Air Pollutants and Breast Cancer Risk: A Parallel Analysis of Five Large US Prospective Cohorts

Abstract

Objectives.

To determine whether outdoor air pollution exposure is associated with breast cancer incidence.

Methods.

Residential-level concentrations of nitrogen dioxide (NO₂, ppb), fine particulate matter (PM_2.5, µg/m³), and ozone (ppb) were estimated for Nurses’ Heath Studies, Women’s Health Initiative Clinical Trials and Observational Study Cohort, and Sister Study participants using high-resolution spatio-temporal models. Cox proportional hazards regression estimated cohort-specific hazard ratios (HRs) and 95% confidence intervals (CIs) and a random effects model determined summary HRs, overall and by estrogen (ER)/progesterone receptor (PR) subtype and census region.

Results.

NO₂ was positively associated with overall breast cancer incidence (N=28,811 cases, HR=1.03, 95%CI: 1.00–1.05), with little variation by subgroups. PM_2.5 was associated with higher incidence of ER-/PR- tumors (N=2,367 cases, HR=1.14, 95%CI: 1.04–1.24; p-heterogeneity=0.0004) and with higher overall incidence in the Midwest (HR=1.15, 95%CI: 1.01–1.32; p-heterogeneity=0.01). Ozone was not associated with overall incidence, but was associated with ER-/PR- tumors (N=3,406 cases, HR=1.10, 95%CI: 1.00–1.21; p-heterogeneity=0.03).

Conclusion.

In this largest US study to date, we confirmed an association between NO₂ and breast cancer, and present novel associations of PM_2.5 and ozone with ER-/PR- tumors.

The incidence of breast cancer in the US has been generally increasing since the mid-1980s.¹ Most established risk factors for breast cancer are related to reproductive and lifestyle factors and largely reflect associations with estrogen receptor (ER)-positive disease.^2,3 Research on environmental risk factors suggests that chemical and pollutant exposures contribute to breast cancer incidence, particularly for women who may experience higher susceptibility.^4,5

Outdoor air pollution exposure (e.g., particulate matter [PM], nitrogen oxides [NO_x, NO₂], ozone [O₃]) has been classified by the International Agency for Research on Cancer (IARC) as a Group 1 carcinogen, based on evidence for fine particulate matter (PM_2.5) and lung cancer.⁶ Air pollution is hypothesized to impact carcinogenesis through altered gene expression, inflammation, and immune and oxidative stress responses.⁶ Air pollution also includes compounds with endocrine disrupting properties⁷ and thus may be particularly relevant to breast cancer given its relationship to hormonal factors.

Previous studies of NO₂ and breast cancer, including three meta-analyses, provide evidence of a modest association (HRs/RRs/ORs of 1.02–1.06 per 10-µg/m³ increase in exposure).^8–13 Although earlier meta-analyses have observed null associations for PM_2.5 and breast cancer,^9,13 more recent studies, including a meta-analysis of 11 cohort studies, have observed positive associations (HR=1.03–1.14 per 5-µg/m³ increase in exposure).^{10,11,14–17} The few studies of ozone have been null.^10,18,19

Despite a growing literature, well-powered studies considering tumor heterogeneity are limited.^{16,17,19–23} Although geographic heterogeneity of air pollution is well documented,^23,24 regional differences in observed associations are not well-established.^16,19,23 The aim of this study was to conduct parallel analyses using data from five large, prospective US cohorts to evaluate the associations between NO₂, PM_2.5, ozone, and incident breast cancer, overall and by tumor subtype and geographic region, and to summarize findings across cohorts.

Methods

Study populations

Study cohorts include Nurses’ Health Study (NHS), Nurses’ Health Study II (NHSII), Sister Study (SIS), Women’s Health Initiative Clinical Trials (WHI-CT), and Women’s Health Initiative Observational Study (WHI-OS). NHS recruited 121,700 female nurses from 11 US States in 1976 who were 30–55 years of age.²⁵ NHSII recruited 116,430 female nurses from 14 US States in 1989 who were 25–42 years of age.²⁵ By the mid-1990s, participants in NHS and NHSII lived in all 50 States. A total of 161,808 postmenopausal women (ages 50–79) enrolled in one or more of the overlapping, randomized WHI-CT of menopausal hormone therapy (HT), calcium/vitamin D supplementation, and diet modification or the WHI-OS from 1993–1998.^26–28 Participants in WHI-CT and WHI-OS were recruited from 40 clinical centers nationwide, located in 23 states and Washington, D.C. WHI-CT participants who participated in the HT arm were initially assigned to one of four treatments: estrogen, estrogen control, estrogen plus progestin, estrogen plus progestin control. From 2003–2009, SIS recruited 50,884 women (ages 35–74) from all 50 states and Puerto Rico with at least one full or half-sister with a previous diagnosis of breast cancer.²⁹

Outcome assessment

In all studies, incident breast cancer was reported by participants during routine follow-up and verified by medical records (as well as physician adjudication for all studies except SIS) and pathology reports when available. This analysis was limited to diagnoses confirmed by medical record or death certificate/National Death Index records; for SIS, unconfirmed diagnoses were censored.

