Ethylene oxide emissions and incident breast cancer and non-Hodgkin lymphoma in a US cohort

Abstract

Background

Ethylene oxide (EtO) is a carcinogenic gas used in chemical production and to sterilize medical equipment that has been linked to risk of breast and lymphohematopoietic cancers in a small number of occupational studies. We investigated the relationship between environmental EtO exposure and risk of these cancers.

Methods

Using the US Environmental Protection Agency’s Toxics Release Inventory, we estimated historical exposures for National Institutes of Health–AARP Diet and Health Study participants enrolled in 1995-1996. We constructed 2 metrics at 3, 5, and 10 km: 1) distance between residences and EtO-emitting facilities, weighted by the proportion of time the home was downwind of each facility, and 2) distance-weighted, wind direction–adjusted average airborne emissions index (AEI=∑[lbs EtO/km²]). We estimated risk (hazard ratio [HR], 95% confidence interval [CI]) of incident breast cancer (in situ and invasive) among postmenopausal women (n = 173 670) overall and by tumor estrogen receptor status and non-Hodgkin lymphoma in the full cohort (n = 451 945).

Results

We observed an increased risk of breast cancer associated with EtO-emitting facilities within 10 km (HR_[≤10vs>10] = 1.05, 95% CI = 1.00 to 1.10) that appeared stronger for in situ (HR_[≤10vs>10] = 1.13, 95% CI = 1.00 to 1.27) than invasive (HR_[≤10vs>10] = 1.03, 95% CI = 0.97 to 1.09) disease. Risk of breast cancer in situ was also increased in the top AEI quartiles, and associations weakened with larger distances (HR_[Q4vs0] = 1.60, 95% CI = 0.98 to 2.61; HR_[Q4vs0] = 1.28, 95% CI = 0.92 to 1.79; HR_[Q4vs0] = 1.25, 95% CI = 1.02 to 1.53 at 3, 5, and 10 km, respectively). No differences in breast cancer risk were observed by estrogen receptor status. We found no clear pattern of increased non-Hodgkin lymphoma risk.

Conclusions

A novel potential association between EtO emissions and risk of in situ, but not invasive, breast cancer warrants additional evaluation.

Ethylene oxide (C₂H₄O; EtO) is a colorless, odorless gas used primarily in the production of ethylene glycol and other chemicals involved in the manufacture of antifreeze, adhesives, detergents, and some plastics and is used to sterilize medical equipment (1). EtO is released into the atmosphere by these facilities and degrades slowly in ambient air (estimated half-life of 1 month to 1 year) by reaction with photochemically produced hydroxyl radicals (2-4). Commercial sterilization operations, chemical plants, and medical facilities are the main occupational and environmental sources of exposure (5). The US Environmental Protection Agency (EPA) monitors EtO and other hazardous air pollutants through the Toxics Release Inventory (TRI), which requires reporting of emissions from specific industrial sectors. EtO has been monitored in the TRI since 1987.

EtO is classified as a human carcinogen by the International Agency for Research on Cancer based on sufficient evidence in experimental animals, strong evidence from mechanistic studies, and limited evidence from studies demonstrating an increased risk for lymphohematopoietic (primarily lymphoma) and breast cancers in occupationally exposed workers (3). The most compelling epidemiologic data come from a large National Institute for Occupational Safety and Health (NIOSH) cohort, which found increased mortality from lymphohematopoietic malignancies (eg, chronic lymphocytic leukemia [CLL], non-Hodgkin lymphoma [NHL]) (6,7) and breast cancer incidence (7) and mortality (8) associated with high cumulative exposure to EtO. Few studies have evaluated cancer risks from environmental exposure. In one US cohort using census tract–level exposure estimates, EtO was weakly and nonstatistically significantly associated with risk of invasive breast cancer (9). The only nonoccupational study of lymphohematopoietic malignancies found statistically significant clustering of diffuse large B-cell lymphoma (DLBCL) cases around 57% of EtO-emitting facilities in Georgia. (10).

The number of EtO point sources and emissions has decreased since the late 1980s (11). Data on EtO levels in the ambient environment are limited, and population exposures have not been well characterized, but those residing close to EtO point sources may be exposed to higher levels (4,12). With the epidemiologic evidence for carcinogenicity (specifically lymphoid and breast cancers) derived from occupational settings and the continued release of this known carcinogen, investigation of environmental EtO exposure and these cancers is needed. We evaluated these associations in a large US cohort using a quantitative point source–based exposure assessment at participant residences.

Methods

Study population and cancer ascertainment

The National Institutes of Health (NIH)–AARP Diet and Health Study cohort was recruited in 1995-1996 from the membership of the AARP aged 50-71 years in 6 states (California, Florida, Louisiana, New Jersey, North Carolina, and Pennsylvania) and 2 metropolitan areas (Atlanta, GA, and Detroit, MI). Study details have been described (13). Briefly, participants self-reported demographics (including race and ethnicity), lifestyle, dietary, reproductive characteristics, and family history of cancer. The study was approved by the NIH institutional review board. Of 566 398 participants at baseline, we excluded those with a pre-enrollment cancer diagnosis except nonmelanoma skin cancer (n = 51 346), whose death was attributed to cancer but not found in registries (n = 4269) and proxy respondents (n = 15 760). Among the remaining cancer-free participants, we excluded those with missing or poorly geocoded addresses (n = 43 078) or for whom person-time could not be estimated (n = 69; Supplementary Methods, available online).

