nonaccidental and cardiovascular mortality in Phoenix. Phoenix is an arid southwestern city with a population of approximately 1 million residents (16). It is an interesting location because of its large proportion of elderly people (9.7% of the population is [is greater than] 65 years of age). The elderly are more susceptible to air pollution than the general public (2). The primary sources of PM in Phoenix are motor vehicles, paved road dust, and vegetative burning (15).
This study focused on the effects of air pollution on cardiovascular mortality for several reasons. First, the association between air pollutants and cardiovascular mortality has been consistent in previous studies (1,2,9,17). Second, a study in Baltimore, Maryland, found that heart rate variability was associated with [PM.sub.2.5] in elderly subjects with cardiovascular conditions (18). Finally, in this study cardiovascular mortality had the largest sample size, accounting for 45% of the total non accidental deaths in the study region (based on zip codes). This may be reflective of the increased size of Phoenix's elderly population, which is more prone to cardiovascular disease.
A unique aspect of this study is that our pollution data include daily information not only on traditional gaseous pollutants, but also on PM in various size fractions and the chemical composition of [PM.sub.2.5]. From 1995 to 1997, the U.S. Environmental Protection Agency (EPA) National Exposure Research Laboratory (NERL) operated a comprehensive monitoring platform in Phoenix. They collected daily [PM.sub.2.5] samples and subsequently analyzed them for various chemical components of PM. This provided an opportunity to examine more specific metrics for PM than simply mass, as well as an opportunity to identify selected chemical components of PM that are associated with mortality.
In addition to PM, this study also evaluated the association between total nonaccidental and cardiovascular mortality and other measured air pollutants: carbon monoxide, nitrogen dioxide, sulfur dioxide, and ozone. These EPA criteria pollutants are also associated with mortality (7,17,19,20).
Methods
Study area and data. Mortality data for all of Maricopa County from 1995 to 1997 were obtained from the Arizona Center for Health Statistics in Phoenix. Death certificate data included residence zip code and the primary cause of death as identified by the International Classification of Diseases, Ninth Revision (ICD-9, World Health Organization, Geneva). Only the deaths of residents in the zip codes located near the air pollution platform were included in this study. This zip-code region was recommended by the Arizona Department of Environmental Quality (Phoenix, AZ). We evaluated total nonaccidental mortality (ICD-9 codes [is less than] 800) and cardiovascular mortality (ICD-9 codes 390-448.9) in this study. Summary statistics for the mortality outcomes are presented in Table 1.
Table 1. Mortality counts for individuals [is greater than or equal to] 65 years of age in Phoenix.
Total Average nonaccidental
Year nonaccidental deaths/day
1995 3,072 8.45
1996 3,201 8.74
1997 3,003 8.45
1995-1997 9,276 8.55
Total Average cardiovascular
Year cardiovascular deaths/day
1995 1,391 3.86
1996 1,473 3.98
1997 1,318 3.73
1995-1997 4,182 3.85
We obtained [PM.sub.2.5], [PM.sub.10], [PM.sub.CF], and [PM.sub.2.5] chemical composition data from the EPA NERL platform in central Phoenix. Chemical composition was only available for [PM.sub.2.5]. The monitoring platform is approximately 10 km west-northwest of downtown Phoenix at a state and local air monitoring station. Standard meteorologic parameters such as wind speed and direction, temperature, and relative humidity were continuously measured. The average temperature in Phoenix from 1995 to 1997 was 23.7 [+ or -] 8.1 [degrees] C. The average relative humidity was 32 [+ or -] 15%.
NERL investigators made hourly [PM.sub.2.5] and [PM.sub.10] measurements each day using two colocated tapered element oscillation microbalance (TEOM) monitors (Rupprecht & Patasnick Co., Albany, NY). The TEOM-[PM.sub.10] was fitted with an EPA-approved federal reference method [PM.sub.10] impactor inlet (model 246b; Andersen Instruments, Smyrna, GA). The TEOM-[PM.sub.2.5] was fitted with a [PM.sub.2.5] cyclone inlet (University Research Glassware, Chapel Hill, NC). The [PM.sub.2.5] cyclone on the TEOM was replaced with a well-impactor ninety-six (WINS) inlet on 20 December 1996. The WINS inlet has a sharper cut point as compared to the cyclone. We averaged the hourly concentrations to create a 24-hr average (0700-0700 hr), and we calculated the concentration of coarse fraction (TEOM [PM.sub.CF]) as TEOM [PM.sub.10] minus TEOM [PM.sub.2.5].
NERL investigators collected the daily gravimetric integrated 24-hr (starting at 0700) fine particle filter samples using a dual fine particle sequential sampler (DFPSS; University Research Glassware). The DFPSS was fitted with a cyclone that was identical to the cyclone on the TEOM-[PM.sub.2.5]. The DFPSS collected daily samples on both Teflon and quartz filters. The Teflon filter was used for mass and elemental analysis, whereas the quartz filter was used for carbon analysis. In addition to the DFPSS, NERL investigators operated a dichotomous sampler (Andersen Instruments, Inc.) every third day beginning 17 June 1996. Both the [PM.sub.2.5] and [PM.sub.CF] samples were collected on Teflon filters. The investigators measured elemental concentrations at the EPA (Research Triangle Park, NC) with energy dispersive X-ray fluorescence. OC and elemental carbon (EC) were measured by Sunset Laboratory (Forest Grove, OR) using thermal optical transmittance (21).
PM and gaseous pollutant concentrations (range and mean [+ or -] SD) from 1995 to 1997 are presented in Table 2. We obtained gaseous criteria pollutant data for CO, [NO.sub.2], [O.sub.3], and [SO.sub.2] from the EPA Aerometric Information Retrieval System (AIRS) database (22) for residential sites in the Phoenix region. We averaged CO values over four monitoring sites and we averaged [NO.sub.2] over two sites. Only one residential monitoring site was available for [SO.sub.2]. We averaged the hourly averages for CO, [NO.sub.2], and [SO.sub.2] over 24 hr from 0700 to 0700. We used the maximum hourly [O.sub.3] ([O.sub.3] max) concentration in the same 24-hr period in the analysis.
