Carcinogenic Air Pollutants and Breast Cancer Risk in the Arkansas Rural Community Health Study: A Nested Case-Control Study

Abstract

Background:

Previous epidemiological studies on the associations between hazardous air pollutants (HAPs) and breast cancer (BC) have largely neglected rural, medically underserved areas in the United States, which differ in exposures and disparities compared to urban areas.

Objectives:

We aimed to examine the associations between carcinogenic HAPs and BC risk in a rural population.

Methods:

Using a nested case-control design, 574 cases and 2,295 incidence density-sampled controls matched 1:4 on index age and race were included from the Arkansas Rural Community Health study. Twelve census-tract level HAPs from the 2005 National Air Toxics Assessment were geocoded to participants’ baseline addresses and categorized into tertiles based on the control distributions. Multivariable conditional logistic regression analysis was used to estimate associations for BC and subgroups. Additionally, weighted quantile sum (WQS) regression was used to assess the effects of pollutant mixtures.

Results:

Adjusted models resulted in significantly increased risk of BC for moderate PAH (OR_T2 = 1.32; 95% CI 1.03 – 1.69) and high chromium (OR_T3 = 1.31; 95% CI 1.00 – 1.70, P trend=0.04) exposure. The WQS index term, which is the regression coefficient of the combined pollutant mixture, was non-significantly positively associated with BC (OR = 1.21; 95% CI 0.96 – 1.53). Chromium (0.45), propylene dichloride (0.15), and polychlorinated biphenyls (0.09) were weighted the most, suggesting these pollutants had the greatest impact on the increased risk of BC. Joint effects models with first-degree family history of BC resulted in increased risk of BC for women with family history compared to women without family history and low PAH exposure.

Conclusion:

This study highlights significant associations between specific HAPs, particularly PAHs and chromium, and increased BC risk in a rural population, emphasizing the need for targeted public health interventions and regulatory efforts to mitigate exposure in medically underserved areas.

Graphical Abstract

Introduction

Breast cancer is the most diagnosed cancer in women in the United States (US) with an incidence rate of 128.1 per 100,000 women from 2015–2019.¹ Breast cancer is a heterogeneous disease with varying subtypes and immunological profiles, each having specific tumor aggressiveness, treatment resistances, and survival probabilities.² Accepted risk factors for breast cancer include hormonally mediated risk factors such as age at menarche or hormone replacement therapy, lifestyle-related factors such as alcohol consumption or being overweight or obese, other breast conditions such as hyperplasia or breast density, and genetic variations.^3,4 Recently, epidemiological studies and reviews have focused on occupational and environmental exposures and their association with breast cancer, with several cohorts being utilized for geographic and biological markers of exposure.^5,6 However, many of the existing cohort studies for breast cancer were conducted among populations well below average risk and not enriched with high-risk participants,⁶ and did not focus on underserved populations in rural communities in the US.

Outdoor air pollution from natural and anthropogenic sources varies within regions of the world, proximity to industrial processes, and individual level factors. In rural areas, common environmental exposures include trash burning,⁷ field burning,⁸ heavy metals,^9,10 pesticides and herbicides,¹¹ occupational exposures,¹² and cooking practices.¹³ Specifically in Arkansas and the Southeastern US,^14,15 industrial and individual processes in rural areas such as oil refining, logging, field burning, and diesel exposure produce hazardous air pollutants (HAPs)¹⁵ that can lead to inflammation, endocrine disruption,¹⁶ and increased risk of disease and mortality.¹⁷ Rural populations may have a unique constituency of HAPs compared to their urban counterparts due to the industrial and individual practices in these areas. Combined with lower access to care,¹⁸ greater health disparities,¹⁹ and a lack of research into this population in the US,²⁰ there is an urgent need for studies regarding outdoor air pollution and cancer outcomes in rural states and communities.

Arkansas is a rural state with low population mobility (3.3% change in population from 2010–2020) and is ranked 45th in life expectancy in the US in 2020.²¹ The state’s heavy reliance on agricultural, manufacturing, and forest industries, combined with low population mobility, provides an opportunity to study the effects of persistent environmental exposures that may be associated with breast cancer in rural, medically underserved communities. This study’s aim was to estimate the association between individual and mixed HAP census-tract level exposure estimates from the 2005 National Air Toxics Assessment (NATA) and risk of incident breast cancer in the Arkansas Rural Community Health (ARCH) study. The hypothesis was that cases would be exposed to greater concentrations of HAPs than controls, leading to increased risk of breast cancer.

Methods

Ethics and Legal Approval

The study was approved by the Institutional Review Boards of University of Arkansas for Medical Sciences (IRB# 89071).

Study Population

The Arkansas Rural Community Health (ARCH) study is a population-based cohort study of breast cancer (survivors) and disease-free participants from Arkansas that was open for enrollment from 2007 to 2013. Initially created as the Spit for the Cure Cohort (SFCC), the study was designed to research genetic biomarkers for women with prevalent breast cancer and women who were disease-free at baseline. Eligible participants included women aged 18–100, gave informed consent, completed a questionnaire, and provided a saliva sample for DNA isolation. Participants were enrolled through community-based convenience sampling at community events including the Susan G. Komen Race for the Cure^© and the American Cancer Society Relays for Life, among other events. Complete SFCC methodology can be found elsewhere.²² Women who provided a baseline residential address in Arkansas and did not elect to withdraw from the study were eligible for analysis in this study. Women who indicated they had a previous breast cancer diagnosis or were linked to a cancer diagnosis prior to enrollment were not eligible for analysis.

