Long-Term Ozone Exposure and Mortality in a Large Prospective Study

Abstract

Rationale: Tropospheric ozone (O₃) is potentially associated with cardiovascular disease risk and premature death. Results from long-term epidemiological studies on O₃ are scarce and inconclusive.

Objectives: In this study, we examined associations between chronic ambient O₃ exposure and all-cause and cause-specific mortality in a large cohort of U.S. adults.

Methods: Cancer Prevention Study II participants were enrolled in 1982. A total of 669,046 participants were analyzed, among whom 237,201 deaths occurred through 2004. We obtained estimates of O₃ concentrations at the participant’s residence from a hierarchical Bayesian space–time model. Estimates of fine particulate matter (particulate matter with an aerodynamic diameter of up to 2.5 μm [PM_2.5]) and NO₂ concentrations were obtained from land use regression. Cox proportional hazards regression models were used to examine mortality associations adjusted for individual- and ecological-level covariates.

Measurements and Main Results: In single-pollutant models, we observed significant positive associations between O₃, PM_2.5, and NO₂ concentrations and all-cause and cause-specific mortality. In two-pollutant models adjusted for PM_2.5, significant positive associations remained between O₃ and all-cause (hazard ratio [HR] per 10 ppb, 1.02; 95% confidence interval [CI], 1.01–1.04), circulatory (HR, 1.03; 95% CI, 1.01–1.05), and respiratory mortality (HR, 1.12; 95% CI, 1.08–1.16) that were unchanged with further adjustment for NO₂. We also observed positive mortality associations with both PM_2.5 (both near source and regional) and NO₂ in multipollutant models.

Conclusions: Findings derived from this large-scale prospective study suggest that long-term ambient O₃ contributes to risk of respiratory and circulatory mortality. Substantial health and environmental benefits may be achieved by implementing further measures aimed at controlling O₃ concentrations.

At a Glance Commentary

Scientific Knowledge on the Subject

Tropospheric ozone (O₃) might be associated with cardiovascular disease risk and premature death. Results from long-term epidemiological studies on O₃ are scarce and inconclusive.

What This Study Adds to the Field

In this study, we examined the association between chronic ambient O₃ exposure and all-cause and cause-specific mortality in an extended analysis of the Cancer Prevention Study II, using new national-level estimates of ambient O₃, fine particulate matter (particulate matter with an aerodynamic diameter of up to 2.5 μm [PM_2.5]), and NO₂ concentrations. Results from this large-scale prospective study suggest that long-term ambient O₃ contributes to risk of respiratory and circulatory mortality. There were also positive mortality associations observed with PM_2.5 (both near source and regional) and NO₂ in multipollutant models.

Epidemiological studies investigating short-term exposures (ranging from hours to a few days) to ambient ozone (O₃) showed positive associations with mortality, exacerbation of respiratory illness, and increased hospital admissions (1). There is suggestive evidence that short-term O₃ is associated with adverse cardiovascular effects (2–4). Epidemiological studies of long-term O₃ exposure are scarce, and the causal nature of associations uncertain (4, 5).

A prior study based on 18 years of follow-up of 448,850 participants, including 118,777 deaths, in the American Cancer Society Cancer Prevention Study-II (CPS-II) showed significant positive associations between long-term (1977–2000) O₃ from available urban government monitors and both respiratory and cardiovascular mortality in single-pollutant models (6). In models adjusted for fine particulate matter (particulate matter with an aerodynamic diameter of up to 2.5 μm [PM_2.5]), only the association with respiratory mortality remained (hazard ratio [HR] per 10 ppb, 1.04; 95% confidence interval [CI], 1.01–1.07). There was a moderately high correlation between the two pollutants (r = 0.64).

The California Teachers Study showed positive associations between year-round O₃ concentrations and mortality due to respiratory causes (HR per interquartile range [11.02 ppb], 1.07; 95% CI, 0.97–1.19) and ischemic heart disease (IHD) (HR per 11.02 ppb, 1.06; 95% CI, 0.99–1.14) (7). In two-pollutant models, the association with IHD was confounded by PM_2.5. No positive mortality associations were observed in a U.K. patient cohort (8).

There was a positive association between long-term (2001–2008) county-level O₃ concentrations and chronic lower respiratory disease mortality in a U.S. ecological study (9). Positive associations were observed with cardiometabolic but not respiratory mortality in multipollutant models adjusted for PM_2.5 and NO₂ in the Canadian CanCHEC study (10). There were no data on potential individual-level behavioral confounding factors, including cigarette smoking, in either study.

The worldwide mean 3-month hourly maximum O₃ concentration in 2005 was 54 ppb (11), and it is increasing in densely populated areas of South Asia and East Asia due largely to growing O₃ precursor emissions (9). O₃ contributes to increased radiative forcing and climate change (12, 13).