Table 2. Annual range of pollutant concentrations (1995-1997). Particulate matter pollutant, year Range Gaseous pollutant [PM.sub.2.5] (DFPSS) CO (ppm) 1995 4-37 1996 3-39 1997 2-35 3-year mean 12.0 [+ or -] 6.6 [PM.sub.10] (TEOM) [NO.sub.2] (ppb) 1995 9-129 1996 5-213 1997 7-186 3-year mean 46.5 [+ or -] 22.3 [PM.sub.2.5] (TEOM) [O.sub.3] max (ppb) 1995 1-40 1996 0-42 1997 1-34 3-year mean 13.0 [+ or -] 7.2 [PM.sub.CF] (TEOM) [SO.sub.2] (ppb) 1995 5-104 1996 5-187 1997 5-159 3-year mean 33.5 [+ or -] 17.3 Particulate matter pollutant, year Range [PM.sub.2.5] (DFPSS) 1995 0.5-4.0 1996 0.3-4.0 1997 0.3-3.7 3-year mean 1.5 [+ or -] 08 [PM.sub.10] (TEOM) 1995 8-64 1996 9-59 1997 8-61 3-year mean 30 [+ or -] 10 [PM.sub.2.5] (TEOM) 1995 10-131 1996 14-112 1997 14-104 3-year mean 57.0 [+ or -] 17.7 [PM.sub.CF] (TEOM) 1995 0-11 1996 1-17 1997 2-12 3-year mean 3.1 [+ or -] 2.2
The [PM.sub.2.5] constituents that we evaluated for effects on mortality were sulfur, zinc, lead, soil-corrected potassium ([K.sub.S]) (23), OC, EC, total carbon (TC), and reconstructed soil. Soil was reconstructed by summing the oxides of Al, Si, Ca, Fe, and Ti using the formula recommended by Maim et al. (24). We also considered [PM.sub.2.5] that was corrected for soil content (nonsoil [PM.sub.2.5] = [PM.sub.2.5] - reconstructed soil). Table 3 presents the percent of the total mass of [PM.sub.2.5] accounted for by each component. The elements aluminum, silicon, calcium, titanium, and iron were not evaluated separately in the mortality analysis because they are the major elemental components of soil.
Table 3. Percent of total mass of [PM.sub.2.5] accounted for by each component.
Component [PM.sub.2.5] (%) S(a) 3.69 Mn 0.05 Zn 0.15 Br 0.03 Pb 0.06 OC(a) 1.4 38.37 EC 10.78 [K.sub.S] 0.52 Soil(b) 17.50
(a) If S is assumed to be in the form of [([NH.sub.4]).sub.2][SO.sub.4], the mass percent would be 15.2%.
(b) 2.20% AI + 2.49% Si + 1.63% Ca + 2.42% Fe + 1.94% Ti (23).
Statistical analysis. In our zip-code regions, we analyzed a total of 9,276 nonaccidental deaths from 1995 to 1997. Poisson regression was used to evaluate the association between the air pollutant exposure variables and the mortality outcomes (2,5).
We used Poisson regression because mortality data are discrete counts and death is a rare event. Poisson regression assumes the variance is equal to the mean. When the variance exceeds the mean, the variance is overdispersed. We adjusted standard errors for overdispersion; however, the amount of overdispersion was small. The overdispersion parameter was 1.05 and 1.00 for nonaccidental and cardiovascular mortality, respectively. We calculated all relative risks (RRs) for an interquartile increase (25th-75th percentile) in pollutant concentration.
The effect of air pollution on mortality is small and can be influenced by confounders. Therefore, base models for total mortality and cardiovascular mortality were constructed by adjusting for day of the week with indicator variables, and time trends, temperature, and relative humidity with smoothing functions (25). We determined degrees of freedom (df) for the function used to smooth time trend by minimizing autocorrelation as well as the Akaike information criterion (AIC) (26). We chose the df and lag for the smoothing functions for temperature and relative humidity to minimize the AIC. The base model for total mortality used indicator variables for day of the week, 10 df for time trends, 2 df for temperature with 2 days lag, and 2 df for relative humidity with 0 days lag. The base model for cardiovascular mortality used indicator variables for the day of the week, 10 df for time trends, 2 df for temperature with 1 day lag, and 2 df for relative humidity with 0 days lag.
We included continuous daily data from 1995 to 1997 (1,097 days) in the study. Each day was coded and included in the model to adjust for time trends. Little autocorrelation was observed after adjusting for day of week, time trends, temperature, and relative humidity. The autocorrelation for days 1-25 for both total and cardiovascular mortality were within the 95% confidence interval for an independent series.
We evaluated air pollution exposure variables by adding them individually as linear terms to the base model. The air pollution exposure metrics that were evaluated in this analysis included CO, [NO.sub.2], [O.sub.3], [SO.sub.2], TEOM [PM.sub.10], TEOM [PM.sub.2.5], TEOM [PM.sub.CF], [PM.sub.2.5] (DFPSS), S, Zn, Pb, soil, [K.sub.S], nonsoil PM, OC, EC, and TC. Lag days ranging from 0 to 4 were investigated. We evaluated the assumption of a linear relationship using a smooth function. This assumption was met if a straight line could be placed within the 95% confidence intervals (CIs). A p-value [is less than] 0.05 associated with the pollution exposure variable was considered significant. We conducted Poisson regression analyses using S-PLUS 4 (Mathsoft, Inc., Seattle, WA)
Factor analysis. We conducted a factor analysis on the daily concentrations of the chemical components of [PM.sub.2.5] from samples collected by the DFPSS (Al, Si, S Ca, Fe, Zn, Mn, Pb, Br, [K.sub.S], OC, and EC). The analysis also included the daily averages of the gaseous species emitted by combustion sources (CO, [NO.sub.2], and [SO.sub.2]). Factor analysis is a technique used to explain the correlations between variables in terms of underlying factors that are nor directly measurable. Each factor is a linear combination of the original variables and all such factors are orthogonal to each other. The factors were extracted using principal component analysis with a varimax rotation. We conducted factor analysis using SAS (SAS Institute Inc, Cary, NC). We used the resultant factor scores as surrogate exposure variables in predicting mortality outcomes with the Poisson regression model. Each factor was evaluated in a single source model. However, because the factor scores formed a set of orthogonal variables, we performed a separate regression analysis with all of the scores included in one multifactor model.
We also conducted a factor analysis on the daily concentrations of the chemical components of [PM.sub.CF] from samples collected by the dicot (Al, Si, Cl, S, K Ca, Mn, Fe, Zn, Br, Pb, Sr, Cu, and Rb). We did not use the scores from this analysis in the time-series analysis because the sampling period started in June 1996 and samples were only collected every third day.