Cases

The ARCH cohort is linked annually with the Arkansas Central Cancer Registry (ACCR) to identify incident cases of in situ and invasive breast tumors. The ACCR is a population-based registry located within the Arkansas Department of Health that has been consistently certified as Registry of Distinction by the National Program of Cancer Registry and a gold-standard registry designated by the North American Association of Central Cancer Registries. Participants were asked at baseline if they had previously been diagnosed with breast cancer, and incident cases are individuals who answered no or left the answer blank and were not identified through registry linkage as having a reportable breast cancer diagnosis prior to enrollment. Cases for the present study were defined as individuals with incident in situ or invasive breast cancer (ICD-O-3 primary site codes C500-C506, C508-C509) identified through registry linkage from participants’ date of enrollment to December 31^st, 2022. Breast cancer cases were classified into overall breast cancer and subgroups: early-onset (≤50 years) or average age-onset (>50 years), non-invasive (in situ) or invasive (local, regional, or distal), hormone receptor (HR) positive (estrogen receptor (ER)+ or progesterone receptor (PR)+) or negative (ER- and PR-), and postmenopausal cases (only women who were postmenopausal at baseline, as menopausal status at diagnosis was not obtained through ACCR).

Controls

Up to four controls from the baseline cohort were matched to each case through incidence density sampling by index age (+/− 1 year) and race (White/European American, “White” or Black/African American, “Black”) and were selected using a random seed at the case diagnosis date within a risk set. Matching was accomplished using code developed by Kiri (2012) that allowed for the minimization of control selection bias due to the randomized selection process.²³ Cases were eligible to serve as controls until their event date (breast cancer diagnosis), and controls were eligible for random selection in multiple risk sets until censorship (death, loss to follow-up).

Exposures

The Environmental Protection Agency (EPA) created the National Air Toxics Assessment (NATA) as a screening tool to determine if areas, pollutants, or types of pollution sources warrant further investigation for public health research. Briefly, emissions data for individual HAPs are collected from data reported from individual facilities (point sources) and estimated through emission inventory models for area and mobile sources. NATA provides ambient air, population exposure, and cancer risks for individual HAPs in each census tract.

The exposures of interest for the current study were population exposure estimates collected from the 2005 NATA, as this year was before study enrollment and resembles outdoor air pollution in a period before the onset of latent cancers. Exposure concentrations were used as these values are modeled to represent the concentration of HAPs a person may breathe over time, rather than ambient air concentrations. Exposure modeling for the 2005 NATA was derived by multiplying each source ambient concentrations by the previous Hazardous Air Pollutant Exposure Model, version 5 (HAPEM5) for the 1999 NATA, yielding tract-level exposure concentrations that are more representative of movements of individuals and microenvironments. Exposure concentrations (μg/m³) were geocoded to census tracts and linked with participants’ self-reported addresses at enrollment. Twelve HAPs were chosen based on their biological effects defined by the National Toxicology Program (NTP) as having adverse immunological, hematological, developmental, endocrinal, or reproductive effects, prior evidence of association with breast cancer in observational studies and known carcinogenic classification (IARC Group 1 or NTP Known Human Carcinogen). Four classes of chemicals were chosen to include pollutants which are present in outdoor air and are often characterized in combustible sources: 1) Polycyclic aromatic hydrocarbons/polycyclic organic matter (PAHs); 2) polychlorinated biphenyls (PCBs); 3) volatile organic compounds (1,2-dichloropropane (propylene dichloride), 1,3-butadiene, benzene, formaldehyde, trichloroethylene, and vinyl chloride); and 4) heavy metals (arsenic compounds, chromium compounds, cadmium compounds, and nickel compounds).

Covariates

The ARCH baseline survey included questions on breast cancer risk factors such as family history, body mass index (BMI), alcohol consumption, reproductive and breastfeeding history, education, and physical activity. The 2015 Arkansas area deprivation index (ADI) based on Census-blocks was also included to represent neighborhood disparities since it was constructed using data from the 2011–2015 American Community Survey 5-year means, including population indicators related to educational attainment, housing, employment, and poverty.²⁴ A minimal set of confounders was identified through a directed acyclic graph (DAG) (Supplemental Figure S1).²⁵ These confounders included BMI (continuous, kg/m²), first-degree family history of breast cancer (with or without having a mother, daughter, or sister with breast cancer), menopausal status at baseline (pre-/peri- or post-menopausal), Arkansas state ADI (continuous, decile), education (college degree, some college/technical degree, high school graduate/less than high school), physical activity (metabolic equivalent of tasks tertiles based on vigorous, moderate, light activity, and time sitting in the control group), and alcohol consumption (non-, light, or moderate/heavy drinker). Arkansas state ADI and education may influence where an individual lives, which is associated with their HAP exposure, and may be associated with an individual’s breast cancer risk. BMI, first-degree family history of breast cancer, menopausal status, physical activity, and alcohol consumption may be associated with HAP exposure metabolism and are known risk factors for breast cancer risk.

