Ozone exposure and pulmonary effects in panel and human clinical studies: Considerations for design and interpretation

ABSTRACT

A wealth of literature exists regarding the pulmonary effects of ozone, a photochemical pollutant produced by the reaction of nitrogen oxide and volatile organic precursors in the presence of sunlight. This paper focuses on epidemiological panel studies and human clinical studies of ozone exposure, and discusses issues specific to this pollutant that may influence study design and interpretation as well as other, broader considerations relevant to ozone-health research. The issues are discussed using examples drawn from the wider literature. The recent panel and clinical literature is also reviewed. Health outcomes considered include lung function, symptoms, and pulmonary inflammation. Issues discussed include adversity, reversibility, adaptation, variability in ozone exposure metric used and health outcomes evaluated, co-pollutants in panel studies, influence of temperature in panel studies, and multiple comparisons. Improvements in and standardization of panel study approaches are recommended to facilitate comparisons between studies as well as meta-analyses. Additional clinical studies at or near the current National Ambient Air Quality Standard (NAAQS) of 70 ppb are recommended, as are clinical studies in sensitive subpopulations such as asthmatics.

Implications: The pulmonary health impacts of ozone exposure have been well documented using both epidemiological and chamber study designs. However, there are a number of specific methodological and related issues that should be considered when interpreting the results of these studies and planning additional research, including the standardization of exposure and health metrics to facilitate comparisons among studies.

Introduction

Ozone, a secondary pollutant generated from nitrogen oxide and volatile organic hydrocarbon precursors, has well-documented pulmonary effects that have been observed consistently in epidemiological, human clinical, and toxicological studies (U.S. Environmental Protection Agency [EPA] 2013). Ozone has been associated with hospital admissions and emergency department visits for respiratory causes in population-based epidemiological studies (for a recent systematic review and meta-analysis see Zheng et al. 2015). In panel studies, ozone has been associated with asthma exacerbation as evidenced by increased symptoms such as cough, wheeze, and chest pain, as well as decrements in pulmonary function (e.g., Schachter et al. 2016; Samoli et al. 2017; also see review by Li et al. 2012). Human clinical studies in which individuals are exposed to ozone in a controlled laboratory setting have similarly shown increased respiratory symptoms and decreased lung function. Animal studies have provided supportive evidence for these effects and have additionally been able to investigate biological mechanisms (Gordon et al. 2016; Kumarathasan et al. 2015). As a result of these findings taken as a whole, during the last review of the National Ambient Air Quality Standards (NAAQS) in the United States, the Integrated Science Assessment associated with that review (EPA 2013) concluded that there is a causal relationship between short-term ozone exposure and respiratory effects.

While the body of published literature on ozone and pulmonary effects is large, there does not yet exist an article that integrates the panel and human clinical study literature, and that focuses on issues relevant to these disciplines that may both influence our understanding of the knowledge base and inform future studies in this area. This paper focuses on design, methodological, and interpretation issues relevant to both panel studies and human clinical studies of ozone and pulmonary effects; health outcomes considered include lung function, symptoms, and inflammation. The paper does not comprehensively or systematically review the totality of all panel and human clinical studies of ozone exposure, although it does initially provide an overview of more recent studies in order to provide the reader with the broad state of the science in this area. These studies were identified by conducting PubMed searches with combinations of search terms “ozone, “panel,” “longitudinal,” “epidemiology,” “pulmonary,” “respiratory,” “lung,” “inflammation,” and “symptoms.” Following the reviews of the panel study and clinical study literature, the paper considers in turn issues of adversity, reversibility, adaptation, variability in ozone exposure metric used and health outcomes evaluated, co-pollutants in panel studies, influence of temperature in panel studies, and multiple comparisons. These issues were selected as they were considered to be important in interpreting and comparing results across studies; however, there are clearly other influential factors, such as age and health status of the study population. In the sections of the paper that address the various issues, relevant citations are provided as illustrative examples.

Panel studies of ozone exposure

Panel studies in the air pollution setting are a type of cohort study in which individual-level measurements of exposure and health outcomes are collected longitudinally. The repeated measurements provide enhanced ability to establish temporality and causation, and because individuals act as their own control some confounding factors are minimized. Even a panel study with a small number of subjects can be informative given a sufficient number of repeated measurements (Weiss and Ware 1996). In fact, panel studies typically have a relatively small number of participants and are conducted over a relatively short period of time (several weeks to several months), largely due to the resource-intensiveness of this type of study and the demands on participants, which make recruitment a challenge. For ozone panel studies of pulmonary effects, study characteristics include the population being studied (e.g., children vs. adults, healthy individuals vs. asthmatics), location, study period (e.g., summer or other season, year-round), duration of study period for a participant (i.e., how long the participant was studied), ozone exposure metric, and specific health outcome. The latter two characteristics (exposure and outcome) are covered in detail later in this paper.

A review of more recent panel study literature (post 2000) using PubMed revealed 22 panel studies in either children or adults that evaluated the pulmonary effects of ozone. Table 1 provides details related to the design and results of these studies; they are also briefly summarized below. Of these studies, 4 were conducted in adults and 18 were conducted in children. Thirteen were of sensitive subjects—most commonly, in subjects with asthma—and 9 were of healthy individuals. Most studies (16) included assessment of lung function, with a majority (14) evaluating respiratory symptoms; seven studies evaluated both outcomes. A small number of studies assessed other outcomes, such as medication use and fractional exhaled nitric oxide (FeNO), a marker of pulmonary inflammation. The sections that follow first summarize studies in children and then in adults; within each group, papers are discussed chronologically. Overall, results are mixed in these 22 studies, with variable and somewhat inconsistent findings.

Table 1. Summary of post-2000 panel studies that evaluated associations between ozone and pulmonary effects in children or adults