Results
Table 4 shows the correlation coefficients between PM, gaseous pollutants, temperature, and relative humidity for Phoenix in 1995-1997. [PM.sub.2.5] (obtained from the DFPSS) was highly correlated with CO (r = 0.85) and [NO.sub.2] (r = 0.79), but less so with [SO.sub.2] (r =0.43). [PM.sub.2.5] from the DFPSS was highly correlated with that measured with the TEOM (r = 0.93). Table 5 shows the correlation coefficients between selected chemical composition components of [PM.sub.2.5] and the other air pollutants. TEOM [PM.sub.10] was correlated with fine soil (r = 0.72), OC (r = 0.58), EC (r = 0.58), and TC (r = 0.59). TEOM [PM.sub.2.5] was highly correlated with OC (r = 0.89), EC (r = 0.84), TC (r = 0.90), and to a lesser extent with Zn (r = 0.61), Pb (r = 0.67), and [K.sub.S] (r = 0.59). The high correlation coefficients between carbon and [PM.sub.2.5] indicate that the majority of the variation in [PM.sub.2.5] is due to combustion products. [PM.sub.CF] was correlated with soil (r = 0.66).
Table 4. Correlation coefficients between PM, gaseous pollutants, temperature, and relative humidity (RH) for Phoenix, 1995-1997.
[PM.sub.10]
[PM.sub.2.5](a) Temp (TEOM)
[PM.sub.2.5](a) 1.00 -0.31 0.69
Temp - 1.00 -0.08
[PM.sub.10] (TEOM) - - 1.00
RH - - -
[PM.sub.2.5] (TEOM) - - -
[PM.sub.CF] (TEOM) - - -
CO - - -
[NO.sub.2] - - -
[O.sub.3] - - -
[SO.sub.2] - - -
[PM.sub.2.5] [PM.sub.CF]
RH (TEOM) (TEOM) CO
[PM.sub.2.5](a) 0.16 0.93 0.50 0.85
Temp -0.55 -0.25 0.00 -0.49
[PM.sub.10] (TEOM) -0.12 0.77 0.97 0.53
RH 1.00 0.09 -0.19 0.23
[PM.sub.2.5] (TEOM) - 1.00 0.59 0.82
[PM.sub.CF] (TEOM) - - 1.00 0.34
CO - - - 1.00
[NO.sub.2] - - - -
[O.sub.3] - - - -
[SO.sub.2] - - - -
[NO.sub.2] [O.sub.3max] [SO.sub.2]
[PM.sub.2.5](a) 0.79 -0.24 0.43
Temp -0.40 0.71 -0.38
[PM.sub.10] (TEOM) 0.53 -0.12 0.41
RH 0.08 -0.54 0.10
[PM.sub.2.5] (TEOM) 0.77 -0.20 0.48
[PM.sub.CF] (TEOM) 0.37 -0.08 0.33
CO 0.87 -0.40 0.53
[NO.sub.2] 1.00 -0.24 0.57
[O.sub.3] - 1.00 -0.37
[SO.sub.2] - - 1.00
(a) Measured with the DFPSS.
Table 5. Correlation coefficient matrix of air pollutants.
S Zn Pb OC EC TC
S 1.00 0.14 0.25 0.12 0.04 0.10
Zn - 1.00 0.63 0.62 0.71 0.65
Pb - - 1.00 0.69 0.69 0.71
OC - - - 1.00 0.91 0.99
EC - - - - 1.00 0.95
TC - - - - - 1.00
[K.sub.S] - - - - - -
[PM.sub.10](a) - - - - - -
[PM.sub.2.5](a) - - - - - -
[PM.sub.CF](a) - - - - - -
Nonsoil [PM.sub.2.5] - - - - - -
Soil - - - - - -
CO - - - - - -
[NO.sub.2] - - - - - -
[O.sub.3] - - - - - -
[O.sub.3] max - - - - - -
[SO.sub.2] - - - - - -
[K.sub.S] [PM.sub.10] [PM.sub.2.5]
S 0.02 0.19 0.27
Zn 0.30 0.46 0.61
Pb 0.39 0.48 0.67
OC 0.65 0.58 0.89
EC 0.57 0.58 0.84
TC 0.64 0.59 0.90
[K.sub.S] 1.00 0.34 0.59
[PM.sub.10](a) - 1.00 0.79
[PM.sub.2.5](a) - - 1.00
[PM.sub.CF](a) - - -
Nonsoil [PM.sub.2.5] - - -
Soil - - -
CO - - -
[NO.sub.2] - - -
[O.sub.3] - - -
[O.sub.3] max - - -
[SO.sub.2] - - -
Nonsoil
[PM.sub.CF] [PM.sub.2.5] Soil CO
S 0.13 0.26 0.25 0.01
Zn 0.33 0.63 0.49 0.65
Pb 0.34 0.71 0.49 0.71
OC 0.38 0.96 0.52 0.89
EC 0.40 0.89 0.52 0.90
TC 0.39 0.96 0.53 0.91
[K.sub.S] 0.19 0.64 0.26 0.52
[PM.sub.10](a) 0.97 0.62 0.72 0.55
[PM.sub.2.5](a) 0.60 0.91 0.64 0.82
[PM.sub.CF](a) 1.00 0.41 0.66 0.37
Nonsoil [PM.sub.2.5] - 1.00 0.54 0.87
Soil - - 1.00 0.48
CO - - - 1.00
[NO.sub.2] - - - -
[O.sub.3] - - - -
[O.sub.3] max - - - -
[SO.sub.2] - - - -
[NO.sub.2] [O.sub.3] [O.sub.3max]
S 0.04 0.13 0.31
Zn 0.62 -0.49 -0.27
Pb 0.63 -0.51 -0.30
OC 0.81 -0.57 -0.32
EC 0.82 -0.64 -0.41
TC 0.83 -0.60 -0.35
[K.sub.S] 0.45 -0.27 -0.14
[PM.sub.10](a) 0.56 -0.25 -0.11
[PM.sub.2.5](a) 0.77 -0.44 -0.19
[PM.sub.CF](a) 0.39 -0.14 -0.07
Nonsoil [PM.sub.2.5] 0.80 -0.54 -0.29
Soil 0.49 -0.17 0.05
CO 0.87 -0.68 -0.39
[NO.sub.2] 1.00 -0.60 -0.24
[O.sub.3] - 1.00 0.81
[O.sub.3] max - - 1.00
[SO.sub.2] - - -
[SO.sub.2]
S -0.07
Zn 0.26
Pb 0.33
OC 0.49
EC 0.46
TC 0.49
[K.sub.S] 0.25
[PM.sub.10](a) 0.42
[PM.sub.2.5](a) 0.47
[PM.sub.CF](a) 0.35
Nonsoil [PM.sub.2.5] 0.46
Soil 0.09
CO 0.51
[NO.sub.2] 0.56
[O.sub.3] -0.46
[O.sub.3] max -0.37
[SO.sub.2] 1.00
(a) Based on TEOM measurements.
