Environmental Quality and Invasive Breast Cancer

Breast cancer is the most common malignancy among women, and it is estimated that one in eight women will develop invasive breast cancer (1). Recent studies suggest that the risk of breast cancer include a combination of genetic and environmental factors (reviewed in ref. 2) with compelling associations for increased risk with exposures to single pesticides, ionizing radiation, solvents, and other environmental contaminants (3–14). However, the specific environmental contributors to disease risk remain poorly characterized and most importantly, there is paucity of studies evaluating the role of everyday environmental exposures, which occur simultaneously and as mixtures rather than single agents (15, 16). Failing to account for cumulative environmental exposure may result in an underestimation of the true impact of the environment on breast cancer risk (17).

Empirically measuring the totality of the environment remains a challenge in epidemiologic research (17) because of demographic characteristics, including race/ethnicity and socioeconomic status (18) are increasingly correlated with higher breast cancer incidence. In particular, African American and younger women and those residing in rural areas are more likely to be diagnosed with aggressive hormonal subtypes like triple-negative and inflammatory breast cancers (19–27), which are highly invasive and frequently present as late-stage tumors. To address this issue, the environmental quality index (EQI) was developed by the U.S. Environmental Protection Agency (USEPA; ref. 28). EQI is a publicly available database that combines assessments of environmental factors across the entire United States (2000–2005) into an overall environmental assessment score and scores for five domains of environmental quality (air, water, land, sociodemographic, and built environments). These data were used in a recent cancer study across the United States; findings suggest that county-level age-adjusted breast cancer incidence rates are associated with indicators of poor environmental quality, such that areas with worst environmental quality appear to have higher rates of breast cancer (17). In addition to cancers, the EQI has also been linked to health outcomes such as preterm birth and mortality (29, 30). Herein, we utilize individual-level data for women diagnosed with breast cancer in North Carolina to build upon the work of Jagai and colleagues (17) by accounting for confounding individual demographic and lifestyle factors.

Breast cancer is not one disease, but rather distinct subgroups that may be separated by factors such as stage, morphology, histology, gene expression, and hormone receptor status (27, 31–34). Therefore, to test the hypothesis that the incidence of specific breast cancer stages can vary by demographics (age, race, reproductive age or history, weight, income, and location; refs. 27, 35–38) in relation to environmental factors, we utilized EQI datasets to investigate associations between environmental factors and breast cancer (separated as localized, regional, or distant metastatic disease) compared with carcinoma in situ.

Breast cancer patient data were obtained from the North Carolina Central Cancer Registry (NC CCR). The NC CCR is a central reporting system for cancer cases that collects all cancer incidence data for the state of North Carolina, in all 100 counties. We selected breast cancer diagnosed between 2009 and 2014 for our analyses (∼10 years after the time period that the EQI was constructed to capture). Our analyses included patients diagnosed with localized, regional, or distant metastatic invasive breast cancer as classified in the summary staging defined by the Surveillance Epidemiology and End Results (SEER) program 2000 (39). This staging is based on International Classification of Diseases-10 tumor histology and behavior coding, tumor locality within the breast, and lymph node involvement, including metastatic spread (Supplementary Table S1). Localized, regional, or distant metastatic patients were considered cases in separate datasets. Because only patients with cancer are included in the NC CCR, we used patients with carcinoma in situ (includes both ductal and lobular, noninvasive breast cancer) as registry-based controls. County at diagnosis, age, and race were available for all patients with breast cancer, while body mass index (BMI) and smoking status were also available for many but not all individuals.

The EQI is available at the county-level and includes a total environmental assessment, as well as estimates of environmental quality in five separate domains—air, water, land, sociodemographic, and built environments. The EQI was developed in four distinct parts, which includes identification of environmental domains, identifying and reviewing sources of data from 2000 to 2005 for individual factors that would make up each domain, constructing variables based on these data, and reduction of data including compilation into domain-specific and a total EQI score (40). Factors comprising each of the five domains and utilized in statistical analyses can be found in Supplementary Fig. S1 or ref. 40. Data were finalized in 2014 and is publicly available on the USEPA website at https://edg.epa.gov/EPADataCommons/public/ORD/NHEERL/EQI/. The EQI is a continuous variable but was categorized into quartiles for these analyses to aid in interpretation.

