Association Between Particulate Matter 2.5 and Breast Cancer Mortality in California—A Place-Based Cross-Sectional Study

Abstract

To investigate the place-based association between BCM and air pollution in middle-aged (45–64) and older-aged women (65+) in California at the zip code level, secondary data were collected from the California Department of Public Health (CDPH) Data and Vital Statistics, CalEnviroScreen 4.0, and the American Community Survey (ACS) from the Census. Multiple linear regression was used to test the significance between air pollution and age-standardized BCM rates. The results indicate a significant association between PM_2.5 and age-standardized BCM rates for both the middle-aged and older-aged groups (β = 3.73, 95% CI [2.89, 4.58]; β = 5.33, 95% CI [2.75, 8.32], respectively). Furthermore, we found evidence of effect modification by the concentration of Hispanic women (β = −6.73, 95% CI [−9.37, −4.08]. This study provides evidence of a significant spatial association between PM_2.5 and BCM rates, which has policy implications for stricter air quality regulations and urban planning policies. Further research is needed to establish causality and the mechanism of action at the population level.

1. Introduction

In the United States, the chance of women developing breast cancer (BC) is 1 in 8 [1]. At the state level, California’s age-adjusted breast cancer mortality (BCM) rate is 18.8. The Healthy People 2030 objective for BCM rates is 15.3 per 100,000, and the current mortality rates in the United States and California are well above the desired level [2]. Prognostic and established risk factors for BC are known, such as age, race/ethnicity, lifestyle factors, tumor stage, treatments, comorbidities, genetic factors, and reproductive risk factors, and impact health outcomes; however, there are emerging factors associated with BC [3,4].

Air pollution is an emerging factor gaining recognition for its potential impact on the health and survival of BC patients [4,5]. One of the air pollutants is particulate matter 2.5 (PM_2.5), a mixture of liquid and solid particles sourced from the atmosphere, car emissions, or other direct sources with a diameter of 2.5 μm [6]. PM_2.5 is classified as a group 1 carcinogen by the International Agency for Research on Cancer (IARC) and the WHO [7,8]. It is well established that air pollution is linked to respiratory and cardiovascular disease with prolonged exposure; studies have investigated the potential association of air pollutants with BCM in the United States [9,10,11]. Using the Surveillance Epidemiology End Results (SEER) program and 255,128 cases of BC, researchers found a strong association between PM_2.5 and PM₁₀ and BCM with localized tumors showing a significant association [12]. The Nurses’ Health Study had 8936 BC cases over a 2-year average, and a significant association was only found between Stage 1 disease and PM_2.5 [13]. More lifestyle factors were included in the Multiethnic Cohort Study, which included 3089 BC cases and 474 BC deaths [4]. This study found a strong association between kriged NO₂ and BCM and a positive association between particulate matter (PM) and gaseous pollutants such as NO₂ among all racial and ethnic groups with BCM. Increased risk of BCM was positively associated with PM_2.5 and PM₁₀.

Despite this, research determining the relationship between air pollution and BCM is still in its early stages [4,9]. Although Californian studies found an association between air pollution and BCM, further research is necessary to compare air pollution levels in different counties and zip codes [4,8,12,14]. Previous studies have been conducted in Los Angeles, a highly polluted city with a unique particulate matter composition typical of large urban areas. However, no other studies have explored other locations with more diverse sources of particulate matter [4,12]. Although a cross-sectional study has looked at the social determinants of BCM and conducted a geospatial association, key criteria air pollutants have not been investigated to determine their association with BCM [15]. Unhealthy air quality places over 40% of individuals residing in U.S. counties in vast danger of adverse health outcomes [4]. Hence, it is crucial to consider whether geographical regions have fluctuating BCM rates due to air pollution [4,15]. Although these studies analyzed PM and NO₂, they did not analyze interactions with ground-level ozone, another air pollutant investigated in Canadian cohort studies [16]. Although there was no statistical significance between ozone and BCM, it is an air pollutant that has not been investigated in California [16].