OC and EC concentrations follow a seasonal pattern--they are high in the colder months and low in the warmer months. This pattern is due to increased com-bustion emissions from space heating and the decreased mixing height during the winter months. Particulate sulfur concentrations peak in the warmer months. Soil concentration also follows a seasonal trend, with higher concentrations in the spring and fall. Measured soil concentrations decreased after 20 December 1996 because of the use of the WINS inlet.
Summaries of the RR between the exposure variables and both total and cardiovascular mortality are presented in Tables 6 and 7, respectively. Because of space limitations, we only present statistically significant (p [is less than] 0.05) and marginally significant (p [is less than] 0.10) results in the tables, although models were run using all of the pollutants listed in Table 5. Tables of all of the nonsignificant results are available from the authors by request. We evaluated the associations between total and cardiovascular mortality and the gaseous pollutants, PM mass metrics, and PM composition metrics using single-pollutant models.
Table 6. RR for total mortality in Phoenix from an interquartile range (IQR) increase in pollutants.
Pollutant Lag days [Beta]
CO 0 4.50 x [10.sup.-2]
1 4.15 x [10.sup.-2]
[NO.sub.2] 0 2.64 x [10.sup.0]
1 3.29 x [10.sup.0]
3 1.80 x [10.sup.0]
4 2.20 x [10.sup.0]
[SO.sub.2] 0 1.17 x [10.sup.-2]
S 3 -1.38 x [10.sup.-4]
4 -1.10 x [10.sup.-4]
Soil 1 -1.75 x [10.sup.-5]
2 -1.76 x [10.sup.-5]
3 -1.75 x [10.sup.-5]
4 -1.47 x [10.sup.-5]
[PM.sub.10] (TEOM) 0 1.06 x [10.sup.-3]
[PM.sub.CF] (TEOM) 0 1.17 x [10.sup.-3]
Pb 3 -2.70 x [10.sup.-3]
Pollutant Lag days SE t
CO 0 1.48 x [10.sup.-2] 3.05
1 1.48 x [10.sup.-2] 2.81
[NO.sub.2] 0 1.15 x [10.sup.0] 2.31
1 1.13 x [10.sup.0] 2.91
3 1.06 x [10.sup.0] 1.69
4 1.07 x [10.sup.0] 2.05
[SO.sub.2] 0 6.37 x [10.sup.-3] 1.84
S 3 6.24 x [10.sup.-5] -2.21
4 6.10 x [10.sup.-5] -1.80
Soil 1 8.67 x [10.sup.-6] -2.01
2 8.59 x [10.sup.-6] -2.05
3 8.56 x [10.sup.-6] -2.04
4 8.54 x [10.sup.-6] -1.72
[PM.sub.10] (TEOM) 0 5.35 x [10.sup.-4] 1.98
[PM.sub.CF] (TEOM) 0 6.99 x [10.sup.-4] 1.68
Pb 3 1.59 x [10.sup.-3] -1.69
Pollutant Lag days IQR RR LCI UCI
CO 0 1.19 1.06 1.02 1.09
1 1.19 1.05 1.01 1.09
[NO.sub.2] 0 0.02 1.05 1.01 1.10
1 0.02 1.07 1.02 1.12
3 0.02 1.04 0.99 1.08
4 0.02 1.04 1.00 1.09
[SO.sub.2] 0 2.78 1.03 1.00 1.07
S 3 280.60 0.96 0.93 1.00
4 279.90 0.97 0.94 1.00
Soil 1 1,767.45 0.97 0.94 1.00
2 1,769.33 0.97 0.94 1.00
3 1,772.48 0.97 0.94 1.00
4 1,775.62 0.97 0.95 1.00
[PM.sub.10] (TEOM) 0 24.88 1.03 1.00 1.05
[PM.sub.CF] (TEOM) 0 18.39 1.02 1.00 1.05
Pb 3 6.00 0.98 0.97 1.00
Abbreviations: [Beta], regression coefficient; LCI, lower 95% confidence interval; t, t-statistic from the regression model; UCI, upper 95% confidence interval.
Table 7. RR for cardiovascular mortality from an interquartile range (IQR) increase in pollutants.
Lag
Pollutant days [Beta] SE
CO 0 4.49 x [10.sup.-2] 2.14 x [10.sup.-2]
1 7.66 x [10.sup.-2] 2.07 x [10.sup.-2]
2 5.79 x [10.sup.-2] 2.00 x [10.sup.-2]
3 5.32 x [10.sup.-2] 2.03 x [10.sup.-2]
4 6.43 x [10.sup.-2] 2.06 x [10.sup.-2]
[NO.sub.2] 1 4.88 x [10.sup.0] 1.59 x [10.sup.0]
2 2.53 x [10.sup.0] 1.54 x [10.sup.0]
3 2.76 x [10.sup.0] 1.55 x [10.sup.0]
4 5.74 x [10.sup.0] 1.57 x [10.sup.0]
[SO.sub.2] 2 1.63 x [10.sup.-2] 8.64 x [10.sup.-3]
3 1.85 x [10.sup.-2] 8.65 x [10.sup.-3]
4 2.49 x [10.sup.-2] 8.58 x [10.sup.-3]
[K.sub.S] 3 5.81 x [10.sup.-4] 2.96 x [10.sup.-4]
[PM.sub.10] 0 1.88 x [10.sup.-3] 7.66 x [10.sup.-4]
(TEOM) 1 1.47 x [10.sup.-3] 7.56 x [10.sup.-4]
[PM.sub.2.5] 0 3.91 x [10.sup.-3] 2.38 x [10.sup.-3]
(TEOM) 1 6.85 x [10.sup.-3] 2.36 x [10.sup.-3]
3 4.86 x [10.sup.-3] 2.35 x [10.sup.-3]
4 5.43 x [10.sup.-3] 2.35 x [10.sup.-3]
[PM.sub.CF] 0 2.50 x [10.sup.-3] 9.88 x [10.sup.-4]
1 1.62 x [10.sup.-3] 9.78 x [10.sup.-4]
Nonsoil 1 5.56 x [10.sup.-6] 3.12 x [10.sup.-6]
[PM.sub.2.5]
OC 1 1.46 x [10.sup.-5] 6.82 x [10.sup.-6]
3 1.39 x [10.sup.-5] 6.89 x [10.sup.-6]
EC 1 4.40 x [10.sup.-5] 1.82 x [10.sup.-5]
TC 1 1.15 x [10.sup.-5] 5.05 x [10.sup.-6]
3 9.71 x [10.sup.-6] 5.10 x [10.sup.-6]
Pollutant t IQR RR LCI UCI
CO 2.10 1.19 1.05 1.00 1.11
3.71 1.19 1.10 1.04 1.15
2.89 1.19 1.07 1.02 1.12
2.63 1.19 1.07 1.02 1.12
3.12 1.19 1.08 1.03 1.13
[NO.sub.2] 3.08 0.02 1.10 1.04 1.17
1.64 0.02 1.05 0.99 1.12
1.78 0.02 1.06 0.99 1.12
3.66 0.02 1.12 1.05 1.19
[SO.sub.2] 1.88 2.78 1.05 1.00 1.10
2.14 2.79 1.05 1.00 1.10
2.90 2.79 1.07 1.02 1.12
[K.sub.S] 1.97 55.62 1.03 1.00 1.07
[PM.sub.10] 2.46 24.88 1.05 1.01 1.09
(TEOM) 1.95 24.88 1.04 1.00 1.08
[PM.sub.2.5] 1.64 8.52 1.03 0.99 1.08
(TEOM) 2.90 8.52 1.06 1.02 1.10
2.07 8.51 1.04 1.00 1.08
2.31 8.47 1.05 1.01 1.09
[PM.sub.CF] 2.54 18.39 1.05 1.01 1.09
1.66 18.39 1.03 0.99 1.07
Nonsoil 1.78 6,601.06 1.04 1.00 1.08
[PM.sub.2.5]
OC 2.15 2,976.50 1.04 1.00 1.09
2.02 2,960.00 1.04 1.00 1.08
EC 2.42 1,165.50 1.05 1.01 1.10
TC 2.28 4,169.00 1.05 1.01 1.09
1.90 4,170.00 1.04 1.00 1.09
Abbreviations:
[Beta], regression coefficient;
LCI, lower 95% confidence interval;
t, t-statistic from the regression model;
UCI, upper 95% confidence interval.
