Does ozone inhalation cause adverse metabolic effects in humans? A systematic review

Abstract

We utilized a practical, transparent approach for systematically reviewing a chemical-specific evidence base. This approach was used for a case study of ozone inhalation exposure and adverse metabolic effects (overweight/obesity, Type 1 diabetes [T1D], Type 2 diabetes [T2D], and metabolic syndrome). We followed the basic principles of systematic review. Studies were defined as “Suitable” or “Supplemental.” The evidence for Suitable studies was characterized as strong or weak. An overall causality judgment for each outcome was then determined as either causal, suggestive, insufficient, or not likely. Fifteen epidemiologic and 33 toxicologic studies were Suitable for evidence synthesis. The strength of the human evidence was weak for all outcomes. The toxicologic evidence was weak for all outcomes except two: body weight, and impaired glucose tolerance/homeostasis and fasting/baseline hyperglycemia. The combined epidemiologic and toxicologic evidence was categorized as weak for overweight/obesity, T1D, and metabolic syndrome,. The association between ozone exposure and T2D was determined to be insufficient or suggestive. The streamlined approach described in this paper is transparent and focuses on key elements. As systematic review guidelines are becoming increasingly complex, it is worth exploring the extent to which related health outcomes should be combined or kept distinct, and the merits of focusing on critical elements to select studies suitable for causal inference. We recommend that systematic review results be used to target discussions around specific research needs for advancing causal determinations.

Keywords:

• Systematic review
• ozone
• metabolic syndrome
• diabetes
• overweight
• obesity

1. Introduction

The Clean Air Act requires the US Environmental Protection Agency (EPA; Agency) to set National Ambient Air Quality Standards (NAAQS) for six air pollutants, including ozone. The standards must be reviewed every five years to ensure that they protect human health and the environment with a sufficient margin of safety. As part of the review process, EPA assesses causal relationships between air pollutant exposures and human health effects using a framework developed specifically for this purpose. The completed reviews are referred to as Integrated Science Assessments (ISAs) (https://www.epa.gov/isa/learn-about-isas).

In the 2013 ISA for ozone (US EPA 2013), EPA conducted a review of the scientific literature on ozone and various health outcomes. EPA did not evaluate metabolic outcomes as an independent endpoint, but rather considered metabolic effects in the context of a mode of action (MOA) for cardiovascular effects (US EPA 2013). In contrast, in the most recent NAAQS review cycle for ozone, EPA (2020a) evaluated metabolic effects as an independent endpoint. The 2020 ozone ISA concluded that there is a likely causal relationship between short-term ozone exposure (e.g. hourly, daily, weekly) and metabolic outcomes and a suggestive, but not sufficient-to-infer, causal relationship for long-term ozone exposure (e.g. months or years) and metabolic outcomes. These conclusions were based on assessments of overall metabolic effects rather than on individual specific outcomes (e.g. diabetes, obesity, metabolic syndrome). Other than the EPA ISA, we are aware of only one other review that explored associations between ozone exposure and metabolic effects (Shore 2019), but this study was a narrative review with a primary focus on a few toxicologic studies. Additional research on ozone exposure and metabolic outcomes has been published since the time of the EPA assessment and the Shore review and it is timely to consider the enhanced body of literature in the context of a revised review.

While the systematic review framework used by EPA in the ISA offers one method for assessing causality for criteria air pollutants and health effects, new approaches for systematic review have been developed. These concepts are worth exploring in the context of ozone and metabolic outcomes as well as for other chemical/outcome pairs. For example, Goodman et al. (2020) proposed modifications to the EPA framework, including a more comprehensive set of criteria for assessing study quality, detailed methods for integrating evidence that consider individual study quality and relevance, and methods for assessing the coherence of results across studies both within and across scientific disciplines. Overall, within and outside of the Agency, processes for conducting systematic reviews are numerous, varied, and evolving (e.g. Guyatt et al. 2008; Moher et al. 2009; Money et al. 2013; US EPA 2018, 2020b; Morgan et al. 2019; NTP 2019; Mengist et al. 2020; WHO 2020).

The approaches differ in terms of the type of evaluative elements as well as in style (e.g. narrative, scoring systems, heat maps). In addition, the number of elements to evaluate vary widely from concise to expansive (Wells et al. 2013; Goodman et al. 2014; US EPA 2018; Morgan et al. 2019; Goodman et al. 2020). In fact, there is no single established “best” approach for conducting systematic reviews within environmental disciplines.

There is, however, general agreement that important features of any systematic review should include a comprehensive literature search and a structured, transparent evaluation of individual study quality (Shea et al. 2007, 2017). These processes are resource-intensive; concerns that the systematic review process has become unduly onerous have been expressed (Wallace et al. 2013; Westgate and Lindenmayer 2017; Collins et al. 2019; National Academies of Sciences Engineering and Medicine 2021). An efficient approach would be especially useful in situations where resources are limited or there are recurring regulatory deadlines. Savitz et al. (2019) suggested that assessments should be focused on identifying the sources of bias most likely to be influential, categorizing each study according to how well it addressed each potential bias, and determining whether results differ across studies in relation to susceptibility to each examined source of bias. This reduces what the authors refer to as “procedural repeatability” and could in fact render the review more transparent and reproducible. While there is no “correct” list of review elements applicable to all exposures and outcomes, inclusion of a large, all-encompassing set of review elements is not necessarily the best approach to systematic review, especially if the elements are not used to support assessments of causality. In this review, we use a practical, transparent approach for systematically reviewing a chemical-specific evidence base.

The goal of this systematic review is to address the question: does ozone inhalation exposure cause adverse metabolic effects (overweight/obesity, Type 1 diabetes [T1D], Type 2 diabetes [T2D], and metabolic syndrome) in humans?

2. Methods

In this review, we used a six-step process (Figure 1) based on the basic features and principles of systematic reviews (Shea et al. 2017; NTP 2019; US EPA 2020b). Each step is discussed in detail below.

Figure 1. Steps in this review of ozone and metabolic outcomes.

2.1. PECO formulation

An essential feature of a systematic review is the identification and grouping of studies according to a clearly stated research question (Rooney et al. 2014). This approach to grouping studies enables a structured assessment of inter-study consistency (LaKind et al. 2017). We therefore developed a distinct PECO question for each risk factor/metabolic outcome pair, rather than combining outcomes. We organized our framework to have different PECO questions for human studies (apical outcomes) and for in vivo/in vitro animal studies (related non-apical metabolic outcomes) (Table 1). Further, there is a distinct PECO question for each apical outcome: overweight/obesity, T1D, T2D, and metabolic syndrome. Metabolic syndrome is considered to be a constellation of interrelated risk factors for cardiovascular disease and T2D (Tune et al. 2017). A recent consensus definition for metabolic syndrome requires at least three of the five following risk factors to be present: elevated waist circumference, elevated triglycerides (TG), reduced high density lipoprotein cholesterol (HDL), elevated blood pressure, and/or elevated fasting glucose (e.g. Alberti et al. 2009).

Table 1. PECO formulation addressing: does ozone inhalation exposure cause overweight/obesity, T1D, T2D, or metabolic syndrome?

Human studies
Population	Humans birth to adult
Exposure	Ozone via inhalation
Comparator	Comparison between different categories of ozone concentration (i.e. dose) Comparison across continuous ozone concentrations
Outcome(s)	Overweight and obesity Type 1 diabetes (T1D) Type 2 diabetes (T2D), including measures of glucose and insulin homeostasis Metabolic syndrome. Metabolic syndrome is the constellation of interrelated risk factors for cardiovascular disease (CVD) and T2D (Tune et al. 2017). A joint statement of several expert organizations describes a diagnosis of metabolic syndrome as requiring three of the five risk factors: elevated waist circumference, elevated triglycerides [TG], reduced high density lipoprotein cholesterol [HDL-C], elevated blood pressure [BP], and elevated fasting glucose [FG]) (Alberti et al. 2009). Thus, studies that examined at least three of these five risk factors are included in this review.
In vivo animal studies
Population	Mammalian laboratory animals
Exposure	Ozone via inhalation
Comparator	Comparison across ozone-exposed groups and to air-exposed groups
Outcome(s)	Increases in body weight (adiposity, elevated body weight, obesity, body mass index (BMI), changes in eating patterns, thyroid effects/changes in thyroid hormones/body temperature) Glucose-related effects (glucose intolerance, insulin resistance, fasting hyperglycemia, changes in hepatic gluconeogenesis, serum/plasma hormone analytes) Dyslipidemia (changes in plasma cholesterol, triglycerides [TG], HDL [high density lipoproteins], LDL [low density lipoproteins])
In vitro studies
Population	Mammalian cells (primary cells or cell lines)
Exposure	Ozone via inhalation mimicking air-liquid interface; internal ozone dose
Comparator	Comparison across ozone-exposed groups and to air-exposed groups
Outcome(s)	Glucose-related effects (changes in glucose homeostasis) Dyslipidemia (changes in cholesterol, TG, HDL, LDL)

2.2. Identification and selection of studies

PubMed and Web of Science were used to conduct literature searches for human, animal, and in vitro studies. In addition, we examined the bibliographies of retrieved articles to identify publications not previously captured by the electronic searches. We also searched the bibliography of the 2020 Ozone ISA (US EPA 2020a) for any additional publications that may not have been identified by previous searches.

Publications identified by the literature searches were analyzed for duplicates and then screened by title and/or abstract according to the inclusion/exclusion criteria for either human or in vivo/in vitro studies (see below). The remaining publications then underwent a full-text screen to identify the studies for final inclusion. All screenings were conducted independently by at least two authors; differences in screening results were resolved by consensus. PRISMA diagrams were used to track the literature search and inclusion/exclusion counts. Complete MeSH terms and lists of the citations considered for inclusion along with reasons for exclusion are provided in the Supplementary Material.

2.2.1. Human studies

We identified publications evaluating the metabolic health effects shown in Table 1 associated with short- and long-term ozone exposures in humans. We further conducted a search for literature from January 2018 to May 2020 to capture recent articles that used “air pollution” along with “ozone” as a key word (Supplementary Material).

PECO questions related to human studies included overweight/obesity, T1D, T2D, and metabolic syndrome. Research on mortality associated with these diseases was excluded as mortality may be due to other related factors, such as cardiovascular disease. In addition, mortality data have known reliability issues (McGivern et al. 2017; Falci et al. 2018). For metabolic syndrome, individual components of the definition, such as blood pressure, are often used as diagnostic criteria for other apical outcomes; a diagnosis of metabolic syndrome requires at least three of the five elements noted above and therefore we only included studies that incorporated at least three of these diagnostic criteria.

The inclusion criteria were:

1. The study was conducted with humans (epidemiology or experimental study).

2. Exposure was determined by measurements or models of ozone concentrations.

3. Comparison groups were included in the study (e.g. controls in experimental study, lower ozone levels in comparator population).

4. Outcomes included diagnoses of T1D, T2D, and/or obesity or at least three of five metabolic syndrome elements (elevated waist circumference, elevated triglycerides (TG), reduced high density lipoprotein cholesterol (HDL), elevated blood pressure, and/or elevated fasting glucose).

5. The publication appeared in English prior to 25 May 2020 (end date of the literature search).

Publications were excluded for the following reasons:

1. The design was a case study, with no controls.

2. The publication was a review, editorial, commentary, or guideline, with no de novo data.

3. The study was only in vitro (e.g. cells and tissues).

4. Ozone was not used as the explanatory variable.

5. The outcome was a biochemical change only (e.g. fatty acid and amino acid metabolism, oxidative stress, C-reactive protein), with the change not used to diagnose disease.

6. Gestational diabetes or mortality of any outcome.

7. Studies that looked at single factors for metabolic syndrome (e.g. blood pressure) rather than at least three of the five factors that constitute the metabolic syndrome definition.

2.2.2. In vivo/in vitro animal studies

We identified toxicologic studies that investigated ozone exposure and the metabolic effects shown in Table 1. MeSH terms for these searches are provided in the Supplementary material. In addition, to cast the search more broadly, we conducted searches of the literature from January 2018 to 24 June 2020 using “air pollution” and “ozone” as key words.

The inclusion criteria were:

1. The study was conducted using a standard mammalian animal model or an in vitro mammalian cell model of ozone exposure.

2. Exposure to ozone in animals was through inhalation or, for in vitro studies, through an inhalation-mimicking (air-liquid interface/internal ozone dose) system.

3. Animal models were standard (wild-type) strains.

4. The study outcomes included metabolic effects, such as those related to glucose or insulin homeostasis (glucose intolerance, insulin resistance, fasting hyperglycemia, changes to hepatic gluconeogenesis, serum/plasma hormone analytes), dyslipidemia (changes in plasma cholesterol, TG, high-density lipoprotein [HDL], or low-density lipoprotein [LDL]), and body weight (adiposity, elevated body weight, obesity, body mass index [BMI], changes in eating patterns, thyroid effects/changes in thyroid hormones/body temperature).

5. The study design utilized a control (air-exposed) animal group or in vitro control exposure experimental paradigm.

6. The publication appeared in English prior to 24 June 2020 (the end date of the literature search).

Publications were excluded for the following reasons:

1. The publication was an abstract only, review, editorial, commentary, or guideline.

2. The study was conducted in humans, a non-mammalian species (e.g. birds), or plants.

3. The study used only susceptible animal models (e.g. obese mice) and/or special conditions (e.g. running wheel and special diet).

4. The study included ozone data but only as part of an exposure mixture (e.g. PM2.5 plus ozone).

5. The study outcomes of interest were not related to the metabolic effects or PECO questions.

2.3. Data extraction

The following data were extracted from each epidemiology publication: cohort name or study population, age group, country, sample size, exposure characterization method, exposure duration (short-term [ST], <30 days exposure; long-term [LT], 1 month and longer), exposure level, outcome(s), funding source, model covariates, and effect estimates. For publications that reported results for both a primary model and stratified models, results from the primary model were extracted. For approaches addressing covariate interactions, we included the result from the main model but noted whether any of the results from the interaction models differed in terms of statistical significance or direction.

The following data were extracted from each toxicology publication: animal model, species/strain, age, sample size, exposure method, exposure level, exposure duration and frequency, ozone generation and monitoring methodology, control exposure, PECO outcomes measured, statistical methods applied, funding source, and outcome results. The outcome results of suitable studies were tabulated by denoting directionality (compared with controls) by using up or down arrows, with underlining denoting a statistically identified change.

2.4. Assessment of study suitability for evidence synthesis

There are many published examples of methods for conducting a quality review of research (Klimisch et al. 1997; Schneider et al. 2009; Woodruff and Sutton 2014; NTP 2019). Some methods include numerous evaluative criteria. Comprehensive inclusion of numerous criteria does not necessarily increase the overall confidence in the review, especially when the quality results are not carried forward into evidence synthesis. Furthermore, quality elements that do not have clear cut definitions, but rather lean heavily on professional judgment, may reduce the review’s transparency and reproducibility. For these reasons, rather than enumerate quality elements this review included a focused set of elements that are generally considered to be essential to study suitability for evidence synthesis. Further, only those studies meeting the element(s) were included in evidence synthesis and integration.

Thus, based on the results of the assessment, studies were judged as either being “Suitable” for the subsequent step of evidence synthesis (examining the body of epidemiology literature or toxicology/mechanistic literature for a specific PECO question) or “Supplemental.” Supplemental studies were not excluded entirely. Results from these studies were examined to ascertain the concordance of results with Suitable studies but were not formally carried forward to the evidence synthesis step. This suitability review was conducted by at least two authors for each publication.

For the epidemiologic studies, evidence of exposure preceding the effect (i.e. temporality) was identified as the single criterion for use in determining whether a study was Suitable for evidence synthesis. Temporality is considered an indisputable property of a study when considering a causal link (Potischman and Weed 1999; Kundi 2007; Adami et al. 2011; Goodman et al. 2014), and can also be consistently and transparently applied across studies. Studies with designs that could not ensure that the outcome ascertainment followed the exposure (e.g. case-control studies with prevalent cases; cross-sectional studies) were categorized as Supplemental. Cohort studies that did not collect baseline information as part of the longitudinal design were also categorized as Supplemental.

For the toxicologic in vivo/in vitro animal studies, five elements were used to distinguish Suitable from Supplemental studies: (i) The study used standard (wild-type) animal strain(s); provided a normal diet; animals were not bred for any specific susceptibility or subjected to any special conditions; (ii) Exposure groups consisted of at least five animals; if not, an explanation or power calculation was required; (iii) The ozone generation and monitoring equipment were fully described; (iv) The study included room air/filtered air-exposed concurrent controls; and (v) The study used standard methods considered reliable and, if statistical results were reported, the associated methods were described. Studies that met all five criteria were considered Suitable for the evidence synthesis step. Studies for which one or more criterion was not met were categorized as “Supplemental.” Studies were excluded entirely (i.e. not considered as Supplemental information) if they were determined to have such critical deficiencies in either study details or design that an accurate review of their findings could not be achieved (this is described in greater detail in the Results section).

2.5. Evidence synthesis/confidence characterization

The Suitable studies formed the basis for characterizing confidence (“strong” vs. “weak”) in the evidence for each outcome from each discipline (in vivo/in vitro animal or human studies). The evidence synthesis/confidence characterization step was carried out by at least two authors. Differences were discussed until consensus was achieved.

To determine whether the evidence was “strong” or “weak,” we employed a modified definition of existing tools for robust, strong bodies of evidence (Guyatt et al. 2008; Deener et al. 2018; Goodman et al. 2020; US EPA 2020b; US Preventive Services Taskforce 2021). In general, a body of evidence for a given outcome was categorized as strong if it consisted of at least three studies (Treadwell et al. 2006; Goodman et al. 2010) and exhibited consistency of results across similar exposures and evidence of a monotonic dose-response relationship, based on previous ozone-induced toxicity data (Nishikawa et al. 1990; Holland et al. 2015).

For the human studies, beyond the number of studies, we also considered study components that contributed to the overall confidence assessment. Weak bodies of evidence were assigned if the results were inconsistent based upon the directionality of results with no explanatory reasons; there was limited or inadequate control of confounding factors and co-exposures; and evaluation of exposure was semi-ecologic or with no accounting for spatial or temporal variability. We did not formally evaluate the strength of associations. While evaluation of a body of evidence by comparing strength of associations across publications is desirable, risk estimates were not always reported by the authors, and beta values were often based on different units, making direct comparison across studies impossible. The issue of lack of comparability of outcomes and its effect on weight of evidence assessment has been discussed in depth by Goodman et al. (2010). For the in vivo/in vitro animal studies, strong evidence also demonstrated an increasing effect with increasing exposure duration/frequency, considering both exposure hours and repeated exposures (e.g. episodic exposures over many weeks). The evidence was considered weak if two or fewer studies were available, if the results across studies were not consistent (e.g. no discernible pattern compared with control measurements), or if no clear exposure-response relationship could be discerned (e.g. either across exposure levels or across exposure duration/frequency) to support the hypothesis that ozone induced the outcome in question. The experimental animal evidence could include several experimental outcomes; thus, the animal streams of evidence could include a series of mixed confidence characterizations.

It should be noted that information from animal and mechanistic studies (including any in vitro data) was combined in this review into a single stream of experimental animal evidence. Regarding the role of mode of action (MOA), one approach used in reviews is to begin by developing a hypothetical MOA for the outcomes of interest. In this review, however, we first assessed the experimental animal evidence stream to determine whether it supported a causal role for ozone for any of the outcomes of interest, and then, if so, considered what the MOA might be for such outcome(s).

2.6. Evidence integration

Evidence integration has been defined as making “judgments regarding the strength of evidence for exposure or health effect developed by looking across evidence streams” (National Academies of Sciences Engineering and Medicine 2021). For ozone and each of the human health effects (T1D, T2D, overweight/obesity, and metabolic syndrome), we examined the conclusions of the evidence synthesis step for the human and in vivo/in vitro evidence streams. An overall causality judgment for each PECO statement was then determined using an adaptation of EPA approaches (US EPA 2020a, 2020b). As shown in Figure 2, we employed four causality determinations: causal, suggestive, insufficient, and not likely.

Figure 2. Evidence integration for causality determination, adapted from EPA (US EPA 2020b).

• Causal: A causal determination would be based on strong positive evidence from both human and animal/mechanistic data streams.

• Suggestive: A conclusion of suggestive would be based on evidence streams that include some studies showing associations with health effects, but with uncertainties in the overall evidence from inconsistent inter- or intra-study results or other limitations.

• Insufficient: If the bodies of evidence were categorized as weak for both streams, we concluded that the evidence was insufficient for a causal determination. This mirrors EPA’s “Inadequate” category which was defined as bodies of evidence with studies “of insufficient quantity, quality, consistency or statistical power to permit a conclusion regarding the presence or absence of an effect.”

• Not likely: This conclusion would be based on strong evidence from both human and animal/in vitro data streams that is null OR significant in the direction opposite to the hypothesized direction.

3. Results

3.1. Literature search and overview of studies and data sources

3.1.1. Human studies

The literature searches identified 2789 unique publications (Figure 3). Screening of the publication titles and abstracts reduced this number to 94. These 94 studies underwent a full-text screen, from which 34 publications were identified for final inclusion in the review. No additional studies were identified from a bibliography review of these publications. Study descriptions, design, approach to exposure assessment, and results are summarized for each of the 34 studies in Table 2. Of the 34 studies, nine (27%) were conducted in the US (Hathout et al. 2006; Chen et al. 2016; White et al. 2016; Miller et al. 2016a; Jerrett et al. 2017; Hernandez et al. 2018; Li et al. 2018; Toledo-Corral et al. 2018; Starling et al. 2020), nine in China (Dong et al. 2014; Li et al. 2015; Song et al. 2018; Yang et al. 2018a, 2018b, 2019a, 2019b; Li et al. 2019; Zhang et al. 2019), six in Europe (Malmqvist et al. 2015; Di Ciaula 2016; Tamayo et al. 2016; Lanzinger et al. 2018; Orioli et al. 2018; Renzi et al. 2018), three in Taiwan (Chuang et al. 2010, 2011; Chin et al. 2018), three in South Korea (Kim and Hong 2012; Kim et al. 2018; Shin et al. 2019), two in Canada (To et al. 2015; Elten et al. 2020), one in Chile (Dales et al. 2012), and one in Malaysia (Wong et al. 2020).

