Outdoor air pollution exposure and uterine cancer incidence in the Sister Study

Abstract

Background

Outdoor air pollution is a ubiquitous exposure that includes endocrine-disrupting and carcinogenic compounds that may contribute to the risk of hormone-sensitive outcomes such as uterine cancer. However, there is limited evidence about the relationship between outdoor air pollution and uterine cancer incidence.

Methods

We investigated the associations of residential exposure to particulate matter less than 2.5 µm in aerodynamic diameter (PM_2.5) and nitrogen dioxide (NO₂) with uterine cancer among 33 417 Sister Study participants with an intact uterus at baseline (2003-2009). Annual average air pollutant concentrations were estimated at participants’ geocoded primary residential addresses using validated spatiotemporal models. Cox proportional hazards models were used to estimate hazard ratios and 95% confidence intervals for the association between time-varying 12-month PM_2.5 (µg/m³) and NO₂ (parts per billion; ppb) averages and uterine cancer incidence.

Results

Over a median follow-up period of 9.8 years, 319 incident uterine cancer cases were identified. A 5-ppb increase in NO₂ was associated with a 23% higher incidence of uterine cancer (hazard ratio = 1.23, 95% confidence interval = 1.04 to 1.46), especially among participants living in urban areas (hazard ratio = 1.53, 95% confidence interval = 1.13 to 2.07), but $$PM_2.5 was not associated with increased uterine cancer incidence.

Conclusion

In this large US cohort, NO₂, a marker of vehicular traffic exposure, was associated with a higher incidence of uterine cancer. These findings expand the scope of health effects associated with air pollution, supporting the need for policy and other interventions designed to reduce air pollutant exposure.

Uterine cancer is the most common and second-deadliest gynecologic cancer in the United States, with more than 67 000 new cases and 13 000 deaths expected in 2024 (1,2). Established risk factors for uterine cancer include obesity (3,4), unopposed estrogen hormone therapy (5-8), and genetic predisposition (eg, Lynch syndrome) (9-11). Many of these risk factors, such as obesity and extended exposure to estrogen, are consistent with an underlying hormonal etiology (5-8,12).

Outdoor air pollution contains a complex mixture of pollutants arising from numerous natural and human-made sources, such as industrial activity and transportation (13). The International Agency for Research on Cancer classified outdoor air pollution as a group 1 carcinogen (carcinogenic to humans) based on the evidence for fine particulate matter (airborne particles <2.5 µm in aerodynamic diameter [PM_2.5]) in lung cancer research (14-21). Epidemiologic evidence regarding how air pollution is related to other cancer types is limited, however (14). Several carcinogenic compounds in outdoor air pollution, such as heavy metals in particulate matter and polycyclic aromatic hydrocarbons, may also act as endocrine disruptors, which are hypothesized to contribute to hormone-mediated female health outcomes (22-24). For example, outdoor air pollution is associated with a higher risk of estrogen-linked health conditions, including breast and ovarian cancers (25-29) as well as uterine fibroids (30,31). Notably, nitrogen dioxide (NO₂), which may serve as a proxy for the near-road traffic–related air pollution mixture, has consistently been shown to have a positive association with breast cancer (26).

Few studies have investigated the relationship between outdoor air pollution and uterine cancer incidence (28,32). Our primary objective was to examine the association between uterine cancer incidence and residential exposure to outdoor air pollution—specifically, PM_2.5 and NO₂—in a large and diverse US-wide prospective cohort.

Methods

Study population and design

The Sister Study is a nationwide prospective cohort of 50 884 women across the United States, including Puerto Rico, recruited between 2003 and 2009 (33). Women were eligible if they were between 35 and 74 years of age, had at least 1 sister diagnosed with breast cancer, and had no history of breast cancer. At baseline, women completed a computer-assisted telephone interview; a home visit during which an examiner collected biological specimens and anthropometric measurements; and written questionnaires to collect information about demographics, residential characteristics, sociomedical history, and lifestyle factors. Participants or their next of kin were contacted to report any cancer or other health-related changes over the past year and complete detailed follow-up assessments every 2 to 3 years, including updates to their primary residential address. Responses rates have remained above 80% throughout follow-up (34). All participants provided signed informed consent, and the Sister Study was approved by the Institutional Review Board of the National Institutes of Health (NIH).

