Short-term exposure to ozone air pollution has been associated with adverse health effects including reduced lung function, exacerbation of chronic respiratory diseases and increases in hospital admissions and mortality rates.1–6 Although some small studies have had divergent results, most multicity studies and meta-analyses have found a positive association between daily increases in ambient ozone and mortality.1,3,4,6 Moreover, the association has been shown to be independent of temperature.7 In general, these associations have been restricted to the warm season, for reasons not yet clear.4,7
To date, little work has been directed toward identifying which subpopulations are more susceptible to death due to the effects of ambient ozone. Some, but not all, studies investigating age differences in the mortality risk associated with ozone have found a larger susceptibility among the elderly.8–12 Little is known about the vulnerability of other demographic groups or the role of medical conditions that may also confer susceptibility to ozone effects. Information on vulnerable subgroups would help guide air quality standards and provide a better understanding of air pollution toxicity.
In the present multicity study, we aimed to identify factors that confer susceptibility to ozone air pollution, including socio-demographic characteristics and medical conditions. To do this, we used the case-only approach, which provides important advantages as compared with traditional analyses, as detailed below. The study was conducted in a large sample of US cities, which allowed for the examination of differences in vulnerability factors across different locations and the determinants of its variability.
Study Design and Population
We used a case-only design to identify modifiers of the effect of ozone on mortality in 48 US cities during the period 1989–2000. Cities were randomly selected from among the most populated Metropolitan Statistical Areas with urban counties near appropriate ozone and weather monitoring stations. A few less-populated cities were also included to cover all regions of the country. We included all counties corresponding to the Metropolitan Statistical Area of cities presented in Figure 1.
The case-only design, originally used to examine gene-environment interactions, was recently proposed by Armstrong13 for the identification of modifiers of the effect of time-varying exposures, such as ambient ozone. The underlying idea is that by restricting the sample to only cases (eg, decedents), susceptibility factors can be identified as those time-fixed characteristics that occur more frequently among cases as exposure increases. To understand the case-only approach, consider first a more traditional Poisson time-series, where we model the number of deaths per day as:
In contrast, the case-only approach models the presence of a time-invariant characteristic (eg, diabetes, sex) on the death certificates on a given day. Since this approach models the prevalence of a time-invariant condition on the death certificates (and not the number of certificates), the time-varying factors that predict the number of certificates in Equation (1) (such as weather and season), are irrelevant and drop out of the analysis. Instead, we only estimate β 3 using a logistic regression that predicts the occurrence of the time-fixed characteristic among cases as a function of the exposure level. This is explained more formally in previous publications.13,14 While a case-crossover approach can control for season by matching, it still requires modeling of weather terms, and has less power than the Poisson regression. In contrast, the case-only analysis has slightly more power than the Poisson model for estimating the interaction term.13 Hence, we have used this approach to identify interactions, that is, modifiers of the effect of ozone on mortality. Here, we used a two-stage analysis with separate logistic regressions in each city; the results were then pooled across the 48 cities. We also examined city characteristics that might modify the association in the second stage.
Daily mortality data were obtained for each city from the National Center for Health Statistics mortality tapes. Individual records included information on primary and secondary causes of death, place of death, and personal characteristics such as age, sex, race, and educational attainment. Deaths due to external causes were excluded from our analyses (International Classification of Diseases [9th revision] codes [ICD-9]: 800-999; ICD-10: S00-Z99). Because increases in ozone-related mortality have been reported mainly for respiratory and cardiovascular diseases,1,3,9,15 we examined as potential effect modifiers the most common chronic respiratory and circulatory system diseases listed as secondary causes of death. (See Table 1 for a complete list of conditions examined and their corresponding ICD codes.) Conditions such as atrial fibrillation or diabetes were also chosen for examination on the grounds that they have been previously identified as modifiers of the association of particulate air pollution and temperature with mortality.16–18 Also, because systemic inflammation has been postulated as a possible mechanism through which air pollution may increase mortality,8,19 we examined inflammatory diseases as a potential effect modifier.
