Associations Between Ozone and Daily Mortality: Analysis... : Epidemiology (original) (raw)
The number of short-term mortality studies that analyzed associations with ozone has increased markedly in the past 10 years. This increase is in part due to studies that examined the effects of particulate matter (PM) on mortality. These studies were apparently in response to a controversy surrounding a series of studies published in the early 1990s that reported associations between PM and mortality at PM levels below the ambient air quality standard.1–4 In many of the studies that followed, ozone and other gaseous pollutants were analyzed as potential confounders for PM. Thus, the increased number of reported ozone-mortality associations may be to some extent “byproducts” of the increased attention to effects of PM on mortality. Although the adverse effects of ozone (eg, airway inflammation and transient lung function changes) have been well established, its mortality effects have been less well accepted. Perhaps this is because ozone has no historical counterpart to the 1952 London smog, in which high levels of PM produced thousands of excess deaths.
The recent focus on PM in air pollution studies also presented issues that complicate the interpretation of ozone–mortality associations. Most studies applied the same regression designs to estimate ozone risks as to estimate PM risks. Because ozone is more highly correlated with temperature, the model to adjust for weather effects in PM studies may not be appropriate. Ozone's correlations with temperature and other pollutants are also expected to change across seasons, and analyzing year-round data, which is common in PM studies, may not be appropriate for ozone. Also, the exposure to ozone is influenced by such factors as air conditioning.5,6 The correlation between personal ozone exposure and ambient levels may be poor, much more so than for PM.7 These complications may make it difficult to interpret ozone effects on mortality. Our aims were to review and summarize the current ozone risk estimates, to identify relevant research issues, and then to conduct an additional analysis to resolve some of the issues using available data from several U.S. cities.
A META-ANALYSIS OF SHORT-TERM ASSOCIATIONS BETWEEN OZONE AND MORTALITY
Methods
We focused on studies published during the period 1990–2003. The sources of literature were: 1) the reference sections of U.S Environmental Protection Agency (EPA) Criteria Documents on ozone and PM8,9; 2) results from a MEDLINE search provided by the EPA; 3) an additional reference list provided by the EPA (Conner, personal communication, 2003); and 4) literature listed in past reviews of ozone or PM mortality studies.10–15 Because of the problem that arose from the default convergence criteria used in the Generalized Additive Model (GAM),16,17 we identified those studies that used GAM with default convergence criteria and more than 1 nonparametric smoothing term,18–44 and those that did not.4,45–81 We conducted analyses both with and without the GAM studies. When the estimates were presented only as figures, the figures were scanned, and the point estimates and confidence bands were electronically read by calibrating a “ruler” against the axis in the figure. When only the point estimate and its statistical significance were presented, we assumed a standard error that met the significance criterion. We did not include risk estimates from multicity studies in the meta-analysis.19,20,82,83 To compare our summary estimates with those from other meta-analysis or multicity studies, we focused on the all-cause (nonaccidental) and all-age or age-65-and-over category to compute combined estimates. Combined effect estimates were obtained using the DerSimonian and Laird approach.84
To summarize the risk estimates using a comparable unit increase of ozone, we converted the estimates obtained from varying exposure indices. Using the nationwide ozone data (Langstaff and Pinto, EPA, personal communication, 2003), the difference between the mean and the 95th percentile (that is, “average” to “high” ozone increment) for 1-hour maximum, 8-hour maximum, and 24-hour average were approximately 40, 30, and 20 ppb, respectively. Therefore, the ozone–mortality excess risks were converted using this ratio.
Choosing the most statistically significant lag may bias the air pollution risk estimates upward.85 Examining a larger number of lags would also increase a chance of finding a statistically significant effect. However, most of the ozone–mortality studies examined relatively small numbers of lagged days (typically 0 through 3 days). An examination of the “most statistically significant” lags in the initially selected studies suggests that the majority of the single-day associations were immediate (0-day lag, 21 studies; 1-day lag, 8 studies; 2-day lag, 3 studies; and, longer lag days: 3 studies), ie, not a random pattern. Furthermore, when associations are found at multiple days, choosing only a single-day's results would underestimate the multiday effects. Thus, using a risk estimate for a single lag day can result in bias in either direction. With this limitation in mind, we considered the ozone mortality risk for a “selected” lag in each study. Lags only up to 3 days were considered for consistency.
We also summarized ozone–mortality risk by season. Ozone is expected to change its relationship with temperature (and with other pollutants) across seasons in urban locations. Clean air during the winter is associated with high-pressure systems, which are also associated with colder temperature. Thus, sunny clear winter days in the urban environment are the days when air pollution levels from primary emissions are low. These primary emissions include nitric oxide (which also “quenches” ozone), sulfur dioxide, and PM from local sources. This can lead to negative correlation between ozone and the primary pollutants. Such relationships were observed in a Philadelphia study.67 The changing relationship between ozone and temperature (and other pollutants) across seasons and the potential implications to health effects modeling appear to be underappreciated in the literature. One obvious way to alleviate this complication is to analyze the data by season. We identified 10 studies that reported ozone risk estimates by season and summarized these estimates. The main confounders of interest for ozone are “summer haze” PM such as sulfate. We identified 15 studies that examined ozone–mortality associations with and without PM indices using year-round data and summarized these estimates.