Figure 3. PRISMA diagram for human studies.

Table 2. Synopsis of human studies used in this review.

# Suitable reference	Cohort name or study population; age group; country, N	Design	Exposure characterization (method) and mean concentration (SD)	Results for each analysis (95% CI) **Statistically significant at p < 0.05 results in bold
OBESITY
Dong et al. (2014)	25 Districts Chinese Study, or Seven Northeastern Cities Study Children; China N = 30 056	Cross sectional	LT: from 25 stations, used monitors within 1 km of school and residence 3-Year mean values were from yearly mean levels which in turn were obtained from daily 8-h average concentrations Mean (SD): 27.4 (8.1) ppb	OR scaled to the IQR (11.5 ppb for O3) Obese/overweight: BMI > 85th % OR = 1.14 (1.06–1.22)** Obese: BMI > 95th % OR = 1.14 (1.04–1.24)** Overweight: BMI > 85th % and <95th % OR = 1.09 (1.03–1.15)**
Li et al. (2015)	33 Communities Chinese Health Study Adults; China N = 24 845	Cross sectional	LT: communities and participant homes; 3-year averages derived from daily 8-h averages Mean: 49.4 (14.07) µg/m^3	Per IQR Overweight/Obese: BMI > 25 kg/m^2 OR = 1.09 (1.05–1.14)** Obese: BMI > 30 kg/m^2 OR = 1.10 (1.01–1.20)** Overweight: BMI > 25 and <30 kg/m^2 OR = 1.09 (1.04–1.13)**
Shin et al. (2019)	Korean Community Health Survey (KCHS) Adults; South Korea N = 100 867	Cross sectional	LT: average concentration of hourly data over 10 years Mean: 23.4 (4.5) ppb	Obese: BMI > 25 kg/m^2 Males OR = 0.98 (0.93, 1.04) Females OR = 1.03 (1.00, 1.06)** Note: quantitative results for stratified analyses only
# Starling et al. (2020)	Healthy Start Study Infants; USA N = 1125 (term birth weight) N = 951 (adiposity at birth) N = 574 (adiposity at 5 months)	Cohort	LT: monitors within 50 km of residential address, using max 8-h average during each 24-h period; these were averaged over each trimester and full pregnancy. Inverse-distanced weighted values used for each participant. Median average 8-h max, ppb (IQR) Full pregnancy: 44.0 ppb (5.9 ppb)	Birthweight – difference (g) per IQR Tri 1: 24.5 (−100.7, 149.8) Tri 2: −18.3 (−140.5, 103.9) Tri 3: 20.3 (−88.0, 128.7) Full pregnancy: −1.0 (−63.0, 61.0) Adiposity at birth – difference (%) per IQR Tri 1: 1.05 (−0.36, 2.46) Tri 2: 0.24 (−1.11, 1.59) Tri 3: 0.40 (−0.80, 1.60) Full Pregnancy: 0.58 (−0.11, 1.26) Adiposity at 5 months – difference (%) per IQR Tri 1: −0.69 (−3.11, 1.74) Tri 2: 0.10 (−2.33, 2.54) Tri 3: 2.16 (0.07, 4.26)** Full pregnancy: 0.31 (−0.86, 1.47)
# White et al. (2016)	Black Women’s Health Study (BWHS) Adults, USA N = 38 374	Cohort	LT: CMAQ model with a resolution of 12 km. Estimates were made at the centroid of each census tract. 8-h max. Daily estimates for 8-h maximum levels were compiled into annual average of daily peak concentrations. Mean: 37.5 (4.5) ppb	Weight: change in weight (kg) per IQR (6.7 ppb O3) 0.16 (−0.11, 0.43)
Yang et al. (2019a)	33 Communities Chinese Health Study (33CCHS) Adults, China N = 15 477	Cross sectional	LT: communities and participant homes within 1 km of monitoring stations; 3-year averages derived from daily 8-h averages Mean: 49.4 (14.1) μg/m^3	Overweight/Obese: BMI ≥ 25 kg/m^2 per 10 µg/m^3 increase OR = 1.00 (0.97, 1.03)
Yang et al. (2019b)	China Health and Retirement Longitudinal Study (CHARLS) Adults ≥ 45 years; China N = 13 414	Cross sectional	LT: monitor near city of household, Hourly levels from 125 cities for 5 months Mean (SD): 67.89 (16.91) (unit unclear)	Overweight: BMI ≥24 and <27.9 kg/m^2 OR = 1.015 (1.01, 1.02)** Obese: BMI > 28 kg/m^2 OR = 1.022 (1.016, 1.028)** BMI score β = 0.021**
Zhang et al. (2019)	China Health and Retirement Longitudinal Study (CHARLS); Adults ≥ 60 years; China N = 4 364	Cross sectional	LT: average concentrations Not reported (exposure reported as days exposed)	Obese: WC ≥90 cm men; ≥80 cm women OR = 1.002 (0.999, 1.005) Obese: BMI > 28 kg/m^2 OR = 1.0004 (0.996, 1.004)
T2D
Chen et al. (2016)	BetaGene Adults; USA N = 1023	Cross sectional	ST and LT: daily and monthly cumulative averaged exposures over various lagged periods up to 1 year Up to 4 air quality monitors spatially mapped to residence locations using inverse distance-squared interpolation with a maximum radius of 50 km or nearest monitor Mean 30-day cumulative (SD): 43.4 (13.9) ppb Mean annual (SD): 40.8 (6.8) ppb	Fasting glucose FSIGT OGTT HOMA-IR “No significant associations” [p. 549]
Chuang et al. (2010)	Taiwanese Survey on Prevalence of Hyperglycemia, Hyperlipidemia and Hypertension (TwSHHH) Adults; Taiwan N = 7578	Cross sectional	ST: used daily concentration from nearest monitor within 1–10 km of household 1 or 2–6-day averages for day of exam Mean (SD): 26.83 (9.70) ppb	HbA1C change (%) per IQR 1-d average 0.06 (0.02–0.10)** 3-d average 0.05 (0.02–0.08)** 5-d average 0.07 (0.04–0.10)** Fasting glucose change (%) per IQR 1-d average 0.40 (−0.53, 1.34) 3-d average 0.20 (−0.62, 1.01) 5-dage average 0.77 (−0.05, 1.59)
Chuang et al. (2011)	Social Environment and Biomarkers of Aging Study (SEBAS) Adults > 54 years; Taiwan N = 1023	Cross sectional	LT: monitors within each county or city of the household were averaged over 365 days before the date of examination and assigned to each individual living in the area. Mean (SD): 22.95 (6.76) ppb	Change (mg/dl) per IQR HbA1C 1.30 (0.97–1.63)** Fasting glucose 21.10 (12.03, 30.17)**
Hernandez et al. (2018)	Behavioral Risk Factor Surveillance System (BRFSS) Adults; USA N = 862 519	Cross sectional	LT: CMAQ (Community Multiscale Air Quality Modeling System) with a resolution of 12 km. Estimates were made at the centroid on a county level using daily 8-h maximum to generate annual average concentrations. Median US: 40.9 ppb	T2D prevalence PR = 1.08 (1.04, 1.12)**
# Jerrett et al. (2017)	Black Women's Health Study (BWHS); Adult women; USA N = 43 003	Cohort	LT: CMAQ model estimates of daily 8-h max with a resolution of 12 km. Daily estimates were averaged at the centroid of each census tract. IQR: 6.7 ppb; Max: 56.4 ppb	T2D incidence Fully adjusted model: HR =1.18 (1.04, 1.34)** Basic model + PM2.5 + NO2: HR = 1.13 (0.97, 1.31)
# Kim and Hong (2012)	Korean Elderly Environmental Panel (KEEP) Adults ≥ 60 years; South Korea N = 560	Cohort	ST: daily mean concentration of monitor nearest residence; average distance <1 km Mean over visit day through lag day 10: 19.38 (SD 7.96) ppb	Change (%) with IQR Fasting glucose All: 0.19% (0.09, 0.28)** Without diabetes: 0.09% (0.02, 0.16)** With diabetes: 0.68% (0.28, 1.08)** HOMA-IR All: 0.30% (0.06, 0.53)** Without diabetes: 0.12% (–0.11, 0.35) With diabetes: 1.22% (0.44, 2.00)** Insulin All: 0.71% (0.02, 1.39)** Without diabetes: 0.32% (–0.39, 1.03) With diabetes: 2.78% (0.79, 4.78)**
# Li et al. (2019)	Chronic Disease Surveillance System (CDSS), All ages, China N = 25 130	Time series	LT: daily maximum 8-h average concentrations for whole city were used to obtain monthly averages Mean (SD): 71.10 (34.36) µg/m^3	T2D incidence +PM10: RR = 1.45 (0.8, 2.31) +PM10+NO2+SO2: RR = 1.07 (0.35,6.81) Overall (single pollutant), lag 40 months: RR = 0.78 (0.68, 0.90)**
Li et al. (2018)	Framingham Offspring Cohort and Third Generation Cohort Adults; USA N = 4116	Cohort without baseline data	ST: daily averages of two ozone monitors in the greater Boston, MA area O3 monitors were not at the Supersite Mean (SD): 23.7 (10.9) ppb	Fasting glucose Insulin HOMA-IR Qualitative results only, “associations between O3 and glucose were negative across all moving averages” [p. 8]
# Miller et al. (2016a)	Adult volunteers, USA N = 24	Controlled human randomized crossover	ST: 0.3 ppm, 2 h (15 min of exercise alternating with 15 min of rest) 300 ppb	Fasting Glucose, HOMA-IR, Insulin Qualitative results only “No significant differences were evident” [p. 1384]
Orioli et al. (2018)	Italian National Institute of Statistics surveys Adults > 45 years; Italy N = 376 157	Cross sectional	LT: annual values derived from daily maximum 8 h running average from April to September, weighted for the population of each municipality (data were available for 2003, 2005, 2007, and 2010) Range of median values: 99.8–106.3 µg/m^3	Adult diabetes prevalence (T1 and T2 diabetes) Main mixed model: OR = 1.06 (1.01, 1.11)** High household income, main model: OR = 1.04 (0.99, 1.09) Low household income, main model: OR = 1.09 (1.04, 1.15)** Females, main model: OR = 1.03 (0.98, 1.07) Males, main model: OR = 1.10 (1.04–1.16)**
# Renzi et al. (2018)	Rome Longitudinal Study; Adults ≥ 35 years; Italy 1 425 580 (prevalence) 1 319 193 (incidence)	Cohort	LT: residential exposure from a 1 km grid dispersion model of summer only data; daily 8-h average levels from May to September Mean (SD): 97.4 (6.5) µg/m^3	T2D incidence per 10 µg/m^3 Model 3, adjusted for NOx HR = 1.023 (1.010, 1.036)** Age < 50 (two pollutant) HR = 1.051 (1.021, 1.083)** Age 50–60 (two pollutant) HR = 1.018 (0.994, 1.043) Age > 60 (two pollutant) HR = 0.999 (0.984, 1.015) Female (two pollutant) HR = 1.031 (1.013, 1.048)** Male (two pollutant) HR = 0.994 (0.977, 1.010) T2D prevalence Model 3, per 10 µg/m^3 OR = 0.999 (0.989, 1.010)
Shin et al. (2019)	Korean Community Health Survey (KCHS) Adults; South Korea N = 100 867	Cross sectional	LT: average concentration of hourly data over 10 years Mean (SD): 23.4 (4.5) ppb	Adult diabetes Females OR = 1.15 (1.04–1.28)** Males OR = 0.93 (0.85, 1.02) Note: quantitative results for stratified analyses only
Wong et al. (2020)	National Health and Morbidity Surveys (NHMS) Adults; Malaysia 2006: 56 710 2011: 28 498 2015: 29 460	Cross sectional	LT: Average annual concentration using daily average data from monitors within each state using the station nearest to the residence. 2006: Mean: 31.83 µg/m^3 2011: 33.58 µg/m^3 2015: Mean 39.34 µg/m^3	Self-reported diabetes prevalence 2006: OR = 1.553 (1.328, 1.816)** 2011: OR = 2.043 (1.572, 2.656)** 2015: OR = 0.916 (0.778, 1.079) Underdiagnosed diabetes, fasting glucose levels > 6.1 mmol/L 2006: OR = 1.498 (1.240, 1.810)** 2011: OR = 1.380 (1.064, 1.790)** 2015: OR = 1.174 (1.032, 1.334)** Self-reported + underdiagnosed diabetes 2006: OR = 1.588 (1.400, 1.800)** 2011: OR = 1.701 (1.387, 2.086)** 2015: OR = 1.062 (0.951, 1.187)
Yang et al. (2018a)	33 Communities Chinese Health Study (33CCHS) Adults; China N = 15 477	Cross sectional	LT: communities and participant homes within 1 km of monitoring stations ; 3-year averages derived from daily 8-h averages Mean (SD): 49.4 (14.1) µg/m^3	Per IQR (22 µg/m^3) increase: Adult diabetes prevalence OR = 1.14 (1.05, 1.25)** Fasting glucose (mmol/L) β = 0.04 (0.01, 0.07)** 2 h glucose (mmol/L) β = 0.13 (0.05, 0.22)** HOMA-IR β = 0.09 (-0.10, 0.28) HOMA-B β = −6.72 (−14.14, 3.70) 2 h Insulin (µU/L) β = 1.43 (0.15, 2.71)** Fasting insulin (µU/L) β = 0.24 (0.20, 0.67)**
Yang et al. (2019a)	33 Communities Chinese Health Study (33CCHS) Adults; China N = 15 477	Cross sectional	LT: communities and participant homes within 1 km of monitoring stations ; 3-year averages derived from daily 8-h averages Mean (SD): 49.4 (14.1) µg/m^3	Adult diabetes prevalence per 10 µg/m^3 increase: OR = 1.04 (0.99, 1.09)
T1D
Di Ciaula (2016)	Italian Hospital Discharge Registry Children; Italy N = 1501 hospitalization among 631 275 subjects	Ecological annual concentrations compared to annual prevalence hospitalizations	LT: monitor(s) within city of household averaged over one year Mean (SE): 67.9 (1.4) µg/m^3	T1D hospitalization (proxy for incidence) correlation = −0.05, p = 0.53 “…T1D incidence rate was significantly and positively correlated with mean yearly PM10, but not with ozone” [p. 39] … Non-significant ORs (data not shown)
# Elten et al. (2020)	Better Outcomes Registry and Network (BORN) Children; Canada N = 754 698	Retrospective cohort	LT: monitors at centroid of residential 6-digit postal code using daily 8-h maximum concentration during warm season (May–October). Monthly estimates were derived using a spatiotemporal interpolation technique for all postal codes within 25 km of a station. Mean during pregnancy (SD): 25.91 (3.21) ppb	Diabetes incidence, Ozone exposure at Tri 1: HR = 2.00 (1.04, 3.86) per 10.31 ppb IQR** Tri 2: HR = 0.59 (0.30, 1.18) per 10.42 ppb IQR Tri 3: HR = 1.43 (0.72, 2.83) per 10.36 ppb IQR Entire pregnancy: HR = 1.14 (0.75–1.74) per 4.33 ppb IQR Childhood: HR = 0.89 (0.69–1.14) per 3.31 ppb IQR
# Hathout et al. (2006)	Children; US N = 402	Case-control study	LT: three or less monitors within zip code of household, using monthly and yearly averages. O3 concentration data by zip code month and year obtained by interpolating month/year statistics from fixed-site monitoring stations (≤3)to zip code centroids. Average exposure for each participant used. Cases: 29.4 (SD 7) ppb; Controls: 25.8 (SD 5) ppb	T1D incidence Basic model (BM) = (O3 + age): OR = 1.73 (1.08, 2.77)** BM + family hx of diabetes: OR = 1.55 (0l93, 2.256) BM + highest parental education: OR = 1.80 (1.11, 2.95)**
Lanzinger et al. (2018)	Diabetes follow-up registry Children; Germany N = 37 372	Cohort, one measurement without baseline data	LT: area-weighted mean per postcode area (O3-accumulated ozone exposure AOT) averaged over 5 years; AOT defined as the sum of the differences between the measured values and the threshold value of 80 µg/m^3 averaged over 5 years between May and July (measurements taken between 8:00 am and 10:00 pm) Mean (SD): 48.7 (5.2) µg/m^3	HbA1c among T1D cases O3-AOT % change mean per IQR increase −3.7% (−4.4, −3.0)** O3 % change mean per IQR increase −0.8% (−1.3, −0.4)**
# Malmqvist et al. (2015)	Skåne Study and Better Diabetes Diagnosis Study Children; Sweden N = 930 controls, 324 cases	Case-control study	LT: nearest monitor within 32 km of residence, average distance 8.5 km; residential coordinates linked to air pollution database (no idea what this means); 24-h average data aggregated into trimester data Q4: > 30.6 ppb	T1D incidence Tri 2 (multivariate analysis by exposure levels O3 43.0–51.0 µg/m^3 OR = 1.10 (0.74, 1.61) O3 52.0–60.0 µg/m^3 OR = 0.99 (0.68, 1.46) O3 > 60 µg/m^3 OR = 1.56 (1.04, 2.35)** Tri 1 and Tri 3, data not shown
Tamayo et al. (2016)	Clinical Course of Type 1 Diabetes in Children, Adolescents and Young Adults with Disease Onset in Preschool Age Children; Germany N = 771	Cross sectional	LT: O3-AOT40 used (the sum of differences between 1 h max mean O3 measurements above 40 pb between 8 am and 8 pm from May-July and averaged over 5 years. Participant exposure calculated by postal code and intersection of 8 × 8 km, with adjustment for population. Mean: 13.9 AOT40 (µg/m^3*h)	HbA1c among T1D cases Fully adjusted model with fixed effects: −1.39 (−2.73, −0.06)** Fully adjusted Model with fixed effects and region as random intercept: −1.49 (−3.58, 0.61)
Toledo-Corral et al. (2018)	Childhood Obesity Research Center (CORC) Air Study Children; USA N = 429	Cross sectional	LT: hourly data were averaged to daily levels, which were then used to calculate monthly averages and cumulative 12-month exposure (prior to clinical visit); monthly exposures were spatially interpolated (maximum radius 50 km) from up to 4 monitors using inverse distance-squared weighting 12-month cumulative mean (SD): 20.9 (3.0) ppb	Percent change per 1 SD in previous cumulative 12-month exposure Fasting insulin: −0.2%, p = 0.96 HOMA-IR: −0.04%, p = 0.99 Fasting glucose: 0.2%, p = 0.69 2-h glucose: 0.5%, p = 0.69
METABOLIC SYNDROME AND RELATED OUTCOMES
Chen et al. (2016)	BetaGene Adults; USA N = 1023	Cross sectional	ST and LT: daily and monthly cumulative averaged exposures over various lagged periods up to 1 year Up to 4 air quality monitors spatially mapped to residence locations using inverse distance-squared interpolation with a maximum radius of 50 km or nearest monitor Mean 30-day cumulative (SD): 43.4 (13.9) ppb Mean annual (SD): 40.8 (6.8) ppb	Fasting glucose, Triglyceride, HDL-C Qualitative results only “…we did not find significant associations of O3 or TRAP with metabolic outcomes.” [p. 552]
Chuang et al. (2010)	Taiwanese Survey on Prevalence of Hyperglycemia, Hyperlipidemia and Hypertension (TwSHHH) Adults; Taiwan N = 7578	Cross sectional	ST: used daily concentration from nearest monitor within 1–10 km of household 1 or 2–6-day averages for day of exam Mean (SD): 26.83 (9.70) ppb	Change (%) per IQR ppb Fasting glucose 1-d average 0.40 (−0.53, 1.34) 3-d average 0.20 (−0.62, 1.01) 5-dage average 0.77 (−0.05, 1.59) Triglyceride 1-d average: 2.70 (−0.04, 5.44) 3-d average: 0.30 (−2.11, 2.72) 5-d average: 2.15 (−0.03, 4.32) HDL-C 1-d average: −0.10 (−0.67, 0.34) 3-d average: −0.11 (−0.64, 0.41) 5-d average: −0.03 (−0.53, 0.47) SBP/DBP 1-d average: 0.10 (−0.39, 0.59)/−0.14 (−0.55, 0.26) 3-d average: 0.25 (−0.23, 0.73)/0.37 (0.04, 0.69)** 5-d average: 0.14 (−0.39, 0.67)/0.30 (−0.03, 0.62)
Chuang et al. (2011)	Social Environment and Biomarkers of Aging Study (SEBAS) Adults 54 years and older; Taiwan N = 1023	Cross sectional	LT: monitors within each county or city of the household were averaged over 365 days before the date of examination and assigned to each individual living in the area. Mean (SD): 22.95 (6.76) ppb	Change (mg/dl) per IQR Fasting glucose 21.10 (12.03, 30.17)** Triglyceride 2.13 (−18.10, 22.36) HDL-C −0.48 (−3.69, 2.73) SBP/DBP 21.51 (16.90, 26.13)**/20.56 (18.14, 22.97)**
# Miller et al. (2016a)	Adult volunteers, USA N = 24	Controlled human randomized crossover	0.3 ppm, 2 h (15 min of exercise alternating with 15 min of rest) 300 ppb	Fasting glucose Triglycerides HDL-C SBP Qualitative results only “These were no significant differences as determined by paired t-test between air and ozone exposure when compared immediately post exposure.” Suppl
Yang et al. (2018b)	33 Communities Chinese Health Study (33CCHS) Adults, China N = 15 477	Cross sectional	LT: communities and participant homes within 1 km of monitoring stations ; 3-year averages derived from daily 8-h averages Mean: 49.40 µg/m^3	Metabolic syndrome per 10 µg/m^3 increase in O3 OR = 1.10 (1.01–1.18)**
DIABETES COMPLICATION
# Chin et al. (2018)	Adults with T2D; Taiwan N = 812	Cohort	LT: yearly concentrations from 72 fixed air quality monitor stations, mapped to participant addresses Mean (SD): 28.8 (1.3) ppb	Annual increase in ACR β = 0.0009, p = 0.92
# Dales et al. (2012)	All ages; Chile N = 5 280 000 (for all sectors examined)	Time series	ST: daily averaging for monitor(s) in a sector; 24-h average. Unclear if location from monitor was based on residence or hospital. Mean (SD): 64.41 (38.13) ppb	Coma/ketoacidosis among patients with T1D or T2D RR per IQR (63.50) RR = 1.069 (95% CI 0.982, 1.163)
# Kim et al. (2018)	Emergency room patients All ages; South Korea N = 3527 visits	Time series semi-ecological	ST: daily averages of ozone monitors in Seoul. 8-h average concentration over 5 years Mean (SD): 18.12 (9.89) ppb	Coma among patients with T1D and T2D, RR per IQR RR = 1.009 (95% CI: 0.932, 1.092)
# Song et al. (2018)	Hospitalizations in Shijiazhuang Adults > 35 years; China N = 69 451	Time series semi-ecological	ST: daily concentrations from fixed monitors. Method not provided. Mean (SD): 80 (52) µg/m^3	Diabetes Hospitalization (% change in daily admissions) “For O3, most of lag structures calculated negative relationship, and a statistically significant relationship was observed at lag 2. However, based on the negative Spearman correlation coefficients between O3 and other pollutants, this observation does not provide the evidence of a negative association between O3 and T2DM; it is most likely that O3 might act as a side reflection of better air quality in severe polluted cities.” pgs. 30153–30154
# To et al. (2015)	Provincial health data All ages, Canada N = 13 million population	Time series semi-ecological	ST: average daily 1 h maximum over 14 areas within province of Ontario Mean (SD): 39.57 (14.15) ppb	Diabetes Emergency Department visit per 10 ppb O3 RR = 1.00 (95% CI 1.00, 1.00) Diabetes Hospitalization per 10 ppb O3 RR = 1.00 (95% CI 0.99, 1.00)** Diabetes outpatient visits per 10 ppb O3 RR = 0.99 (95% CI 0.99, 0.99)**

Tri: trimester; d: day; LT: long term (more than 30 days); ST: short term (30 days or less); ACR: albumin-to-creatinine ratio; HDL-C: high density lipoprotein–cholesterol; SBP/DBP: systolic blood pressure/diastolic blood pressure OR: Odds Ratio: RR: relative risk; IQR: interquartile range; #: studies considered Suitable for causal assessment.

