Fine particulate air pollution and mortality in nine California counties: results from CALFINE.
Over the last decade, studies conducted over five continents have demonstrated associations between daily exposure to particulate matter (PM) < 10 [micro]m in aerodynamic diameter (P[M.sub.10]) and premature mortality [U.S. Environmental Protection Agency (EPA) 2004[. The U.S. EPA promulgated ambient air quality standards for fine particles [those < 2.5 [micro]m in diameter (P[M.sub.2.5])] in 1997 and is currently considering revisions to these standards; however, relatively few studies have examined relationships of this pollutant class with mortality (Burnett et al. 2003; Schwartz 2003; U.S. EPA 2004). In addition, most studies to date have been conducted in the eastern United States, Canada, and Western Europe. Relatively few studies have been conducted in California, where particle sources, chemistry, size distribution, and temporal patterns of exposure are quite different. Specifically, existing evidence suggests that, in California, a) nitrates comprise a larger fraction of P[M.sub.2.5] than they do in other regions, and b) mobile sources represent the predominant source of P[M.sub.2.5], whereas a mix of mobile and stationary sources predominate elsewhere (Blanchard 2003). Moreover, in the Los Angeles air basin, peak P[M.sub.2.5] exposures occur in both winter and nonwinter months.
In 1999, the U.S. EPA and the California Air Resources Board (CARB) embarked on a program to collect daily data on P[M.sub.2.5] in many cities throughout California. We have obtained and linked daily readings of P[M.sub.2.5] with mortality in nine heavily populated counties in California. The ability to explore hypotheses of association with adverse health in multiple cities has several distinct advantages. It enhances the power of the statistical analysis and reduces the likelihood of spurious results or publication bias that might result from the analysis of a single city (Anderson et al. 2005). In this article, we report the results of our analysis of the relationship between mortality and fine particles in California (CALFINE).
Materials and Methods
Mortality data. Data on daily mortality were obtained for all California residents from the California Department of Health Services (CDHS), Health Data and Statistics Branch, for the period 1 January 1999 through 31 December 2002 (CDHS 1999-2002). Our study was limited to deaths occurring in nine California counties (cities where the monitors were located are in parentheses): Contra Costa (Concord), Fresno (Fresno), Kern (Bakersfield), Los Angeles (Los Angeles, North Long Beach, Azusa), Orange (Anaheim), Riverside (Riverside), Sacramento (Sacramento), San Diego (San Diego, Escondido, El Cajun), and Santa Clara (two in San Jose). Data were limited to deaths occurring in the decedents' county of residence. Daily counts of total deaths (minus accidents and homicides) were aggregated. Using the International Classification of Diseases, 10th Revision (ICD-10) (World Health Organization 1993), total daily counts of deaths from respiratory disease (ICD-10 codes J00-J98), cardiovascular disease (ICD-10 codes I00-I99), ischemic heart disease (ICD-10 codes I20-I25), and diabetes (ICD-10 codes E10-E14) were also calculated.
We also calculated daily, all-cause mortality counts for the following subpopulations and mortality categories: a) age > 65 years, b) males, c) females, d) white non-Hispanic, e) black non-Hispanic, f) Hispanic, g) in hospital, h) out of hospital, i) less than high school education, and]') high school graduate.
Pollutant and meteorologic data. We obtained pollution data for the 4-year period 1999 through 2002 from multiple sources. Daily average PM2.5 data were obtained from the U.S. EPA's Aerometric Information Retrieval System (AIRS) database. P[M.sub.2.5] monitors were filter-based, ambient air samplers (model RAAS2.5-300; Andersen Instruments, Inc., Smyrna, GA).
This sequential sampler is designated as a federal reference method sampler for collection of P[M.sub.2.5]. There was only one monitor collecting daily P[M.sub.2.5] data in each of the nine counties, except for Los Angeles, San Diego, and Santa Clara counties, which had three, three, and two monitors, respectively. Data from the nine counties represent nearly all locations of monitors in California that measured P[M.sub.2.5] on a daily basis for large parts of 1999-2002. A substantial number of days were missing data, which varied by county and appeared to be fairly random, with a few exceptions. Specifically, in 1999 several of the counties had no data from January through March, and from March through December, Los Angeles and Riverside counties had data only every third day.
Data on gaseous pollutants, including carbon monoxide, nitrogen dioxide, and ozone, were obtained from the CARB air quality database for all nine counties. Most of the monitors for gases were part of the State and Local Air Monitoring Stations (SLAMS) network. All gases were reported as 24-hr averages, except ozone, which was reported as both an 8-hr average (1000-1800 hr) and as a 1-hr maximum.
