Nitrogen Oxides (NOx/NO2) Epidemiology and Respiratory Disease

RoseyFields evaluates the human epidemiological evidence for NOx and NO2 as follows. This page summarizes the study populations and exposure-assessment approaches used in the NO2/NOx literature, organizes the health outcomes that have been investigated in humans by body system and evidence strength, discusses the study-quality considerations that most affect confidence in specific findings, and describes how RoseyFields uses this evidence, alongside toxicological mechanism data, to support general and specific causation opinions.

A threshold note on exposure metrics: NO2 is a reactive gas that does not persist in blood or tissue, so it is not assessed through a serum or urinary biomarker the way persistent chemicals (e.g., PFAS, metals, some pesticides) are, and it is not part of the ambient chemical panel biomonitored by NHANES. Human NO2 epidemiology instead relies on (a) fixed-site ambient monitoring data, (b) land-use regression (LUR) or dispersion models that estimate residential, workplace, or personal outdoor concentrations from monitoring and traffic/land-use data, and (c) in a smaller number of studies, personal exposure monitors worn by individual subjects. RoseyFields’ summary below reflects that exposure-assessment reality rather than a biomonitoring framework, which does not apply to this chemical.

1. Study Populations and Data Sources

The NO2/NOx epidemiological literature draws on several distinct categories of study population, each with different strengths for causal inference:

1.1 Large General-Population Cohorts with Modeled Residential Exposure

The largest and most influential body of NO2 evidence comes from prospective cohorts that assign long-term NO2 exposure to each participant’s residential address using LUR or dispersion modeling, then follow the cohort for years to decades for incident disease and mortality. Examples include the Rome Longitudinal Study (over one million adults; Cesaroni et al. 2013), the multi-country ESCAPE (European Study of Cohorts for Air Pollution Effects) consortium, which pooled cohort data from nine European countries (Cesaroni et al. 2014; Raaschou-Nielsen et al. 2013), and the American Cancer Society Cancer Prevention Study II cohort in the United States. These studies benefit from large sample sizes, long follow-up, and standardized exposure-modeling protocols applied consistently across sites, which allows formal meta-analysis of harmonized effect estimates rather than simple qualitative comparison of heterogeneous studies.

1.2 Birth and Childhood Cohorts

Prospective childhood cohorts — most prominently the University of Southern California Children’s Health Study, which has followed multiple cohorts of children in Southern California communities since the 1990s — assign NO2/NOx exposure using both regional monitors and near-roadway exposure models, and track lung function growth, asthma incidence, and related outcomes as children age (Gauderman et al. 2007; Gauderman et al. 2015). Birth-outcome studies typically assign trimester-specific NO2 exposure to the maternal residence and examine birth weight, preterm birth, and related endpoints (Stieb et al. 2012).

1.3 Time-Series and Case-Crossover Studies of Acute Morbidity

A separate, methodologically distinct literature examines short-term (day-to-day) fluctuations in ambient NO2 against day-to-day counts of emergency department visits, hospital admissions, or symptom reports, using each subject or community as its own control to remove time-invariant confounding. Multi-city and multi-year studies such as Peel et al. (2005) in Atlanta and Schildcrout et al. (2006) across eight U.S. cities are representative of this design, and systematic reviews such as Zheng et al. (2021) pool dozens of such studies internationally.

1.4 Occupational and Near-Source Cohorts

A smaller literature examines occupationally or environmentally elevated NO2/NOx exposure in specific worker or community populations — for example, traffic police and professional drivers whose duties place them in near-roadway microenvironments for extended periods, and residents living near major roadways or point sources. These studies are valuable for characterizing higher-exposure, real-world scenarios closer to many litigation fact patterns, but individual studies in this category tend to be smaller, more heterogeneous in design, and more susceptible to healthy-worker bias than the large general-population cohorts.

1.5 Why Population Diversity Strengthens Causal Inference

RoseyFields places particular weight on findings that replicate across these different population types and study designs, consistent with the Bradford Hill consideration of consistency and with the Daubert/Reference Manual on Scientific Evidence emphasis on replication across independent investigators and settings. A finding that holds in short-term time-series studies (different day, same city), long-term cohort studies (different person, same city, years of follow-up), and multi-country pooled consortia (different country, different regulatory and monitoring context) is far less likely to be an artifact of a single population’s confounding structure, exposure-assessment method, or local co-pollutant mixture than a finding observed in only one study design or one geographic setting. This is the primary reason RoseyFields’ evidence tables, discussed below, weight cross-design and cross-population replication as heavily as statistical significance within any single study.

