Skip to main content
NCBI home page
NIHPA Author Manuscripts logoLink to NIHPA Author Manuscripts
. Author manuscript; available in PMC: 2026 Feb 6.
Published in final edited form as: Environ Int. 2025 Nov 26;206:109966. doi: 10.1016/j.envint.2025.109966

Air pollution and breast cancer incidence in a United States-wide prospective cohort study: Examining sensitive periods of exposure

Jennifer L Ish a, Judy Y Ou b,c, Alison M Rector-Houze d, Nicole M Niehoff e, Jianzhao Bi f, Joel D Kaufman f,g, Dale P Sandler a, Alexandra J White a
PMCID: PMC12874600  NIHMSID: NIHMS2126893  PMID: 41330098

Abstract

Background:

Mounting evidence supports that air pollution is related to a higher breast cancer risk, yet the importance of exposure timing in this relationship remains unclear.

Methods:

In the Sister Study, a United States-wide prospective cohort (n=50,884, 2003–2009), we estimated time-varying annual concentrations of nitrogen dioxide (NO2) and fine particulate matter (PM2.5) from 1990–2017 at residential addresses using validated spatiotemporal models. Self-reported breast cancer diagnoses were validated using medical records. We used Cox proportional hazards regression to estimate hazard ratios (HRs) and 95% confidence intervals (CIs) for breast cancer incidence in relation to air pollutant concentrations during predetermined windows of susceptibility. We also applied distributed lag non-linear models to estimate adjusted cumulative and lag-specific HRs and 95% CIs for the association between air pollutants and breast cancer across a lag period of 0–15 years. We evaluated breast cancer overall and by estrogen receptor (ER) status and tumor extent [ductal carcinoma in situ (DCIS) versus invasive].

Results:

We found limited evidence that air pollutant exposure during the time of a woman’s first birth, most recent birth, or menopause transition was associated with heightened risk for breast cancer. When examining exposure flexibly over the long-term, a 10-ppb increase in NO2 across lag years 1–11 significantly contributed to the risk of ER-positive (HRcumul= 1.14; 95% CI: 1.03–1.27; n=2,619 cases) and DCIS (HRcumul=1.27, 95% CI: 1.04–1.54; n=706 cases) breast cancer, whereas PM2.5 experienced during lag years 11–13 was associated with ER-negative breast cancer (e.g., HRLag12l=1.36 per 10-µg/m3 increase, 95% CI: 1.02–1.81; n=448 cases).

Conclusions:

We identified unique periods of susceptibility to NO2 and PM2.5 for breast cancer risk by ER status.

Keywords: air pollution, breast cancer, distributed lag non-linear models, windows of susceptibility

1. Introduction

Air pollution is a ubiquitous and complex exposure containing known mammary carcinogens and components that demonstrate estrogenic activity, relevant to the hormonal etiology of breast cancer (1, 2). Breast cancer is the most common invasive cancer diagnosed among women in the United States (US) and worldwide (3). Many established risk factors for breast cancer—including reproductive history, obesity, and use of exogenous hormones (4)—are consistent with the role of hormones in the etiology of the disease and suggest that environmental exposures with endocrine disrupting properties have the potential to influence breast cancer risk (2, 5). Epidemiologic research has demonstrated that increased breast cancer risk is associated with higher exposure to ambient air pollution (6, 7). For example, nitrogen dioxide (NO2), a marker of traffic-related air pollution, has been consistently associated with increased breast cancer risk (6, 7). While the literature regarding particulate matter less than 2.5 microns in diameter (PM2.5) has been less consistent, evidence of positive association, particularly in prospective cohort studies, is accumulating (6–9).

While existing studies make use of sophisticated air pollution exposure models that provide fine spatiotemporal resolution, few consider historic exposures (9) with most relying on simple definitions of exposure during a single period in time, for example, at the time of study enrollment (10). While this may serve as a reasonable proxy of long-term exposure levels, this approach may fail to capture exposure during the most etiologically relevant period for breast cancer and precludes the assessment of delayed effects of exposures during potentially sensitive periods. For instance, during windows of susceptibility (WOS) for breast cancer—key life stages characterized by significant changes in the structure and function of breast tissue as well as hormonal fluctuations—individuals may be most susceptible to the effects of air pollution on the breast (11, 12). Other susceptible periods may exist but remain unidentified, as the lack of detailed past residential histories and historical air pollution estimates have hindered efforts to identify sensitive periods of exposure in relation to breast cancer risk.

Previous reports from the Sister Study have linked increased breast cancer incidence to outdoor concentrations of NO2 and PM2.5 at study baseline (10, 13). However, the role of historic exposures—now available through newly obtained, retrospective residential histories of women in the cohort—has yet to be examined. Understanding how exposure timing influences the relationship between air pollution and breast cancer may shed light on the biologic mechanisms involved and help guide prevention strategies aimed at reducing exposure. Thus, the aim of this study was to examine temporal dependencies in the relationship between NO2 and PM2.5 with breast cancer incidence. To accomplish this, we examined exposure during windows of susceptibility (pregnancy and the menopause transition). We also used distributed lag non-linear models (DLNMs), a novel application to the study of breast cancer, to flexibly identify sensitive periods of exposure to air pollution.

2. Methods

2.1. Study population

The Sister Study is an ongoing nationwide prospective cohort study of 50,884 women designed to investigate environmental and lifestyle risk factors for breast cancer (14). Participants were recruited between 2003 and 2009 across the United States and Puerto Rico through a multifaceted outreach strategy employing multimedia campaigns, volunteer networks, breast cancer professionals, and advocates. Eligibility criteria included women aged 35 to 74 with no prior history of breast cancer themselves but with a sister previously diagnosed with the disease. Upon enrollment, participants completed an extensive computer-assisted baseline telephone questionnaire that collected information on demographic characteristics, lifestyle and reproductive factors, medical and family history, and current residential address. Additionally, trained medical examiners visited participants’ homes to conduct physical assessments. Breast cancer history is updated annually, and every 2–3 years, participants are asked to complete detailed follow-up questionnaires to provide updates to lifestyle and risk factor information, health status, and their residential address.

The Sister Study was approved by the Institutional Review Board of the National Institutes of Health, and all participants provided written informed consent prior to enrollment. Response rates have remained above 84% throughout follow-up. This analysis includes follow-up to September 30, 2021 (Data Release 11.1).

2.2. Ascertainment of breast cancer cases and tumor characteristics

Women who report a breast cancer diagnosis are asked to release their medical records including pathology reports to confirm diagnosis and obtain information on tumor characteristics, including hormone receptor status. Over 80% of self-reported breast cancer diagnoses have been confirmed though medical reports. Further, agreement between medical records and self-reported breast cancer and tumor characteristics is high, with a positive predictive value of over 99% for breast cancer overall (15). Thus, self-reported diagnoses were used when medical records could not be obtained.

The primary outcome of interest was invasive breast cancers or ductal carcinoma in situ (DCIS) diagnosed during follow-up. We also evaluated the outcome according to tumor invasiveness (invasive versus DCIS) and hormone receptor status (estrogen receptor [ER] positive versus negative). We excluded women who have withdrawn from the study (n = 9), had an uncertain timing of diagnosis or who were diagnosed with breast cancer before completing all baseline data collection (n = 79) or who did not contribute any follow-up time (n = 288).

2.3. Residential address history data

Sister Study participants self-reported their primary and longest adult residential addresses at enrollment and provided updates on residential moves throughout follow up. As previously described (16), residential histories prior to enrollment have been reconstructed using information from LexisNexis, a public records database, and previously developed algorithms for data cleaning (17). Briefly, for each participant identified in their database, LexisNexis provided a set of up to 20 addresses and dates during which the individual was associated with each address. Algorithms were then applied to clean the address data (e.g., remove duplicates and business addresses) and generate continuous address histories with no gaps or overlaps in addresses for each participant. This process resulted in detailed pre-enrollment residential histories for 93.5% (n = 47,557) of the cohort, with 96% of this total having complete address history beginning in 1990. A validation sample of 823 participants verified 95% of address locations and 82% of residential durations in the LexisNexis-generated address histories (18).