This study relies on Sister Study data release 10.1, which includes follow-up data through October 12, 2020. We excluded women who withdrew from the study (n = 5), had a prevalent or uncertain uterine cancer history (n = 448), had undergone prebaseline hysterectomy (n = 15 598), had no follow-up time (n = 183), or had missing data for key covariates (n = 576). Further, participants were excluded if they did not reside in the contiguous United States (n = 648) or their baseline address could not be geocoded (n = 9), resulting in 33 417 women eligible for our analysis.

Outcome assessment

Uterine cancer cases were defined as women who reported an endometrial cancer, uterine sarcoma, or other type of cancer in the uterus diagnosed after enrollment (n = 319). We included all cases that were diagnosed before June 30, 2017, which was considered the end of follow-up for this analysis due to air pollution data availability. Women with a self-reported uterine cancer diagnosis were asked to provide authorization for the retrieval of their pathology report. A total of 259 uterine cancer cases (81.2% of all cases) were confirmed using either pathology report (n = 243) or using death certificate or National Death Index Plus data, where uterine cancer was recorded as the primary or underlying cause of death (n = 16). For cases without pathology reports or death records, diagnostic information was obtained through self-report (n = 57) or next of kin (n = 3). Among those with pathology reports or death records, the positive predictive value of self-reported cases was 81.4%. Of the 259 medically confirmed uterine cancers, 244 (94.2%) were endometrial cancer using International Statistical Classification of Diseases, Tenth Revision code C54.1.

Exposure assessment

Annual average concentrations of PM_2.5 and NO₂ were estimated for each woman’s residence using a synthesized suite of regional spatiotemporal models derived from regulatory monitors from the Environmental Protection Agency (EPA) Air Quality System monitoring and interagency monitoring of protected visual environments network data, supplemental monitors at residential and community locations, and a large collection of geographic covariates developed for large-scale epidemiologic research (35). Briefly, the models used pollutant concentration data from more than 900 research monitors, including residential monitoring campaigns, fixed sites, and gradient monitoring near roadways (only for the NO₂ model); 1500 regulatory agency monitors across the United States; and more than 200 geographic covariates calculated at each monitoring location and each residential location (35,36).

Model performance was evaluated using cross-validation (PM_2.5R² = 0.89; NO₂R² = 0.87) (35). Our modeling approach accounts for complex spatiotemporal dependencies in land use in a spatial smoothing framework. Two-week average pollutant concentrations at each participant’s residential address were used to calculate 12-month average air pollutant concentrations across all primary addresses through June 30, 2017, when annual PM_2.5 estimates were last available for the cohort. Pollutants were modeled as time-varying 12-month averages, and person-months were included when at least 11 prior months of PM_2.5 and NO₂ data were available.

Covariate assessment

Covariate information was collected at the baseline interview and across 3 follow-up surveys. Baseline questionnaire information was used to ascertain demographic characteristics, including age and year at baseline enrollment (as integer values) as well as physical activity (metabolic equivalent task hours per week [continuous]). Self-reported race and ethnicity were categorized as African American and/or Black, including Hispanic and/or Latina Black, Hispanic and/or Latina non-Black, non-Hispanic White, and an “other” category that included Asian and/or Pacific Islander, American Indian and/or Alaska Native, or none of the included categories. Educational attainment (high school or less, some college, and college or above), annual household income (≤$49 999, $50 000-$99 999, ≥$100 000), and residence type (urban, suburban, rural/small town/other) were categorical variables. Examiner-measured height and weight at the baseline home visit were used to calculate body mass index (BMI; ≤24.9, 25-29.9, and ≥30 kg/m²).

Based on participants’ baseline geocoded address, we determined the baseline area deprivation index (ADI) and census region of residence. ADI is a composite measure of 17 census indicators related to poverty, education, housing, and employment obtained from the American Community Survey that were multiplied by previously published factor weights (37) and summed for each block group (38-40). ADI scores range from 0 to 100 percentiles, where higher percentiles correspond to higher neighborhood deprivation (37,39). We categorized ADI based on the quartile distribution within our cohort: ADI_Q1 (<12th percentile; least deprived), ADI_Q2 (12th-25th percentile), ADI_Q3 (26th-47th percentile), and ADI_Q4 (≥48th percentile; most deprived).

At each follow-up questionnaire, we collected information about census region of residence (Northeast, Midwest, South, and West); self-reported weight (continuous); smoking status (never, former, current); hormone therapy use (none, estrogen alone, and estrogen plus progestin); parity (nulliparous, 1-2 births, ≥3 births); menopausal status (premenopausal or postmenopausal); and oral contraceptive use (never or ever).

