National Ambient Air Quality Objectives For Ground-Level Ozone- Summary - Science Assessment Document

8. Human Health Effects

The data on health effects of ozone have been examined in human epidemiological studies, controlled human exposure studies, and animal toxicological studies. The impact of ozone on human health is mainly via the respiratory system (Figure 8). The symptoms of ozone exposure are cough, shortness of breath, decrements in spirometric values, increases in airway resistance and bronchial responsiveness to stimuli, and airway inflammation with the potential to result in Emergency Department visits or hospital admissions, or mortality.

Figure 8: Human respiratory system. The centriacinar, or central portion of the acinus (the functional respiratory unit including all structures from the respiratory bronchiole to the alveoli) is around the terminal bronchiole.

8.1 Epidemiological Studies

Types of studies

Epidemiological studies of the effects of ozone on human health explore statistical associations between changes in ambient levels of ozone and changes in the occurrence of cardiorespiratory health problems in the general population. Five basic health effect markers have been examined in epidemiological studies: mortality, hospital admissions, Emergency Department (ED) visits, field (camp and panel) studies, and chronic health effects. The latter two types of studies often investigate respiratory symptoms, medication use, pulmonary function changes, reduced activity days, and elementary school absenteeism.

Most of the recent epidemiological studies considering acute and short-term effects (mortality, hospitalizations and ED visits) of air pollution on human health have been time-series studies. A time-series analysis is by definition longitudinal in nature. In this type of study, the timing of an adverse health event is studied relative to short-term time trends in air pollution within a defined geographic area. Major advantages of the time-series study are that it usually provides many units of observational data (typically 1000 days), and examines the temporal pattern of the event, that is, whether health effects lag peak pollution days. The time-series studies are less likely than cross-sectional studies to be biased by inter-population differences (such as age, gender, life style), since the study population does not change substantially over such a short period of time, and acts as its own 'control'.

Time-series studies are 'ecological', since they consider very large groups of people (thousands or millions) rather than individuals, and no control of the experimental conditions is possible. Usually, no individual exposure data are available due to the impracticality of obtaining information on so many people. Exposure is inferred from centrally located outdoor ambient monitoring stations, an explicit difference from the controlled human exposure (clinical) studies. The lack of a direct link between personal exposure to the toxic agent and the resulting health outcome at the individual level is a weakness of ecological studies with respect to making judgments regarding causality. In compensation for this disadvantage, the strength of the time-series study is its ability to examine the overall population responses of very large numbers of individuals to the agent under investigation and, thereby, gain an understanding of the public health impacts and risks to the population as a whole. Causality must then be judged by making use of the sum of all the information available from all epidemiological, clinical, and toxicological studies in a weight of evidence approach.

Other environmental factors and other causes of illness may confound the results and must be taken into account in the time-series as well as the cross-sectional analysis. Daily mortality and morbidity (hospitalizations, emergency department visits) are usually highly cyclic, and undergo strong seasonal fluctuations, with events such as hospitalizations highest in winter and lowest in summer in North America. Ambient ozone concentrations also are highly seasonal, with highest levels in summer and lowest levels in winter. Such seasonal trends could bias the results, and they require some means of adjustment in order to determine whether there is any association or effect of ozone on these health endpoints. Many of the most recent papers using administrative databases have made use of more sophisticated statistical techniques to correct these confounding factors. The studies that did not, are weighted more lightly in view of the strong possibility that the findings are spurious, due to confounding.

Airborne pollution always occurs as a mixture of pollutant agents. Ozone co-exposure with particulate matter (PM) has been of great concern with respect to confounding the relationship between ozone and adverse health endpoints. Some, but not all, researchers have investigated the modifying effects of one or more co-occurring pollutants, including PM [as PM₁₀, PM_2.5, coefficient of haze (CoH), or total suspended particles (TSP)], NO₂, SO₂, or CO, on outcomes associated with ozone.

Many studies attempt to estimate the quantitative influence of ozone pollution on human health by calculating a parameter such as relative risk (RR) from the concentration-response relationships. This is presented with a measure of the uncertainty of the RR estimate, such as 95% confidence interval (95% CI). Uncertainty decreases as sample size increases, thus combining data sets is expected to yield more reliable estimates of relative risk. Combining data from several comparable studies in order to analyze them together is often referred to as meta-analysis. A meta-analysis of published peer reviewed studies is presented in this document for each category of health outcomes, based on the availability of the data, to evaluate whether they collectively indicate statistically significant associations for that outcome.