2.4. Air pollution exposure assessment

NO2 concentrations [parts per billion (ppb)] from 1990–2017 were estimated from satellite-derived measurements in a validated land-use regression model with kriging (19). The average cross-validated R2 ranged from 0.64–0.85, with similar model performance across years (see Young et al. (19) Figure 4). Historical estimates of PM2.5 (µg/m3, 1980–2017) were developed from a validated regionalized universal kriging model with spatial smoothing which incorporated information from several geographic covariates and PM2.5 concentrations from regulatory monitors after 1999 (20). Temporal trends prior to 1999 were estimated from extrapolation of PM2.5 concentrations from the regulatory monitoring data, PM2.5 sulfate data from the Clean Air Status and Trends Network, and data from the Weather Bureau Army Navy network on visibility. The model was validated using multiple external resources with data collected before 1999. The model performed well for years after 1999 (cross-validated R2 = 0.77–87) and between 1990–1999 (external validation R2 = 0.84–0.91), with lower performance before 1990, likely due in part to the presence of fewer monitoring stations and limitations of the external validation data (20). In this analysis, we only considered years of PM2.5 estimates beginning in 1990 to facilitate comparability of results across pollutants and co-pollutant model adjustment.

Using participants’ geocoded address histories, we assigned average air pollutant concentrations for each year from 1990 (NO2) or 1980 (PM2.5) through 2017. As the exposure model is available only in the contiguous US, we excluded participants who lived outside the continuous US (n = 1,003) or whose address(es) were unable to be geocoded (n = 14).

2.5. Covariates

Baseline covariates for Sister Study participants include self-reported age (years), educational attainment (up to high school diploma or equivalent, some college, four-year degree or more), household income ($49,000 or less, $50,000 to $99,999, $100,000 or more) and race and ethnicity. Participants self-classified their race and ethnicity separately. They were first asked if they considered themselves Hispanic or Latina and then selected one or more racial category. We combined race and ethnicity into one variable with precedence to Hispanic ethnicity and then Black race, resulting in categories of Hispanic (any race), non-Hispanic Black, non-Hispanic White, and additional groups (non-Hispanic American Indian or Alaska Native, non-Hispanic Asian or Pacific Islander, other, unknown or not specified). Family history of breast cancer (number of first-degree female relatives, including half siblings, with breast cancer), breast feeding history (ever versus never), and type of menopause (natural versus surgical) was self-reported at baseline and over follow-up. Based on participants’ residential address(es), we assigned time-varying US Census Region (Midwest, Northeast, South, West). We also assigned time-varying median census tract household income and percent of the census tract population who lived in an urban area to participant’s baseline and follow-up residential address(es). According to the year lived at each address, we assigned 2000 or 2010 Census values. To standardize 2000 and 2010 median household income, we categorized values into quartiles based on the distribution among all participants at baseline. For percent of the census tract population who live in an urban area, we dichotomized values into 100% urban vs <100% urban.

2.6. Statistical analysis

2.6.1. Window of susceptibility-specific models

We evaluated exposure during the first and most recent births. Since breast tissue differentiation is a process that occurs through pregnancy and lactation (11), we defined the first and most recent birth periods as the year of age at the index birth and the year after. We excluded women who had not given birth prior to the end of follow-up (n = 6,142 never pregnant; n = 2,752 without a live birth) or had unknown age at birth (n = 40) from models assessing associations during the pregnancy windows. After excluding participants without air pollution data during the defined exposure periods (largely women who gave birth before 1980 or 1990), 4,378 were included in models evaluating the first birth period, and 7,776 were included in models evaluating the most recent birth period. Details of sequential exclusions for each model are presented in Supplemental Material, Figure 1.

Our definition of the menopausal transition period was informed by the Stages of Reproductive Aging Workshop (21) and consisted of the five-year period leading up to the age at menopause. Although the length of the menopausal transition is variable, the five years before the final menstrual period captures a time when more consistent, biologically measurable changes—such as declining ovarian reserve, fluctuating hormone levels, and increased menstrual irregularity—begin in many women (21). Women were considered postmenopausal if they indicated any of the following at baseline or at annual health updates during follow-up: did not have a menstrual period in the past 12 months (not due to gynecologic surgery or medical procedure); underwent a medical procedure (e.g., chemotherapy, radiation, or other treatment) that permanently stopped menstrual periods; had a bilateral oophorectomy; had a hysterectomy without a bilateral oophorectomy and were ≥ 55 years of age; had a uterine ablation and were ≥ 55 years of age; or had unknown menopausal status and were ≥ 55 years of age. The age at menopause is derived from the age associated with the cause of menopause: age at last menstrual period, age at surgery, or at 55 years. We excluded women who had not reached menopause prior to the end of follow-up or had unknown age at menopause (n = 3,001). A total of 14,685 participants did not have air pollution data available during the defined exposure period. After additionally excluding participants with missing covariate data, 31,758 were included in models evaluating air pollutant exposure during the menopause transition.

To capture a time period not specific to a biologically relevant window of susceptibility as a point of comparison, we also defined and assigned average exposure during the pre-baseline period (1998–2002). We performed sensitivity analyses to assess the robustness of the results to exposure window definition and availability of exposure data. We examined alternative definitions for first and last births (year of index birth) and the menopause transition (3-year period leading up to the year of age at menopause). We also evaluated exposure during the first birth, most recent birth and at age 35 among participants with available exposure data for all three periods.

Separately for each pollutant, we used Cox proportional hazards regression to estimate hazard ratios (HRs) and 95% confidence intervals (CIs) for the association between a 10 unit increase in window-specific concentrations and breast cancer incidence. Models were adjusted for age (timescale), race and ethnicity, household income, and educational attainment. We also evaluated co-pollutant models (NO2 was adjusted for PM2.5 during the index period and vice versa). Person-years accrued from age at study entry through the earliest of age at diagnosis, death, loss-to-follow-up or end of follow-up. We estimated outcome-specific associations for overall, invasive, and ER-positive breast cancer. For overall breast cancer, we evaluated whether estimates varied by family history of breast cancer, age at index birth (first and most recent birth only), breastfeeding history (first and most recent birth only), parity (most recent birth only), and menopause type (menopause transition only). We used an interaction term between exposure and the modifying factor to estimate stratified HRs and used a likelihood-ratio test comparing models with and without cross-product terms to determine the p-value for the interaction.

2.6.2. Distributed lag non-linear models

To evaluate temporal dependencies in the association between air pollutant exposure and incident breast cancer, we used distributed lag non-linear models (DLNMs) to flexibly estimate the exposure-lag-response relationship. DLNMs overcome the limitations of averaging exposures across a long period of time by estimating potentially non-linear variations in the lag dimension, described in more detail in Gasparrini (22).

Age-specific annual exposure values for PM2.5 and NO2 were assigned for each participant. We defined the lag period as 0–15 years to align with the latency of breast cancer (23) and exposure data availability. We excluded participants who contributed less than a year of follow-up (n = 461) or with missing or incomplete exposure histories (n = 1,597). Given our longitudinal cohort setting, we used the extended DLNM methodology for time-to-event data described in Gasparrini (22). After excluding women with missing covariate data (n = 178), there were 47,545 participants included in our analysis. We first created risks sets by splitting the data from 47,545 participants at each failure time, generating 487,827 person-time observations. Next, we constructed the exposure history matrix, which consists of annual exposures during a lag period of 0 to 15 years from age at diagnosis (cases) or end of follow-up defined by the risk set (non-cases) for each person-time observation. Details of sequential exclusions for the distributed lag models are presented in Supplemental Material, Figure 2.