Exposure assessment

Biweekly air pollutant concentrations were generated at all participant residential addresses using a synthesized suite of regional, spatio-temporal models derived from a combination of data from 1,500 regulatory monitors, >900 research monitors at residential and community locations, and >200 geographic covariates developed for large-scale epidemiologic research and calculated at each monitoring and residential location.^30,31 For NO₂, gradient monitoring was additionally conducted near roadways. Air pollutant models accounted for complex spatiotemporal dependencies in a land use and spatial smoothing framework.^32,33 The cross-validation method was used to evaluate air pollutant model predictive performance with R²=0.87 for NO₂, R²=0.89 for PM_2.5, and R²=0.73 for ozone.³⁰ NO₂ and ozone concentrations were available at the earliest starting January 1990, and PM_2.5 concentrations were available beginning in January 1999.

Person-months of follow-up were calculated from cohort- and pollutant-specific analytic baselines (e.g., starting in 1992 for NO₂ and ozone in NHS and NHSII) through breast cancer diagnosis, death, loss to follow-up, or end of follow-up in 2016–2017 (May 2016 for NHS and NHSII, June 2017 for SIS, and 2017 for WHI-CT and WHI-OS; Figure S1). Women with a diagnosis of breast cancer prior to the analytic baseline were excluded, as well as those with uncertain timing (n=3,920 for NHS, n=21 for NHS2, n=79 for SIS, n=101 for WHI-CT, n=5,444 for WHI-OS). Women diagnosed with other types of cancer were retained in the analysis. Air pollutant concentrations were analyzed using a long-term exposure estimate calculated as time-varying 24-month rolling averages weighted by address history. Person-months were included when at least 23 prior months of estimates were available. Follow-up began at the earliest in 1992 for NO₂ and ozone, and 2001 for PM_2.5. For SIS, enrollment started later than the other cohorts (2003–2009); pollutant estimates were available starting at the time of enrollment with the 24-month average at baseline based on self-reported and LexisNexis-derived residential histories.³⁴ For WHI-CT and WHI-OS, pollutant estimates prior to enrollment were extended back to 1990.

Covariate assessment

For each cohort we ascertained information on demographic and socioeconomic characteristics, medical and reproductive history, and behavioral risk factors (e.g., smoking status, alcohol intake, and physical activity) at the time of analytic baseline.^25–29 Routine follow-up questionnaires updated reproductive factors, behavioral risk factors, and medical diagnoses. Menopausal status was updated throughout follow-up for those who were premenopausal at the analytic baseline. Census region was updated throughout follow-up for all participants.

The neighborhood socioeconomic status (nSES) index was developed based on data from the 2000 US Census³⁵ and the 2005–2009 and 2007–2011 cycles of the American Community Survey,^36,37 with a higher value indicating greater neighborhood deprivation.³⁸ This index was estimated for the analytic baseline address and updated for 1999/2000 addresses for WHI, NHSI and NHSII.

Statistical analysis

Cohorts contributed individual-level data and were analyzed in parallel using the same statistical models to produce cohort-specific results that were then summarized across cohorts. The analysis was restricted to women with available data on air pollutant exposure (unavailable for women living outside the contiguous US), race and ethnicity, educational attainment, census region of residence, and nSES (Figure S2). We calculated Spearman correlation coefficients for the associations between each of the pollutants and nSES.

We used Cox proportional hazards models to examine the association between each air pollutant and breast cancer incidence, overall and by tumor invasiveness [invasive; ductal carcinoma in situ (DCIS)], ER status (overall and invasive only), joint ER/PR status (overall and invasive only), and ER/PR status among premenopausal and postmenopausal women separately. Hazard ratios (HRs) and 95% confidence intervals (CIs) were calculated for a 10-ppb increase in NO₂ and ozone and a 5-µg/m³ increase in PM_2.5. Linearity was assessed using likelihood ratio tests comparing the model with only the linear term to the model with linear and restricted cubic spline terms, and no violations were identified. For each cohort, the dataset was structured with one record for each person-month of follow-up, and all models were stratified by age. In NHS and NHSII, age and calendar month were used as stratifying variables and an indicator variable was used as the time scale. In SIS, the time scale was calendar month. In WHI-CT and WHI-OS, the time scale was time on study because sharing of calendar time was restricted, and models were further stratified by masked (non-ordered) enrollment year to adjust for temporal effects.