**Statistically significant results in bold.

Several investigators used the same data source but reported on different outcomes; these included publications from the 33 Communities Chinese Health Study (Li et al. 2015; Yang et al. 2018a, 2018b, 2019a), the China Health and Retirement Longitudinal Study (CHARLS) (Zhang et al. 2019; Yang et al. 2019b), and the US Black Women’s Health Study (BWHS) (White et al. 2016; Jerrett et al. 2017).

3.1.2. In vivo/in vitro animal studies

The literature searches identified 3118 toxicologic publications (Figure 4). Screening of the titles and abstracts reduced this number to 167. From the 167 publications that underwent a full-text screen, 40 studies were identified for inclusion. An additional four studies were identified and included after a review of the bibliographies of these publications. Of the 44 in vivo/in vitro animal studies identified, 35 were performed in rats (Hathaway and Terrill 1962; Clemons and Garcia 1980a, 1980b; Clemons and Wei 1984; Mole et al. 1985; ErkenBrack and Clemons 1988; Mautz and Bufalino 1989; Sen et al. 1993; Watkinson et al. 1993, 1995; Huffman et al. 2001, 2006; Creţu et al. 2010; Martrette et al. 2011; Bass et al. 2013; Gordon et al. 2013, 2014, 2016a, 2016b, 2017a, 2017b; Wagner et al. 2014; Miller et al. 2015, 2016b, 2016c; Ramot et al. 2015; Vella et al. 2015; Thomson et al. 2016, 2018; Cestonaro et al. 2017; Henriquez et al. 2018, 2019, 2020; Snow et al. 2019) including one pregnant rat model (Miller et al. 2017). The remaining studies included seven in mice (Graham et al. 1982; Umezu et al. 1993; Slade et al. 1997; Last et al. 2005; Johnston et al. 2006; Aibo et al. 2010; Mathews et al. 2017), one in guinea pigs (Vaughan et al. 1984), and one in vitro study (Peters et al. 1993).

Figure 4. PRISMA diagram for in vivo/in vitro animal studies.

3.2. Assessment of study suitability for evidence synthesis

Of the 34 epidemiologic studies and 44 animal studies, those that met the suitability criteria formally moved forward to the evidence synthesis step. The remainder were considered Supplemental. While Supplemental studies were not included in evidence synthesis in this review, it is useful to be aware of the full body of literature.

3.2.1. Human studies

As described in the Methods section, the element of temporality, i.e. the demonstration that exposure preceded the outcome, was used to distinguish Supplemental studies from Suitable ones. Results from studies categorized as Supplemental are discussed in the following sections; however, they were not used in evidence synthesis. Of the 34 publications identified, 19 were determined to be Supplemental. Sixteen studies employed a cross-sectional design (Chuang et al. 2010, 2011; Dong et al. 2014; Li et al. 2015; Chen et al. 2016; Tamayo et al. 2016; Hernandez et al. 2018; Orioli et al. 2018; Toledo-Corral et al. 2018; Yang et al. 2018a, 2018b; Shin et al. 2019; Zhang et al. 2019; Yang et al. 2019a, 2019b; Wong et al. 2020). Three studies did not collect baseline data for temporal comparison data (Lanzinger et al. 2018; Li et al. 2018), or relied upon prevalent cases (Di Ciaula 2016). Detailed information for both Suitable and Supplemental studies can be found in Table 2.

Five studies evaluated associations between ozone exposure and complications from diabetes (Dales et al. 2012; To et al. 2015; Chin et al. 2018; Kim et al. 2018; Song et al. 2018). On examination, these investigations were found to not directly address the PECO questions and so were not considered further.

3.2.2. Toxicologic studies

The five criteria described in the Methods section were used to categorize the 44 publications as Suitable for evidence synthesis or Supplemental. Thirty-three studies were considered Suitable for evidence synthesis. Included were two studies that did not fully meet all five criteria. Miller et al. (2016c) met four criteria for the control (SHAM) group but met the required five criteria for the exposure group. Mautz and Bufalino (1989) did not provide a description of statistical tests but did provide p values for the core body temperature (BT) results.

Four studies were categorized as Supplemental. Three reported on ozone exposure and effects on thyroid hormones and had only one critical deficiency (Clemons and Garcia 1980a; Clemons and Wei 1984; Sen et al. 1993). The fourth publication (Last et al. 2005) did not meet the criteria as the numbers of animals in the groups were not defined, although the reported effects of ozone on body weight [BW] were statistically significant, which is suggestive of an adequate sample size.

Excluded from further evaluation and discussion were seven papers with critical deficiencies. These deficiencies included missing information on the generation and monitoring of ozone for exposure groups (Hathaway and Terrill 1962; ErkenBrack and Clemons 1988), missing or inadequate sample size (Peters et al. 1993; Slade et al. 1997), and no control groups (Peters et al. 1993; Slade et al. 1997). The publication by Clemons and Garcia (1980b) was excluded because the authors reported the same data as in Clemons and Garcia (1980a). Creţu et al. (2010) was excluded because, although it met the suitability criteria, it was not possible to determine which treatment group the reported data represented, therefore it was impossible to determine ozone-related effects. Wagner et al. (2014) was excluded because, upon further review, the study did not include PECO outcomes relevant to this review. Note that only one of the 44 publications identified was an in vitro study (Peters et al. 1993), but it had no information regarding the methodology of ozone exposure, the use of control groups, n values, or statistical analyses methods. Accordingly, the remaining sections will discuss only the in vivo animal studies.

3.3. Evidence synthesis/confidence characterization

Confidence in the literature for ozone exposure being associated with metabolic outcomes was categorized as either strong or weak (see Section 2.5). General observations regarding strengths and weaknesses of the associations in the reviewed literature were followed by an assessment of the outcome-specific literature (Sections 3.3.1 and 3.3.2 cover human and animal studies, respectively).

3.3.1. Human studies

For the outcomes of interest in this review (T1D, T2D, overweight/obesity, and metabolic syndrome), all 34 studies relied upon robust ascertainment approaches, such as physician diagnosis and standard laboratory results from fasting blood tests. In addition, where relevant, all the studies reported participant fasting, except for one experimental study (Miller et al. 2016a). However, a few of the reviewed studies were not able to distinguish between T1D vs. T2D (To et al. 2015; Di Ciaula 2016; Hernandez et al. 2018; Orioli et al. 2018). Only one study, from the 33 Communities Chinese Health Study, defined and evaluated “metabolic syndrome” (Yang et al. 2018b), but the authors utilized a cross-sectional approach, so this paper could not be used to evaluate causality.

In the following subsections, we characterize the confidence in the subsets of studies related to each outcome, noting that the general issues (See Section 4.2) for exposure assessment cut across the literature.

3.3.1.1. Studies of overweight/obesity

Eight publications evaluated associations between ozone and effects on weight. Only two studies (White et al. 2016; Starling et al. 2020), both conducted in the US, were considered Suitable for evidence synthesis (Table 3). These cohort studies could not be directly compared as the authors examined different populations, outcomes, and exposures. The Healthy Start Study (Starling et al. 2020) assessed birth weight and adiposity at birth and at five months, whereas the Black Women’s Health Study (White et al. 2016) focused on weight change over several years in adult women. In addition, Starling et al. characterized ozone exposure using monitors within 50 km of a residence, whereas White et al. relied upon a smaller spatial scale of within 12 km of the residence. The results of these studies were not statistically significant, with the exception of adiposity at five months related to maternal exposure during the third trimester (Starling et al. 2020); as shown in Table 3, the directionality of the results related to the trimester(s) of ozone measurement was not consistent.

Table 3. Summary of associations between ozone exposure and effects on weight in human (Suitable studies).

References	Population	Exposure duration	Mean concentration (SD) ppb	Adiposity at birth	Adiposity at 5 months	Birth weight	Weight change
Starling et al. (2020)	Children (in utero exposure)	LT	44.0 (5.9)	↑	↑ ↓ ↑	↑↓	–
White et al. (2016)	Adults	LT	37.5 (4.5)	–	–	–	↑

↑: statistically significant, p < 0.05; ↑↓: increased or decreased association; LT: long-term exposure, 30 days or more.

White et al. (2016) also examined a subset of the cohort that did not change residence during the study period. Among these women (n = 10 307), weight changes per interquartile range were slightly but statistically significantly greater (0.49, 95% CIs: 0.05, 0.93) compared to the overall cohort, although these women were also slightly older, heavier, and less likely to be nulliparous.

The remaining publications related to overweight/obesity were cross-sectional and categorized as Supplemental (Dong et al. 2014; Li et al. 2015; Shin et al. 2019; Zhang et al. 2019; Yang et al. 2019a, 2019b). These studies were conducted in China and Korea and focused on adults, with one exception (Dong et al. 2014). The publications by Yang et al. (2019b) and Zhang et al. (2019) were from the same study but evaluated slightly different outcomes and populations. As shown in Table 2, most of the odds ratios (ORs) for these publications were >1.0, with some statistically significant positive results (Dong et al. 2014; Li et al. 2015; Shin et al. 2019; Yang et al. 2019b). Two publications reported null results (Zhang et al. 2019; Yang et al. 2019a).

The strength of the epidemiology evidence associating ozone exposure with overweight/obesity was considered weak. Only two publications were considered Suitable for evidence synthesis and the results did not indicate a consistent association between ozone exposure and effects on weight (Table 3). Neither population (one on infants and the other on adults) nor the outcomes (adiposity, birth weight, and weight change) allowed for direct comparisons.

3.3.1.2. Studies of T2D and measures of glucose metabolism

Fifteen studies evaluated associations between ozone exposure and diagnosis or indicators of T2D, with five of these categorized as Suitable for evidence synthesis (Table 4). These studies fell into two groups: those that examined the relationship between short-term exposures (i.e. <30 days) to ozone and indicators of T2D, including glucose levels, HOMA-IR (Homeostatic Model Assessment for Insulin Resistance), and insulin levels, and those that assessed long-term exposures to ozone and examined associations with T2D incidence.

Table 4. Summary of associations between ozone exposure and T2D (Suitable studies).

References	Exposure duration	Estimated concentration (SD)	T2D incidence	Fasting glucose	HOMA-IR	Insulin
Kim and Hong (2012)	ST	Mean over visit day through lag day 10: 19.38 (SD 7.96) ppb	–	↑	↑	↑
Miller et al. (2016a)	ST	300 ppb	–	Ø	Ø	Ø
Jerrett et al. (2017)	LT	IQR 6.7 ppb; Max 56.4 ppb	↑	–	–	–
Li et al. (2019)	LT	Mean 71.10 (SD 34.36) µg/m^3	↓ ↑	–	–	–
Renzi et al. (2018)	LT	Mean 97.4 (SD 6.5) µg/m^3	↓↑*	–	–	–

↑ ↓: statistically significant, p < 0.05; ↑↓: increased or decreased association; Ø: null with direction not reported; ST: short-term exposure, <30 days; LT: long-term exposure, 30 days or more; IQR: interquartile range.

*Range between 0.9 and 1.05.

The two short-term studies were conducted in South Korea and the US (Kim and Hong 2012; Miller et al. 2016a), and used different designs. Miller et al. was a controlled study of 24 exercising adults exposed to 300 ppb ozone. This level is substantially higher than typical ambient levels (Vingarzan 2004). In contrast, Kim and Hong (2012) conducted a larger (n = 560) cohort study of adults > 60 years of age; the mean ambient ozone concentration in this study was 19.38 ppb. The results across these studies were not consistent. While Miller et al. (2016a) reported statistically non-significant results for the endpoints of interest, Kim and Hong (2012) reported positive and statistically significant results for glucose, insulin, and HOMA-IR. The directionality of the results across the two studies could not be compared because only qualitative results were given by Miller et al. (2016a).

The three long-term studies evaluating T2D incidence were conducted in the US (Jerrett et al. 2017), China (Li et al. 2019), and Italy (Renzi et al. 2018). The direction of the associations within and across studies was not consistent. For example, Li et al. (2019) reported a relative risk (RR) of 1.45 (95% CI 0.80, 2.31) when controlling for PM₁₀, but reported a significant inverse result for the overall single pollutant model (RR = 0.79, 95% CIs: 0.68, 0.90). In contrast Jerrett et al. (2017) observed a statistically significant hazard ratio (HR) (1.18, 95% CIs: 1.04, 1.34) in their single pollutant model, with attenuation after controlling for PM_2.5 and NO₂ (HR = 1.13, 95% CIs: 0.97, 1.31). The statistical significance of the results for Renzi et al. (2018) varied depending on the model structure and covariates. The RR values of these associations ranged from 0.99 and 1.05, making the direction of these small associations difficult to interpret (see Table 2). In addition, the use of covariates differed across the three studies, complicating cross-study comparisons. For example, Jerrett et al. (2017) included lifestyle covariates related to diet and smoking, Renzi et al. (2018) controlled for basic demographic data including socioeconomic factors, and Li et al. (2019) only included meteorologic data, such as temperature, humidity, and wind speed.

Nine of the remaining epidemiologic studies related to T2D were cross-sectional and thus categorized as Supplemental (Chuang et al. 2010, 2011; Chen et al. 2016; Hernandez et al. 2018; Orioli et al. 2018; Yang et al. 2018a, 2019a; Shin et al. 2019; Wong et al. 2020); one was a cohort design but without baseline data (Li et al. 2018). The directionality of the results of these studies for T2D prevalence was positive for the preponderance of the studies, although not consistently statistically significant. Shin et al. (2019) reported an inverse association for male participants (OR = 0.93, 95% CIs: 0.85, 1.02). There was no consistency of direction or statistical significance for blood measures of glucose and insulin (see Table 2).

For the following reasons, the epidemiology literature on associations between ozone exposure and T2D was categorized as weak for the short-term studies: (i) only two studies were identified that were considered Suitable for evidence synthesis; (ii) the studies were not directly comparable as they examined different populations and used different study designs and ozone concentrations. The epidemiology literature on associations between long-term ozone exposure and T2D was also categorized as weak. Despite having three studies that evaluated the same outcome, the results were inconsistent both in direction and statistical significance with no clear reason for these differences.

3.3.1.3. Studies of T1D

Seven studies evaluated associations between ozone exposure and T1D (Table 2). Some studies were specific to T1D, while other study investigators assumed that all pediatric cases were T1D based on age (Elten et al. 2020).

Three studies were considered Suitable for evidence synthesis (Hathout et al. 2006; Malmqvist et al. 2015; Elten et al. 2020) (Table 5); these were conducted in Canada, the US, and Sweden, respectively. The studies addressed different time periods of exposure (in utero and childhood) and yielded mixed results (Table 5). Elten et al. (2020) observed statistically significant positive associations for T1D and in utero ozone exposure in trimester 1 (HR = 2.00; 95% CIs: 1.04, 3.86), positive but null associations for exposure in trimester 3 (HR = 1.43; 95% CI 0.72, 2.83), but an inverse null association related to exposure in trimester 2 (HR = 0.59; 95% CIs: 0.30, 1.18). Malmqvist et al. (2015) reported a statistically significant increased OR at the highest ozone exposure level in trimester 2 (OR = 1.56; 95% CIs: 1.04, 2.35) but not for trimesters 1 or 3 (directionality not reported by the authors). With respect to exposure during childhood, Elten and colleagues reported a non-significant inverse association (HR = 0.89, 95% CIs: 0.69, 1.14), while a statistically significant positive association was observed by Hathout et al. (OR = 1.73, 95% CIs: 1.08, 2.77) (Hathout et al. 2006; Elten et al. 2020).

Table 5. Summary of associations between LT ozone exposure and T1D (Suitable studies).

		Exposure period
Reference	Estimated concentration (SD)	Trimester 1	Trimester 2	Trimester 3	Entire pregnancy	Childhood
Elten et al. (2020)	Mean during pregnancy 25.91 (SD 3.21) ppb	↑	↓	↑	↑	↓
Hathout et al. (2006)	Cases: 29.4 (SD 7) ppb; Controls: 25.8 (SD 5) ppb	–	–	–	–	↑
Malmqvist et al. (2015)	Q4: > 30.6 ppb	↑	↑ ↓ ↑*	↑	–	–

↑: statistically significant, p < 0.05; ↑↓: increased or decreased association; LT: long-term exposure, 30 days or more.

*These three results correspond to increasing exposure levels.

The remaining four T1D studies were categorized as Supplemental because none was designed to evaluate whether exposure preceded diagnosis due to either the cross-sectional design (Tamayo et al. 2016; Toledo-Corral et al. 2018), reliance on prevalent cases (Di Ciaula 2016), or use of a single HbA1c (hemoglobin A1c) measurement without a baseline comparison (Lanzinger et al. 2018). Statistically significant inverse associations of ozone exposure and HbA1c were reported by Tamayo et al. (2016) and Lanzinger et al. (2018). Di Ciaula (2016) reported no correlation of annual T1D hospitalizations and annual ozone levels. No significant adverse associations were reported for blood levels of fasting insulin, HOMA-IR, fasting glucose, or 2-h glucose testing by Toledo-Corral et al. (2018).

The epidemiology literature on associations between ozone exposure and T1D was categorized as weak. The results by exposure period were not consistent within or across studies and there were no discernable reasons for these differences. The findings were based on different exposure periods (in utero and childhood exposure) (Malmqvist et al. 2015; Elten et al. 2020), and it is unclear whether these two time periods are comparable, particularly for the inhalation route of exposure. All three studies collected and evaluated accepted confounders related to T1D. The exposure assessment scale varied across studies. For example, one study used monitoring data within postal codes for monthly averages (Hathout et al. 2006), another relied on monitoring results during the warm seasons for areas within 25 km of a monitoring station (Elten et al. 2020), and a third used measurements within 32 km of a residence (Malmqvist et al. 2015).

3.3.1.4. Studies of metabolic syndrome

Five studies evaluated associations between ozone and metabolic syndrome (Table 2). As described previously, metabolic syndrome is defined by the presence of at least three of five risk factors (elevated waist circumference, elevated TG, reduced HDL cholesterol, elevated blood pressure, and/or elevated fasting glucose). To be inclusive, we examined publications that specifically examined relationships between ozone exposure and metabolic syndrome, as well as those that assessed at least three of the five diagnostic factors.

The experimental short-term exposure study by Miller et al. (2016a) was the only publication considered Suitable for evidence synthesis (Table 6). Miller et al. did not evaluate “metabolic syndrome” as an outcome, but rather assessed four of the five diagnostic factors (waist circumference was not included). Miller et al. did not observe any statistically significant associations between ozone exposure and the four measures associated with metabolic syndrome.

Table 6. Summary of association between ST ozone exposure and metabolic syndrome (Suitable studies).

Reference	Glucose	Triglycerides	HDL-C	SBP/DBP
Miller et al. (2016a)	Ø	Ø	Ø	Ø

Ø: null with direction not reported; ST: short-term exposure, <30 days.

The remaining four studies related to metabolic syndrome were cross-sectional and therefore were categorized as Supplemental (Chuang et al. 2010, 2011; Chen et al. 2016; Yang et al. 2018b). The 33 Communities Chinese Health Study (Yang et al. (2018b) was the only study to formally assess “metabolic syndrome.” They found a positive, statistically significant association with long-term exposure (OR = 1.10, 95% CI 1.01, 1.18). As shown in detail in Table 2, the other three studies reported mixed results (Chuang et al. 2010, 2011; Chen et al. 2016). Statistically significant results were reported by Chuang et al. (2010) for fasting glucose and blood pressure with year-long exposure estimates, while Chen et al. (2016) found statistically non-significant results. It was not possible to compare the directionality of the results since both Chen et al. (2016) and Miller et al. (2016a) reported only qualitative results.