Reference	Location	Study period	Study population	Ozone metric(s)	Mean ozone concentration (ppb) (SD)	Lag(s)	Outcome(s)	Primary results	Multipollutant models
Altug et al. 2013	3 different areas in Eskisehir, Turkey	2 weeks in 2 seasons in 2008 and 2009	1880 students aged 9–13 years	Average of two 1-week integrated samples collected outside at studied schools	23.9 (6.9) to 55.0 (13.9) depending on season/location	—	FVC, FEV1, PEF, MMEF, FEV1/FVC%, symptoms	Significantly reduced lung function for girls in summer season; no other results significant.	—
Altug et al. 2014	3 different areas in Eskisehir, Turkey	2 weeks in March 2009	695 students aged 9–13 years	1-week avg	23.9 (6.9), 39.3 (13.0), and 43.9 (13.2) for three locations	—	FVC, FEV1, PEF, MMEF, FeNO, upper respiratory tract complaints, medication use	Significant association with upper respiratory tract complaints; significant negative association with PEF in children without respiratory tract complaints. No significant associations for FeNO.	—
Castro et al., 2008	Rio de Janeiro, Brazil	3 months in 2004	118 children aged 6–15 years	24-hr avg	40.5 (18.8)	1-, 2-, and 3-day; 2- and 3-day MAs	PEF	Significant increase in peak flow at lag 1; no significant associations for any other lags.	—
Chan and Wu 2005	Taichung City, Taiwan	November-December 2001	43 mail carriers	8-hr avg, 8-hr max (both during carriers’ working periods)	8-hr avg: 35.6 (12.1); 8-hr max: 52.6 (18.8)	0-, 1-, 2-, and 3-day	Nightime PEFR; nighttime PEFR deviation	8-hr average ozone significantly associated with nighttime PEFR (and deviation) at lags 0, 1, and 2 days; 8-hr max ozone significantly associated with both measures at lags 0 and 1 day.	No change in results with PM10 and NO2.
Dales et al. 2009	Windsor, Canada	28 days in 2005	182 asthmatic children aged 9–14 years	24-hr avg; 24-hr max	24-hr avg: 14.1 (6.0); 24-hr max: 27.2 (8.6)	—	FEV1	No significant associations for either ozone measure.	No change in results with PM2.5, NO2, or SO2.
Delfino et al. 2002	Alpine, CA	March-April, 1996	22 asthmatic children aged 9–19 years	1-hr max, 8-hr max	1-hr max: 69 (16); 8-hr max: 60 (12)	0-, 1-, 2-, 3-, 4-, 5-day; various moving averages selected based on individual lag results	Symptoms (none or not bothersome vs. interfering with daily activities)	No significant associations for either ozone measure at any lag, or in groups of children on or not on medication.	Interactions between ozone and NO2 observed, but risk estimates not reported.
Delfino et al. 2003	Los Angeles, CA	Up to 3 months, 1999–2000	22 Hispanic children with asthma, aged 10–16 years	1-hr max, 8-hr max	1-hr max: 25.4 (9.6); 8-hr max: 17.1 (7.2)	0-, 1-day	Symptoms (0 = none; 1 = not bothersome; 2 = bothersome or more severe), PEF	Significant increase in ORs for symptom scores >2 for 1-hr and 8-hr max ozone at lag 0; no significant associations for symptom scores >1. Results for PEF not reported.	—
Escamilla-Nunez et al. 2008	Mexico City, Mexico	Average of 22 weeks during 2003–2005	147 asthmatic children (6–14 yrs), 50 healthy children (6–14 yrs)	1-hr max	86.5 (34.4)	0 and 1 day; 2–5 day MAs	Presence/absence of cough and wheeze; medication use	Significant increase in cough, wheeze,and bronchodilator use at lag 1 (cough also at lag 0) and cumulative lags 2–5 in asthmatic children. No significant associations in healthy children.	With PM2.5 and NO2, results similar.
Jalaludin et al. 2000	Sydney, Australia	11 months in 1994	125 children with history of wheeze (mean age 9.6 years)	Daytime avg (06:00–21:00); daily max	avg: 12.0 (6.8); max: 26.0 (14.4)	0, 1, 2, 3, 4; 2- and 5- day average	Daily mean variation in PEFR (ΔPEFR)	Significant negative association between daily average ozone and ΔPEFR; no significant association for daily max ozone.	With PM10 and NO2, little change in coefficients.
Jalaludin, O’Toole, and Leeder 2004	Sydney, Australia	11 months in 1994	125 children with history of wheeze (mean age 9.6 years)	Daytime avg (06:00–21:00); daily max	avg: 12.0 (6.8); max: 26.0 (14.4)	0, 1, 2, 3, 4; 2- and 5- day average	Symptoms, medication use, doctor visits for asthma	No significant associations for any outcome.	With PM10 and NO2, little change in coefficients.
Just et al. 2002	Paris, France	3 months in 2006	82 asthmatic children (mean age 10.9 years)	8-hr avg	29.5 (12.3)	0, 1, 2; average of 0–2 days and 0–4 days	Symptoms, PEF	Significant association between increase in PEF variability and 0–2 day average ozone.	BS not significant in model with BS, O3 and int term between O3 and temperature. O3 not significant in model with NO2, O3 and int term between O3 and temperature.
Karakatsani et al. 2012	Amsterdam, The Netherlands; Athens, Greece; Birmingham, UK; Helsinki, Finland	6-month period in 2002–2004	136 adults aged ≥ 35 years diagnosed with asthma, COPD, or chronic non-specific lung disease	24-hr avg	Helsinki 21.3; Athens 23.5; Amsterdam 16.6; Birmingham 18.7 (all medians)	0-, 1-, 2-day; average of lag 0–6	Symptoms	Significant increase in waking with breathing problems at lag 0 in total study group; significant negative association with shortness of breath at lags 1 and 2 in asthmatic group; significant increase in cough at lags 0, 1, and 2 in total study group and asthmatic group	—
Lagorio et al. 2006	Rome, Italy	May–June and November–December 1999	29 adults with COPD (11), asthma (11), or IHD (7)	24-hr avg	33.1 (5.8) in spring; 7.5 (3.8) in winter	1-day, cumulative 2- and 3-day	FEV1, FVC	No significant associations for ozone and any endpoint.	—
Lewis et al. 2005	Detroit, MI	Six 2-week periods in 2001–2002	86 asthmatic children aged 7–11 years	24-hr avg	24-hr avg 27.6 (12.5); peak 8-hr avg 40.4 (18.2) (for Eastside site; concentrations similar at Southwest site)	1- and 2-day; 3–5 day average	FEV1	No significant associations with 24-hr average ozone among children not using maintenance corticosteroids (CS); daily 8-hr peak ozone significantly associated with FEV1 decrements at lag 2.	With PM2.5, significant association at lag 3–5; with PM10, significant association at lags 1, 2, and 3–5.
Mortimer et al. 2002	Eight urban areas in the United States	Summer 1993	846 asthmatic children aged 4–9 years	8-hr avg	Values not provided; Figure 1 in this paper shows mean ranging from ~40–60 depending on urban area. Mean over all 8 areas = 48.	1-, 2-, 3-, 4-, 5-, 6-day; 1–5 day average; 1–4 day average, 0–4 day average, 0–3 day average	Symptoms, PEFR	Significant associations between decline in morning PEFR and 5-day average ozone, increased morning symptoms and 4-day average ozone. No associations for evening measures.	For morning symptoms, addition of SO2 to the model removed ozone significance.
Ostro et al. 2001	Los Angeles and Pasadena, CA	August–November, 1993	138 African-American children with asthma, aged 8–13 years	1-hr max	Los Angeles: 59.5 (31.4); Pasadena: 95.8 (49)	0-, 1-, 2-, 3-day; 4-day moving average	Symptoms, medication use	No significant associations for shortness of breath, cough, or wheeze. In moderate and severe asthmatics, 2-day moving average ozone and lag 1 day significantly assocated with extra medication use.	—
Peacock et al. 2003	3 schools in Medway, England	13 weeks in 1996–1997	179 children aged 7–13	8-hr max moving avg, 1-hr max	8-hr max MA: 21.6 (11.1) for rural; 19.0 (10.6) for urban. 1-hr max: 25.8 (11.1) for rural, 23.1 (10.2) for urban.	0, 1, 2 days; 1–4 day average	PEFR	Either null or significantly protective associations between ozone and PEFR when all schools analyzed together.	—
Samoli et al. 2017	Athens and Thessaloniki, Greece	5 weeks in 2013–2014	186 children aged 10–11 years at schools in low- and high-ozone areas	24-hr avg personal and fixed site	Athens, fixed site: 22.3 (9.4) to 32.1 (11.1); personal: 4.1 (3.9) to 5.4 (3.9). Thessaloniki lower.	0, 1 day	PEF, symptoms	Significantly increased odds of stuffy nose with personal but not fixed-site data; no significant associations with other symptoms or with PEF.	With PM10, no change.
Schachter et al. 2016	New York, NY	Two 2-week periods in summer and winter seasons during 2008–2009	36 asthmatic children aged 6–14 years	24-hr avg	Summer: 25.3 (8.7); winter: 16.6 (9.0)	—	Symptoms, PEF	Ozone significantly associated with symptoms and reduced morning FEV1 in summer but not winter.	—
Schildcrout et al. 2006	8 North American cities	1993–1995; followed for median of 2 months	990 children aged 5–12 years	1-hr max (warm season)	Means not provided; median ranged from 43.0 to 65.8 depending on city	0, 1, 2 days; 0–3 day average	Symptoms, medication use	No significant associations for any outcome.	—
Ward et al., 2002	Birmingham and Sandwell, England	Two 8-week periods (summer and winter) in 1996	162 children aged 9 years	24-hr avg	Means not provided; winter median 13.0, summer median 22.0	0, 1, 2, 3 days; 0–7 day average	Symptoms, PEF	In winter, significant associations for DPEF in morning at 0–7 average lag; some significant associations for symptoms. In summer, significant association for DPEF in morning at 2 day lag and 0–8 day average lag, some significant associations with wheeze.	—
Yoda et al. 2014	Tokyo, Japan	2 weeks during summer 2012	21 healthy female students aged 20–23 years	24-hr avg	Values not provided; Figure 2 in this paper shows range of ~10–60 ppb	lag 0, 1; 2 to 5 day average	PEF, FEV1, EBC pH, FeNO	Significant reduction in EBC pH at all lags; nonsignificant associations for FeNO, PEF, FEV1	Ozone association for EBC pH significantly decreased after adjusting for NO2 or suspended PM.