Statistical analysis

We used Cox proportional hazards models to estimate hazard ratios (HRs) and 95% confidence intervals (CIs) for the association between each air pollutant (time-varying) and incident uterine cancer. The time scale was calendar month, and person-months were accrued from study entry until uterine cancer diagnosis, with censoring at the time of hysterectomy, death, loss to follow-up, or administrative end of follow-up on June 30, 2017. Linearity of the log hazard ratio was assessed using likelihood ratio tests comparing the model with only the linear term to the model with linear and restricted cubic spline terms (41).

All models were stratified by age at baseline. Potential confounders were selected based on a directed acyclic graph and the literature (42,43). When available, we adjusted for time-varying covariates. The primary model represented the minimally sufficient adjustment set based on our directed acyclic graph (Supplementary Figure 1, available online) and included race and ethnicity, education, annual household income, ADI, and the other pollutant (eg, PM_2.5 was adjusted for NO₂ and vice versa). To address the possibility of residual confounding from other factors, we considered as a sensitivity analysis an additional model that further adjusted for urbanicity and physical activity, as measured at baseline, in addition to time-varying census region, BMI, smoking status, parity, hormone therapy use, and oral contraceptive use.

We conducted a sensitivity analysis estimating outcome-specific hazard ratios limited to medical record–confirmed uterine and endometrial cancer cases, censoring unconfirmed or unknown cancer types at time of diagnosis. We evaluated effect measure modification by race and ethnicity, baseline ADI quartile, baseline residence type, baseline BMI, and time-varying census region. Effect measure modification was evaluated using likelihood ratio tests comparing models with and without multiplicative interaction terms. The threshold for statistical significance testing in 2-sided tests was α = .05, determined using SAS, version 9.4, statistical software (SAS Institute Inc, Cary, NC).

Results

Over a median (standard deviation [SD]) follow-up period of 9.8 (2.6) years, 319 uterine cancer cases were reported. Characteristics of the study population overall and stratified by the median NO₂ (Table 1) and PM_2.5 (Supplementary Table 1, available online) at baseline are provided. Mean (SD) age at baseline was 54.1 (8.9) years, and mean (SD) BMI was 27.3 (6.1). The study population consisted of 7.8% Black and/or African American women, 3.2% Hispanic and/or Latina non-Black women, 86.5% non-Hispanic White women, and 2.5% women of other race and ethnicity. The largest proportion of women were college-educated (56%), lived in the South (31%), were postmenopausal (58%), and had previously used oral contraceptives (85%). Participants exposed to higher NO₂ concentrations were more likely to live in urban (30%) or suburban (55%) areas, whereas those exposed to lower NO₂ concentrations lived in rural/small towns/other areas (64%).

Table 1.

Sample enrollment characteristics at baseline, stratified by the median for nitrogen dioxide measurement at baseline, Sister Study (2003-2009)