To create the cross-basis matrix (a bi-dimensional space of functions applied to the exposure history matrix), we applied a combination of a linear exposure-response function and various lag-response functions. Given evidence of a linear association between air pollution and breast cancer in prior literature (6), we chose a priori to define the exposure-response function as linear. For the lag-response function, we considered piecewise constant functions, quadradic B-spline, and natural cubic spline functions. For the spline functions, we determined the number and placement of knots by evaluating multiple models with up to two knots (to avoid overfitting) at equally spaced intervals of the lag distribution. We used Akaike’s Information Criterion (AIC) and Bayesian Information Criterion (BIC) to guide optimal model fit, selecting the model that minimized these values. When the two criteria did not agree, we gave precedence to AIC to avoid underfitting (22). For each pollutant, we aimed for as much consistency as possible in the cross-basis functions across outcome-specific models. The best performing models for NO2 used natural cubic splines for the lag-response function, and the best performing models for PM2.5 used quadratic beta splines. Details of model specification and fit statistics of the various options considered for each exposure-outcome pair are shown in Supplemental Material, Table 5. To estimate associations, we used Cox proportional hazards regression. Final models included the cross-basis exposure matrix, age (timescale), calendar year, race and ethnicity, educational attainment, household income, time-varying census region, time-varying census tract-level urbanicity, and time-varying census tract median household income. We evaluated both single pollutant and co-pollutant models, where the co-pollutant models included the cross-basis exposure matrices for NO2 and PM2.5. We generated lag-response curves that represent outcome-specific hazard ratios (HRs) and 95% confidence intervals (CIs) for a 10 unit increase in NO2 (ppb) or PM2.5 (µg/m3) for each lag year. To summarize the overall impact of the exposure history, we also generated cumulative HRs by summing over the estimated exposure-lag coefficients using a summation vector. The variance of the cumulative effect estimate is derived by propagating the variance of the estimated lag coefficients through the summation operation. We estimated outcome-specific associations for breast cancer overall, by invasiveness (invasive vs. DCIS), and by ER status (ER-positive vs. ER-negative) breast cancer. To facilitate comparison with traditional regression models, we also estimated the overall association using both DLNMs with a single stratum for the lag-response function (i.e., constant lag-response) and cox proportional hazards with a rolling 15-year average exposure concentration with a 1-year exposure lag.

Analyses were performed in R version 2024.12.0 using the package ‘dlnm’ (24). Statistical significance was determined using an alpha of 0.05.

3. Results

3.1. Population and exposure characteristics

On average, participants were 55.1 (standard deviation = 8.9) years of age at study enrollment (Table 1). The majority were non-Hispanic White (84%) and had attended at least some college (85%). Sociodemographic characteristics and reproductive history of participants with window-specific air pollution data are presented in Supplemental Material, Table 1. Women with air pollution data available during the time of their first and/or most recent births were, on average, younger at baseline and older at the first birth than the overall population. A greater proportion also had a household income of $100,000 or greater and have obtained at least a four-year college degree at baseline. The characteristics of women between those with available exposure data during the menopause transition and at baseline were similar.

Table 1.

Baseline sociodemographic characteristics for participants included in dlnm analysis, Sister Study, 2003–2009

Overall Cases
N = 47,723 N = 3,339
Age at baseline (years), mean (SD) 55.1 (8.93) 56.5 (8.75)
Race and ethnicity, n (%)
  Hispanic 1,458 (3.1) 71 (2.1)
  Non-Hispanic Black 4,268 (8.9) 268 (8.0)
  Non-Hispanic White 40,754 (85.4) 2,906 (87.0)
  Additional groups a 1,229 (2.6) 93 (2.8)
  Missing 14 (0.0) 1 (0.0)
Educational attainment, n (%)
  High school graduate or less 7,133 (14.9) 468 (14.0)
  Some college or technical school 16,088 (33.7) 1,073 (32.1)
  Four-year degree or more 24,490 (51.3) 1,797 (53.8)
  Missing 12 (0.0) 1 (0.0)
Household income, n (%)
  Less than $50,000 11,518 (24.1) 782 (23.4)
  $50,000–$99,999 19,807 (41.5) 1,356 (40.6)
  $100,000 or more 16,398 (34.4) 1,201 (36.0)
Census region, n (%)
  Midwest 7,794 (16.3) 545 (16.3)
  Northeast 12,704 (26.6) 899 (26.9)
  South 16,685 (35.0) 1,150 (34.4)
  West 10,540 (22.1) 745 (22.3)
% population of census tract in urban area
  100% 26,073 (54.6) 1,886 (56.5)
  < 100% 21,490 (45.0) 1,445 (43.3)
  Missing 160 (0.3) 8 (0.2)
Median census tract household income b
  Q1 12,081 (25.3) 815 (24.4)
  Q2 12,240 (25.6) 826 (24.7)
  Q3 11,994 (25.1) 820 (24.6)
  Q4 11,250 (23.6) 870 (26.1)
  Missing 158 (0.3) 8 (0.2)
Open in a new tab

SD, standard deviation; Q, quartile

a

Additional groups include non-Hispanic American Indian or Alaska Native, non-Hispanic Asian or Pacific Islander, other, unknown or not specified

b

Median household income (USD$) in 2000 census tracts at baseline residential addresses (Q1: 9,321 to 38,854; Q2: >38,854 to 50,562; Q3: >50,562 to 66,250; Q4: >66,250 to 200,001)

The distribution of exposure concentrations by year shows a gradual and steady decline over time in NO2 (1990–2017) and PM2.5 (1980–2017) (Supplemental Material, Figures 3–4). Average concentrations were higher during the first birth (median [interquartile range, IQR], NO2: 12.3 [8.4, 16.9], PM2.5: 15.4 [12.7, 18.1]) and most recent birth windows (median [IQR], NO2: 11.2 [7.7, 15.6], PM2.5: 14.9 [12.2, 17.7]) than during the menopause transition window (median [IQR], NO2: 9.1 [6.2, 13.2], PM2.5: 12.2 [9.7, 15.0]; Supplemental Material, Table 2). Pearson correlation coefficients for the correlation between air pollutant concentrations at each time period ranged from 0.65–0.83 for NO2 and 0.54–0.87 for PM2.5 (Supplemental Material, Table 3).

3.2. Windows of susceptibility

Table 2 shows the association between WOS-specific air pollutant exposure and breast cancer incidence. For breast cancer overall, a 10-ppb increase in NO2 was associated with an increased risk of similar magnitude across both WOS (First birth: HR = 1.10, 95% CI: 0.95, 1.27; Most recent birth: HR = 1.07, 95% CI: 0.96, 1.21; Menopause transition: HR = 1.05, 95% CI: 0.99, 1.12) and the pre-baseline period (HR = 1.08, 95% CI: 1.02, 1.13). For ER-positive breast cancer, the association was similar for a 10-ppb NO2 exposure during the menopause transition (HR = 1.08, 95% CI: 1.0, 1.17), the most recent birth (HR = 1.11, 95% CI: 0.97, 1.28) and the pre-baseline period (HR = 1.09, 95% CI: 1.03, 1.15). The association of greatest magnitude was observed for ER-positive breast cancer and NO2 levels during the first birth (HR = 1.21, 95% CI: 1.03, 1.42), which was slightly attenuated after adjusting for PM2.5 (HR = 1.19, 95% CI: 0.98, 1.45). However, when including only participants who had exposure data available during the first and most recent birth windows, as well as at age 35, we see associations of similar magnitude at each time period (HRFirst birth = 1.21, 95% CI: 1.02, 1.44; HRLast birth = 1.27, 95% CI: 1.05, 1.53; HRAge35 = 1.20, 95% CI: 1.00, 1.44; Supplemental Material, Table 4).