Note. max, Maximum; avg, average

Jalaludin et al. (2000) found that 1-day lagged daily average ozone exposure was associated with significantly increased variation in peak expiratory flow (PEF) in Australian children with a history of wheeze; no significant associations were observed for other lags or with daily maximum ozone. In a panel of children in Los Angeles, CA, Ostro et al. (2001) found that ozone was not associated with symptom occurrence in African American children in Los Angeles, although it was associated with asthma medication use in some subanalyses. Just et al. (2002) found that 0–2 day average lagged ozone was associated with increased daily PEF variability in asthmatic children in Paris, France. In a study of asthmatic children in eight cities in the United States, Mortimer et al. (2002) reported that 5-day average ozone concentrations were associated with decreased morning PEF; 4-day average ozone concentrations were associated with exacerbated morning asthma symptoms (but not evening asthma symptoms). Ward et al. (2002) studied school children in the United Kingdom and evaluated both lung function (PEF) and symptoms at several different lags of ozone concentrations. They found several significant associations for PEF and symptoms; however, these were inconsistent in direction and there was no discernable pattern with respect to lags. Delfino et al. (2002) reported no significant associations between 1-hr maximum and 8-hr maximum ozone and asthma symptoms at any lag, or in subgroups of asthmatic children taking or not taking anti-inflammatory medications in southern California. Delfino et al. (2003) found significant associations between both 1-hr maximum and 8-hr maximum ozone concentrations at lag 0 (but not lag 1) and bothersome or more severe symptoms in a small panel of asthmatic children; however, while PEF was measured in the study, no associations for that outcome were reported in the paper.

In the United Kingdom, Peacock et al. (2003) studied school children (asthmatic and nonasthmatic) and found mostly null associations, with some significant protective associations, between 1-hr maximum and 8-hr maximum moving average ozone and morning PEF. Jalaludin, O’Toole, and Leeder (2004) studied a group of Australian children with a history of wheeze and reported no significant associations between ozone concentrations and respiratory symptoms, asthma medication use, or doctor visits. Lewis et al. (2005) found no significant associations between 24-hr average ozone and FEV1 decrements among asthmatic children not using maintenance corticosteroids; daily 8-hr peak ozone was significantly associated with FEV1 decrements at lag day 2. Schildcrout et al. (2006) studied children in eight U.S. cities and found nonsignificant associations between warm season ozone and symptoms at lags 0, 1, and 2 days, as well as a 3-day moving average. There were also no significant associations with rescue inhaler use. In Mexico City Escamilla-Nunez et al. (2008) reported that an increase in 1-hr maximum ozone was associated with increased cough and wheeze at lag 1 and cumulative lag 2–5 in asthmatic children. Castro et al. (2009) studied Brazilian children with a history of wheeze and reported a protective effect of ozone, with 1-day lagged ozone associated with a significant increase in PEF; however, other lag structures (lag 2, lag 3, 2-day average, and 3-day average) did not show significant findings. Dales et al. (2009) studied a panel of asthmatic children in Windsor, Canada, and found that maximum ozone was not associated with FEV₁ changes; the results did not differ when considering averaging times of 0–12 hr, 12–24 hr, or 0–24 hr.

Altug et al. (2013) studied children in Turkey and found that 2-week average ozone was significantly associated with impaired lung function in girls in the summer season; no such association was found in boys or in the winter. No associations were reported for symptoms. In a subanalysis of data from this study, Altug et al. (2014) reported significant associations between 1-week average ozone and upper respiratory complaints; a significant negative association was found with PEF in children without respiratory complaints. Associations for exhaled nitric oxide were nonsignificant. Schachter et al. (2016) evaluated asthmatic children in New York and found that ozone was associated with more severe symptom score, increased albuterol puffs, and decreased FEV₁ in summer, but not winter. In Greece, Samoli et al. (2017) followed 186 children for 5 weeks and recorded daily symptoms, absenteeism, and peak expiratory flow (PEF). They found that personal ozone exposure was associated with significantly increased odds of any symptom, largely attributable to the odds of stuffy nose, but not changes in PEF.

As mentioned in the preceding, there are fewer published panel studies of ozone and pulmonary effects in adults. Chan and Wu (2005) examined the relationship between ozone and PEF in 43 mail carriers and found that nighttime PEF and deviation in nighttime PEF were significantly associated with both 8-hr average and 8-hr maximum ozone concentrations (8-hr exposure period was approximately 9 am to 5 pm). Lagorio et al. (2006) studied 29 adults with chronic obstructive pulmonary disease (COPD), asthma, or ischemic heart disease (IHD) and found no significant associations between 24-hr average ozone (lagged 1 day, and cumulative exposure over 2 or 3 days) and either forced expiratory volume in 1 sec (FEV₁) or forced vital capacity (FVC). Karakatsani et al. (2012) followed subjects in four European cities for a 6-month period and collected self-reported information on symptoms and activity restriction. Same-day, lag 1-day, and lag 2-day 24-hr average ozone was significantly associated with cough. In Tokyo, 21 healthy young women were studied for 2 weeks in summer, and pulmonary function was evaluated (Yoda et al. 2014). No significant changes in FEV₁, PEF, or fractional exhaled nitric oxide (FeNO) were reported in association with 24-hr average ozone; however, significant reductions in exhaled breath condensate (EBC) pH were noted at all lags examined (0- and 1-day; 2–5 day moving average).

Although the emphasis of this overview of the panel study literature is on more recent papers, it should be noted that many historical panel studies exist. These include a series of summer-camp studies that reported associations between ozone exposure and pulmonary function decrements in children (e.g., Spektor et al. 1988; Spektor et al., 1991; Higgins et al. 1990; Kinney, Thurston, and Raizenne 1996; Berry et al. 1991). It should be noted that these earlier studies were generally associated with relatively high ozone concentrations, with hourly maxima as high as 150 ppb (Spektor et al. 1991) and 245 ppb (Higgins et al. 1990). Several of the papers (Berry et al. 1991; Higgins et al. 1990) reported adverse impacts on FEV only at concentrations of 120 ppb and above. Nonetheless, the studies overall point to the impact of ambient ozone on respiratory function in children.

Human clinical studies of ozone

There is a rich history of human clinical studies—also called controlled human exposure studies—of ozone exposure. Many of these studies, conducted initially at high ozone concentrations, date back to the 1970s and 1980s (e.g., Silverman et al. 1976; Hazucha 1987; Folinsbee, McDonnell, and Horstman 1988; and review by Sandstrom 1995). These studies consistently demonstrated reversible pulmonary function decrements and transient symptoms. Some studies have also noted enhanced bronchial reactivity (Folinsbee and Hazucha 2000; Foster et al. 2000) and airway inflammation (Kim et al. 2011; Liu et al. 1999). Such studies have also investigated different types of exposure regimens; for example, some studies have involved square wave exposures, where ozone concentration is maintained at a constant value throughout the exposure period. Others have conducted triangular wave exposures, where ozone concentration is gradually increased to a peak level and then reduced similarly in a stepwise fashion. This latter approach is intended to mimic exposure in urban areas in which ozone builds up during the day due to photochemical activity and then gradually declines. Other variables besides the ozone concentration protocol that have been manipulated in chamber studies include exercise (with/without), exposure duration, and exposure method (facemask vs. whole-body exposure in a chamber). The now fairly standard exercise protocol that has been used in most recent chamber studies was developed in the late 1980s in response to interest in establishing an 8-hr ozone NAAQS (Folinsbee, McDonnell, and Horstman 1988). The protocol involves 50 min of exercise at a mean ventilation rate of approximately 40 L/min (target equivalent ventilation rate of 20 L/min/m² of body surface area—actual value depends on subject size), followed by 10 min of rest during each hour for the first 3 hr. This 3-hr period is followed by a 35-minute lunch break, which is followed by a second 3-hr exposure exactly the same as the first. The exercise is intended to reflect individuals who work outdoors in physically demanding professions such as construction.