We found significant associations between both mortality outcomes and selected gaseous air pollutants. CO and [NO.sub.2] were positively associated with total mortality at 0- and 1-day lags. There was evidence of a weak association with [SO.sub.2] at 0 days lag (p [is less than] 0.10). We found several strong associations with cardiovascular mortality. Cardiovascular mortality was positively associated with CO (0-4 days lag). This was the most consistent association because the association was significant for all 5 lag days. Statistically significant associations (p [is less than] 0.05) were also evident with [NO.sub.2] on lag days 1 and 4, although the association was weaker on lag days 2 and 3. Cardiovascular mortality was also associated with [SO.sub.2] (lag days 2, 3, and 4).
We also found significant associations between the mortality outcomes and particulate mass. The associations between [PM.sub.10] and total mortality, and between [PM.sub.CF] and total mortality, were marginal (p [is less than] 0.10). Total mortality was not significantly associated with [PM.sub.2.5]; however, the RR was 1.02 (CI, 1.00-1.05). All PM mass metrics were associated with an excess risk of cardiovascular death. The strongest associations were with [PM.sub.2.5] (TEOM), followed by [PM.sub.10] and [PM.sub.CF]. [PM.sub.2.5] adjusted for soil content (nonsoil [PM.sub.2.5]) was also related with cardiovascular mortality with 1 day lag (p [is less than] 0.10). Table 7 lists all of the statistically significant associations with cardiovascular mortality. Cardiovascular mortality showed a more consistent association with particulate mass concentrations than total mortality. We further investigated the associations between the mortality outcomes and PM by evaluating the association between the mortality outcomes and the PM composition. The [PM.sub.2.5] composition data analysis revealed that EC and TC were significantly associated with cardiovascular mortality (1 day lag). Weaker associations were also evident with OC at 1 and 3 days lag and TC at 3 days lag. [K.sub.S] had a significant positive association with cardiovascular mortality (3 day lag).
We also found that soil, S, and Pb were negatively associated with total mortality. That is, these exposure variables were associated with a decrease in excess deaths.
We further evaluated the associations between the mortality outcomes and sources of both particulate and gas-phase pollutants using the scores from a factor analysis in place of the individual pollutant concentrations. The results from the analysis with five factors are presented in Table 8. Factor 1 probably represents the influence of motor vehicle exhaust and resuspended road dust with high loadings (loading [is greater than] 0.5) on Mn, Fe, Zn, Pb, OC, EC, CO, and [NO.sub.2]. Factor 2 represents soil with high loadings on Al, Si, and Fe. Factor 3 represents vegetative burning with high loadings on OC and [K.sub.S]. Factor 4 represents a local source of [SO.sub.2] with a high loading on [SO.sub.2]. Factor 5 represents predominately regional sulfate with a high loading on S. The RRs associated with an interquartile range increase in each factor are presented in Table 9. Total mortality had both a positive and a negative association with the factor representing regional sulfate, positive on lag day 0 (same day) and negative on lag day 3. The factor representing [SO.sub.2] had a negative association with total mortality. We also found a significant negative association for fine soil on lag days 1 and 2, and a nearly significant negative association on lag days 3 and 4. Cardiovascular disease was significantly associated with the factors representing motor vehicles (lag day 1) and vegetative burning (lag day 3). Regression analysis with all of the factors included in a multisource model produced similar results.
Table 8. Loadings from factor analysis.
Factor Factor Factor Factor Factor
1 2 3 4 5
Al 0.14 0.96 0.08 -0.01 0.07
Si 0.19 0.96 0.11 -0.01 0.10
S 0.04 0.15 0.01 -0.03 0.96
Ca 0.26 0.93 0.15 -0.01 0.09
Mn 0.66 0.62 0.05 0.13 0.07
Fe 0.57 0.76 0.19 0.19 0.05
Zn 0.86 0.24 0.03 -0.03 0.03
Br 0.46 0.31 0.59 0.01 0.28
Pb 0.74 0.21 0.25 0.12 0.26
OC 0.66 0.23 0.55 0.33 0.01
EC 0.76 0.25 0.42 0.28 -0.08
[K.sub.S] 0.20 0.08 0.92 0.08 -0.04
CO 0.76 0.20 0.39 0.35 -0.09
[NO.sub.2] 0.69 0.24 0.31 0.45 -0.05
[SO.sub.2] 0.24 -0.04 0.09 0.93 -0.02
Percent 30.5 27.5 13.7 9.7 7.4
variance
explained
by factor
Table 9. RR for total and cardiovascular mortality from an interquartile range (IQR) increase in each factor.