Menopausal status at baseline was categorized into pre-/peri- or post-menopausal based on the classification from Phipps et al. (2010).²⁶ Women who were 55 or older or indicated their menstrual periods stopped permanently (including natural menopause, surgical procedure, or other condition) were classified as post-menopausal. Pre-/peri-menopausal women were those under 55 years old and answered either no or not sure if their menstrual periods had stopped permanently. Physical activity was quantified using the metabolic equivalent of tasks (METs), calculated by multiplying the time spent in minutes for each activity, the number of days per week spent in each activity, and its corresponding MET weight.²⁷ Alcohol consumption was categorized as heavy, moderate, light, or never drinker based on participants’ self-reported frequency and quantity of alcohol intake from the baseline questionnaire.²⁸

Statistical Analysis

Descriptive analyses were performed between case-control characteristics and HAP exposure distributions. The Wald chi-square test for homogeneity was used for categorical variables (race, education, family history, menopausal status, physical activity, and alcohol use) and Student’s t-test was used for continuous variables (age, BMI, Arkansas state ADI) after confirming normality through visual inspection of histograms. For non-normal variable distributions such as the pollutant exposures, Wilcoxon Rank Sum Test (Wilcoxon two-sample test) was used. In addition to the unadjusted model, multivariable conditional logistic regression was used to estimate the odds ratios (ORs) and 95% confidence intervals (95% CIs) for the association between each individual pollutant and breast cancer risk. Because the outcome is a rare disease in the study population, the rare disease assumption is met. Pollutant exposures were categorized into tertiles (33.33 and 66.67 percentiles) based on the control group distribution, and the lowest exposure category was the reference group for analysis. Trend tests were performed by fitting the logistic regression models with the HAP tertiles as ordinal variables and using the p-value of a Wald chi-square test. Subgroup analyses were performed in the early-onset or average age-onset groups, non-invasive or invasive tumor groups, hormone-receptor positive or negative groups, and postmenopausal cases. Exposure tertiles based on the control distributions were recategorized for each subgroup. Interaction on the multiplicative scale was assessed using a cross-product term in the regression function and comparing the heterogeneity of effects, and interaction on the additive scale was examined using the joint effects approach. A first-degree family history of breast cancer (at least one affected relative or not), state ADI (lesser (≤4) or more (>4) deprived), and body mass index (obese (≥30.0 kg/m²) or non-obese (<30.0 kg/m²)) were assessed for interaction. To assess pollutant mixtures, unconditional weighted quantile sum (WQS) regression was used fitted to a binomial regression model with the HAP mixture categorized into tertiles. Briefly, WQS regression is a constrained regression approach designed to estimate the effect of correlated chemicals and identify which chemicals contributed the most to the outcome of interest.²⁹ The dataset was divided into exposure tertiles with 40% training and 60% validation, and 100 bootstrap samples for parameter estimation were used. A 2-tailed P-value less than 0.05 was used to indicate statistical significance. Descriptive and analytical analyses were performed using SAS v9.4 (SAS Institute Inc., Cary, NC) and R v4.3 (R Development Core Team, Vienna, Austria). WQS regression was accomplished using the gWQS package in R developed by Renzetti et al. (2023).³⁰

Results

Study Participant Characteristics

A total of 19,871 participants were eligible for analysis in the study, after excluding those without an address (n = 30), those who were missing a census tract code (do not currently live in Arkansas, partial or total incomplete address) (n = 3,354), those who self-reported or were registry-confirmed to have a previous breast cancer diagnosis (n = 1,998), those who were matched within the registry but did not have information available (n = 2), and finally those with any missing matching criteria or covariates (n = 1,120). Within the 19,871 total participants left, 574 incident cases were identified, and 2,295 eligible controls were matched 1:4 on baseline age and race through the incidence density sampling procedure. One case had three eligible controls based on the matching criteria of baseline age and race and was retained for analysis.

Case and control characteristics are presented in Table 1 with difference testing of variables between the two groups. The average age of the cases was 52.6 +/− 11.6 years at baseline and 20 percent of the cases were Black. The only significantly different characteristic between the two groups was family history of breast cancer in a first-degree relative, with a higher proportion of cases having an affected first-degree family member than controls (29% vs. 20%; P < 0.001). The full HAP exposure distributions among cases and controls are presented in Table 2 in μg/m³. Cases had significantly higher exposure means for benzene, arsenic, and chromium than controls. Exposure concentrations for benzene, 1,3-butadiene, formaldehyde, and trichloroethylene were orders of magnitude higher than other pollutants.

Table 1.

Study participant characteristics by case-control status.

	Cases	Controls	Difference testinga
Characteristic	(n = 574)	(n = 2295)
	Mean (SD)
Age at baseline, y	52.6 (11.6)	52.6 (11.5)	Matched
Body mass index, kg/m2	29.5 (6.8)	29.3 (7.2)	0.64
State area deprivation index, decile	4.37 (2.92)	4.43 (2.90)	0.67
	Number (%)
Race			Matched
White	461 (80)	1844 (80)
Black	113 (20)	451 (20)
Education			0.14
College degree	268 (47)	982 (43)
Some college/technical degree	182 (32)	737 (32)
High school graduate or lower	124 (22)	576 (25)
Family history of breast cancer			< 0.001
Yes (mother, sister, or daughter)	164 (29)	459 (20)
No/Unsure/Unknown	410 (71)	1836 (80)
Menopausal status at baseline			0.41
Post-menopausal	379 (66)	1557 (68)
Pre-/Peri-menopausal	195 (34)	738 (32)
Physical Activity			0.42
METs Tertile 1	186 (32)	757 (33)
METs Tertile 2	210 (37)	776 (34)
METs Tertile 3	178 (31)	762 (33)
Alcohol use			0.60
Heavy/Moderate consumption	158 (28)	586 (25)
Light consumption	224 (39)	907 (40)
Non-drinker	192 (33)	802 (35)

^aWald chi-square test for homogeneity (Race, Education, Family history of breast cancer, Menopausal status at baseline, Physical activity, and Alcohol use) and Student’s t-test (Age at baseline, Body mass index, and State area deprivation index) p-values between cases and controls.

Table 2.

Distribution of hazardous air pollutant exposure concentrations from the 2005 National Air Toxicity Assessment by case/control status in µg/m³.