This paper focuses on studies conducted at concentrations in the range of previous and current NAAQS (70 and 75 ppb, respectively) in order to provide a contemporary context to our understanding of pulmonary effects at environmentally relevant concentrations. As such, the paper does not review historical studies conducted at substantially higher levels (note, however, that when individual issues are discussed later in the article, older studies are used as illustrative examples). For this reason, the paper also does not review recent human clinical studies of ozone exposure that were conducted at higher concentrations, evaluated endpoints other than lung function, symptoms, or pulmonary inflammation, and/or were designed to answer different research questions (e.g., Bennett et al. 2016). As such, only five studies were identified that were conducted at concentrations below 80 ppb: Adams, (2002), Adams (2006), Schelegle et al. (2009), Kim et al. (2011), and Frampton et al. (2017).

Adams, (2002) exposed 30 healthy, nonsmoking young adults to several different ozone concentrations with the primary goal of comparing responses to facemask and chamber exposures. A secondary goal was to further explore responses to low levels of ozone using facemask exposures. Square-wave exposures to 40 ppb (facemask), 80 ppb (facemask), and 120 ppb (facemask and chamber) were conducted, along with a control exposure to filtered air (chamber). Subjects exercised for six 50-min periods for a total exposure time of 6.6 hr. Outcomes evaluated included changes in FEV₁ and FVC, changes in self-reported symptoms, and pain on deep inspiration. Results indicated statistically significant decrements in pulmonary function at 80 ppb and 120 ppb ozone, but not 40 ppb. Similarly, symptoms were only increased with 80 ppb and 120 ppb ozone. The main research question was adequately answered in that the two methods (facemask vs. chamber) resulted in very similar responses at the concentration that was tested using each exposure approach (120 ppb). However, findings from the facemask exposure at 80 ppb differed in magnitude and temporal pattern of response from earlier chamber studies conducted at the same concentration (e.g., Horstman et al. 1990; McDonnell et al. 1991). Thus, Adams recommended a direct comparison of facemask versus chamber exposures for this concentration using the same subject group, with each subject serving as his or her own control.

Adams (2006) followed up his earlier work with an investigation to determine whether there were differences in response to triangular- and square-wave exposures. As such, he constructed a triangular-wave exposure profile that resulted in an overall average concentration of 80 ppb, the same concentration employed for the square-wave exposure. Exposures were also conducted to triangular- and square-wave mean concentrations of 60 ppb, and an exposure was also conducted to a 40 ppb triangular-wave concentration. Thirty young adults served as subjects and the exercise protocol was the same as in Adams (2002). Exposure to 80 ppb ozone, with either the triangular- or square-wave protocols, caused a significant change in pulmonary function as evidenced by declines in FEV₁ and FVC; symptoms were also significantly increased at this concentration. Responses to the other exposure protocols were not different from one another or from responses to filtered air. Adams found significantly reduced pulmonary function in response to the square-wave protocol at 6.6 hr, versus 4.6 hr for the triangular-wave protocol (when ozone was 150 ppb). Adams concluded that ozone concentration has a greater effect on total inhaled dose than exposure duration.

Schelegle et al. (2009) conducted a chamber study with the objective of identifying the threshold ozone concentration at which effects on pulmonary function and symptoms are induced. Thirty-one healthy adults were exposed to filtered air and four variable hourly protocols (triangular wave) with mean ozone concentrations of 60, 70, 80, and 87 ppb; it should be noted that while those concentrations were the goal, the actual exposures were to 63, 72, 81, and 88 ppb. The standard 6.6-hr exercise protocol was followed. The hourly concentration profiles were designed to evaluate the relative importance of mean ozone concentration versus peak dose rate. The maximum concentrations for each of the four exposure profiles (90, 90, 150, and 120 ppb) occurred in the fourth hour. Schelegle et al. found that significant decrements in FEV₁ and increased symptoms occurred with exposure to the 72, 81, and 88 ppb average ozone protocols, but not with 63 ppb.

Kim et al. (2011) conducted a chamber study in which 59 healthy young adults were exposed to 60 ppb ozone (square wave) and filtered air. Lung function was measured immediately before and after a 6.6-hr exposure with intermittent exercise as previously described. While data were additionally collected during the exposure, these were not analyzed in an effort to reduce the need for control for multiple comparisons; this increased the statistical power of the study. While subjects did show significant decrements in pulmonary function (FEV₁ and FVC) after ozone exposure compared to filtered air, there was no significant change in symptoms. Induced sputum was collected from a subset of 24 participants and polymorphonuclear neutrophils (PMNs) were measured. There was a significant increase in PMNs for all subject combined, indicating an inflammatory response.

Frampton et al. (2017) exposed 87 older adults (ages 55–70 years) to either 0, 70, or 120 ppb ozone (randomized) for 3 hr in the MOSES (Multicenter Ozone Study in Older Subjects) study. During the exposures, participants exercised on a stationary bicycle, alternating 15 minutes of exercise with 15 minutes of rest. While the primary health outcomes were related to cardiovascular impacts, secondary outcomes included lung function decrements, airway inflammation as represented by sputum PMNs and soluble markers, and respiratory symptoms. Spirometry was conducted 30 min prior to exposure, immediately postexposure, and 1 day after exposure. Sputum was induced 1 day after exposure. Symptoms were evaluated 30 min prior to exposure, immediately postexposure, 3–4 hr postexposure, and 1 day after exposure. During clean air exposure, FEV₁ and FVC increased compared with preexposure values, and this effect persisted the following day. These improvements in lung function were shown to be attenuated in a dose-dependent manner with ozone exposure. There were no changes in FEV₁/FVC or FEF_25-75, suggesting that bronchoconstriction was not occurring. A significant increase in the percent PMNs in sputum was noted 22 hr after exposure to 120 ppb ozone; no significant increase was observed after 70 ppb exposure. No significant changes in sputum cluster of differentiation 40 (CD40) ligand, interleukin-6 (IL-6), interleukin-8 (IL-8), tumor necrosis factor-α (TNF-α), or total protein were observed. Ozone exposures did not significantly affect respiratory symptoms.

The following sections discuss current issues or concepts specific to clinical and/or panel studies. These include adversity; reversibility; adaptation; ozone exposure metric and related issues in panel studies; health outcomes in panel studies; co-pollutants and influence of temperature in panel studies; and multiple comparisons.

Adversity

In 2000, the American Thoracic Society (ATS) developed and published a statement on what constitutes an adverse health impact of air pollution. This guidance (ATS 2000) differentiated between reversible and irreversible effects and stated that while healthy persons may experience transient reductions in pulmonary function associated with air pollution exposure, these small transient decrements alone should not be automatically designated as adverse. The committee recommended that reversible loss of lung function in combination with the presence of symptoms should be considered as adverse. It should be noted that while ATS uses the descriptor “small,” it does not provide specific guidance regarding a cutoff value. However, a value of 10% has been used by the U.S. EPA in the rationale for setting the ozone NAAQS (US EPA, 2013).

The ATS committee noted that any degree of permanent lung function loss attributable to air pollution exposure should be considered as adverse. ATS also discusses the population distribution of effects and how a shift in a particular endpoint or effect could result in a net impact that was undesirable. ATS (2000) provided an example of effects on asthmatic individuals; a given population may have lung function values that are not indicative of significant impairment. However, if air pollution shifted the distribution downward—even if no individual experienced clinically relevant consequences—then the population as a whole would have diminished reserve function and therefore could be at greater risk if affected by another risk factor such as an upper respiratory infection. The ATS committee considered such a shift in the distribution of a risk factor to be adverse. Because of this important interpretation, in examining the results of clinical studies it is not only the magnitude of the group mean FEV₁ response that is important, but also the fraction of exposed individuals predicted to experience such a decrement.

Recently, the European Respiratory Society (ERS) and the ATS published a joint statement regarding the adversity of air pollution, which supersedes the prior 2000 ATS statement (Thurston et al. 2016). This new statement updates and broadens the earlier guidance by including effects on other organ systems beyond the respiratory system, such as the cardiovascular and central nervous system effects. The new statement also provides more specific information on biomarkers of effects. In contrast to the previous document, however, the updated ERS/ATS document does not provide substantive guidance for exactly how to interpret research results through the lens of adversity and rather establishes a framework that sets the stage for effects that may be considered adverse in the future. With respect to respiratory effects, the updated guidance includes this statement:

The previous ATS statement addressed the important question of whether small, transient reductions in lung function, as can be seen in susceptible subjects following acute exposure to ozone, should be considered adverse. The document concluded that small transient changes in forced expiratory volume in 1 s (FEV1) alone were not necessarily adverse in healthy individuals, but should be considered adverse when accompanied by symptoms. We support the conclusion that, in otherwise healthy individuals, “a small, transient loss of lung function, by itself, should not automatically be designated as adverse” [46]. However, such small lung function changes should be considered adverse in individuals with extant compromised function, such as that resulting from asthma, even without accompanying respiratory symptoms.