The strength of the evidence on associations between ozone exposure and metabolic syndrome was categorized as weak as only one Suitable study was identified. It is worth noting that the results from this controlled human study may not be generalizable since the study population was composed of 24 healthy adults and the ozone concentration was approximately an order of magnitude higher than ambient concentrations (Vingarzan 2004).

3.3.2. In vivo/in vitro animal toxicologic studies

In this section, we first offer perspective on key strengths and limitations of the overall body of literature. We then provide assessments of the in vivo experimental animal results according to each PECO endpoint. The experimental animal evidence and the conclusions for each experimental outcome are summarized in Table 14.

The studies categorized as Suitable generally used consistent methods and experimental designs (e.g. standard ozone generation and monitoring protocols). It is also worth noting that a variety of rat strains and ages were represented. In addition, while chronic studies were not identified, the studies did attempt to capture short-term exposures, repeated episodic exposures, and sub-chronic exposures. However, few of the toxicology studies evaluated exposure-response in the same study. Accordingly, the analyses in this review involved combining available exposure-response information from studies of different animal strains and sometimes species, with a variety of exposure durations and frequencies. While this approach can add to the strength of an evidence stream if the overall data are coherent, the only outcomes with consistency across studies were glucose tolerance and fasting hyperglycemia tests. The few studies with post-exposure recovery periods allowed for only 18 h or 1 week recovery and were not repeated to demonstrate whether the observed apparent recovery was reproducible following another set of exposures. In addition, the lack of chronic exposure data made it difficult to determine whether adaptation to ozone occurs in exposed animals such that any effect on glucose tolerance and/or hyperglycemia may be considered transient or long-lasting. Finally, it should be noted that the animal exposures were carried out at relatively high levels of ozone, which may not be relevant to typical human exposures.

3.3.2.1. Experimental animal toxicologic studies of glucose-related effects

Potential glucose-related effects of ozone exposure, including impaired glucose tolerance, insulin resistance, fasting hyperglycemia, changes in hepatic gluconeogenesis (GNG), and changes in serum hormone analytes related to glucose homeostasis were evaluated across 18 studies (all 18 were categorized as Suitable for evidence synthesis). Results are discussed below by outcome, with glucose tolerance and fasting glucose levels measured by the glucose tolerance test (GTT, 11 studies), insulin sensitivity by the insulin tolerance test (ITT, two studies), and hepatic GNG by the pyruvate tolerance test (PTT, two studies). Data on serum hormone analytes related to glucose homeostasis, including glucagon, insulin, leptin, and corticosterone (CORT), are also discussed (18 studies).

3.3.2.1.1. Glucose tolerance

In the GTT, glucose is administered to fasted animals and blood samples are taken at various time-points to measure blood glucose concentration, generally over the course of 2 h. Similar to humans, the GTT is used in toxicologic studies to assess the competence of the regulatory mechanisms that determine blood glucose concentration; it provides a physiological overview of any changes in glucose tolerance. Eleven studies evaluated associations between ozone exposure and effects on glucose homeostasis and glucose tolerance as measured by fasting blood glucose concentration and the GTT (Table 7). Rats of various strains were used as the animal model, including Brown Norway (Bass et al. 2013), Long-Evans (Gordon et al. 2017a, 2017b; Snow et al. 2019), Wistar (Vella et al. 2015), Wistar Kyoto (Miller et al. 2015, 2016b, 2016c; Henriquez et al. 2019, 2020), and F344 (Thomson et al. 2018).

Table 7. Glucose tolerance testing (GTT) analysis results after inhalation ozone exposure in rats (Suitable studies).

References	Animal model				O3 exposure					GTT^b (day)	Blood glucose post-glucose injection^b
	Rat strain	Sex	Age^a (month)	N (/grp)	Grp	Level (ppm)	Duration (h/day)	Frequency	Recovery		Min (minutes): 0	<20^d	20^d	30	40^d	60	90	120	AUC
Gordon et al. (2017b)	L-E	F	3	10	–	0.25	5	2 d	–	1	≈	–	–	↑	–	≈	≈	≈	–
Miller et al. (2015)	WIS-KY	M	2.5	6	1	0.25	6	2 d	–	1	≈	–	–	↓	–	↓	↓	↓	–
				6	2		6	2 d	–	2	≈	–	–	↑	–	↑	≈	≈	–
Bass et al. (2013) (Acute)	BN	M	1	4–10	–	0.25	6	2 d	–	–	≈	–	–	≈	–	≈	≈	≈	–
			4	4–10	–	0.25	6	2 d	–	–	≈	–	–	↓	–	↑	≈	↑	–
			12	4–10	–	0.25	6	2 d	–	–	≈	–	–	↑	–	↑	↑	↑	–
			24	4–10	–	0.25	6	2 d	–	–	≈	–	–	≈	–	≈	↓	↓	–
Bass et al. (2013) (Sub-chronic)	BN	M	4^e	4–10	–	0.25	6	2 d/wk, 13 wks	–	–	≈	–	–	≈	–	≈	≈	≈	–
			12^e	4–10	–	0.25	6	2 d/wk, 13 wks	–	–	≈	–	–	↑	–	↑	↑	↑	–
			24^e	4–10	–	0.25	6	2 d/wk, 13 wks	–	–	≈	–	–	↓	–	↑	↓	≈	–
Miller et al. (2016b) (Acute)	WIS-KY	M	2.5	8–10	1	0.25	5	3 d/wk, 13 wks	–	1 wk-d 1	≈	–	–	≈	–	≈	≈	≈	≈
					2		5	3 d/wk, 13 wks	–	1 wk-d 3	≈	–	–	↑	–	↓	≈	≈	≈
Miller et al. (2016b) (Sub-chronic)	WIS-KY	M	2.5	8–10	1	0.25	5	3 d/wk, 13 wks	–	12 wks-d 1	≈	–	–	↓	–	≈	↓	↓	↓
					2		5	3 d/wk, 13 wks	1 wk	13 wks + 1 wk	≈	–	–	↓	–	≈	≈	≈	≈
Miller et al. (2015)	WIS-KY	M	2.5	6	1	0.5	6	2 d	–	1	↑	–	–	↑	–	↑	↑	↑	–
				6	2		6	2 d	–	2	↑	–	–	↑	–	↑	↑	↑	–
Gordon et al. (2017b)	L-E	F	3	10	–	0.5	5	2 d	–	1	≈	–	–	↑	–	↑	↑	↑	–
Thomson et al. (2018)	F344	M	NR^f	8	–	0.8	4	1 d	–	–	↓	–	–	↑	–	≈	↓	↓	≈
Henriquez et al. (2020)	WIS-KY	M	2.5–3	6–8	1	0.8	4	2 d	–	1	↑	–	–	↑^h	–	↑^h	↑^h	↑^h	↑
				6–8	2		4	2 d	–	1	↑	–	–	↑^h	–	↑^h	↑^h	↑^h	↑
Henriquez et al. (2019)	WIS-KY	M	2.5	6–8	1	0.8	4	2 d	–	1	↑	–	–	↑^h	–	↑^h	↑^h	↑^h	↑
				6–8	2		4	2 d	–	1	↑	–	–	↑^h	–	↑^h	↑^h	↑^h	↑
Gordon et al. (2017a)	L-E	M	5	20	–	0.8	4	2 d	–	1	↑	–	–	↑	–	↑	↑	↑	–
		F		20	–		4	2 d	–	1	≈	–	–	↓	–	↑	↑	↑	–
Snow et al. (2019)	L-E	M	3.5	10	–	0.8	5	2 d	–	1	↑	–	–	↑	–	↑	↑	↑	↑
		F		10	–		5	2 d	–	1	↑	–	–	↑	–	↑	↑	↑	↑
Vella et al. (2015)	WIS	NR^g	NR^g	6	–	0.8	16	1 d	–	–	↑	↑	↑	↑	↑	↑	–	–	↑
Miller et al. (2016c)	WIS-KY	M	3	6^i	–	1.0	4	2 d	–	1	↑	–	–	↑	–	↑	↑	↑	↑
Gordon et al. (2017b)	L-E	F	3	10	–	1.0	5	2 d	–	1	↑	–	–	↑	–	↑	↑	↑	–
Bass et al. (2013) (Time-course)	BN	M	4	6	1	1.0	6	1 d	–	–	↑	–	–	↑	–	↑	↑	↑	–
					2		6	2 d	–	–	≈	–	–	↑	–	↑	↑	↑	–
					3		6	2 d	18 h	2 d–18 h	≈	–	–	↑	–	↓	≈	≈	–
Miller et al. (2015) (Time-course)	WIS-KY	M	2.5	8	1	1.0	6	1 d	–	–	↑	–	–	↑	–	↑	↑	↑	–
				8	2		6	2 d	–	–	↑	–	–	↑	–	↑	↑	↑	–
				8	3		6	2 d	18 h	2d–18 h	↓	–	–	↑	–	↑	≈	≈	–
Miller et al. (2015) (Conc/Resp)	WIS-KY	M	2.5	6	1	1.0	6	2 d	–	1	↑	–	–	↑	–	↑	↑	↑	–
				6	2		6	2 d	–	2	↑	–	–	↑	–	↑	↑	↑	–
Bass et al. (2013) (Acute)	BN	M	1	4–10	–	1.0	6	2 d	–	–	↑	–	–	↑	–	↑	↑	↑	–
			4	4–10	–		6	2 d	–	–	↑	–	–	↑	–	↑	↑	↑	–
			12	4–10	–		6	2 d	–	–	↑	–	–	↑	–	↑	↑	↑	–
			24	4–10	–		6	2 d	–	–	↑	–	–	≈	–	↑	↑	↑	–
Bass et al. (2013) (Sub-chronic)	BN	M	4^e	4–10	–	1.0	6	2 d/wk, 13 wks	–	–	↑	–	–	↑	–	↑	↑	↑	–
			12^e	4–10	–		6	2 d/wk, 13 wks	–	–	≈	–	–	↑	–	↑	↑	↑	–
			24^e	4–10	–		6	2 d/wk, 13 wks	–	–	≈	–	–	↑	–	↑	↑	↑	–
Miller et al. (2016b) (Acute)	WIS-KY	M	2.5	8–10	1	1.0	5	3 d/wk, 13 wks	–	1 wk–d 1	↑	–	–	↑	–	↑	↑	↑	↑
					2		5	3 d/wk, 13 wks	–	1 wk–d 3	↑	–	–	↑	–	↑	↑	↑	↑
Miller et al. (2016b) (Sub-chronic)	WIS-KY	M	2.5	8–10	1	1.0	5	3 d/wk, 13 wks	–	12 wks–d 1	↑	–	–	↑	–	↑	↑	↑	↑
					2		5	3 d/wk, 13 wks	1 wk	13 wks + 1 wk	≈	–	–	↓	–	↓	≈	↑	↓

↑ ↓: statistically significant results; ↑↓: increased or decreased results; ≈: approximately equal results; –: no data; BN: Brown Norway; Conc/Resp: concentration–response; d: day; grp: group; GTT: Glucose Tolerance Test; L-E: Long-Evans; NR: not reported; WIS-KY: Wistar Kyoto; wk: week; h: hour.

^aUnless otherwise indicated, age represents the age of the animals at the start of ozone exposure.

^bUnless otherwise indicated, the GTT was performed immediately after ozone exposure.

^cResults are represented as increased, decreased, or approximately equal blood glucose concentrations in ozone-exposed groups compared with control (air)-exposed groups. The vast majority of the results were determined from graphical data. As such, the data symbols/lines were visually distinguished from each other; if they could not be distinguished, the results were considered approximately equal (≈). If numerical data were available for comparison, <5% differences were considered approximately equal (≈).

^dVella et al. (2015), had additional timepoints as part of their GTT analysis, including various times <20 min (but >0), 20 min, and 40 min. However, for the timepoints that were <20 min, the exact timepoint could not be ascertained from the graph provided in the published study and the actual data values from the GTT analysis were not provided.

^eIn Bass et al (2013), the sub-chronic study began when the rats were 1, 9, and 21 months old and concluded when they were 4, 12, and 24 months old.

^fThomson et al. (2018) reported that the rats were studied at 200–250 g bw.

^gVella et al. (2015) did not report the age or sex of the rats; however, the authors reported that the rats were studied at 400–450 g bw, indicating these were adult males.

^hFor two studies, Henriquez et al. (2019) and Henriquez et al. (2020), statistical analyses do not appear to have been performed for all the timepoints of the GTT. In these studies, statistical analyses were only performed for baseline glucose a total AUC; however, both studies show increased glucose concentrations in the O₃ groups compared with air controls at the timepoints indicated.

ⁱMiller et al. (2016c) had six animals in the ozone-exposed group but only 4 animals in the control (air)-exposed comparison group.

The eleven studies represented a total of 48 study groups. Within each study group, fasting blood glucose measurements were taken prior to/at 0 min (baseline) and after glucose injection at 30, 60, 90, and 120 min time-points. Ozone exposure concentrations ranged from 0.25 to 1.0 ppm.

Four studies evaluated glucose homeostasis following ozone exposure at 0.25 ppm (Bass et al. 2013; Miller et al. 2015; Gordon et al. 2017b). For all study groups at 0.25 ppm ozone, baseline blood glucose concentrations (0 min time-point) in ozone-exposed animals were approximately equal to those of air control animals. This was true regardless of rat age, exposure duration, or exposure frequency. Results at the post-glucose GTT time-points (i.e. 30, 60, 90, and 120 min) were mixed, with none of the differences considered statistically significant.

Two studies evaluated ozone exposure at 0.5 ppm (Miller et al. 2015; Gordon et al. 2017b) (Table 7). Miller et al. (2015) included two study groups of Wistar Kyoto rats (n = 6 each); for both groups, baseline glucose concentrations in ozone-exposed animals were higher than those of air controls; however, the differences were not statistically significant. In Gordon et al. (2017b), female Long-Evans rats (n = 10) exposed to 0.5 ppm ozone demonstrated baseline glucose concentrations approximately equal to air controls; however, a consistent trend of higher blood glucose concentrations at the post-glucose time-points of the GTT was observed, with the majority (7/12) demonstrating statistical significance.

Six studies evaluated ozone exposure at 0.8 ppm (Vella et al. 2015; Gordon et al. 2017a; Thomson et al. 2018; Henriquez et al. 2019; Snow et al. 2019; Henriquez et al. 2020). Across these studies, there were ten total study groups composed of various rat strains (Table 7). In seven of the groups, ozone-exposed animals demonstrated significant fasting hyperglycemia at baseline compared with air controls. The remaining three study groups had mixed results, with none of the differences considered statistically significant. For the post-glucose injection time-points of the GTT, a consistent trend of elevated blood glucose concentrations (about 50:50 significant vs. non-significant), as well as significantly higher area under the curve (AUC) values (5 out of 8), were observed in ozone-exposed animals compared with air controls (Vella et al. 2015; Henriquez et al. 2019, 2020).

Five studies evaluated ozone exposure at 1.0 ppm and glucose homeostasis/tolerance (Bass et al. 2013; Miller et al. 2015, 2016b; Gordon et al. 2017b). Among them, there was a total of 21 study groups (Table 7). In 18 of these study groups, the GTT was administered immediately after ozone exposure, and fasting hyperglycemia was observed in 15 of the ozone-exposed groups, with the majority considered statistically significant. In addition, at all post-glucose time-points of the GTT, ozone-exposed animals in the 18 study groups demonstrated blood glucose concentrations that were consistently higher than air controls, with the vast majority statistically significant.

Three studies (Bass et al. 2013; Miller et al. 2015, 2016b) included study groups in which the animals were allowed to recover post-exposure for 18 h or 1 week before the GTT was administered. Among these studies, substantial differences in glucose homeostasis and glucose tolerance were observed. In Bass et al. (2013), a study group of Brown Norway rats (n = 4–10) was allowed to recover for 18 h after 1.0 ppm ozone exposure (6 h/d × 2 d) before the administration of the GTT. In this recovery group, blood glucose concentrations were not statistically different from air controls at any time-point. In contrast, the animals not allowed a recovery period had significantly higher blood glucose concentrations than air controls for every post-glucose GTT time-point studied (Bass et al. 2013). Similar results were obtained in Miller et al. (2015), in which a study group of Wistar Kyoto rats (n = 8) allowed to recover for 18 h after 1.0 ppm exposure (6 h/d × 2 d) had substantially improved glucose tolerance at all but one GTT time-point. In Miller et al. (2016b), Wistar Kyoto rats (n = 8–10) were exposed to 1.0 ppm ozone for 5 h/d × 3 d/week for 13 weeks, with one study group allowed to recover for one week prior to the administration of the GTT. Animals subjected to the GTT immediately following ozone exposure with no recovery period demonstrated blood glucose concentrations that were significantly higher than air controls, including at baseline and all post-glucose time-points of the GTT and AUC (Miller et al. 2016b). In contrast, animals allowed the 1-week recovery period demonstrated baseline and post-glucose blood glucose concentrations that were approximately equal to air controls at all but one time-point, with no differences considered statistically significant.

Among the eleven studies, two (Bass et al. 2013; Miller et al. 2016b) included groups at 1.0 ppm ozone in which the GTT was administered after sub-chronic episodic ozone exposure. The effects on glucose homeostasis from sub-chronic episodic exposures (e.g. 5 h/d × 3 d/week for 12 weeks) were similar to those observed in animals subjected to single short-term exposures (e.g. 6 h/d × 1 day) (Table 7).

The results across the studies indicate a consistent association between ozone exposure at concentrations ≥0.8 ppm and impaired glucose tolerance/homeostasis in rats. The toxicologic evidence was therefore categorized as strong. There were eleven publications representing a total of 48 study groups comprised of various rat strains and ages, with experiments conducted in several different laboratories (Table 14). In addition, the studies had similar methodologies for ozone exposure and GTT administration, thus allowing for direct comparisons. Moreover, there is good evidence of a dose-response relationship between ozone exposure and perturbations in glucose homeostasis. For instance, at 0.25 ppm ozone exposure, no significant changes in glucose tolerance or fasting blood glucose were observed in any of the study groups. This was true regardless of rat strain, age, or exposure duration/frequency. At 0.5 ppm ozone, 7 out of 15 time points showed statistically significant impairment in glucose tolerance out of the three groups. In addition, while fasting blood glucose concentrations were not significantly affected at 0.5 ppm ozone, significant fasting hyperglycemia and greater impairment of glucose tolerance were observed in study groups exposed to 0.8 or 1.0 ppm ozone. Among these groups, greater than half demonstrated significant fasting hyperglycemia and/or ≥3 GTT time-points that were significantly affected.

From the eleven studies evaluated for glucose tolerance and fasting blood glucose, the greatest drivers affecting ozone-induced changes appear to be ozone exposure concentration and the timing of the GTT after exposure (i.e. whether a recovery period was allowed prior to the GTT). The substantial improvements in glucose homeostasis observed in animals allowed a recovery period after ozone exposure, even if only 18 h, suggest that the adverse effects of ozone may be transient and not long-lasting. However, the number of available studies with GTT data following a recovery period is extremely limited.

3.3.2.1.2. Insulin sensitivity

Two studies (Vella et al. 2015; Miller et al. 2016b) evaluated associations between ozone and effects on glucose homeostasis and insulin sensitivity as measured by fasting blood glucose concentration and the insulin tolerance test (ITT) (Table 8). In toxicologic studies, the ITT is used to assess whole-body sensitivity to insulin, specifically the ability of tissues to uptake glucose in response to insulin injection. K_ITT (% min⁻¹) is the rate constant for plasma glucose disappearance.

Table 8. Insulin tolerance test (ITT) results analysis after ozone exposure in rats (Suitable studies).

References	Animal model				O3 exposure					ITT^a (day)	Blood glucose post-insulin injection^b
	Species, strain	Sex	Age (month)	N (/grp)	Grp	Level (ppm)	Duration (h/day)	Frequency	Recovery		Min (minutes): 0	15	30	60	90	120	AUC	KITT
Miller et al. (2016b) (Acute)	WIS-KY rats	M	2.5	8–10	–	0.25	5	3 d/wk, 13 wks	–	1 wk–d 2	≈	–	↑	↑	≈	≈	↓	–
Vella et al. (2015) (Sub-chronic)	WIS rats	NR^c	NR^c	7	–	0.25	12	4 d	–	–	≈	–	↑	↑	↑	↑	–	↓
Miller et al. (2016b) (Sub-chronic)	WIS-KY rats	M	2.5	8–10	–	0.25	5	3 d/wk, 13 wks	–	12 wks–d 1	↓	–	≈	≈	↑	↑	↓	–
Vella et al. (2015) (Acute)	WIS rats	NR^c	NR^c	5	1	0.8	16	1 d	–	–	≈	↑	↑	↑	↑	↑	–	↓
				5	2		16	1 d	0 h	16 h–0 h	≈	↑	↑	↑	↑	↑	–	↓
				5	3		16	1 d	24 h	16 h–24 h	≈	↑	↑	↑	↑	↑	–	↓
				5	4		16	1 d	72 h	16 h–72 h	≈	↑	↑	↑	↑	↑	–	↓
				5	5		16	1 d	96 h	16 h–96 h	≈	↑	↑	↑	↑	↑	–	↓
Miller et al. (2016b) (Acute)	WIS-KY rats	M	2.5	8–10	–	1.0	5	3 d/wk, 13 wks	–	1 wk–d 2	↑	–	↑	↑	↑	↑	↑	–
Miller et al. (2016b) (Sub-chronic)	WIS-KY rats	M	2.5	8–10	–	1.0	5	3 d/wk, 13 wks	–	12 wks–d 1	↑	–	↑	↑	↑	↑	↑	–
											Plasma glucose decrease during ITT in response to O3 (%)
											O3 (ppm): 0		0.1	0.2	0.3	0.5	0.8
Vella et al. (2015) (Dose-response)	WIS rats	NR^c	NR^c	5–8	–	0–0.8	O/N	1 d	–	–		68	65	60	39*	33*	39*

↑ ↓: statistically significant results; ↑↓: increased or decreased results; ≈: approximately equal results; –: no data; AUC: area under the curve; d: day; K_ITT: glucose disappearance rate for an ITT; ITT: Insulin Tolerance Test; M: Male; NR: not reported; O/N: overnight; ppm: parts per million; wk: week; WIS: Wistar; WIS-KY: Wistar Kyoto.