For counties with multiple monitors, the daily average was calculated using all available data. To account for missing data among some of the monitors, we used a process similar to that described by Wong et al. (2001). The average was developed by a) calculating the mean for each monitor, b) subtracting the mean concentration of each monitor from the nonmissing daily values, c) calculating the mean of the available adjusted data, and d) adding back the grand mean of the data.
To allow adjustment for the effect of weather on mortality, we collected daily average temperature and humidity data at weather stations in each of the nine counties. Hourly temperature data were obtained from AIRS for all sites except Contra Costa and Santa Clara counties, for which data were obtained from the Bay Area Air Quality Management District and from Golden Gate Weather Services, respectively. All daily mortality, pollutant, and meteorologic data were converted into a SAS database (SAS Institute Inc., Cary, NC) and merged by date. This resulted in 4 years (1,461 days) of daily time-series data.
Methods. Counts of daily mortality are nonnegative discrete integers representing rare events; such data typically follow a Poisson distribution. Therefore, the analysis relied on Poisson regression, conditional on the explanatory variables. In the basic analytic approach, we used similar model specifications for each city, including smoothing spline functions for time trend and weather. We examined both penalized and natural spline models. The penalized spline model is a flexible, nonparametric approach using cubic splines and a term that penalizes the curvature of the smoothing function (Wood 2000). The "roughness penalty" controls the trade-off between a precise fit of the data and a smoothed function. The model then minimizes the sum of the squared deviations plus the penalty function to determine the amount of smoothing in the fit. The natural spline model is a parametric approach that fits piece-wise polynomial functions joined at knots, which are typically placed evenly throughout the distribution of the variable of concern, such as time. The function is constrained to be continuous at each knot (Ruppert et al. 2003). The model also places two additional knots at the ends of the data, with the function constrained to be linear beyond these points. The number of knots used determines the overall smoothness of the fit. Previous analysis has indicated that different spline models generate relatively similar results (Health Effects Institute 2003). However, depending on the underlying data and model specifications, different splines might produce varying degrees of bias and efficiency in the regression estimates.
For the initial analysis of all-cause, cardiovascular, respiratory, and above-age-65 mortality, a penalized spline regression was used with R (R Development Core Team 2004). We incorporated a smoothed spline function of time, which can accommodate nonlinear and nonmonotonic patterns between time and mortality, offering a flexible modeling tool (Hastie and Tibshirani 1990). In addition, the smooth of time diminishes short-term fluctuations in the data, thereby helping to reduce the degree of serial correlation. Based on previous findings reported in the literature (e.g., Samet et al. 2000), the basic model included a smoothing spline for time with 7 degrees of freedom (df) per year of data. This number of degrees of freedom controls well for seasonal patterns in mortality and reduces and often eliminates autocorrelation. Visual inspection of the data indicated a spike in mortality in several of the cities in southern and central California during a 3-week period starting 17 December 1999. During this period, the actual number of cases exceeded the smoothed estimate. Therefore, for all of the regression models, we added a second smooth of time with 3 knots for this 3-week period.
Other covariates, such as day of the week and smoothing splines of 1-day lags of average temperature and humidity (each with 3 df), were also included in the model because they may be associated with daily mortality and are likely to vary over time in concert with air pollution levels. Previous studies have reported stronger associations of mortality with PM lagged 1 or 2 days or with cumulative exposures over several days. Therefore, in our primary analysis of P[M.sub.2.5], we examined two different apriorilag structures: a 2-day average of lags 0 and 1 (lag 01) and a single-day lag of 2 days (lag 2). The county-specific results were then combined in a meta-analysis using a random effects model in Stata (StataCorp 2003). The meta-analysis focused primarily on all-cause mortality and on cardiovascular, respiratory, and elderly (> 65 years of age) mortality, because these categories have been the focus of previous time-series studies (Health Effects Institute 2003).
We also conducted several sensitivity analyses. First, we examined these same four outcomes using a similar specification, but with a natural spline model. For each county, we used lag 01 for P[M.sub.2.5] and 4, 8, and 12 df/year for the smooth of time. Second, using lag 01 and penalized spline models with 7 df for the smooth of time, we examined other mortality groupings and classifications, including those for males, females, whites, blacks, Hispanics, high school and non-high school graduates, deaths occurring in and out of hospitals, ischemic heart disease, and diabetes. Finally, we examined the impact on the estimated coefficient of P[M.sub.2.5] when gaseous pollutants were added to the penalized spline model (i.e., in two-pollutant models specified with P[M.sub.2.5] and each of the gaseous pollutants).
All final results were calculated using R (version 1.9), and the results are presented as the percent change in daily mortality per 10 [micro]g/[m.sup.3] P[M.sub.2.5]. The percent change per 10 [microg/[m.sup.3] is simply the [beta]-coefficient (times 1,000) from the Poisson regression.