2. Health Outcomes by Body System and Evidence Strength

The following organizes the outcomes most frequently investigated in the human NO2/NOx literature by body system, and classifies each into one of three evidence tiers RoseyFields uses in its case work: (A) strong, consistent, quantitatively modeled evidence; (B) suggestive but inconsistent evidence; and (C) outcomes investigated with null or inadequate findings specific to NO2.

2.1 Respiratory System

Tier A (strong, consistent, quantitatively modeled):

• Asthma exacerbation / emergency department visits and hospital admissions. This is the most consistently replicated NO2 health effect across study designs. Zheng et al. (2021), a systematic review and meta-analysis of short-term NO2, O3, and SO2 exposure and asthma-related emergency visits and hospital admissions, rated the certainty of evidence for 24-hour NO2 exposure as high. Peel et al. (2005) and Schildcrout et al. (2006) are representative of the U.S. multi-city time-series and panel studies underlying this conclusion.
• Childhood lung function development. The Children’s Health Study cohort found that reductions in ambient pollution, including NO2/NOx, over successive calendar periods were associated with improved lung function growth in children, in both asthmatic and non-asthmatic children (Gauderman et al. 2015), building on earlier findings that near-roadway and regional traffic-related pollutant exposure, including NOx, was associated with deficits in lung development from age 10 to 18 (Gauderman et al. 2007).

Tier B (suggestive, less consistent):

• New-onset asthma development in children (as opposed to exacerbation of existing asthma) — the 2016 EPA ISA characterized long-term NO2 exposure as likely to be causally related to asthma development in children, but individual cohort estimates for this specific endpoint are less numerous and more variable in magnitude than the exacerbation literature.
• Chronic obstructive pulmonary disease (COPD) incidence and COPD-related mortality in adults — a growing but still comparatively limited cohort literature, with effect sizes that vary by cohort and co-pollutant adjustment strategy.

2.2 Cardiovascular System

Tier A (strong, consistent, quantitatively modeled):

• Coronary heart disease incidence and mortality. Cesaroni et al. (2013), following more than a million adults in Rome, reported an adjusted relative risk of 1.03 (95% CI: 1.00–1.07) for coronary event incidence per 10 µg/m3 increase in residential NO2, with a stronger association for fatal events (RR 1.07; 95% CI: 1.02–1.12) and out-of-hospital coronary deaths (RR 1.08; 95% CI: 1.02–1.13). The pooled ESCAPE analysis of eleven European cohorts (Cesaroni et al. 2014) similarly examined acute coronary events in relation to long-term NO2 exposure across a substantially larger, multi-country population, adding cross-national replication to the single-city Rome findings.
• All-cause and cardiovascular mortality. The Rome cohort study (Cesaroni et al. 2013) is among the most frequently cited single studies quantifying the association between long-term NO2 exposure and both natural-cause and cardiovascular mortality, and its effect estimates are broadly consistent with subsequent pooled multi-cohort analyses in North America and Europe.

Tier B (suggestive, less consistent):

• Stroke incidence — positive associations have been reported in several European pooled-cohort analyses, generally with smaller and less precise effect estimates than for coronary events, and with more cross-study heterogeneity.
• Hypertension incidence — a mixed literature, with some cohort and cross-sectional studies reporting positive associations with NO2/traffic-related pollution and others null, complicated by the difficulty of separating a chronic air-pollution effect on blood pressure from established risk factors that also cluster geographically with traffic exposure.

2.3 Metabolic / Endocrine

Tier B (suggestive, less consistent):

• Type 2 diabetes incidence. Eze et al. (2015), a systematic review and meta-analysis of ambient air pollution and diabetes in Europe and North America, pooled four NO2 studies and reported an increased risk of type 2 diabetes of 8% per 10 µg/m3 increase in NO2 (RR 1.08; 95% CI: 1.00–1.17). The authors characterized the overall evidence base as indicative of a positive association but noted a high risk of bias across the included studies and called for higher-quality, dose-response-oriented research — language RoseyFields treats as placing this outcome in the suggestive rather than the strong-evidence tier, notwithstanding the statistically significant pooled point estimate.

2.4 Reproductive and Developmental

Tier B (suggestive, inconsistent):

• Preterm birth and low birth weight. Stieb et al. (2012), an early and frequently cited systematic review and meta-analysis, reported a pooled decrease in birth weight associated with NO2 exposure and mixed, generally imprecise associations with preterm birth depending on the trimester of exposure examined. Subsequent, more geographically focused meta-analyses have continued to find directionally positive but often statistically non-significant associations between NO2 and preterm birth, with substantial heterogeneity attributable to differences in exposure windows (first, second, third trimester, or whole pregnancy), outcome definitions, and confounder adjustment across studies. RoseyFields treats birth outcomes as a suggestive but not consistently reproduced NO2 endpoint, in line with EPA’s ISA characterization of the birth-outcomes evidence as suggestive but not sufficient to infer causation.