Results
Figure 1 shows ozone risk estimates for nonaccidental mortality across all ages and with year-round data in single pollutant models from 43 individual studies. The combined random effects estimate was 1.6% excess mortality (95% confidence interval [CI] = 1.1–2.0%) per 20-ppb increase in 24-hour average ozone. Although the majority of the estimates are positive, the heterogeneity across the cities is obvious. Analyses of the same cities by different researchers sometimes resulted in estimates of opposite signs (London, Amsterdam, and Santiago). The index of heterogeneity86,87 for the combined estimate was 77% (69–83%), indicating a high degree of heterogeneity. Excluding the GAM studies reduced the combined estimate only slightly (1.4%; 0.8–2.0%).
Ozone excess total nonaccidental mortality risk estimates for all-season data. Single pollutant model. The studies with asterisks are those that used GAM.
To explore potential sources of the observed heterogeneity, we extracted the mean pollution levels and temperature from the original articles and examined their associations with the estimated risks across studies (Fig. 2). The mean levels of these variables varied markedly across these cities. The ozone risk estimates were regressed on these variables separately and together with inverse–variance weights. All slopes were negative.
Relationships between the mean exposure variables and the ozone percent excess risk estimates per 20-ppb 24-hour average ozone for studies shown in Figure 1. The size of the circle is proportional to the precision of estimate (1/SE). When PM10 was not available (11 of 43 studies), the available PM indices were converted to PM10 using multiplication factors of 2, 2, and 0.5 for PM2.5, black smoke, and total suspended particles, respectively. The ozone risk estimates were regressed on these variables with inverse–variance weights of risk estimates. The lines shown are from these univariate inverse–variance weighted regressions.
We also examined possible publication bias in the pattern of estimates shown in Figure 1, as was done in a recent European analysis.88Figure 3 shows the funnel plot of the excess risk estimates plotted by precision (1/standard error). The test procedure suggested by Egger et al89 resulted in a significant asymmetry. Using the “trim-and-fill” technique,90 the random-effects combined estimate was slightly reduced (1.4%; 0.9–1.9%).
Funnel plot of the single-city studies shown in Figure 1. Solid circles are original estimates. Open circles are the estimates generated by the trim-and-fill method. The long dash line denotes the original random effects combined estimate. The short dash line is the revised random effects estimate with the filled points.
Figure 4 shows the studies that reported ozone risk estimates by season. In all except the Brisbane study, the ozone risk estimates are larger for summer than for winter. It is not surprising that the summer and winter estimates were similar in Brisbane, because the mean ozone levels were similar across season (22 ppb in summer and 27 ppb in winter for 1-hour maximum ozone). The combined random-effects estimate was 2.2% (0.8–3.6%) for the year-round data and 3.5% (2.1–4.9%) for warmer seasons (per 20-ppb increase in 24-hour average ozone). The indices of heterogeneity for these estimates were large (92% and 81%, respectively).
Ozone–mortality risk estimates by season. All-age total nonaccidental mortality, unless otherwise noted. The studies with asterisks are those that used GAM.
Figure 5 shows the ozone risk estimates with and without PM. In general, the ozone mortality risk estimates were not substantially affected by addition of PM indices. The combined random-effects estimates for ozone alone was 1.6% (1.1–2.2%) and with PM 1.5% (0.8–2.2%) per 20-ppb increase in 24-hour average ozone.
Ozone–mortality risk estimates for a subset of studies that examined ozone with and without PM. BS (Black Smoke), KM, and nephelometer are optical measures of particles. TSP, total suspended particles. The studies with asterisks are those that used GAM.
Issues Identified
The review and meta-analysis of the literature on short-term mortality effects of ozone identified several issues. Large heterogeneity in estimated ozone–mortality risk estimates was observed across studies. We found some suggestive evidence that some city-specific factors (eg, mean temperature) can explain some of the heterogeneity. However, the fact that analyses of the same cities’ data (London, Amsterdam, and Santiago) by different researchers resulted in markedly different estimates suggests a large influence of model specification on the results. An analysis of multiple cities using several alternative model specifications would provide information on the extent of model uncertainty.
In limited subsets, the combined estimate for warm seasons was larger than for cold seasons or year-round data. Applying a consistent model to season-specific data in multiple cities would clarify this pattern. Also, very few studies examined copollutant models with PM by season. Because possible confounding by PM during warm season is of particular interest, an analysis of data from multiple cities during the warm season with PM would be useful. Additionally, the relationships between ozone, temperature, and PM across season should be examined, because such information may help explain the negative ozone–mortality associations sometimes reported in the literature.