The covariate adjustment set was selected using a directed acyclic graph (Figure S3) and included self-reported race and ethnicity (Black including Hispanic/Latina; non-Black Hispanic/Latina; non-Hispanic White; “Other,” including Asian/Pacific Islander and American Indian), educational attainment (high school or less; some college/associate’s degree; bachelor’s degree or higher), nSES score (continuous), and time-varying census region (Northeast; Midwest; South; West). In WHI-CT, all models were also adjusted for HT treatment arm (estrogen, estrogen control, estrogen plus progestin, estrogen plus progestin control).

For each air pollutant, summary effect estimates were calculated using a random effects model that pooled study-specific effect estimates that were adjusted for the primary covariate adjustment set. We examined heterogeneity in the hazard ratios estimated across studies using the Cochran Q test.

We evaluated effect measure modification by several covariates (race and ethnicity, analytic baseline nSES quartile, time-varying primary census region, body mass index [BMI], and menopausal status, time-varying for those premenopausal at enrollment). We evaluated modification by race and ethnicity and nSES as air pollution exposure in the US is inequitably distributed, with racial and ethnic minorities and those living in areas of lower socioeconomic status experiencing higher exposure.^39,40 We considered geographic region to evaluate potential heterogeneity by pollution composition²³ and menopausal status given that previous studies observed variability.^15,41,42 We evaluated modification by BMI because particulate matter constituents can act as endocrine disrupting chemicals,⁷ which are lipophilic and accumulate in adipose tissue.⁴³ To assess modification within studies, we used likelihood ratio tests to compare models with and without pollutant by covariate cross-product terms. For the summary estimate, we used the contrast test method⁴⁴ (p-heterogeneity) to evaluate heterogeneity across subtypes and subgroups.

Sensitivity analyses included 1) a 10-year lag analysis and to evaluate the potential of residual confounding, 2) a model co-adjusting for the other air pollutants (e.g., NO₂ model adjusted for PM_2.5 and ozone), 3) an expanded adjustment set, which included breast cancer risk factors: smoking status, alcohol intake, physical activity, BMI, age at first birth, parity, breastfeeding, oral contraceptive use, age at menopause, postmenopausal hormone use (in addition to the HT arm in WHI-CT), and family history of breast cancer (mother or sister), and 4) using cumulative average air pollutant exposure (overall and restricted to time-varying age <60 years). At the start of follow-up, cumulative averages were calculated using all available air pollutant estimates prior to baseline and were updated to reflect the average value throughout follow-up in a time-dependent manner.

Results

Mean follow-up time overall was 17.51 years (range: 9.76 [SIS] to 23.34 [NHSII]). The NO₂ and ozone analyses included 28,811 cases and the PM_2.5 analysis included 20,197 cases; study characteristics are shown in Table 1. Most correlations between NO₂ and PM_2.5 were moderate, whereas ozone was inversely correlated with NO₂ and PM_2.5 (Figure S4).

Table 1.

Selected analytic baseline characteristics of participants in five US-based cohorts (N=416,537 participants for NO₂ and ozone analyses, N=393,547 for PM_2.5 analysis)