We obtained daily ozone levels from the US Environmental Protection Agency's Aerometric Retrieval System.20 For each city, we estimated the 8-hour daily mean concentrations of ozone using an algorithm that averaged levels reported by multiple monitoring locations.21 Most cities in the United States monitor ozone only during the warm months, and studies that have examined ozone mortality in colder months in the United States have failed to find an association.4,7 Hence, we restricted our sample to the warm months (May through September). Overall, ozone data were available for 99% of the sample. Forty-two cities had ozone measurements for least 98% of the days (19 had measurements for all days) and only 2 cities had less than 90% of complete data: Spokane (87.9%) and Kansas City (88.5%), for which a whole year was missing.
For each city, we obtained the daily mean temperature and dew point temperature from the nearest National Weather Service Surface Station (EarthInfo, Boulder, CO). We used this information to calculate the apparent temperature (AT, the perceived air temperature given the ambient humidity) using the formula AT = −2.653 + (0.994 × Ta) + (0.0153 × Tdp 2), where Ta is air temperature and Tdp is dew point temperature.22,23
We explored differences in susceptibility according to several city characteristics (see statistical analysis section below). The city characteristics examined were the mean ozone concentration and mean temperature during the warm months (period 1989–2000); the percentage of households with central air conditioning, calculated using data from the American Housing Survey of the US Census Bureau (period 1994–2002);24 the population density, calculated using data from the 2000 US Census; and the location of the city according to the US Census Divisions.25
City-Specific Case-Only Analysis
In a first stage, we fitted city-specific logistic regression models (PROC LOGISTIC in SAS 9.1, 2003; SAS Institute, Cary, NC) to assess modification of the risk of dying associated to ozone. Analyses were restricted to the warm months (May through September). Models included the hypothesized modifier (ie, a socio-demographic characteristic or a chronic condition) as the dependent variable and the mean ozone level of the last 3 days (lags 0 to 2) as a predictor. Lags 0 to 2 of ozone levels were chosen based on previous evidence that the largest effects of ozone on mortality occur in that timeframe.8 At least 2 of the 3 ozone measurements had to be available for the average ozone level to be computed; otherwise the average was set to missing and not used in the analyses. All models included a sine and cosine term with a 365.24-day period to capture any interactions between season and the characteristic being investigated. While seasonality per se is not a confounder in the case-only design,14 if the proposed modifier of the effect of ozone is a modifier of seasonality, confounding with the interaction of interest could occur if seasonality is not included in the model. For the same reason, all models included a term for apparent temperature at lag 0, since a previous case-only analysis examining susceptibility to temperature in the same population showed that some of the conditions here examined modified the effect of temperature at lag 0.17 Further adjustment of our models for lags 1 and 2 of apparent temperature yielded similar results and are not reported here.
Assessment of the Baseline Mortality Risk
Because we used a case-only design, the ozone coefficient in our logistic model above is equivalent to coefficient β 3 in Equation (1); it represents the additional increase in the risk of dying associated with ozone among subjects with the potential modifier. To gain insight into the magnitude of this “additional increase in risk” attributable to each effect modifier, we assessed the baseline mortality risk associated with ozone for all population groups in the same sample of cities and period included in the case-only analysis. To do this, we conducted a case-crossover analysis, a variation of the matched case-control design where each case subject serves as his own control on days when no event occurs.26 For each city, we fitted a conditional logistic regression model (PROC PHREG in SAS 9.1), choosing control days every third day within the same month of the same year.5 This time-stratified control day selection method has the advantage of providing a larger sample size (and thus more precise estimates) than the more commonly used time-stratified approach that additionally matches for day of the week.5 Models included the mean ozone level of lags 0 to 2 and controlled for day of the week and apparent temperature (lag 0 and average of lags 1 and 2).
