Residential proximity to dioxin emissions and risk of breast cancer in the Sister Study cohort

Abstract

Some dioxins are carcinogenic, but few studies have investigated the relationship between ambient polychlorinated dibenzo-p-dioxins and dibenzofurans (PCDD/F) and risk of breast cancer. We evaluated associations between proximity-based residential exposure to industrial emissions of PCDD/F and breast cancer risk in a large U.S. cohort. Sister Study participants at enrollment (2003–2009) were followed for incident breast cancer through September 2018. After restricting to participants with ≥10 years of residential history prior to enrollment (n=35,908), we generated 10-year distance- and toxic equivalency quotient (TEQ)-weighted average emissions indices (AEI [g TEQ/km²]) within 3, 5, or 10km of participants’ residences, overall and by facility type. Multivariable Cox regression models were used to estimate hazard ratios (HRs) and 95% confidence intervals (CIs) for the association between AEI quartiles (vs. zero AEI) and risk of breast cancer [invasive or ductal carcinoma in situ]. There were 2,670 incident breast cancer cases over 11 years (median) of follow-up. Breast cancer risk was increased for those in the highest quartile [Q] of AEI exposure within 3km (HR_Q4 :1.18, 95% CI: 0.99,1.40, P_trend=0.03). The HR was higher for the 10-year AEI at 3km from municipal solid waste facilities (HR_{≥median.vs.0}:1.50, 95% CI: 0.98, 2.29; P_trend=0.07). Risk was higher among ever smokers (vs. never smokers) in the top quartile of the 3km AEI (HR_Q4:1.41, 95% CI:1.12,1.77, P_trend=0.003; P_interaction=0.03) and higher risk for ER negative tumors was suggested (HR_Q4:1.47, 95% CI: 0.95, 2.28, P_trend=0.07, P_{heterogeneity}=0.17). Our findings suggest that residential exposure to PCDD/F emissions may confer an increased risk of breast cancer.

Introduction

Polychlorinated dibenzodioxins (PCDDs) and polychlorinated dibenzofurans (PCDFs), commonly referred to as dioxins and furans (hereafter, “dioxins”), are persistent organic pollutants that are frequently produced during the combustion of chlorinated organic compounds (1). Industrial facilities, including incinerators of solid, medical and hazardous waste, coal-fired power plants, and smelter operations, are major sources of environmental releases of dioxins (1). Dioxin emissions from industrial point sources contaminate the surrounding environment and have been detected in soil and residential dust from homes within 5km from these facilities (2–4). Although diet is the predominant source of exposure to dioxins, populations residing near dioxin-emitting facilities may be exposed to higher levels of ambient dioxins (2, 5).

Tetrachlorodibenzo-p-dioxin (TCDD) and pentachlorodibenzofuran (PeCDF) have been classified as Group 1 carcinogens by the International Agency for Research on Cancer (IARC) based on the strongest evidence in humans for all cancers combined and positive associations with soft-tissue sarcoma, non-Hodgkin lymphoma, and lung cancer (6). Dioxins are endocrine disrupting compounds and several in vivo studies have described that TCDD increases the susceptibility to breast cancer (6–10). Epidemiologic evidence for dioxin exposure and breast cancer is mostly from studies of accidental industrial releases (11–14) and occupational exposures (15–18), where dioxin exposures are substantially higher than residential ambient exposures. Women with TCDD exposure during the Seveso, Italy accident in 1976, measured as residence in contaminated areas or serum levels of TCDD, had elevated breast cancer incidence but not mortality (11–14). Female workers exposed to TCDD at an herbicide production plant in Hamburg, Germany had an increased risk of breast cancer mortality that became more apparent with increasing years of follow-up (up to 56 years) (15–17). However, few studies have evaluated residential exposure to dioxin and breast cancer risk and the results have been inconsistent (19–22). To clarify this relationship, we evaluated associations between proximity-based residential exposure to industrial air emissions of PCDD/F and breast cancer risk in a large, geographically-diverse U.S. cohort.

Material and methods

Study population

The Sister Study is a nationwide prospective cohort study of 50,884 women recruited in the U.S. and Puerto Rico from 2003–2009 (23). Eligible women were aged 35–75 and had a sister with breast cancer but no personal history of breast cancer. At enrollment, study participants completed a computer-assisted telephone questionnaire that included information on demographics, medical records, and reproductive/lifestyle factors. Participants also provided partial residential histories including residential address at enrollment, duration of residence at the enrollment address, and previous addresses including the longest-lived residence since the age of 20. Our analytic cohort included participants who had a well-geocoded (i.e., to the exact address or intersection) enrollment address and reported the start year of residence at their enrollment address (n=50,020; 98% of the original cohort, Supplementary Figure 1). All participants provided signed informed consent, and the Sister Study was approved by the institutional review boards of the National Institute of Environmental Sciences and the Copernicus Group. This analysis relied on Sister Study Data Release 6.0, which included follow-up data through September 2018.