Generalized estimating equation models, which take into account clustering of individuals within counties, were used to examine the relationship between breast cancer and environmental quality. Separate models were constructed comparing localized, regional, or distant metastatic breast cancer to patients with carcinoma in situ. Models were constructed for overall and domain-specific EQI. For domain-specific models, the other four domains were included in models as potential confounding variables. Models were also adjusted for individual age, BMI, smoking status, and race. Although age and race data were very complete, BMI and smoking data were missing more frequently. To avoid excluding patients with missing data, an “unknown” category was used for BMI and smoking status. To ensure that the inclusion of this category was not biasing associations, we also conducted complete case analyses (including participants with complete data). Results of these analyses were qualitatively very similar; thus, we primarily present analyses including all patients. All statistical analyses were conducted in SAS (Version 9.4). Although we considered several domains of exposure and outcomes in statistical analyses, we did not perform adjustment for multiple comparisons, as this type of adjustment has not been recommended in the epidemiologic literature (41).

Prior research demonstrates that relationships between the EQI and cancer incidence may differ based on urbanicity. To evaluate the possibility of differential impacts in urban and rural datasets, we conducted a series of sensitivity analyses stratifying by urbanity using the rural–urban continuum code (RUCC). RUCC was originally developed as a nine-item categorization code of proximity to/influence of major metropolitan areas in the SEER program (42), and was restructured into four codes for the EQI (40). Because of small sample sizes in some categories, codes were condensed into three categories for our analyses as follows: metropolitan urbanized = codes 1 + 2 + 3; nonmetro urbanized = 4 + 5; and less populated = 6 + 7 + 8 + 9 (Supplementary Table S1). In these analyses, the total EQI and domain-specific EQI were stratified by RUCC categories rather than using the EQI and domains created specifically for each RUCC category in the original EQI composition, to combine the more rural categories with limited sample size.

Among North Carolina women diagnosed with breast cancer between 2009 and 2014, 7,975 were diagnosed with carcinoma in situ, 25,827 with localized breast cancer, 12,371 with regional breast cancer, and 2,073 with distant metastatic breast cancer (defined in Supplementary Table S1). A large portion of these patients with breast cancer inhabited metropolitan urbanized areas (75% carcinoma in situ, 74% localized, 72% regional, and 71% distant metastatic cases), with only 7%–9% of patients residing in less populated rural areas (Supplementary Table S2). In addition, there were “age at diagnosis” differences across rural–urban strata. Mean age at diagnosis was significantly higher in less populated areas for invasive breast cancers versus carcinomas in situ (e.g., 61.4 years carcinoma in situ, 62.8 years distant metastatic; P = 0.03); however, there was no difference in age at diagnosis identified in this dataset related to those residing in metropolitan areas (60.0 years carcinoma in situ, 60.1 years distant metastatic; P = 0.40).

We generated a map of EQI quartiles in ArcGIS, similar to what was presented in the original EQI report (40). Less populated counties generally had overall worse environmental quality, including when separated by domains (Fig. 1). For example, worst land environmental quality was highly concentrated in the eastern region of the state, which is also a less populated region (Fig. 1). As a trend supporting this geographic observation, breast cancer cases in counties with the first EQI quartile (best environmental quality) for total EQI and all EQI domains were underrepresented in the less populated rural–urban category (Supplementary Table S2). For example, only 7%–9% of breast cancer cases regardless of stage were in less populated counties with high overall environmental quality. Conversely, 33%–35% were in less populated counties with the worst overall environmental quality.

Age at diagnosis was significantly associated with all stages of breast cancer, with an increase of 1%–3% in invasive breast cancer odds for every 10-year age increment in localized and distant metastatic breast cancers (Table 1; P < 0.001). A higher percentage of later-stage breast cancer cases (24% regional and 29% distant metastatic) versus patients with carcinoma in situ (20%) were black (Supplementary Table S2), a pattern consistent across rural–urban categories. Black patients had particularly higher odds of regional and distant metastatic breast cancer (Table 1). The highest increase was for distant metastatic breast cancer, where black patients had a 9% increase in odds of having distant metastatic breast cancer versus carcinoma in situ, regardless of rural–urban area [nonstratified OR, 1.09; 95% confidence interval (CI), 1.07–1.12; P < 0.0001]. Those who were current smokers or were a former smoker as opposed to having never smoked had significantly increased odds of breast cancer regardless of stage or rural–urban area. This was most apparent for distant metastatic cases, where current smokers as compared with those who had never smoked had 15% increased odds of presenting with distant metastatic breast cancer versus carcinoma in situ (nonstratified OR, 1.15; 95% CI, 1.10–1.21; P < 0.0001). Finally, there was no association found between breast cancer versus carcinoma and BMI category, although those with an unknown BMI did have significantly different odds compared with normal weight persons, which differed by breast cancer stage.