On the basis of findings from the previous CPS-II study, more than 217,000 deaths caused by chronic obstructive pulmonary disease (COPD) worldwide were attributed to long-term O₃ exposure in 2013 (6, 14). Further evidence for long-term O₃ effects would markedly increase the attributable disease burden. Recent advancements in O₃ exposure assessment integrating air quality data from government monitors with estimates from photochemical models across the United States afford a unique opportunity to further examine O₃ effects in larger national-level studies.

We assessed associations between long-term ambient O₃ exposure and all-cause and cause-specific mortality in an extended analysis of the CPS-II study using new national-level estimates of ambient O₃, PM_2.5, and NO₂ concentrations. The increased number of included participants and extended follow-up period (from 1982 to 2004) resulted in nearly double the number of deaths investigated previously (6). Research designed to disentangle the independent effects of such ambient air pollutants is a key research priority (5). Some results were previously reported in the form of an abstract (15).

Methods

Study Population

CPS-II is a prospective study of nearly 1.2 million participants enrolled in all 50 U.S. states, the District of Columbia, and Puerto Rico by 77,000 volunteers in 1982. Participants were largely friends and family members of volunteers, at least 30 years of age, and had a family member aged 45 years or older. A four-page, self-administered enrollment questionnaire was used to capture data on demographic, lifestyle, medical, and other factors (16). Ethical approval for CPS-II was obtained from the Emory University School of Medicine Human Investigations Committee.

In 1984, 1986, and 1988, vital status was ascertained by study volunteers and confirmed by corresponding death certificates. After 1989, computerized linkage to the National Death Index was used (17). Through 2004, 743,543 (62.8%) participants were alive, 438,123 (37.0%) had died, and 2,921 (0.2%) were lost to follow-up or had follow-up terminated in September 1988 due to insufficient record linkage information. Deaths were classified by underlying cause using the International Classification of Diseases, 9th and 10th revisions (18, 19). More than 99% of known deaths were assigned a cause.

A total of 669,046 participants were analyzed (see Figure E1 in the online supplement). The majority of exclusions were due to missing or invalid residence (n = 385,422) (20) or covariate (n = 130,119) data (Table E1). The study cohort included 237,201 deaths in 12,662,562 person-years of follow-up.

Estimates of Ambient Air Pollution Concentrations

Estimated O₃ concentrations were obtained from the hierarchical Bayesian space–time model (HBM) of the U.S. Environmental Protection Agency and Centers for Disease Control and Prevention Environmental Public Health Tracking Network (Figure 1) (21). The HBM combines ambient measurement data from the National Air Monitoring Stations/State and Local Air Monitoring Stations with gridded estimates from the Models-3/Community Multiscale Air Quality (CMAQ) photochemical model to obtain daily 8-hour maximum O₃ concentrations in 36 × 36-km grids for the entire United States for the years 2001–2006. To coincide with our cohort follow-up, we examined both mean annual and summertime (April–September) estimates for the years 2002–2004; year 2001 estimates were omitted due to differences in model input meteorological parameters. Mean O₃ (2002–2004) values were assigned to the geocoded participant residence at enrollment and used as an indicator of long-term O₃ exposure.

Figure 1.

Distribution of mean annual daily 8-hour maximum tropospheric ozone concentrations based on a hierarchical Bayesian space–time modeling system, United States, 2002–2004.

We also examined O₃ concentrations derived from a Bayesian space–time downscaling fusion model (henceforth termed “Downscaler O₃”) (22). Daily 8-hour maximum O₃ concentrations at the census tract centroid were estimated on the basis of National Air Monitoring Stations/State and Local Air Monitoring Stations and CMAQ model data in 12 × 12-km grids for the years 2001–2008. Downscaler O₃ estimates consider all monitors, as opposed to the most prevalent monitor, where there are multiple monitors per site. We assigned mean monthly daily estimates to 545,302 CPS-II participants, as data were available only for the eastern United States for 2002–2004. Model performance using the predictive mean absolute error showed that Downscaler O₃ outperformed ordinary kriging or CMAQ models alone, with a predictive mean absolute error of 5 based on the square root of daily O₃ values (23). Correlations with holdout locations for daily predictions ranged from 0.61 to 0.86 at three sites in the eastern United States.