Table 2.

Air pollutant concentrations during windows of susceptibility and breast cancer incidence, Sister Study, 2003–2009

Exposure
window		 Overall Invasive ER-positiveb
HR (95% CI) HR (95% CI) HR (95% CI)
First birth a (N = 4,380)
     Cases (n) 334 256 247
     NO2
     Model 1 1.10 (0.95, 1.27) 1.11 (0.94, 1.30) 1.19 (1.01, 1.40)
     Model 2 1.10 (0.95, 1.27) 1.12 (0.95, 1.33) 1.21 (1.03, 1.42)
     M2 + co-pollutant 1.08 (0.91, 1.29) 1.10 (0.91, 1.34) 1.19 (0.98, 1.45)
 
     PM2.5
     Model 1 1.14 (0.86, 1.52) 1.17 (0.85, 1.63) 1.20 (0.86, 1.67)
     Model 2 1.16 (0.86, 1.55) 1.19 (0.85, 1.66) 1.27 (0.91, 1.78)
     M2 + co-pollutant 1.06 (0.76, 1.50) 1.08 (0.73, 1.59) 1.05 (0.71, 1.56)
Most recent birth a (N = 7,778)
     Cases (n) 591 463 433
     NO2
     Model 1 1.08 (0.96, 1.22) 1.06 (0.93, 1.21) 1.10 (0.96, 1.26)
     Model 2 1.07 (0.96, 1.21) 1.05 (0.92, 1.20) 1.11 (0.97, 1.28)
     M2 + co-pollutant 1.07 (0.94, 1.23) 1.05 (0.90, 1.23) 1.11 (0.95, 1.30)
 
     PM2.5
     Model 1 1.09 (0.88, 1.37) 1.07 (0.83, 1.37) 1.06 (0.82, 1.38)
     Model 2 1.08 (0.86, 1.35) 1.05 (0.81, 1.36) 1.12 (0.86, 1.45)
     M2 + co-pollutant 1.01 (0.78, 1.31) 1.00 (0.75, 1.34) 1.01 (0.75, 1.37)
Menopause transition a (N = 31,761)
     Cases (n) 2,574 2,039 1,800
     NO2
     Model 1 1.05 (0.99, 1.12) 1.01 (0.95, 1.09) 1.07 (0.99, 1.15)
     Model 2 1.05 (0.99, 1.12) 1.02 (0.95, 1.10) 1.08 (1.00, 1.17)
     M2 + co-pollutant 1.08 (1.01, 1.16) 1.06 (0.98, 1.15) 1.10 (1.01, 1.20)
 
     PM2.5
     Model 1 0.99 (0.88, 1.11) 0.92 (0.81, 1.05) 1.01 (0.88, 1.16)
     Model 2 0.98 (0.87, 1.10) 0.92 (0.80, 1.05) 1.03 (0.90, 1.19)
     M2 + co-pollutant 0.91 (0.79, 1.04) 0.87 (0.75, 1.01) 0.95 (0.81, 1.11)
Baseline a (N = 45,546)
     Cases (n) 4,169 3,294 3,017
     NO2
     Model 1 1.08 (1.03, 1.14) 1.05 (1.00, 1.11) 1.08 (1.02, 1.14)
     Model 2 1.08 (1.02, 1.13) 1.05 (1.00, 1.11) 1.09 (1.03, 1.15)
     M2 + co-pollutant 1.08 (1.02, 1.14) 1.06 (1.00, 1.13) 1.10 (1.03, 1.17)
 
     PM2.5
     Model 1 1.08 (0.97, 1.20) 1.03 (0.92, 1.16) 1.01 (0.90, 1.14)
     Model 2 1.06 (0.96, 1.18) 1.03 (0.91, 1.16) 1.05 (0.93, 1.19)
     M2 + co-pollutant 0.99 (0.87, 1.11) 0.97 (0.85, 1.11) 0.95 (0.83, 1.10)
Open in a new tab

NO2, nitrogen dioxide; PM2.5, particulate matter < 2.5 µm in diameter; ER; estrogen receptor

a

Exposure period definitions: First birth and most recent birth defined as 2-year period including the year of age of the index birth and following year. Menopause transition defined as 5-year period leading up to year of age at menopause. Pre-baseline period defined as 5-year period from 1998–2002.

Hazard ratios (HRs) and 95% confidence intervals (CIs) for a 10-unit increase in NO2 (parts per billion, ppb) or PM2.5 (µg/m3) estimated via Cox proportional hazards regression

Model 1 adjusted for age in years (timescale)

Model 2 (M2) adjusted for age in years (timescale), race and ethnicity, educational attainment, and household income

b

ER-positive includes both invasive and DCIS tumors (first birth: 60 DCIS, 187 invasive; most recent birth: 98 DCIS, 335 invasive; menopause transition: 359 DCIS, 1,458 invasive; baseline; 612 DCIS, 2,400 invasive, 4 uncertain)

There was little evidence of an association between a 10-µg/m3 increase in PM2.5 during each WOS and breast cancer. An elevated but imprecise association for overall breast cancer in relation to exposure levels during the first birth (HR = 1.16, 95% CI: 0.86, 1.55) did not persist after co-pollutant adjustment (HR = 1.06, 95% CI: 0.76, 1.50). Similarly, co-pollutant adjusted models were null for exposure during the most recent birth (HR = 1.01, 95% CI: 0.78, 1.31) and menopause transition (HR = 0.91, 95% CI: 0.79, 1.04). Associations with invasive and ER-positive breast cancer were also null. When only considering women with exposure data during the first and most recent births, we observed an elevated but imprecise HR for overall breast cancer in relation to exposure during the most recent birth (HRLast birth = 1.22, 95% CI: 0.84, 1.77) compared to the association for overall breast cancer in relation to exposure at age 35 (HRAge35 = 1.22, 95% CI: 0.84, 1.77).

Using alternative definitions for the WOS time periods did not meaningfully change the results (Supplemental Material, Table 5). We examined associations with exposure during the first and most recent births according to age at first birth, parity (most recent birth models only) breastfeeding history, and family history of breast cancer, and with exposure during the menopause transition by menopause type and family history of breast cancer but did not find evidence of heterogeneity (Supplemental Material, Table 6).

3.3. Distributed lag models

Hazard ratios (HRs) and 95% confidence intervals (CIs) for the associations between the co-pollutant adjusted cumulative air pollutant exposure history on breast cancer incidence are shown in Table 3. Accounting for all contributions from lag years 0–15 (i.e., the cumulative association across the entire lag period) in the DLNM-spline model, a 10-ppb increase per year in NO2 was associated with a 10% higher incidence of overall breast cancer (HR = 1.10, 95% CI: 1.00, 1.21), and this association was stronger for DCIS (HR = 1.27, 95% CI: 1.04, 1.54). The association was also evident for ER-positive (HR = 1.14, 95% CI: 1.03, 1.27) but not ER-negative breast cancer (HR = 0.94, 95% CI: 0.71, 1.24). We obtained comparable effect estimates from the DLNM using a single stratum for the lag-response function (i.e., constant lag-response) and the Cox proportional hazards regression model with a rolling 15-year exposure average.

Table 3.

