Below is the uncorrected machine-read text of this chapter, intended to provide our own search engines and external engines with highly rich, chapter-representative searchable text of each book. Because it is UNCORRECTED material, please consider the following text as a useful but insufficient proxy for the authoritative book pages.
4 Contributions of Relevant Health Studies to the Estimation of Reductions in Premature Mortality INTRODUCTION This chapter reviews evidence of ozone-mediated effects on mortality. It builds on the wealth of toxicologic and epidemiologic evidence showing the deleterious effects of ozone on an array of health effects, including oxidative damage to DNA, inflammation, oxidative stress, heart-rate variability, decrease in lung function, such respiratory conditions as bronchitis and wheezing, in- crease in medication use, and exacerbation of respiratory diseases leading to emergency-room visits and hospitalization. Those effects have consistently been found in toxicologic and epidemiologic studies, independently of study design and location, and therefore constitute compelling and well-accepted evidence that ozone adversely affects health. In contrast, until its regulatory impacts assessment (RIA) for the proposed ozone national ambient air quality standards (NAAQS) released in 2007, the Environmental Protection Agency (EPA) had not included mortality results in it primary estimates of the benefits of reductions in acute exposure to ambient ozone, because of uncertainty about the appropriate interpretation of those re- sults. Previously, EPA had included epidemiologic studies that link ambient ozone concentrations with premature deaths in sensitivity analyses. The shift is attributable to consistent findings from multicity time-series studies and meta- analyses conducted over the past few years that suggested a modest but highly significant increase in mortality associated with relatively short-term exposure to ambient ozone. This chapter reviews evidence of ozone-related death with regard to its biologic plausibility and reviews findings from the multicity time-series studies and meta-analyses. It also assesses evidence of the existence of susceptible groups and discusses how such factors as personal and population average- 75
76 Ambient Ozone and Mortality: Estimating Risk-Reduction Benefits exposure error, epidemiologic designs and methods of analysis, measures of the effects of ozone exposure on death, and uncertainty contribute to our ability to estimate ozone-related mortality risk and to apply the estimates to risk and bene- fits analysis. BIOLOGIC PLAUSIBILITY The mechanisms by which ozone can damage health and possibly lead to mortality remain a subject of considerable clinical interest. Toxicologic studies have attempted to identify the mechanisms by using a variety of techniques, from ultrastructural, biochemical, and cytologic analyses to in vivo measure- ment of airflow mechanics. The techniques generally examine the effects of con- trolled ozone exposures of less than 8 h and, to a smaller extent, exposures of several days to increased ambient concentrations. The findings of a number of toxicologic studies provide insights to suggest the potential for a number of events by which ozone exposure could lead to in- creased mortality, including lung inflammation leading to local pulmonary com- promise (Devlin et al. 1991; Holz et al. 1999; Koren et al. 1991; Ratto et al. 2006; Bosson et al. 2007), worsening of pre-existing cardiopulmonary disease, systemic release of mediators that affect adverse cardiovascular events (Hollingsworth et al. 2007), effects on the autonomic nervous system that could contribute to increased airway responsiveness (Chen et al. 2003) or reduced heart rate variability, and an increase in factors that lead to vascular changes (Chuang et al. 2007). Definitive proof of such occurrences leading to mortality are not found in the current scientific literature, however, there is ample evi- dence ozone can induce mechanisms through a sequence of events to trigger oxidative stress or inflammation pathways, which can contribute to release of pulmonary or systemic mediators. The events may further result in the exacerba- tion of a pre-existing respiratory or cardiovascular condition. Chen et al. (2004) found increased sensitivity to aeroallergens in a subgroup of asthmatics follow- ing exposure to 200 ppb ozone. There exists the biological plausibility that for a few individuals with pre-existing cardiopulmonary or chronic respiratory dis- ease, such oxidative or inflammatory events could also cause death, via the onset of an asthmatic attack (Selgrade et al. 2008), myocardial infarction, or vascular event. Inflammatory events have been shown to occur for 20-30 h after inhala- tion of ozone even at ambient concentrations in the range of 0.08-0.10 ppm (Ratto et al. 2006; Chuang et al. 2007). There is less information on the airway inflammatory response to ozone at low concentrations, such as at or below 0.08 ppm. The oxidative stress and inflammation cascade is initiated by an ozone- mediated airway and lung tissue response (Chuang et al. 2007). Ozone can initi- ate the cascade immediately after inhalation when it comes in contact with epithelial lining fluids and cellular membranes, most often at nasal epithelial surfaces and at the junction of the bronchioles and the alveolar region (centriaci-
Study Contributions to the Estimation of Reduced Premature Mortality 77 nar region). At the air-liquid interfaces, ozone can interact with unsaturated fatty acids to form lipid ozonation products, which in turn can activate lipases that lead to production and release of cell-signal transduction molecules and proin- flammatory mediators. With oxygen species and reactive oxygen metabolites generated by exposure, those products of ozone exposure may lead to oxidative stress and the upregulation of transcription factors, such as NF-ÎºB and proin- flammatory genes. Correspondingly, antioxidant enzymes (such as glutathione enzymes, superoxide dismutase, and catalase) and nonenzymatic factors (such as ascorbic acid and Î±-tocopherol) that react with ozone directly have been shown to attenuate oxidant damage to airway lipid membranes. Biochemical evidence of ozone-mediated inflammation is provided by animal and human studies that show increases in downstream activation prod- ucts of the innate immune system, including fibronectin, elastase, plasminogen activator, tissue factor, factor VIII, C3a fragment of complement, prostagland- ins, interleukin-1 (IL-1), tumor-necrosis factor a, IL-6, IL-8, and granulocyte macrophage colony-stimulating factor (GM-CSF) in alveolar lavage fluid (Koren et al. 1989; Aris et al. 1993; Devlin et al. 1994). Inflammatory effects of ozone occur in people who are healthy and in people who have increased vul- nerability because of host (genetic) factors or pre-existing cardiopulmonary dis- ease, such as asthma. For example, ozone exposure at 100 ppb for 2 h leads to an acute neutrophilic inflammation of the airway in allergic asthmatic subjects (Depuydt et al. 2002). In addition, pre-exposure of these subjects (atopic asth- matics) to ozone potentiates the late-phase eosinophilic response to the allergen. The acute lung response to ambient ozone has many clinical features in common with asthma, and ozone exposure also leads to increase in proinflammatory cy- tokines (Koren et al.1989), cellular inflammation in airway tissues, and bron- chial hyperresponsiveness (Foster et al. 2000). Perhaps as a result, exposure to ambient ozone can exacerbate pre-existing allergic asthma (White et al. 1994). Cultured human epithelial cells (HBECs) from asthmatic and nonasthmatic sub- jects exposed for 6 h to ozone at 100 ppb had increases in inflammation- associated mediators IL-8, GM-CSF, and intercellular adhesion molecule 1 (Bayram et al. 2001). HBECs from asthmatic but not nonasthmatic people also showed increased expression of the chemotactic cytokine CCL5. Increased ex- pression of CCL5 suggests that ozone exposure may increase eosinophilic in- flammation, exacerbate asthmatic symptoms, and lead to pathologic changes in the epithelium (Bayram et al. 2001). Support for that suggestion is provided by studies that show human asthmatic subjects to have increased eosinophilic in- flammation of the lower airways after prolonged acute exposure to ozone at 160 ppb. Epithelial cytokine expression of IL-5, GM-CSF, and IL-8 has also been found to be higher in asthmatic subjects after laboratory exposure to ozone for 2 h at 200 ppb (Bosson et al. 2003). Numerous laboratory chamber studies have validated the reproducible na- ture of the inflammatory lung response (Devlin et al. 1991; Koren et al. 1991; Holz et al. 1999). That finding has been supported by additional studies in hu- mans using therapeutic ablation of inflammatory effects, which have also helped
78 Ambient Ozone and Mortality: Estimating Risk-Reduction Benefits uncover essential mechanisms of injury. Although inhalation of the corticoster- oid budesonide was not found to limit ozone-induced airway inflammation in healthy subjects, budesonide reduced neutrophilic infiltration in patients with mild asthma as evidenced by changes in the inflammatory cellularity of induced sputum (Nightingale et al. 2000; Vagaggini et al. 2001). Inhaled and oral corti- costeroids have also been found to ablate ozone-induced neutrophilic inflamma- tion in selected healthy subjects who were characterized as neutrophilic ozone- responsive (Holz et al. 2005). In mildly asthmatic subjects, pretreatment and posttreatment with apocynin aerosol, an inhibitor of the nictotinamide adenine dinucleotide phosphate (NADPH) oxidase complex present in membranes of inflammatory cells (eosinophils and neutrophils) and a source of reactive oxygen species, prevented ozone-induced airway hyperresponsiveness to methacholine (Peters et al. 2001). There has been relatively little research done on cardiovascular effects of ozone as compared with the amount done on pulmonary effects. Gong et al. (1998) used an ozone chamber study to monitored cardio-vascular effects in humans. In response to exposures to 300 ppb ozone, no major acute cardiovas- cular changes were found to be induced in either hypertensive or control sub- jects. It was noted, however, that ozone can increase myocardial work and im- pair pulmonary gas exchange. Together, those findings constitute supporting evidence that ozone expo- sure induces a cascade of events that increase oxidative stress and inflammation. As observed in numerous cohort and toxicologic studies, it is plausible that in- creased oxidative stress and inflammation can influence cardiac risk and ulti- mately mortality through increased autonomic dysregulation, vascular dysfunc- tion, atherosclerosis, and arrhythmogenesis (Ridker et al. 1998, 2000; Pradhan et al. 2001, 2002). Those downstream effects may be most pronounced in people who have pre-existing diseases, in that ozone exposure may worsen a disturbed respiratory, vascular, or cardiac system, making them more vulnerable to ozoneâs adverse effects. The adverse effects can occur after intermittent exposure to ozone, which is a typical pattern of exposure to this pollutant. Although the effects of ozone may accumulate over a lifetime, it is well established that gaps in ozone expo- sure lead to a decrease in innate resistance to its adverse effects. With the onset of each ozone-exposure episode, the injury process will be repeated, initiating once again the biologic events listed above. Such a cellular response to ozone means that each exposure event poses a renewed risk of a cascade of activities that could decrease resiliency and possibly enhance vulnerability, resulting in premature death. OZONE-MORTALITY STUDIES Given the evidence from toxicologic studies, it is biologically plausible that ozone exposures are also related to death in human populations. From a
Study Contributions to the Estimation of Reduced Premature Mortality 79 biologic and epidemiologic perspective, there are two main ways in which air pollutants, such as ozone, may affect mortality: as a consequence of acute ef- fects that cause death in the near term and as a consequence of chronic patho- physiologic changes that ultimately lead to death. The two may well overlap and be interrelated, but the distinction is useful because it is relevant to the interpre- tation and quantification of risks. Acute Studies Acute effects of ozone are those observed within a few hours or days of a rather short exposure. In those cases, exposure is typically defined as the maxi- mum concentration on a given day, the mean across the 8 consecutive hours with the highest concentrations, or the 24-h average concentration. Under this acute-effect model, if death occurs, it is a consequence of the exposure. An acute exposure may also trigger a condition of poor health or frailty that can lead to death within a few days or weeks if intervention is not successful; in this de- layed case, death may be considered a subacute effect of an acute (short-term) exposure. Time-series analysis is by far the most common approach to investigate the acute and subacute effects of ozone on mortality (see Box 4-1). The most systematic and comprehensive time-series analyses were two multicity studies (Bell et al. 2004; Ito et al. 2005) and three meta-analyses (Bell et al. 2005; Ito et al. 2005; Levy et al. 2005) solicited and commissioned by EPA.1 The three re- search teams conducting the meta-analyses were provided with the same data- bases from EPA but conducted the analyses separately and did not communicate with each other about their methods or findings until the studies were com- pleted. Those studies examined ozone-mediated mortality while addressing pre- viously unresolved issues related to confounding, exposure variability, and model specifications. Although the studies used different approaches, their ma- jor results were similar: each found a statistically significant relationship be- tween ozone exposure and premature mortality that appears robust after control- ling for exposure to particulate matter with a diameter of 10 Âµm or less (PM10). The similar estimates from these three meta-analyses suggest not only an indica- tion of stable estimates in the literature, but also indicate that the meta-analytic approach is generally insensitive to analytical decisions, which provides support for the meta-analytical methods. Together, their results were cited by EPA as constituting strong evidence of a link between short-term exposure at concentra- tions below the NAAQS and premature mortality. However, many questions remain, specifically regarding the size and significance of the ozone-mortality effect and the implications of the findings for ozone benefits analysis. Study 1 The study by Ito et al. (2005) included both a meta-analysis and a time-series analy- sis; results from the time-series analysis were intended to help to explain issues identified in the meta-analysis.