Study participant characteristicsa	Eligible cohort (N = 33 417)	NO2 <8.5 ppb (n = 16 708)	NO2 ≥8.5 ppb (n = 16 709)
12-mo average NO2, median (IQR), ppb	8.5 (5.8-11.9)	—	—
12-mo average PM2.5, median (IQR), µg/m3	10.7 (8.7-12.4)	—	—
Age at baseline, mean (SD), y	54.1 (8.9)	54.5 (8.9)	53.8 (8.9)
Race and ethnicity, No. (%)
African American and/or Black, including Hispanic and/or Latina Black	2604 (7.8)	953 (5.7)	1651 (9.9)
Hispanic and/or Latina, non-Black	1082 (3.2)	379 (2.3)	703 (4.2)
Non-Hispanic White	28 900 (86.5)	14 965 (89.6)	13 935 (83.4)
Other	831 (2.5)	411 (2.5)	420 (2.5)
Educational attainment, No. (%)
High school or less	4389 (13.1)	2483 (14.9)	1906 (11.4)
Some college	10 375 (31.1)	5628 (33.7)	4747 (28.4)
College or above	18 653 (55.8)	8597 (51.4)	10 056 (60.2)
Annual household income, No. (%)
≤$49 999	7210 (21.6)	3696 (22.1)	3514 (21.0)
$50 000-$99 999	13 787 (41.3)	7190 (43.0)	6597 (39.5)
≥$100 000	12 420 (37.2)	5822 (34.9)	6598 (39.5)
National ADI percentile, mean (SD)	32.2 (24.1)	37.2 (24.0)	27.2 (23.2)
Quartile 1 (<12th percentile), less deprived	7899 (23.6)	2565 (15.4)	5334 (31.9)
Quartile 2 (12th to 25th percentile)	8389 (25.1)	3949 (23.6)	4440 (26.6)
Quartile 3 (26th to 47th percentile)	8533 (25.5)	4718 (28.2)	3815 (22.8)
Quartile 4 (≥48th percentile), most deprived	8596 (25.6)	5476 (32.8)	3120 (25.7)
Census region, No. (%)
Northeast	6567 (19.6)	2930 (17.5)	3637 (21.8)
Midwest	9233 (27.6)	5076 (30.4)	4157 (24.9)
South	10 406 (31.1)	6259 (37.5)	4147 (24.8)
West	7211 (21.6)	2443 (14.6)	4768 (28.5)
Residence type, No. (%)
Urban	6354 (19.0)	1399 (8.4)	4955 (29.7)
Suburban	13 816 (41.3)	4634 (27.7)	9182 (54.9)
Rural/small town/other	13 247 (39.6)	10 675 (63.9)	2572 (15.4)
BMI (kg/m2), mean (SD)	27.3 (6.1)	27.3 (6.1)	27.2 (6.2)
≤24.9	14 320 (42.8)	6991 (41.8)	7329 (43.9)
25-29.9	10 212 (30.6)	5217 (31.2)	4995 (29.9)
≥30	8885 (26.6)	4500 (26.9)	4385 (26.2)
Physical activity metabolic equivalent task h/wk, mean (SD)	50.8 (31.3)	52.8 (31.8)	48.7 (30.7)
Menopausal status, No. (%)
Premenopausal	14 047 (42.0)	6754 (40.4)	7293 (43.6)
Postmenopausal	19 370 (58.0)	9954 (59.6)	9416 (56.4)
Hormone replacement therapy use,b No. (%)
None	23 209 (69.4)	11 364 (68.0)	11 845 (70.9)
Estrogen alone	2296 (6.9)	1224 (7.3)	1072 (6.4)
Estrogen plus progestin	7912 (23.7)	4120 (24.7)	3792 (22.7)
Smoking status, No. (%)
Never	19 004 (56.9)	9588 (57.4)	9416 (56.3)
Former	11 794 (35.3)	5853 (35.0)	5941 (35.6)
Current	2619 (7.8)	1267 (7.6)	1352 (8.1)
Parity, No. (%)
Nulliparous	6660 (19.9)	2826 (16.9)	3834 (22.9)
1-2 births	17 320 (51.8)	8769 (52.5)	8551 (51.2)
≥3 births	9437 (28.2)	5113 (30.6)	4324 (25.9)
Oral contraceptive use, No. (%)
Never	4967 (14.9)	2264 (13.5)	2703 (16.2)
Ever	28 450 (85.1)	14 444 (86.5)	14 006 (83.8)

^aExcluded women who withdrew (n = 5), had a prevalent or uncertain uterine cancer history (n = 448), had had a prebaseline hysterectomy (n = 15 598), had no follow-up time (n = 183), had missing air pollutant data (n = 657), or had missing covariate information (race and ethnicity [n = 5], education [n = 3], ADI [n = 111], BMI [n = 97], hormone replacement therapy [n = 82], menopause [n = 3], physical activity [n = 250], smoking [n = 2], parity [n = 22], and oral contraceptive use [n = 1]). ADI = area deprivation index; BMI = body mass index; IQR = interquartile range; NO₂ = nitrogen dioxide; ppb = parts per billion; PM_2.5 = particulate matter <2.5 µm in diameter; SD = standard deviation.

^bWomen who ever reported using estrogen alone as hormone replacement therapy were categorized as “estrogen alone.”

A 5-parts-per-billion (ppb) increase in NO₂ was associated with a 23% higher incidence of uterine cancer (hazard ratio [HR] = 1.23, 95% confidence interval [CI] = 1.04 to 1.46) in the fully adjusted model. Associations were similar in the reduced model (HR = 1.18, 95% CI = 1.03 to 1.36); therefore, we report the fully adjusted findings throughout. Results were similar when the analysis was restricted to medically confirmed uterine (HR = 1.25, 95% CI = 1.04 to 1.50) and endometrial cancers (HR = 1.28, 95% CI = 1.06 to 1.54), but a 5 µg/m³ increase in PM_2.5 was not associated with increased incidence of uterine cancer (HR = 0.83, 95% CI = 0.54 to 1.27) or medically confirmed uterine cancer (HR = 0.86, 95% CI = 0.54 to 1.38) or endometrial cancer (HR = 0.89, 95% CI = 0.55 to 1.43) (Table 2). There was no evidence of a nonlinear association for either PM_2.5 or NO₂ (all P > .40).