Hazard ratios (HRs) and 95% confidence intervals CI) for the association between air pollutant exposure history and breast cancer incidence overall, by invasiveness and by estrogen receptor status, using different approaches for modeling the 0–15-year lag period (N = 47,545)

Invasiveness ER status a
Overall Invasive DCIS ER-positive ER-negative
Cases (n) 3,330 2,532 706 2,619 448
NO2 (per 10-ppb increase)
DLNM-spline 1.10
(1.00, 1.21)		 1.05
(0.94, 1.17)		 1.27
(1.04, 1.54)		 1.14
(1.03, 1.27)		 0.94
(0.71, 1.24)
DLNM-strata 1.09
(0.99, 1.20)		 1.06
(0.95, 1.18)		 1.21
(0.99, 1.48)		 1.13
(1.01, 1.26)		 0.96
(0.72, 1.27)
Cox proportional hazards 1.11
(1.01, 1.21)		 1.08
(0.97, 1.19)		 1.22
(1.00, 1.48)		 1.14
(1.03, 1.26)		 0.95
(0.71, 1.26)
 
PM2.5 (per 10-µg/m3 increase)
DLNM-spline 0.88
(0.71, 1.10)		 0.90
(0.71, 1.15)		 0.81
(0.52, 1.29)		 0.84
(0.65, 1.09)		 0.95
(0.54, 1.69)
DLNM-strata 0.92
(0.76, 1.12)		 0.91
(0.73, 1.13)		 0.94
(0.62, 1.43)		 0.90
(0.73, 1.12)		 1.19
(0.71, 2.02)
Cox proportional hazards 0.88
(0.73, 1.06)		 0.87
(0.71, 1.06)		 0.92
(0.61, 1.38)		 0.87
(0.71, 1.07)		 1.17
(0.68, 2.00)
Open in a new tab

NO2, nitrogen dioxide; PM2.5, particulate matter < 2.5 µm in diameter; ppb, parts per billion; ER, estrogen receptor; HR, hazard ratio; CI, confidence interval; DLNM, distributed lag non-linear model

Cox proportional hazards models for 15-year rolling average air pollutant concentrations with a 1-year lag DLNM (strata): lag-response function specified as a single stratum modeling a constant lag-response association

DLNM (strata): lag-response function specified as a single stratum modeling a constant lag-response association

DLNM (spline): lag-response function specified as a natural cubic spline with 1 knot (NO2) or a beta spline with 1 knot (PM2.5) modeling a non-linear lag-response association

Models adjusted for age in years (timescale), calendar year, race and ethnicity, educational attainment, household income, time-varying census region, time-varying median census tract income, time-varying census tract urbanicity, and co-pollutant

a

Includes both invasive and DCIS tumors (ER-positive: 520 DCIS, 2,008 invasive, 4 uncertain; ER-negative: 107 DCIS, 340 invasive, 1 uncertain)

A 10-µg/m3 increase in PM2.5 over the entire lag period from the DLNM-spline model was inversely but imprecisely associated with overall breast cancer (HR = 0.88, 95% CI: 0.71, 1.10). Estimates were similar across modeling approaches and for each outcome. For ER-negative breast cancer, however, when assuming a constant lag-response association, PM2.5 was positively but imprecisely associated with higher incidence (HR = 1.19, 95% CI: 0.71–2.02).

The effect estimates from the single pollutant models are presented in Supplemental Material, Table 7. The results for NO2 are similar to the co-pollutant model estimates, while the estimates for PM2.5 are closer to the null value of 1.

Analysis of the lag structure underlying these relationships revealed unique potential periods of susceptibility for NO2 and PM2.5 on breast cancer risk that differed by subtype. Figures 1–2 show the lag-response curves, which represent the outcome-specific HRs and 95% CIs for a 10-unit increase in NO2 (ppb) or PM2.5 (µg/m3) across the 15-year lag period. For DCIS and ER-positive breast cancer, we observed a small but consistent increase in risk associated with NO2 exposure across lag years 1–11 (Figure 1). For PM2.5, while the HR for the cumulative exposure history was null, the DLNM identified specific lag years that were significantly associated with ER-negative breast cancer. Specifically, higher risk of ER-negative breast cancer was significantly associated with PM2.5 exposure experienced at lag period of 11–13 years. Similar lag-response patterns were observed for overall and invasive breast cancer although they were not statistically significant. Numeric values for the lag-specific HRs and 95% CIs are given in Supplemental Material, Table 9. Single pollutant models yielded similar lag-response associations (Supplemental Material, Table 10).

Figure 1.

Figure 1.

Open in a new tab

Lag-response relationship representing hazard ratios (HRs) and 95% confidence intervals (CIs) for the association between a 10-parts per billion increase in NO2 and breast cancer incidence along a 15-year exposure period, estimated using distributed lag non-linear models (DLNMs) using natural cubic spline functions and adjusted for age (years), calendar year, race and ethnicity, educational attainment, household income, time-varying census region, time-varying median census tract household income, and time-varying census tract urbanicity.

Figure 2.

Figure 2.

Open in a new tab

Lag-response relationship representing hazard ratios (HRs) and 95% confidence intervals (CIs) for the association between a 10-µg/m3 increase in PM2.5 and breast cancer incidence along a 15-year exposure period, estimated using distributed lag non-linear models (DLNMs) using beta spline functions and adjusted for age (years), calendar year, race and ethnicity, educational attainment, household income, time-varying census region, time-varying median census tract household income, and time-varying census tract urbanicity.

4. Discussion

In this large, established cohort study, we estimated historic exposure to NO2 and PM2.5 to evaluate associations between exposure, both during WOS and flexibly over the long-term, in relation to breast cancer incidence. We found limited evidence that air pollutant exposure during the time of a woman’s first birth, most recent birth, or menopause transition was associated with heightened risk for breast cancer. However, when we examined long-term adult exposure flexibly, we found that recent chronic exposure to NO2 was associated with increased incidence of ER-positive breast cancer, while PM2.5 exposure experienced during a lag period of 11–13 years was associated with higher incidence of ER-negative breast cancer.

Our results are in line with current evidence that supports a relationship between NO2 exposure and increased breast cancer risk, particularly for hormone receptor positive breast cancer. For example, a meta-analysis of 21 observational studies reported a 2.3% increased risk of breast cancer (95% CI: 1.005, 1.041) associated with a 10-µg/m3 increase in NO2 concentrations (6). This association was evident for ER+/PR+ tumors (risk ratio [RR] = 1.045; 95% CI: 0.980, 1.114) but not ER-/PR- tumors (RR = 0.987; 95% CI: 0.885, 1.101). The stronger association observed in our analysis (HR = 1.11, 95% CI: 1.04, 1.19) compared to these meta-analyzed estimates (HRs ~ 1.02) and a previous Sister Study report examining exposure at baseline (HR = 1.06, 95% CI: 1.02, 1.11) (10) could reflect the importance of long-term, chronic exposure to NO2 for the development of breast cancer.

The patterns observed in the DLNM lag-response curves for NO2, particularly the elevated risks for ER-positive breast cancers over a 10-year lag period, suggest hormonally mediated mechanisms may underlie these associations. While NO2 itself is not considered carcinogenic, it is a marker of traffic related air pollution and represents a mixture of pollutants including heavy metals and aromatic hydrocarbons such as benzene and polycyclic aromatic hydrocarbons (PAHs) (25). These compounds have been shown to play a role in breast cancer pathophysiology due to their endocrine disrupting properties (26, 27). Given that ER-positive tumors rely on the presence of hormones for growth, our observation that higher incidence of ER-positive breast cancer was associated with NO2 levels across a 10-year lag period is consistent with the hypothesized role of hormonally active compounds in traffic related air pollution as tumor promotors. The similar pattern of results for DCIS breast cancer could reflect that 70% of DCIS tumors express estrogen receptors (28). However, due to the shared profiles of hormone receptor expression and risk factors between DCIS and invasive breast cancer (29), the mechanisms underlying the elevated risk of DCIS cancer are less clear. Given the smaller number of DCIS cases and resulting lower statistical power, it is also uncertain whether this elevated risk meaningfully differs from that of invasive breast cancer. Notably, because that all women in the cohort have a family history of breast cancer and are engaged in regular screening (95% had a mammogram in the two years before study enrollment), we do not expect the association between NO2 and DCIS to be explained by differential screening practices. Given that DCIS is less commonly diagnosed and therefore more difficult to study in cohort settings, our ability to examine this outcome contributes to the limited evidence base.