Lag
Outcome, factor days [Beta] SE t
Total mortality
Factor 2 1 -0.03 0.01 -2.03
2 -0.04 0.01 -2.45
3 -0.02 0.01 -1.67
4 -0.02 0.01 -1.74
Factor 4 2 -0.03 0.01 -2.01
4 -0.03 0.01 -1.72
Factor 5 0 0.03 0.01 2.23
3 -0.03 0.01 -2.22
Cardiovascular
mortality
Factor 1 1 0.05 0.02 2.59
Factor 3 3 0.05 0.02 2.67
Factor 5 0 0.04 0.02 2.03
Outcome, factor IQR RR LCI UCI
Total mortality
Factor 2 1.26 0.96 0.93 1.00
1.26 0.96 0.92 0.99
1.26 0.97 0.94 1.01
1.26 0.97 0.94 1.00
Factor 4 1.09 0.97 0.94 1.00
1.09 0.97 0.94 1.00
Factor 5 1.38 1.04 1.01 1.08
1.39 0.96 0.92 0.99
Cardiovascular
mortality
Factor 1 1.11 1.06 1.01 1.10
Factor 3 1.02 1.05 1.01 1.09
Factor 5 1.38 1.06 1.00 1.12
Abbreviations:
[Beta], regression coefficient;
LCI, lower 95% confidence interval;
t, t-statistic from the regression model;
UCI, upper 95% confidence interval.
Table 10 presents the results from the factor analysis on the daily concentrations of the chemical components of [PM.sub.CF] from samples collected by the dichotomous sampler. Factor 1 represents soil with high loadings on Al, Si, K, Ca, Mn, Fe, Sr, and Rb. Factor 2 represents a source of coarse fraction metals with high loadings on Zn, Pb, and Cu. Factor 13 represents a marine influence with a high loading on Cl. These three factors explain 91.8% of the variance in the [PM.sub.CF] data.
Table 10. Factor analysis results for [PM.sub.CF].
Factor Factor Factor
Element 1 2 3
Al 0.91 0.33 0.22
Si 0.90 0.36 0.24
Cl 0.25 -0.35 0.82
S 0.59 0.55 0.41
K 0.91 0.33 0.23
Ca 0.84 0.41 0.31
Mn 0.88 0.42 0.17
Fe 0.84 0.50 0.19
Zn 0.47 0.83 0.07
Br 0.23 0.30 0.85
Pb 0.40 0.80 -0.02
Sr 0.83 0.42 0.28
Cu 0.41 0.82 -0.02
Rb 0.91 0.27 0.17
Percent 51.1 26.5 14.2
variance
explained
by factor
Sensitivity analysis. As a sensitivity analysis, we analyzed temperature as a cofactor rather than a confounder. That is, we evaluated the effects of temperature on mortality as an independent variable rather than adjusting for it in the model as a confounding variable. We evaluated the significance of temperature after adjusting for day of the week, time trends, and relative humidity. For total and cardiovascular mortality, we found that temperature was not associated with excess deaths. Temperature was not correlated with either [PM.sub.10] (r = -0.08) or [PM.sub.2.5] (r = -0.25). A second analysis examined the effect of extreme temperatures. If the average daily temperature was greater than or equal to the 95th percentile (35.4 [degrees] C), we assigned a 1 to the predictor variable; otherwise we assigned a 0. We did not find an association between extreme temperature and total mortality. However, with cardiovascular mortality, extreme temperature was associated with excess deaths at 0 and 2 days lag (p [is less than] 0.1). To further assess the importance of the high temperature days to our analysis, we evaluated the association between [PM.sub.2.5] and cardiovascular mortality after excluding the days when the temperature was above the 95th percentile. The effect of eliminating the high temperature days was negligible. The RR for cardiovascular mortality associated with [PM.sub.2.5] (1 day lag) including all days was the same as that excluding the hottest days (RR = 1.06; CI, 1.02-1.10).
We also conducted a sensitivity analysis with relative humidity as a cofactor, with the model controlling for time trends and temperature. As a cofactor, relative humidity was not associated with either total mortality or cardiovascular mortality. To further assess the effects of extreme relative humidity, we eliminated the driest days (relative humidity [is less than] 25th percentile) from the data. We then found that the coarse fraction was no longer associated with total mortality. The association between cardiovascular mortality and coarse fraction was statistically significant (p [is less than] 0.05) on the concurrent day, but nonsignificant with 1-day lag.
We also used dew point rather than relative humidity in the base model. Controlling for dew point rather than relative humidity did not alter our results. We obtained similar regression coefficients.
To assess the effect of replacing the [PM.sub.2.5] cyclone on the TEOM with the WINS, we evaluated the association between soil and total mortality from 1 January 1995 to 31 December 1996 and from 1 January 1997 to 31 December 1997. The latter period represented the WINS inlet measurements. The association between soil and mortality was not significant for the cyclone measurements alone. Analysis with only the WINS data revealed that the association between soil and mortality was positive and significant at 0 days lag, but not significant for any of the other days.
We estimated soil-related potassium using a correction ratio = K/Si (23). We then reevaluated the RR for cardiovascular mortality associated with [K.sub.S] using [K.sub.S] calculated from three slightly different values of K/Si. This correction ratio is dependent on where the soil was obtained: [PM.sub.2.5] paved road dust (K/Si = 1.85/13.69), an agricultural field (K/Si = 1.98/14.35), or Phoenix desert soil (K/Si = 1.89/14.00) (27). We found similar RRs for cardiovascular mortality associated with [K.sub.S] when we used any of these three approaches. In contrast, total potassium was not associated with either total or cardiovascular mortality.
Discussion
To our knowledge this is the first time-series analysis that has looked at the association between PM chemical composition and mortality and the association between the underlying factors influencing that composition and mortality. Ozkaynak and Thurston (28) combined source apportionment and epidemiologic methods to assess the effects of air pollution on mortality. However, their study was a cross-sectional analysis rather than a time--series analysis. The present study found significant associations between air pollutants and total nonaccidental and cardiovascular mortality. The association between [PM.sub.10] and cardiovascular mortality is consistent with previous studies. Zmirou et al. (17) reported an RR for cardiovascular mortality from a 50-[micro]g/[m.sup.3] increase in [PM.sub.10] (RR = 1.04) in a study of air pollution in 10 large Western European cities. Pope et al. (5) found an association between respiratory disease death and cardiovascular deaths with [PM.sub.10] in Utah. Schwartz (2) also found that on high-pollution days (increased total suspended particulates) there was an increased risk of death from cardiovascular disease (RR = 1.09) in Philadelphia, Pennsylvania, and Birmingham, Alabama (1). Furthermore, Anderson et al. (29) found that black smoke was associated with a 0.58% increase in cardiovascular deaths in London.