Cohort follow-up and case ascertainment

Sister Study participants were followed from baseline (2003–2009) until the date of first breast cancer, the date of death or the end of study follow-up (23 September 2018), whichever came first. Incident cases of breast cancer were identified through annual self-report of health, follow-up questionnaires (every 2–3 year; >90% response rate) and medical records/pathology reports. High agreement between medical reports and self-reported breast cancer diagnosis and tumor characteristics has been reported (e.g., 99% positive predictive values for invasive breast cancer and 64% for in situ breast cancer) (24). We defined our outcome of interest as self-reported invasive or ductal carcinoma in situ (DCIS) breast cancer. We evaluated heterogeneity in the outcome by estrogen receptor (ER) status.

Dioxin database

We used a historical U.S. Environmental Protection Agency (USEPA) dioxin database (1), which was estimated to account for over 85% of air emissions from PCDD/F-emitting industrial facilities in the U.S. between 1987–2000 (1, 25), as the basis for PCDD/F exposure assessment in the cohort. Dioxin estimates were derived either from an actual release from the source (i.e., points of release from the source were sampled and evaluated) or through calculation based on an emission factor and activity level. There were ten industrial facility types in the dataset: medical waste incinerators, coal-fired electric generating facilities, sewage sludge incinerators, municipal solid waste incinerators (MSWIs), non-hazardous cement kilns, hazardous cement kilns, hazardous waste incinerators, industrial boilers, iron ore sintering plants, and secondary copper smelters. We also used an ESRI (Redlands, CA) database to identify hospital locations assumed to have medical waste incinerators (26). Previously, we verified the location of 84% of facility smokestacks in a subset of the database (n=240) (27). Our database included 4,478 facility smokestack locations (latitude, longitude) and annual emission estimates reported in 1987 (MSWIs and copper smelters only), 1995, and 2000, expressed as the toxic equivalence quotient (TEQ). TEQ, derived from World Health Organization-International Programme on Chemical Safety expert meeting in 1998, is a summed metric that weights congeners relative to the potency of TCDD using toxic equivalency factors that are established for all biologically active PCDD/Fs and dioxin-like polychlorinated biphenyls (PCBs) (28). Annual TEQs from 1987–2000 were estimated by applying a facility-type specific linear rate of change to the facility’s emissions; the applied linear rates of decline beginning in the 1980s are reflective of national temporal trends reported by the USEPA (25, 29).

Air dioxin emissions generally decreased after regulation of particular dioxin-emitting facilities (e.g. MSWIs, medical waste incinerators, hazardous waste incinerators) was promulgated in the U.S. in the late 1990s (30). To estimate annual emissions from these sources after 2000, we examined the secular trend in dioxin emissions by facility types between 2000 to 2009 (the end of enrollment for the cohort) using USEPA’s Toxics Release Inventory (31). We found that dioxin emissions were in the linear trend of slightly decreasing or relatively stable during this period with some fluctuations by facility types (Supplementary Table 1). Therefore, we assumed that 2000 TEQ emissions applied up to the enrollment year.

Residence history estimation

Our goal was to assess long-term (10+ years) exposure to ambient dioxin emissions. The cohort does not have full residence histories but reported their duration at the enrollment address as well as the longest-lived address since age of 20. Approximately 50% of the cohort (n=25,871) lived at the enrollment address for ≥10 consecutive years. To estimate long-term dioxin exposures for the other 50%, we combined information on the enrollment and longest-lived address. Nearly 59% of the cohort had the same enrollment and longest-lived addresses, 39% had different addresses, and 2% had missing information for their longest-lived address. Among those with different addresses, nearly 59% had less than a 3-year gap between the year in which they stopped living at their longest-lived address and the year in which they started living at their enrollment address. Therefore, when the duration at enrollment address was <10 years, we assumed participants with a ≤3-year gap had lived at their longest-lived address until the start year of the enrollment address. With this approach, we could estimate 10+ years of residential history for an additional 11,189 women (22% of the total cohort) with the median years of residence history captured as 19 years from the time of enrollment (interquartile range=14, 28). We then linked geocoded cohort enrollment and longest-lived addresses (32, 33) to the dioxin database of facility locations to estimate residential dioxin exposures.

Exposure metrics

For each participant, we first enumerated the total number of dioxin facilities and the number by each facility type (e.g., coal-fired electric generating facilities, MWSIs) that were within a Euclidean distance of 3, 5, or 10km of the participants’ address(es). The exposure period was defined as during the 10 years prior to enrollment. To calculate a 10-year average dioxin emission index for each participant, we linked annual TEQ values of facilities proximal to participants’ addresses for the years spent at each residence. For each residence-facility pair, we divided the TEQ by the squared distance from the residence. We summed annual distance-weighted TEQs for all facilities and by facility type for each participant and, then averaged over their 10-year exposure period to create a 10-year average emissions index (AEI [g toxic equivalency quotient (TEQ)/10-year]) (29). We created categorical exposures summarizing 10-year AEI for all facilities within a 3, 5, or 10km buffer (0, quartiles above zero (Q1, Q2, Q3, Q4)) and type-specific facilities within the same buffer (0, <median, ≥median; facility-specific AEIs). Interquartile ranges for the 10-year AEI at 3, 5, or 10km were approximately 1 g TEQ/km² (<0.001 g TEQ/km², 0.01 g TEQ/km², 0.05 g TEQ/km², respectively)