Although the literature regarding PM2.5 exposure and breast cancer risk has been mixed, evidence of a positive association is accumulating, particularly in more recently conducted prospective cohort studies (7–9). For example, a 2025 meta-analysis found that a 10-µg/m3 increase in PM2.5 exposure levels were associated with a 6% increased risk of breast cancer (7). Associations with tumor subtypes defined by hormone receptor status, however, remains mixed (6–9). In this study, we did not find strong evidence that long-term exposure to PM2.5 was associated with breast cancer, which is consistent with a previous Sister Study report that found a null association with PM2.5 exposure levels at baseline (10). Notably, the estimate for baseline exposures in our analysis (i.e., the average PM2.5 concentration from 1998–2002; HR = 1.06, 95% CI: 0.96, 1.18) differed from the estimate for time-varying 15-year average PM2.5 concentrations (see Table 3, Cox proportional hazards model; HR = 0.88, 95% CI: 0.73, 1.06). This discrepancy may indicate that specific sensitive windows are important for PM2.5—unlike NO2, for which the effect estimate remained relatively stable across the extended lag period. Consistent with this interpretation, the DLNM results suggest that historic PM2.5 exposure may be particularly relevant for the risk of breast cancer.

The DLMN results for PM2.5, which show elevated risk of ER-negative breast cancer in relation to past exposures, point to potential mechanistic pathways—such as tumor initiation and early carcinogenic changes—that may underlie these associations. Specifically, we observed that PM2.5 exposure during a lag of 11–13 years significantly contributed to the overall risk ER-negative breast cancer. This timing is in line with the minimum latency of breast cancer (23) and could suggest PM2.5 plays a role in the initiation of ER-negative tumors. The carcinogenic mechanisms underlying the effects of outdoor air pollution on breast cancer risk have yet to be fully understood. Still, PM2.5 contains mutagenic and genotoxic compounds and can disrupt several molecular processes by increasing inflammation and oxidative stress, inducing genomic instability, and causing epigenetic modifications that alter the expression of key cancer-related genes including tumor suppressor genes and oncogenes (30). Additionally, while ER-negative tumors do not depend on hormones for growth, endocrine disrupting compounds contained in PM2.5, such as PAHs or heavy metals, may disrupt hormonal pathways that promote their formation (31, 32). Other pathways by which PM2.5 may influence risk include alterations in the breast tissue that increases susceptibility to carcinogenesis. For instance, epidemiologic studies have observed increased PM2.5 levels were associated with early markers of breast cancer risk, including increased mammographic density (33, 34), reduced involution of terminal duct lobular units (TDLUs) (35), and breast tissue composition (36).

Our study builds upon the existing literature on air pollution and breast cancer by providing some of the first evidence on the importance of timing of exposure to NO2 and PM2.5. In the Sister Study, the creation of detailed retrospective residential histories enabled the unique opportunity to assess historic air pollution exposures. Furthermore, we overcome the limitations of assessing long-term exposure averages to identify sensitive periods of exposure by applying DLNMs (22). Although this methodology was originally developed for time series studies of short-term health effects (37, 38), it has since been extended to investigate the impacts of prolonged, time-varying exposures. However, in the decade following the introduction of this extended modelling framework, the application of DLNMs in cancer and chronic disease research has been limited (39, 40). This is likely due to the challenge of meeting the model’s requirement for complete long-term exposure data, which is often unavailable due of the lack of complete, detailed residential histories. By leveraging such data, our study helps bridge this gap and demonstrates the value of DLNMs in uncovering sensitive periods of exposure relevant to the risk of chronic disease, such as breast cancer. Our results provide evidence that NO2 and PM2.5 may have differential effects on breast cancer by ER status, underscoring the need for further research into the biologic mechanisms underlying the associations, particularly the observed association between past PM2.5 exposures and ER-negative breast cancer.

The breast may also be particularly vulnerable to the impacts of environmental chemicals during specific windows of susceptibility. During the four distinct WOS—the prenatal period, puberty, pregnancy, and the menopause transition—significant changes occur in the structure and function of the mammary gland as well as to hormone signaling pathways that may make the breast tissue more susceptible to environmental carcinogenesis (11). For example, evidence that tobacco smoking during early life is associated with heightened risk of breast cancer highlights the vulnerability of the developing breast to environmental insults (41). Previous epidemiologic studies that have found increased risk of breast cancer associated with exposure to traffic-related air pollution at the time of a woman’s first pregnancy (42) and markers of higher breast density associated with combined exposure to ambient PAH and tobacco smoke during pregnancy (43). In our analysis of air pollution exposures during WOS, we did not find evidence of the first or most recent birth nor the menopause transition as a sensitive period of exposure to air pollution. However, our analysis was subject to certain limitations that hindered our ability to comprehensively evaluate this question. First, air pollution levels are correlated within individuals over time—individuals tend to experience relatively high or relatively low exposure levels over the lifecourse—making it difficult to parse out the independent effect of air pollution during specific time periods. Statistical methods have been developed to examine timing of susceptibility to environmental chemicals with high intraindividual correlation (44, 45), however, limited overlap in the groups for which WOS air pollution data was available made the application of more sophisticated methods unfeasible. We determined that the advantages of these approaches, namely the ability to formally test heterogeneity of period-specific effect estimates, did not outweigh the loss of generalizability when restricting the analysis to women with exposure data across both WOS (who tended to be both older at their first birth and younger at menopause). Further complicating analysis of exposure during WOS is evidence suggesting air pollution exposure may impact timing of WOS which in turn influences breast cancer risk (46, 47).

Our analysis of air pollution exposure during WOS is subject to some other limitations. Nearly all first births occurred before study enrollment, and because of limited covariate information on time periods before baseline, we assumed baseline socioeconomic factors to be representative of this earlier period in life. Further, there is possible confounding due to unmeasured area-level socioeconomic status for participants at the time of each WOS. The limited availability of air pollution data for years corresponding to participants’ early life precluded analysis of exposures during the prenatal and pubertal periods. This is an important limitation, as both prenatal development and puberty are recognized windows of susceptibility during which environmental exposures may have lasting effects on biological systems and disease risk (11). Because routine air pollution monitoring in the US did not begin until around 1980, it is not feasible to reliably estimate exposures during these earlier life stages for Sister Study participants. Continued investigation of air pollution exposure during these and earlier WOS is imperative for understanding the role of timing of air pollution exposure in breast cancer etiology, and future studies with extended historical exposure data for younger cohorts will be important for capturing the full range of life course exposures and susceptibility.

By design, all Sister Study participants have a family history of breast cancer. While this does not impact the validity of internal comparisons, it may influence the generalizability of the findings. However, for the WOS models, we did not find evidence that associations varied by extent of family history. Lastly, the effect estimates for PM2.5 across analyses had wider confidence intervals than estimates for NO2. We believe this is due to the fact that while PM2.5 and NO2 concentrations are scaled the same (per 10-unit increase), PM2.5 has a narrower concentration range than NO2 (see Supplemental Material Figures 3–4), which gives larger standard errors and thus larger confidence intervals.

In this large prospective study that considered the timing and intensity of long-term exposure to air pollution in relation to breast cancer incidence, we identified unique potential periods of susceptibility for effects of NO2 and PM2.5 on risk for specific hormone receptor subtypes of breast cancer. These findings shed new light on the role of air pollution exposure in breast cancer etiology, informing the design of future studies of air pollution and breast cancer risk and potential biologic mechanisms underlying this association.

Supplementary Material

1

Highlights.

  • We examined timing of air pollution exposure in relation to breast cancer (BC) risk

  • We found limited evidence of heightened risk during windows of susceptibility

  • NO2 was associated with increased risk across a 10-year exposure lag

  • Risk of ER-negative BC was associated with PM2.5 at an 11–13-year lag

Funding:

This research was supported by the Intramural Research Program of the NIH, National Institute of Environmental Health Sciences (Z01-ES103332, Z01-ES044005).