ANALYSIS OF 7 U.S. CITIES DATA
Methods
To further investigate these issues, we conducted an additional analysis using available data from 7 cities. These are New York City (1985–1994 for mortality analysis; and 1999–2000 for characterization of ozone and PM2.5); Cook County, IL (1985–1994); Detroit, MI (1985–1989, when ozone was measured year-round and daily PM10 data were available; and 1992–1994 when ozone was measured in warm seasons only but PM2.5 was available); Houston, TX (1985–1994); Minneapolis, MN–St. Paul, MO (1985–1994); Philadelphia, PA (1992–1995, PM2.5 was also available); and St. Louis, MO (1985–1994). We also had near-daily measurements of PM10 for all cities except New York City. The average of available multiple monitors were computed for both ozone and PM (both using the 24-hour average values). We examined total nonaccidental mortality for all ages. Our specific aims were to: 1) characterize ozone–PM relationships across seasons; 2) examine the sensitivity of ozone–mortality risk estimates to alternative weather models and the extent of temporal adjustments; 3) obtain ozone–mortality risk estimates by season; and 4) examine the sensitivity of ozone–mortality risk estimates to adjustment for PM by season.
To characterize ozone–PM relationships across seasons, we computed mean ozone for each quintile of PM for summer (June, July, and August) and winter (December, January, and February).
To examine the sensitivity of ozone–mortality risk estimates to alternative weather model specifications and the extent of temporal adjustments, we used a Poisson Generalized Linear Model adjusting for temporal trends, day of the week, and weather effects. Based on the fact that the majority of the reviewed studies showed 0- or 1-day lagged associations, we included the average of 0- and 1-day lagged 24-hour average ozone. To adjust for seasonal cycles and other temporal trends, we included a smoothing function of days using natural splines with 4 sets of degrees of freedom (df): 4, 6, 12, and 26 df/y. This range covers the extent of temporal smoothing used in the studies we reviewed.
Based on the types of weather models used in the studies we reviewed, we considered 4 alternative weather models: 1) quintile indicator variables67; 2) piecewise (V-shaped) linear terms with a cutoff point at median temperature28,29,51,58,80; 3) 2 smoothing terms, including one with natural splines of same-day temperature (df = 3) and another with natural splines of same-day dewpoint (df = 3)91–94; and 4) 4 smoothing terms, including natural splines of same-day temperature (df = 6), natural splines of the average of lag 1 through 3-day temperature (df = 6), natural splines of same-day dewpoint (df = 3), and natural splines of the average of lag 1 through 3-day dewpoint (df = 3).40,83 Because our past analyses of some of these cities also indicated that cold temperature effects were lagged,33,52,59,60 we used indicators for the average of 2- and 3-day lags for the 2 lowest quintiles in the quintile indicator model, and also included an indicator for hot (80th percentile) and humid (80th percentile) days to model possible interactions between temperature and humidity. We were also interested in the extent of collinearity between ozone and the terms in these alternative weather models, because the high collinearity makes it more difficult to interpret ozone coefficients.10 We examined this issue by computing concurvity between ozone and the weather terms.16,17
Using the 4 alternative weather models and the 4 sets of degrees of freedom, we computed ozone–mortality risk estimates per 20-ppb 24-hour average. The regression analysis was then repeated for the situations with 12 df/y for temporal trend adjustment, and for colder (October through March) and warmer (April through September) seasons with 6 df/y, with and without PM10 or PM2.5 (average of 0- and 1-day lag) in the model. The ozone–PM10 copollutant model was not run for New York City, because the PM10 was measured only every sixth day.
Results
Figure 6 shows the relationships between ozone and PM10 for summer and winter, with ozone averaged for quintiles of PM. The slopes are positive for summer but negative (although shallower) in winter, except in Houston, where the slope was positive in winter. The relationship between temperature and ozone examined in the same way showed that the relationships were generally J-shaped except in Houston, where slopes were positive in both seasons (results not shown). Note that the “winter” temperature in Houston is mild (the lowest quintile of temperature in December through February in Houston was approximately 40°F, compared with less than 20°F in the other 6 cities). These results confirm a generally negative correlation of ozone with PM in winter.
Relationship between PM and ozone in warm months (June through August) and cold months (December through February) as sorted and averaged by the quintiles of PM.
Figure 7 shows the estimated ozone mortality excess risks for the 4 weather models and 4 sets of degrees of freedom for temporal trend adjustment. The alternative weather models can make substantial differences in risk estimates. Generally, the quintile temperature model produced the largest estimates, and the 4-smoothing-term model produced the smallest. As expected, the 4-smoothers model also showed the highest concurvity for ozone among the weather models, and the quintile model showed the lowest concurvity, although all the models showed relatively high concurvity (r > 0.8) for all the cities except Houston (r approximately 0.6–0.8). The extent of smoothing for temporal trend adjustment did not make substantial or systematic difference in ozone risk estimates, except for the Cook County data. The large city-to-city variation in the ozone risk estimates was also clear. To summarize the relative importance of these 3 factors (ie, weather model, extent of smoothing, and city identification), all the estimates shown in Figure 7 were regressed (using the original beta coefficients) on the 3 natural splines functions of the levels of these terms. As expected, city-to-city variation was the strongest predictor, with the largest contrast between Detroit and St. Louis (corresponding to 3.4% difference between the largest and smallest fitted estimates per 20-ppb increase in the average of 0- and 1-day lags). This was followed by the weather model (1.1% between the quintile model and 4-smoother model) and the extent of smoothing (0.4% between the models with 12 df/y and 26 df/y).