Combination of the City-Specific Estimates and Assessment of Their Heterogeneity
We calculated a combined estimate for both the baseline mortality risk and the additional mortality risk attributable to each potential effect modifier. We did so in a random-effects meta-analysis using the restricted maximum likelihood method (metareg procedure in Stata, version 8; StataCorp, College Station, TX).27 This procedure calculates the overall effect estimate as a weighted average of the city-specific estimates with weights wi = 1/(vi + τ 2), where νi is the variance within city i and τ 2 is the between-cities variance. The procedure also provides the I 2 statistic, which represents the proportion of total variation in effect estimates that is due to between-cities heterogeneity.28
For those personal characteristics showing evidence of modification of the effect of ozone on mortality (ie, an overall estimate with a P value <0.05), we further investigated whether differences in city-specific estimates could be explained by city characteristics. To do this, we included each city characteristic as a covariate in the meta-regression models and used the estimated model coefficients to predict how each personal characteristic modified the effect of ozone on mortality in cities at the 25th and the 75th percentile of the distribution of each city characteristic. A significant difference between these 2 predicted values would indicate that susceptibility of that subpopulation to ozone effects varies according to the characteristics of the city of residence. Note that, while the focus of our paper is on modification of the association of ozone with mortality (ie, a two-way interaction), the meta-regression across cities is testing a three-way interaction (eg, Ozonex Diabetesx Percentage of Air Conditioning). This is because the city-specific coefficient being analyzed in the meta-regression is already the coefficient of a two-way interaction term.
Our study included a total of 2,729,640 nonaccidental deaths that occurred in 48 US cities from 1989 to 2000 during the months of May through September. As seen in Table 1, the majority of decedents were older than 65 years (71%), nonblack (81%), and with a low educational level (74%). The most common of the chronic health conditions examined was atherosclerotic heart disease (13%), followed by congestive heart failure (9%), and atherosclerotic cardiovascular disease (6%). Forty-one percent of deaths occurred outside the hospital, and the most common primary cause of death was cardiovascular disease (36%).
The median ozone level during the warm season ranged from 16.1 ppb in Honolulu, HI to 58.8 ppb in Charlotte, NC (Fig. 1). Only 4 cities (Albuquerque, Honolulu, San Francisco, and Spokane) did not exceed the National Ambient Air Quality Standard of 80 ppb during the study period. On the other hand, the highest peaks of ozone (>120 pbb) occurred in Los Angeles (14 days), Jersey City (5 days), Atlanta (4 days), and New Haven (3 days). Across all cities, the average difference between the 25th and the 75th percentile of the daily ozone level distributions was 21.5 ppb.
The combined estimate (for the entire population) of the baseline increase in mortality associated with a 10 ppb increase in the mean ozone level of the previous 3 days was 0.65% (95% confidence interval [CI] = 0.38% to 0.93%). Table 2 shows the estimates of the additional increase (or decrease) in ozone-related mortality expected to occur among certain population groups. Note that the additional risk for a specific subpopulation cannot be directly added to the baseline mortality risk to obtain the ozone-related mortality risk for that subpopulation, because the baseline mortality risk was calculated for all population groups (including the subpopulation of interest). Among the socio-demographic groups examined, the elderly presented the highest susceptibility to ozone, with a 1.10% (0.44% to 1.77%) additional increase in mortality (as compared with younger ages) when the mean ozone level of the previous 3 days increased 10 ppb. Other groups that were especially susceptible to die due to the ozone effects were black individuals (additional 0.53% [0.19% to 0.87%] compared with nonblacks) and women (additional 0.58% [0.18% to 0.98%] compared with men). Interestingly, while women over age 60 presented a 0.60% (0.25% to 0.96%) additional increase in mortality as compared with men in the same age group, no sex difference in risk was found among persons under 60 years of age. Of all chronic health conditions examined, only atrial fibrillation significantly increased the risk of death with high ambient ozone concentrations (additional 1.66% [0.03% to 3.32%] compared with those without a secondary diagnosis of atrial fibrillation) (Table 2).