The ATS (2000) and updated statements are thus consistent in their recommendation that in healthy individuals a small reduction in lung function alone is not considered adverse at the individual level. It is a useful exercise to examine the existing panel and clinical study literature in light of the criteria just described for reversible loss of lung function and symptomatology. Overall, of the studies described in the preceding, many only considered changes in lung function or changes in symptoms, not both. From an adversity standpoint, this does not help us determine whether the effects observed would meet ATS criteria. That said, of the seven panel studies that evaluated both endpoints, four showed significant associations with decreased lung function and symptoms, while three reported significant changes in one endpoint but not the other.

We can also consider panel studies that evaluated changes in pulmonary function in terms of the magnitude of these changes. Unfortunately, the majority of studies did not measure FEV₁ but rather PEF; these measures are poorly correlated, especially in certain groups (e.g., Aggarwal, Gupta, and Jindal 2006). However, for illustrative purposes we consider PEF in this context. In the study by Jalaludin et al. (2000), the average decrease in PEF was about 1% for all children and 4% for asthmatic children for a 40-ppb increase in mean daytime ozone concentration; these findings were statistically significant. The authors state that “even for such a large increase in ozone exposure, the decrements in lung function are small and unlikely to be clinically significant.” However, the authors also point out that an increase in ozone concentration equivalent to the IQR was associated with ~20% increase in the prevalence of PEF >20% below the median. Thus, they conclude that in population terms, even this small decrease in average lung function may have adverse effects at the ends of the distribution. Castro et al. (2009) found a significant protective effect of ozone at a 1-day lag; this corresponded to a 0.2 L/min increase in mean PEF and represented a change in PEF of <0.1%. On the other hand, Mortimer et al. (2002) reported a significant decrease in morning PEF of 0.59% per interquartile range (IQR) increase in 5-day average ozone.

In examining the clinical literature, Adams (2002) reported group mean FEV₁ decrements >10% only at 120 ppb ozone, not 40 or 80 ppb; symptoms were noted at 80 and 120 ppb. Adams (2006) did not find any group mean FEV₁ decrements >10% with exposures to 40, 60, or 80 ppb, although increased symptoms were observed at 80 ppb. Schelegle et al. (2009) reported group mean FEV₁ decrements >10% only after exposure to ozone at the highest concentration (88 ppb), but not in response to 63, 72, or 81 ppb. In that study, symptoms were increased after exposure to 72, 81, and 88 ppb. Kim et al. (2011) reported a mean decrease in FEV₁ of 1.7% in response to a 60-ppb ozone exposure; no increase in symptoms was observed. However, as per the ATS (2000) and updated 2016 guidance, it is not simply the magnitude of the change in the group mean that is important, but also the fraction of individuals experiencing a decrement. Thus, of interest in clinical studies is the number of individuals with a decrement in FEV1 of >10%, which is not always reported.

Reversibility of effects

In clinical studies, reversibility can be evaluated by assessing health outcomes at various points post exposure. Of the four studies discussed in the preceding, Schelegle et al. (2009) evaluated lung function and symptoms at time points beyond the 6.6-hr ozone exposure and reported that for all the protocols (at all concentrations), lung function and symptoms had returned to preexposure levels within 4 hr of the end of exposure. Frampton et al. did not find any change in symptoms with ozone exposure; effects on FEV₁ and FVC persisted 22 hr postexposure.

Reversibility has also been evaluated in earlier ozone chamber studies. For example, Horstman et al. (1995) exposed 17 asthmatic and 13 nonasthmatic subjects to 160 ppb ozone (and separately, to filtered air) for 7.6 hr; a light exercise protocol was administered during exposures. On the day of exposure and for 3 days postexposure, asthmatic subjects (but not nonasthmatic subjects) performed spirometry and recorded asthma attacks, symptoms, and medication use in a diary. Symptoms and medication were greater and FEV₁ was lower on the day of ozone exposure as compared to the day of air exposure. For the next 3 days, symptoms remained higher than those following air exposure; medication use was not different. FEV₁ was lower than after air exposure on the first postexposure day, and higher on the second day, while there was no difference on day 3. Overall, Horstman et al. concluded that they found little remarkable evidence of ozone-induced asthma responses during the 3-day postexposure period in their study. In a study with one stated goal to examine recovery after ozone exposure, Adams (2003) found that in a 1.4-hr filtered air recovery period following 80- or 300-ppb ozone exposure for 2 hr at intermittent exercise, the rates of symptom and FEV₁ return to preexposure levels were different and appeared to be related to total ozone dose (product of concentration, ventilation rate, and duration).

Overall, the clinical literature suggests that in healthy individuals and asthmatics, the effects of ozone exposure appear to be reversible. However, it should be noted that clinical studies are only able to evaluate reversibility after acute exposures. It is not clear the extent to which chronic effects of ozone exposure show reversibility, with some epidemiological research suggesting that long-term exposure to ozone results in permanent decrements in lung function (e.g., Rojas-Martinez et al. 2007), and others showing no such association (e.g., Islam et al. 2007). The observational nature of these studies makes determination of a causal effect more challenging, and additional investigation of this issue is warranted.

Adaptation

Evidence from the clinical and epidemiological literature dating back many years, suggests that adaptation, or tolerance, to repeated ozone exposure occurs, demonstrated by attenuation of lung function responses over time. Studies from the 1970s and 1980s first provided indications that adaptation may develop with chronic ozone exposure. Hackney et al. (1977) conducted a small study in which six male volunteers with respiratory hyperreactivity were exposed to 500 ppb ozone, 2 hr/day for 4 consecutive days. One subject showed little response, while 5 showed lung function decrements and increased symptoms on days 1–3, which were largely reversed by day 4. These authors noted that while acute respiratory effects may be prevented with adaptation, other adverse effects of ozone may not be mitigated. Using a similar study design, Folinsbee and Horvath (1986) showed that with daily exposures, healthy adult volunteers showed the most pronounced decreases in lung function on the second day, with almost complete attenuation of responses by days 3–5. Linn et al. (1988) demonstrated that Los Angeles, CA, residents exposed to ozone in the laboratory showed lower sensitivity after the summer months with high ambient ozone concentrations. The following spring, after a winter season with low ozone, responsiveness returned to that of the previous spring. Other studies have shown similar findings (e.g., Bedi, Drechsler-Parks, and Horvath 1985; Horvath, Gleener, and Folinsbee 1981; Linn et al. 1982).

While the studies just described were conducted in healthy subjects, it was of interest to determine whether attenuation similarly occurred in asthmatics. Gong, McManus, and Linn (1997) exposed 10 adults with asthma to 400 ppb ozone for 3 hr/day on 5 consecutive days. Followup ozone exposures were conducted 4 and 7 days after the last consecutive exposure. A moderate exercise protocol was employed. Symptoms and FEV₁ responses were similarly large on the first and second days of consecutive exposure, but subsequent days showed progressive reduction in effects. By the fifth day, responses approached those to filtered air. Four and 7 days later, this adaptation was partially lost. Bronchial reactivity was also evaluated and peaked after the first ozone exposure; it remained somewhat elevated after all subsequent exposures. Overall, Gong et al. concluded that asthmatics develop tolerance in a manner similar to healthy subjects, but the process may be slower and may manifest to a lesser extent.