Estimated PM_2.5 concentrations were obtained using a national-level hybrid land use regression (LUR) and Bayesian maximum entropy (BME) interpolation model (Figure E2) (24). Monthly PM_2.5 monitoring data were collected from 1,464 sites from 1999 through 2008, with 10% reserved for cross-validation. The base LUR model that predicted PM_2.5 concentrations included traffic within 1 km and green space within 100 m³. Residual spatiotemporal variation in PM_2.5 concentrations was interpolated with a BME interpolation model. The two estimates were then combined. The cross-validation R² was approximately 0.79. Mean PM_2.5 (1999–2004) concentrations were used here. To address potential confounding of O₃–mortality associations by PM_2.5, estimates of PM_2.5 were decomposed a priori into near-source (LUR) and regional (LURBME-LUR) components to more accurately account for differences in correlation structure with O₃ (Tables E2 and E3). Results for the overall LURBME PM_2.5 are presented in Table E4 and elsewhere for selected mortality endpoints (20, 25). HBM PM_2.5 data were also examined (above) (26).

NO₂ concentrations were based on a national LUR model using regulatory monitoring (hourly data from 423 monitors) and satellite-based measurements (approximately 4 million measurements, aggregated into annual average values at 81,743 locations [approximately 10 × 10-km grids]) at the census block group level for the year 2006 (27). Additional independent variables included population; satellite-based classification of land uses, impervious surfaces, tree coverage, and distance to roadways (model R² = 0.78).

Statistical Analysis

We used Cox proportional hazards regression models to examine associations between mean O₃ (2002–2004), PM_2.5 (1999–2004), and NO₂ (2006) concentrations and all-cause and cause-specific mortality. Models were stratified by 1-year age categories, sex, and race (white, African American, or other). Follow-up time in days since enrollment was used as the time axis. The survival times of those alive at the end of follow-up were censored.

Models were adjusted a priori for the following covariates assessed at enrollment as in previous work (6, 20, 25, 28–30): education; marital status; body mass index (BMI) and BMI squared; cigarette smoking status; cigarettes per day and cigarettes per day squared; years smoked and years smoked squared; started smoking at younger than 18 years of age; passive smoking (hours); vegetable, fruit, fiber, and fat intake; beer, wine, and liquor consumption; occupational exposures; an occupational dirtiness index; and six sociodemographic ecological covariates at both the postal code and postal code minus county-level mean derived from the 1990 U.S. Census (median household income and percentage of African American residents, Hispanic residents, adults with postsecondary education, unemployment, and poverty) (Table E5).

We examined potential confounding by elevation, metropolitan statistical area size, annual average daily maximum air temperature, and 1980 percentage of air conditioning (and mean county-level residential radon concentrations for respiratory and lung cancer mortality only) (6, 29, 31–34). We also used a proportional hazards model with a random effect for county of residence at enrollment. An interaction term between O₃ and follow-up time was used to assess the proportional hazards assumption.

Threshold models, defined by setting the O₃ concentration to 0 below the threshold and the concentration minus the threshold value otherwise, were examined at 1-ppb increments across the entire exposure range. Potential modification of O₃ associations by age at enrollment, sex, education, BMI, cigarette-smoking status, passive smoking, prior cardiovascular disease (high blood pressure, heart disease, stroke, or diabetes) or respiratory disease (asthma, emphysema, or chronic bronchitis) at enrollment, and temperature was assessed using multiplicative interaction terms. Two-sided P values based on the likelihood ratio statistic were calculated to assess their significance.

For analyses, we used SAS version 9.2 software (SAS Institute, Cary, NC) and specialized software developed for the random effects survival model (30, 35). Ethical approval for our present analysis was obtained from the Ottawa Hospital Research Ethics Board.

Results

Participants were largely between the ages of 40 and 69 years, female, and had more than a high school education (Table 1). There was little variation in O₃ concentrations by participant characteristics.

Table 1.

Distribution of Selected Participant Characteristics at Enrollment (1982), American Cancer Society Cancer Prevention Study II Cohort, United States (n = 669,046)