Characteristicsa	NHS	NHSII	SIS	WHI-CT	WHI-OS
NO2
Analytic N	108,642	113,743	49,126	64,117	80,833
Analytic baseline year(s)	1992	1992	2003–2008	1993–1998	1993–1998
24-month avg (ppb), median (IQR)	11.4 (7.5, 17.2)	11.9 (7.9, 17.4)	8.5 (5.8, 11.9)	14.1 (9.7, 19.2)	14.0 (9.8, 19.0)
PM2.5
Analytic N	97,759	111,597	49,126	59,844	75,315
Analytic baseline year(s)	2001	2001	2003–2008	2001	2001
24-month avg (µg/m3), median (IQR)	12.0 (10.3, 14.0)	12.9 (11.0, 14.5)	10.8 (8.8, 12.4)	12.9 (10.5, 14.9)	13.0 (10.6, 15.0)
Ozone
Analytic N	108,642	113,743	49,126	64,117	80,833
Analytic baseline year(s)	1992	1992	2003–2008	1993–1998	1993–1998
24-month avg (ppb), median (IQR)	24.8 (22.8, 27.0)	24.7 (22.7, 26.9)	26.6 (24.5, 28.7)	24.0 (21.4, 26.1)	24.1 (21.7, 26.2)
Total follow-up time, mean	19.72	23.34	9.76	14.61	13.61
Age, %
<40	0.0	64.4	4.1	0.0	0.0
40–49	14.0	35.6	23.9	0.0	0.0
50–59	41.5	0.0	39.1	34.7	32.3
60–69	39.2	0.0	26.8	46.2	44.0
70–79	5.3	0.0	6.0	19.1	23.7
Age, mean ± SD	58.7 ± 7.2	37.4 ± 4.7	55.7 ± 9.0	62.8 ± 7.0	63.5 ± 7.4
nSES score, mean ± SDb	−0.32 ± 1.0	−0.22 ± 0.9	−0.38 ± 1.0	−0.45 ± 1.1	−0.60 ± 1.1
Census region, %
Northeast	53.8	34.8	17.1	23.6	23.6
Midwest	17.8	33.4	27.4	22.6	22.4
South	14.7	17.3	33.7	25.9	26.1
West	13.7	14.6	21.8	27.9	27.9
Race and ethnicity, %
Black	2.0	1.8	8.5	10.3	8.1
Hispanic/Latina	0.9	0.9	3.0	4.5	4.2
White (not of Hispanic origin)	95.2	93.4	85.3	82.9	85.2
Other	2.0	3.8	3.3	2.3	2.5
Educational attainment, %
High school or less	0.0	0.0	15.1	24.0	21.2
Some college/associate’s	77.5c	0.0	33.8	39.8	36.5
Bachelor’s or higher	22.5	100d	51.1	36.2	42.3
Mother and sisters with breast cancer, %
0	85.2	89.1	0.0	87.7	86.4
1	13.5	10.5	73.7	11.2	12.3
≥2	1.2	0.5	26.3	1.1	1.3
Smoking status, %
Never	43.0	64.7	55.8	50.3	49.9
Former	39.4	22.3	36.0	40.7	42.5
Current	17.4	12.9	8.2	7.8	6.2
Missing	0.2	0.1	<0.1	1.1	1.4
Alcohol intake, %
Non-drinker	33.5	37.8	18.4	28.4	28.8
<1 drink/week	18.1	17.2	35.4	34.5	31.7
1 to <7	21.5	39.5	32.4	25.8	26.1
≥7	12.4	5.2	13.7	10.5	12.7
Missing	14.5	0.3	0.1	0.8	0.7
Total physical activity, %
<5 MET-hours/week	25.7	24.9	36.7	38.7	32.2
5 to <10	14.2	17.5	16.7	16.2	16.8
10 to <20	18.4	22.2	21.9	19.8	25.3
≥20	27.3	35.3	24.7	15.5	24.6
Missing	14.4	0.1	<0.1	9.8	1.2
Body mass index, %
<25 kg/m2	49.6	65.3	40.6	27.1	40.4
25 to 29	31.5	20.5	31.5	35.6	33.7
≥30	18.7	13.9	27.6	36.8	24.8
Missing	0.2	0.3	0.3	0.5	1.1
Age at menarche, %
<12 years	22.4	24.5	20.3	21.7	21.9
12 to 13	56.8	57.3	56.3	54.7	55.0
≥14	20.0	17.9	23.4	23.2	22.6
Missing	0.8	0.3	<0.1	0.4	0.4
Parity, %
Nulliparous	5.9	25.6	18.2	10.7	12.5
1–2 births	34.9	53.6	51.3	31.5	35.1
≥3 births	57.1	20.9	30.4	57.3	51.7
Missing	2.1	0.0	<0.1	0.5	0.7
Age at first birth, %
Nulliparous	5.9	25.6	18.2	10.7	12.5
<20 years	0.8	5.0	13.1	14.6	11.2
20–24	46.9	23.7	31.2	38.7	36.7
≥25	44.3	45.2	37.4	26.7	29.5
Missing	2.1	0.5	0.1	9.4	10.1
Breastfeeding, %
Nulliparous	5.9	25.6	18.2	10.7	12.5
Parous, breastfed <1 month	29.1	12.9	29.3	37.1	36.2
1–6 months	27.6	11.9	18.1	25.5	25.2
≥7 months	22.5	39.6	34.3	25.2	24.3
Missing	15.0	10.0	0.1	1.5	1.8
Oral contraceptive use, %
Never	51.2	15.5	15.6	56.7	59.6
Ever	48.8	84.4	84.4	43.3	40.4
Missing	0.0	0.2	<0.1	<0.1	0.0
Age at menopause, %
Premenopausal	19.6	96.9	33.5	0.0	0.0
<40 years	3.9	1.6	5.2	9.2	8.9
40 to 49	26.5	1.4	21.1	36.1	37.3
≥50	46.2	<0.1	40.2	46.5	49.7
Missing	3.8	0	<0.1	8.3	4.1
Postmenopausal hormone (PMH) use, %e
Never	52.1	97.2	54.4	48.4	39.3
Former	15.8	0.2	34.7	17.6	13.9
Current	28.7	2.6	10.5	34.0	46.7
Missing	3.4	<0.1	0.4	0.1	0.1