*Statistically significant decrease compared with 0 ppm ozone.

^aUnless otherwise indicated, the ITT was performed immediately after ozone exposure.

^bThe two studies represented in this table, Vella et al. (2015) and Miller et al. (2016b), reported the ITT results differently. Vella et al. (2015) reported their ITT results as glycemia, % of baseline, thus all 0 min time-points were normalized to 100% regardless of ozone exposure concentration. In contrast, Miller et al. (2016b) reported their ITT results as blood glucose (mg/dL), thus the 0 min timepoints were reflective of the actual blood glucose concentration at that ozone exposure. Results are represented as increased, decreased, or approximately equal blood glucose concentrations in ozone-exposed groups compared with control (air)-exposed groups. ITT results were determined from graphical data. As such, the data symbols/lines were visually distinguished from each other; if they could not be distinguished, the results were considered approximately equal.

^cVella et al. (2015) did not report the age or sex of the rats; however, the authors reported that the rats were studied at 400–450 g bw, indicating these were adult males.

Between the two studies, a total of 11 study groups were evaluated, with ozone exposure concentrations ranging from 0.25 to 1.0 ppm. In addition, Vella et al. (2015) included a dose-response study group aimed at determining % plasma glucose decrease during the ITT. Among the study groups, the ITT was administered after short-term acute exposure (e.g. 5 h/d × 3 d/week) in eight groups and after sub-chronic exposure (5 h/d × 3 d/week) of 12 weeks in two groups. In addition, one study group (n = 7) in Vella et al. (2015) was used to evaluate ozone exposure at 12 h/d for 4 days (considered sub-chronic exposure by the authors).

Vella et al. (2015) and Miller et al. (2016b) evaluated ozone exposure at 0.25 ppm (Table 8). Baseline blood glucose concentrations (0 min ITT time-point) in 0.25 ppm ozone-exposed animals were approximately equal to or lower than control (air)-exposed animals, with no differences considered statistically significant. Results at the post-insulin ITT time-points (i.e. 30, 60, 90, and 120 min) were mixed.

Vella et al. (2015) evaluated a single short-term ozone exposure (16 h) at 0.8 ppm in five different study groups (n = 5 each). In this series of experiments, the ITT was administered immediately after exposure or after various assigned recovery periods ranging from 0 to 96 h. In all the study groups, baseline blood glucose concentrations were approximately equal to those of control (air)-exposed animals and none of the ITT glucose values were statistically increased. However, K_ITT values were significantly decreased, indicating decreased insulin sensitivity in the five study groups (Table 8).

In Miller et al. (2016b), short-term (5 h/d × 3 d/week for 1 week) and subchronic ozone exposure (5 h/d × 3 d/week for 12 weeks) at 1.0 ppm (n = 8–10 each group) resulted in significantly increased baseline blood glucose concentrations compared with control (air)-exposed animals, but the ITT showed mixed results with 3 out of 10 statistically significantly increased (Table 8).

To determine plasma glucose decreases (%) during the ITT, Vella et al. (2015) conducted a dose-response study in Wistar rats (n = 5–8) exposed overnight (16 h) to ozone at 0 to 0.8 ppm. Control animals (0 ppm) demonstrated a 68% decrease in plasma glucose in response to insulin injection during the ITT (Table 8). Animals exposed to 0.1 and 0.2 ppm ozone overnight demonstrated plasma glucose decreases (65 and 60%, respectively) during the ITT that were not significantly different from control (0 ppm) animals. However, animals exposed to 0.3, 0.5, and 0.8 ppm ozone overnight demonstrated plasma glucose % decreases during the ITT that were significantly lower (39, 33, and 39%, respectively) than those observed in control (0 ppm) animals, indicating significantly decreased insulin sensitivity in these animals.

Taken together, these results do not indicate a consistent association between ozone exposure and insulin sensitivity (Table 8). While one study (Vella et al. 2015) reported a consistent and significant decrease in the K_ITT rate constant across five study groups exposed to 0.8 ppm ozone, indicating that ozone exposure may adversely affect insulin sensitivity, additional reliable and reproducible data are needed to fully characterize this association. In addition, although data from the same study for the post-insulin ITT time-points appear to demonstrate a consistent trend in increased blood glucose concentrations at 0.8 ppm ozone, it is unclear whether statistical analyses were performed for these data, making a final determination of their significance difficult. Further, while the results from the dose-response study group in Vella et al. (2015) indicate that an adverse dose-response relationship between ozone exposure concentration and insulin sensitivity may exist, a single study group is not adequate evidence, thus more data are needed.

Therefore, for this endpoint, the strength of the toxicologic evidence for an association between ozone exposure and insulin sensitivity was categorized as weak. First, the evidence is based upon only two publications and a limited number of study groups (Table 14). In addition, inconsistencies across the two studies in reporting ITT results further limit any direct comparisons that can be made, particularly regarding baseline (0 min ITT time-points) blood glucose measurements. For example, Vella et al. (2015) reported the ITT results of their study as “glycemia, % of baseline,” indicating that all 0 min ITT time-points were normalized to 100% regardless of control or ozone exposure concentration. In contrast, Miller et al. (2016b) reported the ITT results of their study as “blood glucose (mg/dL)”; thus, the 0 min ITT time-points in this study were reflective of the actual blood glucose concentration in that group of animals at the given ozone concentration. Furthermore, no data are available for female animals.

3.3.2.1.3. Hepatic gluconeogenesis (GNG) and the pyruvate tolerance test (PTT)

Two studies (Vella et al. 2015; Miller et al. 2016b) evaluated associations between ozone and effects on glucose homeostasis and hepatic GNG as measured by fasting blood glucose concentration and the PTT (Table 9). A third study (Aibo et al. 2010) qualitatively evaluated histopathological staining as an indirect measure of hepatic GNG following ozone exposure. During the fasting state, GNG is an important hepatic process that prevents hypoglycemia by generating glucose from non-carbohydrate substrates, including pyruvate. In the PTT, pyruvate is injected to fasted animals and the resulting glycemic response is reflective of the level of hepatic GNG.

Table 9. Pyruvate tolerance test (PTT) results analysis after ozone exposure in rats (Suitable studies).

References	Animal model				O3 exposure				PTT^a (day)	Blood glucose post-pyruvate injection^b
	Species, strain	Sex	Age (month)	N (/grp)	Grp	Level (ppm)	Duration (h/day)	Frequency		Min (minutes): 0	15	30	60	90	120	AUC
Miller et al. (2016b) (Acute)	WIS-KY rats	M	2.5	8–10	–	0.25	5	3 d/wk, 13 wks	1 wk–d 1	≈	–	↑	↑	↑	≈	↓
Miller et al. (2016b) (Sub-chronic)	WIS-KY rats	M	2.5	8–10	1	0.25	5	3 d/wk, 13 wks	13 wks–d 1	≈	–	≈	↓	↓	↓	↓
					2		5	3 d/wk, 13 wks	13 wks–d 3	≈	–	≈	↓	↓	↓	≈
Vella et al. (2015)	WIS rats	NR^c	NR^c	6	–	0.8	16	1 d	–	≈	↓	↓	↓	↓	↓	↓
Miller et al. (2016b) (Acute)	WIS-KY rats	M	2.5	8–10	–	1.0	5	3 d/wk, 13 wks	1 wk–d 1	↑	–	↑	↑	↑	↑	↑
Miller et al. (2016b) (Sub-chronic)	WIS-KY rats	M	2.5	8–10	1	1.0	5	3 d/wk, 13 wks	13 wks–d 1	↑	–	↑	↑	↑	↑	↑
					2		5	3 d/wk, 13 wks	13 wks–d 3	≈	–	≈	↓	↓	↓	↓

↑ ↓: statistically significant results; ↑↓: increased or decreased results; ≈: approximately equal results; –: no data; AUC: area under the curve; d: day; grp: group; M: male; NR: not reported; ppm: parts per million; PTT: Pyruvate Tolerance Test; WIS: Wistar; WIS-KY: Wistar Kyoto; wk: week.

^aUnless otherwise indicated, the PTT was performed immediately after ozone exposure.

^bResults are represented as increased, decreased, or approximately equal blood glucose concentrations in ozone-exposed groups compared with control (air)-exposed groups. PTT results were determined from graphical data. As such, the data symbols/lines were visually distinguished from each other; if they could not be distinguished, the results were considered approximately equal.

^cVella et al. (2015) did not report the age or sex of the rats; however, the authors reported that the rats were studied at 400–450 g bw, indicating these were adult males.

Across the two studies, a total of seven study groups were evaluated, with ozone exposure concentrations ranging from 0.25 to 1.0 ppm. Among the study groups, the PTT was administered after short-term acute exposure (i.e. ≤2 weeks) or repeated, sub-chronic exposure (5 h/d × 3 d/week for 13 weeks).

Miller et al. (2016b) evaluated ozone exposure at 0.25 ppm. For all study groups, baseline blood glucose concentrations (0 min PTT time-point) in ozone-exposed animals were approximately equal to those of control (air)-exposed animals. Results at the post-pyruvate PTT time-points (i.e. 30, 60, 90, and 120 min) and the AUC were mixed, with none statistically significant. Vella et al. (2015) showed no significant effects on baseline blood glucose or hepatic GNG following a single (16 h) exposure to 0.8 ppm ozone. Miller et al. (2016b) also examined short-term and subchronic ozone exposure at 1.0 ppm using three study groups (n = 8–10). Short-term ozone exposure (5 h/d × 3 d/week for 1 week) resulted in significantly increased baseline blood glucose concentrations compared with control (air)-exposed animals. In addition, hepatic GNG was significantly induced at all post-pyruvate PTT time-points and the AUC was significantly increased. However, the results were mixed for animals subjected to subchronic ozone exposure (5 h/d × 3 d/week for 13 weeks). One study group (d1 of week 13) demonstrated similar results to the short-term exposure group and the other (d3 of week 13) demonstrated baseline and post-pyruvate blood glucose concentrations that were approximately equal to control (air)-exposed animals, possibly indicating a rapid adaptive response to ozone exposure. In the third study identified, Aibo et al. (2010) qualitatively evaluated liver glycogen staining in mice after ozone exposure (0, 0.25, 0.5 ppm ozone for 6 h) and found no changes.

Overall, the results do not indicate a consistent association between ozone exposure and hepatic GNG (Table 9). Thus, the strength of the toxicologic evidence was categorized as weak. First, the evidence is based on only two publications and a limited number of study groups. While the use of similar animal models and consistent methodology in administration of the PTT allow for direct comparisons, the results are inconsistent, and no data are available for female animals. In addition, while a dose-response relationship may be present (e.g. significant changes in hepatic GNG were observed at 1.0 ppm ozone compared with 0.25 and 0.8 ppm ozone), the data are limited.

3.3.2.1.4. Serum/plasma hormone analytes related to glucose homeostasis

Seventeen studies evaluated certain glucose-related hormonal serum/plasma analytes, including glucagon, insulin, leptin, and corticosterone (CORT), across 47 study groups (Table 10) (Martrette et al. 2011; Bass et al. 2013; Gordon et al. 2013, 2016b, 2017a, 2017b; Miller et al. 2015, 2016b, 2016c, 2017; Vella et al. 2015; Thomson et al. 2016, 2018; Mathews et al. 2017; Henriquez et al. 2018, 2019, 2020; Snow et al. 2019). Forty-six of the study groups used various rat strains as the animal model (Brown Norway, Fischer 344, Long-Evans, Wistar, Wistar-Kyoto) and one used a mouse model (C57BL/6J). Eleven study groups were comprised of female animals (two with pregnant rats; Miller et al. 2017), while 36 were comprised of male animals. The data cover a range of ozone exposure levels (from 0.12 to 2.0 ppm ozone), durations (3–16 h), and frequencies (single acute exposures to 1–3 d/week for up to 17 weeks).

Table 10. Animal studies: blood/serum/plasma hormone analytes after ozone inhalation exposure (Suitable studies).

References	Animal model				O3 exposure				Outcomes
	Rat strain	Sex	Age (months)	N (per grp)	Level (ppm)	Duration (h/day)	Frequency	Sampling time/recovery	Baseline glucose	Time 0 glucose			Gluc-agon	Insulin	Lep-tin	Cortico-sterone
										GTT	PTT	ITT
Vella et al. (2015)	WIS	NR^a	NR^a	4–6	0.1	16	1 d	–	≈	–	–	–	–	–	–	–
Martrette et al. (2011)	WIS	F	1.5	12	0.12	6	15 d	–	–	–	–	–	–	–	–	↑ (s)
Vella et al. (2015)	WIS	NR^a	NR^a	4–6	0.2	16	1 d	–	↑	–	–	–	–	–	–	–
Gordon et al. (2017b)	L-E	F	3	10	0.25	5	2 d	–	–	≈	–	–	–	≈ (s)	↑ (s)	–
Miller et al. (2015)	WIS-KY	M	2.5	6	0.25	6	2 d	1 d–0 h	–	≈	–	–	–	–	–	–
						6	2 d	2 d–0 h	–	≈	–	–	–	–	–	–
Bass et al. (2013) (Acute)	BN	M	1	4–10	0.25	6	2 d	–	–	≈	–	–	–	–	–	–
			4	4–10		6	2 d	–	–	≈	–	–	–	–	–	–
			12	4–10		6	2 d	–	–	≈	–	–	–	–	–	–
			24	4–10		6	2 d	–	–	≈	–	–	–	–	–	–
Miller et al. (2016b) (Acute)	WIS-KY	M	2.5	8–10	0.25	5	3 d/wk, 13 wk	1 wk–d 1/2/3	–	≈/≈	≈	≈	–	–	–	–
Vella et al. (2015)	WIS	NR^a	NR^a	4–6	0.25	12	4 d	–	–	–	–	≈	–	–	–	–
Bass et al. (2013) (Subchronic)	BN	M	1^b	4–10	0.25	6	2 d/wk, 13 wk	–	–	≈	–	–	–	–	–	–
			9^b	4–10		6	2 d/wk, 13 wk	–	–	≈	–	–	–	–	–	–
			21^b	4–10		6	2 d/wk, 13 wk	–	–	≈	–	–	–	–	–	–
Miller et al. (2016b) (Subchronic)	WIS-KY	M	2.5	8–10	0.25	5	3 d/wk, 12 wk	12 wk–d 1	–	≈	–	↓	–	↑ (s)
						5	3 d/wk, 13 wk	13 wk–d 1/3	–	–	≈/≈	–	–	↑ (s)	↑ (s)	↓ (p)^c
				8–10		5	3 d/wk, 13 wk	1 wk	–	≈	–	–	–	↑ (s)	↓ (s)	↓ (p)^c
Vella et al. (2015)	WIS	NR^a	NR^a	4–6	0.4	16	1 d	–	↑	–	–	–	–	–	–	–
Gordon et al. (2017b)	L-E	F	3	10	0.5	5	2 d	–	–	≈	–	–	–	↑ (s)	↑ (s)	–
Miller et al. (2015)	WIS-KY	M	2.5	6	0.5	6	2 d	1 d–0 h	–	↑	–	–	–	–	–	–
						6	2 d	2 d–0 h	–	↑	–	–
Vella et al. (2015)	WIS	NR^a	NR^a	4–6	0.6	16	1 d	–	↑	–	–	–	–	–	–	–
Thomson et al. (2016)	F344	M	NR^d	5	0.8	4	1 d	–	–	–	–	–	↓^e (p)	↓ (p)	↑ (p)	↑ (p)
Thomson et al. (2018)	F344	M	NR^d	8	0.8	4	1 d	–		≈	–	–	↓^e (p)	≈ (p)	≈ (p)	↑ (p)
Henriquez et al. (2018)	WIS-KY	M	4–5	6–8	0.8	4	1 d	–	–	–	–	–	–	–	–	≈ (p)
Gordon et al. (2016b) (Acute)	BN	M	5	10	0.8	5	1 d	–	↓	–			–	NS (s)	–	–
		F	5	10	0.8	5	1 d	–	↓	–			–	NS (s)	–	–
Henriquez et al. (2018)	WIS-KY	M	4–5	6–8	0.8	4	1 d	–	–	–	–	–	–	↑ (s)	–	–
							2 d	–	–	–	–	–	–	≈ (s)	–	–
Henriquez et al. (2019)	WIS-KY	M	2.5–3	6–8	0.8	4	2 d–S 1	–	–	↑	–	–	–	–	–	↑ (p)
							2 d–S 2	–	–	↑	–	–	–	–	–	↓ (p)
Henriquez et al. (2020)	WIS-KY	M	2.5–3	6–8	0.8	4	2 d–S 1	–	–	↑	–	–	–	↓ (s)	–	↑ (p)
							2 d–S 2	–	–	↑	–	–	–	↓ (s)	–	≈ (p)
Gordon et al. (2017a)	L-E	M	5	4	0.8	4	2 d	–	–	↑	–	–	–	NS (s)	–	NS (s)
Gordon et al. (2017a)	L-E	F	5	4	0.8	4	2 d	–	–	≈	–	–	–	NS (s)	–	NS (s)
Snow et al. (2019)	L-E	M	3	10	0.8	5	2 d	–	–	↑	–	–	–	–	–	–
		F	3	10	0.8	5	2 d	–	–	↑	–	–	–	–	–	–
Vella et al. (2015)	WIS	NR^a	NR^a	4–6	0.8	16	1 d	–	↑	↑	≈	≈^f	–	↑ (p)	–	↑ (p)
						16	1 d	–	↑	–	–	–	–	–	–	–
						16	1 d	0 h	–	–	–	≈ ^f
						16	1 d	24 h	–	–	–	≈ ^f
						16	1 d	72 h	–	–	–	≈ ^f
						16	1 d	96 h	–	–	–	≈ ^f
Gordon et al. (2016b) (Acute)	BN	M	5	10	0.8	5	1 d	–	↓	–	–	–
		F	5	10	0.8	5	1 d	–	≈	–	–	–
Gordon et al. (2016b) (Subacute)	BN	M	5	10	0.8	5	1 d/wk, 4 wk	–	↑	–	–	–	–	NS (s)	–	–
		F	5	10	0.8	5	1 d/wk, 4 wk	–	≈	–	–	–	–	NS (s)	–	–
Gordon et al. (2013)	BN	M	4	7–10	0.8	6	1 d/wk, 17 wk	–	–	–	–	–	–	≈ (s)	≈ (s)	–
Gordon et al. (2013)	BN	M	20	7–10	0.8	6	1 d/wk, 17 wk	–	–	–	–	–	–	↑ (s)	≈ (s)	–
Miller et al. (2016c)	WIS-KY	M	1.5–2	4–6	1.0	4	1 d	–	–	↑	–	–	–	–	–	↑ (s)
				4–6		4	2 d	–	–	–	–	–	–	–	–	↓ (s)
Gordon et al. (2017b)	L-E	F	3	10	1.0	5	2 d	–	–	↑	–	–	–	↑ (s)	↑ (s)	–
Bass et al. (2013) (Time-course)	BN	M	4	6	1.0	6	1 d	–	≈	↑	–	–	–	↓ (s)	↑ (s)	–
				6		6	2 d	–	≈	≈	–	–	–	↓ (s)	≈ (s)	–
				6		6	2 d	18 h	≈	≈	–	–	–	≈ (s)	↓ (s)	–
Miller et al. (2015) (Time-course)	WIS-KY	M	2.5	8	1.0	6	1 d	0 h	–	↑	–	–	–	↑ (s)	↑ (s)	–
				8		6	2 d	0 h	–	↑	–	–	–	↓ (s)	↑ (s)	–
				8		6	2 d	18 h	–	↓	–	–	–	↑ (s)	≈ (s)	–
Bass et al. (2013) (Acute)	BN	M	1	4–12	1.0	6	2 d	18 h	↑	↑	–	–	↑ (s)	↓ (s)	–	–
			4	4–12		6	2 d	18 h	↑	↑	–	–	≈ (s)	↓ (s)	–	–
			12	4–12		6	2 d	18 h	↑	↑	–	–	↑ (s)	↑ (s)	–	–
			24	4–12		6	2 d	18 h	↑	↑	–	–	≈ (s)	↑ (s)	–	–
Miller et al. (2015) (Conc/Resp)	WIS-KY	M	2.5	6	1.0	6	2 d	1 d–0 h	–	↑	–	–	–	–	–	–
						6	2 d	2 d–0 h	–	↑	–	–	–	–	–	–
Miller et al. (2016b) (Acute)	WIS-KY	M	2.5	8–10	1.0	5	3 d/wk, 13 wk	1 wk–d 1/2/3	–	↑/↑	↑	↑	–	≈ (s)	–	–
Bass et al. (2013) (Subchronic)	BN	M	1^b	4–10	1.0	6	2 d/wk, 13 wk	–	↑	↑	–	–	≈ (s)	↓ (s)	–	–
			9^b	4–10		6	2 d/wk, 13 wk	–	≈	≈	–	–	↑ (s)	↓ (s)	–	–
			21^b	4–10		6	2 d/wk, 13 wk	–	≈	≈	–	–	↑ (s)	↓ (s)	–	–
Miller et al. (2016b) (Subchronic)	WIS-KY	M	2.5	8–10	1.0	5	3 d/wk, 13 wk	12 wk–d 1	–	↑	–	↑	–	↓ (s)	–	–
						5	3 d/wk, 13 wk	13 wk–d 1/3	–	–	↑/≈	–		↓ (s)	↓ (s)	↓ (p)^c
						5	3 d/wk, 13 wk	1 wk	–	≈	–	–	–	↓ (s)	↓ (s)	↑ (p)^c
Mathews et al. (2017)	C57BL/6J mice	F	2.5	5–8	2.0	3	1 d	24 h	–	–	–	–	–	≈ (s)	–	–
Study conducted in pregnant rats
Miller et al. (2017)	L-E	F	NR^g	9	0.4	4	GD 5 and 6	GD21^d	↓	–	–	–	–	↑ (p)	–	↑ (s)
				10	0.8	4	GD 5 and 6	GD21^d	↓	–	–	–	–	↑ (p)	–	↑ (s)

↑ ↓: statistically significant results; ↑↓: increased or decreased results, not statistically different; ≈: results of treated approximately equal to control based on visual assessment for graphic data or <5% difference for numeric data, and not statistically different; –: no data; BN: Brown Norway; Conc/Resp: concentration–response; d: day; F: female; F334: Fischer 344; GD: gestational day; h: hour; L-E: Long-Evans; M: male; NR: not reported; NS: not significant; (p): plasma; ppm: parts per million; (s): serum; SD: Sprague–Dawley; WIS-KY: Wistar Kyoto; wk: week.