Incident cancer cases were identified through linkage to cancer registries in the catchment area states and those to which participants most frequently moved (Arizona, Nevada, and Texas). Vital status was determined by linkage with the National Death Index. Person-years were calculated from enrollment until the earliest date of cancer diagnosis, relocation from the registry areas as determined through routine tracing efforts, death, or the end of follow-up (December 31, 2011). Over a median 16 years of follow-up, 12 222 breast cancer cases were diagnosed among postmenopausal women, including 7666 (62.7%) with known hormone receptor status. Breast cancers were grouped as either ductal carcinoma in situ (DCIS; n = 1974 [16.2%] cases) or invasive (n = 9918 [81.1%] cases) (14); 2.7% were missing this information. During the same period, 6484 NHL cases were diagnosed. Major subtypes (15) included CLL/small lymphocytic lymphoma (CLL/SLL; 25.4%), DLBCL (20.0%), follicular lymphoma (12.2%), and multiple myeloma (19.8%). Five subtypes with lower frequency comprised the remaining 22.6%, including 2.1% NHL not otherwise specified.

Exposure assessment

Estimated EtO emissions were derived from TRI, which compiles industry-reported annual releases of 500 or more chemicals to air, water, and land from approximately 53 000 facilities. Most (>90% annually) EtO emissions reported to TRI are released to air from stacks or as fugitive emissions; therefore, we included only airborne emissions.

EtO has a long half-life in air and is heavier than air, and transport over multiple kilometers with prevailing winds has been demonstrated (4,16,17). Concentrations near EtO-emitting sources are higher than background levels (17); therefore, we constructed residence-based exposure metrics reflecting a distance-based decline in ambient concentrations. We assessed exposures during the 9-year period before enrollment (1987-1995), assuming participants lived at their enrollment address in this period based on residence histories in a subset of California participants [median 13 years prior to study start (18); Supplementary Methods, available online]. First, we linked enrollment addresses to all facilities reporting EtO releases during the exposure period (n = 118). We then classified participants as residing within 3, 5, and 10 km of at least 1 of these facilities in nonmutually exclusive categories (simple proximity metric). We calculated an inverse distance-weighted annual average emissions index within each distance threshold by dividing the yearly emissions per facility by the square of the distance between the facility and the residence. We adjusted both metrics for prevailing wind direction (Supplementary Methods and Supplementary Figure 1, available online) and averaged the year-specific wind-adjusted average emissions indexes (AEIs) across the exposure period.

Statistical analyses

We used Cox proportional hazards models scaled on follow-up time to estimate hazard ratios (HRs) and 95% confidence intervals (CIs) of the relationship between EtO exposure and risk of breast cancer in postmenopausal women (n = 173 648) and NHL in the full cohort (n = 451 876). We evaluated relationships overall, and for breast cancer, by intraductal vs invasive disease and estrogen receptor (ER) status (ER positive and negative [ER+; ER-) irrespective of progesterone receptor status, and for NHL by histologic type (major subtypes, n cases = 5016), sex, and smoking status. Associations were estimated for the simple proximity metric and, for the AEI metric, in quartiles or at the median (< and ≥) among the exposed; reference groups were those with no exposure within the distance threshold. We treated the median values of exposure categories as a continuous variable and evaluated Wald tests for linear trend (P_trend). We evaluated interactions by sex and smoking with Wald tests (P_interaction) and heterogeneity in hazard ratios by intraductal vs invasive disease and ER+ and ER- status with joint Cox models (P_{heterogeneity}) (19). We evaluated the proportional hazards assumption with exposure-time interaction terms. The threshold for statistical significance in 2-sided tests was α = .05.

We ran age-adjusted models and evaluated potential confounders based on the literature (20,21) using a 10% change in hazard ratios from stepwise selection. Final, fully adjusted multivariable models included age, state of residence, race and ethnicity (Hispanic, non-Hispanic Black, non-Hispanic White, Other [American Indian or Alaskan Native, Asian, and Pacific Islander], Unknown), smoking status (never, former, current, unknown), and sex (NHL models only). Additional covariates evaluated included educational attainment, body mass index, physical activity, alcohol use, and a census tract–level deprivation index (22). For breast cancer models, we also considered age of menarche and menopause, marital status, age at first child’s birth, oral contraceptive use, menopausal hormonal therapy use, number of breast biopsies, having a first-degree female relative with breast cancer, and any mammograms in the previous 3 years. None of these adjustments resulted in a more than 10% change in risk estimates (most were approximately 1%-3%). Most (88%) women reported a mammogram, and we conducted sensitivity analyses restricted to these women.

We also evaluated confounding by other environmental risk factors for both cancers, including polychlorinated dibenzo-p-dioxins and dibenzofurans (23) and several hazardous air pollutants (9) (Supplementary Methods, available online). Correlations with EtO exposures were weak (Supplementary Table 1, available online), and none of these pollutants were retained in final models.