Tables 1 and 2 provide the descriptive statistics for population, air quality, mortality, and meteorologic data from the nine counties. The populations in 2000 ranged from 661,645 in Kern County to 9,519,338 in Los Angeles County; the total in these nine counties accounted for 65% of California's population in 2000. Mean daily mortality varied from 146 in Los Angeles County to 11 in Kern County. Mean daily P[M.sub.2.5] levels ranged from 14 [micro]g/[m.sup.3] in Sacramento and Contra Costa Counties to 29 [micro]g/[m.sup.3] in Riverside County, exceeding the U.S. EPA annual average P[M.sub.2.5] standard of 15 [micro]g/[m.sup.3] in six of the nine counties. Temporally, among the cities, P[M.sub.2.5] was highly correlated with both nitrogen dioxide (mean r = 0.56; range, 0.38-0.66) and carbon monoxide (mean r = 0.60; range, 0.37-0.83), but only moderately and often inversely correlated with both 1-hr ozone levels (mean r = -0.14; range, -0.39 to 0.17) and 8-hr ozone levels (mean, -0.22; range, -0.47 to 0.12).
Table 3 summarizes the basic results for the meta-analyses for four mortality categories using penalized splines with two different lag structures. The results suggest associations between P[M.sub.2.5] and all-cause, cardiovascular, respiratory, and elderly mortality. Point estimates of risk were particularly elevated for respiratory-specific mortality. Also, cumulative exposures of 2 days generated larger pooled effect estimates than did the single-day lags that were examined. Diagnostics indicated that autocorrelation was present over the entire data series for many of the counties when a simple smooth of time was used. The autocorrelation was eliminated, however, when the second smooth of time was included for the 3-week period starting 17 December 1999.
Table 4 summarizes the results for the meta-analyses for four mortality categories when similar models were used with lag 01 for P[M.sub.2.5] and natural splines for the smoothers of temperature and humidity and three alternative smoothers of time. The results generally support, but are slightly lower than, those observed using penalized splines (Table 3), indicating associations with all-cause, respiratory, and elderly mortality and more modest associations with cardiovascular mortality. In addition, greater degrees of freedom for time trend tended to lower the effect estimates.
Table 5 summarizes the meta-analytic results for P[M.sub.2.5] for different mortality categories and subpopulations using a penalized spline model and lag 01. The results suggest somewhat stronger associations of daily P[M.sub.2.5] concentrations with mortality for diabetics, females, and whites. The association for deaths occurring outside of hospitals was demonstrated with greater precision than for those occurring inside hospitals. In addition, the point estimate for mortality among those who had not graduated from high school was more than twice that of those who had, with an association that was of marginal statistical significance (p < 0.10). Finally, in multipollutant models (using lag 01), the estimated P[M.sub.2.5] coefficient was attenuated when the highly correlated pollutants--nitrogen dioxide and carbon monoxide--were added to the model but was not affected by the inclusion of either 1-hr or 8-hr ozone. However, for mortality among those > 65 years of age, the inclusion of any of the gaseous pollutants to the model did not affect the P[M.sub.2.5] coefficient (data not shown).
In this time-series analysis in nine California counties, short-term exposures to P[M.sub.2.5] were associated with increased daily mortality. These results appear to be relatively insensitive to the use of natural versus penalized spline model and the degrees of freedom in the smoothing functions for time, although both of these factors alter the effect estimates. Specifically, P[M.sub.2.5] was associated with all-cause, cardiovascular, and respiratory mortality, as well as with deaths in persons > 65 years of age. P[M.sub.2.5]-mortality associations were particularly elevated among females, whites, persons who did not graduate from high school, diabetics, and those who died out of hospital.
Several earlier studies that examined associations between daily mortality and either P[M.sub.10] or P[M.sub.2.5] were reanalyzed for the Health Effects Institute (Health Effects Institute 2003). The reanalyses were conducted after the generalized additive models had been found to produce biased effect estimates and standard errors when default convergence criteria were used in S-Plus (Dominici et al. 2003). Regarding P[M.sub.2.5], Schwartz et al. (1996) found statistically significant increases in mortality in their reanalysis of the Six Cities study using both natural spline [1.29% per 10 [micro]g/[m.sup.3] P[M.sub.2.5]; 95% confidence interval (CI), 0.88-1.70] and penalized spline (1.13%; 95% CI, 0.70-1.56) models with 4 dr/year for time. Burnett et al. (2003) reexamined nonaccidental mortality from 1986 to 1996 in eight Canadian cities, using natural spline models with 2 dr/year for time, and reported a 1.10% increase in mortality (95% CI, 0.35-1.85) per 10 [micro]g/[m.sup.3] of P[M.sub.2.5]. A reanalysis of another Canadian study found a nonsignificant increase in mortality (0.46% per 10 [micro]g/[m.sup.3] P[M.sub.2.5]) in Montreal from 1984 to 1993 (Goldberg and Burnett 2003). In a reanalysis of a time-series study in Santa Clara, California, Fairley (2003) reported a 2.75% increase (95% CI, 0.61-4.89) in nonaccidental mortality per 10 [micro]g/[m.sup.3] P[M.sub.2.5] using a natural spline model with 9 df/year. The reanalyses of data from Detroit (Ito 2003) and Los Angeles (Moolgavkar 2003) using natural spline models demonstrated positive but nonsignificant increases in mortality of 0.79 and 0.55%, respectively, per 10 [micro]g/[m.sup.3] P[M.sub.2.5]. Finally, in a study in Mexico City, Mexico, P[M.sub.2.5] was associated with a 1.4% (95% CI, 0.2-2.5) increase in daily mortality per 10 [micro]g/[m.sup.3] (Borja-Aburto et al. 1998).