Table 2.

Association between time-varying 12-month air pollution exposure and incident uterine cancer in Sister Study participants, 2003-2017

	Cases, No.	Person-years	Age- and time-adjusted HRa (95% CI)	Model 1 HRa,b (95% CI)	Model 2 HRa,c (95% CI)
PM2.5 (5 µg/m3 increase)
All uterine cancer cases	319	313 469	1.17 (0.87 to 1.58)	0.98 (0.70 to 1.36)	0.83 (0.54 to 1.27)
Confirmed uterine cancer cases	259	313 469	1.26 (0.90 to 1.75)	1.07 (0.74 to 1.54)	0.86 (0.54 to 1.38)
Confirmed endometrial cancer cases	244	313 469	1.27 (0.90 to 1.79)	1.09 (0.75 to 1.59)	0.89 (0.55 to 1.43)
NO2 (5-ppb increase)
All uterine cancer cases	319	313 469	1.18 (1.05 to 1.33)	1.18 (1.03 to 1.36)	1.23 (1.04 to 1.46)
Confirmed uterine cancer cases	259	313 469	1.19 (1.04 to 1.35)	1.19 (1.03 to 1.38)	1.25 (1.04 to 1.50)
Confirmed endometrial cancer cases	244	313 469	1.19 (1.04 to 1.36)	1.20 (1.03 to 1.39)	1.28 (1.06 to 1.54)

^aAll models were stratified by age at baseline and time-scaled using calendar month. CI = confidence interval; HR = hazard ratio; NO₂ = nitrogen dioxide; ppb = parts per billion; PM_2.5 = particulate matter <2.5 µm in diameter.

^bAdjusted for race and ethnicity (non-Hispanic White; African American and/or Black, including Hispanic and/or Latina Black, Hispanic and/or Latina non-Black, other), education (high school or less, some college, college or above), income (≤$49 999, $50 000-$99 999, ≥$100 000), baseline area deprivation index (continuous), and time-varying co-air pollutant (eg, PM_2.5 was adjusted for NO₂).

^cAdditionally adjusted for time-varying census region (Northeast, Midwest, South, West), baseline residence type (urban, suburban, rural/small town/other), time-varying body mass index (continuous), baseline physical activity (metabolic equivalent task hours per week; continuous), time-varying smoking status (never, former, current), time-varying menopause (pre- and postmenopausal), time-varying hormone replacement therapy (none, estrogen alone, estrogen plus progestin), and time-varying parity (nulliparous, 1-2 births, ≥3 births).

We observed a positive association among those who lived in urban areas at enrollment (HR = 1.53, 95% CI = 1.13 to 2.07) and negligible associations among women living in rural areas (HR = 0.92, 95% CI = 0.61 to 1.40) and suburban areas (HR = 1.06, 95% CI = 0.78 to 1.45), but there was no evidence of statistical interaction (P = .13 for heterogeneity). We observed evidence of modification by race and ethnicity (P = .01 for heterogeneity), but only 21 African American and/or Black women with uterine cancer were included, limiting inference and resulting in instability in the effect estimates and extremely wide confidence intervals after adjustment for confounders (Supplementary Table 2, available online). The associations with NO₂ did not vary by ADI (P = .19 for heterogeneity), region (P = .46 for heterogeneity), or BMI (P = .27 for heterogeneity) (Table 3). No effect modification was observed for PM_2.5 in stratified analyses.

Table 3.