2.5 Neurological

Tier B (suggestive, emerging):

• Dementia incidence and cognitive decline. Chen et al. (2017), a population-based cohort study of more than two million adults in Ontario, Canada, found that residential proximity to major roads and exposure to ambient air pollutants, including NO2, was associated with increased dementia incidence, with the association strongest for residents living nearest to heavy traffic. This is a comparatively recent area of investigation relative to the respiratory and cardiovascular literature, and replication across additional cohorts and longer follow-up periods is ongoing; RoseyFields treats this endpoint as an emerging, suggestive association rather than a well-established one at this time.

2.6 Cancer

Tier C (null/inadequate findings specific to NO2):

• Lung cancer incidence. This is a body system in which NO2 findings notably diverge from findings for co-emitted particulate pollutants. Raaschou-Nielsen et al. (2013), the pooled ESCAPE analysis of seventeen European cohorts, found a statistically significant association between long-term PM10 and PM2.5 exposure and lung cancer incidence, but did not find a corresponding association between NO2 exposure and lung cancer incidence in the same cohorts. This distinction matters for case work: NO2/NOx is frequently emitted alongside PM2.5, black carbon, and polycyclic aromatic hydrocarbons from the same combustion sources, and it is common in both the epidemiological literature and in litigation for NO2 to be used as a convenient marker or indicator of the overall traffic-related or combustion-related pollution mixture. RoseyFields does not treat a documented mixture-level cancer association as evidence of an NO2-specific cancer effect, and flags this distinction explicitly whenever a case record cites general “air pollution and cancer” literature in connection with NO2/NOx exposure specifically.

No IRIS cancer classification exists for NO2 (Section II of the IRIS Chemical Assessment Summary lists the carcinogenicity assessment as “not available”), and the epidemiological record above is consistent with, rather than contradicting, that regulatory posture.

3. Study Quality Considerations Affecting Confidence in Specific Findings

3.1 Co-Pollutant Confounding

NO2 is emitted from the same combustion sources — principally motor vehicle traffic — as PM2.5, black carbon, ultrafine particles, and polycyclic aromatic hydrocarbons, and ambient concentrations of these pollutants are frequently correlated at the same monitoring location. Because of this correlation, single-pollutant NO2 models can partly capture the health effect of co-pollutants rather than an independent NO2 effect, and adjustment for co-pollutants in multi-pollutant models tends to attenuate NO2 effect estimates to varying degrees across studies. The 2016 EPA ISA specifically addressed this issue for the respiratory endpoint, concluding that recent evidence strengthens the case that short-term NO2 respiratory effects are at least partly independent of other traffic-related pollutants — a conclusion RoseyFields treats as the best-supported current position for that endpoint, while continuing to flag co-pollutant confounding as an open, endpoint-specific issue for cardiovascular, metabolic, birth-outcome, neurological, and cancer endpoints, where the independent-effects literature is thinner.

3.2 Exposure Misclassification

Most long-term NO2 cohort studies assign a single modeled outdoor concentration to a participant’s baseline residential address for the entire follow-up period. This approach does not capture indoor infiltration, personal time-activity patterns (time spent at work, in transit, or in other microenvironments), residential mobility during follow-up, or intra-urban gradients steeper than the spatial resolution of the underlying LUR or dispersion model. Because true individual exposure is measured with error relative to the assigned residential estimate, this form of misclassification generally biases effect estimates toward the null in cohort studies, meaning the true effect size may be larger than reported rather than smaller — an important point for RoseyFields to raise where a defendant argues that a modest reported relative risk understates a plaintiff’s actual exposure-response relationship, and an equally important caution against inflating modest cohort-level relative risks when reconstructing an individual’s exposure history.

3.3 Reverse Causation

Reverse causation is a more significant concern for cross-sectional studies (where exposure and outcome are measured at the same time) than for the prospective cohort designs that dominate the strong-evidence tier above, where NO2 exposure is assigned at baseline before outcome ascertainment. RoseyFields gives comparatively less weight to cross-sectional NO2 findings — for example, some of the earlier diabetes and hypertension literature — precisely because temporality between exposure and disease onset cannot be established with the same confidence as in the prospective cohort literature.

3.4 Study Design Heterogeneity and Publication Bias

Several of the meta-analyses discussed above explicitly report meaningful heterogeneity across included studies (Zheng et al. 2021, for pollutants other than 24-hour NO2 specifically) or a high assessed risk of bias across the pooled literature (Eze et al. 2015, for diabetes). RoseyFields treats reported heterogeneity and risk-of-bias findings from the underlying meta-analyses themselves as part of the evidence record, rather than independently re-litigating study quality, and calibrates the strength of any general-causation opinion accordingly — treating a low-heterogeneity, high-certainty pooled estimate (e.g., the asthma exacerbation literature) differently from a heterogeneous, high-risk-of-bias pooled estimate (e.g., the diabetes literature), even where both report a statistically significant point estimate.