Results

Breast cancer analyses

The breast cancer analyses included 2% Hispanic women, 5.9% non-Hispanic Black, 89% non-Hispanic White, 1.6% Other, and 1.5% with unknown race and ethnicity (Table 1). Although only 13% of the women lived in New Jersey, they represented more than 30% of those within 3 km of an EtO facility. The opposite was observed in Florida, where the proportion highly exposed was smaller (5.9%) than the residents (21.5%). The proportion of White women closest to EtO sources was similar to the proportion living more than 10 km away (88% vs 90%), whereas Hispanic women were more likely to live closest to sources (3.5% vs 1.8%), and Black women were more often within 3 to no farther than 5 km (6.3%) and 5 to no farther than 10 km (12.1%) vs farther than 10 km (5%). The proportion (42%) of individuals with a high school education or less was highest among those no farther than 3 km away. There were few other notable exposure patterns across sociodemographic or reproductive risk factors except for current menopausal hormonal therapy use, which was lowest (34.9%) for those in closest proximity to sources. The greatest proportion (88%) of women reporting a mammogram lived more than 10 km away from a source.

Table 1.

Demographic and other baseline characteristics of postmenopausal female NIH-AARP Diet and Health Study participants included in analyses of ethylene oxide and breast cancer risk, by residential distance to any ethylene oxide-emitting facility

Characteristicsa	All participants	≤3 km	3 to ≤5 km	5 to ≤10 km	>10 km
	(n = 173 648)	(n = 3555)	(n = 6362)	(n = 19 747)	(n = 143 984)
State, %
California	32.7	26.1	29.8	32.1	33.1
Florida	21.5	5.9	5.1	5.1	24.8
Georgia	3.0	1.0	1.7	2.9	3.1
Louisiana	3.7	3.9	4.1	5.0	3.5
Michigan	5.5	7.4	6.6	7.7	5.1
North Carolina	7.0	2.7	3.2	3.8	7.7
New Jersey	12.6	34.1	35.7	29.9	8.7
Pennsylvania	14.0	19.0	13.7	13.6	14.0
Urban residence, %	87.8	97.8	97.6	97.7	85.7
Age, mean (SD)	62.2 (5.2)	62.0 (5.2)	62.3 (5.2)	62.1 (5.2)	62.2 (5.2)
BMI, mean (SD)	26.8 (6.0)	27.8 (6.5)	27.3 (6.0)	27.3 (6.2)	26.7 (6.0)
Race and ethnicity, %
Hispanic	2.0	3.5	3.2	2.7	1.8
Non-Hispanic Black	5.9	4.9	6.3	12.1	5.0
Non-Hispanic White	89.0	88.0	86.9	81.2	90.2
Otherb	1.6	1.6	1.9	2.1	1.5
Unknown	1.5	1.9	1.7	2.0	1.5
Smoking status, %
Never	43.8	44.9	45.4	44.6	43.6
Former	38.7	36.5	36.8	37.5	39.0
Current	14.4	15.2	14.8	14.5	14.3
Unknown	3.1	3.5	3.0	3.4	3.1
Highest schooling achieved, %
High school or less	31.8	42.0	40.0	34.5	30.8
Post-high school	10.6	11.1	10.0	9.6	10.8
Some college	24.8	21.9	22.3	24.3	25.0
College and postgraduate	29.5	21.6	24.4	28.1	30.1
Unknown	3.3	3.4	3.4	3.5	3.3
First-degree relative with breast cancer, %	12.4	11.9	11.1	12.2	12.4
Unknown	4.6	5.6	5.2	5.0	4.5
Mammogram in past 3 years, %c	87.9	83.8	85.0	87.2	88.2
Age of menarche, %, y
Younger than 10	6.8	7.5	7.2	7.1	6.7
11-12	42.0	41.3	42.0	42.0	42.0
13-14	41.5	41.5	41.3	41.0	41.6
15 and older	9.3	9.3	9.1	9.6	9.3
Unknown	0.4	0.4	0.4	0.4	0.4
Menopausal hormone therapy use, %
Never	45.9	56.9	55.5	52.5	44.3
Current	44.6	34.9	36.0	38.9	46.0
Former	9.2	7.9	8.2	8.3	9.4
Unknown	0.3	0.3	0.2	0.2	0.3
Years using replacement hormones, %, y
Never	44.1	55.1	53.2	50.4	42.5
<5	19.1	16.5	16.9	18.3	19.3
5-9	13.6	10.8	10.9	11.7	14.1
≥10	21.3	15.6	16.5	17.3	22.1
Unknown	2.0	2.1	2.4	2.3	2.0
Age at menopause, %, y
Younger than 45	33.9	32.5	31.0	32.8	34.2
45-49	24.9	25.6	25.3	25.2	24.9
50-54	32.2	33.1	34.5	33.2	32.0
55 and older	7.1	7.6	7.3	7.1	7.1
Still menstruating	1.3	0.6	1.3	1.1	1.3
Unknown	0.6	0.7	0.7	0.6	0.6
Oral contraceptive use, %
Never (or <1 y)	60.6	65.7	65.1	63.4	59.9
1-4 y	17.1	14.9	15.9	16.4	17.3
≥5 y	21.4	18.7	18.0	19.3	21.9
Unknown	0.9	0.7	1.1	1.0	0.9
Age at first live birth, %, y
Never gave birth	14.2	14.9	15.3	15.1	14.1
Younger than 20	17.2	17.9	16.2	16.4	17.4
20-24	43.7	43.8	42.8	42.1	44.0
25 to 29	18.0	16.9	18.6	18.9	17.8
30	5.9	5.7	6.3	6.5	5.8
Unknown	1.0	0.8	0.9	1.0	1.0