Our effect estimate of about 0.6% per 10 [micro]g/[m.sup.3] P[M.sub.2.5] for all-cause mortality is in the lower end of the range of these previous estimates. There are several possible explanations for the lower effect estimates. First, large exposure measurement errors were likely, owing to the use of one to three monitors to represent exposure in these counties, some of which extend over thousands of square miles.
Therefore, assuming such measurement errors were nondifferential with respect to the populations at risk, the effect estimates would likely be biased downward. Second, the composition of P[M.sub.2.5] in California, which in several of these counties is dominated by nitrates, may be less toxic, particularly to the cardiovascular system (Schlesinger and Cassee 2003). However, this hypothesis contrasts with the findings of one of the few studies to explicitly examine the effects of nitrates, which were associated with significant increases of mortality in Santa Clara County (Fairley 2003). Third, California residents may be less susceptible to the cardiovascular effects of air pollution, possibly due to differences in exercise and dietary patterns, or to active and passive smoking rates that are lower than national averages. Fourth, there may be geographic confounding related to some unknown and therefore unmeasured spatially varying factors. Finally, this could be a chance finding. The likely potential importance of measurement error, geographic confounding, and chance is suggested by the large variability in effect estimates among the nine counties. Such heterogeneity has also been reported in the analysis of the 90 largest U.S. cities (Samet et al. 2000). There is no obvious explanation for the different P[M.sub.2.5]-mortality associations in each county. This merits further study.
Of additional interest is the strength of the association of P[M.sub.2.5] with respiratory mortality relative to that for cardiovascular mortality. Many previous studies [reviewed by Ostro et al. (1999)] report stronger effects for cardiovascular mortality, which may be due to a) the greater prevalence of circulatory disease (and therefore increased statistical power) and b) the likely attribution of cause of death as cardiovascular when there is uncertainty or when there is an underlying respiratory condition. It is often more difficult to detect associations between air pollution and respiratory deaths because the latter generally represent a small fraction of total mortality and are more likely to be ascribed to cardiovascular causes than vice versa. However, it is clear that P[M.sub.2.5] and other PM metrics are associated with daily mortality from respiratory causes. For example, Penttinen et al. (2004), Zanobetti et al. (2003), Braga et al. (2001), and Ostro et al. (1999) all report stronger associations of PM with respiratory than with cardiovascular mortality. De Leon et al. (2003) reported that those with an underlying respiratory condition were more susceptible to the impacts of air pollution on nonrespiratory (e.g., circulatory or cancer-related) mortality. Associations have also been reported between P[M.sub.2.5] and respiratory morbidity, including hospitalizations and emergency department visits for respiratory disease (Delfino et al. 1997; Ito 2003; Peel et al. 2005).
Our analysis also suggests that diabetics and those with less than a high school education may be at increased risk from exposure to P[M.sub.2.5]. Several previous time-series studies have reported that diabetics may be at increased risk from exposure to PM (Goldberg et al. 2001; Zanobetti and Schwartz 2002). Pope et al. (2002) reported that educational attainment was an important effect modifier in the association between long-term exposure to P[M.sub.2.5] and survival. However, susceptibility to PM pollution is not likely to be affected by education per se, but rather by factors that might be associated with education, such as nutritional status, access to health care, occupation, psychosocial stress, and residential proximity to heavy traffic. On the other hand, most time-series studies to date have not reported a significant effect modification by socioeconomic status (Samet et al. 2000; Schwartz 2000). We also found, as have others, a better model fit for P[M.sub.2.5] for deaths occurring out of hospital (Schwartz 2000). We found that when copollutants highly correlated with P[M.sub.2.5] were included in the model, they tended to attenuate the magnitude and significance of its coefficient, except for mortality for those > 65 years of age. The latter finding suggests that, at least for deaths occurring in the elderly, gaseous copollutants do not confound the P[M.sub.2.5]-mortality associations. The gaseous pollutants, however, are spatially heterogeneous and may involve significant exposure misclassification. The separate effects of the gaseous pollutants on mortality will be the focus of subsequent analyses.