Stratified adjusted hazard ratios of the association between time-varying 12-month air pollution exposure and uterine cancer risk using baseline covariate information from Sister Study participants, 2003-2017

			PM2.5 (5-µg/m3 increase)			NO2 (5-ppb increase)
Strataa	Cases, No.	Person-years	Age- and time-adjusted HRb (95% CI)	Model 1 HRb,c (95% CI)	Model 2 HRb,d (95% CI)	Age- and time-adjusted HRb (95% CI)	Model 1 HRb,c (95% CI)	Model 2 HRb,d (95% CI)
ADI
Quartile 1 (less deprived)	67	76 033	1.19 (0.63 to 2.26)	0.94 (0.45 to 1.92)	0.96 (0.38 to 2.41)	1.21 (0.95 to 1.53)	1.26 (0.95 to 1.67)	1.30 (0.90 to 1.86)
Quartile 2	76	79 669	1.11 (0.61 to 2.02)	0.88 (0.46 to 1.68)	0.97 (0.40 to 2.36)	1.39 (1.07 to 1.80)	1.42 (1.07 to 1.89)	1.37 (0.95 to 1.98)
Quartile 3	98	79 631	1.06 (0.58 to 1.93)	0.83 (0.43 to 1.61)	0.80 (0.35 to 1.84)	1.26 (0.99 to 1.61)	1.31 (0.99 to 1.73)	1.28 (0.91 to 1.80)
Quartile 4 (most deprived)	78	78 135	1.14 (0.59 to 2.19)	1.26 (0.60 to 2.65)	0.97 (0.40 to 2.36)	0.95 (0.70 to 1.28)	0.86 (0.60 to 1.23)	0.97 (0.62 to 1.51)
P for heterogeneity			.34	.38	.30	.25	.23	.19
Census region
Northeast	80	59 973	1.53 (0.81 to 2.90)	0.76 (0.28 to 2.07)	0.77 (0.27 to 2.23)	1.25 (1.06 to 1.48)	1.34 (1.02 to 1.76)	1.30 (0.96 to 1.76)
Midwest	88	85 443	1.08 (0.47 to 2.48)	1.21 (0.45 to 3.26)	1.14 (0.40 to 3.24)	0.96 (0.70 to 1.33)	0.97 (0.65 to 1.47)	0.96 (0.60 to 1.54)
South	92	99 295	1.98 (0.82 to 4.80)	1.66 (0.63 to 4.40)	1.48 (0.54 to 4.04)	1.37 (0.94 to 2.00)	1.17 (0.76 to 1.82)	1.38 (0.84 to 2.26)
West	59	68 758	0.86 (0.46 to 1.60)	0.74 (0.31 to 1.77)	0.77 (0.31 to 1.94)	1.02 (0.76 to 1.37)	1.08 (0.71 to 1.64)	1.08 (0.69 to 1.71)
P for heterogeneity			.74	.70	.70	.44	.43	.46
Residence type
Urban	66	59 264	1.29 (0.68 to 2.46)	0.76 (0.36 to 1.61)	0.53 (0.21 to 1.33)	1.32 (1.10 to 1.57)	1.41 (1.12 to 1.76)	1.53 (1.13 to 2.07)
Suburban	122	131 516	1.02 (0.59 to 1.75)	0.91 (0.50 to 1.64)	0.84 (0.39 to 1.78)	1.09 (0.86 to 1.39)	1.09 (0.84 to 1.41)	1.06 (0.78 to 1.45)
Rural/small town/other	131	122 689	1.15 (0.72 to 1.83)	1.11 (0.67 to 1.85)	1.05 (0.54 to 2.05)	0.98 (0.70 to 1.38)	1.01 (0.69 to 1.46)	0.92 (0.61 to 1.40)
P for heterogeneity			.68	.94	.94	.23	.20	.13
BMI
≤24.9 kg/m2	81	138 266	1.04 (0.59 to 1.82)	0.70 (0.37 to 1.32)	0.78 (0.35 to 1.72)	1.29 (1.06 to 1.58)	1.35 (1.07 to 1.71)	1.23 (0.91 to 1.67)
25-29.9 kg/m2	92	95 055	1.34 (0.76 to 2.38)	1.32 (0.70 to 2.48)	1.49 (0.66 to 3.36)	1.10 (0.87 to 1.40)	1.05 (0.79 to 1.39)	1.16 (0.81 to 1.64)
≥30 kg/m2	146	80 148	0.95 (0.59 to 1.55)	0.93 (0.55 to 1.59)	0.59 (0.30 to 1.16)	1.09 (0.89 to 1.35)	1.11 (0.88 to 1.41)	1.23 (0.91 to 1.65)
P for heterogeneity			.21	.24	.22	.21	.21	.27

^aStratified hazard ratios are presented for categories of baseline ADI, baseline residence type, and baseline BMI as well as time-varying census region. ADI = area deprivation index; BMI = body mass index; CI = confidence interval; HR = hazard ratio; NO₂ = nitrogen dioxide; PM_2.5 = particulate matter <2.5 µm in diameter; ppb = parts per billion.