Despite clear attenuation of pulmonary function responses and symptoms with repeated ozone exposure, some clinical studies have indicated that changes in the inflammatory response may be more nuanced. Devlin et al. (1997) exposed 16 young adult males to 400 ppb ozone for 2 hr/day on 5 consecutive days, and again once 10 or 20 days after the initial exposure. Bronchoalveolar lavage results showed that some cellular and biochemical markers of inflammation and injury were diminished after 5 days of repeated exposure, while others—especially those related to injury (i.e., epithelial cells, lactate dehydrogenase, and total protein)—were not. Similarly, Jorres et al. (2000) exposed 23 healthy subjects to 200 ppb ozone for 4 hr on one occasion, as well as on 4 consecutive days. Results showed attenuation of lung function responses and some, but not all, markers of inflammation and injury; furthermore, mucosal biopsy results suggested morphological changes that did not diminish over time. However, in contrast, Christian et al. (1998) reported attenuation of the inflammatory response to ozone over 4 consecutive days of exposure to 200 ppb, as evidenced by reductions in polymorphonuclear luekocytes (PMNs), fibronectin, and interleukin-6 after repeated exposure versus a 1-day exposure. Toxicological studies also generally report attenuation of inflammatory responses. For example, Van Bree et al. (2002) reported that after 5 consecutive days of exposure to 400 ppb ozone in Wistar rats, the effects observed after a single exposure (marked permeability and inflammatory responses in the lower airway) completely disappeared. Wiester et al. (1995) exposed Fischer 344 rats to a simulated urban ozone profile (60 ppb for 13 hr/day, 7 days/week; peak to 250 ppb over 9 hr, Mondasy–Friday) for 12 or 18 months. Rats were challenged with high ozone concentrations (either 0.5 or 1 ppm) for 1 or 2 hr, and potential adaptation to ozone was assessed by monitoring breathing pattern. Adaptation was found in the rats within 8 hr after the chronic ozone exposure, but not at 4 months postexposure.

Ozone exposure metric and related issues in panel studies

Considerations related to ozone exposure metric can include temporal (averaging time, i.e., 24-hr average, 8-hr maximum, 1-hr maximum), spatial (central site monitor, spatial average based on several monitors, or model-based approach), microenvironmental (i.e., personal, residential indoor or outdoor), and lags. In time series and long-term cohort studies, different metrics have been shown to lead to different results (e.g., Li et al. 2015; Darrow et al. 2011; Smith, Xu, and Switzer 2009; Atkinson et al. 2016). For example, Darrow et al. (2011) examined six different ozone metrics and their associations with emergency department visits for respiratory causes: 8-hr maximum, 1-hr maximum, 24-hr average, commute (7–10 a.m., 4–7 p.m.), daytime (8 a.m.–7 p.m.), and nighttime (12 a.m.–6 a.m.). They found positive, significant associations for the 8-hr maximum, 1-hr maximum, and daytime metrics, nonsignificant associations for 24-hr average and commute metrics, and a negative, significant association for nighttime ozone. The metrics were generally strongly correlated with one another (r = 0.68–0.95), except the nighttime metric, which was uncorrelated with the other five metrics except for the 24-hr average (r = 0.46). The highest correlations were between 8-hr maximum and 1-hr maximum, and between 8-hr maximum and daytime ozone. This study, albeit with a different design from the panel studies reviewed herein, illustrates how the choice of ozone metric can impact results.

Ozone panel studies have employed a variety of temporal metrics, including 24-hr average (Castro et al. 2009; Ward et al. 2002), 8-hr maximum (Delfino et al. 2003), 8-hr maximum moving average (Peacock et al. 2003), 8-hr average (Just et al. 2002; Mortimer et al. 2002), and 1-hr maximum (Dales et al. 2009; Delfino et al. 2003; Escamilla-Nunez et al. 2008; Peacock et al. 2003; Schildcrout et al. 2006). Virtually all the studies reviewed utilized central site monitoring data obtained from public sources, with the exception of Altug et al. (2013; 2014), who conducted passive sampling campaigns outside the primary schools included in the study, and Samoli et al. (2017), who collected personal exposure measurements.

Several panel studies examined different temporal metrics and were thus able to compare results. Delfino et al. (2003) included both 1-hr maximum and 8-hr maximum ozone concentration in analyses and found little difference in results. Peacock et al. (2003) evaluated both 1-hr maximum and 8-hr maximum moving averages and found small changes in effect estimates, with some variation in level of significance. Jalaludin et al. (2000) analyzed 1-hr maximum and daytime average ozone; while no significant associations were reported for 1-hr maximum concentrations and PEF, significant positive associations were found between daytime average ozone and PEF in a subgroup of children with a history of wheeze, doctor-diagnosed asthma, and positive histamine challenge, as well as in all children.

As mentioned, only one study (Samoli et al. 2017) conducted personal monitoring. People spend most of their time indoors, where ozone concentrations are very low in the absence of indoor ozone generation sources or open windows. As a reactive gas, ozone is removed as it passes through the building envelope or enters the HVAC (heating, ventilation, and air conditioning) system. Comparisons of personal exposure to ozone and ambient concentrations show poor correlations (e.g., Geyh et al. 2000; Sarnat et al. 2001), and some epidemiological studies have shown differing results using these metrics (e.g., Delfino et al. 1996). Of the studies reviewed herein, Samoli et al. (2017) used central site measurements and personal monitoring data in analyses and were thus able to compare and contrast results. These investigators found significantly increased odds of stuffy nose using personal measurements, but a nonsignificant effect estimate with central site data. They also found a significant negative association (protective effect) between personal ozone and absenteeism, but a nonsignificant association with central site data.

Lags are another aspect of study design for which there is no common approach. Table 1 shows the lags considered in the panel studies reviewed earlier in this paper. There is wide variability in the number of lags investigated, whether or not a moving average was considered, and how cumulative lags were incorporated into analyses.

Health outcomes in panel studies

The majority of panel studies evaluating lung function have measured PEF, likely because of the availability of simple, low-cost peak flowmeters. Of the 16 studies reviewed earlier in this paper that considered lung function, 13 evaluated PEF and 6 evaluated FEV₁ (several studies measured both outcomes). As mentioned earlier, FEV₁ and PEF do not show good correspondence, even when expressed as percent of predicted values (Giannini et al. 1997; Llewellin et al. 2002; Aggarwal, Gupta, and Jindal 2006). Since clinical guidelines are based on FEV₁ values and not PEF (e.g., Pellegrino et al. 2005; Reddel et al. 2009), it would be more informative to examine this endpoint in panel studies. Even among the studies that evaluated PEF, the specific metric included in epidemiological models differed. For example, Mortimer et al. (2002) examined PEF as daily percent change from the diary-specific median PEFR as well as incidence of ≥10% decline from the diary-specific median PEFR. Peacock et al. (2003) also evaluated PEF data in several ways: odds of a 20% decrement below the median, and as a continuous variable in regression analyses. Schachter et al. (2016) calculated odds ratios (ORs) for having lower percent predicted PEF (or FEV₁) in the morning. Ward et al. (2002) analyzed PEF as deviations from an individual’s mean for morning and afternoon separately.

Consistency in the assessment of symptomatology is similarly problematic in panel studies. Though standardized instruments are available—for example, the International Study of Allergy and Asthma in Childhood (ISAAC) questionnaires (Ellwood et al. 2000; ISAAC 1998)—they have not often been utilized in the air pollution setting. Not only are the specific symptoms evaluated highly variable, but the way in which the symptoms are included in epidemiological models differs from study to study. Of the 14 papers reviewed that evaluated symptoms earlier in this article, the number of symptoms evaluated ranged from one to seven, and the approaches included:

• Overall symptom severity: Individual symptoms are not evaluated but rather questions are posed about the severity of symptoms in general (or asthma episodes). For example, Schildcrout et al. (2006) asked children to record the severity of their symptoms at the end of each day. Severity was coded as follows: 0—no asthma symptoms; 1—one to three mild asthma episodes, each lasting 2 hr or less; 2—four or more mild asthma episodes or one or more episodes that temporarily interfered with activity, play, school, or sleep; and 3—one or more asthma episodes lasting longer than 2 hr or resulting in shortening of normal activity, seeing a doctor for acute care, or going to a hospital for acute care. The symptom variable was then dichotomized for analysis (any symptom vs. no symptoms). Other investigators have used different scales (and evaluated different symptoms); for example, Delfino et al. (2003) had participants report the daily combined severity of asthma symptoms (cough, wheeze, sputum production, shortness of breath, and chest tightness) using a 6-level scale: 0: no asthma symptoms present; 1: asthma symptoms present but caused no discomfort; 2: asthma symptoms caused discomfort but did not interfere with daily activities or sleep; 3: asthma symptoms interfered somewhat with daily activities or sleep; 4: asthma symptoms interfered with most activities and may have required that the participants stay home in bed, return home early from school or work, or call a doctor or nurse for advice; 5: asthma symptoms required going to a hospital, emergency room, or outpatient clinic. The analysis was based on two dichotomous outcome variables with different cutoff points across the symptom score: (a) no symptoms or symptoms not bothersome (score 0 or 1) versus bothersome or more severe symptoms (symptom scores > 1), and (b) none-to-bothersome symptoms but no interference with daily activities (score 0–2) versus symptoms that interfered with daily activities (symptom scores > 2). Karakatsani et al. (2012) asked subjects to grade shortness of breath, wheeze, and cough as absent (0), slight (1), or moderate/severe (2), and then conducted the analyses with dichotomized variables (0 vs. 1 + 2), as both incident and prevalent symptoms.