^aVella et al. (2015) did not report the age or sex of the rats; however, the authors reported that the rats were studied at 400–450 g bw, indicating these were adult males.

^bIn Bass et al. (2013), the data for the sub-chronically exposed animals are reported as post-exposure ages: 4, 12, and 24 months.

^cIn Miller et al. (2016b), the authors reported that corticosterone were measured in plasma in section 3.5 and Figure 6; however, section 2.10 states serum corticosterone was measured.

^dThomson et al. (2016, 2018) reported that the rats were studied at 200–250 g bw.

^eThomson et al. (2016, 2018) also measured glucagon-like peptide-1 (GLP-1) and ghrelin levels. Those levels were not significantly changed after O₃ exposure, except a statistically significant decrease in ghrelin levels was observed in Thomson et al. (2016).

^fVella et al. (2015) reported their ITT results as glycemia, % of baseline, thus all 0 min time-points were normalized to 100% regardless of ozone exposure concentration.

^gMiller et al. (2017): time-mated female Long-Evans rats; no age information provided; outcome data collected on GD21, which was 15 d following the 2-d O₃ exposure.

Close review of the results (Table 10) indicates no obvious patterns of change for most of the hormonal data, with no clear associations with ozone exposure concentration or acute vs. repeated exposure, duration, or frequency. Glucagon showed both increases and decreases, along with roughly equivalent values and non-significant results. While there were some statistically significant changes in leptin values, there was a similar pattern of both increases and decreases and several data points were roughly equivalent to controls. Acute exposure to ozone at 0.8 ppm ozone demonstrated some increased CORT values (both statistically significant and non-significant) compared with controls; however, the expected similar effect following acute and repeated exposures to 1.0 ppm ozone was not present. In addition, two studies by Henriquez et al. (2019, 2020) reported CORT data from four study groups of identical ozone exposure; however, the results were inconsistent. Two study groups demonstrated increased (non-significantly) CORT values, one study group had decreased (non-statistically) CORT values, and one study group had CORT values approximately equal to control (air)-exposed animals. For insulin, there appeared to be a trend for increasing levels, although non-significantly, following 13 weeks of exposure to 0.25 ppm ozone. In contrast, insulin levels were decreased, although not statistically significantly, following 13 weeks of exposure to 1.0 ppm ozone. However, since none of these changes were statistically significant compared with control (air)-exposed animals, it is difficult to draw any conclusions on how insulin levels may be affected by acute or repeated exposure to ozone.

There were no obvious patterns or associations related to changes in serum/plasma values for glucagon, insulin, leptin, or CORT in response to acute or repeated exposure to ozone at concentrations ranging from 0.12 to 2.0 ppm, and for up to 17 weeks of repeated exposures. The possible pattern for insulin is not sufficiently supported with the available data to have confidence in its potential association with ozone exposure. Therefore, the strength of this body of evidence was categorized as weak.

3.3.2.1.5. Glucose-related outcomes summary

Overall, the strength of the evidence was categorized as strong for the outcome related to glucose intolerance, including the related outcome of fasting/baseline hyperglycemia (Tables 7 and 14). For the remaining glucose-related outcomes, including insulin resistance, hepatic GNG, and related serum hormone analytes, the strength of the evidence was categorized as weak (Tables 8–10 and 14).

3.3.2.2. Dyslipidemia

Fifteen publications were identified that reported serum lipid values in mammalian animal models following inhalation exposure to ozone, comprising 68 study groups (Table 11). The studies reported on six different rat strains (BN, F344, L-E, SD, WIS, and WIS-KY) (Mole et al. 1985; Bass et al. 2013; Miller et al. 2015, 2016b, 2016c; Ramot et al. 2015; Vella et al. 2015; Thomson et al. 2016, 2018; Gordon et al. 2016b, 2017a, 2017b; Snow et al. 2019) and one guinea pig strain (Hartley) (Vaughan et al. 1984). One study used pregnant Long-Evans rats (Miller et al. 2017). Nine of the rat study groups were female and two of the guinea pig study groups were female; all the remaining study groups were comprised of males. Several datasets included dose-response data, ranging from 0.25 ppm ozone up to 3.0 ppm ozone, with the exposure duration varying from a single 4-h exposure up to 22 h/d. Several studies included repeated exposures, from 1 d/week for 4 weeks to 3 d/week for 13 weeks.

Table 11. Animal studies: serum lipids results after ozone inhalation exposures (Suitable studies).

References	Animal model				O3 Exposure				Serum lipids
	Species, strain	Sex	Age (months)	N	Level (ppm)	Duration (h/day)	Frequency	Sampling time/recovery	CHOL	TG	LDL	HDL
Ramot et al. (2015)	WIS-KY rats	M	2.5–3	8	0.25	4	1 d	0 h	≈	–	↓	↓
	WIS rats	M	2.5–3	8	0.25	4	1 d	0 h	≈	–	≈	≈
	SD rats	M	2.5–3	8	0.25	4	1 d	0 h	↓	–	≈	≈
	WIS-KY rats	M	2.5–3	8	0.25	4	1 d	20 h	≈	–	≈	↓
	WIS rats	M	2.5–3	8	0.25	4	1 d	20 h	↑	–	≈	≈
	SD rats	M	2.5–3	8	0.25	4	1 d	20 h	↓	–	↓	≈
Gordon et al. (2017b)	L-E rats	F	3	10	0.25	5	2 d	–	NS	≈	≈	NS
Bass et al. (2013) (Acute)	BN rats	M	1	5–12	0.25	6	2 d	18 h	≈	↓	≈	≈
			4	5–12	0.25	6	2 d	18 h	↑	↑	↑	≈
			12	5–12	0.25	6	2 d	18 h	↑	↓	↑	↑
			24	5–12	0.25	6	2 d	18 h	≈	↓	≈	≈
Bass et al. (2013) (Sub-chronic)	BN rats	M	1^#	5–12	0.25	6	2 d/wk, 13 wks	18 h	≈	≈	≈	↑
			9^#	5–12	0.25	6	2 d/wk, 13 wks	18 h	≈	↓	↓	↑
			21^#	5–12	0.25	6	2 d/wk, 13 wks	18 h	↓	↓	↓	↓
Miller et al. (2016b)	WIS-KY rats	M	6	8–10	0.25	5	3 d/wk, 13 wks	–	↑	↓	↑	≈
			6	8–10	0.25	5	3 d/wk, 13 wks	1 wk	↓	↓	↓	↓
Ramot et al. (2015)	WIS-KY rats	M	2.5–3	8	0.5	4	1 d	0 h	↑	–	↓	↑
	WIS rats	M	2.5–3	8	0.5	4	1 d	0 h	↑	–	≈	↑
	SD rats	M	2.5–3	8	0.5	4	1 d	0 h	≈	–	↓	≈
	WIS-KY rats	M	2.5–3	8	0.5	4	1 d	20 h	↑	–	↑	≈
	WIS rats	M	2.5–3	8	0.5	4	1 d	20 h	↑	–	≈	≈
	SD rats	M	2.5–3	8	0.5	4	1 d	20 h	≈	–	↑	↑
Gordon et al. (2017b)	L-E rats	F	3	10	0.5	5	2 d	–	NS	≈	↓	NS
Thomson et al. (2016)	F344 rats	M	NR^a	5	0.8	4	1 d	–	NS	–	NS	NS
Thomson et al. (2018)	F344 rats	M	NR^a	8	0.8	4	1 d	–	–	↑	–	–
Gordon et al. (2016b) (Acute)	BN rats	M	5	10	0.8	5	1 d	18 h	↓	↑	≈	↑
		F	5	10	0.8	5	1 d	18 h	↑	↑	↓	↑
Gordon et al. (2017a)	L-E rats	M	5	10	0.8	4	2 d	–	NS	NS	NS	NS
		F	5	10	0.8	4	2 d	–	NS	NS	NS	NS
Snow et al. (2019)	L-E rats	M	3.5	10	0.8	5	2 d	–	≈	≈	≈	≈
		F	3.5	10	0.8	5	2 d	–	≈	≈	≈	↑
Vella et al. (2015)	WIS rats	NR^b	NR^b	12	0.8	16	1 d	–	↑	≈	–	–
Gordon et al. (2016b) (Sub-acute)	BN rats	M	6	10	0.8	5	1 d/wk, 4 wks	18 h	↓	↓	≈	≈
		F	6	10	0.8	5	1 d/wk, 4 wks	18 h	↑	≈	≈	≈
Ramot et al. (2015)	WIS-KY rats	M	2.5–3	8	1.0	4	1 d	0 h	↑	–	↑	≈
	WIS rats	M	2.5–3	8	1.0	4	1 d	0 h	↑	–	↑	↑
	SD rats	M	2.5–3	8	1.0	4	1 d	0 h	↑	–	↑	≈
	WIS-KY rats	M	2.5–3	8	1.0	4	1 d	20 h	↑	–	↑	↑
	WIS rats	M	2.5–3	8	1.0	4	1 d	20 h	↑	–	↓	≈
	SD rats	M	2.5–3	8	1.0	4	1 d	20 h	↓	–	≈	≈
Miller et al. (2016c)	WIS-KY rats	M	1.5–2	4–6	1.0	4	1 d	–	≈	↑	–	–
Miller et al. (2015)	WIS-KY rats	M	2.5	6–8	1.0	6	1 d	0 h	↑	–	↓	↑
Miller et al. (2016c)	WIS-KY rats	M	1.5–2	4–6	1.0	4	2 d	–	↑	↓	–	–
Gordon et al. (2017b)	L-E rats	F	3	10	1.0	5	2 d	–	NS	↑	≈	NS
Miller et al. (2015)	WIS-KY rats	M	2.5	6–8	1.0	6	2 d	0 h	↑	–	↑	↑
		M	2.5	6–8	1.0	6	2 d	18 h	↑	–	↑	↑
Bass et al. (2013) (Time Course)	BN rats	M	4	6	1.0	6	1 d	–	NS	NS	NS	NS
			4	6	1.0	6	2 d	–	NS	NS	NS	NS
			4	6	1.0	6	2 d	18 h	NS	NS	NS	NS
Bass et al. (2013) (Acute)	BN rats	M	1	5–12	1.0	6	2 d	18 h	↑	≈	≈	↓
			4	5–12	1.0	6	2 d	18 h	↑	↓	≈	≈
			12	5–12	1.0	6	2 d	18 h	≈	≈	↓	↑
			24	5–12	1.0	6	2 d	18 h	≈	↑	↓	≈
Mole et al. (1985) (Conc/Resp)	SD rats	M	2.5–3	9	1.0	5	10 d over 14 d	20 h	≈^c	≈	–	↓^c
Mole et al. (1985) (Sampling time study)	SD rats	M	2.5–3	4–5	1.0	5	15 d over 19 d	0–44 h (pooled)	NS^c	NS	–	↑^c,d
Bass et al. (2013) (Sub-chronic)	BN rats	M	1^#	5–12	1.0	6	2 d/wk, 13 wks	18 h	≈	↓	≈	≈
			9^#	5–12	1.0	6	2 d/wk, 13 wks	18 h	↑	↓	↓	↑
			21^#	5–12	1.0	6	2 d/wk, 13 wks	18 h	≈	↓	≈	≈
Miller et al. (2016b)	WIS-KY rats	M	6	8–10	1.0	5	3 d/wk, 13 wks	–	↑	↓	↓	↑
		M	6	8–10	1.0	5	3 d/wk, 13 wks	1 wk	↓	↓	↓	↓
Vaughan et al. (1984)	Hartley guinea pigs	M^e	3–5	18–36	1.0	22	14 d	–	↑	↓	↑	–
			3–5	18–36			14 d	30 d	↓^e	↑^e	↓^e	–
		F^e	3–5	18–39	1.0	22	14 d	–	↑	↑	↑	–
			3–5	18–39			14 d	30 d	↓^e	↓^e	↓^e	–
Mole et al. (1985)	SD rats	M	2.5–3	8	1.75	5	10 d over 14 d	20 h	↑^c	↓	–	≈^c
Mole et al. (1985)	SD rats	M	3.5–4	9	3	5	10 d over 14 d	20 h	↑^c,f	↓	–	↑^c
Study conducted in pregnant rats
Miller et al. (2017)	L-E rats	F	NR^g	9–10	0.4	4	GD 5 and 6	GD21^g	≈	↓	≈	↓
				9–10	0.8	4	GD 5 and 6	GD21^g	≈	↓	≈	≈

↑ ↓: statistically significant results; ↑↓: increased or decreased results; ≈: results of treated approximately equal to control based on visual assessment for graphic data or <5% difference for numeric data, and not statistically different; –: no data; BN: Brown Norway; CHOL: cholesterol; Conc/Resp: concentration–response; d: day; F: female; F344: Fischer 344; GD: gestational day; h: hour; HDL: high-density lipoprotein; LDL: low-density lipoprotein; L-E: Long-Evans; M: male; NR: not reported; NS: not significant; ppm: parts per million; SD: Sprague–Dawley; WIS-KY: Wistar Kyoto; TG: triglycerides; wk: week.

^#Bass et al. 2013 discussed these groups based on age at end exposure, 3 months older than age at start exposure, which is shown here.

^aThomson et al. (2016, 2018) reported that the rats were studied at 200–250 g bw.

^bVella et al. (2015) did not report the age or sex of the rats; however, the authors reported that the rats were studied at 400–450 g bw, indicating these were adult males.

^cIn Mole et al. (1985), both free and esterified components of cholesterol and HDL were measured.

^dIn Mole et al. (1985), total HDL was significantly increased post-exposure, but not free HDL.

^e30-d post-end exposure values compared to values at end-exposure; thus, these 30-d values (30-d post-end of 14-d exposure) are not compared to controls but to the corresponding O₃-treated value at end of 14-d exposure.

^fIn Mole et al. (1985), free cholesterol was significantly increased post-exposure, but not esterified cholesterol.

^gMiller et al. (2017): time-mated female Long-Evans rats, with no age information provided; exposure on GD5 and GD6 but outcome data collected on GD21.

In analyzing the results on serum lipids, it is important to recognize that while increases in cholesterol (CHOL), triglycerides (TG), and low-density lipoprotein (LDL) are considered detrimental, increases in high density lipoprotein (HDL) are typically considered beneficial. Thus, it is important to not merely consider the directionality (i.e. the “up” or “down”) of the results, but rather whether the direction correlates with a health benefit.

Close review of the available results did not reveal any clear patterns to demonstrate potential consistent effects of ozone inhalation exposure, either acute or repeated, on serum lipids in exposed rats or guinea pigs compared with controls. For example, while there were more increases in CHOL values (11 out of 19, but only two study groups with statistically significant increases) than other responses (e.g. decreases or roughly equivalent) in rats acutely exposed to 1.0 ppm ozone, this possible pattern of ozone-induced increased CHOL was contradicted by the results for repeated exposures to 1.0 ppm ozone, where only two CHOL values (only one statistically significant) showed increases. Similar mixed results were found for the three study groups of rats exposed to 1.0 ppm ozone in a repeated exposure paradigm, where the number of roughly equivalent CHOL values (3) matched the number of increased CHOL values (3; one statistically increased). In addition, following a 1-week recovery, CHOL values were reported to be statistically significantly decreased for one study group at 1.0 ppm in the repeated exposure group. Similarly, neither TG, LDL, nor HDL showed any clear pattern in terms of changes, whether up or down, following ozone exposure.

Overall, these study results did not indicate any reproducible pattern of significant association between inhalation ozone exposure and effects on the serum lipids studied. These findings indicate no consistent or coherent evidence of increased CHOL, TG, or LDL values following inhalation exposure to ozone. Similarly, no consistent or coherent evidence was found for decreased HDL values following ozone exposure. With no consistent pattern in changes demonstrated for any of the serum lipids analyzed, the strength of the toxicologic evidence was categorized as weak; it provides no evidence for ozone-induced dyslipidemia in rats or guinea pigs (Table 14).

3.3.2.3. Effects on body weight, food consumption, body temperature, and thyroid hormone homeostasis

It is well-understood that thyroid hormones regulate body temperature. In addition, it is known that reduced body temperature is associated with decreased metabolic function, which can lead to bodyweight increases and obesity.

Sixteen publications evaluated potential associations between inhalation exposure to ozone and effects on body weight (BW; occasionally expressed as body mass), body composition (BC), and food consumption (FC) (Table 12). Not all studies evaluated all three parameters, but every study assessed at least one. In the 16 studies, rats, mice, and guinea pigs of various strains were used as the animal model, including the following rat strains: Brown Norway (Gordon et al. 2013, 2016b), Long-Evans (Miller et al. 2017), Wistar (Martrette et al. 2011; Cestonaro et al. 2017), Wistar-Kyoto (Miller et al. 2016c; Henriquez et al. 2018, 2019, 2020), Sprague-Dawley (Mole et al. 1985; Gordon et al. 2016a), and Fisher 344 (Watkinson et al. 1995). The mouse strains used were ICR (Umezu et al. 1993), CD-1 (Graham et al. 1982), and C57BL/6J (Johnston et al. 2006). Hartley guinea pigs were used in one study (Vaughan et al. 1984).

Table 12. Animal studies: outcomes of body weight, body composition, and food consumption after ozone inhalation exposure (Suitable studies).

References	Animal model				O3 exposure			Outcomes
	Species, strain	Sex	Age (months)	N (/grp)	Level (ppm)	Duration (h/day)	Frequency	BW	BC	FC
Cestonaro et al. (2017) (Acute)	WIS rats	M	2.5	6	0.05	3	14 d	NS	–	NS
						24	14 d	NS	–	NS
Cestonaro et al. (2017) (Sub-acute)	WIS rats	M	2.5	6	0.05	3	28 d	NS	–	NS
						24	28 d	NS	–	NS
Umezu et al. (1993)	ICR mice	M	2–6.5	6	0.1	22	7 d	≈^a	–	≈^a
Martrette et al. (2011)	WIS rats	F	1.5	12	0.12	6	15 d	≈	–	↓
Gordon et al. (2016a)	SD rats	F	6	7–10	0.25	5	1 d/wk, 6 wks	≈	≈/≈/≈^b	–
Umezu et al. (1993)	ICR mice	M	2–6.5	6	0.2	22	7 d	↓≈^a	–	≈^a
Umezu et al. (1993)	ICR mice	M	2–6.5	6	0.4	22	7 d	↓≈^a	–	↓≈^a
Watkinson et al. (1995)	F344 rats	M	3–4	4–6	0.5	6	5 d; 7 d recovery	≈	–	–
Gordon et al. (2016a)	SD rats	F	6	7–10	0.5	5	1 d/wk, 6 wks	≈	≈/≈/≈^b	–
Watkinson et al. (1995)	F344 rats	M	3–4	4–6	0.5	23	5 d; 7 d recovery	↓	–	–
Henriquez et al. (2018)	WIS-KY rats	M	3.5–4	8	0.8	4	1 d	≈	–	–
							2 d	↓	–	–
Henriquez et al. (2019) S1	WIS-KY rats	M	2.5–3	6–8	0.8	4	2 d	≈	–	–
Henriquez et al. (2019) S2	WIS-KY rats	M	2.5–3	6–8	0.8	4	2 d	≈	–	–
Henriquez et al. (2020) S1	WIS-KY rats	M	2.5–3	6–8	0.8	4	2 d	≈	–	–
Henriquez et al. (2020) S2	WIS-KY rats	M	2.5–3	6–8	0.8	4	2 d	≈	–	–
Gordon et al. (2016b)	BN rats	M	5	9–10	0.8	5	1 d/wk, 4 wks	–	↓/↑/↓^b	↑
		F	5	9–10	0.8	5	1 d/wk, 4 wks	–	≈/↓/↑^b	≈
Gordon et al. (2013)	BN rats	M	4	8–10	0.8	6	1 d/wk, 17 wks	NS	↓/≈/≈^b	–
		M	20	8–10	0.8	6	1 d/wk, 17 wks	NS	≈/≈/↑^b	–
Umezu et al. (1993)	ICR mice	M	2–6.5	6	0.8	22	7 d	↓^a	–	↓≈^a
Miller et al. (2016c)	WIS-KY rats	M	1.5–2	4–6	1.0	4	2 d	↓	–	–
Graham et al. (1982)	CD-1 mice	F	1.5	20	1.0	5	1 d	↑^c	–	–
Gordon et al. (2016a)	SD rats	F	6	7–10	1.0	5	1 d/wk, 6 wks	↓	↓/↑/≈^b	–
Mole et al. (1985)	SD rats	M	2.5–3	9	1.0	5	10 d over 14 d	≈	–	↑
Vaughan et al. (1984)	Hartley guinea pigs	M	3–5	3	1.0	22	14 d	↓	–	↓
			3–5	10	1.0	22	14 d	↓	–	–
		F	3–5	5	1.0	22	14 d	≈	–	↓
			3–5	10	1.0	22	14 d	↓	–	–
Mole et al. (1985)	SD rats	M	2.5–3	8	1.75	5	10 d over 14 d	↓	–	↑
Johnston et al. (2006)	C57BL/6J mice	B	3.5	5–7	2.0	3	1 d	≈	–	–
Mole et al. (1985)	SD rats	M	2.5–3	9	3.0	5	10 d over 14 d	↓	–	↑
Study conducted in pregnant rats and fetuses from exposed dams
Miller et al. (2017)	L-E rats (pregnant)	F	NR^d	9–10	0.4	4	GD 5 and 6	≈^e	–	≈^e
							GD21 (15 d post-exp)	≈^e	–	≈^e
			NR^d	9–10	0.8	4	GD 5 and 6	↓^e	–	↓^e
							GD21 (15 d post-exp)	≈^e	–	≈^e
	L-E rats^e (GD21 fetuses)	M	GD21	7–8	0.4	4	GD 5 and 6	↓	≈/≈/≈^f	–
			GD21	7–8	0.8	4	GD 5 and 6	↓	↓/↓/↓^f	–
		F	GD21	7–8	0.4	4	GD 5 and 6	↓	↓/↓/≈^f	–
			GD21	7–8	0.8	4	GD 5 and 6	↓	↓/↓/↓^f	–

↑ ↓: statistically significant results; ↑ ↓: results direction compared to control (not statistically different); ≈: results of treated approximately equal to control based on visual assessment for graphic data or <5% difference for numeric data, and not statistically different; –: no data; BC: body composition; BN: Brown Norway; BW: body weight; d: day; F: female; F344: Fischer 344; FC: food consumption; GD: gestational day; L-E: Long-Evans; M: male; NS: not significant, data not shown; SD: Sprague–Dawley; WIS: Wistar; WIS-KY: Wistar Kyoto; wk: week; S1: Study 1; S2: Study 2.