Abbreviations: IQR, interquartile range; MET, metabolic equivalent; NHS, Nurses’ Health Study; NHSII, Nurses’ Health Study II; NO₂, nitrogen dioxide; nSES, neighborhood socioeconomic status; PM_2.5, fine particulate matter; PMH, postmenopausal hormone; ppb, parts per billion; SD, standard deviation; SIS, Sister Study; WHI-CT, Women’s Health Initiative Clinical Trials; WHI-OS, Women’s Health Initiative Observational Study

^aAnalytic population excludes those with missing air pollution, census region, nSES score, race and ethnicity, or educational attainment

^bHigher scores represent greater neighborhood deprivation

^cRegistered nurses are required to have, at minimum, an associate’s degree. Therefore, NHS participants who did not report their educational attainment were categorized as having received at least some college.

^dEducational attainment was not collected by NHSII; participants were assumed to have at least a bachelor’s degree

^ePostmenopausal hormone use excluded hormone use related to the treatment arm in WHI-CT

A small positive association with overall breast cancer was observed for a 10-ppb increase in NO₂ exposure (HR_summary=1.03, 95% CI: 1.00–1.05; Table 2). Although the HR was most consistently elevated for ER/PR positive tumors, there was no significant heterogeneity by ER/PR status or by invasiveness (Table S1). Further, little variability was observed by menopausal status, race and ethnicity, geographic region, nSES, or BMI (Figure 1; Table S2) although elevated associations were observed for those living in the Midwest (HR_summary=1.07, 95% CI: 1.00–1.14) and in the lowest category of nSES (HR_summary=1.06, 95% CI: 1.01–1.11; Table S2).

Table 2.

Individual cohort and meta-analysis hazard ratios and 95% confidence intervals for the association between spatio-temporally modeled NO₂, PM_2.5, and ozone and incident breast cancer, overall and by hormone receptor status with follow-up through 2016–2017

	Overall				ER+ or PR+		ER-/PR-
	Person-years	Cases	Crude HR (95% CI)a	Adj HR (95% CI)b	Cases	Adj HR (95% CI)b	Cases	Adj HR (95% CI)b	pc
10-ppb increase in NO2
NHS	2,142,217	9,267	1.04 (1.01, 1.08)	1.03 (0.99, 1.07)	5,898	1.00 (0.96, 1.05)	1,053	0.92 (0.83, 1.02)	0.14
NHSII	2,654,911	5,772	1.05 (1.01, 1.10)	1.05 (0.99, 1.10)	3,610	1.04 (0.97, 1.10)	737	0.95 (0.83, 1.09)	0.24
SIS	479,594	3,044	1.04 (0.95, 1.13)	1.04 (0.95, 1.14)	2,528	1.09 (0.99, 1.20)	387	0.87 (0.67, 1.12)	0.09
WHI-CT	937,031	4,879	0.99 (0.94, 1.04)	0.98 (0.93, 1.04)	3,639	0.97 (0.91, 1.03)	541	1.12 (0.97, 1.29)	0.06
WHI-OS	1,100,157	5,907	1.02 (0.98, 1.07)	1.03 (0.98, 1.08)	4,410	1.03 (0.98, 1.09)	700	0.93 (0.82, 1.07)	0.15
Summary				1.03 (1.00, 1.05)		1.02 (0.99, 1.05)		0.96 (0.89, 1.04)	0.15
Q test				0.54		0.29		0.19
5-µg/m3 increase in PM2.5
NHS	1,277,322	5,242	0.98 (0.93, 1.04)	1.00 (0.94, 1.06)	3,523	0.97 (0.90, 1.05)	557	1.06 (0.89, 1.26)	0.39
NHSII	1,669,694	4,326	0.96 (0.89, 1.02)	0.97 (0.90, 1.04)	2,869	0.89 (0.82, 0.98)	522	1.25 (1.03, 1.52)	0.002
SIS	479,594	3,044	0.97 (0.88, 1.06)	0.96 (0.86, 1.06)	2,528	0.95 (0.84, 1.06)	387	1.06 (0.81, 1.38)	0.43
WHI-CT	648,537	3,518	0.98 (0.91, 1.04)	0.99 (0.92, 1.06)	2,780	0.98 (0.91, 1.07)	391	1.16 (0.96, 1.41)	0.12
WHI-OS	762,475	4,077	1.00 (0.94, 1.07)	1.02 (0.95, 1.09)	3,183	0.99 (0.91, 1.07)	513	1.16 (0.97, 1.38)	0.10
Summary				0.99 (0.96, 1.02)		0.96 (0.93, 1.00)		1.14 (1.04, 1.24)	0.0004
Q test				0.85		0.46		0.75
10-ppb increase in ozone
NHS	2,142,217	9,267	0.97 (0.91, 1.04)	0.98 (0.92, 1.05)	5,898	1.05 (0.97, 1.14)	1,053	1.10 (0.92, 1.32)	0.64
NHSII	2,654,911	5,772	0.96 (0.89, 1.04)	1.00 (0.92, 1.08)	3,610	0.99 (0.89, 1.10)	737	1.27 (1.02, 1.58)	0.04
SIS	479,594	3,044	0.93 (0.84, 1.03)	0.93 (0.84, 1.03)	2,528	0.91 (0.81, 1.02)	387	1.05 (0.79, 1.39)	0.35
WHI-CT	937,031	4,879	0.99 (0.91, 1.06)	0.99 (0.91, 1.07)	3,639	0.99 (0.90, 1.09)	541	1.02 (0.82, 1.29)	0.79
WHI-OS	1,100,157	5,907	0.97 (0.90, 1.04)	0.97 (0.90, 1.04)	4,410	0.94 (0.86, 1.02)	700	1.04 (0.85, 1.28)	0.34
Summary				0.98 (0.94, 1.01)		0.98 (0.93, 1.03)		1.10 (1.00, 1.21)	0.04
Q test				0.82		0.22		0.67