• Presence of individual symptoms: Ward et al. (2002) asked questions about the presence or absence of each symptom, that is, “Did you cough today?,” “Were you ill today?,” “Were you short of breath today?,” “Did you wake up last night with a cough or wheeze?,” and “Did you wheeze today?” The daily proportion of subjects reporting each symptom (percent prevalent symptoms) or reporting a new episode of each symptom (percent incident symptoms) was included in the analysis. Altug et al. (2013) utilized the ISAAC questionnaire (ISAAC 1998) and focused their analysis on current wheeze (“Has your child had wheezing or whistling in the chest in the past 12 months?”), while Altug et al. (2014) asked about more recent wheeze (attack of shortness of breath or wheeze in the last 7 days; attack of shortness of breath or wheeze today). Just et al. (2002) had parents record the presence or absence of asthma attacks, and the severity of nocturnal cough, wheeze, and symptoms of irritation on a 3-point scale (0 = none, 1 = moderate, 2 = severe). However, the severity of symptoms was not used in the data analysis; rather, the presence of symptoms on a given day was used to create “incident” and “prevalent” variables for each of the outcomes, similar to Ward et al. (2002). Escamilla-Nunez et al. (2008) also employed an incident symptom approach, including cough and wheeze in separate models. Yet another approach was employed by Mortimer et al. (2002), who simply evaluated the presence of cough, chest tightness, and wheeze and used “any incident symptom” in analyses.

• Calculated symptom index: Some studies collected information on the severity of individual symptoms and then combined them into a total score. For example, Schachter et al. (2016) had children record daily cough and wheeze symptoms by severity (none = 0, mild = 1, moderate = 2, severe = 3). Daily total symptom scores were calculated by adding the severity scores of both cough and wheeze.

It is difficult to compare and contrast results among studies that have employed such dramatically different approaches to both evaluating and analyzing symptom data, and some degree of standardization of approaches is recommended.

Co-pollutants in panel studies

Most panel studies examined a number of pollutants in addition to ozone; most commonly these included a measure of particulate matter (PM)—either PM₁₀ or PM_2.5—along with NO₂ and SO₂. Because people are not exposed to only one pollutant in isolation, single-pollutant models may not be the most accurate reflection of potential health impacts due to a complex mixture. Thus, there has been a move toward multipollutant modeling in recent years (e.g., Snowden, Reid, and Tager 2015; Tolbert et al. 2007). Despite the fact that most air pollution panel studies include measurements of multiple pollutants, only about half of the studies reviewed earlier in this article conducted multipollutant modeling. In a number of studies, including (an)other pollutant(s) did not appreciably alter effect estimates for ozone (e.g., Chan and Wu 2005, [PM₁₀, NO₂]; Dales et al. 2009, [PM_2.5, NO₂, SO₂]; Escamilla-Nunez et al. 2008, [PM_2.5, NO₂]; Jalaludin et al. 2000 and 2004, [PM₁₀, NO₂]; Samoli et al. 2017, [PM₁₀]). However, in some cases results did change. For example, Just et al. (2002) ran multipollutant models including black smoke (BS) and NO₂ with ozone, and found that ozone was not significantly associated with symptoms in a model with these two pollutants and an interaction term between O₃ and temperature, whereas ozone was significant in a single-pollutant model. Lewis et al. (2005) found several additional significant associations (at different lags) with FEV1 when PM_2.5 and PM₁₀ were included separately in multipollutant models. Mortimer et al. (2002) conducted multipollutant models for morning symptoms in association with ozone plus SO₂, NO₂, and PM₁₀. While ozone was significant in a single-pollutant model, the effect estimate became nonsignificant with the addition of SO₂ into the model. There was little change in the ozone effect estimate with the addition of NO₂, or the addition of both NO₂ and SO₂. There was also little change when PM₁₀ or PM₁₀ + NO₂ + SO₂ was added to the model. In general, the strongest associations in single-pollutant models were observed for NO₂, followed by PM₁₀, SO₂, and then ozone; however, it should be noted that the lag structures were different for each pollutant and the number of cities also varied due to availability of monitoring data, making comparisons across pollutants difficult. Finally, Yoda et al. (2014) found that the association for EBC pH significantly decreased after adjusting for NO₂ or suspended PM.

It should be noted that multipollutant models have their own issues with interpretation (Bergen et al. 2016; Dionisio, Baxter, and Chang 2014). One issue is that of exposure error, for which there is a large literature related to air pollution epidemiology studies. When measuring a pollutant at a central monitoring location, exposure error can arise from spatial variability. Pollutants that are locally generated, such as NO₂ and SO₂, are more spatially heterogeneous than secondary pollutants such as ozone that are more regional in nature. This can result in greater exposure error for these more spatially variable pollutants. Exposure error also results from the use of central site monitoring data to reflect individual-level exposures. In this case, pollutants that tend to infiltrate more efficiently into homes and other buildings will be associated with less exposure error than pollutants whose concentrations are significantly attenuated upon being transported indoors. As discussed earlier, personal ozone exposures tend to be poorly correlated with ambient measures. Correlations between pollutants also need to be considered; for example, secondary PM and ozone tend to be highly correlated, at least in the summer months, because they arise from similar photochemistry.

Influence of temperature in panel studies

Panel studies of ozone and respiratory health have applied a variety of approaches for controlling or accounting for temperature. Most studies controlled for ambient temperature as a confounder but did not directly evaluate the effect of temperature on lung function and symptoms. For example, Castro et al. (2009) included minimum, mean, and maximum ambient temperatures in models separately as confounders, while Dales et al. (2009), Peacock et al. (2003), Karakatsani et al. (2012), and Altug et al. (2014) included daily mean temperature. Delfino et al. (2003) included 1-hr maximum temperature in models, but this variable did not have a significant impact on effect estimates. Minimum temperature was considered by Escamilla-Nunez et al. (2008) and Ward et al. (2002). Jalaludin et al. (2000) included mean daily temperature, and as expected noted a positive correlation between ozone and temperature (r = 0.64; p = 0.0001). Just et al. (2002) included mean daily temperature and reported a Pearson correlation coefficient of 0.65 (p < 0.001). Just et al. also investigated various temperature and humidity lags; in the final model the meteorological variables with the lag showing the strongest association with health outcomes were included (mostly lag 0, 1, or 2). Mortimer et al. (2002) included 12-hr average wet-bulb temperature in the model and found that inclusion increased the magnitude of the ozone effect estimate by >30%, suggesting it was a strong confounder. Schachter et al. (2016) included mean daily temperature and evaluated this variable as a potential confounder by including it as a single predictor in the model. In that analysis, temperature was not shown to be a strong confounder as evidenced by an odds ratio of 0.99 [0.96–1.02]; p = 0.5525 per 1 degree increase in temperature. Schildcrout et al. (2006) included maximum hourly temperature in models.

In contrast to other studies that only included temperature as a possible confounder, Just et al. (2002) evaluated possible interactions between temperature and ozone. Statistically significant interactions between these variables for asthma attacks and lung function decrements were observed, and the authors concluded that ozone and temperature have a synergistic effect on these outcomes. Overall, the panel studies reviewed have employed widely varying approaches to evaluating confounding by temperature, as evidenced by the use of different measures. From a comparability perspective, having consistency in study design and approach would be valuable. To this end, investigators could conduct sensitivity analyses with a range of approaches for temperature control and include results in publications. Another common approach could be to include ozone–temperature interaction terms in models, although it should be noted that insufficient statistical power may be an issue with smaller studies.