Covariates

We obtained information on demographic, health and lifestyle characteristics from the baseline questionnaire including age, self-reported race/ethnicity, examiner measured body mass index (BMI), attained education, smoking status, household income per person, menopause status, age at menarche and enrollment year. We also obtained information on urbanicity of residence from the 2000 U.S. Census (34) and residential outdoor air pollutants (NO₂, PM_2.5) which were estimated using previously described methods (33, 35). PM_2.5 and NO₂ have both been associated with increased risk of breast cancer (overall and DCIS) (35). We calculated 5-year averages of annual NO₂ and PM_2.5 levels through enrollment (e.g., average between 1999–2003, 2005–2009) for covariate adjustment. PM_2.5 data were available only starting in 1999, which did not allow us to compute 10-year average levels to correspond with dioxin exposure metrics. However, NO₂ measurements were available starting in 1993 and we found strong correlations between 5- and 10-year averages of this pollutant (rho=0.97).

Statistical analysis

Given that the expected latency of breast cancer is 10–20 years (36), we restricted analyses to participants with an estimated ≥10 years of residential history (n=35,972, Supplementary Figure 1.) then excluded participants missing the covariates (n=64, 5 cases). Our analytic cohort included 35,908 participants and 2,670 breast cancer cases. Selected participants were slightly older (median age=57 years) than excluded participants (median age=53 years) as expected given that we restricted women to those with ≥10 years of residential history, however, there were no significant differences in other characteristics.

We summarized demographic and lifestyle characteristics associated with breast cancer by quartiles of the 10-year AEI and conducted ANOVA tests for continuous variables and chi-squared tests for categorical variables to generate p-values for differences between groups. We fit Cox regression models to compute the hazard ratio (HR) and 95% confidence intervals (CIs) for the risk of breast cancer associated with the 10-year AEI (continuously per 1 g TEQ/km² increase and quartiles with zero AEI as the referent group) using age as the underlying time metric. Models were stratified by age group at baseline (<50, ≥50-<55, ≥55-<60, ≥60-<65, ≥65 years) and adjusted for a priori potential confounders (21, 35): self-reported race/ethnicity (non-Hispanic White, non-Hispanic Black, Hispanic, other), BMI (<24.9, 25–29.9, ≥30 kg/m³), attained education (≤high school degree or equivalent, some college/technical degree, four-year college or more), smoking status (never, ever), menopause status at baseline (pre, post), age at menarche (continuous), enrollment year, and urbanicity of residence (percent of the population whose residence is in an urban area in 2000, census tract level). Residential outdoor air pollutants (NO₂, PM_2.5) and household income were excluded from the covariates adjustment as they were not significant confounders in the association. To test for trend, we modelled a continuous AEI variable based on the midpoint of AEI quartiles and calculated a Wald statistic.

We conducted stratified analyses by several factors (baseline age group, race/ethnicity, smoking status, menopause status at baseline) and tested for multiplicative interaction using cross-product model terms. In the analyses by facility type, we computed 10-year facility-specific AEIs and adjusted the models for the other facility AEIs. We focused on six facility types (MSWIs, medical waste incinerators, coal-fired electric generating facilities, sewage sludge incinerators, hazardous waste incinerators, non-hazardous cement kilns); there was a limited number of other facility types across the U.S. and few cohort participants lived in close proximity to these sources. We also conducted analyses stratified by extent of disease (invasive, DCIS) and tumor ER status (ER positive, ER negative) and conducted joint Cox regression models by stratifying on these factors to test heterogeneity in associations (37). We further examined associations with the 15-year AEI in the subset of women with continuous residential history ≥15 years (n=26,300). We tested the proportional-hazards assumption using statistical tests based on the scaled Schoenfeld residuals and observed no indication of a violation after stratifying the baseline hazard by age category (3km, P=0.71; 5km, P=0.46; 10km, P=0.98). All statistical analyses were conducted in R Version 4.1.1 (38). P values less than 0.05 were considered statistically significant.

Results

A total of 2,670 incident breast cancer cases (2,078 invasive, 587 DCIS) were diagnosed over a median of 11 years of follow-up through September 2018. Participants with a higher 10-year AEI within 3km were more likely to be of non-Hispanic Black, ever smokers, and have a higher BMI (Table 1). Nearly 70% of women were postmenopausal at baseline. Of the reported breast cancer cases, 72% were ER positive (n=1,929), and 13% were ER negative (n=355).

Table 1.