80 Ambient Ozone and Mortality: Estimating Risk-Reduction Benefits BOX 4-1 Definitions of Time-Series Analysis and Meta-Analysis Time-series analysis describes and models the behavior of observa- tions that are occurring sequentially and is analysis of the temporal relation between ozone and mortality data at different times. Moreover, time series analysis in air pollution studies is typically an approach in which daily mor- tality or morbidity counts are expressed as a function of covariates, includ- ing at least one indicator of air pollution in regression models, which can control for confounding effects of seasonality, long term trend and weather. The multicity time-series model has several advantages over multicity meta-analyses. Instead of having to rely on the different analytic ap- proaches and assumptions used by researchers of completed studies, the multi-city time-series model provides the ability to use an identical model structure, exposure lags, control for confounders, and statistical criteria for data from all cities. Furthermore, the time-series model can include factors that explain differences in effect estimates among the cities, such as re- gion, housing characteristics, and measurement method. Meta-analysis is a statistical technique used to aggregate, summa- rize, and review previous quantitative research. Through meta-analysis, a wide variety of questions can be investigated, assuming that a representa- tive body of primary research exists. Selected parts of the reported results of primary studies (effect sizes or other characteristics) are analyzed as other data are: descriptively and then inferentially to test hypotheses. The appeal of meta-analysis is that it in effect combines research on a topic into one large study with many subjects. The danger is that in aggregating a large set of studies, the construct definitions can become imprecise and the results difficult to interpret fully (Neill 2006). Moreover, a major issue with meta-analysis is publication bias, which refers to the greater likelihood that studies with statistically significant positive results will be published as compared with studies reporting negative results. Thus, estimates from pooled meta-analysis may overestimate the association between ozone and mortality. Like multicity time-series analyses, meta-analyses use a multilevel model to examine pollution-effect associations. City specific estimates (level 2) are obtained from time series data, (level 1) for each city with a generalized linear model analysis. These are then pooled into a national average estimate using a Bayesian hierarchical model in which stages of the hierarchy have assumed normal distributions. descriptions, findings, limitations, and remaining issues related to the five com- prehensive analyses are presented below. Multicity Time-Series Studies Two multicity time-series studies of ozone and mortality were conducted. Bell et al. (2004) used the National Morbidity, Mortality, and Air Pollution
Study Contributions to the Estimation of Reduced Premature Mortality 81 Study (NMMAPS) database to conduct a large time-series analysis with data from 95 U.S. cities over the period 1987-2000. Ito et al. 2005 conducted a time- series study with data from seven U.S. metropolitan areas over the period 1985- 1995. Bell et al. (2004). As shown in Table 4-1, Bell et al. used data and meth- ods developed for the NMMAPS study to estimate a national average relative rate of mortality (noninjury mortality and cardiovascular and respiratory mortal- ity) associated with short-term average ambient ozone concentrations in 1987- 2000 in 95 large U.S. urban communities made up of almost 40% of the U.S. population. The 95 areas were chosen on the basis of availability of daily ozone data from EPAâs Aerometric Information Retrieval Service (now known as the Air Quality System database). Descriptive statistics on each community are pro- vided at (iHAPSS 2005). For ozone, the 24-h average, maximum 8-h, and maximum hourly concentrations were calculated for each day. In several loca- tions, ozone was measured only during the peak ozone season, April-October. A two-stage statistical model was used to estimate a national average association between short-term ambient ozone concentrations and mortality risks, account- ing for weather, seasonality, long-term trends, and PM10. In the first stage, dis- tributed-lag overdispersed Poisson regression models were used for estimating community-specific relative rates of mortality associated with exposure to ozone in the week prior to death. In the second stage, the community-specific relative rates were combined to generate a national average estimate of the association between ozone and mortality that accounts for within-community and across- community variability. Analyses were performed with multiple ozone-concentration metrics (1-h maximum, 8-h maximum, and 24-h average), which were highly correlated with each other; variations were most likely due to weather and sources of ozone. Temperature was modeled as a natural cubic spline (a type of mathematical function) of a dayâs temperature and the average of the three previous daysâ temperatures. Analyses were based on single-day lags and constrained and un- constrained distributed-lag models for all 95 communities, for all communities in warm months (April-October), and for the 55 communities on which yearly ozone data were available. The sensitivity of model results to choice of model, temperature, age (all ages and less than 65, 65-74, and over 75 years), and PM10 was also assessed. The sensitivity analysis with PM10 was limited to days for which PM10 and ozone data were available (1 in 6 days). Results showed that average short-term changes in ozone are significantly associated with premature mortality and that the statistical association is robust to adjustments for PM10, weather (temperature), and seasonality. It should be noted that robustness for PM10 may have been hindered by lack of daily moni- toring data. Ozone-mediated risks were greater for cardiovascular and respira- tory mortality than for total mortality, and effects were larger at lag 0 d than lag 1 or 2 d. Bell et al. examined the 95 communities using a constrained distrib- uted-lag model and found a statistically significant 0.52% (95% posterior
82 TABLE 4-1 Summaries of Recent Studies of Acute Effects of Ozone on Mortality Risk Design/ (95% Confidence Number of Exposure Health Interval) Author Source Studies or Cities Period Metrics Outcomes per 10 ppb Bell, McDermott, JAMA 2004; Time-series study 1987-2000 Cumulative Noninjury 0.52% Zeger, Samet, 292(19): of 95 U.S. large exposure of mortality (0.27-0.77%) and Dominici al. 2372-2378 urban areas previous week Cardiovascular 0.64% and respiratory (0.31-0.98%) mortality Ito, Leon, Epidemiology Meta-analysis 1990-2003 24-h average Nonaccidental 0.8% and Lippmann 2005;16(4): of 43 studies mortality (0.55 -1.0%) 446-457 (international) Time-series 1985-1995 0- and 1-d lag, 0.52-1.0%a analysis of 7 U.S. 24-h average cities Levy, Chemerynski, Epidemiology Meta-regression Pre-October 2003 1-h maximum All-causes 0.41% and Sarnat 2005;16(4): of 28 time-series mortality (0.32-0.52%) 458-468 studies 1-h maximum: 0.84% (international) summer; (0.57-1.09%) 1-h maximum: -0.04% winter (-0.34 to 0.28%) Bell, Dominici, Epidemiology Meta-analysis 1990-June 2004 0-, 1-, and 2-d lag Total mortality 0.87% and Samet 2005;16(4): of 39 time-series or 2-d average of (0.55-1.18%) 436-445 studies lags (0, 1, and 2 d) a Range of risk estimates without confidence intervals.
Study Contributions to the Estimation of Reduced Premature Mortality 83 interval: 0.27%-0.77%) increase in non-accidental mortality for a 10 ppb in- crease in the previous weekâs daily average ozone concentration. The increase in cardiovascular and respiratory mortality was somewhat higher at 0.64% (95% PI, 0.31%-0.98%). Interestingly, Bell et al. found a lower increase in total mor- tality of 0.39% (95% posterior interval: 0.13%-0.65%) in those communities during the ozone high season (April through October). Results were similar for the unconstrained models. The mortality effect of a single day of ozone exposure appeared to be dis- tributed over several days. The study further suggested that ozone-mortality ef- fects can be calculated separately from PM-mortality effects because ozone risk estimates were robust to adjustment for PM10. Bell et al. (2007) also reported the 95 Bayesian city-specific estimates of risk and the standard deviation of these city-specific estimates. The standard deviation for the constrained distributed lag model was 0.64%. Thus 95% of the true city-specific risk effects lie in the inter- val -0.73% to 1.77%. Bell at al. (2007) further investigated whether PM is a confounder of ozone and mortality using data for 98 U.S. urban communities from 1987 to 2000. They concluded that neither PM10 nor PM2.5 is a likely con- founder of observed ozone and mortality relationships. They recommended that further investigation is needed of potential confounding of the short-term effects of ozone on mortality by PM chemical composition. As part of a larger study of ozone and mortality, Ito et al. (2005) conducted a time-series analysis with data from seven U.S. cities (Chicago, Detroit, Houston, Minneapolis-St Paul, New York City, Philadelphia, and St Louis) for the period 1985-1994 (Table 4-2). Those cities were chosen because of the availability of data on daily (or nearly daily) PM10, which went back to 1985 in most cities, and their relatively large populations. It was necessary to obtain daily rather than every-6th-day samples (the usual sampling frequency for most U.S. cities) of PM to obtain reasonable statistical power (Ito, personal communication, 12/04/07). In each of the cities, ambient ozone was measured year-round, and daily PM10 data were available for all cities except New York. PM2.5 data were available for selected years for Philadelphia (1992-1995) and St. Louis (1985-1989). Ito et al. (2005) used the data specifically to examine the sensitivity of ozone-mortality risk estimates to several factors, including season, alternative weather models and adjustments (of which four were examined), and confounding by PM. Results of the sensitiv- ity analyses were intended to supplement a corresponding meta-analysis. To characterize the ozone-PM relationships across seasons, the authors computed the mean ozone concentration for each quintile of PM in summer and winter. To examine the sensitivity of ozone-mortality risk estimates to alterna- tive weather-model specifications and temporal adjustments, they used a Poisson generalized linear model that adjusted for temporal trends, day of week, and weather effects. The average of 0- to 1-d lagged 24-h average ozone concentra- tion was included. A smoothing function of days using natural splines was in- cluded to adjust for seasonal cycles and other temporal trends, and various weather models from the literature were investigated.
84 Ambient Ozone and Mortality: Estimating Risk-Reduction Benefits TABLE 4-2 Summaries of Recent Time Series Studies of Acute Effects of Ozone on Mortality Authors Bell, McDermott, Zeger, Ito, Leon, and Lippmann Samet, and Dominici. Study design (methods) Time-series study Time-series study Type of model(s) Hierarchic (mixed-effects), Poisson generalized linear distributed Lag model Numbers of studies or cities 95 U.S. large urban areas 7 U.S. large urban areas Study periods 1987-2000 1985-1995 Exposure level County level Metropolitan areas Exposure metric 24-h average 24-h average 1. Maximum 8-h average 2. Maximum hourly concentration Time of study 1. April-October Whole year 2. Whole year Health outcomes 1. Non-injury-related Total nonaccidental mortality mortality 2. Cardiovascular and respiratory mortality Lag time days in days 0, 1, 2, 3, up to a week Adjustment for temperature Yes Yes Adjustment for humidity No No Adjustment for season Yes Yes Consideration of age strata Yes Yes Adjustment for PM10 Yes Yes Adjustment for PM2.5 Yes Yes Adjustment for other No No pollutants RESULTS Non-injury-related 0.39% (95% posterior 1.0% (0 .55-1.40%) with 0-, mortality (per 10-ppb interval, 0.13-0.65%) with 1-d lag.(year-round) change) up to a week lag (April- 1-h maximum October) 8-h maximum 24-h average Cardiovascular and 0.64% (0.31-0.98%) with respiratory mortality up to a week lag (per 10-ppb change) 1-h maximum 8-h maximum 24-h average
Study Contributions to the Estimation of Reduced Premature Mortality 85 Results from the multicity analysis show short-term associations between ozone and mortality with substantial heterogeneity across cities. Pooled analysis for the six cities with corresponding daily PM data resulted in an all-year, ozone-only mortality effect estimate of 2.0% (1.1-2.9%) for the quintile model and 1.0% (0.0-2.0%) for the model that used four smoothing terms per 20-ppb increase in the average of 0- and 1-d lag 24-h ozone. When stratified by season, the excess risk estimates were higher in summer than in the whole year and in cold seasons, when the estimates were low or null. The potential confounding between ozone and PM did not substantially affect ozone risk estimates. Results further showed that the use of different weather models could ac- count for a twofold difference in overall effects estimates (0.24-0.49%) in analy- ses that included yearly data, with the quintile temperature model producing the largest estimate and the model that used four smoothing terms producing the smallest. Regardless of the adjustment for weather and for PM10, PM2.5, and sea- son, however, large city-to-city variation in ozone-associated mortality risk es- timates persisted. The heterogeneity was thought to be due to corresponding heterogeneity in factors known to vary with city, such as air-conditioning use, study population, ozone, and other photochemical oxidants. The relationship between ozone and PM10 was characterized by a positive slope in summer and a negative, shallower slope in winter. The relationships between temperature and ozone were generally J-shaped. Smoothing for tempo- ral trend did not generally alter the risk estimates. In addition to the largely U.S.-based multicity studies (Bell et al. 2004 and Ito et al. 2005), Gryparis et al. (2004) collected data on daily ozone concentra- tion, daily number of deaths, confounders, and potential effect modifiers for 23 European cities for at least three years since 1990. Effect estimates were ob- tained for each city with city-specific models and were then combined using second stage regression models. No significant effects were noted during the cold half of the year but for the warm season, an increase in the one hour ozone concentration by 10 ug/m3 was associated with a 0.33% increase in total daily number of deaths. When considering the number of cardiovascular deaths, there was a 0.45% increase. When considering respiratory deaths, there was a 1.13% increase. The corresponding figures for 8-h ozone levels were similar. Meta-analyses Three meta-analyses were carried out to obtain a summary or composite estimate of ozone-associated mortality risks while explaining observed hetero- geneity in risk estimates. The meta-analyses used different statistical techniques and datasets to aggregate results of multiple time-series studies of changes in ozone and mortality (Tables 4-1, 4-2, and 4-3). Ito et al. (2005) considered 43 time-series studies in the U.S. and abroad that were conducted in 1990-2003. Levy et al. (2005) considered 48 risk estimates from 28 time-series studies
86 TABLE 4-3 Summary of Meta-analyses % Change in Daily Mortality Number of Adjustment per 10 ppb Studies Study Season of Exposure for Other change in ozone Reference (or Cities) Areaa Period Study Metric Lag (d) Pollutants Mortality (95% CI) Levy et al. 46 Inter 1973-1999 All 1-h maximum 0-2 NA All causes 0.41 2005 (4:3:2 1,8,24 (0.31-0.51) conversion) 27 N. Am All 1-h maximum 0-2 NA All causes 0.41 (0.29-0.53) 31 Inter All 1-h maximum 0 NA All causes 0.51 (0.37-0.66) 15 Inter All 1-h maximum 1-2 NA All causes 0.25 (0.18-0.31) 14 Inter Summer 1-h maximum 0-2 NA All causes 0.84 (0.57-1.10) 10 Inter Winter 1-h maximum 0-2 NA All causes -0.04 (-0.33 to 0.27) 23 Inter All 1-h maximum 0-2 NA All causes 0.55 23 Inter All 1-h maximum 0-2 (0.37-0.72) 0.31 (0.20-0.45)
Ito et al. 43 Inter 1990-2003 Year round 24-h average Up to 3 NA All causes, 0.80 2005 non-accidental (0.6-1.0) Maximum 1-h All causes, 0.39 average non-accidental (0.26-0.51) 15 Am 1983-1999 Year round 24-h average PM10 All causes, 0.75 non-accidental (0.4-1.1) Maximum 1-h All causes, N/A average non-accidental Bell et al. 39 Various 24-h average PM10 All causes, 0.87 2005 PM2.5 non-accidental (0.55-1.18) Cardiovascular 1.11 (0.68-1.53) Respiratory 0.47 (0.51-1.47) Maximum 8-h NA average Maximum 1-h NA average a Inter = International; Am = America. 87
88 Ambient Ozone and Mortality: Estimating Risk-Reduction Benefits with study data from 1973 to 1999, excluding publications from NMMAPS. The meta-analysis conducted by Bell et al. (2005) used 144 effect estimates from 39 time-series studies published from 1990 to June 2004, also excluding studies based on NMMAPS. Each of the meta-analyses examined statistical concerns (confounding, collinearity, and possible interaction effects) related to modeling of a dependent response variable. Ito et al. (2005). Ito et al. conducted a meta-analysis of single-city studies that investigated the short-term association between ozone and nonaccidental mortality (at all ages or over the age of 65 y). From each study, a âpreselectedâ exposure lag (up to 3 d for consistency) was used in the meta-analyses to avoid selecting the most statistically significant lag and artificially biasing the estimate upward. Ozone-related changes in mortality risk were summarized by season. The authors conducted analyses to examine the sensitivity of the results to using different number of nonparametric smoothing terms in the generalized additive model. The combined random-effects estimate from the 43 studies included in the meta-analysis was 1.6% (95% CI, 1.1-2.0%) excess mortality per 20-ppb in- crease in 24-h average ozone (roughly 0.80% excess mortality per 10-ppb in- crease). There was large heterogeneity in ozone-mortality risk estimates across studies that was potentially related to city-specific factors, such as mean tem- perature, or model specifications. Correspondingly, seasonal differences in ozone-mortality estimates were found. For the 10 studies that reported ozone risk by season, the ozone risk estimates were typically larger in summer than in winter, with random-effects estimates of 2.2% (95% CI, 0.8-3.6%) and 3.5% (95% CI, 2.1-4.9%) per 20-ppb increase in 24-h average ozone for all year and the warmer season, respectively. Although adjustment for PM10 had little effect on the ozone-mortality association for the whole year, the study showed poten- tial confounding of ozone effects by PM10 in the warm season. That season- specific confounding by PM10 or temperature might explain the reported ob- served negative ozone-mortality associations. Levy et al. (2005). Levy et al. applied an empiric Bayes metaregression to estimate the health effects of ozone on all-causes mortality and to assess predic- tors of between-study variability. Their meta-analysis included 48 city-specific relative-risk estimates from 28 studies conducted in North America and Europe and published before October 2003. In the metaregression, a hierarchic linear model with known level-1 variances was applied, and exposures were assessed with the 1-h maximum ambient ozone concentration. The meta-analysis found a significant overall 0.41% (95% CI, 0.32-0.52%) increase in daily mortality per 10-ppb increase in 1-h maximum ozone concentrations; results in North Ameri- can and European cities were similar. Results from the metaregression also demonstrated seasonal heterogeneity in ozone health effects, with a significant 0.84% (95% CI, 0.57-1.09%) increase per 10-ppb increase in 1-h maximum ozone in summer and a nonsignificant -0.04% (95% CI, -0.34 to 0.28%) de- crease in winter for the same increase in maximum ozone. Furthermore, results with the metaregression model suggested that between-study variability in