Sensitivity of ozone mortality excess risk estimates to weather model specifications and the extent of smoothing for temporal trend adjustment. y-axis is percent excess risk per 20-ppb increase in the average of 0- and 1-day lagged 24-hour average ozone. x-axis is the extent of smoothing for temporal trends.
Using the same 4 weather models, ozone–mortality risk estimates were obtained by season with and without PM indices. Because this above sensitivity analysis showed that the extent of smoothing for temporal trends did not substantially affect the estimated risks, we estimated season-specific effects estimates using 6 df/y and compared with the year-round estimates for the 12 df/y model. Figure 8 shows the estimated ozone–mortality excess risks by weather model, season, and with and without PM10 (PM2.5 for Philadelphia and the 1992–1994 Detroit summer data) in the model. Although only Philadelphia and Detroit had PM2.5 data, PM2.5 and PM10 were highly correlated (r = 0.91 and 0.89, respectively) in these cities, where sulfate, the presumed confounder for ozone, was prevalent. Strong contrasts in ozone risk estimates between colder and warmer seasons are seen in New York City, Cook County, and Detroit, but the contrasts also varied across the weather models, with the 4-smoother model generally showing the least contrast. In the Houston data, ozone risk estimates in colder seasons were larger than those for the year-round or warmer seasons. This winter positive slope may be in part due to the positive association between ozone and PM10 in winter in Houston, which was not seen in other cities (Fig. 6). Including PM in the model did not substantially reduce ozone risk estimates in most cases. In Cook County, Detroit, and Philadelphia, PM in single pollutant models was also associated with mortality in both all-year and warm months. Including both ozone and PM in the regression models in these cities tended to attenuate both pollutants’ coefficients, but not substantially. The models with both ozone and PM in these cities often showed better fits (ie, lower Akaike's Information Criteria) than those with either pollutant alone. These results suggest that ozone and PM contribute independently to mortality. In the other 3 cities, PM was less strongly associated with mortality and did not influence ozone–mortality associations.
Estimated ozone mortality percent excess risks per 20-ppb increase in the average of 0- and 1-day lags for all year (X), colder months (white circle), and warmer month (black circle) in a single pollutant model (solid line) and with PM10 or PM2.5 (dashed line) for 4 weather-adjustment models. The upper confidence bands for Detroit 1992–1994 (denoted with asterisks) are cut off in the scale shown.
The ozone risk estimates were further combined across cities for each weather model (Fig. 8). Because we were interested in comparing the estimates with and without PM, the New York City data were not included in the combined estimates. In the combined estimates, the difference across the weather models is clear, especially for the year-round and warm-season results. The quintile temperature indicator model resulted in the largest estimates, which were approximately twice those for the 4-smoother temperature model. For example, the estimates for the all-year ozone-only case were: 2.0% (1.1–2.9%) for the quintile model versus 1.0% (0.0–2.0%) for the 4-smoother model per 20-ppb increase in the average of 0- and 1-day lag 24-hour ozone. Corresponding numbers for the warm seasons with PM case were: 2.0% (0.6–3.4%) and 1.1% (−0.1–2.2%).
DISCUSSION
Figure 9 shows a comparison of the results of this review and of other recent meta-analyses and multicity studies. Our summary estimates, which included more studies than past meta-analyses, are fairly consistent with the results from other recent meta-analyses. However, these estimates are approximately twice as large as the combined estimates from the largest U.S. 90 cities (the National Morbidity, Mortality, and Air Pollution Study [NMMAPS]) study,83 which were 0.8% in the single pollutant model for the copollutant subset and 0.6% in the model with PM10. These differences may be partly due to the more aggressive adjustment model for weather effects (4 smoothing terms for temperature and dewpoint) in the NMMAPS study. Our additional analysis of 6 U.S. cities also indicates that a weather model similar to the one used in the NMMAPS study tends to show the smallest estimates among the 4 weather models. Another possible explanation is that the combined estimates from various single-city studies may be biased upward because the “optimal” or “best” lags were chosen from each study, whereas in the NMMAPS results, the estimate being compared is for the fixed 0-day lag for all the 90 cities.
A comparison of ozone–mortality risk estimates from recent meta-analyses, multicity studies (this review and additional analyses). Those with asterisks include GAM-affected studies. The upper confidence bands for APHEA and Canadian 8-city studies (denoted with 2 asterisks) are cut off in the scale shown.
We also computed combined estimates using the subsets of studies (10 studies) that reported estimates by season. We found that the estimates for warm seasons were generally larger than the estimates for year-round data. A similar pattern was seen the NMMAPS study, although again, the estimates were smaller than our combined estimates. In our additional analysis of 6 U.S. cities, the estimate for warm seasons (2.9%) was larger than that for all year (2.0%) in single pollutant model using the quintile indicator weather model, but the estimates and their contrast were reduced once PM was included in the model (2.0% vs. 1.6%). Interestingly, the estimates from the 4-smoother model were comparable (0.9–1.1%) regardless of season or inclusion of PM. This may be due to the more flexible fits of the 4-smoother model, attributing more of the season-specific variation of mortality to the weather, rather than to ozone.