Selected characteristics of Sister Study participants with 10 years of residential history prior to enrollment by quartile [Q] of 10-year AEI within 3km of participants’ residential addresses (n=35,908)

	10-year AEI (g TEQ/km2)
	0 (n=29,697)		>0-<0.04 (Q1, n =1,553)		≥0.04-<0.13 (Q2, n=1,552)		≥0.13-<0.41 (Q3, n=1,553)		≥0.41 (Q4, n=1,553)		Pdifference
Age at baseline, n, %											0.65
<50	6873	23.14%	370	23.82%	375	24.16%	373	24.02%	370	23.82%
≥50–<55	5935	19.99%	323	20.80%	338	21.78%	300	19.32%	299	19.25%
≥55–<60	6284	21.16%	321	20.67%	326	21.01%	340	21.89%	308	19.83%
≥60–<65	4903	16.51%	259	16.68%	245	15.79%	243	15.65%	260	16.74%
≥65	5702	19.20%	280	18.03%	268	17.27%	297	19.12%	316	20.35%
Age at menarche, years, median	13		13		13	0.84%	13	0.84%	13	0.84%	0.7
Race/ethnicity, n, %											<0.0001
non-Hispanic White	25973	87.46%	1316	84.74%	1320	85.05%	1273	81.97%	1246	80.23%
non-Hispanic Black	2184	7.35%	161	10.37%	180	11.60%	218	14.04%	249	16.03%
Hispanic	794	2.67%	44	2.83%	27	1.74%	28	1.80%	31	2.00%
Other	746	2.51%	32	2.06%	25	1.61%	34	2.19%	27	1.74%
Body mass index (BMI) (kg/m2), n, %											0.01
<24.9	11342	38.19%	640	41.21%	577	37.18%	577	37.15%	569	36.64%
25.0–29.9	9495	31.97%	478	30.78%	516	33.25%	456	29.36%	499	32.13%
≥30.0	8860	29.83%	435	28.01%	459	29.57%	520	33.48%	485	31.23%
The highest education completed, n, %										0.00%	<0.0001
≤High school degree or equivalent	4725	15.91%	181	11.65%	202	13.02%	223	14.36%	226	14.55%
Some college/technical degree	10115	34.06%	472	30.39%	481	30.99%	468	30.14%	481	30.97%
Four-year college or more	14857	50.03%	900	57.95%	869	55.99%	862	55.51%	846	54.48%
Smoking status, n (%)											<0.0001
Never smoker	16926	57.00%	826	53.19%	828	53.35%	846	54.48%	819	52.74%
Ever smoker	12771	43.00%	727	46.81%	724	46.65%	707	45.52%	734	47.26%
Menopause status at baseline, n, %											0.36
No	8745	29.45%	477	30.71%	479	30.86%	479	30.84%	476	30.65%
Yes	20952	70.55%	1076	69.29%	1073	69.14%	1074	69.16%	1077	69.35%

We conducted ANOVA tests for continuous variables and chi-squared tests for categorical variables to generate P for difference between groups. AEI: average emissions index.

Most women (83%) did not live within 3km of dioxin emitting facilities. Compared to the referent group (zero AEI at 3km; 83% of participants), we observed HRs of 0.76 (95% CI 0.61, 0.94), 0.90 (0.74, 1.09), 1.11 (0.93, 1.33) and 1.18 (0.99, 1.40) for increasing quartiles of non-zero 10-year AEI exposure within 3km (P_trend=0.03, Table 2). HRs were not consistently elevated at 5km (HR_Q4.vs.0: 1.07, 95% CI: 0.93, 1.23, P_trend=0.18) or 10km (HR_Q4.vs.0: 1.03, 95% CI: 0.92, 1.16, P_trend=0.26). For continuous exposure, a 1 g TEQ/km² increase in the 10-year AEI was not associated with breast cancer risk at any distance (3km, 5km, or 10km; Table 2). In the analyses by facility type, we observed that 10-year dioxin emissions from MSWIs were marginally associated with increased breast cancer (3km: HR_{≥median vs.0}:1.50, 95% CI: 0.98, 2.29; P_trend=0.07, Table 3) and medical waste incinerators (3km: HR_{≥median vs.0}:1.14, 95% CI: 0.99, 1.32; P_trend=0.08). Associations with dioxin emissions from other facility types were null.

Table 2.

Hazard ratios and 95% confidence intervals of invasive breast cancer or ductal carcinoma in situ among Sister Study participants with 10 years of residential history prior to enrollment by quartile [Q] of 10-year AEI within 3km, 5km, and 10km of participants’ residential addresses (n=35,908)

10-year AEI (g TEQ/km2)	N cases	Age-stratified HR (95% CI)	Multivariable-adjusted HR (95% CI)*
3km
0	2204	1	1
>0-<0.04 (Q1)	91	0.78 (0.63, 0.96)	0.76 (0.61, 0.94)
≥0.04-<0.13 (Q2)	107	0.93 (0.76, 1.13)	0.90 (0.74, 1.09)
≥0.13-<0.41 (Q3)	131	1.15 (0.97, 1.38)	1.11 (0.93, 1.33)
≥0.41 (Q4)	137	1.22 (1.02, 1.45)	1.18 (0.99, 1.40)
Ptrend		0.01	0.03
per 1 g/ TEQ/km2 increase	2670	1.01 (0.99, 1.02)	1.00 (0.99, 1.02)
5km
0	1799	1	1
>0-<0.02 (Q1)	199	0.92 (0.80, 1.07)	0.89 (0.77, 1.04)
≥0.02-<0.07 (Q2)	198	0.92 (0.79, 1.06)	0.89 (0.77, 1.03)
≥0.07-<0.25 (Q3)	238	1.12 (0.98, 1.28)	1.07 (0.94, 1.23)
≥0.25 (Q4)	236	1.12 (0.98, 1.28)	1.07 (0.93, 1.23)
Ptrend		0.05	0.18
per 1 g/ TEQ/km2 increase	2670	1.00 (0.99, 1.02)	1.00 (0.99, 1.02)
10km
0	1082	1	1
>0-<0.009 (Q1)	388	0.98 (0.87, 1.10)	0.95 (0.84, 1.07)
≥0.009-<0.03 (Q2)	385	0.97 (0.86, 1.09)	0.93 (0.83, 1.05)
≥0.03-<0.16 (Q3)	388	0.99 (0.88, 1.11)	0.94 (0.84, 1.06)
≥0.16 (Q4)	427	1.10 (0.98, 1.23)	1.03 (0.92, 1.16)
Ptrend		0.05	0.26
per 1 g/ TEQ/km2 increase	2670	1.00 (0.99, 1.02)	1.00 (0.99, 1.02)