Characteristic	n (%)*	HBM O3 (ppb)	LURBME PM2.5 (μg/m3)	LUR NO2 (ppb)
Age, yr
<40	29,615 (4.4)	38.0 (3.9)	12.8 (2.9)	12.4 (5.7)
40–49	137,618 (20.6)	38.0 (3.9)	12.5 (2.8)	11.4 (5.1)
50–59	245,195 (36.7)	38.1 (3.9)	12.6 (2.8)	11.6 (5.0)
60–69	178,062 (26.6)	38.3 (4.1)	12.5 (2.9)	11.6 (5.1)
70–79	66,527 (9.9)	38.4 (4.2)	12.6 (2.9)	11.8 (5.0)
≥80	12,029 (1.8)	38.3 (4.2)	12.7 (2.9)	12.2 (5.2)
Race
White	632,919 (94.6)	38.2 (3.9)	12.5 (2.8)	11.5 (5.0)
African American	25,508 (3.8)	38.1 (3.3)	13.7 (2.5)	13.3 (5.2)
Other	10,619 (1.6)	38.3 (5.9)	12.9 (4.3)	15.6 (6.4)
Sex
Male	292,772 (43.8)	38.2 (4.0)	12.5 (2.8)	11.5 (5.1)
Female	376,274 (56.2)	38.2 (4.0)	12.6 (2.9)	11.7 (5.1)
Education
Less than high school	78,391 (11.7)	38.1 (3.8)	12.8 (2.8)	11.6 (5.3)
High school	207,710 (31.1)	38.1 (3.7)	12.6 (2.8)	11.4 (5.1)
High school or above	382,945 (57.2)	38.2 (4.1)	12.5 (2.9)	11.7 (5.0)
BMI, kg/m2
<18.5	11,904 (1.8)	38.4 (4.0)	12.6 (2.9)	11.7 (5.0)
18.5–24.9	338,528 (50.6)	38.2 (4.0)	12.5 (2.9)	11.6 (5.1)
25–29.9	242,144 (36.2)	38.1 (3.9)	12.6 (2.8)	11.6 (5.1)
≥30	76,470 (11.4)	38.1 (3.8)	12.8 (2.8)	11.7 (5.2)
Marital status
Single	21,966 (3.3)	37.5 (3.8)	13.0 (2.8)	13.1 (5.5)
Married	564,186 (84.3)	38.2 (4.0)	12.5 (2.8)	11.5 (5.0)
Other	82,894 (12.4)	38.1 (4.0)	12.9 (2.9)	12.3 (5.3)
Cigarette-smoking status
Never	299,530 (44.8)	38.4 (4.0)	12.6 (2.9)	11.5 (5.1)
Current	129,876 (19.4)	38.0 (3.8)	12.7 (2.8)	11.8 (5.1)
Former	172,689 (25.8)	38.0 (4.0)	12.5 (2.9)	11.7 (5.1)
Ever pipe and/or cigar	66,951 (10.0)	38.0 (3.8)	12.5 (2.8)	11.5 (5.0)
1990 ecological covariates, mean (SD)
Median household income, $10,000s	3.5 (1.3)	—	—	—
African American, %	8.9 (15.8)	—	—	—
Hispanic, %	5.9 (10.9)	—	—	—
Postsecondary education, %	38.6 (13.3)	—	—	—
Unemployment, %	5.6 (2.9)	—	—	—
Poverty, %	10.5 (7.8)	—	—	—

Definition of abbreviations: BMI = body mass index; HBM = hierarchical Bayesian space–time model; LUR = land use regression; LURBME = land use regression Bayesian maximum entropy; O₃ = tropospheric ozone; PM_2.5 = particulate matter with an aerodynamic diameter of up to 2.5 μm.

^*Except where otherwise indicated.

Year-round O₃ concentrations ranged from 26.7 to 59.3 ppb, with a mean (SD) of 38.2 (4.0) ppb (Table E2). For PM_2.5 and NO₂, average concentrations ranged from 1.4 to 27.9 μg/m³ (mean [SD], 12.6 [2.9] μg/m³) and 1.0 to 37.6 ppb (mean [SD], 11.6 [5.1] ppb), respectively. Correlations between year-round O₃ and both near-source and regional PM_2.5 were weak (Pearson’s r = −0.13 and +0.23, respectively) (Table E3).

In single-pollutant models, significant positive associations were present between year-round O₃ and all-cause (HR per 10 ppb, 1.02; 95% CI, 1.01–1.04), circulatory (HR per 10 ppb, 1.03; 95% CI, 1.02–1.05), and respiratory (HR per 10 ppb, 1.14; 95% CI, 1.10–1.18) mortality (Table E6). Significant positive associations were observed for both regional PM_2.5, ranging up to HR per 10 μg/m³ of 1.16 (95% CI, 1.10–1.23) for respiratory mortality, and near-source PM_2.5, ranging up to HR per 10 μg/m³ of 1.45 (95% CI, 1.35–1.57) for circulatory mortality. There were significant positive associations between NO₂ and all-cause (HR per 10 ppb, 1.04; 95% CI, 1.03–1.06) and circulatory (HR per 10 ppb, 1.08; 95% CI, 1.06–1.09) mortality but not respiratory mortality.