Study Contributions to the Estimation of Reduced Premature Mortality 89 ozone-related mortality could be partially explained by between-study differ- ences in exposure lag times, air-conditioning prevalence, and ozone and other air-pollutant relationships, with effect estimates greatest for same-day expo- sures, studies conducted in cities with low air-conditioning prevalence, and a positive association between ozone and NO2. In addition, studies using 8-h maximum ozone concentrations to estimate exposures had slightly lower esti- mates than studies using 1-h maximum or 24-h averages. Bell et al. (2005). Bell et al. combined 144 effect estimates (38 from the U.S. and 106 from outside the U.S.) from 39 time-series studies to generate a pooled estimate of how ozone affects mortality. Included in the meta-analysis were peer-reviewed time-series studies that were in English, were published and indexed from 1990 to June 21, 2004, provided numerical estimates and 95% confidence intervals or t-values of the ozone-mortality relationship (total, car- diovascular, or respiratory mortality), and were not based on NMMAPS. The authors combined information across locations and estimated the pooled effect by using a two-stage Bayesian hierarchic model. Pooled estimates were gener- ated on the basis of exposure lags, age groups, cause-specific mortality, concen- tration metrics, study location (U.S. or elsewhere), and potential confounding by PM. Pooled estimates from the time-series studies not based on NMMAPS were compared with pooled estimates from NMMAPS studies. The meta-analysis showed that overall a 10-ppb increase in ozone in the few preceding days (lags of 0, 1, or 2 d or a 2-d average of lags of 0 and 1 d or lags of 1 and 2 d) was associated with a 0.87% (95% posterior interval, 0.55- 1.18%) increase in total mortality. The ozone-associated risk for cardiovascular- disease mortality was larger, 1.11% (95% PI, 0.68-1.53%); and that for respira- tory mortality was lower and insignificant, 0.47% (95% PI, -0.51 to 1.47%). Findings were similar for U.S. and non-U.S. cities. The statistical association was robust to the type of adjustment made for PM, weather, and seasonality. When PM10 or PM2.5 was included in the model, for example, effect estimates for ozone-mediated total mortality were similar, albeit insignificant, at 0.97% (95% PI, 0.03-1.98%). Pooled estimates were similar for the meta-analysis and NMMAPS results. Both studies showed larger effects for cardiovascular mortality (meta-analysis) and cardiovascular and respiratory mortality (for NMMAPS) than for total mor- tality, larger effects for exposures at lag 0 d than lag 1 or 2 d, and lack of con- founding by PM. However, the estimated pooled effects from the meta-analysis were higher than those from NMMAPS, and the difference was attributed to possible publication bias (see below). The authors therefore recommend caution in the use of single-city studies, whether individual or pooled. Summary of Multicity Time-Series Studies and Meta-analyses Each of the multicity time-series studies and meta-analyses found strong evidence of a statistically significant association between short-term ozone ex-
90 Ambient Ozone and Mortality: Estimating Risk-Reduction Benefits posure and mortality, and overall effect estimates were generally similar (Figure 4-1). Effect estimates, however, were higher for the meta-analyses, and this sug- gested the influence of publication bias on the ozone mortality effect estimates. The publication bias was attributed to the fact that meta-analyses relied solely on published studies, which are more likely to contain statistically significant find- ings of an ozone-mortality association. As a result, the pooled meta-analysis estimates may overstate the true ozone-mortality relationship. Moreover, inas- much as the meta-analyses are based on individual study results rather than uni- form analysis of the ârawâ data from each city, it is possible that meta-analyses preferentially included the most significant results (obtained with different ozone exposure matrices and statistical models), and this, too, would result in higher effect estimates. 1.5 % Increase in Mortality per 10 ppb Daily O3 L0 1 1.0 L0 6 L0 0.5 1 U.S. Europe U.S. 0.0 U.S. U.S. -0.5 Europe Anderson et al. (04) Thurston & Ito (01) Gryparis et al. (04) Stieb et al. (02) Bell et al. (05a) Levy et al. (01) Levy et al. (05) Bell et al. (04) Bell et al. (06) Schwartz (05) Ito et al. (05) FIGURE 4-1 Percentage increase in ozone-associated mortality in meta-analyses and time-series studies. Results from all meta-analyses (first seven bars from left), multicity time-series (next three bars), and case-crossover study (last bar on right). âL06â refers to exposures estimated with distributed-lag models of week-long ambient ozone concentra- tions, âL01â to exposures estimated with average of same-day and previous-day ambient concentrations. Results presented in the papers were converted to the percent increase in mortality risk per 10 ppb increase in daily ozone levels. If both summer and yearly results were presented, yearly results were used. Several studies presented results as the 1-hour or 8-h daily maxima. To convert to the 24-hour ozone, conversion factors as noted in the studies were used when such values were provided. Otherwise, 1.73 was used for the ratio of the 1-h max to the 24-h average and 1.53 was used for the ratio of the 8-h max to the 24-h average, based on analysis of ozone levels at 78 U.S. cities. For studies that pre- sented results based on Âµgm/m3, the increment of ozone was converted to ppb based on 1.96 Âµgm/m3 = 1 ppb. Source: Adapted from Bell (2007) presentation to committee. Re- printed with permission from author; copyright 2007.
Study Contributions to the Estimation of Reduced Premature Mortality 91 Each of the investigations found heterogeneity of risk across cities or counties, particularly with regard to the time-series analyses. A recent paper by Bell and Dominici (2008) investigated whether this heterogeneity could be ex- plained by community specific characteristics using NMMAPS. These included race, income, education, urbanization, transportation use, particulate matter and ozone concentrations, the number of ozone monitors, weather and the use of air conditioning. National relative rate did not vary greatly after adjusting for com- munity specific variables (0.46% to 0.54% increase in mortality per 10 ppb in- crease in the previous weekâs ozone). Between-city heterogeneity in effect esti- mates (which is assessed in meta-analyses by using between-study variability) was attributed to several factors, consistently including seasonality and exposure lag periods. Effect estimates were generally higher in summer than in whole year and winter, irrespective of analysis or model choice, although Bell et al. (2004) found higher effects for all-year analyses compared to warm season analyses. Furthermore, ozone mortality effects were greater after same-day than after previous-day exposures in Levy et al. (2005) and Bell et al. (2004, 2005) and after exposures averaged over the week preceding death than after a single day in Bell et al. (2004). Those results suggest that ozoneâs mortality effects are largest on the day or days immediately preceding death but cannot be explained fully by these exposures as they accumulate over the week. Another important explanatory factor was air-conditioning prevalence: effect estimates were lower in cities that had higher air-conditioning prevalence (Levy et al. 2005; Bell and Dominici 2008). Ozone mortality effects were generally not confounded by temperature or PM when year-round data were examined. However, the potential for confound- ing by PM was shown to vary by season: confounding was shown in summer but not winter by Ito et al. (2005). The seasonal confounding pattern is consistent with the seasonal pattern in ambient-ozone and ambient-PM associations in many U.S. cities. Moreover, an additional limitation of the multi-city studies is that they were not able to evaluate any co-pollutants for confounding other than PM10 and PM2.5. Although temperature was not shown to explain the ozone-mortality asso- ciations completely, the associations were found to be sensitive to adjustments for the mathematical form of temperature in models. As found by Ito et al. (2005), the use of more aggressive temperature-smoothing models resulted by itself in a 50% smaller effect estimate than other, simpler weather models. The effect of temperature adjustment was offered as an explanation for the observed doubling of the effect estimate found in the Ito et al. (2005) meta-analysis com- pared with Bell (2004) NMMAPS, which used four smoothing terms to adjust for temperature and dew point. Given those potentially large effects and the fact that the appropriate or best method to adjust for temperature is not known, risk estimates should consider uncertainties introduced by different model choices to account for weather. Also, care must be taken not to select the lag time with the largest risk estimate so as not to bias the estimate upwards.
92 Ambient Ozone and Mortality: Estimating Risk-Reduction Benefits Chronic Studies Long-term effects of ozone on mortality are considered to be the result of cumulative effects on chronic pathologic conditions caused by repeated expo- sure to ozone, that is, not the consequence of exposure that occurred yesterday or recently but the cumulated result of conditions that develop over longer peri- ods. These chronic conditions may lead to subclinical or clinical effects and ul- timately to death. In such deaths, very recent exposure is not relevant; what mat- ters is the long-term history of exposure that increases frailty and thus shortens life expectancy because of related chronic conditions. The standard approach to investigate such effects is the cohort study, in which large numbers of subjects are followed for several years. Ideally, the cu- mulated long-term exposure of each subject to ambient ozone is estimated. The cohort study accumulates time to death. If there is no effect of ozone on mortal- ity, subjects with high long-term or lifetime exposure to ozone will have the same life expectancy as subjects with low ozone exposure. That is, mortality rates, after all other factors that might affect mortality are taken into account, are expected to be the same among populations that have different ozone exposure histories. Cohort studies use variation in ambient air-pollution concentrations over space to create contrasts in personal exposure, whereas time-series studies use variation in time. Chronic Ozone Exposure and Life Expectancy Several air-pollution cohort studies have been published in North America and Europe. The mortality cohort studies have focused on ambient particulate matter or markers of local traffic to characterize exposure of the cohorts. In con- trast, the assessment of the association between life expectancy and ambient ozone has been addressed to a much lesser extent. None of the cohort studies available at this time were designed to investigate chronic effects of ozone, and contrasts in estimated long-term exposure were rather limited. Results of the few cohort studies with ozone data available and that did report associations between long-term mean concentrations of ozone and death rates were not consistent. The American Cancer Society study (ACS)âthe largest cohort study of allâand the Harvard Six City study initially found a nonsignificant negative association of ozone with mortality on the basis of ozone recorded year-round (Dockery et al. 1993; Pope et al. 1995). The Health Effects Institute (HEI) re- analysis based on cohort followup from 1982 to 1989, however, reported a sig- nificant association of the third-quarter (July-September) average daily 1-h maximum ozone concentration for 1980 only and % increase in cardiopulmon- ary mortality of 1.026 (1.003-1.051) per 10 ppb increase in ozone (Krewski et al. 2000). The extended analysis of the ACS cohort from 1982 to 1998 (Pope et al. 2002) observed increased mortality from cardiopulmonary diseases with the 1982-1998 third-quarter average of the daily 1-h maximum ozone concentration
Study Contributions to the Estimation of Reduced Premature Mortality 93 of 1.011 % increase in mortality (0.998-1.026) per 10 ppb change in ozone. A positive but not statistically significant relative risk was observed between ozone and cardiovascular events on the basis of hospital records and deaths in the sub- set of 28,402 women followed for an average of 6 y in the Womenâs Health Ini- tiative Observational Study (Miller et al. 2007). A negative but not statistically significant association with ozone risk was observed after adjustment for several copollutants, including PM2.5. The 15-y followup of the Adventist Health on Smog (AHSMOG) population showed that lung cancer was significantly associ- ated with ozone in men (Abbey et al. 1999). For other causes of death, associa- tions were positive but did not reach statistical significance. Those few cohort studies have not demonstrated a clear positive association between long-term average ozone concentrations and cardiopulmonary mortality after controlling for PM2.5 A persistent problem in the analyses of long-term associations between ambient ozone and health is the characterization of exposure. Factors such as correlations with copollutants, the far lower concentrations of ozone indoors than outdoors, the relevance of ventilation rates, and the influence of time- activity patternsâhamper characterization of the effects of long-term exposure to ozone. Those factors are likely sources of nonsystematic errors that are ex- pected to lead to an underestimation of effects. Long-Term Effects of Ozone on Lung Function A relevant although indirect approach to evaluation of the evidence of long-term effects of pollution on life expectancy studies the effects of pollution on the development of lung function. Measures of lung function, such as forced vital capacity (FVC, the vital capacity expired during a forced spirometric test) or the forced expiratory volume (such as FEV1, the volume expired during the first second of a spirometry test), are strong predictors of life expectancy and correlate in particular with various chronic inflammatory cardiorespiratory dis- eases and related deaths. Associations between long-term exposure to ambient ozone and the development of lung function throughout life can provide useful complementary evidence about the role of ozone as a risk factor for shorter life expectancy. Lung function increases during childhood and adolescence and steadily declines with age after a few years of a plateau phase in early adult- hood. Reduction in growth or accelerated decline can result in clinically relevant impairment of lung function which in turn strongly correlates with premature death. As in the case of mortality studies, PM-related investigations of lung func- tion are more abundant and extensive than those dealing with ozone. However, some studies were either designed to or did report associations between ambient ozone and lung function in children. Recent relevant examples are the Southern California Childrenâs Health Study (CHS) (Gaudermann et al. 2004), the Uni- versity of California, Berkeley (UCB) ozone studies (KÃ¼nzli et al. 1997; Tager
94 Ambient Ozone and Mortality: Estimating Risk-Reduction Benefits et al. 2005), the Yale college-student study (Galizia and Kinney 1999), and a cohort study in Mexico City (Rojas-Martinez et al. 2007). Both lung function and growth correlated with ambient pollution in those investigations, but the specific contribution of ozone was less clearly established. Interpretation of findings of effects of chronic ozone exposure on mortal- ity is complicated by an unexplained inconsistency reported in the CHS, the largest study, with 8 y of annual followup already published. At the beginning of the cohort study, elementary-school childrenâs lung function correlated with the long-term mean of the 1-h maximum ozone across 12 communities (Peters et al. 1999). Associations were particularly pronounced with respect to small-airway function. That would be expected in that tissue dose of inhaled ozone is known to be highest in the small airways (Hu et al. 1994; Kabel et al. 1994), but the findings were significant only in girls and in boys who spent more time out- doors. During the cohort followup, however, lung-function growth rate showed significant associations with a set of urban pollutants, although findings for ozone were not significant and were inconsistent among age groups and meas- ures of function (Gauderman et al. 2000, 2002, 2004). Growth rates in small- airway function were inversely (although not significantly) associated with ozone in the youngest cohort (Gauderman et al. 2002) but not in the 8-y fol- lowup from the age of 10y to 18 y (Gauderman et al. 2004). The comparisons were based on community-level assignment of exposure to ozone, that is, spatial differences in ozone concentrations within communities and among homes were not available. As emphasized by the same research group, indoor:outdoor ratios of ozone among houses in some of the CHS communities are very heterogene- ous (Avol et al. 1998), and strong negative correlations between (measured) ozone and NO2 along busy roads have been reported (McConnell et al. 2006). In contrast to PM, indoor levels of ozone are far lower then outdoors, and the use of outdoor monitors to characterize ozone exposure has been shown to be far more challenging than in case of PM. In the main analyses of the CHS, ambient ozone has been measured only at the central monitor, so the negative correla- tions may be a serious problem in light of the association observed in the CHS between proximity to traffic and lung-function development (Gauderman et al. 2007). Compared to other pollutants, the cross-community contrasts in ozone concentrations was smaller and the same study found rather strong effects re- lated to organic and inorganic acids. The formation of the latter is closely tied to ozone photochemistry. Moreover, Gauderman et al. (2004) also showed that the community (mainly Long Beach) with the lowest ozone concentrations ranked highest for elemental carbon, which correlated strongly with lung-function growth. Control of such strong negative correlations between pollutants is chal- lenging in a 12-community design and highlights the inherent difficulties faced in all chronic-effects studies in the assessment of personal exposure to ambient ozone. As discussed elsewhere in this report, one may need an individual-level rather than community-level characterization of exposure to ozone. Several cohort studiesâincluding those with UCB students (KÃ¼nzli et al. 1997; Tager et al. 2005) and Yale freshmen (Galizia and Kinney 1999) men-
Study Contributions to the Estimation of Reduced Premature Mortality 95 tioned aboveâassigned individual ozone exposures for each study subject. Re- sults of the studies suggest adverse long-term effects of ozone on lung function, which were robust in multipollutant models and among studies. The finding of the first UCB pilot study (KÃ¼nzli et al. 1997), for example, was fully confirmed in a second investigation (Tager et al. 2005) with a larger sample that showed lifetime home outdoor ozone exposures significantly associated with small- airway function. The studies were cross-sectional, so their findings can be com- pared only with the cross-sectional (partly positive) findings of the CHS men- tioned above (Peters et al. 1999). Correspondingly, a recent followup study of a Mexican cohort of children (Rojas-Martinez et al. 2007) reported significant associations of ozone with lung growth, and adjustment for copollutants did not remove these associations. Each child lived within 2 km of a monitoring station. The stations were placed at the schools and so captured the ambient conditions during school hours and in par- ticular exposure during times spent outdoors at school. The study was based on 36 clusters of children who shared schools and residential neighborhoods and thus had similar exposures (Rojas-Martinez et al. 2007). Ihorst et al. (2004) repeated lung-function measurements twice a year for 3.5 y in 2,153 schoolchildren in 15 towns in Germany and Austria. They con- cluded that ozone may be related to seasonal changes in functional growth. No effects were detectable if the measurements were integrated over the entire 3.5 y of followup. The latter finding may be due to partial reversal of acute seasonal ozone-mediated effects on lung capacity, as described in chamber studies, but it is also of note that contrasts in the long-term concentrations of ozone were rather small among the towns. Effects of ozone on adult lung function have been poorly investigated in epidemiologic studies. Recent studies had insufficient ranges of ambient ozone concentrations or did not use ambient ozone data at all. In a study conducted more then 20 y ago in the Los Angeles area, both lung function and its decline correlated with oxidant pollutants (Detels et al. 1987; Tashkin et al. 1994). However, although the findings were suggestive, firm conclusions could not be drawn, given that the comparison relied on only three Southern California com- munities; it was impossible to disentangle effects of various components of the pollution or of community-specific confounders from those expected to be due to ozone. According to animal studies, adverse effects of long-term exposure to ozone on pulmonary function are biologically plausible. As discussed previ- ously, chronic ozone exposure results in the formation of fewer airways, hyper- plasia of bronchial epithelium, increased mucous cells, and shifts in airway smooth-muscle orientation and abundance, all of which increase airway hyper- reactivity in monkeys. Lung-function decreases are both a direct result of those effects and a secondary result of the increase in insults that the lungs will ex- perience because of increased susceptibility to infection and exacerbation of respiratory problems later in life.