Abbreviations: CI, confidence interval; ER, estrogen receptor; HR, hazard ratio; HT, hormone therapy; NHS, Nurses’ Health Study; NHSII, Nurses’ Health Study II; NO₂, nitrogen dioxide; nSES, neighborhood socioeconomic status; PM_2.5, fine particulate matter; ppb, parts per billion; PR, progesterone receptor; SIS, Sister Study; WHI-CT, Women’s Health Initiative Clinical Trials; WHI-OS, Women’s Health Initiative Observational Study

^aCrude model adjusted for age (as a stratifying variable) and calendar month (or calendar year for WHI-CT and WHI-OS)

^bAdjusted model additionally adjusted for census region (time-varying; Northeast, Midwest, South, West), nSES (continuous), race and ethnicity (Black, Hispanic, non-Hispanic White, Other), educational attainment (high school or less, some college/associate’s, bachelor’s or higher), HT arm (WHI-CT only: estrogen, estrogen control, estrogen plus progestin, estrogen plus progestic control)

^cFor individual cohort estimates, the p-value is for the likelihood ratio test comparing models with and without a pollutant*covariate cross-product term; for the summary estimate, the p-value is the contrast test method p-heterogeneity

Figure 1.

Summary hazard ratios and 95% confidence intervals for the associations between NO₂ (panel A), PM_2.5 (panel B), and ozone (panel C) with incident breast cancer, stratified by menopausal status (time-varying for all except WHI-CT and WHI-OS), race and ethnicity, census region (time-varying), and neighborhood SES (higher score represent increasing neighborhood deprivation). P-heterogeneity was determined using the contrast test method.

Abbreviations: NO₂, nitrogen dioxide; p-het, p-heterogeneity; PM_2.5, fine particulate matter; SES, socioeconomic status

Overall, a 5-µg/m³ increase in PM_2.5 exposure was not associated with breast cancer overall (HR_summary=0.99, 95% CI: 0.96–1.02; Table 2) or with ER+ and/or PR+ breast cancer. However, we observed a positive association with ER-/PR- tumors (HR_summary=1.14, 95% CI: 1.04–1.24; p-heterogeneity=0.0004). When stratifying by region, we observed an overall breast cancer association in the Midwest (HR_summary=1.15, 95% CI: 1.01–1.32), but not other regions (p-heterogeneity=0.01; Figure 1, Table S2). No differences were observed by invasiveness (Table S1) or other factors (Figure 1, Table S2).

No association was observed for ozone (10-ppb increase) and breast cancer overall (Table 2). However, an elevated association was observed for ER-/PR- tumors (HR_summary=1.10, 95% CI: 1.00–1.21; p-heterogeneity=0.04). No significant differences were observed by tumor invasiveness (Table S1) or other factors (Figure 1); although an inverse association was observed among women in the Midwest (HR_summary=0.88, 95% CI: 0.79–0.98; p-heterogeneity=0.13; Table S2).