	Case-Control	Mean (SD)	Min	Q1	Median	Q3	Max	P a
Polycyclic aromatic hydrocarbons	Cases	3.96E-03 (2.74E-03)	2.10E-04	1.82E-03	3.44E-03	5.55E-03	2.22E-02	0.41
	Controls	3.93E-03 (2.98E-03)	2.10E-04	1.71E-03	3.29E-03	5.68E-03	2.22E-02
Polychlorinated biphenyls	Cases	3.10E-05 (2.49E-05)	1.90E-06	1.34E-05	2.13E-05	4.06E-05	1.54E-04	0.77
	Controls	3.18E-05 (2.83E-05)	1.89E-06	1.31E-05	2.13E-05	4.27E-05	2.03E-04
Benzene	Cases	5.57E-01 (3.37E-01)	1.23E-01	3.32E-01	4.61E-01	7.44E-01	2.58E+00	0.04
	Controls	5.31E-01 (3.42E-01)	1.23E-01	3.15E-01	4.30E-01	7.09E-01	2.58E+00
1,3-Butadiene	Cases	2.88E-02 (1.85E-02)	3.49E-03	1.41E-02	2.77E-02	4.12E-02	1.11E-01	0.05
	Controls	2.73E-02 (1.86E-02)	2.37E-03	1.37E-02	2.34E-02	3.98E-02	1.11E-01
Formaldehyde	Cases	1.90E+00 (2.77E-01)	1.30E+00	1.70E+00	1.90E+00	2.10E+00	2.80E+00	0.48
	Controls	1.89E+00 (2.74E-01)	1.30E+00	1.70E+00	1.90E+00	2.00E+00	2.80E+00
Propylene dichloride	Cases	3.65E-04 (4.04E-05)	2.34E-04	3.67E-04	3.71E-04	3.76E-04	6.16E-04	0.68
	Controls	3.61E-04 (4.43E-05)	2.34E-04	3.67E-04	3.71E-04	3.76E-04	6.49E-04
Trichloroethylene	Cases	2.62E-02 (1.80E-02)	9.31E-03	1.39E-02	2.11E-02	3.12E-02	2.14E-01	0.16
	Controls	2.64E-02 (3.52E-02)	9.27E-03	1.28E-02	1.97E-02	3.02E-02	1.04E+00
Vinyl Chloride	Cases	6.50E-05 (3.48E-04)	0.00E+00	7.32E-06	1.37E-05	2.98E-05	5.52E-03	0.83
	Controls	5.51E-05 (2.69E-04)	0.00E+00	6.46E-06	1.39E-05	3.51E-05	6.23E-03
Arsenic	Cases	1.90E-04 (1.32E-04)	6.58E-05	7.74E-05	9.74E-05	3.35E-04	5.95E-04	0.04
	Controls	1.77E-04 (1.27E-04)	6.58E-05	7.63E-05	8.98E-05	3.30E-04	5.95E-04
Cadmium compounds	Cases	3.11E-05 (9.32E-06)	1.95E-05	2.45E-05	3.02E-05	3.38E-05	8.62E-05	0.55
	Controls	3.17E-05 (1.18E-05)	1.93E-05	2.40E-05	3.04E-05	3.46E-05	1.87E-04
Chromium compounds	Cases	3.51E-04 (3.23E-04)	3.45E-05	7.18E-05	2.49E-04	5.51E-04	2.22E-03	0.03
	Controls	3.43E-04 (4.39E-04)	3.45E-05	5.98E-05	1.69E-04	5.37E-04	5.75E-03
Nickel compounds	Cases	1.39E-04 (5.31E-04)	3.30E-05	4.63E-05	8.04E-05	1.04E-04	1.18E-02	0.32
	Controls	1.34E-04 (4.46E-04)	3.29E-05	4.57E-05	6.50E-05	1.03E-04	1.18E-02

^aWilcoxon Rank Sum p-value comparison between case and control distributions.

Association Between Breast Cancer and Individual Pollutants

Unadjusted and adjusted multivariable conditional logistic regression results between individual HAPs and breast cancer risk are presented in Table 3. Unadjusted models for PAH, arsenic, and chromium resulted in an increased risk of breast cancer for those exposed to high arsenic and chromium and moderate PAH. Upon adjustment for confounders, moderate PAH (OR_T2 = 1.32; 95% CI 1.03 – 1.69) and high chromium (OR_T3 = 1.31; 95% CI 1.00 – 1.70, P trend=0.04) exposure retained a significant positive association with overall breast cancer. High arsenic exposure was positively associated with breast cancer risk, but this OR was not statistically significant (OR_T3 = 1.27; 95% CI 0.98 – 1.64, P trend=0.06). Other non-significant positive associations include moderate (OR_T2 = 1.16; 95% CI 0.90 – 1.49) and high (OR_T3 = 1.20; 95% CI 0.91 – 1.57) benzene, moderate trichloroethylene (OR_T2 = 1.17; 95% CI 0.93 – 1.48), and moderate vinyl chloride (OR_T2 = 1.16; 95% CI 0.92 – 1.45) exposure. High cadmium had a non-significant inverse association with overall breast cancer (OR_T3 = 0.85; 95% CI 0.67 – 1.08).

Table 3.

Unadjusted and adjusted^a conditional logistic regression estimated odds ratios (OR) and 95% confidence intervals (CI) for the associations of individual hazardous air pollutants^b and overall breast cancer risk.