^aUnless otherwise noted, variables were derived from the baseline questionnaire. BMI = body mass index; NIH-AARP = National Institutes of Health–AARP.

^bIncludes American Indian and Alaskan Native, Asian, and Pacific Islander.

^cInformation collected from a separate risk factor questionnaire conducted in 1996-1997 (n = 111 531 women).

In fully adjusted models, we observed a statistically significantly increased risk for breast cancer for the simple proximity metric within 10 km (HR = 1.05, 95% CI = 1.00 to 1.10); risk was not statistically significantly elevated at 3 or 5 km (Table 2). The AEI metrics showed a nonmonotonic pattern of nonstatistically significantly increased risk in the top exposure quartile that was strongest at 3 km (HR_{AEI3km, Q4vs0} = 1.14, 95% CI = 0.91 to 1.45; P_trend = .28) and weakened with increasing distance from the home (HR_{AEI5km, Q4vs0} = 1.09, 95% CI = 0.95 to 1.26; P_trend = .24; and HR_{AEI10km, Q4vs0} = 1.06, 95% CI = 0.97 to 1.15; P_trend = .22).

Table 2.

Hazard ratios (HRs) and 95% confidence intervals (CIs) for wind-adjusted ethylene oxide exposures and risk of breast cancer among women in the NIH-AARP Diet and Health Study (n = 173 648)

Exposure metric	Age adjusted		Fully adjusteda
	No. cases	HR (95% CI)	HR (95% CI)
Simple proximity metric, kmb
≤3	242	0.96 (0.85 to 1.10)	0.98 (0.86 to 1.11)
≤5	717	1.03 (0.95 to 1.11)	1.04 (0.96 to 1.12)
≤10	2139	1.03 (0.98 to 1.08)	1.05 (1.00 to 1.10)
AEI ≤3 kmc
No emissions	11 980	1.0 (Referent)	1.0 (Referent)
Q1	59	0.96 (0.74 to 1.24)	0.97 (0.75 to 1.26)
Q2	52	0.81 (0.62 to 1.07)	0.83 (0.63 to 1.09)
Q3	60	0.96 (0.75 to 1.24)	0.96 (0.75 to 1.24)
Q4	71	1.12 (0.89 to 1.42)	1.14 (0.91 to 1.45)
Ptrend		.36	.28
AEI ≤5 kmc
No emissions	11 505	1.0 (Referent)	1.0 (Referent)
Q1	176	1.01 (0.87 to 1.18)	1.02 (0.88 to 1.18)
Q2	181	1.04 (0.90 to 1.21)	1.06 (0.91 to 1.23)
Q3	172	0.98 (0.84 to 1.14)	0.99 (0.85 to 1.15)
Q4	188	1.08 (0.93 to 1.24)	1.09 (0.95 to 1.26)
Ptrend		.35	.24
AEI ≤10 kmc
No emissions	10 083	1.0 (Referent)	1.0 (Referent)
Q1	537	1.03 (0.94 to 1.12)	1.04 (0.95 to 1.13)
Q2	517	0.99 (0.91 to 1.09)	1.01 (0.93 to 1.11)
Q3	546	1.05 (0.96 to 1.14)	1.08 (0.98 to 1.17)
Q4	539	1.04 (0.95 to 1.13)	1.06 (0.97 to 1.15)
Ptrend		.37	.22

^aAdjusted for state of residence, age, race and ethnicity, and smoking status. AEI = average emissions index; EtO = ethylene oxide; NIH-AARP = National Institutes of Health–AARP; Q = quartile.

^bWind direction–adjusted distance to nearest facility. Referent group is individuals with no EtO emissions within this distance

^cAverage emissions index—calculated as the sum of inverse distance squared- and wind-weighted Toxics Release Inventory–reported emissions of EtO 1987-1995 for all facilities within distance threshold (∑[lbs EtO/km²]).