Overall, this large, multicounty analysis provides evidence of significant associations of P[M.sub.2.5] with daily mortality among nearly two-thirds of California's population.
We thank F. Forastiere and M. Stafoggia for their technical assistance.
The opinions expressed in this article are solely those of the authors and do not represent the policy or position of the State of California or the California Environmental Protection Agency.
Anderson HR, Atkinson RW, Peacock JL, Sweeting M J, Marston L. 2005 Ambient particulate matter and health effects. Publication bias in studies of short-term associations. Epidemiology 16:155-163.
Blanchard C. 2003. Spatial and temporal characterization of particulate matter. In: Particulate Matter Science for Policy Makers: A NARSTO Assessment (McMurry PH, Shepherd MF, Vickery JS, eds). Cambridge, UK:Cambridge University Press, 191-231.
Borja-Aburto VH, Castiilejos M, Gold DR, Bierzwinski S, Loomis D. 1998. Mortality and ambient fine particles in southwest Mexico City, 1993-1995. Environ Health Perspect 106:849-855.
Brage AL, Zanobetti A, Schwartz J. 2001. The lag structure between particulate air pollution and respiratory and cardiovascular deaths in 10 US cities. J Occup Environ Med 43:927-933.
Burnett RT, Goldberg MS. 2003. Size-fractionated particulate mass and daily mortality in eight Canadian cities. In: Revised Analyses of Time-Series Studies of Air Pollution and Health. Special Report. Boston:Health Effects Institute, 85-90.
CDHS [California Department of Health Services). 1999-2002. Death Statistical Master Files, 1999-2002. Sacramento, CA:Center for Health Statistics.
De Leon SF, Thurston GD, Ito K. 2003. Contribution of respiratory disease to nonrespiratory mortality associations with air pollution. Am J Respir Crit Care Med 167:1117-1123.
Delfino RJ, Murphy-Moulton AM, Burnett RT, Brook JR, Becklake MR. 1997. Effects of air pollution on emergency room visits for respiratory illnesses in Montreal, Quebec. Am J Respir Crit Care Med 155:568-576.
Dominici F, Daniels M, McDermott A, Zeger SL, Samet JM. 2003. Shape of the exposure-response relation and mortality displacement in the NMMAPS database. In: Revised Analyses of Time-Series Studies of Air Pollution and Health. Special Report. Boston:Health Effects Institute, 91-96.
Fairley D. 2003. Mortality and air pollution for Santa Clara County, California, 1989-1996. In: Revised Analyses of Time-Series Studies of Air Pollution and Health. Special Report. Boston:Health Effects Institute, 97-106.
Goldberg MS, Burnett RT. 2003. Revised analysis of the Montreal time-series study. In: Revised Analyses of Time-Series Studies of Air Pollution and Health. Special Report. Boston:Health Effects Institute, 113-132.
Goldberg MS, Burnett RT, Bailar JC III, Brook J, Bonvalot Y, Tamblyn R, et al. The association between daily mortality and ambient air particle pollution in Montreal, Quebec. 2. Cause-specific mortality. Environ Res 86:26-36.
Hastie TJ, Tibshirani RJ. 1990. Generalized Additive Models. London:Chapman & Hall.
Health Effects Institute. 2003. Revised analyses of the National Morbidity, Mortality, and Air Pollution Study (NMMAPS), part II. In: Revised Analyses of Time-Series Studies of Air Pollution and Health. Special Report. Boston:Health Effects Institute, 9-72.
Ito K. 2003. Associations of particulate matter components with daily mortality and morbidity in Detroit, Michigan. In: Revised Analyses of Time-Series Studies of Air Pollution and Health. Special Report. Boston:Health Effects Institute, 143-156.
Moolgavkar SH. 2003. Air pollution and daily deaths and hospital admissions in Los Angeles and Cook counties. In: Revised Analyses of Time-Series Studies of Air Pollution and Health. Special Report. Boston:Health Effects Institute, 183-198.
Ostro B, Chestnut L, Vichit-Vadokan N, Laixuthai A. 1999. The impact of particulate matter on daily mortality in Bangkok, Thailand. J Air Waste Manag Assoc 49:100-107.
Peel JL, Tolbert PE, Klein M, Metzger KB, Randers WD, Todd K, et al. 2005. Ambient air pollution and respiratory emergency department visits. Epidemiology 16:164-174.
Penttinen P, Tiittanen P, Pekkanen J. 2004. Mortality and air pollution in metropolitan Helsinki, 1988--1996. Scand J Work Environ Health 30(suppl 2):19-27.