Results for all three pollutants remained similar when using a 10-year lag, adjusting for the other pollutants or for the model adjusting for breast cancer risk factors (Table S3) and when cumulative average exposures were used instead of the rolling average (Table S4). Including the Sister Study self-reported breast cancer cases did not change findings.

Discussion

In this parallel analysis of 5 large US cohorts, with over 28,000 breast cancer cases combined, we observed a 3% increase in overall breast cancer incidence for a 10-ppb increase in NO₂ outdoor concentrations. We also found that a 5-µg/m³ increase in PM_2.5 outdoor concentrations was associated with a higher incidence of hormone receptor negative breast cancer as well as overall breast cancer among participants living in the Midwest.

Our finding of a modest increase in the incidence of breast cancer in relation to NO₂ is consistent with the literature. Three recent meta-analyses all concluded that a 10-µg/m³ increase in NO₂ was associated with a 2% increase in breast cancer risk (95% CI: 1.01–1.04^9,13; 95% CI: 1.00–1.03).¹² Gabet et al. (2021) estimated that a decrease in average NO₂ from 17.4 µg/m³ to 6.3 µg/m³ (fifth percentile) based on a HR=1.02 would result in a 3% decrease in the number of breast cancer cases, or 1,677 fewer cases in France annually.⁹ Based on an estimated 316,950 cases of female breast cancer expected to be diagnosed in the US in 2025,¹ a 3% reduction would result in 9,500 fewer cases. NO₂, which results from combustion processes, is a component of traffic-related air pollution. While not considered a mutagen, NO₂ may serve as a cancer promoter or as a proxy for other traffic-related chemicals such as polycyclic aromatic hydrocarbon (PAH) exposure which are not routinely monitored.^12,45 PAHs can induce inflammation and oxidative stress and some have demonstrated estrogen agonist or antagonist (i.e., endocrine disrupting) properties.^7,46

With more than 2,300 ER-/PR- cases, we observed an approximately 14% higher incidence of ER-/PR- tumors per 5-µg/m³ increase in PM_2.5. Of the few studies that have evaluated PM_2.5 and breast cancer subtype heterogeneity, results have been largely underpowered,^19–23 including previous analyses in both NHSII²¹ and SIS.²³ In the Multiethnic Cohort, a 26% increase in ER-/PR- breast cancer (case N=543; 95% CI: 0.83–1.91) was observed for PM_2.5, although the estimate was also elevated for ER+ or PR+ cancer (case N=2,413; HR=1.17, 95% CI: 0.96–1.42) .¹⁷ Other studies found no associations for either hormone receptor positive or negative tumors,^21–23 and one observed an association only for ER+ tumors.¹⁶

We also observed a 15% increase in overall breast cancer risk for PM_2.5 exposure among Midwestern women, which is similar to results from the Black Women’s Health Study (18% increase per IQR [2.87 µg/m³]).¹⁹ In an earlier analysis in the Sister Study, an association between PM_2.5 and invasive breast cancer was observed only for participants living in the West.²³ Although not shown, with the additional cases and new exposure model used in this analysis, results from SIS were consistent with those from the combined cohorts.

PM_2.5 is a heterogeneous mixture with demonstrated genotoxic potential,⁶ that has been associated with early markers of breast cancer susceptibility.^47,48 Components include metals (e.g., arsenic, cadmium); PAHs; secondary inorganic aerosols such as nitrate (NO₃^-), ammonium (NH₄⁺), and sulfate (SO₄^2-); and other compounds and elements.²⁴ There is some limited evidence suggesting an association between airborne cadmium^49,50 and arsenic⁵⁰ and ER-/PR- breast cancers. Further well-powered studies to consider both PM_2.5 components, including from specific sources such as wildfires,⁵¹ and hormone receptor-negative tumors are needed.

Regional variation in the PM_2.5 component mixture could explain in part the geographic variability observed here. There is evidence that both nitrate and ammonium concentrations were higher in the Midwest compared to the other census regions in 2000.⁵² Both nitrate and ammonium have been associated with higher levels of inflammatory markers^53,54 and with a higher proportion of breast epithelial-to-stroma tissue,⁴⁷ both of which are risk factors for breast cancer.^55–57

While our observed lack of association for ozone and breast cancer overall is consistent with previously published studies,^10,18,19 our finding of an association between ozone and ER-/PR- breast cancer is novel. Ozone exposure may induce oxidative stress, although the carcinogenicity of ozone remains uncertain.⁵⁸

Our sensitivity analyses, evaluating a 10-year lag and a cumulative average exposure, produced very similar results to those for our overall models, with a small association for NO₂ with overall breast cancer. Given that carcinogenesis is a multi-stage process, including tumor initiation and tumor promotion,⁵⁹ and that associations were not stronger when we explored a 10-year lag, our findings suggest that NO₂ likely acts as a promoter but not an initiator. These findings could also be reflective of a high correlation in exposure across time which may make it difficult to disentangle the impact of timing of exposure.