Analyses comparing intraductal and invasive disease showed a significant association with the 10 km simple proximity metric for DCIS (HR = 1.13, 95% CI = 1.01 to 1.27) but not for invasive cancer (HR = 1.03, 95% CI = 0.97 to 1.09; P_{heterogeneity} = .04; Table 3). Likewise, an association with the 3 km AEI was observed for DCIS (HR_Q4vs0 = 1.60, 95% CI = 0.98 to 2.61; P_trend = .08) but not invasive tumors (HR_Q4vs0 = 1.06, 95% CI = 0.81 to 1.38; P_trend = .70), although this difference was not significant (P_{heterogeneity} = .53). We saw a similar pattern of elevated risk in the highest AEI quartile for DCIS but not invasive cancer at 5 km (HR_Q4vs0 = 1.28, 95% CI = 0.92 to 1.79 vs HR_Q4vs0 = 1.04, 95% CI = 0.88 to 1.22; P_{heterogeneity} = .39) and 10 km (HR_Q4vs0 = 1.25, 95% CI = 1.02 to 1.53 vs HR _Q4vs0 = 1.02, 95% CI = 0.93 to 1.13; P_{heterogeneity} = .03). This pattern persisted among women reporting a mammogram in the previous 3 years (Supplementary Table 2, available online). We saw no clear differences in associations with AEIs for ER+ vs ER- breast cancers (P_{heterogeneity} = .23-.44; Supplementary Table 3, available online).

Table 3.

Hazard ratios (HRs) and 95% confidence intervals (CIs) for wind adjusted–ethylene oxide exposures and risk of breast cancer among women in the NIH-AARP Diet and Health Study, by intraductal and invasive disease (n = 173 648)

Exposure metric	DCIS		Invasive		P heterogeneity d
	No. cases	HR (95% CI)a	No. cases	HR (95% CI)a
Simple proximity metric, kmb
≤3	39	0.96 (0.70 to 1.32)	196	0.99 (0.85 to 1.14)	.99
≤5	111	0.98 (0.81 to 1.19)	584	1.06 (0.97 to 1.15)	.65
≤10	376	1.13 (1.01 to 1.27)	1697	1.03 (0.97 to 1.09)	.04
AEI ≤3 kmc
No emissions	1935	1.0 (Referent)	9722	1.0 (Referent)
Q1	7	0.70 (0.33 to 1.48)	49	1.01 (0.76 to 1.33)
Q2	8	0.76 (0.38 to 1.53)	44	0.88 (0.65 to 1.18)
Q3	8	0.79 (0.39 to 1.58)	50	1.00 (0.76 to 1.32)
Q4	16	1.60 (0.98 to 2.61)	53	1.06 (0.81 to 1.38)
Ptrend		.08		.70	.53
AEI ≤5 kmc
No emissions	1863	1.0 (Referent)	9334	1.0 (Referent)
Q1	29	1.03 (0.71 to 1.49)	142	1.02 (0.86 to 1.21)
Q2	24	0.84 (0.56 to 1.26)	151	1.10 (0.94 to 1.30)
Q3	22	0.77 (0.50 to 1.17)	147	1.06 (0.90 to 1.25)
Q4	36	1.28 (0.92 to 1.79)	144	1.04 (0.88 to 1.22)
Ptrend		.18		.62	.39
AEI ≤10 kmc
No emissions	1598	1.0 (Referent)	8221	1.0 (Referent)
Q1	100	1.18 (0.96 to 1.45)	418	1.00 (0.91 to 1.11)
Q2	95	1.14 (0.92 to 1.41)	401	0.98 (0.89 to 1.09)
Q3	78	0.95 (0.75 to 1.20)	456	1.11 (1.01 to 1.23)
Q4	103	1.25 (1.02 to 1.53)	422	1.02 (0.93 to 1.13)
Ptrend		.06		.55	.03

^aAdjusted for state of residence, age, race and ethnicity, and smoking status. AEI = average emissions index; DCIS = ductal carcinoma in situ; EtO = ethylene oxide; NIH-AARP = National Institutes of Health–AARP; Q = quartile.

^bWind direction–adjusted distance to nearest facility. Referent group is individuals with no EtO emissions within this distance.

^cAverage emissions index—calculated as the sum of inverse distance squared- and wind-weighted Toxics Release Inventory–reported emissions of EtO 1987-1995 for all facilities within distance threshold (∑[lbs EtO/km²]).

^dFrom the test for heterogeneity in hazard ratios in quartile analyses.

NHL analyses

Patterns of EtO exposure across population characteristics in the NHL analyses were similar to the breast cancer analyses (Supplementary Table 4, available online). In fully adjusted models, we observed a weak association for NHL overall with the simple proximity metric only at the 10 km distance (HR = 1.07, 95% CI = 1.00 to 1.14; Table 4). Compared with participants with no exposure, positive associations with AEIs at all 3 distances were generally non-statistically significant and lacked monotonic trend. We found no evidence of multiplicative interaction with smoking status (P_interaction = .47, .19, .46 for 3, 5, and 10 km AEIs, respectively). Sex-specific analyses showed inconsistent elevations in risk, although some point estimates were higher for women, and the interaction was statistically significant for the simple proximity metric at 10 km (P_interaction = .01) and the 5 and 10 km AEI (P_interaction = .03 and .04, respectively; Supplementary Table 5, available online).

Table 4.