Pope CA III, Burnett RT, Thun MJ, Calle EE, Krewski D, Ito K, et al. 2002. Lung cancer, cardiopulmonary mortality, end long-term exposure to fine particulate air pollution. JAMA 287:1132-1141.
R Development Core Team. 2004. R: A Language and Environment for Statistical Computing, version 1.9. Vienna: R Foundation for Statistical Computing.
Ruppert D, Want MP, Carroll RJ. 2003. Semiparametric Regression. Cambridge, UK:Cambridge University Press.
Samet JM, Zeger SL, Dominici F, Curriero F, Coursac I, Dockery DW, et al. 2000. The National Morbidity, Mortality, and Air Pollution Study. Part II: Morbidity and Mortality from Air Pollution in the United States. Boston:Health Effects Institute, June. Available: http://healtheffects.org/Pubs/Samet2.pdf [accessed 17 December 2004].
Schlesinger RD, Cassee F. 2003. Atmospheric secondary inorganic particulate matter: the toxicological perspective as a basis for health effects risk assessment. Inhal Toxicol 15:197-235.
Schwartz J. 2000. Assessing confounding, effect modification, and thresholds in the association between ambient particles and daily deaths. Environ Health Perspect 108:563-568.
Schwartz J. 2003. Daily deaths associated with air pollution in six US cities and short-term mortality displacement in Boston. In: Revised Analyses of Time-Series Studies of Air Pollution and Health. Special Report. Boston:Health Effects Institute, 219-226.
Schwartz J, Dockery DW, Neas LM. 1996. Is daily mortality associated specifically with fine particles? J Air Waste Manag Assoc 46:927-939.
StataCorp. 2003. Stata, version 8. College Station, TX:StataCorp.
U.S. EPA. 2004. Air Quality Criteria for Particulate Matter. EPA 600/P-99/002aF-bF. Washington, DC:U.S. Environmental Protection Agency.
Wong CM, Ma S, Hedley AJ, Lam TH. 2001. Effect of air pollution on daily mortality in Hong Kong. Environ Health Perspect 109:335-340.
Wood SN. 2000. Modeling and smoothing parameter estimation with multiple quadratic penalties. J R Stat Soc B 62:413-428.
World Health Organization. 1993. International Classification of Diseases, 10th Revision. Geneva:World Health Organization.
Zanobetti A, Schwartz J. 2002. Cardiovascular damage by airborne particles: are diabetics more susceptible? Epidemielogy 13:588-592.
Zanobetti A, Schwartz J, Samoli E, Gryparis A, Touloumi G, Peacock J, et al. 2003. The temporal pattern of respiratory and heart disease mortality in response to air pollution. Environ Health Perspect 111:1188-1193.
Bart Ostro, (1) Rachel Broadwin, (1) Shelley Green, (1) Wen-Ying Feng, (2) and Michael Lipsett (3)
(1) California Office of Environmental Health Hazard Assessment, Oakland, California, USA; (2) University of California Davis, Davis, California, USA; (3) University of California San Francisco, San Francisco, California, USA
The authors declare they have no competing financial interests.
Address correspondence to B. Ostro, Air Pollution Epidemiology Section, California Office of Environmental Health Hazard Assessment, 1515 Clay St., 16th Floor, Oakland, CA 94612 USA. Telephone: (510) 622-3157. Fax: (510) 622-3210. E-mail: email@example.com
Received 18 May 2005; accepted 1 September 2005.