The criteria used in this assessment to select studies for inclusion in the quantitative meta-analysis, and, from within selected studies, to select from among several reported results are that the study must:

1. measure daily mortality or hospitalization (i.e., is a time series study);
2. report quantitative results for ozone;
3. be an original study (rather than a review paper or an abstract) in a peer-reviewed publication;
4. consider the entire population (rather than only a subset of the population) in the study location;
5. adjust for effects of some measure of seasonal cycle, temperature and relative humidity;
6. report results from a co-pollutant model, including PM or some proxy for PM in the model with ozone; PM₁₀ or PM_2.5 is preferable to other measures of particulate matter, and more pollutants in the model is preferable to fewer pollutants; and
7. consider summer results when there are results from a whole year and/or from several seasons in the same study.

Reporting a statistically significant positive result for ozone is not a criterion for study selection, nor does statistical significance or size of the relative risk estimate affect the evaluation in selecting studies. In a meta-analysis, it is reasonable that a pooled estimate, that combines estimates from all selected studies, should give more weight to those estimates from the studies with the smaller variance. This gives greater weight to those estimates with lower associated uncertainties. Variance takes into account both the consistency of data and the sample size used to obtain the estimate, two key factors that influence the reliability of results.

Acute effects: Mortality

Overall mortality studies indicate that there was a significantly positive association between ozone pollution and non-accidental mortality [Table 3 (SAD Table 12.1b)]. Seventeen of the 23 mortality studies (Table 3) reported consistent and significant associations between increases in mortality and ozone air pollution. These associations could not be explained on the basis of yearly trends, day-to-day variations, epidemics, or weather. The latter is the most important source of variation with respect to the ozone mortality association, because ozone is often moderately correlated with temperature in summer (correlation coefficient up to 0.35), while extreme temperatures have also been associated with increased mortality. All studies took temperature into account in some way in their regressions, and also included all other cyclic factors that were shown to influence the results during preliminary analyses. These associations were found in cities across North America, in four U.S. and 13 Canadian locations, in Santiago Chile and three European cities, and in a meta-analysis including seven European cities, demonstrating consistency of results despite widely varying climatic conditions and pollutant mixtures.

The mortality association was found at mean ozone concentrations (daily one-hour maximum) between 20 and 75 ppb, below the current Canadian National Ambient Air Quality Objective for ozone of 82 ppb, across widely varying climatic conditions and pollutant mixtures in the study locations (see Table 3). A pooling of ten studies which had adjusted for seasonal cycles, weather terms and co-pollutants, reveals that the weighted estimate of risk of total non-accidental mortality is a 0.40% increase in mortality (95% CI 0.19 - 0.60%) per 10 ppb increase in ozone (daily 1-hour maximum). Several studies indicate that the ozone concentration-response relationship is approximately linear, in one case down to less than 10 ppb. In addition, these studies did not show evidence of thresholds at low concentrations. Only a few studies provided evidence of a more specific cause of death; cardiovascular deaths were associated with ozone increases in several studies while respiratory deaths were not, possibly because of proportionally fewer deaths in the latter category.

The results from the most recent reanalysis of a large Canadian database by Burnett (1998, details in Appendix A of the SAD) demonstrate a strong consistency with the previous mortality studies. This study was carried out in response to a request from the Working Group on Air Quality Objectives and Guidelines. The non-accidental mortality data from 13 cities were reviewed. The regression analysis shows that the risk for non-accidental mortality was 0.79% (95% CI: 0.59-0.99%) for every 10 ppb increase in ozone (daily 1-hour maximum). Since previous studies using similar databases have shown that CO, NO₂ and SO₂ did not confound ozone effects on mortality, it is expected that the ozone mortality impact is likely to be independent of other gaseous air pollutants. The advantage of this study is that it provided an estimate of the lowest observed adverse effect level (LOAEL) of ozone with statistical significance for mortality. The LOAEL for non-accidental mortality was 20 ppb (1-hr maximum) (p<0.05). The data continue to show a trend of positive association with ozone values as low as 10 ppb. There appears to be no threshold for mortality.

Six studies (7 locations) did not find any association between ozone and mortality. Some differences in the statistical treatment and/or data limitations (such as exposure data from only one monitor) were identified. These differences may explain the lack of association between ozone and mortality for these studies.

Table 3. Summary of relative risk estimates in daily mortality for each 10 ppb increase in ozone, in univariate and multi-variate models (full references in Science Assessment Document).