Pollutant	Concentration	Cases/Controls	Unadjusted	Adjusted
	(µg/m3)		OR (95% CI)	OR (95% CI)
Polycyclic aromatic hydrocarbons
Tertile 1	< 2.27E-03	165/758	1.00 (ref.)	1.00 (ref.)
Tertile 2	2.27E-03 – 4.95E-03	218/767	1.31 (1.04, 1.64)	1.32 (1.03, 1.69)
Tertile 3	> 4.95E-03	191/770	1.14 (0.91, 1.45)	1.13 (0.88, 1.44)
			Ptrend = 0.28	Ptrend = 0.47
Polychlorinated biphenyls
Tertile 1	< 1.48E-05	180/747	1.00 (ref.)	1.00 (ref.)
Tertile 2	1.48E-05 – 3.43E-05	215/797	1.12 (0.90, 1.40)	1.11 (0.88, 1.39)
Tertile 3	> 3.43E-05	179/751	0.99 (0.79, 1.25)	0.98 (0.77, 1.24)
			Ptrend = 0.93	Ptrend = 0.84
Benzene
Tertile 1	< 3.46E-01	172/773	1.00 (ref.)	1.00 (ref.)
Tertile 2	3.46E-01– 6.26E-01	196/762	1.15 (0.92, 1.45)	1.16 (0.90, 1.49)
Tertile 3	> 6.26E-01	206/760	1.23 (0.97, 1.55)	1.20 (0.91, 1.57)
			Ptrend = 0.08	Ptrend = 0.21
1,3-Butadiene
Tertile 1	< 1.47E-02	175/755	1.00 (ref.)	1.00 (ref.)
Tertile 2	1.47E-02 – 3.36E-02	188/770	1.05 (0.83, 1.32)	1.03 (0.79, 1.35)
Tertile 3	> 3.36E-02	211/770	1.19 (0.95, 1.50)	1.14 (0.86, 1.51)
			Ptrend = 0.13	Ptrend = 0.35
Formaldehyde
Tertile 1	< 1.76E+00	185/764	1.00 (ref.)	1.00 (ref.)
Tertile 2	1.76E+00 – 2.01E+00	197/774	1.05 (0.84, 1.32)	1.03 (0.82, 1.31)
Tertile 3	> 2.01E+00	192/757	1.05 (0.83, 1.32)	0.99 (0.77, 1.27)
			Ptrend = 0.68	Ptrend = 0.92
Propylene dichloride
Tertile 1	< 3.68E-04	198/759	1.00 (ref.)	1.00 (ref.)
Tertile 2	3.68E-04 – 3.74E-04	194/766	0.97 (0.77, 1.22)	0.96 (0.77, 1.21)
Tertile 3	> 3.74E-04	182/770	0.90 (0.72, 1.14)	0.94 (0.74, 1.18)
			Ptrend = 0.38	Ptrend = 0.58
Trichloroethylene
Tertile 1	< 1.52E-02	172/764	1.00 (ref.)	1.00 (ref.)
Tertile 2	1.52E-02 – 2.78E-02	206/781	1.17 (0.94, 1.47)	1.17 (0.93, 1.48)
Tertile 3	> 2.78E-02	196/750	1.17 (0.93, 1.48)	1.12 (0.87, 1.44)
			Ptrend = 0.19	Ptrend = 0.41
Vinyl chloride
Tertile 1	< 8.93E-06	188/779	1.00 (ref.)	1.00 (ref.)
Tertile 2	8.93E-06 – 2.51E-05	210/747	1.16 (0.93, 1.45)	1.16 (0.92, 1.45)
Tertile 3	> 2.51E-05	176/769	0.95 (0.75, 1.20)	0.94 (0.74, 1.19)
			Ptrend = 0.66	Ptrend = 0.60
Arsenic
Tertile 1	< 7.99E-05	172/768	1.00 (ref.)	1.00 (ref.)
Tertile 2	7.99E-05 – 2.42E-04	181/760	1.06 (0.84, 1.34)	1.05 (0.82, 1.34)
Tertile 3	> 2.42E-04	221/767	1.30 (1.04, 1.63)	1.27 (0.98, 1.64)
			Ptrend = 0.02	Ptrend = 0.06
Cadmium compounds
Tertile 1	< 2.63E-05	193/772	1.00 (ref.)	1.00 (ref.)
Tertile 2	2.63E-05 – 3.29E-05	213/760	1.13 (0.91, 1.41)	1.09 (0.86, 1.37)
Tertile 3	> 3.29E-05	168/763	0.88 (0.70, 1.11)	0.85 (0.67, 1.08)
			Ptrend = 0.30	Ptrend = 0.17
Chromium compounds
Tertile 1	< 8.58E-05	170/766	1.00 (ref.)	1.00 (ref.)
Tertile 2	8.58E-05 – 4.97E-04	185/773	1.08 (0.85, 1.36)	1.10 (0.85, 1.41)
Tertile 3	> 4.97E-04	219/756	1.32 (1.05, 1.66)	1.31 (1.00, 1.70)
			Ptrend = 0.02	Ptrend = 0.04
Nickel compounds
Tertile 1	< 5.13E-05	187/766	1.00 (ref.)	1.00 (ref.)
Tertile 2	5.13E-05 – 9.54E-05	180/760	0.97 (0.77, 1.22)	0.93 (0.73, 1.18)
Tertile 3	> 9.54E-05	207/769	1.11 (0.88, 1.39)	1.03 (0.81, 1.32)
			Ptrend = 0.37	Ptrend = 0.76

^aModel adjusted for family history, state area deprivation index, education, menopausal status, physical activity, alcohol use, and body mass index.

^bHazardous air pollutant tertiles categorized based on the control distribution for each individual pollutant.