In two-pollutant models adjusted for PM_2.5, significant positive associations between O₃ and all-cause (HR per 10 ppb, 1.02; 95% CI, 1.01–1.04), circulatory (HR per 10 ppb, 1.03; 95% CI, 1.01–1.05), and respiratory (HR per 10 ppb, 1.12; 95% CI, 1.08–1.16) mortality were observed (Table E7). Results were unchanged with further adjustment for NO₂ (Table 2). The strongest O₃ association was noted for diabetes mortality specifically (HR, 1.16; 95% CI, 1.07–1.26), followed by mortality due to dysrhythmias, heart failure, and cardiac arrest; COPD; and pneumonia and influenza. Significant positive mortality associations also remained for both regional and near-source PM_2.5 in the multipollutant model. For NO₂, the association with circulatory mortality attenuated (HR per 10 ppb, 1.03; 95% CI, 1.01–1.05) and that with all-cause mortality was not apparent.

Table 2.

All-Cause and Cause-Specific Mortality Hazard Ratios in Relation to Each 10-Unit Increase in Air Pollutant Concentrations, 1982–2004 Follow-up in American Cancer Society Cancer Prevention Study II Cohort, United States (n = 669,046)

Cause of Death	ICD-9 Codes; ICD-10 Codes	Deaths (n)	Multipollutant Model Data, Fully Adjusted HR* (95% CI)
			HBM O3	Regional PM2.5	Near-Source PM2.5	LUR NO2
All-cause mortality	All	237,201	1.02 (1.01–1.04)	1.04 (1.02–1.06)	1.26 (1.19–1.34)	1.01 (1.00–1.03)
Diseases of the circulatory system (plus diabetes) (48)	390–459, 250; I00–I99, E10–E14	105,039	1.03 (1.01–1.05)	1.07 (1.04–1.10)	1.41 (1.29–1.54)	1.03 (1.01–1.05)
Cardiovascular	410–440; I20–I25, I30–I51, I60–I69, I70	84,132	1.03 (1.01–1.05)	1.07 (1.04–1.10)	1.35 (1.23–1.49)	1.03 (1.01–1.06)
Ischemic heart disease	410–414; I20–I25	45,644	0.98 (0.95–1.00)	1.06 (1.02–1.11)	1.40 (1.23–1.60)	1.09 (1.06–1.12)
Dysrhythmias, heart failure, cardiac arrest	420–429; I30–I51	18,314	1.15 (1.10–1.20)	1.06 (1.00–1.13)	1.15 (0.93–1.42)	0.99 (0.95–1.04)
Cerebrovascular disease	430–438; I60–I69	17,085	1.03 (0.98–1.07)	1.13 (1.06–1.21)	1.50 (1.21–1.87)	0.92 (0.88–0.97)
Diabetes	250; E10–E14	4,890	1.16 (1.07–1.26)	1.01 (0.90–1.15)	2.02 (1.33–3.07)	1.01 (0.92–1.10)
Diseases of the respiratory system	460–519; J00–J98	20,484	1.12 (1.08–1.16)	1.11 (1.05–1.18)	1.17 (0.96–1.42)	0.99 (0.95–1.04)
Pneumonia and influenza	480–487; J10–J18	6,599	1.10 (1.03–1.18)	1.24 (1.12–1.37)	1.01 (0.71–1.42)	1.07 (0.99–1.15)
COPD and allied conditions	490–496; J19–J46	9,967	1.14 (1.08–1.21)	1.06 (0.97–1.15)	1.24 (0.94–1.64)	0.97 (0.91–1.04)
Lung cancer	162; C33–34	16,432	0.96 (0.91–1.00)	1.13 (1.06–1.21)	1.31 (1.05–1.63)	0.94 (0.90–0.99)

Definition of abbreviations: CI = confidence interval; COPD = chronic obstructive pulmonary disease; HBM = hierarchical Bayesian space–time model; HR = hazard ratio; ICD-9 = International Classification of Diseases, Ninth Revision; ICD-10 = International Classification of Diseases, 10th Revision; LUR = land use regression; O₃ = tropospheric ozone; PM_2.5 = particulate matter with an aerodynamic diameter of up to 2.5 μm.

^*HRs are derived from multipollutant models including all air pollutants simultaneously. HRs are age, race, and sex stratified and adjusted for education; marital status; body mass index; body mass index squared; cigarette smoking status; cigarettes smoked per day and cigarettes per day squared; years smoked and years smoked squared; age started smoking younger than 18 years; passive smoking; vegetable, fruit, and fiber intake; fat intake; beer, wine, and liquor intake; industrial exposures; an occupational dirtiness index; and 1990 ecological covariates: median household income and percentage of African Americans, Hispanics, postsecondary education, unemployment, and poverty.

Results for O₃ strengthened slightly for respiratory mortality with adjustment for percentage with air conditioning (Table E8). Results were slightly attenuated for both for circulatory (HR, 1.03; 95% CI, 1.00–1.05) and respiratory (HR, 1.11; 95% CI, 1.06–1.16) mortality with inclusion of a county-level random effect.