Hazard ratios (HR) and 95% confidence intervals (CI) for wind-adjusted ethylene oxide exposures and risk of non-Hodgkin lymphoma in the NIH-AARP Diet and Health Study (n = 451 876)

Exposure metric	No. cases	Age adjusted	Fully adjusteda
		HR (95% CI)	HR (95% CI)
Simple proximity metric, kmb
≤3	125	0.97 (0.81 to 1.16)	0.95 (0.79 to 1.13)
≤5	374	1.04 (0.94 to 1.16)	1.02 (0.92 to 1.13)
≤10	1138	1.08 (1.02 to 1.15)	1.07 (1.00 to 1.14)
AEI ≤3 kmc
No emissions	6359	1.0 (Referent)	1.0 (Referent)
Q1	29	0.91 (0.63 to 1.31)	0.90 (0.62 to 1.30)
Q2	39	1.19 (0.87 to 1.63)	1.16 (0.84 to 1.59)
Q3	32	1.00 (0.71 to 1.42)	0.97 (0.68 to 1.37)
Q4	25	0.77 (0.52 to 1.14)	0.76 (0.52 to 1.13)
Ptrend		.20	.18
AEI ≤5 kmc
No emissions	6110	1.0 (Referent)	1.0 (Referent)
Q1	94	1.03 (0.84 to 1.27)	1.02 (0.83 to 1.25)
Q2	88	0.99 (0.80 to 1.22)	0.96 (0.78 to 1.19)
Q3	108	1.21 (1.00 to 1.46)	1.18 (0.97 to 1.42)
Q4	84	0.93 (0.75 to 1.16)	0.92 (0.75 to 1.15)
Ptrend		.69	.60
AEI ≤10 kmc
No emissions	5346	1.0 (Referent)	1.0 (Referent)
Q1	283	1.07 (0.95 to 1.20)	1.06 (0.94 to 1.20)
Q2	293	1.12 (0.99 to 1.26)	1.11 (0.98 to 1.25)
Q3	302	1.16 (1.03 to 1.30)	1.13 (1.01 to 1.28)
Q4	260	0.99 (0.88 to 1.13)	0.98 (0.86 to 1.11)
Ptrend		.96	.75

^aAdjusted for state of residence, age, race and ethnicity, and smoking status. AEI = average emissions index; EtO = ethylene oxide; NIH-AARP = National Institutes of Health–AARP; Q = quartile.

^bWind direction–adjusted distance to nearest facility. Referent group is individuals with no EtO emissions within this distance.

^cAverage emissions index—calculated as the sum of inverse distance squared- and wind-weighted Toxics Release Inventory–reported emissions of EtO 1987-1995 for all facilities within distance threshold (∑[lbs EtO/km²]).

Analyses of major histologic subtypes showed a nonstatistically significantly increased risk for CLL/SLL in association with the 10 km simple proximity metric (HR = 1.11, 95% CI = 0.97 to 1.27); we observed some positive, but mostly not statistically significant, associations with 10 km AEI quartiles 1-3 (Table 5). An association with follicular lymphoma was observed with the 5 km (HR = 1.30, 95% CI = 0.98 to 1.71) and 10 km (HR = 1.24, 95% CI = 1.03 to 1.49) simple proximity metrics. An increased risk for this subtype was also suggested across the 10 km AEI quartiles, but none of these associations reached statistical significance, and there was no exposure response (P_trend = .25).

Table 5.

Hazard ratios (HRs) and 95% confidence intervals (CIs) for wind-adjusted ethylene oxide exposures and risk of non-Hodgkin lymphoma in the NIH-AARP Diet and Health Study, by major histologic subtype (n = 451 876)

Exposure metric	Chronic lymphocytic leukemia and small lymphocytic lymphoma		Diffuse large B-cell lymphoma		Follicular lymphoma		Multiple myeloma
	No. cases	HR (95% CI)a	No. cases	HR (95% CI)a	No. cases	HR (95% CI)a	No. cases	HR (95% CI)a
Simple proximity metric, kmb
≤3	31	0.95 (0.66 to 1.35)	29	1.08 (0.75 to 1.57)	12	0.73 (0.41 to 1.30)	21	0.82 (0.53 to 1.26)
≤5	81	0.88 (0.70 to 1.10)	73	0.97 (0.77 to 1.24)	57	1.30 (0.98 to 1.71)	70	0.96 (0.75 to 1.23)
≤10	289	1.11 (0.97 to 1.27)	203	0.91 (0.78 to 1.06)	153	1.24 (1.03 to 1.49)	229	1.05 (0.90 to 1.22)
AEI ≤3 kmc
No emissions	1615	1.0 (Referent)	1268	1.0 (Referent)			1264	1.0 (Referent)
Q1; <median	6	0.77 (0.35 to 1.72)	17	1.28 (0.79 to 2.07)	—d	—	8	0.61 (0.30 to 1.22)
Q2; ≥median	14	1.67 (0.98 to 2.83)	12	0.89 (0.50 to 1.58)	—	—	13	1.04 (0.60 to 1.79)
Q3	5	0.60 (0.25 to 1.45)
Q4	6	0.73 (0.33 to 1.62)
Ptrend		.38		.69		—		.90
AEI ≤5 kmc
No emissions	1565	1.0 (Referent)	1224	1.0 (Referent)	731	1.0 (Referent)	1215	1.0 (Referent)
Q1; <median	22	0.96 (0.63 to 1.47)	18	0.96 (0.60 to 1.53)	33	1.50 (1.05 to 2.14)	16	0.85 (0.52 to 1.40)
Q2; ≥median	23	1.00 (0.66 to 1.51)	19	1.03 (0.65 to 1.62)	24	1.10 (0.73 to 1.65)	17	0.94 (0.58 to 1.53)
Q3	21	0.90 (0.59 to 1.39)	22	1.17 (0.77 to 1.79)			19	1.06 (0.67 to 1.67)
Q4	15	0.65 (0.39 to 1.09)	14	0.75 (0.44 to 1.26)			18	1.01 (0.63 to 1.61)
Ptrend		.09		.32		.71		.94
AEI ≤10 kmc
No emissions	1357	1.0 (Referent)	1094	1.0 (Referent)	635	1.0 (Referent)	1056	1.0 (Referent)
Q1	81	1.25 (1.00 to 1.57)	36	0.66 (0.47 to 0.93)	39	1.30 (0.94 to 1.81)	53	0.92 (0.69 to 1.22)
Q2	74	1.14 (0.90 to 1.45)	56	1.01 (0.77 to 1.33)	34	1.09 (0.77 to 1.55)	71	1.32 (1.03 to 1.70)
Q3	80	1.22 (0.97 to 1.54)	68	1.22 (0.95 to 1.56)	42	1.34 (0.97 to 1.85)	52	0.97 (0.73 to 1.29)
Q4	54	0.83 (0.63 to 1.09)	43	0.76 (0.56 to 1.03)	38	1.22 (0.88 to 1.71)	53	1.00 (0.75 to 1.32)
Ptrend		.16		.13		.25		.90