Air pollution is a significant factor that should be investigated to ensure public health policies are updated to protect women at risk and lower BCM rates. Studies analyzing the association between air pollution and BCM are limited [4,12,13,17]. California has hazardous air pollution levels, with six of the ten most polluted cities present in the United States [4,17]. Hence, it is crucial to explore how it could contribute to BCM rates and add to the literature so that public health policies can be altered. This study aims to investigate the potential place-based association between BCM and air pollution, as any significant finding would shed light on how to protect women at risk of BCM through public health policies.

2. Materials and Methods

2.1. Study Design

A cross-sectional ecological study design that examined mortality data from 2016–2018 determined the relationship between age-standardized BCM rates and air pollution while controlling for sociodemographic factors using secondary data. The Institutional Review Board of California State University, Fresno, approved the research study (Protocol # 1965).

2.2. Variables

2.2.1. Outcome of Interest

The outcome variables in this study were age-standardized BCM rates for two age-stratified groups. The death counts by zip code and female population in each zip code from 2016–2018 were utilized to calculate age-standardized mortality rates (ASMRs) per 100,000 women by referencing the 2000 U.S. Standard Population. The outcome variable was stratified into two age groups: 45–64 (middle-aged) and 65+ (older-aged).

2.2.2. Exposure Variable

The exposure variable was PM_2.5. For PM_2.5, the annual mean concentrations from the average quarterly mean in μg/m³ were collected over the six years of 2012 to 2017 by estimating the ASMR mean concentration at the geographic center of the Census tract through ordinary kriging. Mean values of the air pollution variables were computed between CalEnviroScreen 3.0 and 4.0. The data were available at the zip code level and were a continuous variable transformed into z-scores for analysis.

2.2.3. Covariates

To assess the direct impact of air pollution, the following predictor variables were controlled: population density, race/ethnicity (African American and Hispanic), insurance coverage (public insurance), and poverty. Poverty ranged from 0% to 20%. Outliers and extreme outliers were eliminated based on the evaluation of box plots by poverty.

2.3. Data Sources/Measurements

The databases utilized in these data were the CDPH death data files, CalEnviroScreen 3.0 and 4.0, and the American Community Survey (ACS). The mortality data from 2016–2018 were collected from the California Department of Public Health’s (CDPH) Vital Source database. The database included the final cause of death (ICD-10 code), individual zip codes, years of residence, marital status, education level, comorbidities, years in the county, and age at death. The ICD-10 codes for BC (C500-C509) were filtered out to select the BCM cases, and ASMRs were calculated using population data from the American Community Survey. The ASMRs were continuous variables that were transformed into z-scores for analysis.

The air pollution data were collected from CalEnviroScreen 3.0 and 4.0, sourced from the California Air Resource Board (CARB). For the CalEnviroScreen 3.0 data, the years of air pollution data were from 2012–2014; for CalEnviroScreen 4.0, the years of data were from 2015–2017.

The covariates were collected from the Census for the 5-year estimates of 2018. Population density was sourced from the 2010 Census data from the FourFront website at the zip code level. It was recoded into a dichotomous variable by assigning 0 to rural and 1 to urban (70% percentile and above was assigned 1 for urban). The ACS was utilized to obtain zip code-level data of females over 45 for the percentage of African Americans, Hispanics, public insurance, and those below 200% of the poverty line in California. The sociodemographic variables of Hispanic, public insurance, and poverty were continuous and transformed into z-scores for analysis.

2.4. Study Size

This study focused on the female population who died of BC in 2016–2018. The 2016–2018 death data files from CDPH had 730,050 cases and 360,094 female cases. Out of the female population, 14,442 cases were of BCM. A total of 9763 women out of the 14,442 resided in the county for at least 15 years or more and were 45 or older. Events of mortality that occurred prior to age 45 were removed from the analysis to remove genetic biases and those who may be more likely to change their residence. After missing variables were omitted, 1007 zip codes were analyzed with completed mortality rates, air pollution, and sociodemographic data.