For those susceptibility factors identified on Table 2, there was substantial heterogeneity in city-specific estimates (I 2 > 60%), except for blacks (I 2 = 24%). As seen in Figure 2, the extent to which some individual characteristics increased susceptibility to ozone varied according to some city characteristics. Overall, susceptibility factors had a more marked effect in cities with a low average ozone level. For instance, in a city with a mean ozone level of 42 ppb (ie, in the 25th percentile of the distribution) the additional increase in ozone-related mortality for the elderly was 1.48% (0.81% to 2.15%), whereas in a city with a mean ozone level of 51 ppb this additional increase was 0.45% (−0.27% to 1.19%). A low ozone level in the city also increased the susceptibility of blacks. The percentage of households with central air conditioning did not significantly modify the susceptibility of subpopulations except for those with atrial fibrillation. These had an additional increase in ozone-related mortality only in cities where air conditioning was common (additional 2.91% [0.86% to 5.01%] for a city in the 75th percentile of the air conditioning distribution). No important differences in susceptibility were observed according to the mean temperature of the city during the warm months, its population density, or its geographical location (results not shown).
In this large, multicity study we identified several subpopulations that appear particularly susceptible to the mortality effects of ambient ozone. These vulnerable groups were the elderly, women, black people, and those with atrial fibrillation. Most of these susceptibility factors had a more marked effect in cities with a low ozone level, suggesting that at high concentrations the effects of ozone are more uniform among the population.
Of all the socio-demographic groups investigated, the elderly were the most susceptible to the ozone effects. Their 1.1% additional increase in ozone-related mortality was actually quite large if compared with the overall mortality increase of 0.65% in our case-crossover analysis using the entire population, or with similar estimates from earlier studies.1 Thus, in our study the mortality risk associated with ozone was over 2-fold greater in the elderly than in younger people. This increased vulnerability of the elderly to ozone is consistent with studies among older populations in the United States,8 Latin America,9,11 and China.29 In a Chilean time-series study,9 the increase in susceptibility was monotonic across 3 increasingly older age groups; whereas in the United States, only those between 65- and 74-year-old showed an increased susceptibility as compared with the total population.8 Time-series studies in Europe, however, have not found age-group differences in the association between daily mortality and ozone.10,12
Little research has been conducted to assess differences in sex and race susceptibility to ambient ozone. We found increased susceptibility of women and blacks to mortality from ozone air pollution. Similarly, a study in China found a suggestion of higher ozone mortality among women.29 Results from a laboratory-based study investigating pulmonary responses to ozone among young volunteers showed no meaningful differences in response to 6 different concentrations of ozone among sex-race groups.30 However, when pulmonary responses were grouped across all ozone concentrations, larger decrements in the forced expiratory volume in one second (FEV1) were found for black subjects. Even though no sex differences were observed in young volunteers, a similar study among persons older than 54 years found a greater pulmonary responsiveness for women.31 Similarly, we found that sex differences in susceptibility to ozone occurred only among those 60 years or older. A possible explanation for these findings could be the involvement of hormonal changes in enhancing the effects of ozone in women after menopause. This hypothesis needs further investigation.
To our knowledge, our study is the first to report that persons with atrial fibrillation are particularly susceptible to die of the effects of ozone. Even though this subpopulation represents only a small fraction of all deaths, its increased susceptibility may offer some hints as to the mechanisms involved in the association between ozone and mortality, since alterations of the cardiac autonomic function have been postulated as one possible pathway.32 In addition, our findings are in agreement with results from other studies reporting an association between elevated ozone levels and the occurrence of episodes of supraventricular arrhythmias.33,34 In Boston (Massachusetts), a study among patients with implanted cardioverter defibrillators found that episodes of paroxysmal atrial fibrillation were associated with high ambient ozone concentrations in the concurrent hour.33 Thus, it seems possible that in patients with permanent atrial fibrillation, exposure to ambient ozone can have even more dramatic consequences.