This study is the largest analysis of air pollution and breast cancer in the US, with >28,000 cases across 5 cohorts, including the largest number of ER-/PR- cases. However, US air pollution levels are lower than other populous countries⁶⁰; indeed, the average concentrations observed here are lower than current regulatory guidelines (53 ppb annually).⁶¹ In Asia, where pollution levels are higher, studies of air pollution and breast cancer are limited.^14,62 More research in these areas is needed to determine whether observed associations with breast cancer, assessed at largely low concentrations of air pollutants, extend to higher exposures.⁵¹

An important strength of this study is the use of the same high-resolution exposure model across all cohorts³⁰ for estimating pollutant concentrations. We used the same statistical models with the same covariates in each of the five cohorts to improve comparability across studies. Many prior studies, including some of the largest ones,^10,15,16,23 have relied solely on a participants’ baseline addresses to evaluate air pollution exposure which likely results in non-differential misclassification of exposure and potentially biased effect estimates whereas here we have considered exposure at all addresses throughout follow-up.^63,64 We considered different ways to model time-varying exposure (rolling vs. cumulative average), and the results were similar for both, lending support to the robustness of our findings. We also included multiple US-wide cohort studies that provided both a wide range of exposures due broad geographic coverage and permitted the evaluation of regional heterogeneity of exposure-effect relationships, which addresses the potential impact of differences in PM_2.5 composition across US census regions.^16,19,23

The PM_2.5 data were not available at the time of recruitment for all of the cohorts because extensive monitoring by the US Environmental Protection Agency was not established until 1999.³⁰ Thus, the follow-up period for the PM_2.5 analysis, which began in 2001 at the earliest, post-dated recruitment by 3–8 years for WHI-CT and WHI-OS, by 12 years for NHSII, and by 25 years for NHS. It is therefore possible that more rapidly growing tumors were diagnosed prior to the start of analytic follow-up and were not included here. This could result in a bias towards the null for our findings due to a depletion of susceptibles.⁶⁵ However, our positive findings for PM_2.5 were limited to an association for ER-/PR- tumors, which tend to be more aggressive than ER+ or PR+ tumors.⁶⁶ Additionally exposure data were available for the entire Sister Study follow-up period due to the later study enrollment dates and we observed similar results across studies.

We were unable to account for exposures at locations other than the residence, nor did we have information on the time spent outdoors. We anticipate that any misclassification of exposure would be non-differential with respect to the outcome. We were also not able to capture exposures during several critical windows of susceptibility (e.g., puberty, pregnancy) related to breast development.⁶⁷ For example, cigarette smoke, which contains similar constituents to air pollution such as PAHs and heavy metals,^24,68,69 has been associated with breast cancer risk more robustly in women who started smoking as teenagers⁷⁰ or before their first pregnancy.⁷¹ This suggests that exposure to PM_2.5 before first birth may be of particular importance, although this has been challenging to address in current cohorts due to the timing of availability of air pollution exposure models, which largely draw on routine EPA monitoring data. Younger cohorts will be better poised to address the impacts of early life exposure.

Public health implications

This is the largest analysis of outdoor air pollution concentrations and breast cancer incidence in the US and the largest study globally to consider tumor subtypes. Consistent with prior studies, NO₂ was associated with a small increase in breast cancer incidence overall. Higher concentrations of PM_2.5 and ozone were both associated with ER-PR- tumors, which have few established risk factors and higher case-fatality compared to hormone receptor positive tumors. Although our risk estimates were modest, air pollution may contribute to large numbers of breast cancer cases given high incidence rates and ubiquitous exposures. Further, the potential impact of exposure could be greater in countries or regions with higher concentrations of air pollutants.

Human Participant Protection:

Women’s Health Initiative Clinical Trials and Observational Study:

All women provided written informed consent at baseline, and the protocol and consent forms were approved by the institutional review board for each participating institution.

Nurses’ Health Study and Nurses’ Health Study II:

The Institutional Review Board of Brigham and Women’s Hospital approved both studies, and all participants provided informed consent through the return of the initial questionnaire.

Sister Study:

The institutional review board (IRB) of the National Institute of Environmental Health Sciences and the Copernicus Group IRB approved the study. All participants provided written consent.