Location and reference	Ozone mean, ppb (range)(1-h max., unless indicated)	Percent increase (95% CI) per 10 ppb ozone (1-h max., unless indicated otherwise). Ozone only models	% increase (95%CI) per 10 ppb ozone (1-h max., unless indicated otherwise). Multi-pollutant models	Inclusion or exclusion in meta-analysis S: Significant NS: Non- significant
Detroit, MI (Schwartz 1991)	not given	no increase	----	Results NS Excluded, no quantitative results.
Los Angeles Co. CA (Kinney, Ozkaynak 1991)	75 ± 45	Increased, % not given	4%, O3 and NO2	Results S Excluded, no quantitative results for multi- variate model.
St. Louis MO Harriman TN (Dockery et al. 1992)	22.5 ± 18.5 (24-h mean) 23.0 ± 11.4 (24-h mean)	24-h ozone: St. Louis: 0.29% ( -1.18% to 2.94%),. Harriman: -0.64% (-3.93% to 3.29%)	---	Results NS Excluded, no quantitative results for multi- variate model.
New York City NY (Kinney, Ozkaynak 1992)	56 (range not given)	[5.5 deaths per 10 ppb]	10%, O3 and COH	Results S Excluded, no quantitative results for multi- variate model
Philadelphia PA (Li & Roth 1995)	19.8 ± 14.4 (24-h mean)	Increased for age 65y+ not increased for age <65	---	Results S Excluded, no quantitative results for multi- variate model
Philadelphia PA (Moolkavgar et al. 1995)	19.9 (year) (24-h mean) 35.5 (summer) (1.3- 159)	Yearly data not given separately Summer: 1.5% (0.9-2.1%), 24- h ozone	Yearly: 0.62% (0.18- 1.04%) Summer: 1.5% (0.7-2.4%) +TSP, SO2, 24-h ozone	Results S Included
Los Angeles CA (Kinney et al 1995)	70 ± 41 (3-201), yearly	0.2% (0% - 0.5%), yearly ozone	0% (-0.44 to 0.4%) +PM10	Results S Included
Toronto ON (Ozkaynak et al. 1995)	36 (95th percentile 66)	(not given separately)	0.34% to 0.42% (+TSP)	Results S Excluded, an abstract
Sao Paulo Brazil (Saldiva et al. 1995)	38.3 + 29.7 (1-h max.), 12.5 + 11.5 (24-h mean)	For age 65+ years. 1-h: 0.4% (-0.42% to 1.03%); 24-h: -1.31% (-4.15% to 1.75%)	---	Results NS Excluded, no data
Philadelphia PA (Dockery et al. 1996)	not given	(not given separately)	0.9% (+PM2.5)	Results NS Excluded, an abstract
Chicago IL (Ito & Thurston 1996)	38 ± 19.9, yearly, 2- d average.	1.0% (0.6-1.5%), yearly ozone, 2-d average	2-day average ozone. 0.68% (0.08-1.16%) (+PM10)	Results S Included
Santiago Chile (Ostro et al. 1996)	52.8 (11-264)	Yearly: OLS model: -0.57% (-0.75 to - 0.38%); Poisson: 0% (-0.19% to 0.38%) Summer: OLS model: 0.20% (0-0.50%) Poisson: 0.4% (0-1.0%)	Yearly: OLS model: -0.56% (- 0.92 to 0%); Poisson: - 0.20% (-0.56 to 0.20) Summer: OLS model: 0.38% (-0.57 to 1.32%); Poisson: 0.4% (0-0.9%) +PM10	Results S Included
Mexico City DF (Loomis et al. 1996)	154 (26-319) (1-h max) 62 (12-130) (24-h mean), yearly	Yearly, 1-h max. ozone: 0.24% (0.11-0.39%) Yearly, 24-h max. ozone: 0.58% (0.22-0.94%)	Yearly, 1-h max. ozone: -0.18% (-0.52 to 0.16%) (+TSP and SO2)	Results S Included
London UK (Anderson et al. 1996)	Yearly: 20.6 ± 13.2 (1-h max.), 15.5 ±10.9 (8-h mean); summer: 7-36 (8-h) 11-45 (1-h)	Yearly ozone: 1-h max: 0.83% (0.42-1.25%) 8-h mean: 1.01% (0.46-1.57%); Summer: 1-h max. 1.03%(0.53-1.53%) 8-h mean: 1.2%(0.6-1.8%)	Yearly, 8-h average ozone: 1.14% (0.59 -1.69%) with black smoke. Summer, 8-h average ozone: 1.45% (0.7-2.19%) with BS. No 1-h max. data reported.	Results S Included