Our findings are consistent with numerous studies aimed at identifying subpopulations susceptible to other air pollutants. The effects of particulate matter, black smoke, sulfur dioxide (SO2), nitrogen dioxide (NO2), and carbon monoxide (CO) on mortality have been reported to be more pronounced among the elderly.10,18,35–37 Women also appear to be particularly susceptible to the effects of particulate matter and black smoke,35,38 with the largest sex differences observed among individuals of advanced age.16,18 Race differences have been less apparent,38 with some studies observing differences in the effect of particulate matter only for specific mortality causes.18 Atrial fibrillation has not been previously examined as a modifier of the effect of air pollution, but was found to marginally increase susceptibility to extreme temperatures.17 On the other hand, our study failed to identify as effect modifiers certain health conditions that appear to confer susceptibility to particulate matter, such as diabetes,16,18 stroke,18 and congestive heart failure.16,18,36 A low educational attainment or a low socio-economic status has also been related to a larger susceptibility to die of the effects of particulate matter and SO2,18,39 but did not appear to modify the effect of ozone in our study.
Interestingly, the susceptibility factors identified in our study played a more important role in cities with low levels of ambient ozone. This suggests that ozone, at low concentrations, has a particularly detrimental effect on vulnerable populations, whereas at high concentrations differences between subgroups become less apparent, indicating equally large or small ozone effects for all populations. This pattern could explain why those with atrial fibrillation showed an increased susceptibility only when the prevalence of air conditioning in the city was high (ie, in cities where the population is expected to be less exposed to ozone). However, no such pattern was observed for the rest of the subgroups examined, suggesting that either air conditioning does not effectively protect against ozone effects or it protects homogeneously all population groups. In other words, even though it is possible that air conditioning protects against dying from ozone (as suggested by other studies40), the lack of a three-way interaction in our study suggests that the potential protection afforded by air conditioning did not differ among subpopulations.
A limitation of our study is the failure to control for particulate matter air pollution due to the limited number of measurements available for the study period. In most of our cities, particulate matter concentrations were available only from 1 day in 6. Even though previous studies have shown that particulate matter is not an important confounder of the association between mortality and ozone,6,8 if concentrations of particulate matter and ozone peaked simultaneously and some of the subpopulations studied were particularly vulnerable to the effects of particulate matter, then our results may be overestimations. Another limitation of the study is that data from mortality registries may be inaccurate, resulting in misclassification of mortality cause and probably under-reporting of contributing causes of death. If present, these problems will likely not vary daily with air pollution levels and thus our estimates would be expected to be biased downwards.
In conclusion, we confirmed in a large multicity study that the elderly are particularly susceptible to die of the effects of ambient ozone. We identified other vulnerable subpopulations to ozone including women, blacks, and those with atrial fibrillation. Our study suggests that in less polluted cities the adverse effects of ozone are more manifest among the vulnerable subpopulations, whereas in cities with high ozone concentrations differences between subgroups become less apparent. Identifying vulnerable subpopulations and the impact of ozone on these subpopulations will help in establishing air quality standards that will better protect these groups.
1. Bell ML, Dominici F, Samet JM. A meta-analysis of time-series studies of ozone and mortality with comparison to the national morbidity, mortality, and air pollution study. Epidemiology
2. Brunekreef B, Holgate ST. Air pollution and health. Lancet
3. Gryparis A, Forsberg B, Katsouyanni K, et al. Acute effects of ozone on mortality from the “air pollution and health: a European approach” project. Am J Respir Crit Care Med
4. Levy JI, Chemerynski SM, Sarnat JA. Ozone exposure and mortality: an empiric bayes metaregression analysis. Epidemiology