2.5. Statistical Methods

Univariate analysis was conducted for statistical tests to assess the frequencies and descriptive statistics of each continuous variable. Mean, standard deviation, and range were reported for the continuous variables of PM_2.5, population density, percent African American, percent Hispanic, public insurance, and poverty. In contrast, the number of cases and the percentage were identified for the categorical population density variable.

Bivariate analysis investigated the associations between BCM and the independent variables (air pollution) and covariates (population density, race, insurance, and poverty). A threshold of 0.2 for significance and an alpha value of 0.05 were utilized to determine significance.

For multivariate analysis, multiple linear regression was conducted to predict the impact of air pollution exposure on BCM for both the middle- and older age groups while controlling for covariates (i.e., population density, percent African American, Hispanic, public insurance, and poverty). Independent variables were entered stepwise into the model if they met the inclusion criteria of having a VIF of 3 or below. Goodness-of-fit tests were assessed to evaluate the final regression models. SPSS 28 was utilized to conduct the data analysis.

3. Results

3.1. Descriptive Data

Out of the 13,398 women who died of BC from 2016–2018, 9763 women were included in the analysis. Cases were dropped if the deceased did not reside in the county for at least 15 years prior to death.

Table 1. Frequency and percentage of women included in the study by demographic characteristic from 2016–2018 (N = 1007).

Category	N	%
Age
45–64	3416	35
65+	6347	65
Race/Ethnicity
Non-Hispanic White	6077	62.2
Non-Hispanic Black	834	8.5
Non-Hispanic Asian	936	9.6
Hispanic	1693	17.3
Other	224	2.4
Educational Level
Less than High School	1236	12.7
High School/GED	3085	31.6
Some College	2635	27.0
Bachelor’s	1816	18.6
Master’s or Higher	920	9.4
Unknown	71	0.7
Marital Status
Married	4075	41.7
Single	1004	10.3
Divorced/Widowed	4684	48.0

presents a detailed description of demographic variables. Regarding age, most females were over 65 (65%) and were white (62.2%). For educational level, 31.6% of women had completed high school/GED, and 18.6% completed a bachelor’s degree. Regarding marital status, 41.6% of women were married, and 48% were divorced/widowed.

3.2. Descriptive Statistics for Zip Code-Level Variables

Table 2. Descriptive statistics of zip code-level variables from 2016–2018 (N = 1007).

Variable	Mean/%	SD	Range
Air Pollution
PM2.5 (µg/m3)	9.78	1.70	5.05–13.66
Covariates
Population Density
Urban	34.70	-	-
Rural	65.30	-	-
% Public Insurance
45–64	22.61	13.55	0–73.28
65+	92.02	17.91	0–100
% African American Women
45–64	12.08	9.70	0–50.00
65+	-	-	-
% Hispanic Women
45–64	-	-	-
65+	3.87	4.89	0–34.37
% Women in Poverty
45–64	7.58	4.62	0–33.33
65+	7.10	5.70	0–50.00

Note. This table highlights the descriptives for zip code-level environmental and demographic variables. The mean is indicated for air pollution and covariates whereas the percentage is for all demographic variables.

describes the exposure variable and covariates. The average value of PM_2.5 across zip codes was 9.8 mg/m³. In total, 34.7% of the zip codes were urban, whereas 65.3% were rural, as classified by the population density. The percentage of women with public insurance was higher among women 65+ (92.0%) compared to middle-aged women (65.3%). The average percentage of middle-aged African American women residing in a zip code was 12.1%, whereas the average of older Hispanic women was 3.9%. The average percentage of women in poverty was relatively the same for both age groups (7.6%, 7.1%).

3.3. Association Between ASMR and Women 45–64 and 65+

Table 3. Multiple linear regression results for ASMR 45–64 and ASMR 65+ (N = 1007).