Barcelona Spain (Sunyer et al. 1996)	28 (3.6-96) winter 44 (4.8-144) summer	Yearly: 0.96% (0.24-1.72%) Summer: 1.16% (0.34-2.02%)	---	Results S Excluded, no quantitative results for multi- variate model
Amsterdam NL (Verhoeff et al. 1996)	21.9 (4-41; 10-90th %), yearly	yearly ozone 0.98% (0.02-2.0%),	Yearly: 0.58% (-0.67 to1.9%) + BS 1.0% (-1.1 to 3.3%) + PM10	Results S Included
Lyon France (Zmirou et al. 1996)	7.75 (0-72) (1-h max) 5.1 (0-40.2) (8-h mean), yearly	Increase not significant 1-h max.: 1.6% (-2.4% to 6.4%) 24-h avg.: 1.2% (-2.0% to 4.8%), yearly ozone	--- ---	Results NS Excluded, no quantitative results for multi- variate model
Paris France (Dab et al. 1996)	22.3 (3.1-74.8) (1-h max 14.1 (10-56) (8-h mean), yearly	Respiratory mortality only; Yearly ozone; 1-h max.: 0.8% (-1.3% to3.1%) 24-h avg.: 1.5% (-1.2% to 4.6%),	--- ---	Results NS Excluded, no quantitative results for multi- variate model
4 APHEA cities: Athens Barcelona London Paris 3 additional cities: Amsterdam Basel Zurich (Touloumi et al. 1997)	47.7 ± 21.8 (all 1-h max) 36.8 ± 17.8 21.0 ± 13.2 23.5 ± 16.7 21.9 (4-41; 10-90th %) no 1-h values given for Basle or Zurich (8-h means 12 (B) and 14 (Z)), yearly ozone	Meta-analysis pooled estimate (random effects): For a single day: 1.16% (0.40-1.96%); Average of 2-5 day cumulative ozone: 0.96% (0.48 - 1.48%) Plus 4 non-APHEA cities: 1.16% (0.40% to 1.96%) (1-h max.) All yearly ozone data,1-h max. ozone.	Meta-analysis (random effects): 1.12% (0.20-2.00%) +BS 1.28% (-0.12% to 2.72%) +NO2 Pooled estimates with non-APHEA cities not included in bivariate analysis	Results S Included for 4 APHEA cities.
Philadelphia PA (Kelsall et al. 1997, Samet et al., 1997)	19.9 ± 14.6 (2-day average) 8.3-28.5 IQR (=20.2)	2-d average ozone: 1.13% (0.4% - 1.9%)	2-d average ozone: 1.01% (0.02-2.0%)+TSP 1.11% (0.38- 1.84%)+SO2 1.12% (0.73- 1.84%)+NO2 1.17% (0.40- 1.94%)+CO2 0.96% (0.33-1.59%) + all four	Results S Included
Zurich, Basle and Geneva, Switzerland, 1984 - 1989. (Wietlisbach et al., 1996)	Mean ± SD (presumably 24-h average): Zurich: 13.4 ± 10.5; Basle: 12.0 ± 9.7; Geneva: 0	Data are expressed as regression coefficients; no SD or confidence interval reported. No ozone data for Geneva. For total mortality, no association in Zurich or Basle. For mortality of people >65 years of age, no association in Zurich, significant association in Basle (p<0.05). For respiratory and cardiac mortalities, no association in any cities.	---	Results NS Excluded, no relative risk data reported; exposure data not detailed (1-h max. vs. 24-h average).
11 Canadian cities (Burnett et al. 1998)	16.2 (24-h mean of 11 cities) IQR 13 3-35 (5-95 th centile), yearly ozone	24-h ozone: 0.86% (0.35-1.37%), yearly ozone	24-h average ozone: 1.11% (0.66-1.56%) (+NO2. SO2, CO) (particles contribute another 1%)	Results S Included
13 Canadian cities [Burnett 1998 (special analysis for WGAQOG)]	31 (1-h max) (25-38 range of means, 13 cities) 32.9 ± 16.7 (16 cities)	0.79% (0.59-0.99%), yearly ozone	---	Results S Excluded, no data on multi- variate analysis; not in a peer reviewed journal.
Multi-pollutant analyses, mean increase (%) in mortality ± SD = 0.523% ± 0.444% (n=10), per 10 ppb increase in ozone (1-h max.), 95% CI: -0.347% to 1.39%, p>0.05. Meta-analysis of multi-pollutant studies, weighted mean increase (%) in mortality ± SD = 0.399% ± 0.105% (n=10), per 10 ppb increase in ozone (1-h max.), 95% CI: 0.193-0.604%, p<0.05.