Although the combined estimates from this study and other past meta-analyses are fairly consistent, there appears to be considerable heterogeneity in ozone–mortality risk estimates across studies. Some of this heterogeneity may come from the difference in model specifications, as seen in the conflicting results from the analyses of data from the same city by different researchers. City-to-city variation may also be due to several city-specific factors, including the housing characteristics (eg, air-conditioning rate), population demographics (eg, percent of underprivileged population), and the pattern of air pollution (eg, correlation between ozone and PM). In our examination of the possible sources of heterogeneity in the past studies, the average city temperature was negatively associated with ozone risk estimates. This observation and the apparent lack of positive association between mean ozone level and ozone risk estimates are counterintuitive, but housing characteristics or air-conditioning rates may override or complicate the influence of these factors. Unfortunately, we did not have data on air-conditioning use.
As shown in our analysis of multiple U.S. cities, the difference in weather adjustment model alone can make a difference in the combined estimates by a factor of two. In terms of statistical fits, the models with smoothers tend to give better fits than piecewise linear or indicator models, but they also show higher concurvity with ozone, making the interpretation of ozone risk estimates more difficult. From an epidemiologic point of view, model validations of these weather models are needed. Most of these models fit much of the mortality variations in the mild temperature range. Statistical properties aside, it is not clear whether these models actually adjust for weather effects. Because daily fluctuations of weather and air pollution are related, it is possible that these models may ascribe at least some of the real pollution effects to weather. Therefore, the process of risk assessment using these estimates will need to take into consideration the model uncertainties.
In summary, our meta-analysis and additional analysis of multiple cities, other meta-analyses, and multicity studies collectively suggest short-term associations between ozone and mortality, although the estimates are heterogeneous across cities. The excess risk estimates were higher in summer when ozone is high and people spend more time outdoors, and lower or null in cold seasons when ozone is low and exposures are expected to be low. The potential confounding between ozone and PM does not appear to substantially affect ozone risk estimates. Risk assessment should take into consideration model uncertainties that can make a 2-fold difference in estimates.
ACKNOWLEDGMENTS
The 1992–1994 PM2.5 data used in Detroit analysis were originally provided from Dr. Jeff Brook of Environment Canada for Dr. Lippmann's project funded by Health Effects Institute. The PM2.5 data for Philadelphia were obtained from the National Exposure Research Laboratory (NERL) at EPA.
REFERENCES
1. Fairley D. The relationship of daily mortality to suspended particulates in Santa Clara county, 1980–86. Environ Health Perspect. 1990;89:159–168.
2. Schwartz J, Dockery DW. Particulate air pollution and daily mortality in Steubenville, Ohio. Am J Epidemiol. 1992a;135:12–24.
3. Schwartz J, Dockery DW. Increased mortality in Philadelphia associated with daily air pollution concentrations. Am Rev Respir Dis. 1992b;145:600–604.
4. Dockery DW, Schwartz J, Spengler JD. Air pollution and daily mortality: associations with particulates and acid aerosols. Environ Res. 1992;59:362–373.
5. Brauer M, Brook JR. Personal and fixed-site ozone measurements with a passive sampler. J Air Waste Manag Assoc. 1995;45:529–537.
6. Avol EL, Navidi WC, Colome SD. Modeling ozone levels in and around Southern California homes. Environ Sci Technol. 1998;32:463–468.
7. Sarnat JA, Schwartz J, Catalano PJ, et al. Gaseous pollutants in particulate matter epidemiology: confounders or surrogates? Environ Health Perspect. 2001;109:1053–1061.
8. US Environmental Protection Agency. Air Quality Criteria for Ozone and Other Photochemical Oxidants; 1996, EPA /600/P-93/004aF.
9. US Environmental Protection Agency. Air Quality Criteria for Particulate Matter; 1996, EPA/600/P-95/001.
10. Thurston GD, Ito K. Epidemiological studies of ozone exposure effects. In: Holgate S, Samet JM, Koren H, et al., eds. Air Pollution and Health. San Diego: Academic Press; 1999:485–510.
11. Thurston GD, Ito K. Epidemiological studies of acute ozone exposures and mortality. J Expo Anal Environ Epidemiol. 2001;11:286–294.
12. Levy JI, Carrothers TJ, Tuomisto JT, et al. Assessing the public health benefits of reduced ozone concentrations. Environ Health Perspect. 2001;109:9–20.
13. Levy JI, Hammitt JK, Spengler JD. Estimating the mortality impacts of particulate matter: what can be learned from between-study variability? Environ Health Perspect. 2000;108:109–117.
14. US Environmental Protection Agency. Regulatory Impact Analysis for the Particulate Matter and Ozone National Ambient Air Quality Standards and Proposed Regional Haze Rule. Research Triangle Park, NC: Office of Air Quality Planning and Standards; 1997.
15. Stieb DM, Judek S, Burnett RT. Meta-analysis of time-series studies of air pollution and mortality: effects of gases and particles and the influence of cause of death, age, and season. J Air Waste Manag Assoc. 2002;52:470–484.