Glossary:

NO2

nitrogen dioxide

PM2.5

particulate matter < 2.5 µm in diameter

ppb

parts per billion

DLNMs

distributed lag non-linear models

ER

estrogen receptor

DCIS

ductal carcinoma in situ

HR

hazard ratio

CI

confidence interval

WOS

windows of susceptibility

Footnotes

Publisher's Disclaimer: This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to our customers we are providing this early version of the manuscript. The manuscript will undergo copyediting, typesetting, and review of the resulting proof before it is published in its final form. Please note that during the production process errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain.

Declaration of competing interests

The authors have nothing to declare

Data availability statement:

Requests for data, including the data and code used in this manuscript, are welcome. De-identified data are made available upon request as described on the public study website (https://sisterstudy.niehs.nih.gov/English/data-requests.htm). The data sharing policy was developed to protect the privacy of study participants and is consistent with study informed consent documents as approved by the NIH Institutional Review Board.

References

  • 1.Loomis D, Grosse Y, Lauby-Secretan B, Ghissassi FE, Bouvard V, Benbrahim-Tallaa L, et al. The carcinogenicity of outdoor air pollution. The Lancet Oncology. 2013;14(13):1262–3. doi: 10.1016/S1470-2045(13)70487-X. [DOI] [PubMed] [Google Scholar]
  • 2.Hiatt RA, Brody JG. Environmental Determinants of Breast Cancer. Annual Review of Public Health. 2018;39(1):113–33. doi: 10.1146/annurev-publhealth-040617-014101. [DOI] [PubMed] [Google Scholar]
  • 3.Arnold M, Morgan E, Rumgay H, Mafra A, Singh D, Laversanne M, et al. Current and future burden of breast cancer: Global statistics for 2020 and 2040. The Breast. 2022;66:15–23. doi: 10.1016/j.breast.2022.08.010. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 4.American Cancer Society. Breast Cancer Facts and Figures 2022–2024. Atlanta, GA: 2022. [Google Scholar]
  • 5.Rodgers KM, Udesky JO, Rudel RA, Brody JG. Environmental chemicals and breast cancer: An updated review of epidemiological literature informed by biological mechanisms. Environmental Research. 2018;160:152–82. doi: 10.1016/j.envres.2017.08.045. [DOI] [PubMed] [Google Scholar]
  • 6.Gabet S, Lemarchand C, Guénel P, Slama R. Breast Cancer Risk in Association with Atmospheric Pollution Exposure: A Meta-Analysis of Effect Estimates Followed by a Health Impact Assessment. Environmental Health Perspectives. 2021;129(5):57012. doi: 10.1289/EHP8419. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 7.Wang R, Wang P, Zhou Y, Wang Y, Xu C, Wang Z, Wang W. Association between long-term ambient air pollution exposure and the incidence of breast cancer: A meta-analysis based on updated evidence. Ecotoxicology and Environmental Safety. 2025;289:117472. doi: 10.1016/j.ecoenv.2024.117472. [DOI] [PubMed] [Google Scholar]
  • 8.Wu AH, Wu J, Tseng C, Stram DO, Shariff-Marco S, Larson T, et al. Air Pollution and Breast Cancer Incidence in the Multiethnic Cohort Study. J Clin Oncol 2025;43(3):273–84. doi: 10.1200/jco.24.00418. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 9.White AJ, Fisher JA, Sweeney MR, Freedman ND, Kaufman JD, Silverman DT, Jones RR. Ambient fine particulate matter and breast cancer incidence in a large prospective US cohort. JNCI: Journal of the National Cancer Institute. 2023. doi: 10.1093/jnci/djad170. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 10.White AJ, Keller JP, Zhao S, Carroll R, Kaufman JD, Sandler DP. Air Pollution, Clustering of Particulate Matter Components, and Breast Cancer in the Sister Study: A U.S.-Wide Cohort. Environmental Health Perspectives. 2019;127(10):107002. doi: 10.1289/EHP5131. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 11.Terry MB, Michels KB, Brody JG, Byrne C, Chen S, Jerry DJ, et al. Environmental exposures during windows of susceptibility for breast cancer: a framework for prevention research. Breast Cancer Res 2019;21(1):96. doi: 10.1186/s13058-019-1168-2. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 12.Martinson HA, Lyons TR, Giles ED, Borges VF, Schedin P. Developmental windows of breast cancer risk provide opportunities for targeted chemoprevention. Exp Cell Res 2013;319(11):1671–8. doi: 10.1016/j.yexcr.2013.04.018. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 13.Reding KW, Young MT, Szpiro AA, Han CJ, DeRoo LA, Weinberg C, et al. Breast Cancer Risk in Relation to Ambient Air Pollution Exposure at Residences in the Sister Study Cohort. Cancer Epidemiol Biomarkers Prev 2015;24(12):1907–9. doi: 10.1158/1055-9965.EPI-15-0787. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 14.Sandler DP, Hodgson ME, Deming-Halverson SL, Juras PS, D’Aloisio AA, Suarez LM, et al. The Sister Study Cohort: Baseline Methods and Participant Characteristics. Environmental Health Perspectives. 2017;125(12):127003. doi: 10.1289/EHP1923. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 15.D’Aloisio AA, Nichols HB, Hodgson ME, Deming-Halverson SL, Sandler DP. Validity of self-reported breast cancer characteristics in a nationwide cohort of women with a family history of breast cancer. BMC Cancer. 2017;17(1). doi: 10.1186/s12885-017-3686-6. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 16.Ish JL, Daniel M, Ringwald P, Niehoff NM, Jones RR, White AJ. Accuracy of LexisNexis-derived retrospective address histories in the Sister Study cohort. J Expo Sci Environ Epidemiol 2025:Under Review. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 17.Stinchcomb DG, Roeser A. NCI/SEER Residential History Project. Rockville, MD: 2016. [Google Scholar]
  • 18.Ish JL, Daniel M, Ringwald P, Niehoff NM, Jones RR, White AJ. Accuracy of LexisNexis-derived retrospective address histories in the Sister Study cohort. J Expo Sci Environ Epidemiol 2025. doi: 10.1038/s41370-025-00802-1. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 19.Young MT, Bechle MJ, Sampson PD, Szpiro AA, Marshall JD, Sheppard L, Kaufman JD. Satellite-Based NO2 and Model Validation in a National Prediction Model Based on Universal Kriging and Land-Use Regression. Environmental science & technology. 2016;50(7):3686–94. doi: 10.1021/acs.est.5b05099. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 20.Kim SY, Olives C, Sheppard L, Sampson PD, Larson TV, Keller JP, Kaufman JD. Historical Prediction Modeling Approach for Estimating Long-Term Concentrations of PM2.5 in Cohort Studies before the 1999 Implementation of Widespread Monitoring. Environmental health perspectives. 2017;125(1):38–46. doi: 10.1289/ehp131. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 21.Harlow SD, Gass M, Hall JE, Lobo R, Maki P, Rebar RW, et al. Executive Summary of the Stages of Reproductive Aging Workshop + 10: Addressing the Unfinished Agenda of Staging Reproductive Aging. The Journal of Clinical Endocrinology & Metabolism. 2012;97(4):1159–68. doi: 10.1210/jc.2011-3362. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 22.Gasparrini A Modeling exposure–lag–response associations with distributed lag non-linear models. Stat Med 2014;33(5):881–99. doi: 10.1002/sim.5963. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 23.Olsson H, Baldetorp B, Fernö M, Perfekt R. Relation between the rate of tumour cell proliferation and latency time in radiation associated breast cancer. BMC Cancer. 2003;3:11. doi: 10.1186/1471-2407-3-11. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 24.Gasparrini A Distributed Lag Linear and Non-Linear Models in R: The Package dlnm. Journal of Statistical Software. 2011;43(8):1 – 20. doi: 10.18637/jss.v043.i08. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 25.Beckerman B, Jerrett M, Brook JR, Verma DK, Arain MA, Finkelstein MM. Correlation of nitrogen dioxide with other traffic pollutants near a major expressway. Atmospheric Environment. 2008;42(2):275–90. doi: 10.1016/j.atmosenv.2007.09.042. [DOI] [Google Scholar]