96 Ambient Ozone and Mortality: Estimating Risk-Reduction Benefits It should be noted that an indirect argument for plausibility in this context leaves open the possibility for impaired lung function being only a marker of poor health rather than a definite step in the biologic pathway of chronic effects of ozone. Another possibility is that ozone could affect mortality without any effects on lung function. However, chronic subclinical systemic inflammation due to ozone may well contribute to damages in the lung or the vascular system of the lung, resulting in lower lung function, which is a typical feature of many chronic inflammatory diseases. Thus, the indirect argument for coherence link- ing chronic effects on lung function with long-term effects on life expectancy has some appeal despite not being stringent proof. Unfortunately, as discussed below, neither chronic effects on lung function nor on mortality are as well and consistently established in ozone studies as they are for urban air pollution marked with ambient particulate matter due the lack of studies designed for the purpose to understand chronic effects of ozone. Summary of Chronic Effects Some inconsistencies remain to be clarified in the epidemiologic findings regarding ozoneâs effects on lung function during childhood and adolescence, but the observed associations between ozone and small-airway function should not be downplayed. Ozone tissue dose is highest in the small airways, so the findings are in line with expected ozone-related pathophysiologic conditions that result in reduced lung function. The evidence of an effect of long-term exposure to ozone on lung-function growth increases the plausibility of an effect of the same exposure on mortality. The association between poor lung function and reduced life expectancy is strong and well established. SUSCEPTIBILITY It is well accepted that susceptibility to adverse effects of environmental exposures depend on endogenous factors (such as sex, age, and genes) and ex- ogenous factors (such as diet, physical activity, and time spent outdoors). Such co-determinants of a personâs response to ozone would lead to either attenuation or amplification of the effects. Exposure to increased concentrations of ozone has been shown to be associated with adverse health effects in some seemingly healthy individuals, as well as in individuals who are members of susceptible groups, including asthmatics (Romieu et al. 1997; Bernstein et al. 2004; Finkel- stein and Johnston 2004; Maynard 2004; Wilson et al. 2004; McCunney 2005; Trasande and Thurston 2005). Studies have estimated that about 30 to 50% of the population may experience measurable adverse effects after short-term ex- posure to ozone at concentrations such as about 100 to 200 ppb (McDonnell 1991; Corradi et al. 2002; Arjomandi et al. 2005). Observations of different re- sponses suggest that individual host factors, are likely to be important determi- nants of adverse health effects associated with ozone exposure (Kleeberger
Study Contributions to the Estimation of Reduced Premature Mortality 97 1995; Drazen and Beier 1997; Arjomandi et al. 2005; Kleeberger 2005; McCun- ney 2005). Although susceptibility varies among people, questions remain about the factors that make people more or less susceptible. Answers to those questions would be helpful in the evaluation of risks. As summarized below, it is recognized that host factors regulate the bio- logic and physiologic responses to ambient ozone in rodents (Kleeberger et al. 1997; Savov et al. 2004) and humans (Weinmann et al. 1995; Balmes et al. 1996). Studies have focused mainly on acute rather then chronic effects of ozone, but modifiers of acute effects may to some extent be markers of suscepti- bility to chronic effects, given that some chronic pathologic conditions may be the cumulated results of repeated acute effects. The following discussion of fac- tors in the susceptibility to acute effects of ozone may therefore be of broader relevance. Age Whether age modifies the effect of ozone has not been extensively inves- tigated. Studies that have addressed the issue have suggested lower relative risks in the young. Because of the much higher baseline risk of death in the elderly, the absolute contribution of ozone may be substantially larger in them. The NMMAPS analyses of Bell et al. (2004) reported larger effects in people 65-74 y old (0.7% change per 10-ppb increase in the previous weekâs ozone concentra- tion) than in younger and older people (about 0.5%). Medina-Ramon and Schwartz (in press) reported that individuals greater than 65 y old presented an additional increase in mortality as compared to younger individuals. Normal growth during adolescence and aging of the adult lung are suggested as host factors that moderate the functional changes induced by ozone exposure. A normal physiologic aspect of growth and aging is a predictable increase fol- lowed by a decline in lung function. In laboratory-based studies designed to evaluate age as an important variable in response to ozone exposure, two inves- tigations (Drechsler-Parks et al. 1987; Reisenauer et al. 1988) demonstrated a reduced acute functional response (such as FEV1) to ozone in subjects over 40 y old. However, the decline in the functional response with age does not necessar- ily imply that effects of ozone on mortality would follow the same pattern, in- asmuch as different pathophysiologic pathways may play a role. The negative correlation between functional response to ozone exposure and age may be in part attributable to the beneficial role of the mucous lining of the conducting airways in lung defense against irritant gases (Emmons and Foster 1991). It is important to note that early exposure to ozone may have effects over the life span of a person. Recent studies have demonstrated ozoneâs substantial impact on a variety of respiratory conditions in children, including the genesis of asthma in young people (McConnell et al. 2002), which may have later effects on health and pollution-mediated responses. Furthermore, the response to ozone
98 Ambient Ozone and Mortality: Estimating Risk-Reduction Benefits in children is clearly different from that in adults. Children spend more time outdoors, where ozone concentrations are highest, and have higher levels of ac- tivity, which contribute to greater ventilatory rates relative to body mass than observed in adults; these factors logically are associated with greater ozone doses in children than in adults. Those conditions, coupled with the fact that children have an immature immune system and are in the process of developing the capacity to metabolize a wide variety of xenobiotic compounds, accentuate their susceptibility to environmental injury. Studies conducted in nonhuman primate neonates (rhesus monkeys) have demonstrated that early exposure to ozone is associated with a substantial reduc- tion in airway formation and morphogenesis that results in the formation of fewer airways, hyperplasia of bronchial epithelium, increased mucous cells, and shifts in airway smooth-muscle orientation and abundance, all of which increase airway hyperreactivity (Joad et al. 2006; Fanucchi et al. 2006). Other alterations in postnatal lung development include interrupted differentiation of airway basement membranes, modified airway epithelial nerve-fiber distribution, and reorganization of the airway vascular and immune systems (Kajekar et al. 2007; Plopper et al. 2007). Such alterations in early life can have lasting consequences, such as increased susceptibility to infection, respiratory compromise, and exac- erbation of insults to the lungs later in life. Sex and Ethnic Background Epidemiologic studies of the acute effects of ozone on mortality have not paid much attention to sex or ethnicity as modifiers of effects. Medina-Ramon and Schwartz (in press) report that blacks are more susceptible to ozone associ- ated mortality than non-blacks, and females are more susceptible than men. The laboratory-based studies mentioned above (Drechsler-Parks et al. 1987; Reisenauer et al. 1988) have demonstrated not only a reduced functional re- sponse (such as FEV1) to ozone in the elderly but a larger response in female subjects than in males. In both studies, the effective ozone exposure dose (the product of concentration, duration of exposure, and minute ventilation) inhaled by the female subjects was lower than that inhaled by the males. It seems that older, even postmenopausal women are selectively more responsive to ozone exposure than age-matched men: the reasons for the difference remain unknown, but there may be interactions between hormonal factors and mortality. That has not been specifically investigated. In laboratory-based studies (Seal et al. 1993) designed specifically to evaluate ozone-mediated responses by race and sex, the pattern of response was not significantly different among four sex-race groups (male and female, white, and black). Changes in FEV1, airway resistance, and cough in response to ozone at various concentrations (1, 120, 180, 240, 300, and 400 ppb) were compared; a significant decrease in FEV1 was observed after exposure to ozone at each con- centration. When the FEV1 responses were collapsed across all ozone concentra-
Study Contributions to the Estimation of Reduced Premature Mortality 99 tions and subjected to multiple comparison analysis, it was found that black men and black women had significantly larger decrements in FEV1 than white men and that black men had significantly larger decrements than white women. The interrelation between acute functional response and acute risk of death was not established. Pre-existing Diseases For biologic reasons, it is plausible that pre-existing morbidity can modify the effects of exposure to ozone.2 For example, a subject in the critical phase of a bout of pneumonia will have lower capacity to deal with additional exogenous stressors than will a healthy person. However, formal tests of interactions of ozone exposure with pre-existing conditions are not much published yet for ozone effects on mortality. Bell et al. 2004 observed similar effects for total mortality and death from cardiorespiratory causes. Medina-Ramon and Schwartz (in press) reported that among a range of chronic diseases considered to modify the mortality effect of ozone exposure, only atrial fibrillations showed signifi- cantly stronger associations with ozone. The additional percent increase in mor- tality risk per 10 ppb change in ozone was 1.66% (0.03, 3.32). Diabetes, stroke and other conditions did not modify the association. Although effects were also 1.35% larger per 10 ppb change in ozone among those with asthma listed as the second cause of death, this modification was not statistically significant (-0.31, 3.03). The preliminary findings of mortality studies are supported by ozone stud- ies that used outcomes other than death, such as those reporting an increase in asthma hospitalizations during days with high ozone concentrations. Subjects with respiratory diseases (such as asthma) and those prone to cardiovascular death (such as those with atherosclerotic plaque) may be considered âsuscepti- ble.â Genetic Susceptibility Functional differences in the genes that play a role in the pathogenesis of the adverse effects of ozone are the primary candidates for gene-ozone interac- tions. Several studies, summarized next, confirm that concept. The issue has not been addressed with respect to acute effects of ozone on mortality. Results from both animal and human studies support a role of genetic sus- ceptibility in modulating ozone-induced lung inflammation and other pathologic effects and suggest important roles for genes associated with both oxidative stress and innate immunity (reviewed in Backus-Hazzard et al. 2004). In hu- mans, genes associated with oxidative stress, redox balance, and innate immu- 2 The concept of pre-existing diseases or frailty does not imply that susceptible people are necessarily very ill and destined to die soon.