Having a first-degree family history of breast cancer was not a significant multiplicative interaction term and did not produce non-overlapping confidence intervals in the heterogeneity of effects models (Supplemental Table S1). Using joint effects, however, revealed that women with a family history of breast cancer had a significantly increased risk of breast cancer across HAP tertiles compared to those in the first tertile without a family history of breast cancer (Table 4). Using arsenic as an example, those with a family history of breast cancer had significantly increased risk of breast cancer in all tertiles compared to those with the lowest exposure and no family history (OR_T3|FX+ = 2.20; 95% CI 1.52 – 3.19, OR_T2|FX+ = 1.50; 95% CI 1.02 – 2.21, OR_T1|FX+ = 1.97; 95% CI 1.35 – 2.88). Having no family history and living in areas with moderate or high arsenic exposure was not significantly associated with breast cancer (OR_T3|FX- = 1.31; 95% CI 0.98 – 1.75, OR_T2|FX- = 1.17; 95% CI 0.89 – 1.55). A significant interaction of state ADI on the multiplicative scale was observed for propylene dichloride (P_interaction = 0.01), indicating that living in high exposure to propylene dichloride was inversely associated with breast cancer risk in less deprived regions, whereas it was positively associated with breast cancer risk in more deprived regions (Supplemental Table S2). Body mass index was not a significant interaction term on either scale for any pollutants (Supplemental Table S3).

Table 4.

Joint effects of first-degree family history of breast cancer and hazardous air pollutants in relation to overall breast cancer risk^a

	Without Family History			With Family History
Pollutant	Cases/Controls	OR	95% CI	Cases/Controls	OR	95% CI
Polycyclic aromatic hydrocarbons
Tertile 1	124/614	1.00	(ref.)	41/144	1.46	(0.98, 2.18)
Tertile 2	145/615	1.21	(0.91, 1.60)	73/152	2.46	(1.72, 3.52)
Tertile 3	141/607	1.15	(0.87, 1.52)	50/163	1.54	(1.04, 2.26)
Polychlorinated biphenyls
Tertile 1	132/607	1.00	(ref.)	48/140	1.59	(1.09, 2.33)
Tertile 2	150/634	1.09	(0.83, 1.42)	65/163	1.85	(1.30, 2.62)
Tertile 3	128/595	0.99	(0.76, 1.31)	51/156	1.51	(1.04, 2.20)
Benzene
Tertile 1	131/639	1.00	(ref.)	41/134	1.52	(1.02, 2.26)
Tertile 2	134/596	1.13	(0.85, 1.50)	62/166	1.91	(1.32, 2.76)
Tertile 3	145/601	1.19	(0.88, 1.61)	61/159	1.88	(1.29, 2.75)
1,3, Butadiene
Tertile 1	134/626	1.00	(ref.)	41/129	1.53	(1.03, 2.28)
Tertile 2	128/598	1.01	(0.75, 1.37)	60/172	1.67	(1.14, 2.44)
Tertile 3	148/612	1.12	(0.82, 1.53)	63/158	1.83	(1.25, 2.68)
Formaldehyde
Tertile 1	133/610	1.00	(ref.)	52/154	1.59	(1.10, 2.29)
Tertile 2	140/616	1.03	(0.79, 1.35)	57/158	1.64	(1.14, 2.36)
Tertile 3	137/610	0.98	(0.73, 1.30)	55/147	1.63	(1.12, 2.37)
Propylene dichloride
Tertile 1	141/603	1.00	(ref.)	57/156	1.57	(1.11, 2.24)
Tertile 2	136/605	0.96	(0.73, 1.25)	58/161	1.54	(1.08, 2.20)
Tertile 3	133/628	0.92	(0.71, 1.21)	49/142	1.53	(1.05, 2.24)
Trichloroethylene
Tertile 1	129/622	1.00	(ref.)	43/142	1.50	(1.01, 2.22)
Tertile 2	143/622	1.13	(0.86, 1.48)	63/159	1.96	(1.36, 2.81)
Tertile 3	138/592	1.10	(0.83, 1.47)	58/158	1.75	(1.20, 2.53)
Vinyl chloride
Tertile 1	142/619	1.00	(ref.)	46/160	1.29	(0.89, 1.88)
Tertile 2	146/607	1.05	(0.80, 1.36)	64/140	1.98	(1.39, 2.82)
Tertile 3	122/610	0.88	(0.67, 1.15)	54/159	1.49	(1.04, 2.14)
Arsenic
Tertile 1	122/633	1.00	(ref.)	50/135	1.97	(1.35, 2.88)
Tertile 2	132/590	1.17	(0.89, 1.55)	49/170	1.50	(1.02, 2.21)
Tertile 3	156/613	1.31	(0.98, 1.75)	65/154	2.20	(1.52, 3.19)
Cadmium compounds
Tertile 1	143/611	1.00	(ref.)	50/161	1.34	(0.93, 1.94)
Tertile 2	151/608	1.03	(0.79, 1.35)	62/152	1.71	(1.20, 2.43)
Tertile 3	116/617	0.78	(0.59, 1.02)	52/146	1.48	(1.02, 2.15)
Chromium compounds
Tertile 1	130/626	1.00	(ref.)	40/140	1.40	(0.94, 2.08)
Tertile 2	128/610	1.05	(0.79, 1.39)	57/163	1.78	(1.22, 2.59)
Tertile 3	152/600	1.25	(0.93, 1.68)	67/156	2.10	(1.46, 3.02)
Nickel compounds
Tertile 1	133/628	1.00	(ref.)	54/138	1.89	(1.31, 2.74)
Tertile 2	129/594	1.00	(0.76, 1.32)	51/166	1.41	(0.97, 2.05)
Tertile 3	148/614	1.08	(0.81, 1.42)	59/155	1.71	(1.18, 2.48)

^aConditional logistic regression model with PAH tertile x family history cross, product term adjusted for body mass index, education, state area deprivation index, menopausal status, physical activity, and alcohol consumption.