Models were constructed to estimate odds of having localized, regional, or distant metastatic invasive breast cancer (compared with noninvasive carcinoma in situ) as a function of the EQI. ORs greater than one thus represent greater odds of having localized, regional, or distant metastatic breast cancer, whereas ORs less than one represent greater odds of having carcinoma in situ versus the comparison breast cancer stage. Our results show that environmental quality is associated with breast cancer differentially by breast cancer stage, environmental domain, and rural–urban area.

Taken together, we observed little evidence of association between invasive breast cancer and total environmental quality as measured by the EQI. However, poor total environmental quality was associated with a 5% increase in the odds of having both regional and distant metastatic breast cancer versus carcinoma in situ (OR 1.05; 95% CI, 1.01–1.09; P = 0.003) but in most cases, this association was seen in the third quartile (poor, but not the worst environmental quality; Table 1; Figs. 2–4).

The air environmental quality domain was comprised of criteria and hazardous air pollutants such as particulate matter, ozone, carbon monoxide, and numerous volatile compounds, among others. Air quality tended to be inversely associated with localized and regional breast cancer, with a suggestion of stronger effects in nonmetro urbanized and less populated areas. For example, the worst air quality was associated with 20% decreased odds of regional breast cancer versus carcinoma in situ in less populated regions (OR, 0.80; 95% CI, 0.72–0.88; P < 0.0001), and similar associations were seen for localized breast cancer (Figs. 2–4; Supplementary Table S3).

The water environmental quality domain was comprised of estimates of domestic, agricultural, and industrial water use; drought status; and water quality and contaminant levels in natural sources, precipitation monitors, and public water supplies. In general, water quality associations were null; however, we did see an association in metropolitan urbanized counties for localized and regional breast cancer (Figs. 2–4; Supplementary Table S3). We caution against overinterpretation of these results because they suggest opposite effects of water environmental quality.

The land environmental quality domain was comprised of agricultural information including crops, livestock, and pesticides used; toxic release and cleanup sites on the national priority list; geochemical data (such as arsenic and lead, among others); and areas with potentially elevated indoor radon levels (Supplementary Fig. S1). Land environmental quality was consistently associated with increased odds of localized, regional, and distant metastatic breast cancers compared with carcinoma in situ. This association was most consistent in metropolitan urbanized and less populated areas. Overall, patients residing in metropolitan urbanized counties with the worst land environmental quality had 3%–5% increased odds of having breast cancer versus carcinoma in situ (localized OR: 1.03, 95% CI: 1.01–1.06, P = 0.003; regional OR: 1.08, 95% CI: 1.03–1.12, P < 0.001; and distant metastatic OR: 1.05, 95% CI: 1.00–1.10, P = 0.04), while the risk increased to 5%–8% in less populated counties (localized OR: 1.05, 95% CI: 1.01–1.09, P = 0.01; regional OR: 1.06, 95% CI: 1.01–1.12, P = 0.012; and distant metastatic OR: 1.08, 95% CI: 1.01–1.16, P = 0.02; Supplementary Table S3).

The sociodemographic environmental quality domain is comprised of U.S. census population, housing, and economic data, as well as community and crime data. In this analysis of North Carolina counties, we observed associations between sociodemographic environmental quality and invasive breast cancer in unadjusted models. However, it should be noted that these associations, after accounting for individual demographic characteristics became statistically insignificant but remained comparable with unadjusted models (Supplementary Table S4). For example, the worst sociodemographic environmental quality increased odds of distant metastatic breast cancer by 10% in nonmetro urbanized counties (OR, 1.10; 95% CI, 1.00–1.20; P = 0.035), an association which was not significant in adjusted models (adjusted OR, 1.07; 95% CI, 0.98–1.16: P = 0.18), suggesting that this environmental domain may be at least partially confounded by individual factors like age, race, BMI, and smoking status that are used to adjust models.