^aAdjusted for state of residence, age, race and ethnicity, smoking status, and sex. AEI = average emissions index; EtO = ethylene oxide; NIH-AARP = National Institutes of Health–AARP; Q = quartile.

^bWind direction–adjusted distance to nearest facility. Referent group is individuals with no EtO emissions within this distance.

^cAverage emissions index—calculated as the sum of inverse distance squared- and wind-weighted Toxics Release Inventory–reported emissions of EtO 1987-1995 for all facilities within distance threshold (∑[lbs EtO/km²]).

^d“—” signifies fewer than 5 cases per exposure category; hazard ratio not estimated.

Discussion

In this first US analysis of environmental EtO emissions using point-level exposure data, we observed an increased risk of breast cancer in the highest category of exposure to distance-weighted emissions that appeared to be driven by in situ rather than invasive cancers. The strength of these associations declined modestly with increasing distance between the participant and EtO sources, consistent with a presumed distance-based reduction in ambient levels. We found no differences in these associations by ER status and limited evidence of a relationship with NHL risk.

Our finding of an increased risk of intraductal breast cancer associated with EtO is the first such observation in a prospective cohort. One prior analysis of environmental EtO exposure was conducted in the Nurses’ Health Study II (NHS II) cohort, which yielded a nonstatistically significant association (n = 833 cases) with invasive disease of similar magnitude to our overall breast cancer association (HR_Q4vsQ1 = 1.04, 95% CI = 0.94 to 1.15) (9). Their analysis did not evaluate DCIS and included mostly (72%) premenopausal women, whereas female AARP cohort participants are predominantly postmenopausal. The NHS II exposure assessment was also limited to census tract–level estimates of EtO concentrations for a single year more than 10 years after enrollment. Other investigations come from occupational settings, the largest of which was conducted among women in a NIOSH cohort of more than 18 000 workers at 14 industrial facilities using EtO (24) and a cohort of hospital-based sterilization workers in the United Kingdom (25,26). These studies did not report excess breast cancer mortality. However, extended follow-up in the NIOSH cohort found breast cancer mortality associated with a 20-year cumulative exposure lag (8), and in a separate nested case-control analysis, risk of incident breast cancer was greatest in the highest quintile of 15-year lagged exposure (n = 230 cases; odds ratio = 1.87, 95% CI = 1.12 to 3.10) (7). These risks increased modestly when in situ cases (n = 20) were excluded, in contrast to our stronger associations for DCIS. The 2005 EPA-estimated average background concentration of ambient EtO in the United States was 0.0065 µg/m³ when including near sources, considerably lower than the Occupational Safety and Health Administration airborne permissible exposure limit (1 ppm averaged over an 8-hour shift or 5 ppm per 15-minute period) (27). Given that exposure levels from environmental pathways are likely much lower than in the workplace, directly comparing the magnitude of our observed associations with those in occupational settings may not be appropriate. However, the increased breast cancer risk pattern observed among the most highly exposed was consistent with the NIOSH study (7,8).