^aIn Umezu et al. (1993), BW and FC were measured daily at the end of the 22-h O₃ exposure. BWs were statistically significantly decreased at day 2 (at 0.2 ppm), day 1–3 (at 0.4 ppm), and all 7 days (at 0.8 ppm) after O₃ exposure; all but the 0.8 ppm BWs recovered by the last exposure. FCs were significantly decreased at day 2 (at 0.4 ppm) and day 1–3 (at 0.8 ppm) after O₃ exposure; all FCs had recovered by the final exposure.

^bIn Gordon et al. (2013, 2016a, 2016b), body composition was measured as % fat mass, % lean mass, and % fluid.

^cGraham et al. (1982) measured body weight loss (body weight difference between pre- and post-O₃ exposure); while there was a significant decrease in both control and O₃-exposed mean BW, the O₃-exposed animals had less BW loss than controls; where controls had a 9.8% BW loss, O₃-exposed had a 5.6% BW loss.

^dMiller et al. (2017) used time-mated dams but their age was not reported.

^eIn Miller et al. (2017), BW gain and FC were measured during the 48-h exposure period and then for the entire gestation period in pregnant rats. Significantly reduced body weight gain and food consumption were observed in dams at 0.8 ppm during the 48-h exposure period, but not at 0.4 ppm. No significant effects were observed for total gestational weight gain and overall food consumption during gestation at any dose.

^fIn Miller et al. (2017), fetuses were not directly exposed to O₃ as it was in utero; data were included for completeness. Body composition was measured as fat mass, lean mass, and fat to lean mass ratio in the fetuses: fat/lean/fat-to-lean ratio.

There were 42 study groups, organized in Table 12 according to ozone exposure level, duration, and frequency. BW and FC data generally corresponded to the end of exposure or following a recovery time; some data were provided for interim time points. Of the 42 study groups, 34 were comprised of male animals and eight of females. Ozone exposure levels ranged from 0.05 to 3 ppm, with exposure durations ranging from 3–24 h/d and exposure frequency from a single 1-d exposure up to 28 consecutive days. Seven study groups had exposures that were repeated over 4 to 17 weeks; for two study groups there were also 7-d recovery periods. Eleven datasets reported on 0.8 ppm ozone exposures and eight on exposures at 1.0 ppm ozone. Four studies included a dose-response assessment (Mole et al. 1985; Umezu et al. 1993; Gordon et al. 2016a; Miller et al. 2016b).

Overall, acute (1 or 2 d) exposure to ozone concentrations ranging from 0.05 to 0.1 ppm appeared to have minimal effect on BW; data are not available to address acute exposure and effects on BC or FC. However, based on the limited data, BW effects were not observed to be consistent across species following acute exposures to 0.8 or 1.0 ppm ozone. For example, in Wistar-Kyoto male rats, one 4-h/d, 2-d exposure to 0.8 ppm ozone showed a statistical decrease in BW (Henriquez et al. 2018) and one 4-h/d 1-d exposure to 1.0 ppm ozone showed a non-statistical decrease in BW (Miller et al. 2016c), while another 5-h, 1-d exposure to 1.0 ppm ozone in male CD-1 mice showed a statistical increase in BW (Graham et al. 1982). A single 5-h exposure to 3 ppm ozone did not affect BW (roughly equivalent) compared with control (air)-exposed Sprague-Dawley rats (Mole et al. 1985). Therefore, acute exposures of rats and mice to ozone up to 3.0 ppm did not appear to affect BW consistently.

Repeated inhalation exposures to ozone showed some indications of dose-response in its effect on BW based on duration and frequency, with little effect seen for exposures from 0.05 to 0.8 ppm for durations of 6 h or less. Three studies “continuously” exposed the animals to ozone, with a duration of 22–24 h/d, and reported roughly equal BW at 0.1 ppm ozone but decreased BW from 0.2 to 1.0 ppm ozone, with 7 to 14 d of exposure [ICR mice: (Umezu et al. 1993); Fischer 344 rats: (Watkinson et al. 1995); Hartley guinea pigs: (Vaughan et al. 1984)]. However, the decrease in BW recovered to control values during the repeated exposures for the 22 h/d, 7-d exposures in mice, except for the 0.8 ppm ozone exposures. Thus, repeated lengthy (22–24 h) exposures to 0.2 ppm ozone and above seemed to result in decreased BW, with recovery during exposures for some subgroups, but not all, despite continued exposure.

There was only a single study group where BW increased following ozone exposure. As mentioned above, a 5-h, 1-d exposure to 1.0 ppm ozone in male CD-1 mice showed a statistically significant increase in BW (Graham et al. 1982). However, in general, at ozone exposures below 0.8 ppm, it was observed that exposures must be lengthy (e.g. 22–24 h) in order to produce any effect on BW or BC (limited data).

For the remaining 41 study groups, including those comprised of rats, mice, and guinea pigs, when there was a change observed with ozone exposure compared with controls, it was a decrease in BW. This result is the opposite to the hypothesized direction of body weight and ozone exposure in terms of metabolic syndrome and diabetes. This observation is echoed in the limited BC data, where % fat was consistently either unchanged or decreased compared with controls. A consistent response of BW loss in animal models following inhalation exposure to ozone was observed across several laboratories and in different species. It should be noted that the study groups were subjected to high exposures of ozone. The recovery demonstrated by the limited number of study groups that included data after the termination of ozone exposure or even while the exposure continued in some cases indicated that this BW loss was likely a transitory effect. The available FC data were limited, and the results were mixed, with no obvious pattern related to ozone exposure level or duration/frequency.

Overall, there is no evidence to support an association between inhalation exposure to ozone and increased BW. Limited data on female animals and the lack of any long-term data on BW/BC/FC also hinders the analysis and any confidence in a conclusion on a possible association with increased BW/BC. However, there is evidence to support an association between ozone exposure and decreased BW. This is considered a more expected toxicologic outcome of inhalation exposure to high levels of a reactive gas, such as ozone. Thus, while collectively, these results do not indicate any significant association between inhalation ozone exposure and increased BW, they strongly support an association between ozone inhalation exposure and decreased BW. Thus, the evidence was categorized as a strong inverse association, as the observed effect was opposite to the hypothesis (Table 14).

Five studies evaluated associations between ozone and effects on body temperature (BT) (Mautz and Bufalino 1989; Watkinson et al. 1993, 1995; Gordon et al. 2014; Henriquez et al. 2018) (Table 13). The studies evaluating ozone exposure and effects on BT exclusively used male rats, with various strains represented, including Sprague-Dawley, Fischer 344, Wistar Kyoto, and Brown Norway. Ozone concentrations ranged from 0.2 to 1.0 ppm. Single exposures (2–3 h) to 0.2 (n = 8), 0.25 (n = 4–6), or 0.4 ppm (n = 8) ozone (Mautz and Bufalino 1989; Watkinson et al. 1993) did not change BT as compared with control (air)-exposed animals. However, in two additional study groups, single exposures (2 h) to 0.37 (n = 4–6) or 0.5 ppm (n = 4–6) ozone resulted in significant decreases in BT in ozone-exposed animals during the second hour of exposure (Watkinson et al. 1993). BT measurements in these animals during the first hour of exposure were not different from control (air)-exposed animals. No changes in BT were observed in animals (n = 4–5) subjected to a longer exposure regimen of 6 h/day for 5 days to 0.5 ppm ozone (Watkinson et al. 1995). In the same study, continuous exposure of the animals (n = 4–5) to 0.5 ppm for 23 h/day for 5 days resulted in inconsistent results. BT in ozone-exposed animals was significantly decreased on exposure day 1 but was unchanged from control (air)-exposed animals at days 2–5.

Table 13. Thyroid and body temperature results analysis after ozone exposure in rats (Suitable studies).

References	Animal model				O3 exposure				Body temp^a	Thyroid measurements^a
	Rat strain	Sex	Age (month)	N (/grp)	Grp	Level (ppm)	Duration (h/day)	Frequency		TSH	T4	T3	FT4	FT3	T3 uptake	ThyWt
Martrette et al. (2011)	WIS	F	1.5	12	–	0.12	6	15 d	–	–	–	–	↑	↑	–	–
Mautz and Bufalino (1989)^b	SD	M	1.5	8	–	0.2	3	1 d	≈	–	–	–	–	–	–	–
Watkinson et al. (1993)^c	F344	M	3–4	4–6	–	0.25	2	1 d	≈	–	–	–	–	–	–	–
Watkinson et al. (1993)^c	F344	M	3–4	4–6	1	0.37	2	1 d, T at 15–75 min	≈	–	–	–	–	–	–	–
Watkinson et al. (1993)^c	F344	M	3–4	4–6	1	0.37	2	1 d, T at 90–120 min	↓	–	–	–	–	–	–	–
Mautz and Bufalino (1989)^b	SD	M	1.5	8	–	0.4	3	1 d	≈	–	–	–	–	–	–	–
Watkinson et al. (1993)^c	F344	M	3–4	4–6	1	0.5	2	1 d, T at 15–45 min	≈	–	–	–	–	–	–	–
Watkinson et al. (1993)^c	F344	M	3–4	4–6	1	0.5	2	1 d, T at 60–120 min	↓	–	–	–	–	–	–	–
Watkinson et al. (1995)	F344	M	3–4	4–5	–	0.5	6	5 d	≈	–	–	–	–	–	–	–
Watkinson et al. (1995) (Continuous Study)	F344	M	3–4	4–5	1	0.5	23	5 d, T at d 1	↓	–	–	–	–	–	–	–
					1		23	5 d, T at d 2–5	≈	–	–	–	–	–	–	–
Mautz and Bufalino (1989)^b	SD	M	1.5	8	–	0.6	3	1 d	↓	–	–	–	–	–	–	–
Mautz and Bufalino (1989)^b	SD	M	1.5	8	1	0.8	3	1 d	↓	–	–	–	–	–	–	–
					2		3	1 d, T at 0–100 min	≈	–	–	–	–	–	–	–
					2		3	1 d, T at 120–180 min	↓	–	–	–	–	–	–	–
Henriquez et al. (2018)	WIS-KY	M	4–5	6–8	1	0.8	4	1 d	↓	–	–	–	–	–	–	–
					2		4	2 d	↓	–	–	–	–	–	–	–
Watkinson et al. (1993)^c (Study 1)	F344	M	3–4	4–6	1	1.0	2	1 d, T at 15–30 min	≈	–	–	–	–	–	–	–
					1		2	1 d, T at 45–120 min	↓	–	–	–	–	–	–	–
Watkinson et al. (1993)^c (Study 2)	F344	M	3–4	4–6	1	1.0	2	1 d, T at 30 min	≈	–	–	–	–	–	–	–
					1		2	1 d, T at 60–120 min	↓	–	–	–	–	–	–	–
					2		2	1 d, T at recovery, 150–240 min	↓	–	–	–	–	–	–	–
					2		2	1 d, T at recovery, 270–300 min	≈	–	–	–	–	–	–	–
Gordon et al. (2014)	BN	M	9	5–6	1	1.0	6	2 d/wk, 13 wks; avg T from final 2 h of exposure days each wk	↓/↓ 9/4	–	–	–	–	–	–	–
				5–6	2		6	2 d/wk, 13 wks, hourly T: 1st 12 h of recovery d1 of each exp wk	↑/↑ 4/8	–	–	–	–	–	–	–
			21	6	1		6	2 d/wk, 13 wks; avg T from final 2 h of exposure days each wk	↓/↓/≈ 2/9/2	–	–	–	–	–	–	–
				6	2		6	2 d/wk, 13 wks, hourly T: 1st 12 h of recovery d1 of each exp wk	↑/↑ 1/11	–	–	–	–	–	–	–
Huffman et al. (2006)^d	SD	M	214–268 g	5–6		1.0	4	1 d	–	–	↓	–	–	–	–	–
Clemons and Garcia (1980a)^d,e	SD	M	300–350 g	6–8		1.0	24	1 d	–	↓	↓	↓	–	–	≈	↑
Clemons and Wei (1984)^d,e	SD	M	100–250 g	8	1	1.0	24	1 d	–	–	↓	↓	–	–	–	–
				8	2		24	1 d	–	–	↓	–	–	–	–	–
				17	3		24	1 d	–	↓	↓	–	–	–	–	–
Sen et al. (1993)^e	SD	M	3–4	NR		1.0	24	1 d	–	–	–	–	↓	↓	–	–
Huffman et al. (2001)	SD	M	1.5	5–6	–	3.0	3	1 d	–	–	≈	–	–	–	–	–

↑ ↓: statistically significant results; ↑↓: increased or decreased results; ≈: results of treated approximately equal to control based on visual assessment for graphic data or <5% difference for numeric data, and not statistically different; –: no data; d: day; grp: group; M: male; NR: not reported; ppm: parts per million; SD: Sprague–Dawley; BN: Brown Norway; WIS: Wistar; WIS-KY: Wistar Kyoto; F344: Fischer-344; wk: week; TSH: thyroid stimulating hormone; T4: thyroxin; T3: triiodothyronine; FT4: free T4; FT3: free T3.

^aResults are represented as increased, decreased, or approximately equal measurements in ozone-exposed groups compared with control (air)-exposed groups. Bold lines separate different levels (ppm) of ozone exposure.

^bUnless otherwise indicated, In Mautz and Bufalino (1989), body temperature measurements were taken and averaged during the last hour of exposure (120–180 min).

^cIn Watkinson et al. (1993), body temperature measurements were taken during the 2 h of exposure (15–120 min).

^dClemons and Garcia (1980a), Clemons and Wei (1984), and Huffman et al. (2006) did not report the age of the rats, however, the authors reported the body weights.

^eClemons and Garcia (1980a), Clemons and Wei (1984), and Sen et al. (1993) did not meet the quality criteria for WOE review due to critical deficiencies in their study design and reporting. Their data were NOT used for causal determination.

At higher ozone concentrations, including 0.6, 0.8, and 1.0 ppm ozone, short-term exposure (<1 week) resulted in some significant decreases in BT in ozone-exposed animals, particularly during the latter half of the exposure period (e.g. 120–180 vs. 0–100 min), while other study groups remained unchanged compared with control (air)-exposed animals (Mautz and Bufalino 1989; Watkinson et al. 1993; Henriquez et al. 2018). Gordon et al. (2014) evaluated longer-term ozone exposure (6 h/day × 2 day/week for 13 weeks) to 1.0 ppm ozone in both adult (9-month-old) and senescent (21-month-old) animals. BT measurements taken in the final 2-h of each exposure day, averaged across the two consecutive exposure days for each exposure week, indicated that ozone exposure resulted in some significant decreases in BT during the exposure, depending on the week, with more of an effect observed in adult animals (n = 5–6) compared with senescent animals (n = 6). For both age groups, BT measurements taken during the first 12 h of the first recovery day of each exposure week were increased (often significantly) compared with control (air)-exposed animals, indicating that the effects of ozone on BT may be transient (Gordon et al. 2014).

Collectively, the available data on BT are limited and, although the results do indicate a tendency toward a decrease in BT during inhalation exposure to ozone at or above 0.8 ppm, the limited recovery data (single dataset) show an increase in BT compared with control (air)-exposed animals during each early post-exposure recovery period following inhalation exposure to 1.0 ppm ozone. Thus, with such a limited dataset, there is not adequate evidence to support any association between ozone exposure and BT (Table 13). The strength of the toxicologic evidence was therefore categorized as weak, because it is inconsistent in direction and is based upon only five publications and a limited number of study groups (Table 14). In addition, no data were available for female animals.

Table 14. Summary of animal toxicological PECOS evidence.

Outcome	Method	No. of studies	Animal species	No. study groups	O3 exposure range (ppm)	Duration range (h/d)	Frequency range	Animal evidence synthesis
Glucose-related
Glucose intolerance/fasting hyperglycemia	GTT test	11	Rats	48	0.25–1.0	4–16	Single exposure (1 d) to 3 d/wk for 13 wks	Strong evidence for O3-induced glucose intolerance and fasting hyperglycemia
Insulin resistance	ITT test	2	Rats	11	0.25–1.0	5–16	Single exposure (1 d) to 3 d/wk for 13 wks	Weak (limited) evidence for ozone-induced insulin resistance
Hepatic gluconeogenesis	PTT test	2	Rats	7	0.25–1.0	5–16	Single exposure (1 d) to 3 d/wk for 13 wks	Weak (inconsistent) evidence for O3-induced changes to hepatic gluconeogenesis
Serum/plasma hormone analytes (glucagon, insulin, leptin, CORT)	Serum, plasma sampling	17	Rats, mice	47	0.1–2.0	3–16	Single exposure (1 d) to 1–3 d/wk for 13–17 wks	Weak (inconsistent) evidence for O3-induced changes in serum hormone analytes
Dyslipidemia
Dyslipidemia (CHOL, TG, LDL, HDL)	Serum lipid sampling	15	Rats, guinea pigs	68	0.25–3.0	4–22	Single exposure (1 d) to 3 d/wk for 13 wks	Weak (inconsistent) evidence for O3-induced dyslipidemia
Overweight/obesity-related
Body weight (BW)/ body composition (BC), food consumption (FC)	BW/BC, food weight	16	Rats, mice, guinea pigs	BW: 34 BC: 11 FC: 18	0.1–3.0	3–24	Single exposure (1 d) to 28 d	Strong Negative: evidence for O3-induced decreases in BW (opposite to the hypothesis) Weak (limited and inconsistent) evidence for changes to FC
Body Temperature	Telemetry, rectal temp	5	Rats	26	0.12–1.0	2–23	Single exposure (1 d) to 2 d/wk for 13 wks	Weak (limited and inconsistent) evidence for O3-induced decrease in body temperature
Thyroid (TSH, T4, T3, FT4, FT3, T3 uptake, thyroid weight)	Serum, plasma sampling	3	Rats	9	0.12–3.0	3–24	Single exposure (1 d) to 15 d	Weak (limited and inconsistent) evidence for O3-induced thyroid effects

No.: number; ppm: parts per million; h: hours; d: days; wk: week; wks: weeks; GTT: glucose tolerance test; ITT: insulin tolerance test; PTT: pyruvate tolerance test; CORT: corticosterone; CHOL: cholesterol; TG: triglycerides; LDL: low-density lipoprotein; HDL: high-density lipoprotein; BW: body weight; BC: body composition; FC: food consumption; temp: temperature; TSH: thyroid stimulating hormone; T4: thyroxine; T3: triiodothyronine; FT4: free T4; FT3: free T3.

Six studies (Clemons and Garcia 1980a; Clemons and Wei 1984; Sen et al. 1993; Huffman et al. 2001, 2006; Martrette et al. 2011) evaluated associations between ozone and effects on thyroid hormone homeostasis (Table 13). Three (Huffman et al. 2001, 2006; Martrette et al. 2011) were considered Suitable for evidence synthesis, while data from the remaining three studies (Clemons and Garcia 1980a; Clemons and Wei 1984; Sen et al. 1993) were included as Supplemental only (see above). Aside from Martrette et al. (2011), which used female Wistar rats, all the studies used male Sprague-Dawley rats as the animal model. Among them, ozone concentrations ranged from 0.12 to 3.0 ppm. At 0.12 ppm ozone, Martrette et al. (2011) reported that ozone exposures of 6 h/day for 15 days resulted in significantly increased free T3 (FT3) levels in female rats. While not statistically significant, an increase in free T4 (FT4) levels was also reported for ozone-exposed animals. Huffman et al. (2006) reported a non-significant decrease in T4 levels (n = 5–6) after exposure to ozone at 1.0 ppm for 4 h. In addition, Huffman et al. (2001) reported no change in T4 levels (n = 5–6) after exposure to ozone at 3.0 ppm for 3 h. From the studies in which the data were included as Supplemental, Clemons and Garcia (1980a), Clemons and Wei (1984), and Sen et al. (1993) reported that ozone exposure at 1.0 ppm resulted in significant decreases in various thyroid hormones, including TSH, T4, T3, FT4, and FT3, in ozone-exposed animals compared with control (air)-exposed animals. However, as noted, these studies omitted critical information (e.g. missing n values, no information on ozone generation or exposure methods).

Overall, the toxicologic evidence is quite limited and inconsistent, and therefore the information available was categorized as weak (Table 14).

3.4. Evidence integration

In this section, judgments are presented about the strength of the combined human and animal evidence, as derived from the evidence synthesis step, for the association between ozone inhalation exposure and each PECO statement health outcome (overweight/obesity, T1D, T2D, and metabolic syndrome) (see Figure 2, Methods section).