The association between [PM.sub.2.5] and cardiovascular mortality is similar to that of Schwartz et al. (3), who found that a 10-[micro]g/[m.sup.3] increase in [PM.sub.2.5] was associated with a 1.5% increase in total mortality and 2.1% increase in mortality from ischemic heart disease in a study of six eastern U.S. cities. In contrast to Schwartz et al. (3), the present study also found a significant association between [PM.sub.CF] and total and cardiovascular mortality. Although Schwartz et al. (3) did not find a significant association between coarse fraction and mortality when the results from all six cities were combined, there was an association in Steubenville, Ohio, alone. Such observed differences may have been due to differences in regional coarse fraction composition. In Spokane, Schwartz et al. (12) also found no association between coarse particle concentration and total mortality. However, that study only looked at high episodes of coarse particle concentrations resulting from dust storms. Our findings are in agreement with Ostro et al. (13), who found a significant association between daily [PM.sub.10] dominated by coarse particles and mortality.
We investigated the possibility that [PM.sub.CF] was a surrogate for dryness by eliminating the days with humidity less than the 25th percentile. Although the association with total mortality was no longer significant, we found a significant association with cardiovascular mortality.
The reason for the negative association between soil and total mortality is unclear. One possible explanation for this observation is related to the fact that the [PM.sub.2.5] cyclone on the DFPSS was replaced with a WINS on 20 December 1996. The sharper cut point reduced the amount of soil intrusion into the [PM.sub.2.5] sample, which could produce soil data that are essentially different between the 1995-1996 period and 1997. To assess this effect, we eliminated all 1997 soil data and reevaluated the RR for total mortality. After removing the WINS data, the association between reconstructed soil and total mortality was not significant. These observations are similar to that of Ozkaynak and Thurston (28), who in a study of the association between U.S. mortality rates and particle pollution levels in 1980 found that soil was the least significant predictor of mortality. We also evaluated the association between soil and mortality with only soil data obtained with the WINS. The association was positive and significant (p [is less than] 0.05) on the concurrent day, but not significant on any other lag days. However, this observation may be due to the low number of days used to evaluate the association between WINS [PM.sub.2.5] soil and total mortality (n = 377).
With respect to the elemental components of [PM.sub.2.5], we found that EC was significantly associated with cardiovascular mortality. EC is found in combustion-derived particles, most notably diesel exhaust (21). We found that Ks [potassium from vegetative burning (23)] was also associated with cardiovascular mortality.
We found several associations that are potentially spurious. The associations with these variables were found with only total mortality and not with cardiovascular mortality. Lead was negatively associated with total mortality at lag day 3, although this may be reflective of the moderate correlation between Pb and soil in Phoenix (r = 0.49). Pb may have accumulated in the soil or in road dust from the past use of leaded gasoline. S was also negatively associated with total mortality on lag day 3. At present, the reason for the negative association with S is unclear. However, S accounts for a rela-tively small percentage of the mass of [PM.sub.2.5] (15%). The significant negative associations between total mortality and Pb and S were not consistent with the lack of association between these exposure variables and cardiovascular mortality.
For the gaseous species, we found that total nonaccidental mortality and cardiovascular mortality were strongly associated with CO and [NO.sub.2]. These observations are similar to those of Burnett et al. (30), who found associations between CO and [NO.sub.2] and total nonaccidental mortality in Toronto, Canada. Burnett et al. (30) also found that cardiac mortality was associated with CO. CO exacerbates cardiac conditions (10). CO concentrations are also associated with hospital admissions for cardiovascular disease (31). In Phoenix the primary sources of CO and [NO.sub.2] are motor vehicles.
The association between [SO.sub.2] and cardiovascular mortality was similar to that of Zmirou et al. (17), who also found that an increase in [SO.sub.2] was associated with an increase in cardiovascular deaths (RR = 1.04). In addition, Zmirou et al. (17) found weak but significant association between 1-hr maximum [O.sub.3] concentrations and cardiovascular mortality (RR = 1.02). Hoek et al. (32) also found an association between total mortality and O.sub.3] in the Netherlands. We found no significant associations with [O.sub.3].
The present study demonstrated the use of factor analysis in an epidemiologic study. Using factor analysis, we were able to identify those underlying factors of measured air pollution composition variability that were associated with excess mortality. Poisson regression with factor scores as exposure variables revealed that combustion-related pollutants associated with motor vehicles and vegetative burning as well as fine particulate [SO.sub.4] concentrations were significantly associated with cardiovascular mortality. The soil factor, however, was associated with fewer than expected total deaths. These results are consistent with our time-series results for individual pollutants, specifically CO, [NO.sub.2], [K.sub.S], EC, OC, and reconstructed soil. It is interesting to note that the factor repre-senting S was significantly associated with cardiovascular mortality, whereas S alone in an individual pollutant model was not associated with cardiovascular mortality. This may be reflective of the contribution of Pb and Br to the S factor.
A unique aspect of this study was the use of the chemical composition data of [PM.sub.2.5] Using such data, we found positive associations between cardiovascular mortality and [K.sub.S], OC, and EC as well as the more traditionally measured pollutants CO, [NO.sub.2], [SO.sub.2], [PM.sub.10], [PM.sub.2.5], and [PM.sub.CF]. Significant associations were also found with factors associated with incomplete combustion products and particulate S compounds. A limitation of this study is that the factor analysis results are only in terms of the variance explained by each factor, rather than in terms of the quantitative contribution from a specific source category. Although methods are available to include quantitative source apportionments in a time-series framework (33), such an analysis is beyond the scope of this initial investigation.
REFERENCES AND NOTES
(1.) Schwartz J. Air pollution and daily mortality in Birmingham, Alabama. Am J Epidemiol 137:1136-1146 (1993).
(2.) Schwartz J. What are people dying of on high air pollution days? Environ Res 64:26-35 (1994).
(3.) Schwartz J, Dockery D, Neas L. Is daily mortality associated specifically with fine particles? J Air Waste Manag Assoc 46:927-939 (1996).
(4.) Fairley D. The relationship of daily mortality to suspended particulates in Santa Clara County, 1980-1986. Environ Health Perspect 89:159-168 (1990).
(5.) Pope CA, Schwartz J, Ransom MR. Daily mortality and [PM.sub.10] pollution in Utah Valley. Arch Environ Health 47:211-217 (1992).
(6.) Dockery DW, Pope CA III. Acute respirator effects of particulate air pollution. Annu Rev Public Health 15:107-132 (1994).
(7.) Ostro B. Fine particulate air pollution and mortality in two Southern California counties. Environ Res 70:98-104 (1995).