100 Ambient Ozone and Mortality: Estimating Risk-Reduction Benefits nity have been examined for associations between genetic polymorphisms, oxi- dant exposure, and adverse respiratory outcomes. Several of those studied are related to oxidative stress and redox balance, for example, glutathione S- transferase M1 (GSTM1), glutathione S-transferase P1 (GSTP1), nicotinamide adenosine dinucleotide (phosphate) reduced: quinone oxidoreductase (NQO1), glutathione peroxidase-1 (GPX1), glutathione reductase (GR), and superoxide dismutase-2 (SOD2). The GSTM1 null and NQO1 wild-type (NQO1wt) variants have shown in- teresting results. In studies in Italy, Bergamaschi et al. (2001) reported small decrements in both FEV1 (about 3%) and peak expiratory flow (PEF) (about 9%) in 24 healthy subjects with GSTM1 null and NQO1wt genotypes after a 2-h outdoor bike ride when ambient ozone was over 0.080 ppm. Functional changes were associated with small increases in serum CC16 (interpreted as reflecting an increase in epithelial permeability). In a later laboratory-exposure study using similar ozone concentrations, Corradi et al. (2002) reported increased exhaled- breath condensate (EBC) 8-isoprostane, leukotriene B4 (LT-B4), and thiobarbi- turic acid-reacting substances (TBARs) in 22 healthy nonsmoking subjects in the same genetic groups (GSTM1 null and NQO1wt) after 2 h of intermittent exercise. With respect to responses to ozone in the presence of both airway disease and genetic factors, Romieu and colleagues have observed asthmatic children exposed to high ambient ozone concentrations in Mexico City (Romieu et al. 2002, 2004; David et al. 2003). In about 150 asthmatic children, both decre- ments in forced expiratory flow rate at 25-50% lung volume (FEF25-75) (5.2% decrease per 0.05-ppm increase in ozone concentration) (Romieu et al. 2004) and increased breathing difficulties (8% increase per 0.020-ppm increase in the ozone 1-h daily maximum averaged over 7 d) were found in GSTM1 null asth- matics; similar deficits in lung function growth were reported earlier by Gilliland et al. (2002) for asthmatic children who were homozygous for the GSTP1val 105 allele. In a case-control study of 218 case-parent triads, David et al. (2003) observed that among GSTM1 null children with high ozone exposure in Mexico City the children who were NQO1ser187ser homozygotes appeared to be protected from the development of asthma compared with NQO1pro187pro homozygotes. The finding of an interaction of ozone effects with those genes supports the notion that oxidative stress is a relevant pathway of action. Socioeconomic Status and Other Factors Socioeconomic status (SES) may be a marker of a variety of factors that modify the effects of ozone. Modification of effects by SES has been investi- gated in the case of urban PM pollution rather than ozone. There is a high corre- lation between SESâincluding diet and the intake of antioxidants, physical ac- tivity, time spent outdoors, strenuous work-related outdoor work, the use of air
Study Contributions to the Estimation of Reduced Premature Mortality 101 conditioning, stress, and exposure to environmental copollutantsâand various health-relevant (both adverse and protective) effects (OâNeil et al. 2003). The intervention study of Romieu et al. (2004) supplied antioxidants dur- ing a âtreatmentâ period. The intervention period resulted in lower effects of ozone on lung function than in periods of supplementation with placebo. The protective effect of antioxidants was particularly strong in those genetically de- ficient in the ability to oppose oxidative stress (Romieu et al. 2008). Relevance of Susceptibility to the Interpretation of Mortality Studies The fact that occurrence or degree of adverse effects of ozone depends on other host factors has important ramifications in the interpretation of epidemi- ologic findings. It is relevant for risk assessors and policy-makers. The few stud- ies summarized above lead to the preliminary conclusion that the effects of ozone on acute death rates are likely to be larger in those with pre-existing pathologic conditions and the list of plausible candidates modifying effects is rather long, although poorly investigated at this point in time (see above). Al- though the role of genes has not been investigated for mortality, they undoubt- edly play an important role as modifiers of various pathways involved in prema- ture mortality. Interactions among combinations of all those factors have not been investigated. Although some susceptibility factors may explain heterogene- ity in effects only within populations, other marker of susceptibility may very across cities, thus partly explain heterogeneity in ozone effects observed be- tween cities such as, for example, the use of air conditioning (Zanobetti and Schwartz 2008). Medina-Ramon and Schwartz (in press) reported that suscepti- bility factors had a larger effect in cities with lower ozone concentrations. For instance, the additional increase in ozone-related mortality for the elderly was 1.48% (0.81, 2.15) in a city with a mean ozone concentration of 42 ppb versus 0.45% (-0.27, 1.19) in a city with a level of 51 ppb. One can infer that the occurrence of effects depends on subjectsâ suscepti- bility profile, which consists of a wide range and combination of single factors. âSusceptibilityâ is most likely not a dichotomous trait but a dynamic characteris- tic that follows a wide distribution in the population, from no to high susceptibil- ity. Given the involvement of various complex mechanisms, susceptibility may be outcome-specific or pathway-specific. The level of susceptibility of any sub- ject at any time may depend on the joint distribution of various exogenous and endogenous determinants of susceptibility (and their interactions) and may change over time. Another view of or consequence of the variation in suscepti- bility is that the dose (or the ambient concentrations) necessary to trigger an event (death in this case) and the time between exposure and event must follow some distribution among people. The observable association between ozone and mortality thus reflects the weighted average of all true (unobservable) individual concentration-response functions. The latter cannot be established, but the risk of dying because of a 10-
102 Ambient Ozone and Mortality: Estimating Risk-Reduction Benefits ppb increase in ozone concentrations is likely to be substantially larger among the most susceptible population. We emphasize that the concept of âfrailtyâ discussed later in this chapter in the section on short-term mortality displacement can be seen as a complemen- tary way of describing aspects of âsusceptibility.â Very short displacement of death (or âharvestingâ) may reflect the most extreme tail in the continuum of losing just a few days of life due to ozone exposure versus the loss of months or years at the other end of the distribution. Those carrying the worst combination of susceptibility factors at any point in time may be viewed as the faction in the population with the highest frailty and the shortest remaining life expectancy. While âheavily diseasedâ people may belong to the âfrailty pool,â âhigh susceptibilityâ or âhigh frailtyâ may as well be a clinically silent condition of an apparently âhealthyâ subject. For example, substantial evidence supports the notion that pollutants such as urban particulate matter or diesel exhaust activate an array of partly interrelated mechanisms that are all involved in triggering myocardial infarction (MI) and arrhythmia (Brook et al. 2004; Mills et al 2007). Subjects prone to suffer an MI or arrhythmia would belong to the pool of âfrailâ (or susceptible) subjects, but many of those who suffer lethal or nonlethal MI or arrhythmia would not be considered âheavily diseasedâ until the occurrence of the first event (the MI or arrhythmia), because often the underlying pathologic condition, such as atherosclerosis, is silent. Thus, being susceptible does not mean being severely ill. It is also likely that frailty varies in any person because although acute illness (such as pneumonia) might put a person in the pool of those susceptible to death from ozone exposure, if the person survives and re- covers the frailty would be decreased. Although susceptibility factors certainly matter, the distribution of the ozone-mortality effect estimates across the categories of susceptibility is not known; that is, the quantitative details of the heterogeneity of effects are not readily available. Therefore, the overall (population-weighted average) effect in the total population is the only currently scientifically supportable entity to use in risk assessment. Its use is appropriate, even though the population-weighted mean effect may not be a valid estimate for any specific sub-population, because the âtrueâ effect may be much larger among the susceptible individuals but much smaller or zero among the less susceptible. To determine the change in risk as- sociated with changes in ambient ozone concentrations for a specific location, the estimated posterior distribution of the city-specific ozone-mortality risks should be reported, including the mean and standard deviation of the distribu- tion, and the Bayesian shrunk city-specific risks. OTHER FACTORS THAT AFFECT INTERPRETATION OF OZONE MORTALITY EFFECTS Errors in Exposure Estimates The use of daily, 1-h maximum, or 8-h maximum ambient concentrations
Study Contributions to the Estimation of Reduced Premature Mortality 103 to estimate exposure in ozone-mortality studies clearly results in error. Results of exposure studies suggest that the error is large, with consistently weak (and often insignificant) associations found between 24-h ambient ozone concentra- tions and corresponding personal exposures and with personal exposure concen- trations at or below the measurement limit of detection. As discussed in Chapter 3, those weak associations (based on the results of exposure studies relying mainly on passive ozone monitors) are attributed primarily to peopleâs spending most of their time indoors, where ozone concentrations tend to be low and un- correlated with outdoor concentrations. For ozone-mortality studies, which are time-series and case-crossover in design, the use of 24-h ambient concentrations to estimate exposures will result in reduced power to examine associations be- tween ozone exposure and mortality (Carroll et al. 1995; Zeger et al. 2000) and can lead to bias in estimates of ozone mortality effects (Zeger et al. 2000). The magnitude of exposure errors probably depends on several factors, in- cluding season, home ventilation characteristics, and exposure averaging time. Ozoneâs mortality effects, for example, were shown to be greater in the warm season than in winter or the entire year (Ito et al. 2005; Levy et al. 2005). Greater ozone effects in the warm months are consistent with peopleâs spending more time outdoors and tending to live in homes that have better ventilation in the warm months. As evidenced by results of exposure studies, both those fac- tors should lead to stronger 24-h ambient-personal associations, and thus lower exposure error, in the warm seasons, through increased contributions of ambient ozone to personal exposures and through greater infiltration of ozone from out- door to indoor environments, respectively. Those factors may also result in ef- fective doses that vary by season, for example, because of differential physical- activity patterns and breathing rates, which would result in seasonal differences in exposure error. The seasonal differences may be mitigated by behavior pat- terns that are modified by air-quality alerts, which lead people to spend more time indoors on high-ozone days. The effect of behavior during ozone air- quality alerts warrants further examination, as discussed in Chapter 3. The importance of season and ventilation has been further shown by Levy et al. (2005), who provided preliminary evidence of decreasing city-specific ozone mortality effects with increasing prevalence of central air conditioning, presumably because of weaker personal-ambient ozone associations and thus greater exposure error in homes with air-conditioning use. Further support is provided by Ito et al. (2005), who found mean city-specific temperature to be negatively associated with corresponding city-specific ozone-mortality risk es- timates, possibly because of the influence of air-conditioning prevalence in the cities. Those findings suggest that error due to the use of 24-h ambient concen- trations as the exposure metric is greater in winter than in summer, and this of- fers a potential explanation of the lower ozone mortality effects in cold seasons than in warm seasons. Even in the warm months in non-air-conditioned homes, when exposure errors are lower, it probably remains a concern for health studies and their inter- pretation. As discussed in Chapter 3, when 24-h ambient concentrations are sig-
104 Ambient Ozone and Mortality: Estimating Risk-Reduction Benefits nificantly related to 24-h personal ozone exposures, the slope of their association remains small and substantially lower than those found for PM2.5 and sulfate (Sarnat et al. 2000; Sarnat et al. 2006). The low slopes may raise questions about biologic plausibility. Taken at face value, they suggest that some minute change in the true (unknown) 24-h mean exposure to ozone of only 1 ppb or a few parts per billion could cause death. That appears hard to accept, but other interpreta- tions of the studies may lead to less paradoxical conclusions. Given the observed consistency in ozone-mortality findings and the low 24-h personal-ambient ozone associations, it is possible that 24-h ambient ozone concentrations are representative not to 24-h personal ozone exposures, but to some correlated toxic agent or agents. These correlated toxic agent(s) could in- clude photochemical pollutant mixtures, ozone byproducts, or ozone exposure of shorter or longer duration. For exposures of shorter duration, for example it is well established that 24-h ambient concentrations are strongly correlated with 1- h and 8-h maximum ambient concentrations. Although it has not been much studied, it is also possible that maximum ambient ozone concentrations averaged over 1-h or 8-h are better proxies (with respect to both strength and magnitude of association) of corresponding 1-h and 8-h ozone exposures than 24-h ambient concentrations and exposures. In a scripted-exposure study using a trained technician, for example, 1-h ambient concentrations were strong proxies of corresponding personal exposures when the technician spent the hour outdoors. During the time outdoors, 1-h ambient ozone concentrations explained most of the variability in personal exposures and were about equal to personal exposures (Chang et al. 2000). The generalizability of those study findings to at-risk populations, such as the elderly, after their typical activities is not known and warrants future study. People are outdoors primarily when maximum 1-h and 8-h ozone concentrations occur in the afternoon, so it is possible that ozone-mortality studies are reflect- ing maximum ozone exposures averaged over 1 h or 8 h rather than the daily ambient ozone concentration. Given the high correlation between these ambient short-term concentrations and the 24-h mean associations, use of the 24-h con- centrations in time-series studies may serve as an equally good proxy for peak exposures. If so, the differences in the true (unknown) day-to-day peak exposure may indeed be far more than 1 ppb or a few parts per billion and reach some 20- 40 ppb. Such an interpretation would resolve the apparent paradox of a few parts per billion causing death. Confounding Co-linearity among ambient pollutants raises concerns about possible con- founding of ozone mortality effects by correlated co-pollutants, such as PM2.5 (and its components), NO2, and SO2. This concern centers upon the possibility that ambient effects associated with ozone concentrations may represent not only consequences of ozone but also of correlated pollutants not included in the
Study Contributions to the Estimation of Reduced Premature Mortality 105 health effects model. This possibility is critical to address, as our understanding of confounding is central to our ability to ascertain the effects of ozone on mor- tality and any benefits that will result from ozone control. Of the possible con- founders, ambient PM2.5 its components, and weather have raised the most con- cerns about confounding. This concern results from several factors, including the strong correlation of ambient ozone with temperature and ambient PM2.5 as well as between mortality and both temperature and ambient PM2.5. For PM2.5, concerns over confounding also arise from the relatively weak association be- tween personal and ambient ozone, especially in relation to the much stronger association between personal and ambient PM2.5. As noted in Chapter 3, rela- tions between ambient ozone, temperature and ambient PM2.5 vary substantially by season and geographical location, suggesting that the potential impact of con- founding must also be considered by season and location. Also, as noted in Chapter 3, the association between ozone and PM2.5 components is less well understood, in particular as to how this association varies by location, season, and component and how it differs for ambient concentrations and actual per- sonal exposures. Even with these seasonal and geographical considerations, controlling for confounding in ozone mortality studies is a considerable challenge. To date, ozone mortality studies generally have addressed confounding by weather (gen- erally temperature and/or dew point) using a variety of approaches, including those based on indicator variables, linear terms, V-shaped linear terms, and non- linear smooth terms. The ability of these approaches to control for confounding and their corresponding impact on ozone mortality risk estimates have been shown to differ. In a reanalysis of data from seven cities, for example, Ito et al. (2005) examined the impact of control for weather using four approaches (quin- tile indicator variables, V-shaped linear terms, and two combinations of non- linear smooth terms) and showed that overall ozone mortality risk estimates dif- fered by as much as 100% when different approaches were used to control for weather. Models using indicator variables had the highest risk estimates, while models with four non-linear terms had the lowest estimates. Further, the model with the four non-linear terms had the best statistical fit but also the greatest concurvity with ozone, suggesting that ozone and temperature effects from these models are the most difficult to separate. Consistent with these findings, Levy et al. (2005) (using a meta-analytic approach) found that studies with non-linear temperature terms had higher ozone mortality effect estimates than studies using linear terms, although results were limited by the fact that only a small fraction of the examined studies used linear terms. Despite non-linear terms having the lowest effect estimates, Levy et al. (2005) and Ito et al. (2005) are consistent in that the ozone risk was influenced by the manner in which temperature was modeled. Implications of these findings to other ozone mortality studies are not known. At a minimum, the observed large impact of model choices on ozone mortality risk estimates suggests that control for weather is an important source of uncertainty in ozone mortality risk estimates.