^*Stratified by age-group and adjusted for race, body mass index, education, smoking status, menopause status, age of menarche, enrollment year, percent of the population whose residence is in an urban area (tract level)

AEI: average emissions index, HR: Hazard ratio, CI: confidence interval

Table 3.

Hazard ratios and 95% confidence intervals of invasive breast cancer or ductal carcinoma in situ for dioxin emissions within 3km by facility type among Sister Study participants with 10 years of residential history prior to enrollment (n=35,908)

10-year AEI (g TEQ/km2)	N cases	Multivariable-adjusted HR (95% CI)*
Municipal solid waste incinerators
0	2639	1
<0.34 (median)	9	0.61 (0.32, 1.17)
≥0.34 (median)	22	1.50 (0.98, 2.29)
Ptrend		0.07
Medical waste incinerators
0	2290	1
<0.13 (median)	166	0.86 (0.73, 1.00)
≥0.13 (median)	214	1.14 (0.99, 1.32)
Ptrend		0.08
Coal-fired electric generating facilities
0	2613	1
<0.11 (median)	30	1.11 (0.77, 1.59)
≥0.11 (median)	27	0.91 (0.62, 1.34)
Ptrend		0.63
Sewage sludge incinerators
0	2642	1
<0.007 (median)	9	0.55 (0.29, 1.06)
≥0.007 (median)	19	1.14 (0.72, 1.79)
Ptrend		0.83
Hazardous waste incinerators
0	2662	1
<0.002 (median)	<5	0.97 (0.36, 2.58)
≥0.002 (median)	<5	0.94 (0.35, 2.52)
Ptrend		0.83
Non-hazardous cement kilns
0	2664	1
<0.02 (median)	<5	1.12 (0.28, 4.49)
≥0.02 (median)	<5	1.73 (0.65, 4.62)
Ptrend		0.27

^*Stratified by age-group and adjusted for race, body mass index, education, smoking status, menopause status, age of menarche, enrollment year, percent of the population whose residence is in an urban area (tract level), 10-year AEIs for other facilities

AEI: average emissions index, HR: Hazard ratio, CI: confidence interval

In the analyses stratified by smoking status, we found a positive association among ever smokers in the highest quartile of dioxin emissions (HR_Q4.vs.0:1.41, 95% CI: 1.12,1.77, P_trend=0.003, Table 4) but not among never smokers (HR_Q4.vs.0:0.95, 95% CI: 0.73,1.25, P_trend=0.91, P_interaction=0.03). We found a slightly stronger relationship among non-Hispanic Black women (HR_Q4.vs.0:1.21, 95% CI: 0.77, 1.90, P_trend=0.31) compared to White women (HR_Q4.vs.0:1.13, 95% CI: 0.93, 1.37, P_trend=0.15), but the difference was not statistically significant (P_interaction=0.32). The associations also did not differ by age at baseline (P_interaction=0.92) or menopausal status at baseline (P_interaction=0.71). We observed higher HRs among ER negative tumors associated with the 10-year AEI at any distance although differences by ER status were not statistically significant (Table 5). Associations with 10-year AEI were similar for invasive breast cancer and DCIS at all distances (Supplementary Table 2). We observed similar associations with the 15-year AEI (HR_Q4.vs.0:1.13, 95% CI: 0.93, 1.38, P_trend=0.13, Supplementary Table 3) among participants with an estimated residential history ≥15 years (n=26,300) compared to those with 10-year AEI (Table 2).

Table 4.