5. Medina-Ramon M, Zanobetti A, Schwartz J. The effect of ozone and PM10 on hospital admissions for pneumonia and chronic obstructive pulmonary disease: a national multicity study. Am J Epidemiol
6. Thurston GD, Ito K. Epidemiological studies of acute ozone exposures and mortality. J Expo Anal Environ Epidemiol
7. Schwartz J. How sensitive is the association between ozone and daily deaths to control for temperature? Am J Respir Crit Care Med
8. Bell ML, McDermott A, Zeger SL, et al. Ozone and short-term mortality in 95 US urban communities, 1987–2000. JAMA
9. Cakmak S, Dales RE, Vidal CB. Air pollution and mortality in Chile: susceptibility among the elderly. Environ Health Perspect
10. 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
11. O'Neill MS, Loomis D, Borja-Aburto VH. Ozone, area social conditions, and mortality in Mexico City. Environ Res
12. Parodi S, Vercelli M, Garrone E, et al. Ozone air pollution and daily mortality in Genoa, Italy between 1993 and 1996. Public Health
13. Armstrong BG. Fixed factors that modify the effects of time-varying factors: applying the case-only approach. Epidemiology
14. Schwartz J. Who is sensitive to extremes of temperature? A case-only analysis. Epidemiology
15. 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
16. Bateson TF, Schwartz J. Who is sensitive to the effects of particulate air pollution on mortality? A case-crossover analysis of effect modifiers. Epidemiology
17. Medina-Ramon M, Zanobetti A, Cavanagh DP, et al. Extreme temperatures and mortality: assessing effect modification by personal characteristics and specific cause of death in a multi-city case-only analysis. Environ Health Perspect
18. Zeka A, Zanobetti A, Schwartz J. Individual-level modifiers of the effects of particulate matter on daily mortality. Am J Epidemiol
19. Corradi M, Alinovi R, Goldoni M, et al. Biomarkers of oxidative stress after controlled human exposure to ozone. Toxicol Lett
20. Nehls GJ, Akland GG. Procedures for handling aerometric data. J Air Pollut Control Assoc
21. Schwartz J. Assessing confounding, effect modification, and thresholds in the association between ambient particles and daily deaths. Environ Health Perspect
22. Kalkstein LS, Valimont KM. An evaluation of summer discomfort in the United States using a relative climatological index. Bull Am Meteorol Soc
23. Steadman RG. The assessment of sultriness. Part II: effects of wind, extra radiation and barometric pressure on apparent temperature. J Appl Meteorol
26. Maclure M. The case-crossover design: a method for studying transient effects on the risk of acute events. Am J Epidemiol
27. Thompson SG, Sharp SJ. Explaining heterogeneity in meta-analysis: a comparison of methods. Stat Med
28. Higgins JP, Thompson SG. Quantifying heterogeneity in a meta-analysis. Stat Med
29. Zhang Y, Huang W, London SJ, et al. Ozone and daily mortality in Shanghai, China. Environ Health Perspect
30. Seal E Jr, McDonnell WF, House DE, et al. The pulmonary response of white and black adults to six concentrations of ozone. Am Rev Respir Dis
31. Reisenauer CS, Koenig JQ, McManus MS, et al. Pulmonary response to ozone exposures in healthy individuals aged 55 years or greater. JAPCA
32. Routledge HC, Ayres JG. Air pollution and the heart. Occup Med (Lond)
33. Rich DQ, Mittleman MA, Link MS, et al. Increased risk of paroxysmal atrial fibrillation episodes associated with acute increases in ambient air pollution. Environ Health Perspect
34. Sarnat SE, Suh HH, Coull BA, et al. Ambient particulate air pollution and cardiac arrhythmia in a panel of older adults in Steubenville, Ohio. Occup Environ Med
35. Filleul L, Baldi I, Dartigues JF, et al. Risk factors among elderly for short term deaths related to high levels of air pollution. Occup Environ Med
36. Goldberg MS, Burnett RT, Bailar JC III, et al. Identification of persons with cardiorespiratory conditions who are at risk of dying from the acute effects of ambient air particles. Environ Health Perspect
. 2001;109(suppl 4):487–494.
37. Katsouyanni K, Touloumi G, Samoli E, et al. Confounding and effect modification in the short-term effects of ambient particles on total mortality: results from 29 European cities within the APHEA2 project. Epidemiology
38. Zanobetti A, Schwartz J. Race, gender, and social status as modifiers of the effects of PM10 on mortality. J Occup Environ Med
39. Finkelstein MM, Jerrett M, DeLuca P, et al. Relation between income, air pollution and mortality: a cohort study. CMAJ
40. Zanobetti A, Schwartz J. Mortality displacement in the association of ozone with mortality: an analysis of 48 cities in the United States. Am J Respir Crit Care Med