106 Ambient Ozone and Mortality: Estimating Risk-Reduction Benefits To address uncertainties based on model choices, Schwartz (2004) exam- ined ozone mortality effects in 14 cities using a case-crossover approach, which controlled for temperature mainly by matching control days with the tempera- ture on the day of death. Using this approach, Schwartz showed a significant overall effect of ambient ozone on mortality that was greater in warm as com- pared to cold seasons, with these effects relatively unchanged when control days were matched on temperature. Similar effects when control days were matched or unmatched on temperature suggest that observed ozone mortality effects are not due to confounding by weather. However, these findings need to be repli- cated in additional studies. Further research examining the potential for con- founding by weather and the impact of methods used to control for this con- founding are clearly needed. For PM2.5, ozone mortality studies have addressed concerns about con- founding focusing on PM2.5 as mainly as a whole (and not its components). The potential for confounding has been examined using multiple pollutant models, showing for example that yearly ozone mortality estimates are robust to adjust- ment for ambient particles and suggesting that confounding of ozone mortality effects by ambient particles is not substantial (Bell et al. 2007). Interpretation of the findings from these multi-pollutant models, however, is complex. For exam- ple, the use of two-stage regressions (as was used in the ozone meta-analyses) to examine confounding by correlated pollutants have raised statistical concerns about the potential for misleading results (Marcus and Kegler 2001). Further, measurement error can differ substantially by pollutant, raising further issues regarding interpretability of findings. Franklin and Schwartz (2008) found a 0.89% (95% CI: 0.45%, 1.33%) increase in non-accidental mortality with a 10 ppb increase in same-day 24-h summertime ozone across the 18 communities. After adjustment for PM2.5 mass or nitrate this estimate decreased slightly but when adjusted for particle sulfate, the estimate was substantially reduced to 0.58% (95% CI: -0.33%, 1.49%). The authors concluded that the association between ozone and mortality is confounded by particle sulfate, suggesting that some PM components could be responsible for part of the observed ozone mor- tality effect. Interpretation of multi-pollutant results is further complicated by the strong correlations among the ambient pollutants, which make their associations with health effects difficult to separate. For multi-pollutant analyses based on yearly rather than seasonal data, this separation is particularly difficult, given the seasonal variability in ambient pollution relationships (see Chapter 3). Further, correlations with ambient ozone in the eastern U.S. generally are strongly posi- tive in the summer but weak or negative in the winter. Evidence from Ito et al. (2005), which examined associations for seven cities located primarily in the eastern U.S., suggests that this seasonal variability in ambient pollutant correla- tions is important, as ozone mortality estimates for the warm and cold seasons were lower and were more similar by season when measured ambient fine parti- cles were included in the season-specific models. Since the relation between ambient ozone and ambient PM2.5 differ substantially by season and location, the
Study Contributions to the Estimation of Reduced Premature Mortality 107 observed similar warm and cold season associations suggest observed ozone mortality effects in the eastern U.S. were not due to confounding by ambient particles. The observed seasonal differences demonstrates the importance and need for season-specific analyses to examine potential confounding of ozone mortality effects by PM2.5 and other pollutants. These types of analyses should be repeated for other regions of the United States. The implications of findings based on multipollutant ambient concentra- tions for actual exposures and their impact on ozone mortality estimates is not clear. When ambient ozone and PM2.5 concentrations are included in the same model, they are assumed to be proxies of their own exposures. However, results from exposure studies suggest that this assumption may not be valid in some geographic areas, as 24-h ambient ozone concentrations in the eastern U.S. have been shown to be better measures of exposures to PM2.5 than to ozone (Sarnat et al. 2001), while 24-h ambient PM2.5 concentrations have been related to personal exposures to ozone in addition to PM2.5. These findings from the U.S. East Coast may suggest that ambient ozone and ambient PM2.5 are serving as proxies for the same pollutant or pollutant mixtures, although results are somewhat limited by the fact that 24-h personal ozone exposures measured in these studies are often near or below the method limit of detection. Despite this, 24-h personal ozone exposures have been shown to be poorly correlated with corresponding personal PM2.5 exposures, suggesting that if ozone mortality studies were to include more direct measures of ozone and PM2.5 exposures, issues related to confounding of ozone mortality effects by PM2.5 may be better resolvedâat least for the as- sessment of daily exposure periods. For shorter exposure averaging periods, confounding of ozone mortality effects by PM2.5 may still be possible, as 1-h personal ozone and PM2.5 exposures were strongly correlated in one study when the technician was outdoors and away from roads (Chang et al. 2000). Whether exposure study results from the East Coast and for PM2.5 can be generalized to other locations and to specific PM2.5 components, respectively, is of yet un- known. Air-Pollutant Mixtures Results of toxicologic studies provide conflicting evidence on whether the mortality risks posed by ozone differ in the presence of other air-pollutant expo- sures. Follinsbee and coauthors (1981) have shown in young healthy subjects that exposure to a gas-phase mixture of ozone and NO2 induced about the same degree of pulmonary-function decrease and breathing-pattern changes as were observed previously in a separate dataset on subjects exposed to ozone alone (Follinsbee et al. 1977). In contrast, a mixture of ozone and NO2 was found to elicit greater pulmo- nary-function decrements in both older (about 50-75 y old) and younger subjects than ozone alone (Drechsler-Parks et al. 1987, 1989). Older subjects were less responsive than younger subjects, but both cohorts exhibited a fairly broad range
108 Ambient Ozone and Mortality: Estimating Risk-Reduction Benefits of pulmonary-function changes. Exposure to the mixture, but not exposure to the individual pollutants, led to decreases in the older subjects in stroke volume, heart rate, and cardiac output during exercise periods of the exposure. That ob- servation suggests not only that exercise performance may be limited by expo- sure to the pollutant gases as a mixture but that it may put people with pre- existing cardiovascular disease at increased risk of exacerbation of their disease. Cardiac-function was not studied in the younger cohort. Healthy adults (56-85 y old) showed significantly smaller exercise-induced cardiac output with exposure to the combination of nitrogen dioxide (NO2) + ozone than with exposure to ozone or filtered air alone (Drechsler-Parks 1995). Also, as discussed earlier in this chapter, Levy et al. (2005) found the estimates of ozone-mortality effects estimates were higher where there was a positive correlation between ozone and NO2. Controlled exposure to mixtures has also been evaluated in susceptible populations. For example, exposure of mild atopic asthmatics for 3 h to ozone at 200 ppb, NO2 at 400 ppb, or a combination of the two pollutants significantly increased sensitivity to inhaled allergen compared with exposure to air, without additive effects. Restudy of the same subjects for 6 h at half the concentrations did not have significant effects. The results suggest that the pollutant-induced changes in sensitivity to allergen in mild allergic asthmatics may depend on a threshold concentration of gas-phase pollutants rather than on the total amount of pollutants inhaled (Jenkins et al. 1999). Gas-phase copollutants can exert their own direct effects on the respiratory tract; however, they may also influence the deposition fraction of solid-phase air pollutantsâthat is, fine and coarse particlesâand localize and intensify injury to epithelial tissues. For example, ozone-induced airway obstruction has been shown to develop differentially even in healthy subjects (in a current database on 140 subjects exposed to ozone in controlled chamber studies), and as many as one-third of the subjects have a significant degree of airflow obstruction during and immediately after short periods of exposure to a concentration similar to an ambient concentration of ozone. Ozone-induced airflow obstruction has been shown to increase bronchial deposition of respirable particles (PM2.5) in the lower respiratory tract (Foster et al. 1993). The pulmonary inflammation and increased epithelial permeability that follow ozone exposure may also lead to increased transepithelial transport of deposited particles, which may increase the risk of PM-related systemic effects. Just as ozone can have direct effects on airway epithelial tissues, PM2.5 exposure is known to induce systemic inflammation and proinflammatory cyto- kine production that may result from free-radical activity of components in PM. Controlled chamber studies have also been evaluated with 2-h exposures to con- centrated ambient fine particles (at about 150 Âµg/m3) in combination with ozone at 120 ppb. In healthy subjects, those exposures induced an immediate alteration in vascular function as indexed by brachial arterial vasoconstriction (Brook et al. 2002). That suggestive study has shown that acute exposure to fine-particle and
Study Contributions to the Estimation of Reduced Premature Mortality 109 ozone mixtures at concentrations often found in the urban environment can lead to acute peripheral arterial constriction. Thresholds To summarize the association between daily variations in ambient ozone concentration and daily fluctuations in deaths, a linear model is used in which it is assumed that the change in mortality risk is constant across levels of pollution. It is unlikely that the association between exposure and response at the individ- ual level follows that simple mathematical formulation. People have their own susceptibility, which is characterized by a unique exposure-response association. The association may be further characterized by a unique âthresholdâ value, an ozone exposure concentration below which there is no risk of death. Each per- sonâs threshold value will also vary over time, depending on the frailty of the person at any given moment, and thresholds may depend on the averaging pe- riod used to assess exposure. The time-series design, however, relates individual exposure not to indi- vidual risk but to the average of personal exposures among the at-risk population on any given day. Thus, the appropriate concentration-response function for the time-series studies is based on an aggregation of individual exposure-response curves. Aggregation of a large number of complex functions can yield smoother and more nearly linear curves at the population level, so we would expect the shape of the concentration-response function based on time-series studies to have a form that is relatively simple. The shape of the concentration-response function for ozone and mortality has been examined in detail (Bell et al. 2006). The association between a 2-d moving average of daily average ozone concentrations in 98 U.S. communities based on time-series data from 1987-2000 was examined. Bell et al. restricted their analysis to all nonaccidental causes of death and the entire year of observa- tion. Three methods were considered to examine the shape of the concentration- response function: restricting ozone concentrations below specified values, pos- tulating a threshold function with no risk below the threshold and a linear asso- ciation above it, and a natural spline formulation of association. A positive asso- ciation between ambient daily average ozone concentrations and mortality was observed down to 15 ppb for the restriction and spline approaches, and the threshold model could not show any substantial improvement in statistical fit down to concentrations of 5 ppb, the lowest value examined. We note that posi- tive and statistically significant (p < 0.05) associations were observed down to 25 ppb using the restriction approach. The risks based on 24-h mean ozone con- centrations below either 20ppb or 15ppb were positive, but did not reach formal statistical significance. Threshold values based on 8-h running daily maxima would be somewhat higher. Those results suggest a near-linear association between ambient concen- tration of 24-h mean ozone and daily mortality count in the United States. Mis-
110 Ambient Ozone and Mortality: Estimating Risk-Reduction Benefits classifying exposure (by using an average of ambient fixed-site monitoring data to estimate population average personal exposure) makes it more difficult to distinguish between linear and threshold models (Cakmak et al. 1999). Estimates of the concentration-response curve based on epidemiologic studies with impre- cisely measured exposure should therefore be viewed with caution. A sensitivity analysis of the shape of the curve may be required to capture the uncertainties in this procedure more fully. Moreover, approaches based on 24-h averaging may cloud thresholds related to actual exposures which may be better represented with shorter averaging times, such as by using an 8-h maximum or daily 1-h maximum. The analysis of Bell et al. (2006) examined data from the entire year. However, it is not clear whether there is an association between ozone and mor- tality in the cooler months (Levy et al. 2005). The warmer months should there- fore be examined separately. There is likely to be less exposure error during that time because people tend to spend more time outdoors or with their windows open. However, the magnitude and nature of the error could depend on location. Furthermore, confounding by other pollutants, particularly PM2.5 in the warmer season (Levy et al. 2005), may alter the shape of the concentration-response curve. Future approaches to further examining concentration-response relation- ships should consider using alternative algorithms for computing nationally av- eraged effects, using regional nonlinear or threshold estimates, and repeating all analyses using 8-h and 1-h maxima, instead of 24-h ozone data. On the basis of its review of the evidence, the committee concluded that the association between ozone concentration and mortality is generally linear throughout most of the concentration range although a number of uncertainties make it difficult to know whether there is a threshold for the association at the lower end of the range. If there is a threshold, it is probably at a concentration below the current ambient air quality standards. Short-Term Mortality Displacement Investigations of acute effects of pollution assess the association between the daily number of events and ambient air quality during the period shortly be- fore death. By the design of the studies, the time of life lost or the âprematurityâ of the deaths is not directly observable. In the absence of knowledge about the amount of time lost because of these acute effects, one may hypothesize that air pollutants were able to trigger death only among a pool of very frail people, namely, those already in very bad health. Assuming that the remaining life ex- pectancy among those frail subjects was short even in the absence of pollution, the effect of air pollution would consist of only a minor shift of the time of death, that is, a short âadvancementâ of death. The concept that air pollution affects mortality only in frail people already near death is referred to as short-
Study Contributions to the Estimation of Reduced Premature Mortality 111 term mortality displacement.3 A period with increased mortality would be coun- terbalanced by a period with lower then expected mortality because deaths among the pool of susceptible people have already occurred (the âpremature deathsâ have been displaced by a few days). It has not been defined how âshortâ the advancement of death must be to qualify as short-term displacement, but the committee assumes that this concept implies losing a few days rather then weeks or months. The issue of short-term mortality displacement is important in the judg- ment of the public-health relevance and quantification of the public-health im- pact of the acute effects of air pollution if one is willing to weight the deaths of people destined to die within a very short time differently from the deaths of people who would have a longer life expectancy if there had been no exposure to ozone. The significance of effects will be determined by the size of the frail pool and the extent of displacement among the pool (Fung 2007). Air pollutants may affect the pool in various ways. The size of the pool, the rates at which people join or leave the pool, and the causes of their joining or leaving all affect the dynamics in mortality patterns. Ozone could increase the death rate outside the risk pool, increase the recruitment rate into the pool, or delay the recovery rate outside the pool. The net acute impact would depend on the relative size of each effect, so the temporal pattern of increased deaths after exposure might be com- plex. If ozone affects only the death rate outside the risk pool, one would expect to see fewer people die after an ozone episode because the risk pool is smaller. If instead ozone increases recruitment or delays recovery, the risk pool may enlarge, and the number of deaths in the period after the episode would be larger rather than smaller. Rabl (2006) has emphasized that the main effects in the mortality-displacement hypothesis are in essence unobservable because ob- served death rates are the net result of all the changes that affect the dynamics of the pool of frail subjects. He has shown that the increase in rates due to prema- ture displacement and the decrease in deaths after the depletion of the suscepti- ble pool cannot be observed or disentangled. Short-term mortality displacement has been investigated (with ambient PM as the marker of daily pollution) in several ways with a variety of statistical methods, including assessment of the displacement with generalized additive models (Zanobetti 2000), decomposition of the time series into time components (Zeger et al. 1999; Schwartz 2000), Gaussian-state space models (Murray and Nelson 2000), frequency-domain log-linear regressions (Kelsall et al. 1999), and time-scaled Fourier analyses (Dominici et al. 2003; Fung et al. 2005). Questions remain about the methods and assessment of mortality displacement, and the time lost because of acute effects of ambient PM cannot be quantified, but the 3 The term harvesting is sometimes used instead of short-term mortality displacement to refer to the concept that air pollution leads to the death of people who are highly sus- ceptible and near death (and die a few days earlier than they would have without air- pollution exposure) rather than death of people who are not otherwise near death.
112 Ambient Ozone and Mortality: Estimating Risk-Reduction Benefits main conclusion from all the work is that the observed acute effects cannot be explained only by short-term displacement of death. Apart from those findings, one should emphasize that the idea that all acute effects of ozone can be explained by short-term mortality displacement involves a biologically implausible dichotomization of the distribution in the time between some adverse acute exposure and death. Instead, the time between exposure and death and the time lost are expected to follow some distributions that depend on the pattern of susceptibility factors or the frailty of the popula- tion. For example, survival time after an MI follows a wider distribution. Some MIs are immediately lethal, but survivors experience a wide distribution of in- creased risk of death as compared to those who have not suffered an MI, and risks are particularly increased at the onset of an MI but also in the early phase of recovery. Thus, if air pollution triggers MIs, the natural distribution of time to death after an MI should most likely be mirrored in MI-related deaths due to air pollution in that the severity of an MI and its survival pattern depend less on the trigger than on the underlying pathology, such as location or degree of athero- sclerosis. A similar case that demonstrates the inherent limitation of the hypothesis of short-term mortality displacement can be made for the frail pool of patients with chronic obstructive pulmonary disease (COPD). Air pollution may trigger acute exacerbations due to, for example, the interference of particles with de- fense mechanisms in the lung, but whether and when decompensation and death occur depend on various host and exogenous factors, including type and time of access to health care, that affect the distribution of probability of and time to death. During an exacerbation of COPD due to, for example, a pneumonia, the patient may be particularly frail during the critical phase of the pneumonia (typi- cally at the end of the first week) and have a transient inability to deal with the additional stress of increased ozone exposure. That may lead to death; but in the absence of the additional hazard posed by ozone, the patient might have sur- vived and lived for as long as an âaverageâ COPD patient of the same age. This example underscores that frailty is not necessarily a stable condition but may change substantially over short periods in a given subject. Some of the acute effects of ozone (see earlier discussion in this chapter), in fact, contribute to moving subjects to a higher level of frailty. In contrast, the terminal phase of lung cancer may serve as an example in favor of air pollutionâs being in the role of short-term mortality displacement. As shown by Goldberg et al. (2001) in patients with lung cancer, death rates correlate with ambient concentrations of urban air pollution. However, lung can- cer continues to be by and large an incurable disease with short survival time after diagnosis. Thus, the role of acute exposure to air pollution in the terminal phase of this disease may be considered less relevant despite some additional displacement of lung-cancer death due to air pollution. The above examples and discussion explain well the observation made in the literature on short-term mortality displacement by PM that risk estimates may increase if the statistical models take extended lag periods into account.