16. Dominici F, McDermott A, Zeger S, et al. On generalized additive models in time-series studies of air-pollution and health. Am J Epidemiol. 2002;156:1–11.
17. Ramsay T, Burnett R, Krewski D. The effect of concurvity in generalized additive models linking mortality and ambient air pollution. Epidemiology. 2003;14:18–23.
18. Anderson HR, Bremner SA, Atkinson RW, et al. Particulate matter and daily mortality and hospital admissions in the west midlands conurbation of the United Kingdom: associations with fine and coarse particles, black smoke and sulphate. Occup Environ Med. 2001;58:504–510.
19. Burnett RT, Brook J, Dann T, et al. Association between particulate- and gas-phase components of urban air pollution and daily mortality in eight Canadian cities. Inhal Toxicol. 2000;12(suppl 4):15–39.
20. Burnett RT, Cakmak S, Brook JR. The effect of the urban ambient air pollution mix on daily mortality rates in 11 Canadian cities. Can J Public Health. 1998;89:152–156.
21. Burnett RT, Cakmak S, Raizenne ME, et al. The association between ambient carbon monoxide levels and daily mortality in Toronto, Canada. J Air Waste Manag Assoc. 1998;48:689–700.
22. Cifuentes LA, Vega J, Kopfer K, et al. Effect of the fine fraction of particulate matter versus the coarse mass and other pollutants on daily mortality in Santiago, Chile. J Air Waste Manag Assoc. 2000;50:1287–1298.
23. Conceicao GMS, Miraglia SGEK, Kishi HS, et al. Air pollution and child mortality: a time-series study in Sao Paulo, Brazil. Environ Health Perspect. 2001;109:347–350.
24. Fairley D. Daily mortality and air pollution in Santa Clara County, California: 1989–1996. Environ Health Perspect. 1999;107:637–641.
25. Fischer P, Hoek G, Brunekreef B, et al. Air pollution and mortality in The Netherlands: are the elderly more at risk? Eur Respir J Suppl. 2003;40:34s–38s.
26. Goldberg MS, Burnett RT, Brook J, et al. Associations between daily cause-specific mortality and concentrations of ground-level ozone in Montreal, Quebec. Am J Epidemiol. 2001;154:817–826.
27. Gwynn RC, Burnett RT, Thurston GD. A time-series analysis of acidic particulate matter and daily mortality and morbidity in the Buffalo, New York, region. Environ Health Perspect. 2000;108:125–133.
28. Hoek G, Schwartz JD, Groot B, et al. Effects of ambient particulate matter and ozone on daily mortality in Rotterdam, The Netherlands. Arch Environ Health. 1997;52:455–463.
29. Hoek G, Brunekreef B, Fischer P, et al. The association between air pollution and heart failure, arrhythmia, embolism, thrombosis, and other cardiovascular causes of death in a time series study. Epidemiology. 2001;12:355–357.
30. Hong YC, Leem JH, Ha EH, et al. PM10 exposure, gaseous pollutants, and daily mortality in Inchon, South Korea. Environ Health Perspect. 1999;107:873–878.
31. Kelsall JE, Samet JM, Zeger SL, et al. Air pollution and mortality in Philadelphia, 1974–1988. Am J Epidemiol. 1997;146:750–762.
32. Kotesovec F, Skorkovsky J, Brynda J, et al. Daily mortality and air pollution in northern Bohemia: different effects for men and women. Cent Eur J Public Health. 2000;8:120–127.
33. Lippmann M, Ito K, Nadas A, et al. Association of Particulate Matter Components With Daily Mortality and Morbidity in Urban Populations, Report 95. Health Effects Institute; 2000.
34. Mar TF, Norris GA, Koenig JQ, et al. Associations between air pollution and mortality in Phoenix, 1995–1997. Environ Health Perspect. 2000;108:347–353.
35. Michelozzi P, Forastiere F, Fusco D, et al. Air pollution and daily mortality in Rome, Italy. Occup Environ Med. 1998;55:605–610.
36. Moolgavkar SH. Air pollution and daily mortality in three US counties. Environ Health Perspect. 2000;108:777–784.
37. Ostro BD, Broadwin R, Lipsett MJ. Coarse and fine particles and daily mortality in the Coachella Valley, California: a follow-up study. J Expo Anal Environ Epidemiol. 2000;10:412–419.
38. Ostro BD, Hurley S, Lipsett MJ. Air pollution and daily mortality in the Coachella Valley, California: a study of PM10 dominated by coarse particles. Environ Res. 1999;81:231–238.
39. Saez M, Ballester F, Barcelo MA, et al. A combined analysis of the short-term effects of photochemical air pollutants on mortality within the EMECAM project. Environ Health Perspect. 2002;110:221–228.
40. Samet J, Zeger S, Dominici F, et al. The National Morbidity, Mortality, and Air Pollution Study Part II: Morbidity, Mortality, and Air Pollution in the United States. Cambridge, MA: Health Effects Institute; 2000.
41. Simpson R, Denison L, Petroeschevsky A, et al. Effects of ambient particle pollution on daily mortality in Melbourne, 1991–1996. J Expo Anal Environ Epidemiol. 2000;10:488–496.