Finally, we assessed the built environmental quality domain, which is comprised of commercial business information, roads, motor vehicle crash fatalities, low-rent and section-eight housing, and public transportation use. Our analysis suggests an inverse association with built environmental quality for regional breast cancer in nonmetro urbanized counties (OR, 0.93; 95% CI, 0.87–0.99; P = 0.04) and increased odds of distant metastatic breast cancer for patients residing in counties with the worst built environmental quality, particularly in metropolitan urban areas (P_trend = 0.03).

Counties across North Carolina had a wide range of EQI values, representing anywhere from 27% to 54% the range of EQI values across all counties in the United States, depending on the EQI domain (Supplementary Table S5).

Investigating incidence of breast cancers of all types in total has the potential to mask differences in the impact of environmental factors on the development of different subsets of breast cancer, particularly those influencing development of later stages or more aggressive subsets. Hormone receptor subtype information was available for only 68% of patients within the dataset, so summary staging as “localized,” regional,” and “distant” was chosen to differentiate patients as it allows for assessment of increasing invasiveness and severity. Patients with distant metastatic breast cancer have poor prognosis despite aggressive, multidisciplinary treatment regimens compared with carcinoma in situ or early-stage breast cancer (43). This reinforces the unmet need to identify risk factors associated with advanced breast cancers to reduce incidence and improve overall breast cancer survival.

This study demonstrates the strongest positive association for poor land environmental quality and distant metastatic breast cancer. In particular, patients residing in less populated counties with the worst land environmental quality were 8% more likely to have distant metastatic breast cancer than carcinoma in situ, an association which was greater in rural areas but also persisted in metropolitan areas. This association was also present for localized and regional breast cancer, with 5%–6% increased odds in worst land environmental quality rural areas. While 8% is a small increase, a large proportion of the population lives in these communities and at the population level, the impact of 8% is quite substantial. Most importantly, while our data evaluated EQI and breast cancer in the state of North Carolina, greater variation in land quality is present in other areas of the United States (Supplementary Table S5), thereby highlighting the broader applicability of this work. Moreover, our data indicate that the association between distant metastatic breast cancer and broad land environmental quality is dependent on rural–urban context, with the major effects occurring in less populated areas, which suggests the need to critically evaluate specific rural versus urban environmental factors. For example, hog farms and resultant toxic waste lagoons, which frequently are in rural areas associated with land EQI domains, have recently been studied in eastern North Carolina and found to have significant potential human health effects (44). This underscores the importance of our sensitivity analysis stratifying by rural–urban context and should be used in future studies considering associations between disease and environmental exposures.

A strength of our analysis lies in using North Carolina as a study area, as it has a range of population densities and diverse demographics and environmental conditions. In maps merging both urbanicity and environmental quality, less populated and/or more rural areas had consistently worse environmental quality across all environmental domains. This is of great interest as recent reports increasingly suggest the role of socioeconomic and environment as breast cancer risk factors (3–14, 18, 45). However, association between rural–urban location and breast cancer stage are controversial and less certain (19), and the association between breast cancer stages, urbanicity, and environmental factors has not been studied simultaneously. Our maps showing different patterns in urbanicity and environmental quality reinforce the idea that environmental risk factors may impact breast cancer incidence differentially in urban and rural environments. In addition, North Carolina has previously been used as a study area to spatially associate urbanicity with receipt of radiotherapy in Medicare-receiving patients with breast cancer (20) and to study incidence of higher stage basal-like breast cancer risk in premenopausal African-American patients (21).

In our unadjusted statistical models, patients residing in a county with the worst sociodemographic environment were 4% more likely to have regional or distant metastatic breast cancer than carcinoma in situ, which increased to 9%–10% in nonmetro urbanized areas. It has been noted previously that socioeconomic status has significant associations with breast cancer incidence, in both location- and stage-specific models (18, 25, 26). However, individual factors such as age, BMI, smoking status, and race accounted for this significance in adjusted models for distant metastatic breast cancer. Individual race as a covariate in adjusted models also had a significant association with distant metastatic breast cancer. Those self-identifying as Black had 9% greater odds of distant metastatic breast cancer versus carcinoma in situ regardless of rural–urban area. This has been seen in prior studies investigating race/ethnicity associations with spatial incidence and mortality of breast cancer, where non-Whites and more specifically non-Hispanic Blacks are consistently at higher risk for total as well as advanced- and late-stage breast cancer subtypes at both state and national levels of analysis (22–24, 27, 46, 47). In addition, epidemiologic studies report that racial and ethnic minorities, as well as those living in poverty, are exposed to higher levels of various environmental pollutants compared with other populations (48, 49). This suggests that, after accounting for individual race and other factors, sociodemographic characteristics of a county are less important in distant metastatic breast cancer and strengthen previous studies reporting both individual- and county-based heterogeneity in breast cancer incidence and outcomes (19, 25).