Study Contributions to the Estimation of Reduced Premature Mortality 113 The observations after the classical London smog further support the concept of distributions of events with long tails of subacute effects on mortality rates. The latter remained above expected values for several months after the extreme smog episode without observable evidence of short-term mortality displacement. One published study (Zanobetti and Schwartz 2008) addresses this impor- tant issue related to the acute effects of ozone. It investigated the effects of short-term ozone exposure on mortality in summer by using distributed-lag models. With that seasonal restriction, it was not possible to assess the longer tail of the effects beyond 21 d. As in the case of PM, the study does not support the notion that only short-term mortality displacement occurs. However, in con- trast with the evidence from PM studies, these preliminary results suggest that death rates normalize within about 1 wk, that is, faster than in the case of PM. Accordingly, the effect estimates (or odds ratios) of distributed-lag models were not much larger than those of the acute 1- to 2-d model. As mentioned, the de- tails of the interrelated effects of the size of the frail pool and changes in size due to persons becoming frail or recovering and leaving it cannot be disentan- gled in these approaches. The apparent difference between results of the PM distributed-lag models and those in the ozone studies is difficult to interpret. The 20-d distributed-lag effect of ozone on mortality was at most some 60% larger than the immediate lag-0 effectâa substantially smaller difference than that reported for PM. There is a need for further investigations of short-term mortal- ity displacement and the distribution of lagged subacute effects of ozone, but one can conclude, on the basis of the preliminary analyses and on conceptual grounds, that acute effects of ozone based on the published meta-analytic slopes underestimate to some degree the total acute and subacute effects. The underes- timation may be smaller than in case of PM. It is also of note that the preliminary findings regarding the probable pat- tern of acute and subacute effects of ozone on mortality do not provide further information about the occurrence or size of chronic effects. On the basis of the above background from the PM literature, the biologic concepts of acute effects, and the preliminary results from Zanobetti and Schwartz (2008), one can conclude the following: â¢ Short-term displacement of death is a likely explanation for the sub- group in the upper tail of the susceptibility of the frailty distribution. The size of the fraction explained by short-term displacement is not known. â¢ Short-term mortality displacement is not a plausible explanation of all effects. In fact, the short-term-displacement-only hypothesis conflicts with the notion that underlying mechanism, frailty, severity of outcomes, and time be- tween exposure and event all follow some distributions in populations. Although the amount of life time lost is not observed in acute-effects studies, the upper tail of the distribution of time lost may be months or years. The distributed-lag effect estimates from time-series studies may be used to estimate the population average lifetime lost because of acute and subacute
114 Ambient Ozone and Mortality: Estimating Risk-Reduction Benefits effects of ozone if one assumes a common hazard or probability of instantaneous death in all members of the population of any given age. More likely, though, hazard rates follow a wide distribution, thus the assumption reflects a simplifica- tion. That âpopulation meanâ approach, however, does not reflect the distribu- tion of the amount of time lost because of the deaths that occur as a consequence of increased ozone concentrations or the distribution of pre-existing frailty among those affected. These issues highlight the difficulties in translating the time-series findings into an estimate of life time lost. Sensitivity analysis should be conducted on the number of degrees of freedom used to analyze the time trend in models with distributed lags because the risk estimates for the longer lagging times can be influenced by how temporal trends in mortality effects are modeled. Uncertainty and Variability in Ozone-Mortality Analysis4 One of the major barriers to the broad acceptance of recent EPA health- benefits analysis is the large amount of inherent uncertainty. How the agency deals with that uncertainty is therefore critical for acceptance. Assessment of uncertainty is not the same as assessment of variability. Uncertainty is a consequence of imperfection in knowledge or data and can (in theory) always be reduced by getting better data. Variability is an inherent prop- erty of an exposed population and cannot be changed by getting better data. A fair and balanced characterization of uncertainty in risk estimation is important because most risk estimation is not highly precise and many people are tempted to over interpret the resulting values. EPA (1997c) recognizes that well- performed uncertainty analysis helps decision-makers and the public to place risk estimates in the proper perspective and facilitates informed decision- making. The need for uncertainty analysis is also recognized in other countries as evidenced by the European Commission report (Bickel and Friedrich 2005) that emphasizes the importance of proper communication of uncertainties to ensure that users understand the limitations of analyses and their results. Other reports discuss the issue of uncertainty in risk assessment and bene- fits estimation (NAE 1972; NRC 1975, 1982, 1983, 1994, 1996, 2002, 2007a; PCCRARM 1997). All those reports found that proper and adequate characteri- zation of uncertainty is essential. Almost all expressed concern that most risk assessments and health-benefits analyses tend to underestimate uncertainties and leave decision-makers with unwarranted confidence in the risk estimates pro- vided. To address that concern, the reports recommended the use of formal ap- proaches to characterize uncertainty, such as Bayesian analysis or Monte Carlo analysis (Gilks et al. 1996). Less consistent were opinions about characterization 4 This section concerns the uncertainty and variability in the analysis of mortality asso- ciated with ozone exposure. The potential uncertainty in the monetary valuation of mor- tality risk associated with ozone is considered in Chapter 5.
Study Contributions to the Estimation of Reduced Premature Mortality 115 of model uncertainty: some reports discussed and recommended the use of ex- pert judgment, and others recommended that such scientific uncertainty be thor- oughly described but not quantified. Efforts to characterize model uncertainty, however, are critical, as evidenced by the most recent report (NRC 2007b), which states that model uncertainty in particular tends to be understated or ig- nored. That report states that modelers often assume that their models are correct and base estimates of the modelsâ parameter values on single studies; because a model may be incorrect or incomplete, the uncertainties that they produce may be significant. Sources of Uncertainty There are many sources of uncertainty in mortality-risk assessment related to ozone exposure, including random sampling error in a random sample of data, measurement error (systematic error or random error), data nonrepresentative- ness, surrogate data, lack of relevant data, problem and scenario specification, and model uncertainty. Of those, only sampling error is captured in ordinary statistical measurers of uncertainty (p-values and confidence bounds), so total uncertainty is necessarily greater and perhaps much greater than can be directly measured. Several methods exist to reduce uncertainty, including iteration of model- building and input distributions, in combination with sensitivity analysis, which can help to focus resources on the most important model inputs and components. The uncertainty sources that probably have the greatest influence in ozone- mortality analysis are the epidemiologic models. Much of this uncertainty re- sults from unavoidable and expected variability, from estimation of concentra- tion-response functions with epidemiologic studies, from possible lagged ef- fects, and from baseline statistical variation. Baseline statistical variation, which tends to be inherent in data, could be reduced in some cases with control meas- ures and in others with better models. To deal with model uncertainty, it is pos- sible to compare alternative models but not combine them, weight predictions of alternative models (for example, with probability trees), or use metamodels that degenerate into alternative models. Uncertainty is also associated with ozone concentrations and with the reliability of ambient ozone-monitoring data to indi- cate ozone exposure. As stated above, the formal approaches to uncertainty analysis in mortality models include Bayesian analysis and Monte Carlo analysis. The use of expert judgment, which could be considered an empiric Bayesian approach, provides another framework for uncertainty analysis. EPA used expert elicitation in the 2006 final regulatory impact analysis for the PM NAAQS (EPA 2006). How- ever, the best way to use expert judgment remains to be determined. A key step in moving forward would be to agree on conditions under which expert judg- ment is an acceptable input, even for rule-making situations. The World Health Organization (WHO 2000) has developed guidelines to identify a set of proc-
116 Ambient Ozone and Mortality: Estimating Risk-Reduction Benefits esses and general approaches to assess available epidemiologic information in a clear, consistent, and explicit manner. In particular, it recommended the use of expert assessment according to an explicit protocol, defined in advance. The essential components of the protocol are: specification of the question to be ad- dressed, justification of the expertise represented, and specification of the meth- ods to be used. Before EPA makes a decision about the use of expert judgment in the context of the impact of ozone on mortality and for benefit analysis, fur- ther evaluation is needed to assess the relevance and usefulness of the recent expert elicitation results for PM. In particular, in view of the time and cost asso- ciated with an expert-judgment procedure, the agency needs to understand whether, in the case of PM, the expert elicitation in the presence of other sensi- tivity analysis added substantial value or could have been replaced with a pub- lished formal or âqualitativeâ meta-analysis. EPA could also consider other options for incorporating expert judgment into its probabilistic uncertainty analysis. The agency possesses considerable internal expertise, which should be used as fully as possible. If it continues to use expert elicitation for uncertainty analysis, it should consult outside experts as needed, individually or in panels. The experts and the rationales and empirical bases of their judgments should be made known. It is important to distinguish between uncertainty due to projecting the fu- ture and uncertainty inherent in estimating mortality on the basis of information on the magnitude of ozone exposure, which varies across space and time and among individuals. Thus, a good characterization of temporal and spatial correlation is also needed. Sensitivity analysis and model diagnostics would help to determine the appropriateness of the spatial and temporal characterization of the dependence structure (and other assumptions) incorporated in the models under considera- tion. The ability to quantify and propagate uncertainty throughout the analysis is still in development. However, uncertainty analysis has developed further and faster than our ability to use it in decision-making. A National Research Council report (2007a) recommends that more attention be paid to questions of how to use uncertainty analysis to set action levels and make regulatory decisions. Ef- fective communication of uncertainty requires a high level of interaction with the relevant decision-makers to ensure that they have the necessary information about the nature and sources of uncertainty and its consequences. Uncertainty in Air-Quality Numeric Models Air-quality numeric models, such as the Community Multiscale Air Qual- ity (CMAQ) model, are potentially valuable tools that can extend the ozone- mortality analysis to places and times on which the desired data are not avail- able. Air-quality models, which are based on the dynamics and mechanics of atmospheric processes, typically provide information on larger regions than data
Study Contributions to the Estimation of Reduced Premature Mortality 117 from observational networks. Errors and biases in these deterministic models are still inevitable because physical processes are simplified or neglected and be- cause of the mathematical approximations used in assigning parameter values and inaccurate inputs. As a result of their own uncertainties, the models will add more uncertainty to the ozone-mortality risk-assessment estimates if they are used to characterize ozone exposures in future studies. Evaluation of these models and their uncertainties can help to quantify and characterize the magnitude of errors in the models. Although a full Bayesian analysis that incorporates all sources of information may be desirable in princi- ple, it will be necessary in practice to make strategic choices about which sources of uncertainty justify such treatment and which sources are handled bet- ter through less formal means, such as consideration of how model outputs might change as some of the inputs vary through ranges of plausible values. A National Research Council report (NRC 2007b) describes in more detail different sources of uncertainty in air-quality numeric models and addresses the question of how to judge whether these models and their results are adequate for supporting regulatory decision-making. Effective communication of the differ- ent sources of uncertainty in the models requires a high level of interaction with the relevant decision-makers to ensure that they have the necessary information about the nature and sources of uncertainty and its consequences. Thus, if such models are used to assist in mortality risk assessment, uncertainty analysis and extensive discussion between analysts and decision-makers are needed. USE OF EPIDEMIOLOGIC INFORMATION IN OZONE-RELATED RISK AND BENEFITS ASSESSMENT Because death is inevitable, the question relevant to policy-makers, risk assessors, and the public is whether people who live in areas more highly ex- posed to air pollution experience death at an earlier age than people who live in communities that are less exposed. The effect of pollution on death may be quantified in terms of number of deaths in a given period or life months (or years) lost. Number of Deaths The âacute-effect studiesâ directly estimate the change in daily death rates due to short-term exposure (or a short pulse of exposure). Most models investi- gate the change in death rates for only one or a few days, whereas distributed- lag models look further ahead to capture delayed acute effects, often referred to as subacute effects. The expansions are useful in understanding the statistical distribution of time between an exposure pulse and the time pattern of occur- rence of death. On a population level, it is extremely unlikely that all ozone- related deaths occur either immediately or within 1-2 d of exposure. A more plausible model assumes that susceptibility to death (or frailty) and the success
118 Ambient Ozone and Mortality: Estimating Risk-Reduction Benefits of intervention strategies follow a distribution wherein some people die immedi- ately or within 1-2 d and others first suffer acute health ailments (such as an MI or pneumonia) and their deaths follow a period of unsuccessful treatment or as a result of the decompensation of defense mechanisms. The most complete as- sessment of the association between acute exposure and death originates in dis- tributed-lag models that integrate the distribution of the time between exposure and death. However, both the usual time-series model and the distributed-lag models focus on a short window of exposure. Effects of long-term cumulated exposure are, by design, ignored in such models. In contrast, cohort studies focus on the associations between some metric of long-term exposure and death rates (as in the ACS study and the Harvard Six Cities Study). As a consequence, the increases in death rates associated with pollution may reflect the total effect of pollution on death, including life time lost because of acute or subacute effects and because of chronic conditions. For example, if pollution increases systemic inflammation, it may contribute to re- modeling of airways and to obstruction or contribute to the development of atherosclerosis, and these chronic ailments increase the risk of premature death independently of any acute effects of pollution. One may argue that the current way of reconstructing long-term exposure in cohort studies is insensitive to short-term peaks and changes in exposure and that cohort studies may therefore miss some acute effects. Although cohort studies are likely to capture a large fraction of the effects of pollution on deathâincluding most acute, chronic, and combined effectsâsome fraction of the immediate acute effects may not be con- tained in the effect signal, so the total effect of ozone might be larger than that observed in cohort studies. Years of Life Lost If one assumes that baseline probability of death or hazard is common to all members of the population at any given age, one can treat estimates of pollu- tion-related risk based on time-series mortality studies in a manner similar to estimates obtained from cohort studies (Miller and Armstrong 2001; Burnett et al. 2003; Rabl 2006). That is, the risk estimates from time-series studies can be used to estimate the life years lost because of acute effects of pollution in the same way that risk estimates from cohort studies are used to derive the time lost because of the chronic, if not total, effects of pollution. However, there is proba- bly a wide distribution of likelihood of death in any population on any given day. Current ambient ozone concentrations probably will not increase risk to a point at which death is likely for most of the population, and only people in a frail state (for example, having a chronic disease) would clearly be at risk of death from ozone exposure. It is assumed that large portions of the population are at risk of death be- cause of longer-term exposure to air pollution, so it is more reasonable to apply a common baseline hazard function within the cohort design. Relative risks from
Study Contributions to the Estimation of Reduced Premature Mortality 119 the cohort studies are then combined with common baseline hazard functions, in the form of life tables, for selected populations to estimate the amount of life lost because of exposure to air pollution (Coyle et al. 2003; Miller and Hurley 2003). The same life-table calculations can be used to estimate the number of people who will die over a fixed period because of changes in air pollution ex- posure. If the hazard and risk distributions in the population are not independent, estimates of excess deaths and life years lost in the entire population because of changes in ozone exposure based on time-series studies will not equal the aver- age of estimates across all risk groups.5 The size of the difference is not known but could be estimated if risks for susceptible groups are obtained (Goldberg et al. 2005). Translation to Risk and Benefits Assessment Acute and chronic effects of urban air pollution, such as PM, on mortality have been well established. Risks based on cohort studies are some 10 times larger than those based on acute-effects studies that investigated only the influ- ence of yesterdayâs concentrations of PM2.5 on todayâs death rates and some 3-5 times larger than the total acute effects, including death, distributed over several weeks. More recent risk assessments provided separate estimates of acute effects that occur within a few days of exposure and total effects. Such separation may be useful, given that the time between pollution abatement and health benefits may follow a distribution, with acute effects being reduced more quickly than effects due to chronic conditions (RÃ¶Ã¶sli et al. 2005). Ozone-mortality risk assessment faces major challenges. As summarized above, evidence of long-term effects of ozone on mortality (or survival time) is presently weak, so the derivation of a relative risk to describe the association between ozone exposure and death is more difficult than in the case of PM. The translation of the results into number of attributable deaths is possible but leaves us with two main challenges. First, the assessment of acute effects would be incomplete if based solely on the usual time-series studies, given that subacute (delayed) effects are not captured with these studies; only one study has pub- lished estimates for distributed-lag models. Second, in the absence of abundant quantifiable evidence of chronic effects of ozone on mortality, total life time lost because of both acute and chronic effects cannot be estimated. The current evi- dence from cohort studies, although it is weak, supports the notion that the esti- mates of the risk of death due to ozone may be largely underestimated if based 5 In this context, the distribution of hazards among the population is the likelihood of death at any age; the distribution of risk is the association between ozone and death at any age.