Similar results were observed using summer O₃ concentrations, except for mortality due to dysrhythmias, heart failure, and cardiac arrest; diabetes; and respiratory causes. The latter results were attenuated (Table E9). Results were similar for O₃ when adjusted for HBM PM_2.5 compared with decomposed LURBME PM_2.5 concentrations (Table E10). Results were slightly stronger using estimated Downscaler O₃ concentrations (HR per 10 ppb, 1.05; 95% CI, 1.03–1.07 [all-cause mortality]; HR per 10 ppb, 1.06; 95% CI, 1.03–1.09 [circulatory mortality]; HR per 10 ppb, 1.14; 95% CI, 1.07–1.21 [respiratory mortality]), due largely to the subsample of included participants (Tables E2, E3, and E11).

The proportional hazards assumption was violated (P < 0.05) for associations between O₃ and all-cause, circulatory, cardiovascular, and IHD mortality, with positive associations (except null results for IHD) observed in the middle 1990–1999 and later 2000–2004 time periods only, although the magnitude of the differences was small (Table E12). There was some evidence that a threshold model improved model fit for respiratory mortality at 35 ppb (P = 0.002) compared with a linear model using year-round but not summertime O₃ (HR per 10 ppb using threshold O₃ indicator at 35 ppb for respiratory mortality, 1.17; 95% CI, 1.11–1.22) (Figures E3 and E4). Results were somewhat suggestive of a threshold for circulatory mortality at 35 ppb (P = 0.07).

O₃ circulatory and respiratory mortality associations varied according to temperature and prior cardiovascular or respiratory disease at enrollment, respectively (Table E13). Positive respiratory associations were also stronger in those younger than 65 years of age at enrollment.

For comparability across pollutants, results according to each fifth percentile–mean increment are presented in Table 3. Results were somewhat stronger with near-source PM_2.5 for both all-cause (HR per 1.6 μg/m³, 1.04; 95% CI, 1.03–1.05) and circulatory (HR per 1.6 μg/m³, 1.06; 95% CI, 1.04–1.07) mortality and with O₃ for respiratory mortality (HR per 7.1 ppb, 1.08; 95% CI, 1.05–1.11).

Table 3.

All-Cause and Cause-Specific Mortality Hazard Ratios in Relation to Each Fifth Percentile–Mean Increment in Air Pollutant Concentrations, 1982–2004 Follow-up in American Cancer Society Cancer Prevention Study II Cohort, United States (n = 669,046)

Cause of Death	ICD-9 Codes; ICD-10 Codes	Deaths (n)	Multipollutant Model Data, Fully Adjusted HR* (95% CI)
			HBM O3 (per 7.1 ppb)	Regional PM2.5 (per 4.5 μg/m3)	Near-Source PM2.5 (per 1.6 μg/m3)	LUR NO2 (per 6.5 ppb)
All-cause	All	237,201	1.02 (1.01–1.03)	1.02 (1.01–1.03)	1.04 (1.03–1.05)	1.01 (1.00–1.02)
Diseases of the circulatory system (plus diabetes) (48)	390–459, 250; I00–I99, E10–E14	105,039	1.02 (1.01–1.03)	1.03 (1.02–1.04)	1.06 (1.04–1.07)	1.02 (1.00–1.03)
Cardiovascular	410–440; I20–I25, I30–I51, I60–I69, I70	84,132	1.02 (1.01–1.03)	1.03 (1.02–1.04)	1.05 (1.03–1.07)	1.02 (1.01–1.04)
Ischemic heart disease	410–414; I20–I25	45,644	0.98 (0.97–1.00)	1.03 (1.01–1.05)	1.06 (1.03–1.08)	1.06 (1.04–1.08)
Dysrhythmias, heart failure, cardiac arrest	420–429; I30–I51	18,314	1.10 (1.07–1.14)	1.03 (1.00–1.06)	1.02 (0.99–1.06)	1.00 (0.97–1.03)
Cerebrovascular disease	430–438; I60–I69	17,085	1.02 (0.99–1.05)	1.06 (1.02–1.09)	1.07 (1.03–1.11)	0.95 (0.92–0.98)
Diabetes	250; E10–E14	4,890	1.11 (1.05–1.18)	1.01 (0.95–1.06)	1.12 (1.05–1.20)	1.01 (0.95–1.07)
Diseases of the respiratory system	460–519; J00–J98	20,484	1.08 (1.05–1.11)	1.05 (1.02–1.08)	1.03 (0.99–1.06)	1.00 (0.97–1.02)
Pneumonia and influenza	480–487; J10–J18	6,599	1.07 (1.02–1.12)	1.10 (1.05–1.15)	1.00 (0.95–1.06)	1.04 (0.99–1.10)
COPD and allied conditions	490–496; J19–J46	9,967	1.10 (1.05–1.14)	1.03 (0.99–1.07)	1.04 (0.99–1.08)	0.98 (0.94–1.02)
Lung cancer	162; C33–34	16,432	0.97 (0.94–1.00)	1.06 (1.03–1.09)	1.04 (1.01–1.08)	0.96 (0.93–0.99)