We evaluated the EtO–breast cancer relationship by disease characteristics because differences in etiologic pathways underlying tumor and disease types remain poorly understood. DCIS is a nonobligate precursor to invasive breast cancer; approximately 40% of women with DCIS will go on to develop invasive disease (28), and these cancers share many nonenvironmental risk factors (29). Because they comprise just 20% of new breast cancers, there are relatively few studies of environmental exposures and in situ disease. Our evaluation of invasive and pre-invasive forms may indicate the potential role of EtO in breast cancer initiation and/or progression, with stronger associations for DCIS providing some evidence supporting the former. We evaluated whether factors related to cancer screening and earlier diagnosis (21) may have confounded these associations; results were consistent when accounting for mammography, and our models included adjustment for race and ethnicity because non-White women tend to be diagnosed with breast cancer at later stages (30). We found little impact from adjustment for reproductive risk factors, family history of breast cancer, educational attainment, and neighborhood deprivation. Associations with hormone receptor–defined breast tumor subtypes were also of interest, as some exogenous compounds have shown an affinity for the ER (31). Experimental evidence that EtO has endocrine-disrupting properties is mixed (32), and although genotoxic and mutagenic mechanisms have been clearly established, little data support other modes of action for EtO-induced carcinogenicity (5). We found no differential associations in breast cancer risk by ER status, similar to the NHS II (9).

We observed no consistent pattern of increased NHL risk overall, only sporadic and mostly non-significant elevated risks, and no trend with increasing exposure. In subtype analyses, we saw some suggestion of increased risk for follicular lymphoma and CLL/SLL. A spatial analysis in Georgia found that mean distance to 7 EtO-emitting TRI sites was associated with risk of DLBCL with no differences by sex (10), but other lymphoma types were not evaluated, risk estimates were not adjusted for individual-level covariates, and analyses did not incorporate quantitative emissions data. In our study, associations for DLBCL lacked clear patterns. The NIOSH cohort found cumulative exposure associated with increased mortality for lymphoid malignancies (6,24) and, in extended follow-up, found excess mortality from lymphoid cell tumors specifically (including NHL, myeloma, and CLL), but only among men (8). In contrast, we observed an increased risk of NHL associated with the 10 km simple proximity and AEI metrics among women, but not men. Several other occupational cohorts had mixed findings for lymphoid tumors, but most had small numbers of exposed cases (3). The NIOSH analysis included more than twice as many exposed cases than our study, and lymphohematopoietic cancer classifications have changed over time (3). Lower exposure levels and small numbers of exposed cases may also explain the absence of an observed association with NHL in our data.

Strengths of this work include the prospective design and large study population, which afforded large numbers of cancer cases to evaluate heterogeneity in relationships by disease characteristics. The catchment area is geographically spread, includes a large number of EtO point sources, and all are states ranked within the top 20 for population-weighted average ambient EtO concentrations (33). We demonstrated no confounding by known environmental risk factors for these cancers.

Because EtO concentrations are highest close to sources, our proximity-based metrics likely reflect residential exposure better than available alternatives, like EPA’s National Air Toxics Assessment estimates of census tract–level concentrations (33). Although they incorporate point and nonpoint sources, National Air Toxics Assessment data are semi-annual and are not recommended for risk analyses at small spatial scales, to compare risks between states, or to compare levels over time (34). We acknowledge some limitations with TRI, including that facilities with lower emissions are not included and the data are self-reported. The long-term averages in our quantitative metrics may not reflect cumulative exposures. However, we expect they capture the contrast in exposures among participants residing within proximity of industrial EtO sources and that exposure misclassification is likely nondifferential with respect to these cancers.

Although our analyses attempted to account for the long latent period for the development of these cancers, the latency following EtO exposure is unknown. Lacking occupational information, we could not assess whether participants living close to sources were more likely to work at these facilities. Despite the cohort’s large size, exposure prevalence was low and resulted in less-than-optimal statistical power, particularly for analyses by disease subtype. Therefore, we cannot discount the possibility that observed associations are due to chance.

Our results suggest that airborne EtO emissions near the home may be associated with risk of breast cancer in situ. We did not find consistent evidence of increased NHL risk. Replication of these novel findings will be an important next step.

Funding

This work was supported by the Intramural Research Program of the National Cancer Institute.

Notes

Role of the funder: The views expressed by the authors are their own and do not necessarily represent the views of the U.S. Department of Health and Human Services, the National Institutes of Health, or the National Cancer Institute.

Author disclosures: All authors report no potential conflicts of interest.

Author contributions: Rena R. Jones (Conceptualization, Data curation, Methodology, Project administration, Writing – original draft, Writing – review & editing); Jared A. Fisher (Conceptualization, Data curation, Formal analysis, Writing – review & editing); Danielle N. Medgyesi (Conceptualization, Data curation, Formal analysis, Writing – review & editing); Ian D. Buller (Conceptualization, Formal analysis, Writing – review & editing); Linda M. Liao (Data curation, Writing – review & editing); Gretchen Gierach (Methodology, Writing – review & editing); Mary H. Ward (Conceptualization, Methodology, Writing – review & editing); Debra T. Silverman (Conceptualization, Methodology, Writing – review & editing).

Prior presentations: Preliminary findings from this work were presented at the International Society for Environmental Epidemiology conference in August 2021.

Data availability

Data described in the manuscript, code book, and analytic code will be made available upon request, pending approval from the NIH-AARP Diet and Health Study Steering Committee. Further details are provided at https://www.nihaarpstars.com/.