Hazard ratios and 95% confidence intervals of invasive breast cancer or ductal carcinoma in situ among Sister Study participants with 10 years of residential history prior to enrollment by quartile [Q] of 10-year AEI within 3km of participants’ residential addresses stratified by age, race, smoking status, and menopausal status at baseline (n=35,908)

Multivariable-adjusted HR (95% CI)
	0	>0-<0.04 (Q1)	≥0.04-<0.13 (Q2)	≥0.13-<0.41 (Q3)	≥0.41 (Q4)	Ptrend	Pinteraction
Age							0.92
<50 (n=8361)	1	0.68 (0.41, 1.12)	0.94 (0.61, 1.44)	1.13 (0.76, 1.69)	1.28 (0.88, 1.87)	0.15
≥50-<55 (n=7195)	1	0.83 (0.52, 1.33)	1.07 (0.72, 1.61)	0.78 (0.47, 1.29)	1.07 (0.69, 1.66)	0.85
≥55-<60 (n=7579)	1	0.58 (0.34, 0.97)	0.93 (0.62, 1.41)	1.35 (0.96, 1.91)	1.20 (0.82, 1.76)	0.19
≥60-<65 (n=5910)	1	0.80 (0.51, 1.27)	0.79 (0.48, 1.28)	1.23 (0.82, 1.83)	1.24 (0.84, 1.83)	0.20
≥65 (n=6863)	1	0.92 (0.60, 1.40)	0.76 (0.48, 1.20)	0.99 (0.67, 1.46)	1.12 (0.78, 1.62)	0.54
Race							0.32
White, non-Hispanic (n=31,128)	1	0.73 (0.58, 0.92)	0.94 (0.76, 1.15)	1.08 (0.89, 1.32)	1.13 (0.93, 1.37)	0.15
Black, non-Hispanic (n=2,992)	1	0.89 (0.47, 1.69)	0.70 (0.35, 1.37)	1.39 (0.87, 2.20)	1.21 (0.77, 1.90)	0.31
Smoking status							0.03
Never smoker (n=20,245)	1	0.82 (0.62, 1.09)	0.99 (0.77, 1.29)	1.31 (1.04, 1.64)	0.95 (0.73, 1.25)	0.91
Ever smoker (n=15,663)	1	0.69 (0.50, 0.94)	0.81 (0.60, 1.08)	0.92 (0.69, 1.22)	1.41 (1.12, 1.77)	0.003
Menopausal status at baseline							0.71
Premenopausal women (n=10,656)	1	0.90 (0.62, 1.31)	0.87 (0.59, 1.27)	1.01 (0.71, 1.45)	1.31 (0.95, 1.82)	0.10
Postmenopausal women (n=25,252)	1	0.70 (0.55, 0.91)	0.91 (0.73, 1.15)	1.15 (0.94, 1.41)	1.13 (0.92, 1.39)	0.13

^*Stratified by age-group and adjusted for race, body mass index, education, smoking status, menopause status, age of menarche, enrollment year, percent of the population whose residence is in an urban area (tract level)

Race/ethnicity category includes non-Hispanic White, non-Hispanic Black, Hispanic, and other. Number of breast cancer cases by average emissions index at 3km were too few to evaluate Hispanic (n=924) and other (n=864) group separately

AEI: average emissions index, HR: Hazard ratio, CI: confidence interval. Q1: >0-<0.04 g TEQ/km², Q2: ≥0.04-<0.13 g TEQ/km², Q3: ≥0.13-<0.41 g TEQ/km², Q4: ≥0.41 g TEQ/km²

Table 5.

Hazard ratios and 95% confidence intervals of invasive breast cancer or ductal carcinoma in situ among Sister Study participants with 10 years of residential history prior to enrollment by quartile [Q] of dioxin emissions from facilities within 3, 5, and 10 km of participants’ residential addresses stratified by estrogen receptor (ER) status

		ER status				Pheterogeneity
	ER+		ER−
10- year AEI (g TEQ/km2)	N cases	Multivariable-adjusted HR (95% CI)	N cases	Multivariable-adjusted HR (95% CI)
3km					0.17
0	1585	1	303	1
>0-<0.04 (Q1)	71	0.81 (0.64, 1.03)	6	0.38 (0.17, 0.86)
≥0.04-<0.13 (Q2)	81	0.94 (0.75, 1.18)	10	0.63 (0.33, 1.18)
≥0.13-<0.41 (Q3)	92	1.09 (0.88, 1.34)	14	0.92 (0.54, 1.58)
≥0.41 (Q4)	100	1.20 (0.98, 1.47)	22	1.47 (0.95, 2.28)
Ptrend		0.05		0.07
per 1 g/ TEQ/km2	1929	1.01 (0.99, 1.02)	355	1.01 (0.98, 1.04)
increase
5km					0.47
0	1298	1	242	1
>0-<0.02 (Q1)	152	0.93 (0.79, 1.10)	20	0.71 (0.45, 1.13)
≥0.02-<0.07 (Q2)	143	0.88 (0.74, 1.04)	28	0.99 (0.67, 1.47)
≥0.07-<0.25 (Q3)	169	1.05 (0.89, 1.24)	29	1.05 (0.71, 1.55)
≥0.25 (Q4)	167	1.05 (0.89, 1.23)	36	1.33 (0.93, 1.90)
Ptrend		0.42		0.08
per 1 g/ TEQ/km2	1929	1.01 (0.99, 1.02)	355	1.01 (0.98, 1.04)
increase
10km					0.69
0	795	1	148	1
>0-<0.009 (Q1)	272	0.89 (0.78, 1.03)	45	0.88 (0.62, 1.23)
≥0.009-<0.03 (Q2)	275	0.89 (0.78, 1.03)	49	0.94 (0.68, 1.32)
≥0.03-<0.16 (Q3)	285	0.92 (0.80, 1.06)	51	1.01 (0.72, 1.40)
≥0.16 (Q4)	302	0.98 (0.86, 1.13)	62	1.24 (0.90, 1.70)
Ptrend		0.58		0.08
per 1 g/ TEQ/km2 increase	1929	1.00 (0.99, 1.02)	355	1.01 (0.98, 1.03)