	ASMR 45–64		ASMR 65+
Variable	β	95% CI	β	95% CI
Intercept	16.0 ***	15.00, 16.96	62.0 ***	58.86, 65.09
Air Pollution
PM2.5	3.7 ***	2.89, 4.58	5.3 ***	2.75, 8.32
Covariates
Population Density	2.8 **	1.02, 4.49	6.3 *	0.99, 11.54
African American	2.3 ***	1.44, 3.06	-	-
Hispanic	-	-	9.4 ***	6.11, 12.78
Public Insurance	0.2	−0.57, 1.01	8.5 ***	6.25, 10.85
Poverty	1.0	−0.07, 2.05	2.6 *	0.29, 4.98
Hispanic × PM2.5	-	-	−6.7 ***	−9.37, −4.08

Note. Adjusted R² value: 0.14; 0.17. *** p < 0.001; ** p < 0.01; * p < 0.05.

shows the final multivariate analysis model for women aged 45–64 and 65+. The overall regression for women aged 45–64 was statistically significant (R² = 0.14, F (5,1007) = 34.95, p ≤ 0.001). PM_2.5 was identified as a significant predictor of the ASMR for women 45–64 (β = 3.73, 95% CI [2.89, 4.58]). For every unit increase in PM_2.5, there will be an expected increase in BCM by 3.7 per 100,000 in the population. The overall regression for women aged 65+ was also statistically significant (R² = 0.17, F (6,1007) = 34.36, p ≤ 0.001). In this analysis, PM_2.5, population density, and public insurance were the most significant predictors of the ASMR for women 65+ (β = 5.33, 95% CI [2.75, 8.32]); (β = 6.27, 95% CI [0.99, 11.54]); (β = 8.55, 95% CI [6.25, 10.85]). In the case of PM_2.5, for every unit increase in PM_2.5, there will be an expected increase in BCM by 5.3 per 100,000 in the population. Additionally, being Hispanic and living in poverty were also significant predictors (β = 9.44, 95% CI [6.11, 12.78]); (β = 2.63, 95% CI [0.29, 4.98]).

The interaction between Hispanic and PM_2.5 was significant (β = −6.73, 95% CI [−9.37, −4.08]); hence, binned variables were created to visualize this interaction. The Hispanic population variable was binned into three tertiles (33%), and its relationship with breast cancer mortality (BCM) rates was examined alongside levels of PM_2.5. The mean unstandardized predicted values of BCM rates were plotted against the binned Hispanic variable, with the colors representing levels of PM_2.5.

Figure 1 shows that BCM rates tend to increase with increasing levels of Hispanic concentrations in a community. Furthermore, the figure illustrates that BCM rates tend to positively increase with increasing levels of PM2.5 and the effect is attenuated among communities with high concentrations of Hispanic women. This indicates that across communities with a high concentration of Hispanic women, BCM rates are high despite varying levels of PM2.5.

4. Discussion

This study shows a significant place-based relationship between PM_2.5 and BCM. Moreover, we found evidence of effect modification, where the relationship between PM_2.5 and BCM is modified by the concentration of Hispanics in the zip code. The multiple regression analysis showed that PM_2.5 significantly predicted BCM for the women 45–64 (middle-aged) and 65+ (older age) groups. These results indicate that exposure to particulate matter increases the risk of BCM for middle-aged and older-aged groups, and a positive association exists. This finding is consistent with previous studies suggesting a plausible association between PM_2.5 and increased BCM rates [4,5,7,9]. Researchers found a moderately positive association between PM_2.5 and BCM rates among women with Stage I BC [9]. Similarly, it was found PM_2.5 was positively associated with BCM among all racial and ethnic groups [4]. However, this study goes beyond these studies by showing that PM_2.5 was significant across zip codes in California for middle-aged and older women instead of just at the county level or among a cohort. The deceased women that were analyzed in this study lived in the county for at least 15 years, and the analysis of the older age group showed the strongest association between PM_2.5 and BCM. This adds to the geospatial analysis of the impact of particulate matter on BCM at the community level and how the built environment, not just individual risk factors, impacts BCM. These findings have important implications for policies and interventions to reduce exposure to particulate matter and address environmental health disparities.