It is also important to note that our study identifies inverse effects for some environmental domains. This requires careful interpretation because the comparison is not between breast cancer and no breast cancer, but rather between invasive breast cancer and carcinoma in situ, that is, preinvasive breast cancer. In particular, poor air quality was inversely associated with localized and regional breast cancer in more rural areas, also signifying that it was associated with increased carcinoma in situ versus breast cancer. As a trend, this suggests that contaminants within the air domain that constitute air quality are associated with noninvasive breast cancer rather than any single stage of invasive breast cancer. More work is needed to further elucidate these associations; however, in a recent study, air contamination particulates PM2.5 and NO₂ were found to be associated with breast cancer overall and with ductal carcinoma in situ but not invasive breast cancer (50). Interestingly, associations were seen with invasive breast cancer in certain geographic regions, again indicating that rural–urban sensitivity analysis is paramount to these types of studies of breast cancer and environment associations, including investigations between different breast cancer stages.

Our results should be interpreted in the context of several important limitations. First, not all breast cancer cases may be reported to the NC CCR; however, all health care providers are required by law to report cases to the CCR, so this is not expected to skew results. In addition, exposure at diagnosis may not be the most etiologically relevant timepoint. To address this issue, we obtained breast cancer data approximately 10 years after the time period that the EQI is intended to assess. Nonetheless, women in our study population could have moved in the years preceding their diagnosis. While this could result in exposure misclassification, it is likely nondifferential with respect to cases status and would likely bias associations toward the null. Another limitation lies in use of carcinoma in situ as controls; ideally, we would have had noncancer cases for controls if these data were available. If environmental quality was related to both carcinoma in situ and distant metastatic breast cancer, using our method of control selection could mask important trends, as may be the case for our null results. However, it should be noted that this would likely obscure trends, not create them. We further did not perform adjustment for multiple comparisons for statistical significance, but instead chose to focus on comparing patterns and precision of estimates to better analyze trends and not overinterpret the significance of results. An important feature of the EQI is that it is available for the entire United States; thus, it is feasible to extend our analysis in other geographic areas with cancer registry data and further test the robustness of our findings. Finally, our analysis assigned environmental quality at the county level, which may hide smaller scale trends. Ongoing work is focusing on the acquisition of both patient and environmental data with more granular geographic information, to fully understand the influence of geographically distributed environmental factors on the incidence rates of late-stage invasive breast cancer.

This project provides insight into the association between environmental quality and different stages of invasive breast cancer versus noninvasive carcinoma in situ. We report significant positive association between all stages of breast cancer, particularly distant metastatic breast cancer and poor land environmental quality, highlighting the need for additional research. In addition, our work has implications for future epidemiologic studies investigating the influence of the environment on disease; our findings suggest that the EQI is a highly relevant measure for controlling for diverse environmental exposures in these studies.

No potential conflicts of interest were disclosed.

Conception and design: L.M. Gearhart-Serna, K. Hoffman, G.R. Devi

Development of methodology: K. Hoffman, G.R. Devi

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): L.M. Gearhart-Serna

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): L.M. Gearhart-Serna, K. Hoffman, G.R. Devi

Writing, review, and/or revision of the manuscript: L.M. Gearhart-Serna, K. Hoffman, G.R. Devi

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): K. Hoffman

Study supervision: K. Hoffman, G.R. Devi

This work was supported in part by developmental research funds to G.R. Devi and K. Hoffman from the Duke Cancer Institute (Cancer and Environment Program), Duke Environmental Health Scholars Award (predoctoral) to L.M. Gearhart-Serna, NIH under grant P20 CA202925-01A1, and the Department of Surgery Bolognesi award to G.R. Devi. The authors thank Dr. Kristen Rappazzo at Environment Protection Agency for providing expertise in the environmental quality index and for her insightful review of this article, and Brittany Mills for assistance during the Duke University Summer Research Opportunity Program. In addition, the authors thank Dr. Soundarya Radhakrishnan, Sohrab Ali, and Dr. Gary Leung at the North Carolina Central Cancer Registry for providing the data used in the analyses.