^*Stratified by age-group and adjusted for race, body mass index, education, smoking status, menopause status, age of menarche, enrollment year, percent of the population whose residence is in an urban area (tract level)

AEI: average emissions index, HR: Hazard ratio, CI: confidence interval, ER: estrogen receptor

Discussion

In this U.S.-wide study, we found an increased risk of breast cancer associated with the highest quartile of estimated PCDD/F exposure within 3km of the home. We observed that associations were attenuated for dioxin exposure within 5 or 10km, indicating risks are increased the most for individuals living in close proximity to these sources. Associations were strongest for emissions from MSWIs than any other facility type and elevated among ever smokers relative to never smokers but were not different by race/ethnicity. Stronger associations were observed among ER negative tumors although differences were not statistically significant.

A small number of studies have evaluated breast cancer risk and exposure to ambient dioxin exposures, and results have been inconsistent. A recent investigation in a US cohort of female nurses using the same facility emissions database to estimate dioxin exposures did not find an association between invasive breast cancer risk and emissions from all facilities located within 3km (HR_{>0.09 (median) vs.0}:1.00, 95% CI: 0.90, 1.11), 5km, or 10km of residences (21). However, the study reported positive associations for emissions from MSWIs located within 3km (HR_{>4.70 (median) vs.0}:1.40, 95% CI: 1.03, 1.92) as well as 10km (HR_{>0.48 (median) vs.0}:1.16, 95% CI: 1.03, 1.32). In our analysis restricting to invasive breast cancer, we found a positive relationship in the highest quartile of 10-year AEI within 3km without monotonic trend. Similar to VoPham et al (21), we did not find statistically significant associations with 10-year AEI (from all facilities) within 5 or 10km. These findings may reflect a decline in risk, which suggests that associations with breast cancer risk declines with increasing distance between a participant and the source. Our results of weak associations at >3km were reasonably expected given that half-lives of many dioxins in air are relatively short (<10 days) (39). However, airborne PCDD/Fs may be transported long distances away from the source and their physiochemical properties result in slow degradation in the environment - PCDD/Fs from the facilities have been detected within about 10km in soil samples (e.g. deposition of air PCDD/Fs to soils), and higher levels of air PCDD/Fs have been found close to emitting facilities (39–41). A nested case-control study in France found no association between invasive breast cancer and quintiles of cumulative dioxin exposures (P_trend=0.81) (19), although exposure levels in that study were lower than ours (highest quintile cutoff: 0.10 μg/ TEQ/m²). Another French case-control study assessing dioxin emissions from MSWIs (highest vs. lowest exposures) also found no association with invasive breast cancer among women aged 20–59 years but reported a reduced risk among women aged ≥60 years, although the number of cases in the highest exposure category was small (22). In our study with a larger number of cases, we found positive associations across all age groups, including women aged ≥60 years.

Our findings can also be compared with studies investigating breast cancer risk with surrogates of environmental dioxin exposures. An ecologic analysis in Michigan, U.S. found that ZIP code-level PCDD/F concentrations in soil were positively associated with incident breast cancer rates (42). A case-control study in Spain observed positive associations with breast cancer among women residing within 3km of organic chemical industries (OR 1.68, 95% CI: 0.98–3.76), which release dioxins to air and water (20). Women residing in Chapaevsk, Samara Region, Russia, where a large chemical plant releasing PCDD/Fs was located, had an increased risk for breast cancer mortality than the average mortality in the Samara Region (SMR 2.1, 95% CI: 1.6–2.7) (18). However, in a U.S. hospital-based case-control study, breast cancer risk was not associated with adipose levels of PCDD/PCDFs (43).

Our findings of stronger associations among dioxin emissions from MSWIs are consistent with findings from the Nurses’ Health Study cohort (21). MSWIs were the leading source of dioxin emissions in the U.S. in 1987 and 1995, prior to emissions controls (1). We observed a non-significantly increased risk of breast cancer associated with dioxin emissions from medical waste incinerators (P_trend=0.08) but no associations with other facility types (coal-fired electric generating facilities, sewage sludge incinerators, hazardous waste incinerators) were found. A positive association with wide confidence intervals was observed for emissions from non-hazardous cement kilns (>median vs. 0) due to small case numbers (n<5).