^bAll models were stratified by age at baseline and time-scaled using calendar month.

^cAdjusted for race and ethnicity (non-Hispanic White; African American and/or Black, including Hispanic and/or Latina Black, Hispanic and/or Latina non-Black, other), education (high school or less, some college, college or above), income (≤$49 999, $50 000-$99 999, ≥$100 000), baseline ADI (continuous), and time-varying co-air pollutant (eg, PM_2.5 was adjusted for NO₂).

^dAdditionally adjusted for time-varying census region (Northeast, Midwest, South, West), baseline residence type (urban, suburban, rural/small town/other), time-varying BMI (continuous), baseline physical activity (metabolic equivalent task hours per week; continuous), time-varying smoking status (never, former, current), time-varying menopause (pre- and postmenopausal), time-varying hormone replacement therapy (none, estrogen alone, estrogen plus progestin), time-varying parity (nulliparous, 1-2 births, ≥3 births), and time-varying oral contraceptive use (never, ever).

Discussion

In this large, prospective, US-wide cohort study, we observed that women exposed to higher NO₂ concentrations had a greater incidence of uterine cancer, particularly among participants living in urban areas compared with those living in suburban and rural areas. PM_2.5 was not associated with a higher incidence of uterine cancer. Our findings highlight air pollution as a plausible risk factor for uterine cancer, particularly pollution arising from vehicular traffic emissions.

Experimental studies support the biological plausibility of the effect of air pollution on breast cancer, another hormone-driven outcome (44-49). Benzene, a non-methane volatile organic compound present in traffic exhaust, was observed in mammary tumors in mice models (49). Polycyclic aromatic hydrocarbons (PAHs) from traffic emissions can cause breast cancer through DNA damage (47), aberrant DNA methylation (48), and estrogenic and antiestrogenic activities that give rise to tumor development (46,50). These findings support a similar biological pathway for the effects of air pollution on other hormone-sensitive cancers, particularly uterine cancer (51-54). Our study expands on prior findings from epidemiologic studies on the role of air pollution—in particular, NO₂,—in the incidence of hormone-sensitive outcomes, notably breast cancer (23,26,28,55-62).

In our cohort, we observed that NO₂ was significantly associated with uterine cancer incidence. In a retrospective cohort study by Wei et al. (28) of 2010-2016 Medicare beneficiaries, no association was observed across all NO₂ exposure levels (0.0-127.6 ppb). There was some evidence of low dose effects of NO₂ exposure (<10 ppb) per 1-ppb increase, however, resulting in 141 additional (95% CI = 10 to 272) incident endometrial cancer diagnoses per year (28). Our study population was similar to the sample of Medicare beneficiaries with regard to the extensive follow-up time and calendar time period but was, on average, younger and not yet eligible for Medicare. The exposure range for NO₂ reported by Wei et al. (28) was higher than what we observed in the Sister Study (interquartile range [IQR] = 9.7-21.9 ppb) across follow-up. Despite a large sample size, the Medicare beneficiaries’ study was limited by using zip code–level measurements of air pollution, which may have substantial exposure misclassification compared with our time-varying estimates at the individual residence level over the follow-up period.

We observed a largely null association for PM_2.5, with wide confidence intervals, consistent with findings from Wei et al. (28). PM_2.5 arises from a variety of sources that vary geographically across the United States (63,64). Previous studies of breast cancer have highlighted that geographic region and PM_2.5 component mixtures may significantly modify associations (24,29,65). Therefore, our largely null findings may be due to our inability to estimate region-specific PM_2.5 associations or consider PM_2.5 component mixtures because of small case sample sizes.

Our findings are in contrast with an ecologic study that observed a positive correlation between PM_2.5 and uterine cancer incidence using Japan’s National Cancer Registry data (32). Similarly, a Japanese cross-sectional study reported a positive association between PM_2.5 and uterine cancer mortality using Japanese vital statistics data (66). Our findings may differ from these Japanese studies because of inconsistent air pollutant composition and exposure assessment, data-collection periods, and varying air pollution quality based on the prevalence of industrial and vehicle emissions (67).