As noted in the preceding, in some panel studies temperature was shown to be a strong confounder, while in others it was not. Indeed, even in the time-series literature where temperature/air pollution effects have been better studied, there are conflicting findings about the role of temperature in asthma exacerbation. While some studies report increased asthma risk at high temperatures (e.g., Chen et al. 2016), others report lower risk at high temperatures (e.g., Vaneckova and Bambrick 2013), increased risk at lower temperatures (e.g., Guo et al. 2012), or increased risks at both extremes (e.g., Lam et al. 2016). The question of temperature adjustment in epidemiological studies was recently discussed by Atkinson et al. (2016) in the context of a long-term ozone mortality study. In that study, both warm-season and. all-year ozone data were used in analyses, and associations were generally stronger when using the former. Atkinson et al. noted that ozone production during summer episodes is closely related to temperature and solar radiation and therefore that it was not clear whether ozone associations observed during the warmer months are independent of the effects of temperature. In another long-term study by Jerrett et al. (2009), no evidence of confounding by temperature was observed, but a modifying effect of temperature was reported on the ozone effect estimates. There is very little information available in the panel study literature regarding possible interactions between ozone and temperature in adverse respiratory effects. Some panel studies have examined this issue (e.g., Just et al. 2002), but clearly further research is needed to evaluate the effects of temperature in epidemiological studies, as well as the possible interactive effects between air pollutants and temperature on asthma exacerbation.

Multiple comparisons

It should be noted that almost all the panel studies considered multiple pollutants. For example, Ward et al. (2002) examined 13 different pollutants, 7 different health endpoints, and 5 different lags, for a total of 455 statistical tests. At a p value of 0.05 we would expect about 23 significant findings to be due to chance alone. That said, investigators usually try to interpret their findings holistically, such that more weight is given to findings that are consistent with each other (e.g., across lags, health endpoints representing a similar biological mechanism, or pollutants that may be associated with the same source), and less weight is given to extraneous significant results. Such a careful analysis is not universally conducted, however.

This issue of multiple comparisons could be mitigated by the use of a formal correction technique; however, few if any air pollution epidemiology studies have employed such a method. The Bonferroni method may be overly conservative (Bender and Lange 1999; Perneger 1998), but other techniques could be considered, such as control of the false discovery rate (e.g., Jones, Ohlssen, and Spiegelhalter 2008).

The clinical studies generally have fewer statistical tests, but multiple comparisons remain an issue. Kim et al. (2011) specifically did not include data collected at individual time points during the 6.6-hr exposure in the analysis—and only evaluated outcomes pre- and postexposure—in an effort to eliminate the need for correction for multiple comparisons. While this increased the statistical power of the study, it also resulted in the omission of valuable data from the primary published analysis (the intermediary data were later used by McDonnell et al. [2012] in a modeling effort). The Adams (2006) data, which showed a nonsignificant change in lung function at 60 ppb ozone, were later reanalyzed by Brown, Bateson, and McDonnell (2008) using a statistical approach that also omitted the intermediary time points; in other words, the data were reanalyzed as though the study only collected pre- and postexposure measurements. Brown et al. found the same average magnitude of FEV change; however, the finding was statistically significant. Brown et al. discuss the reasons why their findings differ so dramatically from those of Adams (2006), including the method used to control for multiple comparisons (as well as the number of comparisons to make), illustrating the critical importance of statistical approach in analyzing human clinical study data.

Conclusion

Ozone is one of the most well-studied air pollutants, with initial speculation about health effects dating to the mid-nineteenth century (e.g., Leading Articles 1853). A century later, more focused reviews and studies were being conducted (e.g., Elford and Van Den Ende 1942; Stokinger, Wagner, and Wright 1956). The first NAAQS for ozone was promulgated in 1971 and was based on total photochemical oxidants; the 1-hr standard was set at 80 ppb, not to be exceeded more than 1 hr per year. Subsequent NAAQS were based on ozone as the indicator, and in 1997 the averaging time was increased to 8 hr. The level has been variable, with a 1-hr maximum of 120 ppb in 1979, decreasing to 8-hr averages of 80 ppb (1997), 75 ppb (2008), and 70 ppb (2015). At the time of writing this paper, the ozone NAAQS is just beginning the next review process.

While the extensive study of ozone over decades has resulted in comprehensive understanding of its health effects, some unique characteristics and behavior of this pollutant result in some remaining uncertainties. As noted in the preceding and in Table 1, panel study results are inconsistent; in contrast, clinical studies generally present a clearer picture One possible explanation for these divergent results is that the complex pollutant mixture to which panel study subjects are exposed differs substantially both temporally and spatially and contributes to the differential impact of ozone observed in these studies. Ozone is a reactive gas that oxidizes volatile organic compounds (e.g., terpenes) to secondary products that may have enhanced health impacts compared with their precursors (for a review of this topic see Rohr, 2013). Ambient concentrations of ozone thus may reflect ozone that has already reacted with other atmospheric components, as well as unreacted ozone. The secondary reaction products present at any given point in space and time will depend on the initial mixture, and the presence (and concentration) of ozone may be an indicator that such secondary reactions have occurred. This is not to suggest that ozone itself does not have respiratory effects—that has been clearly established—but that reaction products may contribute to observed health effects in panel studies and at least in part account for the wide variability in results.

Exposure issues are also particularly important for ozone. Because of its reactivity, it is largely removed upon infiltration through the building envelope, and as a result indoor concentrations are very low in the absence of ozone-generating equipment such as certain kinds of air purifiers. Occupant behaviors such as opening windows and the use of air conditioning may modify indoor ozone exposures. As a result, correlations between personal and central site ozone measurements are very low. Of the panel studies reviewed herein, only Samoli et al. (2017) conducted personal monitoring. These investigators found significantly increased odds of stuffy nose using personal measurements, but a nonsignificant effect estimate with central site data. They also found a significant negative association (protective effect) between personal ozone and absenteeism, but a nonsignificant association with central site data. Because of the large degree of exposure error introduced due to population time–activity patterns and the fraction of time an individual spends indoors, personal measurements of ozone are the gold standard and if possible should be employed in panel studies; however, cost may present a challenge in this regard.

As discussed in detail in the preceding, variability in key panel study design characteristics makes comparisons among studies, as well as meta-analyses, difficult. Variables including ozone exposure metrics, lags, health outcomes, and control for temperature are not at all standardized. Some of this is dependent upon availability of equipment for lung function assessment. Spirometers that measure FEV₁ are more expensive than the small personal peak flowmeters often used in panel studies; however, FEV₁ is the preferable metric. For symptom evaluation, a more standardized questionnaire or approach would be helpful in increasing interstudy comparability. A priori selection of variables should always be conducted in order to avoid any tendency to “cherry pick” data and thus lead to bias. In panel studies that consider multiple pollutants—virtually all published studies—multipollutant modeling should be undertaken; however, caution must be used in the interpretation of the results of these models due to differential exposure error among pollutants and variable correlations between pollutants.

Finally, clinical studies are an important tool for evaluating ozone-related pulmonary effects in a controlled setting. Such studies eliminate the exposure error inherent in panel studies due to spatial variability in ozone concentrations, lack of correlation between central site and personal measurements, and spatiotemporal differences in the oxidant mixture as a whole. However, related to the latter consideration, they are also artificial in that real-world air pollution exposures are not to ozone in isolation. With the NAAQS for ozone at 70 ppb, there is a critical need to understand responses at or near this concentration. Only four clinical studies have conducted exposures at concentrations below 80 ppb; additional studies would be informative. Such studies should also include exposures to sensitive subpopulations such as asthmatics. Importantly, as with panel studies, a priori analytical plans, including control for multiple comparisons, should be laid out to avoid post hoc testing that increases the number of statistical comparisons being made in a given data set.

Additional information

Notes on contributors

Annette C. Rohr

Annette C. Rohr is a Principal Project Manager in the Energy and Environment sector at EPRI, where she conducts research on the health effects of air pollution and other environmental exposures.