120 Ambient Ozone and Mortality: Estimating Risk-Reduction Benefits solely on the main time-series studies that were at the center of EPAâs request for this report. While the current lack of a sufficient number of ozone cohort studies pre- cludes a respective assessment, lessons learned from PM would suggest that estimations based on time-series study alone are far smaller than the overall long-term cumulated effects. The question of how to translate epidemiologic information into risk and benefit assessment is the subject of debates. Many risk assessments derive, as one intermediate step, the number of deaths attributable to air pollution on an annual basis. That approach has several limitations that are of concern particu- larly for chronic effects but conceptually also apply to acute effects, as is the case for the acute effects of ozone on death. Although it is true that pollution increases the mortality risk and thus the number of deaths, one has to emphasize that death is inevitable and that improvement in air quality only postpones death and does not prevent it. The concept of âattributable casesâ is often interpreted as the corollary of âpreventable cases,â which in the matter of death is certainly wrong. Another problem faced with âattributable deathâ originates in the fact that changes in death rates in any population lead to changes in the age structure of the population. That is, a reduction in air pollution will lead to longer life expec- tancy and increase the number of elderly people. Age-adjusted death rates are expected to decrease under cleaner conditions, but the absolute number of deaths will steadily increase as the population ages. As a consequence, the âattributable deathsâ (or benefits) will not be the same throughout the years after an im- provement in air quality. Instead, there will be a complex interplay between the distribution of acute vs chronic causes of premature death in which reduction in death rates caused by chronic exposures will take longer to materialize than re- duction in death rates from acute effects and longer than the changes in the population age structure due to the removal or reduction of a risk factor. The net result of those dynamics is a continual adjustment of the expected annual change in the number of deaths attributable to a change in pollution. As shown by Miller and Hurley (2006), although the benefit (in absolute number of deaths) may steadily increase during a few years, it will later decrease for a cou- ple of years and ultimately reach a time when the absolute number of deaths (per year) is larger than it was in the more polluted condition. Therefore, expression of benefits due to reduction in mortality rates can ultimately not be correctly expressed as a stable annual number of deaths prevented. For all those reasons, risk assessors omit the expression of âattributable cases of deathâ but prefer to derive years of life lost (YLL) due to some policy change. Although the derivation of YLL is based on the same input informa- tionâthe risk changes derived from epidemiologic studiesâthe use of YLL is more appealing because one can derive the total YLL for the entire remaining life time of any cohort, whether dynamic or static, and express benefits in terms of annual YLL (Brunekreef et al. 2007). However, even though it may be more appealing to express mortality-risk change in YLL, the concept may be more
Study Contributions to the Estimation of Reduced Premature Mortality 121 problematic for valuation purposes than that of âattributable deathsâ (see Chap- ter 5). Another potential approach to the translation of epidemiologic information into risk and benefit assessment is to stop short of calculating annual numbers of deaths prevented and simply report the change in the annual mortality rate ex- pected as a result of the change in ozone. Rabl (2006) shows the mathematical relationships between the change in mortality rate, the annual reduction in num- ber of deaths, and the years of life saved. For an acute mortality effect as a result of short-term (daily) pollution-exposure changes and a stable population (in which births equal deaths), he shows that a permanent reduction in pollution temporarily reduces the number of deaths over some period because people who would have died from pollution live longer. Eventually, however, they die from something else, so the annual number of deaths returns to its previous level. However, there is a permanent increase in life expectancy, a larger population, and thus a permanently lower mortality rate. The valuation implications of this are discussed in Chapter 5. CONCLUSIONS AND RECOMMENDATIONS Overall Conclusions and Recommendations Human chamber and toxicologic studies have yielded strong evidence in- dicating that short-term exposure to ozone can exacerbate lung conditions, caus- ing illness and hospitalization, and potentially lead to death. Although it is less abundant, the available evidence on ozone exposure and exacerbation of heart conditions points to another area of concern. Panel and epidemiologic studies have also found that exposure to ozone (as an indicator of a broad mix of photo- chemical oxidants) has those effects. The committee found that the four recent time-series analyses and meta-analyses of the relationship between exposure to ozone (and other photochemical oxidants) and premature mortality add to the evidence by providing robust statistical evidence of an association. Overall Conclusions On the basis of the broader available evidence and the additional insights obtained from its review of the new time-series studies, the committee concludes that short-term exposure to ambient ozone and the larger photochemical-oxidant mixture is likely to contribute to premature deaths. Although it is rarely possible to exclude the possibility of zero effect in such analyses, the committee con- cludes that an absence of any effect is highly unlikely. Despite some continuing questions about the evidence, the committee concludes that it is strong enough to be used in the estimation of the expected benefits of reductions in population mortality risk that would result from reducing exposure to ozone and/or the pho- tochemical-oxidant mixture.
122 Ambient Ozone and Mortality: Estimating Risk-Reduction Benefits Overall Recommendation The committee recommends that ozone-related mortality, estimated on the basis of the results of the recent time-series analyses, be included in future esti- mates of health benefits of reducing exposure to ambient ozone. The committee further recommends that the greatest emphasis be placed on estimates based on systematic new multicity analyses using national databases of air pollution and mortality, such as was done in the National Morbidity, Mortality, and Air Pollu- tion Study database, without excluding consideration of meta-analyses of previ- ously published studies. Emphasis should also be placed on risk estimates ob- tained from analyzing data on multiple days so as to include delayed acute effects. Such health-benefits estimates should be accompanied by a broad array of analyses of uncertainty, while at the same time understanding that a zero value is highly unlikely. Strengths and Limitations of the Evidence In reaching its overall conclusions and recommendation, the committee identified a number of strengths and limitations of the evidence. On the basis of available evidence, the committee believes that deaths re- lated to exposure to ozone (and other photochemical oxidants) are not likely to be restricted to people who are at very high risk of death within a few days. The evidence is provided by the recent analysis of time series in several U.S. cities that focused specifically on a mortality-displacement pattern in the time course of exposure and death (Zanobetti and Schwartz 2008). In that analysis, it was clear that short-term mortality displacement (or âharvestingâ) could not fully explain the observed increase in death. The estimates steadily decreased with increasing lags, reaching the level of no effect approximately seven days after exposure but without reaching negative associations at any of these days as one would expected if all events were explained by short-term displacement among the pool of extremely sick and frail people. As discussed by Rabl (2006), if pol- lution affects both the extremely frail and others, the portion related to dis- placement cannot be quantified as one can only observe the net difference be- tween short-term harvesting and all other acute effects. The pattern observed in Zanobetti and Schwartz (2008) indicates that at no lag until the disappearance of any effect (day 7) was the portion explained by mortality displacement larger than the rest. This evidence, albeit from only one study provides evidence that ozone-mortality effects are associated not just with those already near to death. In contrast with the PM literature, very few data are available from the use of distributed-lag models. Those available suggest that in the case of either ozone or PM, effect estimates steadily increase with increasing time of the in- vestigated effects. Specifically, subacute (longer-term) effects that combine ef- fects of several days or weeks of exposure are larger than immediate short-term effects, whereas estimates of effects from cohort studies are the largest. How-
Study Contributions to the Estimation of Reduced Premature Mortality 123 ever, the increase in estimates is far larger for PM than for ozone. That may be a consequence of attenuation due to large errors in characterizing exposure to ozone, or it may reflect a more dominant role of acute pathologic conditions related to ozone exposure whereas the role of chronic pathologic conditions is more important in the effects of PM. As these interpretations rely on very few studies, further confirmation is warranted. In addition to the specific analyses in the time-series studies, there is only very weak, evidence from cohort studies of an association of premature mortal- ity with longer-term exposure. Thus, the committee could not conclude at this time that exposure to long term ozone concentrations is related to mortality. Although the associations in the recent time-series studies appear suffi- ciently robust to provide a basis for estimating benefits, several factors need to be considered in interpretation: â¢ The committee found that short-term ozone exposure is likely to con- tribute to premature mortality in addition to the risks posed by weather and PM, but studies have not been sufficient to control fully for potential confounding by or interactions with condensed-phase constituents of airborne PM, such as sul- fates, acids, carbon, and other elements. â¢ Controlling for such confounding is further complicated by the data from personal-exposure studies, which have found low correlations of monitored ambient ozone with personal ozone exposure. Detailed Conclusions and Recommendations In addition to its conclusions about the strengths and limitations of the new time-series evidence, the committee identified several kinds of additional needed research. Conclusion: On the basis of available evidence, the committee believes that deaths related to exposure to ozone (and other photochemical oxidants) are not likely to be restricted to those in people who are at very high risk of death within a few days. Because the evidence is based on results from only one study, it warrants confirmation by other studies. Recommendation: EPA and the scientific community should conduct ad- ditional studies to investigate short-term mortality displacement and include the use of alternative methods. An example of such methods is investigation of peo- ple who have diseases, such as diabetes and heart disease, which are known to induce high mortality risk associated with air pollution. Conclusion: If further confirmed, the weak current evidence from cohort studies of an association of premature mortality with longer-term exposure would tend to support the assumption that not all the effects seen in time-series studies are due to short-term mortality displacement.
124 Ambient Ozone and Mortality: Estimating Risk-Reduction Benefits Recommendation: The committee recommends that EPA, the National Institutes of Health (NIH), and the scientific community conduct further work with cohort studies to examine the association between long-term ozone expo- sure and mortality. The use of very large cohorts with long followup periods may be required because the long-term ozone-exposure mortality risk appears to be much smaller than that posed by fine PM. This may require pooling informa- tion from several cohorts to obtain sufficient statistical power. Long-term ozone- exposure models also need to be further developed to distinguish between varia- tions in exposure not only on the between-cities and within-cities scales but at the individual level. Recommendation: To the extent that new cohort evidence is strength- ened, EPA should consider including estimates based on that evidence in its benefits assessments. Conclusion: Epidemiologic studies have found that ozone-mediated risks were greater for cardiovascular mortality than for total mortality. However, there have been only a few toxicologic or human research studies directly evaluating cardiovascular effects of ozone exposure. Recommendation: EPA, NIH, and the scientific community should con- duct further studies on the cardiovascular effects of ozone exposure, both in hu- man and animal models. Studies should be designed to identify genetic suscepti- bility factors. Conclusion: Although the committee found that short-term ozone expo- sure is likely to contribute to premature mortality, the optimal way to quantify the effect is unavailable. The method that EPA uses now is appropriate only in very restricted situations that are not likely to be realistic. Recommendation: EPA should study emerging concepts and evaluate their use and implications in benefits assessments, including relationships be- tween changes in mortality rates, annual deaths prevented, and years of life saved. The alternative approaches for expressing ozone mortality effects will lead to rather similar results if one is supposed to express only the most immedi- ate (acute) effects of pollution. However, with the integration of subacute effects estimates and in particular in the case of use of long-term chronic-effect esti- mates, the discrepancies between the approaches increase, and the conceptual flaws of the âattributable-casesâ model become more pronounced. Conclusion: Expected differences in susceptibility causes substantial un- certainty in estimating mortality risks on the basis of results of epidemiologic studies of total populations because of the lack of independence between pollu- tion-related risk and baseline hazard. Recommendation: EPA and NIH should encourage more studies on po- tentially susceptible population groups, and EPA should explore the effects of uncertainty on risk assessments. To the extent that data are not available, models and assumptions can be used for sensitivity analysis.
Study Contributions to the Estimation of Reduced Premature Mortality 125 Conclusion: Although the recent time-series studies have, to the extent possible, included analyses of alternative ozone metrics (such as 1-h maximum and 8-h maximum), it is important to further examine the relationship between those averaging times and to identify additional ozone metrics based on person exposures. That is important both to examine dose-response relationships, test potential confounding more effectively and to inform future regulatory choices among actions that might have differential effects on peak and multi-hour aver- ages. Recommendation: EPA and the scientific community should design and conduct studies that use the full array of potential ozone metrics, including those that estimate population personal exposures. Conclusion: Control for confounding has not sufficiently accounted for geographic and seasonal variability in the relationship between ozone and is possible confounders. Further, data on PM speciation have not been sufficient to include in analyses of potential confounding; however such data are increasingly available from the Speciation Trends Network. Recommendation: EPA and the scientific community should account for seasonal and geographic variability in the relation between ozone and its con- founders and should increasingly include the growing STN database in all future analyses of potential confounding of the ozone associations. To this end, EPA (through its PM Centers program), the Health Effects Institute (through its Na- tional Particle Component Toxicity Initiative [or NPACT]), and others (such as the Electric Power Research Institute) are supporting major new efforts in both time series and cohort epidemiology, to address this need. Further EPA should work with the scientific community to ensure that STN collects data frequently enough on the particle components most relevant to the potential for confound- ing. Conclusion: On the basis of its review of the evidence, the committee concludes that the association between short-term ozone changes in ozone con- centrations and mortality is generally linear throughout most of the concentra- tion range, although uncertainties make it difficult to determine whether there is a threshold for the association for the association at the lower end of the range. If there is a threshold, it is probably at a concentration below the current ambient air quality standards. Recommendation: EPA and the scientific community should further ex- plore how individual thresholds may vary and the extent to which thresholds depend on the frailty of the individual at any given moment. The research should involve panel studies of individuals considered to be susceptible to premature death from ozone exposure, such as those with impaired lung function. Because it is not clear whether there is an association between ozone and mortality in the cooler months, warmer months should be examined separately. A sensitivity analysis should be conducted on concentration-response relation-
126 Ambient Ozone and Mortality: Estimating Risk-Reduction Benefits ships to capture more fully the uncertainties contributed by reliance on average fixed-site monitoring data to estimate population-average personal exposure. Conclusion: There is a lack of an observed association between ozone and mortality during the periods when exposure to ozone is expected to be low, and the ability to understand the lack is inhibited in part by the fact that monitoring is more limited during the winter periods. Better understanding of ozone- mortality relationships during winter is important for full exploration of (1) sea- sonal differences in risk, (2) how these seasonal risk differences vary spatially between communities with warmer and cooler winters, and (3) ozone-mortality relationships at lower ozone concentrations. Recommendation: EPA and the states should extend operation of ozone monitoring into winter and report the measurements. Conclusion: As has been the case with PM, ozone analyses conducted with distributed-lag models over several days appear to capture overall effects better, but there have been relatively few of them. Recommendation: EPA and the scientific community should seek out appropriate databases and conduct distributed-lag analyses as part of future epi- demiologic investigations to understand the statistical distribution of time be- tween an âexposure pulseâ and the time pattern of occurrence of death. Conclusion: Uncertainty in the epidemiologic models is likely to intro- duce substantial uncertainty into ozone-mortality analyses. Recommendations: The committee identified several approaches to ad- dressing the uncertainty: â¢ Results of the models should be presented with discussion of their re- liability and of the estimated uncertainty about which model (if any) is reasona- bly correct. â¢ EPA should consider Bayesian approaches, including additional expert elicitation once the recent experience with PM has been evaluated, to uncer- tainty analysis. â¢ As EPA develops computational models and input distributions, it should intermittently conduct sensitivity analyses to focus resources on the most important inputs and parts of the models. â¢ EPA should distinguish between data-derived estimates of some com- ponents (such as the concentration-response function) and expert opinions about other components that are lacking in scientific data to achieve a better under- standing of how existing data and expert judgment combine to produce esti- mates and of where new data would be most valuable. â¢ Time-series studies and meta-analyses should conduct additional sen- sitivity calculations, in particular to examine sensitivity of results to the structure of the cessation lag and sensitivity of the premature-mortality estimate to the
Study Contributions to the Estimation of Reduced Premature Mortality 127 presence of a potential threshold concentration of ozone below which mortality effects are not observable. EPA should present multiple results from each study and characterize the different sources of error in the studies in a final uncertainty estimation on their benefits analysis. â¢ Air-quality numeric models (such as CMAQ) should be considered for use in ozone epidemiologic studies to extend the spatial scale of available data. However, the uncertainty associated with the models needs to be carefully con- sidered before inferences are drawn from the simulation models for mortality risk assessment.