Definition of abbreviations: CI = confidence interval; COPD = chronic obstructive pulmonary disease; HBM = hierarchical Bayesian space–time model; HR = hazard ratio; ICD-9 = International Classification of Diseases, Ninth Revision; ICD-10 = International Classification of Diseases, 10th Revision; LUR = land use regression; O₃ = tropospheric ozone; PM_2.5 = particulate matter with an aerodynamic diameter of up to 2.5 μm.

^*HRs are derived from multipollutant models including all air pollutants simultaneously. HRs are age, race, and sex stratified and adjusted for education; marital status; body mass index; body mass index squared; cigarette smoking status; cigarettes smoked per day and cigarettes per day squared; years smoked and years smoked squared; age started smoking younger than 18 years; passive smoking; vegetable, fruit, and fiber intake; fat intake; beer, wine, and liquor intake; industrial exposures; an occupational dirtiness index; and 1990 ecological covariates: median household income and percentage of African Americans, Hispanics, postsecondary education, unemployment, and poverty.

Discussion

We observed significant positive associations between long-term O₃ and all-cause, circulatory, and respiratory mortality with 2%, 3%, and 12% increases in risk per 10 ppb, respectively, in this large-scale study with 22 years of follow-up. In a smaller prior study, researchers first reported long-term O₃–mortality associations, but only that for respiratory mortality remained with adjustment for PM_2.5 (6). We hypothesized that with improved exposure models and increased statistical power derived from longer follow-up, robust associations between O₃ and both respiratory and circulatory mortality would be observed with adjustment for other copollutants. Results supporting our hypothesis were robust after adjustment for PM_2.5 and NO₂.

We used new national-level O₃ exposure estimates extending our earlier work based on measured regional levels of air pollutants. Although there were also differences in the time period (2002–2004 vs. 1977–2000) and season (year-round vs. summertime) of O₃ metrics used in the present study versus the previous study, their correlation was moderately strong (r = 0.70) (6). Analysis linking the current HBM O₃ estimate to the previous analytic cohort showed stronger respiratory mortality HRs of 1.16 (95% CI, 1.09–1.24) per 10 ppb and 1.11 (95% CI, 1.06–1.17) per each fifth percentile–mean increment in two-pollutant models adjusted for PM_2.5 (6). In comparison, HRs of 1.04 (95% CI, 1.01–1.07) per 10 ppb and 1.07 (95% CI, 1.02–1.12) per each fifth percentile–mean increment were observed using measured O₃ data in previous work, indicating larger associations with more refined O₃ estimates (6).

Our findings support previous associations with respiratory mortality in U.S. studies (6, 7). Possible biological mechanisms include oxidative stress and inflammatory pathways as well as adverse neural, epithelial, smooth muscle, and immune system impacts (4). In contrast, no positive association was observed in a large (n = 800,000) U.K. patient cohort (8). There O₃ was negatively correlated with PM_2.5 (r = −0.39), and regional patterns in O₃ and mortality may explain the findings observed. There was no positive association with respiratory mortality in the CanCHEC study (10).

We observed a positive association with circulatory mortality that remained after adjustment for PM_2.5 and NO₂. Results from some previous studies were confounded by PM_2.5 (6, 7). The researchers in the prior U.K. study reported no positive association with incident myocardial infarction, stroke, arrhythmia, or heart failure (36). Analysis of California CPS-II participants revealed a positive association with IHD mortality (HR per interquartile range [24.2 ppb], 1.10; 95% CI, 1.02–1.19), which remained after adjusting for PM_2.5 and NO₂ (29). Although we observed no association with IHD mortality here, upon restriction to California participants there was a weak positive association (HR per 10 ppb, 1.07; 95% CI, 0.99–1.14). Regional differences in these findings may relate to participant characteristics, death rates, death certificate coding, or air pollution composition. Positive associations with cardiometabolic disease mortality were observed in multipollutant models in the CanCHEC study, with the strongest findings being for diabetes (11% increase per 9.5 ppb) and IHD (6% increase) using a 21 × 21-km grid surface (10).

Findings for respiratory mortality were stronger among those with no prior respiratory disease at enrollment, suggesting a role for O₃ in the development and exacerbation of disease. In a previous Medicare-based study, researchers reported positive associations between long-term O₃ and mortality in those previously hospitalized for COPD, as well as in those with diabetes, congestive heart failure, or myocardial infarction (37). Findings were also stronger in younger participants. Little age difference was observed in short-term studies (38).