We hypothesized that the air pollution–uterine cancer association may be modified by BMI (68,69) and residential geographic characteristics such as ADI (70), census region (24), and urbanicity (71) based on prior literature. We observed that women who resided in urban areas had a higher incidence of uterine cancer than did suburban and rural women. These findings suggest that urban NO₂ may be a marker for a different mixture of air pollutants compared with suburban or rural areas because NO₂ itself is not carcinogenic. NO₂, a prominent air pollution concern in urban areas (72-75), is considered a proxy for vehicular traffic exposure as well as other constituents of air pollutants that may act as endocrine disruptors and carcinogens (76-81). NO₂ and its respective chemical compounds are associated with a higher incidence of breast cancer (82,83) and lung cancer (84), presumably due to the carcinogenic traffic-related air pollutants with which it is correlated. Urban activities, such as fossil fuel combustion and electric power generation, contribute to greater ambient NO₂ and other air pollutants (85-87) and thus may play a role in this observed association. For example, nitric oxide, a carcinogenic compound associated with combustion sources that are oxidized to form NO₂, can influence the development of hormone-sensitive cancers through estrogen and progestogen pathways (83,88-90).

Important strengths of this study include the nationwide prospective study design with more than a decade of follow-up. The study provides detailed information about uterine cancer diagnoses, including key confounders at baseline and over the follow-up period. Validated spatiotemporal models assessed annual exposure to air pollutants from baseline to end of follow-up, accounting for residential mobility, allowing us to incorporate information about air pollutant levels over time and minimize exposure misclassification. The potential still exists for exposure misclassification, however, including from air pollutant exposure at nonresidential locations and the extent to which participants spend time outside their homes. The time-varying Cox proportional hazard model approach enabled a more robust consideration of the impact of air pollution exposure over an extended period before cancer development, which is important given the established temporal decline in air pollution exposure.

Uterine cancer incidence and mortality rates have increased in recent decades (91), especially among non-Hispanic Black women, who are more likely to be diagnosed with aggressive, nonendometrioid subtypes (92). We observed that Black participants were more likely to live in areas of higher NO₂, but we were not statistically powered to evaluate the association stratified by race and ethnicity because of the small number of exposed cases among non-White groups. Our findings highlight the need for additional studies inclusive of racially, ethnically, and geographically diverse populations.

This study population, based on enrollment criteria, included only women with a family history of breast cancer, which may limit the external generalizability of our findings. It is worth noting, however, that the degree of family history and underlying risk varies considerably among the cohort (93). Further, there is limited evidence that a family history of breast cancer is related to a higher incidence of uterine cancer (94).

In this large, prospective cohort of US women, we found that higher residential exposure to NO₂ but not PM_2.5 was associated with a higher incidence of uterine cancer, and the association was elevated among women who resided in urban areas. Our findings support the need for future studies of the role of environmental factors—in particular, traffic-related air pollution—on uterine cancer incidence, particularly in study populations that can explore racial and ethnic disparities.

Data availability

All data necessary to reproduce the current analysis are available following procedures described on the Sister Study website (https://sisterstudy.niehs.nih.gov/English/data-requests.htm).

Author contributions

Jordyn A. Brown, MPH (Conceptualization; Formal analysis; Investigation; Writing—original draft; Writing—review & editing), Jennifer L. Ish, PhD (Conceptualization; Investigation; Supervision; Writing—review & editing), Che-Jung Chang, PhD (Conceptualization; Investigation; Supervision; Writing—review & editing), Deborah B. Bookwalter, ScD, MS (Formal analysis; Methodology; Writing—review & editing), Katie M. O’Brien, PhD (Writing—review & editing), Rena R. Jones, PhD (Writing—review & editing), Joel D. Kaufman, MD, MPH (Resources; Writing—review & editing), Dale P. Sandler, PhD (Resources; Writing—review & editing), Alexandra J. White, PhD, MSPH (Conceptualization; Investigation; Resources; Supervision; Writing—review & editing).

Funding

This research was supported by the Intramural Research Program of the NIH, National Institute of Environmental Health Sciences (grant No. Z01-ES044005, Z1AES103332). The air pollution models and related efforts of Dr Kaufman were supported by the NIH (R01ES027696), the University of Washington Interdisciplinary Center for Exposures, Diseases, Genomics, and Environment (P30ES007033), and the EPA (RD831697 and RD-83830001). This work has not been formally reviewed by the EPA. Ms Brown is currently supported by the NIH National Cancer Institute Cancer Care Quality Training Program (T32CA116339).

Conflicts of interest

The authors declare no conflict of interest.