3.4.1. Ozone inhalation exposure and overweight/obesity

The strength of the epidemiology evidence was categorized as weak based on the limited number of studies (two) with different study designs, and inconsistent results. The animal studies related to this outcome were those that addressed effects on body weight, temperature increase, thyroid effects, and dyslipidemia. The available data on body temperature and thyroid hormone changes were quite limited and there was insufficient evidence to support an association between ozone exposure and these outcomes. The available data on body weight, based on 16 publications, comprising 42 study groups evaluated in three species, demonstrated a strong inverse effect (decreased body weight), which may have been transient; the strength of the body weight evidence was categorized as strongly inverse, associating inhalation exposure to ozone with a decrease in body weight. Finally, for dyslipidemia, the study results collectively did not indicate any significant association between inhalation ozone exposure and effects on serum lipids, and the strength of the toxicologic evidence was categorized as weak. Thus, the overall strength of the experimental animal evidence stream in supporting increased body weight with ozone exposure was characterized as weak.

Because the evidence streams for both epidemiology and toxicologic studies were categorized as weak, we concluded that there is insufficient evidence to draw conclusions regarding a causal association for ozone inhalation and overweight/obesity (Figure 5).

Figure 5. Integration of the human and animal streams of evidence for assessing causality for the four metabolic outcomes. *For T2D, the glucose tolerance test/fasting hyperglycemia toxicology evidence was strong and the other toxicology evidence was weak. There was no definitive basis for concluding that the combined toxicology evidence was strong or weak. Therefore, the strength of the toxicology evidence for T2D is considered unresolved. This led to an overall conclusion of insufficient or suggestive for T2D.

3.4.2. Ozone inhalation exposure and T2D

Both short-term human exposure studies, which examined biochemical indicators of T2D, and long-term human studies for which T2D was the assessed outcome, were categorized as weak due to the limited number of studies (short-term) and inconsistent results with no clear reasons for discrepancies (both short-term and long-term). In contrast, for the experimental animal studies, the strength of the evidence was categorized as strong for outcomes related to glucose intolerance and related fasting/baseline hyperglycemia. However, for animal studies of insulin resistance, hepatic gluconeogenesis, and glucose-related serum hormone analytes, the strength of the evidence was categorized as weak. Reasons for these outcome-specific differences are not apparent and there is no clear approach for determining whether this combined toxicology evidence should be categorized as weak or strong. Until further research clarifies this issue, the evidence is considered unresolved.

These mixed conclusions regarding the strength of the evidence for the two evidence streams (and across the toxicologic endpoints) led to the conclusion that the association between inhalation exposure to ozone and T2D is insufficient or suggestive (Figure 5).

3.4.3. Ozone inhalation and pediatric diabetes (T1D)

The epidemiology literature on associations between ozone exposure and pediatric diabetes was categorized as weak due to inconsistent results and inter-study exposure periods that may not be comparable. There were no toxicologic endpoints that were considered specific to T1D so only one evidence stream was considered. From this, it was concluded that there is insufficient evidence from which to draw conclusions regarding whether ozone inhalation causes T1D (Figure 5).

3.4.4. Ozone inhalation exposure and metabolic syndrome

The epidemiology research on associations between ozone exposure and metabolic syndrome was categorized as weak based on null results from a single study. Similar to the overweight/obesity outcome, the experimental animal studies’ outcomes related to some aspects of metabolic syndrome (e.g. dyslipidemia) were limited and categorized as weak. For dyslipidemia, the study results did not indicate any significant association between ozone inhalation and effects on serum lipids and the strength of the experimental animal evidence was again categorized as weak. Evidence from animal body weight data was considered strong for the association between decreased body weight and ozone exposure, which is opposite of the hypothesized direction (i.e. strong inverse association). However, evidence from studies examining ozone exposure and glucose intolerance in animals was considered strong, although other glucose-related outcomes (insulin sensitivity, hepatic GNG, and serum hormone analytes) were considered weak. In summary, two lines of evidence were strong, but one was strongly supporting while the other was strongly opposing; the remaining four lines of evidence were weak. Further, the clinical definition of metabolic syndrome requires that three of five specific criteria be met. Thus, it was concluded that the overall animal evidence is weak.

The combination of weak epidemiology and animal evidence resulted in a conclusion of insufficient evidence regarding a causal association for ozone inhalation and metabolic syndrome (Figure 5).

4. Discussion

This review had the goal of conducting a systematic review of the evidence for associations between ozone exposure and metabolic effects in humans. Here we offer perspectives on the lessons learned from the framework used here, common human study considerations, our findings in comparison to past reviews, the relationship of our results to other mechanistic information, and data gaps.

4.1. Framework lessons

In this review we followed the now-standard approach of PECO formulation, the use of robust methods for identifying and selecting studies, and the extraction of key data from studies to enhance side-by-side comparisons (Liberati et al. 2009; NTP 2019; US EPA 2020b). For evidence synthesis and confidence characterization, we utilized a more concise and less algorithmic approach compared to other systematic review approaches. For example, we began this process by identifying elements considered to be essential for evaluating whether the identified epidemiologic and in vitro/in vivo animal studies were of sufficient suitability to be carried forward to evidence synthesis and integration. This approach essentially incorporated a “bright line” by dividing Suitable from Supplemental studies. While possible signals from the Supplemental studies were described, these studies were not incorporated into the final body of evidence for causal inference. With respect to the epidemiologic literature, the Supplemental studies could not discern if the exposure preceded that health outcome and are better suited for hypothesis-generation. The toxicological papers that were judged to be Supplemental were deficient in at least one of the five required elements (already fairly minimal) and some were deficient in several elements. Thus, any dataset missing even just one of these critical five elements was not considered reliable enough to use to support extrapolation of any results for causality. We acknowledge that excluding studies can result in reducing the breadth and depth of the literature. In fact, using our approach, 19 of the 34 epidemiologic studies and 11 of 44 of the in vitro/in vivo studies were determined to be Supplemental. However, we emphasize that certain aspects of a study render it inappropriate for causal inference and that this should be recognized. In fact, for several of the papers reviewed here, the primary authors themselves noted the inappropriateness of using their results for evidence integration (e.g. Yang et al. 2019b).

We further argue that completing an extensive rubric of quality elements but then treating all studies equally in evidence synthesis regardless of their limitations is not an efficient use of resources nor does it provide a clear understanding regarding how the different studies were used in evidence synthesis. In addition, including only those quality elements considered to be essential reduces the need for a “scoring” type approach, which becomes necessary when large numbers of quality elements are included; the limitations of scoring approaches have been discussed previously (Jüni et al. 1999; Whiting et al. 2005; LaKind et al. 2014).

4.2. Common human study considerations

Generally, an assessment of a body of literature for its suitability for evidence synthesis would include an evaluation of the quality of the exposure assessments in each study. For example, EPA (2020b) considered the following: Does the exposure measure capture the variability in exposure among the participants, considering intensity, frequency, and duration of exposure? Does the exposure measure reflect a relevant time window? If not, can the relationship between measures in this time window and the relevant time window be estimated reliably?

For ozone and metabolic health outcomes, these questions present a challenge. First, when considering the temporal relationship between exposure and outcome, it is important that the study design includes an appropriate duration of exposure and lag time to support a biologically plausible association between exposure and outcome. This issue was noted in the ISA (US EPA 2020a) as follows: “Inference is stronger when the duration or lag of the exposure metric corresponds with the time course for physiological changes in the outcome (e.g. up to a few days for symptoms) or latency of disease (e.g. several years for cancer).” However, at issue for the ozone exposure/metabolic outcome association is that neither the necessary, i.e. “correct,” duration nor the latency period was well-established. A review of air pollution studies and chronic disease found no consistent relationship between exposure duration (ranging from 60 days to 35 years) and outcomes including diabetes (Lipfert 2018). For another review of studies on air pollution, smoking, and lung cancer mortality, Lipfert and Wyzga (2019) found that studies did not include latency, exposure duration, or temporal trends; similar issues impact the ozone/metabolic outcome literature. Further complicating the determination of appropriate exposure assessment time and duration is the proposed link between prenatal/postnatal exposures and later development of obesity, diabetes, and metabolic syndrome (Wang et al. 2014). These exposure windows were, for the most part, not included in the studies reviewed here. In general, as noted by McKone (2009), “[s]electing an optimum measurement or modeling-based exposure prediction strategy has to be based on consideration of the timescale over which the biological effect occurs.”

Second, the exposure assessments (with the exception of the single experimental study by Miller et al. 2016a) were generally based on data from one or more monitoring stations, with the assumption that ozone concentrations can serve as an exposure proxy for an individual (or population) (Bravo and Bell 2011). It is also often assumed that ozone concentrations are relatively homogenous over large areas, except for locations near roadways (US EPA 2020a). However, numerous studies have described temporal and spatial variability in outdoor ozone concentrations for both short-term and long-term exposures (Liu et al. 1997; US EPA 2001; Bell 2006; Bravo and Bell 2011; Bravo et al. 2012; Cheadle et al. 2017; Sadighi et al. 2018; Wells et al. 2019; Liu et al. 2020). Further, monitoring locations can introduce bias as they may over- or under-represent some populations, particularly if they do not consider peak concentrations (Bravo and Bell 2011; Wells et al. 2019). Liu (1997) observed that personal ozone exposures may be markedly different from outdoor stationary ozone measurements, and that personal ozone exposure is not well-predicted by outdoor measurements. Spatial and temporal variability is further complicated by indoor ozone exposures, as indoor and outdoor levels differ, and many people spend most of their time indoors. For example, in a recent study examining the relationship between indoor and outdoor exposures to ozone, Barkjohn et al. (2020) reported that “[a]verage personal exposures to O₃ ranged from below the limit of detection (6 ppb) to 22 ppb with a median exposure of 10 ppb (average = 11 ppb); the median outdoor O₃ during this period was 34 ppb (average = 37 ppb).” They also observed large variations in indoor/outdoor ozone concentrations when examining different microenvironments, with 1-h levels ranging from the limit of detection (<6 ppb) to 69 ppb. Daily inhalation intakes of indoor ozone have been estimated at around 25–60% of total daily ozone intake (Weschler 2006) and studies have shown wide ranges of indoor ozone exposure in schools and offices (Salonen et al. 2018) that are unaccounted for in the studies reviewed here. Further, the studies included in this review did not account for movement of people throughout the day (e.g. between home and work).

Additional issues may impact the strength of the exposure assessment, including whether co-pollutants are correlated with ozone (e.g. NO₂) (Kerckhoffs et al. 2015), the extent to which meteorological data were included in the assessment, and the spatial unit (i.e. granularity) of the measurement data. It is also worth noting that the combined effect of air pollution exposure and exercise on health outcomes warrants study and was generally not considered in most of the reviewed publications (Qin et al. 2019).

For all 33 observational studies and their relevant outcomes reviewed here, there was clear potential for exposure misclassification, reducing the overall confidence in the results. Moreover, none of the papers included a quantitative characterization of the error in their exposure assessments due to spatial or temporal variability. Thus, the extent to which these issues introduced the likelihood of exposure misclassification is not known. Finally, whether inter- or intra-study consistency or inconsistency of the results was related to the temporal or geographic scale of measurements, averaging times, exposure model used, latency times, or co-pollutant considerations cannot be discerned for the studies included in this review.

4.3. Conclusions in this review in comparison to other reviews

In this review, the human and the experimental animal literature were examined to determine the strength of the evidence regarding associations between ozone inhalation and four outcomes related to metabolic effects. For three of the four PECO questions relating to T1D, overweight/obesity and metabolic syndrome, we concluded that the evidence is currently insufficient for causal determination. Interestingly, for T2D, a single conclusion could not be easily drawn from the evidence because the animal evidence was mixed and the authors differed on whether a single strong stream of evidence, coupled with four weak streams and a strong inverse stream, yielded an overall strong or weak signal. This, in combination with weak epidemiology evidence, led to a finding of either insufficient or suggestive evidence for the combined animal and human streams.

In contrast, the recent EPA ISA review concluded that there is a likely causal relationship with short-term ozone exposure and a suggestive, but not sufficient to infer, causal relationship with long-term ozone exposure and metabolic effects. There are important differences in this current approach to the review of the literature as compared to the EPA (2020a) review.

First, EPA treated metabolic outcomes as an overall category but considered exposure duration separately. In the current review, we developed PECO questions to differentiate the literature associated with each of the four outcomes and reviewed the duration of exposure within each PECO (we recognize that some reviews are designed to support short- vs. long-term limits on pollutants and for those, the PECO statements may need to formally evaluate exposure duration). This is in concordance with the importance of a well-formulated research question (Morgan et al. 2019). By considering apical outcomes separately, we could carefully consider the strengths, weaknesses, and comparability of the literature. The concept of combining vs. separating outcomes for PECO questions in the systematic review process warrants further discussion not just for metabolic outcomes but for a wide range of outcomes (e.g. this issue was briefly described for neurodevelopmental outcomes in LaKind et al. 2017).

Second, EPA conducted its quality assessment with many review elements using a narrative approach (https://hawcprd.epa.gov/study/assessment/100500031/). EPA stated that it sought high quality studies for assessing causal relationships between ozone and adverse health effects, but it is not clear how the judgment of high quality was made or used. While it appears that EPA considered all study results equally, our review applied a “bright line”; evidence integration was limited to those studies considered “Suitable” based on fundamental quality criteria. Several studies used by EPA in evidence integration were not determined to be Suitable for causal inference in this review.

Similarities between this review and the EPA ISA (US EPA 2020a) should be noted. Specifically, regarding the experimental animal data, EPA stated: “…animal toxicologic studies demonstrate that ozone exposure impaired glucose tolerance, increased triglycerides in serum, fasting hyperglycemia, and increased hepatic gluconeogenesis” (US EPA 2020a). The results of the current review agree that there is strong evidence of an association between ozone exposure at high concentrations (≥0.8 ppm) and greater incidence of glucose intolerance and fasting hyperglycemia. These findings were true regardless of whether the ozone exposure was short-term or subchronic; however, increasing duration and/or repeated exposure did not result in an increased effect. Still, our conclusions differed from the EPA findings for the other related outcomes in the toxicology studies; we concluded that the evidence was weak for any association between ozone exposure and changes to serum triglycerides and/or hepatic gluconeogenesis. Collectively, the data regarding these outcomes were either extremely limited or the results were inconsistent across studies and exposure levels/durations, with no clear pattern.

The difference in overall conclusions between the current review and EPA is likely due to a combination of factors. The literature search in this review was inclusive of publications appearing in 2020 and included additional publications beyond those reviewed in the ISA, some of which conflicted with prior data (e.g. Li et al. 2019; Elten et al. 2020; Starling et al. 2020). As mentioned above, studies labeled as Supplemental in the quality evaluation were not part of the evidence integration for the current review. Using distinct outcomes in the PECO statement rather than combining the literature across outcomes under the umbrella of “metabolic outcomes” clearly results in differences in conclusions as this review was able to draw distinct conclusions for four health effects. Further, we cannot predict with any certainty the impact on EPA’s conclusions had it combined studies of short-term and long-term exposures. We note that our conclusions would not have changed had we treated short-term and long-term exposure epidemiology studies independently. Perhaps most importantly, we raise the question: while our overall conclusions differ from the EPA's, are these differences important? Neither review concluded there was sufficient evidence for causality nor was there evidence to state that ozone is not likely causal for long-term exposure to ozone. EPA did conclude that there was sufficient evidence to support a likely causal relationship between short-term ozone exposure and metabolic effects, but in this review, we did not examine these exposure durations separately.

4.4. Relationship of the results of this review to mechanistic information

An MOA analysis, including a detailed review of available mechanistic data, can be helpful for determining the biological plausibility of evidence streams that demonstrate a possible association. In addition, qualitative and/or quantitative MOA differences between experimental species and humans can be evaluated to provide important insights into the human relevance of experimental animal data. However, given the lack of any conclusions supporting a causal role for ozone-induced effects related to the PECO statements in this review, there was no obvious hypothesis to drive such an MOA analysis. Nonetheless, our findings indicate a strong association between ozone exposure and certain glucose-related effects in experimental animals. Glucose intolerance and hyperglycemia increased with increasing exposure level beginning consistently at 0.8 ppm ozone and above; however, there was less confidence that increased duration of exposure or longer exposure frequency resulted in consistently increased ozone-induced metabolic effects. While limited, the available experimental animal evidence suggests that these acute metabolic effects may be short-lived, with homeostasis restored shortly (18 h) after cessation of exposure. In addition, repeated episodic exposures may result in an attenuation of the ozone-induced effects, possibly due to the initiation of an adaptive response.

Some evidence indicates that neurohumoral sympathetic modulation may be involved in the mechanism of ozone-induced fasting hyperglycemia and glucose intolerance (Bass et al. 2013; Kodavanti et al. 2015; Kodavanti 2019). In particular, the hypothesis suggests that ozone induces a neuroendocrine stress response, activating the hypothalamic-pituitary-adrenal (HPA) and sympathetic-adrenal-medullary (SAM) axes, which can lead to increased circulating epinephrine and glucocorticoids, including corticosterone, although the available data are not consistent (Miller et al. 2015; Snow et al. 2018; Henriquez et al. 2019, 2020). In this process, increased epinephrine and glucocorticoid release would upregulate hepatic GNG and glycogenolysis, likely leading to an increase in blood glucose and an inhibition of beta cell insulin secretion and impaired glucose uptake through the glucocorticoid receptors (Kuo et al. 2015; Miller et al. 2016b). In support of this notion, while some studies have shown that high concentrations of ozone can induce systemic metabolic effects in rats, such as fasting hyperglycemia and glucose intolerance, sometimes also with increased circulating levels of epinephrine and corticosterone, these effects are partially or fully blocked in rats subjected to bilateral adrenal demedullation or total bilateral adrenalectomy (Miller et al. 2016c). In addition, exacerbation of ozone-induced systemic effects has been reported in rats treated with β₂-adrenergic and glucocorticoid receptor agonists, whereas these effects are mitigated in rats treated with adrenergic and glucocorticoid receptor antagonists (Snow et al. 2018; Henriquez et al. 2019, 2020). Emerging evidence suggests that the tissue distribution of these receptors, along with the presence of individual variability, may represent a critical factor in these specific ozone-induced metabolic effects (Kodavanti 2019). Although some data report that ozone has been shown to activate the HPA and SAM axes, the initiating events of autonomic activation remain unclear, particularly the role of vagal sensory nerves and/or circulating mediators that may act on different brain regions, such as the hypothalamus. However, we did not find consistent evidence of increased corticosterone following ozone exposure, nor consistent evidence of increased hepatic GNG, to support this proposed MOA.

4.5. Data gaps

As stated above, certain gaps in the toxicology and epidemiology literature likely contributed to the conclusions reached in this review. We describe a few of these gaps here.

The animal toxicologic studies lacked data related to chronic ozone exposure and only a limited number of datasets were available to evaluate whether the effects of ozone were transient or long-lasting. The few studies with post-exposure recovery periods allowed for only 18 h or 1 week recovery and were not repeated to demonstrate whether the apparent recovery was reproducible following another set of exposures. As a result, future research that includes chronic exposure data will assist in the understanding of whether adaptation to ozone occurs in exposed animals such that any effect on glucose tolerance and/or hyperglycemia may be considered transient or long-lasting.

Within the epidemiologic research, issues around properly characterizing short- and long-term exposures are well-known (and were described earlier in this paper). Improved information on temporal and spatial variability and indoor vs. outdoor exposures is needed. Information on critical windows of exposure would improve researchers’ ability to obtain biologically relevant exposure information. Additionally, while difficult to accomplish, future studies would be improved by the collection of information on study subjects that may correlate with both exposure and disease risk. At present, it is not known whether the lack of this information would result in an over- or under-estimate of risk.

5. Conclusions

We evaluated the epidemiology and toxicology literature on ozone and metabolic outcomes using a streamlined systematic review approach. For T1D, overweight/obesity, and metabolic syndrome, the available evidence is insufficient for causal assessment. For T2D, the evidence is conflicting and the overall conclusion unresolved, with the authors of this paper supporting a conclusion of insufficient evidence or suggestive evidence. The differences in these conclusions compared to those of the US EPA are likely explained by differences in the approach to developing PECO statements, the literature included in the review, and the process used for evidence integration.

The growing body of systematic review guidelines is becoming increasingly complex. As systematic review processes continue to evolve, it is worth exploring the following challenging questions:

• To what extent should related health outcomes be combined or kept distinct?

• Can the elements used to select studies suitable for causal inference be tightly focused to enhance transparency in the evaluation?

• How fine a gradation should be included in evidence integration tiers and what purposes do additional categories serve?

Acknowledgments

Our thanks to Josh Naiman, LaKind Associates, LLC, for literature database management and assistance with quality assurance. We also appreciate Jon Hotchkiss’ (Inhalation Toxicology Consultants) insights into inhalation toxicology. In addition, we thank Wenchao Li and Jieqiong Zhou (both with Gradient Corporation) for valuable discussions and assistance with quality assurance. Finally, we thank the three anonymous reviewers for their careful reviews of the manuscript. API was given a courtesy copy of the draft manuscript but made no comments. The review, synthesis, and conclusions reported in this paper are the exclusive professional work product of the authors and do not necessarily represent the views of the API or the member companies.

Declaration of interest

Judy LaKind, Carol Burns, Lynn Pottenger, Satori Marchitti, and Julie Goodman consult to governmental and/or private sectors. Dr. Burns and Dr. Pottenger were previously employed by The Dow Chemical Company; Dr. Pottenger was also employed by Olin Corporation. Dr. Marchitti was previously employed by the US EPA. Currently, Drs. Burns and Pottenger have established independent consulting practices. Dr. Goodman has reviewed and commented on EPA evaluations of ozone as part of the NAAQS process with funding from API. She has also published several papers on ozone funded by API and TCEQ.

This work was supported by the American Petroleum Institute (API), which issued an RFP requesting proposals to conduct this research. DQN was supported through a contract with the Johns Hopkins University; funding was from API via a contract with LaKind Associates, LLC. API is the major trade association of corporations in the petroleum sector, from discovery through production and refining. API was not involved in the design, collection, management, analysis, or interpretation of the data, or in the preparation or approval of the manuscript. The API was given a courtesy copy of the draft manuscript but made no comments. The authors retain sole responsibility for the writing and content of this paper, which represent the professional opinions of the authors and not necessarily those of API or its member companies. The review, synthesis, and conclusions reported in this paper are the exclusive professional work product of the authors and do not necessarily represent the views of the API or the member companies.

Supplemental material

Supplemental material for this article is available online here.