42. Wong CM, Ma S, Hedley AJ, et al. Effect of air pollution on daily mortality in Hong Kong. Environ Health Perspect. 2001;109:335–340.
43. Wong TW, Tam WS, Yu TS, et al. Associations between daily mortalities from respiratory and cardiovascular diseases and air pollution in Hong Kong, China. Occup Environ Med. 2002;59:30–35.
44. Zeghnoun A, Czernichow P, Beaudeau P, et al. Short-term effects of air pollution on mortality in the cities of Rouen and Le Havre, France, 1990–1995. Arch Environ Health. 2001;56:327–335.
45. Anderson HR, Ponce de Leon A, Bland JM, et al. Air pollution and daily mortality in London: 1987–92. BMJ. 1996;312:665–669.
46. Borja-Aburto VH, Loomis DP, Bangdiwala SI, et al. Ozone, suspended particulates, and daily mortality in Mexico City. Am J Epidemiol. 1997;145:258–268.
47. Borja-Aburto VH, Castillejos M, Gold DR, et al. Mortality and ambient fine particles in southwest Mexico City, 1993–1995. Environ Health Perspect. 1998;106:849–855.
48. Bremner SA, Anderson HR, Atkinson RW, et al. Short term associations between outdoor air pollution and mortality in London 1992–4. Occup Environ Med. 1999;56:237–244.
49. Chock DP, Winkler S, Chen C. A study of the association between daily mortality and ambient air pollutant concentrations in Pittsburgh, Pennsylvania. J. Air Waste Manage Assoc. 2000;50:1481–1500.
50. Dab W, Medina S, Quenel P, et al. Short term respiratory health effects of ambient air pollution: results of the APHEA project in Paris. J Epidemiol Community Health. 1996;50:S42–S46.
51. Diaz J, Garcia R, Ribera P, et al. Modeling of air pollution and its relationship with mortality and morbidity in Madrid, Spain. Int Arch Occup Environ Health. 1999;72:366–376.
52. DeLeon SF, Thurston GD, Ito K. Contribution of respiratory disease to non-respiratory mortality associations with air pollution. Am J Respir Crit Care Med. 2003;167:1117–1123.
53. Fairley D. 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; 2003:97–106. Available at: http://www.healtheffects.org/pubs-special.htm. Accessed May 16, 2003.
54. Gamble JL. Effects of ambient air pollution on daily mortality: a time-series analysis of Dallas, Texas, 1990–1994. The Air and Waste Management Association's 92nd Annual Meeting and Exhibition; June 14–18, 1998; San Diego, CA, Paper No. 98-MP26.03.
55. Garcia-Aymerich J, Tobias A, Anto JM, et al. Air pollution and mortality in a cohort of patients with chronic obstructive pulmonary disease: a time series analysis. J Epidemiol Community Health. 2000;54:73–74.
56. Goldberg MS, Burnett RT, Valois MF, et al. Associations between ambient air pollution and daily mortality among persons with congestive heart failure. Environ Res. 2003;91:8–20.
57. Gouveia N, Fletcher T. Time series analysis of air pollution and mortality: effects by cause, age, and socioeconomic status. J Epidemiol Community Health. 2000;54:750–755.
58. Hoek G. Daily mortality and air pollution in The Netherlands. In: Revised analyses of time-series studies of air pollution and health. Special report. Boston: Health Effects Institute; 2003:133–142. Available at: http://www.healtheffects.org/pubs-special.htm. Accessed May 16, 2003.
59. Ito K, Thurston GD. Daily PM10/mortality associations: an investigation of at-risk subpopulations. J Expos Anal Environ Epidemiol. 1996;6:79–95.
60. Ito K. 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; 2003:143–156. Available at: http://www.healtheffects.org/pubs-special.htm. Accessed May 16, 2003.
61. Kinney PL, Ozkaynak H. Associations of daily mortality and air pollution in Los Angeles County. Environ Res. 1991;54:99–120.
62. Kinney PL, Ito K, Thurston GD. A sensitivity analysis of mortality/PM10 associations in Los Angeles. Inhal Toxicol. 1995;7:59–69.
63. Klemm RJ, Mason RMJr. Aerosol research and inhalation epidemiological study (ARIES): air quality and daily mortality statistical modeling-interim results. J Air Waste Manag Assoc. 2000;50:1433–1439.
64. Lee JT, Schwartz J. Reanalysis of the effects of air pollution on daily mortality in Seoul, Korea: a case–crossover design. Environ Health Perspect. 1999;107:633–636.
65. Lee JT, Shin D, Chung Y. Air pollution and daily mortality in Seoul and Ulsan, Korea. Environ Health Perspect. 1999;107:149–154.
66. Lipfert FW, Morris SC, Wyzga RE. Daily mortality in the Philadelphia metropolitan area and size-classified particulate matter. J Air Waste Manage Assoc. 2000;1501–1513.
67. Moolgavkar SH, Luebeck EG, Hall TA, et al. Air pollution and daily mortality in Philadelphia. Epidemiology. 1995;6:476–484.