  • 26.Santodonato J Review of the estrogenic and antiestrogenic activity of polycyclic aromatic hydrocarbons: Relationship to carcinogenicity. Chemosphere. 1997;34(4):835–48. doi: 10.1016/S0045-6535(97)00012-X. [DOI] [PubMed] [Google Scholar]
  • 27.Choe S-Y, Kim S-J, Kim H-G, Lee JH, Choi Y, Lee H, Kim Y. Evaluation of estrogenicity of major heavy metals. Science of The Total Environment. 2003;312(1–3):15–21. doi: 10.1016/S0048-9697(03)00190-6. [DOI] [PubMed] [Google Scholar]
  • 28.Burstein HJ, Polyak K, Wong JS, Lester SC, Kaelin CM. Ductal Carcinoma in Situ of the Breast. New England Journal of Medicine. 2004;350(14):1430–41. doi: doi: 10.1056/NEJMra031301. [DOI] [PubMed] [Google Scholar]
  • 29.Virnig BA, Wang S-Y, Shamilyan T, Kane RL, Tuttle TM. Ductal Carcinoma In Situ: Risk Factors and Impact of Screening. JNCI Monographs. 2010;2010(41):113–6. doi: 10.1093/jncimonographs/lgq024. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 30.Turner MC, Andersen ZJ, Baccarelli A, Diver WR, Gapstur SM, Pope CA, et al. Outdoor air pollution and cancer: An overview of the current evidence and public health recommendations. CA: A Cancer Journal for Clinicians. 2020;70(6):460–79. doi: 10.3322/caac.21632. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 31.Gupta PB, Kuperwasser C. Contributions of estrogen to ER-negative breast tumor growth. The Journal of Steroid Biochemistry and Molecular Biology. 2006;102(1):71–8. doi: 10.1016/j.jsbmb.2006.09.025. [DOI] [PubMed] [Google Scholar]
  • 32.Wenger D, Gerecke AC, Heeb NV, Schmid P, Hueglin C, Naegeli H, Zenobi R. In vitro estrogenicity of ambient particulate matter: contribution of hydroxylated polycyclic aromatic hydrocarbons. Journal of Applied Toxicology. 2009;29(3):223–32. doi: 10.1002/jat.1400. [DOI] [PubMed] [Google Scholar]
  • 33.White AJ, Weinberg CR, O’Meara ES, Sandler DP, Sprague BL. Airborne metals and polycyclic aromatic hydrocarbons in relation to mammographic breast density. Breast Cancer Res 2019;21(1):24. doi: 10.1186/s13058-019-1110-7. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 34.Yaghjyan L, Arao R, Brokamp C, O’Meara ES, Sprague BL, Ghita G, Ryan P. Association between air pollution and mammographic breast density in the Breast Cancer Surveilance Consortium. Breast Cancer Res 2017;19(1):36. doi: 10.1186/s13058-017-0828-3. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 35.Niehoff NM, Keil AP, Jones RR, Fan S, Gierach GL, White AJ. Outdoor air pollution and terminal duct lobular involution of the normal breast. Breast Cancer Res 2020;22(1):100. doi: 10.1186/s13058-020-01339-x. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 36.Ish JL, Abubakar M, Fan S, Jones RR, Niehoff NM, Henry JE, et al. Outdoor air pollution and histologic composition of normal breast tissue. Environment International. 2023;176:107984. doi: 10.1016/j.envint.2023.107984. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 37.Gasparrini A, Armstrong B, Kenward MG. Distributed lag non-linear models. Stat Med 2010;29(21):2224–34. doi: 10.1002/sim.3940. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 38.Armstrong B Models for the Relationship Between Ambient Temperature and Daily Mortality. Epidemiology. 2006;17(6):624–31. doi: 10.1097/01.ede.0000239732.50999.8f. [DOI] [PubMed] [Google Scholar]
  • 39.Kriit HK, Andersson EM, Carlsen HK, Andersson N, Ljungman PLS, Pershagen G, et al. Using Distributed Lag Non-Linear Models to Estimate Exposure Lag-Response Associations between Long-Term Air Pollution Exposure and Incidence of Cardiovascular Disease. International journal of environmental research and public health. 2022;19(5):2630. doi: 10.3390/ijerph19052630. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 40.Vanoli J, de la Cruz Libardi A, Sera F, Stafoggia M, Masselot P, Mistry MN, et al. Long-term Associations Between Time-varying Exposure to Ambient PM2.5 and Mortality: An Analysis of the UK Biobank. Epidemiology. 2025;36(1):1–10. doi: 10.1097/ede.0000000000001796. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 41.Andersen ZJ, Jørgensen JT, Grøn R, Brauner EV, Lynge E. Active smoking and risk of breast cancer in a Danish nurse cohort study. BMC Cancer. 2017;17(1):556. doi: 10.1186/s12885-017-3546-4. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 42.Nie J, Beyea J, Bonner MR, Han D, Vena JE, Rogerson P, et al. Exposure to traffic emissions throughout life and risk of breast cancer: the Western New York Exposures and Breast Cancer (WEB) study. Cancer Causes & Control. 2007;18(9):947–55. doi: 10.1007/s10552-007-9036-2. [DOI] [PubMed] [Google Scholar]
  • 43.Kehm RD, Walter EJ, Oskar S, White ML, Tehranifar P, Herbstman JB, et al. Exposure to polycyclic aromatic hydrocarbons during pregnancy and breast tissue composition in adolescent daughters and their mothers: a prospective cohort study. Breast Cancer Res 2022;24(1). doi: 10.1186/s13058-022-01546-8. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 44.Buckley JP, Hamra GB, Braun JM. Statistical Approaches for Investigating Periods of Susceptibility in Children’s Environmental Health Research. Curr Environ Health Rep 2019;6(1):1–7. doi: 10.1007/s40572-019-0224-5. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 45.Sánchez BN, Hu H, Litman HJ, Téllez-Rojo MM. Statistical Methods to Study Timing of Vulnerability with Sparsely Sampled Data on Environmental Toxicants. Environmental Health Perspectives. 2011;119(3):409–15. doi: doi: 10.1289/ehp.1002453. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 46.Li H, Hart JE, Mahalingaiah S, Nethery RC, Bertone-Johnson E, Laden F. Long-term exposure to particulate matter and roadway proximity with age at natural menopause in the Nurses’ Health Study II Cohort. Environmental Pollution. 2021;269:116216. doi: 10.1016/j.envpol.2020.116216. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 47.Hood RB, Hart JE, Laden F, Rosner B, Chavarro JE, Gaskins AJ. Exposure to Particulate Matter Air Pollution and Age of Menarche in a Nationwide Cohort of U.S. Girls. Environmental Health Perspectives. 2023;131(10):107003. doi: doi: 10.1289/EHP12110. [DOI] [PMC free article] [PubMed] [Google Scholar]

Associated Data

This section collects any data citations, data availability statements, or supplementary materials included in this article.

Supplementary Materials

1
NIHMS2126893-supplement-1.docx (898KB, docx)

Data Availability Statement

Requests for data, including the data and code used in this manuscript, are welcome. De-identified data are made available upon request as described on the public study website (https://sisterstudy.niehs.nih.gov/English/data-requests.htm). The data sharing policy was developed to protect the privacy of study participants and is consistent with study informed consent documents as approved by the NIH Institutional Review Board.

RESOURCES