(8.) Vedel S. Ambient particles and health: lines that divide. J Air Waste Manag Assoc 47:551-581 (1997).
(9.) Wordley J, Walters S, Ayres JG. Short term variations in hospital admissions and mortality and particulate air pollution. Occup Environ Med 54:108-116 (1997).
(10.) Pope CA III, Hill RW, Villegas GM. Particulate air pollution and daily mortality on Utah's Wasatch Front. Environ Health Perspect 107:567-573 (1999).
(11.) Bascom R. Health effects of outdoor air pollution. Am J Respir Crit Care Med 153:477-498 (1996).
(12.) Schwartz J, Norris G, Larson T, Sheppard L, Claiborne C, Koenig J. Episodes of high coarse particle concentrations are not associated with increased mortality. Environ Health Perspect 107:339-342 (1999).
(13.) Ostro BD, Hurley S, Lipsett MJ. Air pollution and daily mortality in the Coachella Valley, California: a study of [PM.sub.10] dominated by coarse particles. Environ Res 81:231-238 (1999).
(14.) U.S. EPA. The Air Quality Criteria for Particulate Matter. EPA/600/P-95/001bF. Washington, DC:U.S. Environmental Protection Agency, 1996.
(15.) Norris GA, Zweidinger RB, Bloemen HJTh, Purdue LJ, Bowser J, Stevens RK. Unpublished data.
(16.) U.S. Census Bureau. U.S. Census 1990. Available: http://www.census.gov/cgi-bin/gazetteer [cited 18 January 2000].
(17.) Zmirou D, Schwartz J, Saez M, Zanobetti A, Wojtyniak B, Touloumi G, Spix C, Ponce de Leon A, Le Moullec Y, Bacharova L, et al. Time-series analysis of air pollution and cause-specific mortality. Epidemiology 9:495-503 (1998).
(18.) Liao D, Creason J, Shy C, Williams R, Watts R, Zweidinger R. Daily variation of particulate air pollution and poor cardiac autonomic control in the elderly. Environ Health Perspect 107:521-525 (1999).
(19.) Burnett RT, Cakmak S, Raizenne ME, Stieb D, Vincent R, Krewski D, Brook JR, Philips O, Ozkaynak H. The association between ambient carbon monoxide levels and daily mortality in Toronto, Canada. J Air Waste Manag Assoc 48:689-700 (1998).
(20.) Touloumi G, Katsouyanni K, Zmirou D, Schwartz J, Spix C, de Leon AP, Tobias A, Quennel P, Rabczenko D, Bacharova L, et al. Short-term effects of ambient oxidant exposure on mortality: a combined analysis within the APHEA project. Am J Epidemiol 146:177-185 (1997).
(21.) Birch ME, Cary RA Elemental carbon-based method for monitoring occupational exposures to particulate diesel exhaust. Aerosol Sci Technol 25:221-241 (1996).
(22.) U.S. Environmental Protection Agency. Aerometric Information Retrieval System. Available: http:// www.epa.gov/airs/airs.html [cited 14 December 1999].
(23.) Lewis CW, Baumgardner RE, Stevens RK, Russworm GM. Receptor modeling study of Denver winter haze. Environ Sci Technol 20:1126-1136 (1986).
(24.) Maim WC, Sisler JF, Huffman D, Eldred RA, Cahill TA. Spatial and seasonal trends in particle concentration and optical extinction in the United States. J Geophys Res 99:1347-1370 (1994).
(25.) Hastie T, Tibshirani R. Generalized Additive Models. London:Chapman and Hall, 1990.
(26.) Akaike H. Statistic predictor identification. Ann Inst Stat Math 22:203-217 (1970).
(27.) Chow JC, Watson JG, Richards DW, Haase DL, McDade C, Dietrich DL, Moon D, Sloane C. The 1989-90 Phoenix [PM.sub.10] Study, Vol II. Source Apportionment Appendices, DRI Document No. 8931 6F2. Reno, NV:Desert Research Institute, 1991.
(28.) Ozkaynak H, Thurston G. Associations between 1980 U.S. mortality rates and alternative measures of airborne particle concentration. Risk Anal 7:449-461 (1987).
(29.) Anderson HR, Ponce de Leon A, Bland JM, Bower JS, Strachan DP. Air pollution and daily mortality in London: 1987-92. Br Med J 312:665-669 (1996).
(30.) Burnett RT, Smith-Doiron M, Stieb D, Cakmak S, Brook JR. Effects of particulate and gaseous air pollution on cardiorespiratory hospitalizations. Arch Environ Health 54:130-139 (1999).
(31.) Schwartz J. Air Pollution and Hospital Admissions for Cardiovascular Disease in Tucson. Epidemiology 8:371-377 (1997).
(32.) Hoek G, Schwartz JD, Groot B, Eilers P. Effects of ambient particulate matter and ozone on daily mortality in Rotterdam, the Netherlands. Arch Environ Health 52:455-463 (1997).
(33.) Henry RC, Lewis CW, Collins JF. Vehicle-related hydrocarbon source compositions from ambient data: the GRECE/SAFER Method. Environ Sci Technol 28:823-832 (1994).
Address correspondence to T.F Mar, Department of Environmental Health, Box 357234, University of Washington, Seattle, WA 98195-7234 USA. Telephone: (206) 685-1596. Fax: (206) 685-3990. E-mail: therese@u.washington.edu
We thank D. Bates for his advice and comments regarding the manuscript. We also thank T. Moore for advice regarding the representative spatial scale of the platform particulate matter measurements, C. Mrela for the mortality data, and B. O'Brian for help with assembling the data.
This publication was made possible in part by grant 5T32 ES07262 from the NIEHS, NIH. The U.S. EPA Office of Research and Development partially funded and collaborated in the research described here under assistance agreement R 827355 to the University of Washington.
The contents of this paper are solely the responsibility of the authors and do not necessarily represent the official views of the NIEHS, NIH. This paper has been subjected to EPA review and approved for publication. Mention of trade names or commercial products does not constitute an endorsement or recommendation for use.
Received 8 September 1999; accepted 9 November 1999.
Therese F. Mar,(1) Gary A. Norris,(2) Jane Q. Koenig,(1) and Timothy V. Larson(3)
(1) Department of Environmental Health, University of Washington, Seattle, Washington, USA
(2) U.S. Environmental Protection Agency, Research Triangle Park, North Carolina, USA
(3) Department of Civil and Environmental Engineering, University of Washington, Seattle, Washington, USA
COPYRIGHT 2000 National Institute of Environmental Health Sciences
COPYRIGHT 2004 Gale Group