68. Moolgavkar SH, Luebeck EG. A critical review of the evidence on particulate air pollution and mortality. Epidemiology. 1996;7:420–428.
69. Morgan G, Corbett S, Wlodarczyk J, et al. Air pollution and daily mortality in Sydney, Australia, 1989 through 1993. Am J Public Health. 1998;88:759–764.
70. Ostro B. Fine particulate air pollution and mortality in two Southern California counties. Environ Res. 1995;70:98–104.
71. Ostro BD, Sanchez JM, Aranda C, et al. Air pollution and mortality: results from a study of Santiago, Chile. J Expo Anal Environ Epidemiol. 1996;6:97–114.
72. Peters A, Skorkovsky J, Kotesovec F, et al. Associations between mortality and air pollution in central Europe. Environ Health Perspect. 2000;108:283–287.
73. Ponka A, Savela M, Virtanen M. Mortality and air pollution in Helsinki. Arch Environ Health. 1998;53:281–286.
74. Prescott GJ, Cohen GR, Elton RA, et al. Urban air pollution and cardiopulmonary ill health: a 14.5 year time series study. Occup Environ Med. 1998;55:697–704.
75. Roemer WH, Van Wijnen JH. Daily mortality and air pollution along busy streets in Amsterdam, 1987–1998. Epidemiology. 2001;12:649–653.
76. Saldiva PH, Pope CA III, Schwartz J, et al. Air pollution and mortality in elderly people: a time-series study in Sao Paulo, Brazil. Arch Environ Health. 1995;50:159–163.
77. Simpson RW, Williams G, Petroeschevsky A, et al. Associations between outdoor air pollution and daily mortality in Brisbane, Australia. Arch Environ Health. 1997;52:442–454.
78. Sunyer J, Castellsagué J, Sáez M, et al. Air pollution and mortality in Barcelona. J Epidemiol Community Health. 1996;50(suppl 1):S76–S80.
79. Vedal S, Brauer M, White R, et al. Air pollution and daily mortality in a city with low levels of pollution. Environ Health Perspect. 2003;111:45–51.
80. Verhoeff AP, Hoek G, Schwartz J, et al. Air pollution and daily mortality in Amsterdam. Epidemiology. 1996;7:225–230.
81. Zmirou ZD, Barumandzadeh T, Balducci F, et al. Short term effects of air pollution on mortality in the city of Lyon, France, 1985–1990. J Epidemiol Community Health. 1996;50(suppl 1):S30–S35.
82. 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;146:177–185.
83. Dominici F, McDermott A, Daniels M, et al. Mortality among residents of 90 cities. In: Revised analyses of time-series studies of air pollution and health. Special report. Boston: Health Effects Institute; 2003:9:–24. Available at: http://www.healtheffects.org/pubs-special.htm. Accessed May 16, 2003.
84. DerSimonian R, Laird N. Meta-analysis in clinical trials. Control Clin Trials. 1986;7:177–188.
85. Lumley T, Sheppard L. Assessing seasonal confounding and model selection bias in air pollution epidemiology using positive and negative control analyses. Environmetrics. 2000;11:705–717.
86. Higgins JPT, Thompson SG. Quantifying heterogeneity in a meta-analysis. Stat Med. 2002;21:1539–1558.
87. Higgins JPT, Thompson SG, Deeks JJ, et al. Measuring inconsistency in meta-analyses. BMJ. 2003;327:557–560.
88. World Health Organization. Meta-Analysis of Time-Series Studies and Panel Studies of Particulate Matter and Ozone. Report of a WHO Task Group, Anderson HR, Atkinson RW, Peacock, JL, et al. Copenhagen: WHO Regional Office for Europe; 2004.
89. Egger M, Smith GD, Schneider M, et al. Bias in meta-analysis detected by a simple, graphical test. BMJ. 1997;315:629–634.
90. Duval S, Tweedie R. A nonparametric “trim and fill” method of accounting for publication bias in meta-analysis. J Am Stat Assoc. 2000;95:89–98.
91. Schwartz J, Dockery DW, Neas LM. Is daily mortality associated specifically with fine particles? J Air Waste Manag Assoc. 1996;46:927–939.
92. Schwartz J. 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; 2003:219–226. Available at: http://www.healtheffects.org/pubs-special.htm. Accessed May 16, 2003.
93. Klemm RJ, Mason RM, Heilig CM, et al. Is daily mortality associated specifically with fine particles? Data reconstruction and replication of analyses. J Air Waste Manag Assoc. 2000;50:1215–1222.
94. Klemm RJ, Mason RM. Replication of reanalysis of Harvard Six-City mortality study. In: Revised analyses of time-series studies of air pollution and health. Special report. Boston: Health Effects Institute; 2003:165–172. Available at: http://www.healtheffects.org/pubs-special.htm. Accessed May 16, 2003.
95. Canadian Environmental Protection Act Federal–Provincial Advisory Committee Working Group on Air Quality Objectives and Guidelines. National Ambient Air Quality Objectives for Ground-Level Ozone, Science Assessment Document. Ottawa: Health Canada and Environment Canada; 1999.
© 2005 Lippincott Williams & Wilkins, Inc.