4. Weight-of-Evidence Determinations

Unlike some contamination litigation contexts — for example, the C8 Science Panel’s independent “probable link” determinations for PFOA — no comparable independent expert panel has been convened to issue formal weight-of-evidence or probable-link findings specific to NO2/NOx. The operative weight-of-evidence framework for NO2 is instead EPA’s own Integrated Science Assessment (ISA) causal-determination process, conducted periodically as part of the NAAQS review cycle described in RoseyFields’ companion risk assessment methodology page for this chemical. The most recent NO2 ISA (finalized January 2016) reached the following causal determinations, which RoseyFields treats as the current agency weight-of-evidence conclusions for this chemical group:

• Short-term NO2 exposure and respiratory effects (particularly asthma exacerbation): causal or likely to be a causal relationship;
• Long-term NO2 exposure and respiratory effects (including asthma development in children): likely to be a causal relationship;
• Short- and long-term NO2 exposure and cardiovascular and metabolic effects: suggestive, but not sufficient to infer a causal relationship;
• Long-term NO2 exposure and birth outcomes: suggestive, but not sufficient to infer a causal relationship; and
• Long-term NO2 exposure and cancer: suggestive, but not sufficient to infer a causal relationship.

These agency determinations are broadly consistent with the tiering RoseyFields applies to the primary literature in Section 2 above, and RoseyFields cross-references both — the primary peer-reviewed studies and the ISA’s causal-determination language — when characterizing the state of the science in an expert report, rather than relying on the ISA’s summary characterization alone.

5. Use of Epidemiological Evidence in Litigation: General and Specific Causation

5.1 General Causation

For endpoints in the strong-evidence tier (respiratory, particularly asthma exacerbation, and cardiovascular events and mortality), RoseyFields supports general causation opinions by combining: (a) consistency of the epidemiological association across independent study designs (time-series, panel, and cohort) and populations (Section 1.5); (b) a coherent dose-response gradient, where reported across studies; (c) biological plausibility from toxicological and controlled human exposure data on airway inflammation, oxidative stress, and endothelial dysfunction mechanisms; and (d) the concurrence of EPA’s own ISA causal determinations (Section 4). This combination addresses the Bradford Hill considerations most frequently scrutinized in Daubert-type challenges to general causation opinions — strength of association, consistency, temporality, biological plausibility, and coherence with existing scientific knowledge.

For endpoints in the suggestive tier (metabolic, birth outcomes, neurological), RoseyFields is correspondingly more circumspect in any general causation opinion, typically characterizing the evidence as supportive of an association but not, on the current literature, sufficient on its own to establish general causation with the same confidence as the respiratory and cardiovascular endpoints — language calibrated to track the ISA’s own “suggestive, but not sufficient” characterization rather than overstating the epidemiological record. For lung cancer specifically, RoseyFields does not support a general causation opinion attributing lung cancer risk to NO2/NOx exposure itself, given the ESCAPE finding of no NO2-lung cancer association net of the co-emitted particulate matter effect (Section 2.6).

5.2 Specific Causation

Specific causation — whether NO2/NOx exposure caused or contributed to a particular individual’s disease — requires reconstructing that individual’s exposure history (using the dispersion modeling and monitoring-data methods described in RoseyFields’ companion risk assessment methodology page) and comparing it to the concentration-response relationships established in the general causation literature above. RoseyFields flags an important practical limitation for counsel at the outset of any specific-causation engagement: most of the well-replicated NO2 relative risks summarized in Section 2 — including the strong-evidence cardiovascular and respiratory associations — are modest in magnitude (typically in the range of a few percent increased risk per 10 µg/m3 increment, rather than a doubling or more of risk). Because courts applying a “more likely than not” (probability of causation greater than 50%) standard to specific causation frequently look for a relative risk exceeding 2.0 before an epidemiological association alone can support that inference, NO2/NOx epidemiology by itself will often be insufficient to support specific causation for an individual plaintiff without additional case-specific evidence — for example, an unusually high modeled or monitored personal exposure level, a differential diagnosis excluding competing causes, or a combination with other co-exposures established in the same matter. RoseyFields addresses this limitation directly and early in any engagement involving individual specific-causation opinions, rather than allowing a general-causation association to be presented as sufficient, standing alone, to meet an individual plaintiff’s specific-causation burden.

References