Association Between Breast Cancer and Pollutant Mixture

The Spearman correlation matrix for the hazardous air pollutants is shown in Supplemental Table S4 Correlations ranged from weak inverse correlations to highly positive correlations, with intraclass chemicals highly correlated (e.g., 1,3-butadiene & benzene ρ = 0.94). Using the fitted binomial unconditional logit model for the WQS index adjusted for confounders (state ADI, education, menopause, alcohol use, BMI, family history, physical activity, race, and age) produced a non-significant positive association of the weighted sum of exposures in tertiles for overall breast cancer. For a one tertile increase in the WQS index, representing the combined effects of the HAP mixture, the odds of having breast cancer increased by 1.21 times, though the OR was not statistically significant (OR_WQS = 1.21; 95% CI 0.96 – 1.53). Assessing the pollutant weights in the positive direction, chromium was weighted the most (0.47) followed by propylene dichloride (0.15) and polychlorinated biphenyl (0.09), representing the most important exposures in the HAP mixture in the association with breast cancer (Figure 1).

Figure 1.

Weighted quantile sum regression hazardous air pollutant weights in the positive direction.

Subgroup Analyses

Multivariable conditional logistic regression for the individual HAPs and breast cancer risk among early-onset (n = 129) or average age-onset (n = 445) (Supplemental Table S5), non-invasive (n = 133) or invasive (n = 441) (Supplemental Table S6), HR+ (n = 261) or HR- (n = 58) (Supplemental Table S7), and postmenopausal (n = 379) cases (Supplemental Table S8) revealed similar associations with overall findings, though slightly attenuated in some categories while intensified in others. Moderate trichloroethylene was positively associated with early-onset breast cancer, while moderate PAH and high chromium levels were positively associated with average age-onset breast cancer. Higher chromium levels tended to be associated with increased risk of invasive cancer. Finally, chromium was associated with increased risk of both HR+ and HR- cancers, while PAH and PCB tended to be positively associated with HR- cancer. WQS regressions were further performed within the average age-onset (OR = 1.04; 95% CI 0.81 – 1.33), invasive (OR = 1.11; 95% CI 0.87 – 1.43), HR+ (OR = 1.17; 95% CI 0.86 – 1.59), and postmenopausal (OR = 1.02; 95% CI 0.77 – 1.34) subgroups. Chromium was the highest weighted within all subgroups except for invasive cases, in which arsenic was weighted the most.

Discussion

This nested case-control study found increased risk of overall breast cancer for moderate PAH and high chromium airborne exposure, and found that chromium, propylene dichloride, and polychlorinated biphenyls contributed the most in the mixture regression. Joint effects models with first-degree family history of breast cancer provide evidence that underlying susceptibility may play a key role in the relationship between HAP exposure and breast cancer risk and should be investigated further. This is among the few cohorts that have examined NATA exposure estimates with breast cancer risk and focuses on a rural population in a southern state.

In the analysis, chromium was both significantly associated with breast cancer and was the highest weighted pollutant in the WQS regression. Heavy metals and metalloids are ubiquitous in the environment and occur both naturally in the earth’s crust and as by-products of industrial and agricultural processes.³¹ These pollutants possess both carcinogenic and xenoestrogen effects, making them doubly important for environmental epidemiology breast cancer investigations.³² In the Sister Study, researchers found null associations for chromium and overall breast cancer risk, and instead found associations for mercury, cadmium, and lead.¹⁰ An investigation in the California Teachers Study (CTS), another cohort of urban and rural participants in a single state, observed elevated risk of HR- tumors for those exposed to high cadmium and inorganic arsenic.³³ The current study did not find any significant associations for cadmium or arsenic in the receptor subgroup analysis, but did find associations for high chromium exposure and both HR+ and HR- tumors. Due to the small number of cases with receptor information available, results for this subgroup analysis may be hindered by reduced statistical power, and further efforts to capture this information in the cohort are warranted. In the current study, arsenic, cadmium, chromium, and nickel were chosen out of other metalloids for their known breast cancer carcinogenicity and were reported with adverse immunological, hematological, developmental, endocrinal, or reproductive effects.³⁴ The act of choosing specific chemicals for a potential HAP mixture is subjective, and the current analysis is by no means all-encompassing for the plethora of exposures faced by a population. A similar method was used in the CTS, in which researchers chose HAPs due to estrogen disruption evidence,³³ or other studies using broad variables such as particulate matter <2.5μm, nitrogen dioxide, and ozone to represent air pollutants.³⁵ The mean concentrations of heavy metals in the current study were similar orders of magnitude to the Sister Study but were lower on average.¹⁰

The second class of significant HAPs was polycyclic aromatic hydrocarbons, which are established carcinogens that are most widely associated with industrial practices. A recent meta-analysis focusing on PAH exposure and breast cancer risk found a suggestive positive association between PAH measured by geographical modeling and breast cancer risk.³⁶ However, women are not exposed to solely airborne PAH, and other sources of unmeasured exposure may confound the association observed currently, such as smoking, diet, and indoor air pollution.³⁷ PAH was also weighted below the significance threshold of 0.10 for the WQS weights, which may suggest that its individual association is driven by correlations with other chemicals. Measuring PAH exposure by NATA misses specific pollutants, such as benzo[a]pyrene, which could be solved by using newer sources of exposure, such as the EPA’s Air Toxics Screening Assessment (AirToxScreen) started in 2017 in a future study.