Stronger associations were observed in areas with lower temperature as well as in the highest temperature category. The relationship between ambient and personal O₃ exposure is complex, weaker than with particles, and varies with time spent outdoors, indoor infiltration, and season (39). Time spent outdoors and engaged in sports may modify associations for asthma formation (40). Differences in short-term O₃ mortality coefficients across 18 U.S. cities were related to differences in total O₃ (outdoor plus indoor) exposure (41). Mitigation of short-term O₃-related mortality by air conditioning prevalence was observed at high temperatures in 97 U.S. cities (42). Though results suggest a threshold for respiratory and, to a lesser extent, circulatory mortality at 35 ppb of O₃ compared with that of 56 ppb (P = 0.06) in previous work (6), it remains unclear if the results represent a biological threshold per se or other such temperature-related, behavioral, or region-related processes. There were no data on time–activity patterns here.

We observed positive mortality associations with estimated LURBME PM_2.5 concentrations (30). The LURBME PM_2.5 model outperformed other remote sensing, geostatistical, and HBM models in CPS-II (25). Results were somewhat stronger with regional PM_2.5 for respiratory mortality and near-source PM_2.5 for circulatory mortality.

Correlations between O₃ and PM_2.5 were weak. Though air pollutants were estimated at different time periods using different methods and geographic units of scale, possibly complicating interpretation of their correlation structure, results for O₃ were robust in two-pollutant models adjusted for HBM PM_2.5 and estimated for the same time period and unit of scale. Their correlation was also weak (r = 0.04) (Tables E3 and E10). Results for O₃ were similar using Downscaler O₃ concentrations, estimated with a finer spatial resolution (12 × 12 km) (Table E11), compared with HBM O₃. The correlation between HBM and Downscaler O₃ was strong (r = 0.89) (Table E3).

We adjusted findings for decomposed LURBME PM_2.5 and NO₂ to address potential confounding in areas with low O₃ but high near-source pollution (43). Negative correlations between Downscaler O₃ and both near-source PM_2.5 and NO₂ (r = −0.41 and −0.42, respectively) were stronger than those for HBM O₃ (r = −0.13 and −0.08, respectively). Downscaler HRs for all-cause and circulatory mortality also increased to a greater extent in multipollutant models compared with those for HBM O₃ (Tables E3, E6, and E11).

Positive associations between NO₂ and circulatory mortality were attenuated with adjustment for PM_2.5 and O₃. The association with all-cause mortality was no longer apparent. Currently, there is suggestive evidence for NO₂-associated cardiovascular effects but uncertainty regarding its independent role (5, 44, 45). Results in two European studies revealed small positive associations with nonaccidental mortality that remained in two-pollutant models with PM_2.5 (46, 47). There were also small positive mortality associations in multipollutant models with both PM_2.5 and O₃ in two North American studies (10, 29). Williams and colleagues (43) noted close linkages between NO₂ and O₃. Correlations between NO₂ and HBM O₃ (r = −0.08) were weak, possibly due to different units of scale and broad regional patterns of pollutants.

Limitations of this study include a lack of updated data on residential history, leading to potential misclassification of both air pollution concentrations and sociodemographic ecologic covariates over time, as well as of individual covariate data. There were no data on residential history before enrollment. Accounting for residential mobility, however, had little impact on long-term O₃ or PM_2.5 mortality associations in the CanCHEC study but strengthened those for more spatially resolved NO₂ (10). Though there may be some selection bias due to the exclusion of participants with missing or invalid residence data, excluded participants were similar in terms of baseline sociodemographic factors (approximately 61% between 50 and 69 years of age, approximately 57% female, approximately 47% with more than high school level of education, and approximately 21% current cigarette smokers). Though we lacked historical O₃ data, there was little difference in respiratory mortality HRs in previous work when researchers examined specific exposure time windows or O₃ exposures matched more closely in time (5). Recent O₃ concentrations are correlated with past estimates. Correlations between 1998–2000 concentrations and those from 1988–1990 and 1978–1980 were 0.80 and 0.58, respectively (6). Little is known regarding the most relevant exposure time window. Finally, multiple comparisons were performed, and some results may be due to chance.

In summary, findings derived from this large-scale prospective study suggest that long-term ambient O₃ contributes to risk of respiratory and circulatory mortality. Results were robust after adjustment for PM_2.5 and NO₂. There were also positive mortality associations observed between PM_2.5 (both near source and regional) and NO₂ in multipollutant models. Substantial health and environmental benefits may be achieved by implementing further measures aimed at controlling O₃ concentrations.