Table 1. Descriptive statistics for air pollutants and mortality in nine California counties, 1999-2002. Days with data for Mean daily P[M.sub.2.5], P[M.sub.2.5] 2000 temperature, ([micro]g/ County population (a) and RH (n) [m.sup.3]) (b) Contra Costa 949 698 14 (1-77) Fresno 799 1,024 23 (1-160) Kern 662 1,186 22 (1-155) Los Angeles 9,519 1,221 21 (4-85) Orange 2,846 682 21 (4-114) Riverside 1,545 976 29 (2-120) Sacramento 1,223 1,214 14 (1-108) Santa Clara 1,683 717 15 (2-74) San Diego 2,814 1,333 16 (0-66) Mean daily Mean daily temperature Mean daily all-cause County ([degrees]F) (b) RH (%) (b) mortality (b) Contra Costa 60 (34-91) 64 (10-96) 16 (4-32) Fresno 65 (35-94) 55 (18-96) 13 (3-28) Kern 65 (36-95) 56 (13-100) 11 (2-25) Los Angeles 64 (46-89) 57 (15-88) 146 (99-242) Orange 63 (46-84) 67 (6-95) 40 (20-75) Riverside 65 (43-90) 58 (6-100) 28 (9-63) Sacramento 61 (36-91) 66 (13-100) 22 (7-45) Santa Clara 59 (40-89) 69 (22-96) 22 (9-44) San Diego 61 (43-84) 74 (16-100) 49 (26-87) RH, relative humidity. (a) In thousands. (b) Values in parentheses indicate minimum-maximum. Table 2. Mean daily deaths by mortality category in nine California counties, 1999-2002. Contra Los Mortality category Costa Fresno Kern Angeles Orange Age > 65 years 12.2 10.0 8.1 108.6 31.4 Male 7.2 6.4 5.4 70.7 18.3 Female 8.4 6.8 5.6 75.6 21.5 White non-Hispanic 12.3 9.3 8.6 86.0 32.7 Black non-Hispanic 1.5 0.8 0.6 20.6 0.4 Hispanic 0.9 2.4 1.5 26.7 3.6 In-hospital death 6.4 6.2 5.6 79.8 17.4 Out-of-hospital death 9.2 7.0 5.4 66.5 22.3 High school graduate 12.3 7.9 6.6 99.0 30.8 Not high school graduate 3.1 5.1 4.1 40.7 8.2 Diabetes 0.4 0.5 0.3 5.1 1.1 Cardiovascular disease 6.5 5.7 4.9 67.0 17.7 Ischemic heart disease 3.3 3.2 3.1 42.6 10.8 Respiratory disease 1.7 1.5 1.4 15.0 4.3 River- Sacra- Santa San Mortality category side mento Clara Diego Age > 65 years 22.2 16.1 16.5 38.7 Male 13.8 10.4 10.3 23.7 Female 14.2 11.3 11.4 25.5 White non-Hispanic 23.1 16.8 15.5 39.6 Black non-Hispanic 1.3 2.0 0.5 2.1 Hispanic 3.0 1.3 2.5 4.9 In-hospital death 11.5 9.7 9.9 18.6 Out-of-hospital death 16.5 12.1 11.7 30.6 High school graduate 20.4 15.9 15.9 37.6 Not high school graduate 6.8 5.3 5.4 10.2 Diabetes 0.6 0.6 0.6 1.2 Cardiovascular disease 13.0 9.2 9.1 20.4 Ischemic heart disease 8.0 5.3 4.8 11.4 Respiratory disease 3.2 2.6 2.4 5.7 Table 3. Percent change in daily mortality categories and 95% CIs per 10-[micro]g/[m.sup.3] increment in P[M.sub.2.5] using penalized splines and alternative lags [percent change (95% CI)]. All-cause Cardiovascular County, lag (a) mortality mortality Contra Costa 2 0.8 (-1.0 to 2.6) 0.6 (-2.1 to 3.3) 01 0.4 (-1.9 to 2.7) -0.6 (-4.1 to 2.9) Fresno 2 0.3 (-0.8 to 1.4) 0.5 (-1.1 to 2.2) 01 0.2 (-1.1 to 1.5) -0.1 (-2.1 to 1.9) Kern 2 -0.4 (-1.5 to 0.7) 0.8 (-0.6 to 2.3) 01 -0.3 (-1.5 to 0.9) 1.3 (-0.4 to 3.0) Los Angeles 2 -0.1 (-0.5 to 0.4) 0.1 (-0.6 to 0.8) 01 0.6 (0.1 to 1.1) 0.4 (-0.3 to 1.2) Orange 2 1.7 (0.6 to 2.9) 0.8 (-0.9 to 2.6) 01 2.3 (1.0 to 3.6) 1.8 (-0.2 to 3.8) Riverside 2 -0.2 (-1.1 to 0.6) 0.0 (-1.2 to 1.2) 01 0.2 (-0.9 to 1.2) -0.1 (-1.6 to 1.3) Sacramento 2 0.8 (-0.4 to 2.0) 1.1 (-0.7 to 2.8) 01 0.5 (-1.0 to 1.9) 0.9 (-1.2 to 3.0) Santa Clara 2 0.0 (-1.1 to 1.1) -0.2 (-1.8 to 1.4) 01 1.1 (-0.1 to 2.3) 1.1 (-0.6 to 2.9) San Diego 2 0.7 (-0.8 to 2.2) 1.0 (-1.3 to 3.2) 01 0.8 (-1.0 to 2.6) 0.3 (-2.2 