We found a significant positive trend in the associations among ever smokers but not among never smokers (P_interaction=0.03). Other than our study, only one nested case-control study presented results stratified by smoking status at baseline but found no difference in the association between ever and never smokers (P_interaction=0.92) (19). A study of 341 Taiwanese found that active smokers had slightly lower serum levels of the total WHO TEQ from PCDD/F than either passive or nonsmokers (44). Studies using data from the U.S. National Health and Nutrition Examination Survey confirmed that the TEQ from PCDD/F and PCBs for smokers was lower than nonsmokers for both sexes (45, 46). An earlier study among 48 workers occupationally exposed to dioxins showed that some PCDD/F congeners in the blood of smokers decayed significantly faster (e.g. 30% for TCDD, 100% for 1,2,3,4,7,8- Hexachlorodibenzo-p-dioxin) than those in the blood of non- and ex-smokers (47). These findings suggest that the induction of cytochrome P-450 (CYP)1A2 by aryl hydrocarbons (Ah) from exposure to tobacco smoke increases the elimination of PCDD/F and PCB due to common metabolic pathways for these chemicals (45, 47). Given that limited evidence of smoking-stratified results is available, our smoking findings need replication.

Evidence from experimental studies shows the biological plausibility of our observed associations between dioxin exposure and breast cancer risk. Dioxins are endocrine disrupting compounds binding to the Ah receptor (AhR), which leads to changes in gene expression, cell replication, and apoptosis (6). AhR activation by TCDD in both tumorigenic (MCF-7) and non-tumorigenic (MCF-10A) breast cells suppress the apoptotic response induced by UV-irradiation and chemotherapeutic agents (9). M13SV1 human breast luminal epithelial cells treated with TCDD induce tumorigenicity by activating extracellular signal-regulated kinase 2 (ERK2) and protein kinase B (AKT) pathways (48). Several in vivo studies have shown the effects of TCDD on the mammary glands of rats in different life stages. Pre- and perinatal exposure to TCDD delays mammary gland development by increasing number of terminal end buds, which increases the susceptibility to breast cancer (7, 8). In utero exposure to TCDD also reduces CpG methylation of the breast cancer-1 (BRCA-1) expression in mammary tissue of offspring (10).

There are some limitations to our study. First, misclassification of exposure from imputation could not be avoided since emissions data were not available annually, especially after 2000. However, air dioxin emissions were generally lower in the 2000s after regulation of their sources (30) and we also confirmed that the trend of emissions between 2000 and 2009 were slightly decreasing or relatively constant across all facilities (31) (Supplementary Table 1). Second, we could not investigate lifetime exposures to dioxin since we did not have full residential address histories on the cohort and the dioxin emissions database was developed in the 1980s, although the 10-year AEI likely captured some of higher dioxin emissions prior to regulations. Third, we were not able to conduct well-powered analyses stratified by ER status within each racial/ethnic group as most participants were White (87%) and most cases were ER positive (72%). Hispanic and non-Hispanic Black women are more likely to be diagnosed with ER negative breast cancer (49) and were observed to have higher exposure to dioxin emissions compared to White women in this study. Lastly, we did not examine dietary exposure of dioxins(5), however, airborne dioxins are not likely correlated with dietary intake of dioxin and therefore our analyses are unlikely to be confounded by dietary exposures.

The prospective cohort design is the main strength of our study since most studies investigating residential exposures to dioxins and breast cancer risk have been case-control studies. Our results add to the limited evidence supporting the association between environmental dioxin exposures and breast cancer risk. We assessed dioxin exposures by incorporating 10-year residential address histories and a nationwide regulatory database of dioxin-emitting facilities, for which locations were previously validated (27). Although the latency between dioxin exposure and breast cancer is unknown, our exposure metrics for historical 10-year exposure to dioxin allowed for some latent period (36). We captured lower ambient dioxin exposures (post regulations, late 1990s and 2000s) in our exposure metrics but, because of the wide geographic spread of participants, our analyses included sufficient heterogeneity in exposure to observe an association among those with higher dioxin emissions within 3km of their residences. We also examined dioxin exposures from specific types of facilities to capture the unique mixture(s) emanating from different emission sources. Lastly, we were able to conduct analyses stratified by smoking status, extent of disease, ER status, and race/ethnicity, which were not fully addressed in previous studies.

Conclusions

Our findings suggest that increased residential exposure to PCDD/F emissions may be associated with an increased risk of breast cancer. This analysis contributes to additional evidence of a positive association at lower levels of air dioxin emissions. Additional studies to follow up our suggestive findings of differential associations by ER status and race/ethnicity are warranted.

Highlights.

• Dioxins are persistent organic pollutants, and some are known carcinogens

• Research on air emissions of dioxins and breast cancer risk is limited

• Airborne dioxins within 3km were associated with increased breast cancer risk

• This association was strongest for emissions from municipal solid waste facilities

Abbreviations:

AhR

Aryl-hydrocarbon

AhR

Aryl-hydrocarbon receptor

AEI

Average emissions index

BMI

Body mass index

CIs

Confidence intervals

ER

Estrogen receptor

ERK

Extracellular signal-regulated kinase

HRs

Hazard ratios

MSWI

Municipal solid waste incinerator

PeCDF

Pentachlorodibenzofuran

PCBs

Polychlorinated biphenyls

PCDDs

Polychlorinated dibenzodioxins

PCDFs

Polychlorinated dibenzofurans

PCDD/F

Polychlorinated dibenzo-p-dioxins and dibenzofurans

AKT

Protein kinase B

TCDD

Tetrachlorodibenzo-p-dioxin

TEQ

Toxic equivalence quotient

USEPA

U.S. Environmental Protection Agency