PCBs and VOCs resulted in insignificant individual effects, but propylene dichloride and polychlorinated biphenyls were the two highest weighted pollutants after chromium in the WQS regression. The Sister Study found that airborne propylene dichloride was suggestive of increased risk of overall breast cancer and ER+ cancer, along with statistical interaction with BMI.³⁸ The current study did not find any statistical interaction between BMI and any HAP examined. A Danish study focusing on indoor air PCB exposure on multiple cancer sites found null associations for breast cancer risk.³⁹ In the CTS, researchers found that benzene was significantly associated with increased risk for HR- tumors, and vinyl chloride was significantly associated with increased risk for HR+ tumors.⁴⁰ As discussed previously, the small sample size of those with receptor information available in the current study limits statistical power. Evidence for residential and indoor air exposure routes on breast cancer risk is still limited for these pollutants compared to biological measurements in tissue samples. However, the current study supports the argument that these HAPs should be studied further in other cohorts, especially if they are contributing highly to a specific pollutant mixture. Regarding the magnitude of exposure concentrations, VOCs had the highest averages among all the HAPs. Benzene, butadiene, formaldehyde, and trichloroethylene were orders of magnitude higher than PAHs, PCBs, and heavy metals. Even though the VOCs did not have a significant association with breast cancer, it does not mean they do not serve an important role in identifying potential public health concerns in Arkansas.

An area of focus for current environmental breast cancer research has been that of underlying susceptibility.⁶ The observed associations in the joint effects models indicate that carcinogenic air pollutants may be interacting with a familial history of breast cancer, even at low concentrations, which has been theorized previously.⁶ In the Sister Study, researchers found that familial risk, measured using the Breast and Ovarian Analysis of Disease Incidence and Carrier Estimation Algorithm, was a significant interaction term with nitrogen dioxide on both the additive and multiplicative scales.⁴¹ The authors discuss the plausibility that reduced expression of DNA repair genes among women with high familial risk could explain the synergistic findings.⁴¹ Regarding other groups of underlying susceptibility proposed by Zeinomar et al., the current study found only that moderate trichloroethylene was associated with early-onset cancer (<50 years old), though this group suffers from small sample size and results should be interpreted cautiously.⁶ While genetic susceptibility was not investigated in this study, on-going work examining multi-generational exposure among mother-daughter pairs in this cohort will provide opportunities to further investigate the associations of diet, environmental and occupational exposure, use of personal care products, and genetic susceptibilities with breast cancer risk in a rural population.

Strengths and Limitations

Strengths of the current study include minimal control selection bias due to the randomized selection process, though generalizability to the target population is at risk of selection bias due to the convenience sampling nature of the baseline cohort.²² Because women were recruited from breast cancer fundraising events, rates of family history of the disease, screening rates, and/or lifestyle choices may be different from the general population of Arkansas. However, this motivated cohort in a rural state, where environmental risk factors for cancer are severely understudied, offers a unique opportunity for investigation. Case identification is also a strength, as all hospitals in the state report cancer diagnoses to the Arkansas Central Cancer Registry,⁴² allowing for efficiency of data collection compared to self-reported diagnoses and medical record attainment.

There is potential for exposure misclassification using the EPA NATA data, which is the same risk of information bias as prior studies that have investigated outdoor air pollution and risk of breast cancer.^10,33 While NATA is a reliable source of information regarding exposure concentrations in large areas such as census-tracts, individuals living in the same tract will be assigned the same concentration value, leading to possible misclassification bias since cases may have the same value as controls, or vice versa. Rural census tracts represent larger land areas than urban census tracts and have lower exposure concentrations of air pollution from mobile and stationary sources, and may not accurately reflect individual practices such as wood/coal burning in rural households.^43,44 NATA also cannot be modeled in a longitudinal fashion, which poses a major limitation for epidemiological studies to assume concentrations stay constant over time.

The current study is also missing key covariates that could shed light on uncontrolled confounding or other sources of exposure. These include lifestyle factors such as smoking, personal care products, and cancer screening history/frequency, household factors such as trash, field, and leaf burning, and occupational information such as employment in industrial or agricultural sectors. These missing questions are planned to be included in future efforts in the cohort to obtain deep environmental and lifestyle exposure histories of participants. This analysis identified only cases that resided in Arkansas and controls that also lived in the state and provided an address at study enrollment, which limits transportability of results to other states. Multiple comparison was not corrected for in the analysis, which may increase the likelihood of Type I errors for the individual pollutant associations, which differs from previous research in the California Teachers Study,⁴⁰ but is similar to research in the Sister Study which also utilized weighted quantile sum regression.¹⁰ The small number of cases with hormone receptor information limits statistical power, and future efforts to collect this information from medical records are required, as discussed previously. Limitations also exist within the study design, as a prospective cohort design with repeated measurements could shed light on both the changes of exposure concentrations over time, and whether participants change residences. While the nested case-control design is efficient and allows for prospective findings regarding outcomes, it is not as effective as a prospective cohort design investigating risk factors of disease.

Conclusion

In conclusion, this nested case-control study within the Arkansas Rural Community Health Study found several statistically significant associations between carcinogenic HAPs and breast cancer risk, suggesting that outdoor air pollutants may increase breast cancer risk. Investigating joint effects, participants with a previous family history of breast cancer had significantly increased risk of breast cancer relative to those without familial history of cancer and low HAP exposure. The WQS regression was non-significantly associated with an increased risk of breast cancer for the pollutant mixture, with chromium, propylene dichloride, and polychlorinated biphenyls weighted the most. Further research is needed to understand the biological role of HAPs, their sources of exposure, dose-response effects, and potential interactions with underlying susceptibility in relation to breast cancer among rural populations.

Data Availability:

The data underlying this article cannot be shared due to the privacy of individuals that participated in the study. Summary level data will be shared on request to the corresponding author with permission of the advisory committee from the Arkansas Rural Community Health Study.