to 2.9) Pooled results 2 0.2 (-0.2 to 0.7) 0.3 (-0.1 to 0.7) 01 0.6 (0.2 to 1.0) 0.6 (0.0 to 1.1) Respiratory Mortality County, lag (a) mortality > 65 years of age Contra Costa 2 0.4 (-5.1 to 6.0) 0.5 (-1.5 to 2.5) 01 6.9 (0.1 to 13.8) 0.2 (-2.4 to 2.8) Fresno 2 1.2 (-1.8 to 4.2) 0.8 (-0.5 to 2.0) 01 2.0 (-1.6 to 5.6) 0.4 (-1.1 to 1.9) Kern 2 -1.2 (-3.9 to 1.5) -0.1 (-1.4 to 1.1) 01 -1.2 (-4.3 to 1.9) -0.1 (-1.5 to 1.3) Los Angeles 2 1.2 (-0.2 to 2.6) -0.3 (-0.8 to 0.3) 01 2.1 (0.6 to 3.6) 0.5 (-0.1 to 1.1) Orange 2 5.7 (2.4 to 9.0) 1.2 (-0.1 to 2.5) 01 7.6 (3.7 to 11.5) 2.3 (0.9 to 3.8) Riverside 2 -0.5 (-2.7 to 1.7) -0.3 (-1.3 to 0.6) 01 -0.4 (-3.1 to 2.3) 0.1 (-1.0 to 1.3) Sacramento 2 3.5 (0.3 to 6.7) 0.9 (-0.4 to 2.3) 01 4.0 (-1.6 to 6.4) 1.1 (-0.6 to 2.8) Santa Clara 2 1.7 (-1.6 to 5.0) -0.2 (-1.4 to 1.0) 01 1.7 (-1.9 to 5.3) 1.2 (-0.1 to 2.6) San Diego 2 1.4 (-2.8 to 5.6) 1.4 (-0.3 to 3.0) 01 4.0 (-1.0 to 9.0) 1.2 (-0.8 to 3.2) Pooled results 2 1.3 (0.1 to 2.6) 0.2 (-0.2 to 0.7) 01 2.2 (0.6 to 3.9) 0.7 (0.2 to 1.1) CI, confidence interval. (a) Lag 01, average of 0- and 1-day lags of P[M.sub.2.5]; lag 2, 2-day lag of P[M.sub.2.5]. Model also includes day of week, spline smoothers of temperature and humidity, and two spline smoothers for time. Pooled results based on meta-analysis using a random-effects model. Table 4. Pooled estimates of percent changes in daily mortality categories and 95% CIs per 10-[micro]g/[m.sup.3] increment in P[M.sub.2.5] using natural splines. Mortality category df/year % Change (95% CI) All cause 4 0.5 (-0.1 to 1.1) 8 0.4 (-0.1 to 0.9) 12 0.3 (-0.1 to 0.7) Cardiovascular 4 0.4 (-0.2 to 0.9) 8 0.1 (-0.5 to 0.6) 12 0.0 (-0.6 to 0.6) Respiratory 4 2.1 (0.2 to 4.1) 8 1.6 (-0.5 to 3.6) 12 1.3 (-0.3 to 2.9) Older than 65 years 4 0.7 (0.0 to 1.3) 8 0.4 (-0.1 to 0.9) 12 0.3 (-0.1 to 0.8) Model includes average of 0- and 1-day lags of P[M.sub.2.5], day of week, spline smoothers of temperature and humidity, and two spline smoothers of time. Pooled results based on meta-analysis using a random-effects model. Table 5. Pooled estimates of percent changes in daily mortality categories and 95% Cls per 10-[micro]g/[m.sup.3] increment in P[M.sub.2.5] using penalized splines. Mortality category % Change (95% CI) All-cause 0.6 (0.2 to 1.0) Cardiovascular 0.6 (0.0 to 1.1) Respiratory 2.2 (0.6 to 3.9) Age > 65 years 0.7 (0.2 to 1.1) Ischemic heart disease 0.3 (-0.5 to 1.0) Diabetes 2.4 (0.6 to 4.2) Males 0.5 (-0.2 to 1.2) Females 0.8 (0.3 to 1.3) Whites 0.8 (0.2 to 1.3) Blacks 0.1 (-0.9 to 1.2) Hispanics 0.8 (-0.1 to 1.6) In hospital 0.6 (-0.1 to 1.3) Out of hospital 0.6 (0.1 to 1.1) High school graduates 0.4 (0.0 to 0.8) Non-high school graduates 0.9 (-0.1 to 1.9) Model includes average of 0- and 1-day lags of P[M.sub.2.5], day of week, spline smoothers of temperature and humidity, and two spline smoothers of time. Pooled results based on meta-analysis using a random-effects model.
|Printer friendly Cite/link Email Feedback|
|Publication:||Environmental Health Perspectives|
|Date:||Jan 1, 2006|
|Previous Article:||Rapidly measured indicators of recreational water quality are predictive of swimming-associated gastrointestinal illness.|
|Next Article:||Perinatal environmental tobacco smoke exposure in rhesus monkeys: critical periods and regional selectivity for effects on brain cell development and...|