Effect modification was apparent in the relationship between PM_2.5 and BCM by the concentration of Hispanics in a zip code. The concentration of Hispanics in a zip code was positively associated with BCM. This finding contrasts with the Hispanic Paradox, suggesting that despite their socioeconomic disadvantage, Hispanics tend to demonstrate equal or better health outcomes than their white counterparts [18]. In alignment with our finding, a study in Texas found that foreign-born Hispanics tend to have higher rates of breast cancer mortality than U.S.-born Hispanics [18]. These findings are strengthened by a study analyzing metropolitan areas in the USA and Hispanic isolation, which showed an increase in poor health outcomes with increased Hispanic segregation along with limited access to healthcare [19,20]. Residential segregation is a potential explanation for the effect of the modification noticed. Hispanics are also less likely to receive adequate healthcare for breast cancer treatment as influenced by residential segregation, leading to higher mortality rates at the individual level; however, this finding was not supported in the multilevel models [19,20]. No effect modification was present for African American women; however, previous studies reported higher mortality rates due to residential isolation and access to healthcare, supporting our findings for this study [19,20].

For PM_2.5 levels, the EPA recommends levels below 12 µg/m³ to prevent the hazardous effects of PM_2.5. The average PM_2.5 levels in California from 2010–2017 were around 9.6 µg/m³, with areas in the Central Valley ranging over 12 µg/m³. Although California has strict air pollution regulations, these high levels of PM_2.5 in current areas are concerning and require extensive measures to help reduce air pollution levels. These measures could include stricter regulations on industrial emissions, promoting cleaner transportation options, and encouraging public awareness and education about the potential impact of air pollution on BC incidence and mortality. Urbanization goes hand in hand with stricter air pollution regulations since as cities become more densely populated, there is an increase in industrial activities. Urban planners and policymakers need to ensure that neighborhoods are within a safe distance from industries and encourage green space, so the effects of air pollution are not exacerbated by trapping air pollutants in areas with increased impervious surfaces.

Limitations

The limitations listed in this study are essential to consider while interpreting the results. Using secondary data from multiple sources, such as the Vital Source database, CalEnviroScreen 3.0 and 4.0, and the ACS, could result in potential reporting errors, incomplete data, or missing variables. Additionally, this was an ecological study, so several essential variables that could influence the results, such as tumor stage/subtype, genetic predispositions, estrogen receptor/progesterone receptor (ER/PR), family history, alcohol consumption, and smoking status, cannot be included as covariates. The listed risk factors are established at the individual level; however, emerging community-level risk factors of air pollution are the focus of this study and should be considered simultaneously with individual-level risk factors. The issue of misclassification can lead to biased results and closer to null estimates of effects. The potential misclassifications include exposure misclassifications with ambient air pollution exposure since most people spend time indoors and the short time span of PM_2.5 levels from the dataset may not directly represent the PM_2.5 levels that may have contributed to BCM. The latency period between exposure to a causative factor and the clinical appearance of breast cancer is another issue but beyond the scope of this study. Outcome misclassifications from the Vital Statistics Database and ICD-10 codes are another potential misclassification that could be missing cases of women who had BC but were not accurately identified as cases. Furthermore, the study design is cross-sectional and only includes community-level factors, so it cannot establish causality between the variables.

Mortality data from 2016–2018 were specifically selected to overlap with CalEnviroScreen data and to avoid the COVID-19 pandemic’s impact on air pollution due to reduced commuting, ensuring the accuracy of exposure estimates. Although variability exists in the timeframes of the datasets used, this is a recognized challenge in ecological studies. Annual averages of particulate matter concentrations tend to remain relatively stable over short periods, suggesting that the effect of timing mismatch is likely minimal. Nonetheless, this temporal variability may introduce a degree of bias and should be considered when interpreting the findings.

Future research could improve the findings of this study by addressing the limitations mentioned above. Researchers could collect primary data with a larger sample size, including individual-level data and community-level factors. Using primary data with the proper funding sources could help reduce the issue of misclassification and increase the accuracy of the results through a prospective or retrospective cohort study. Researchers can also include additional individual-level variables such as tumor stage, ER/PR hormone receptor and other biomarkers, family history, alcohol consumption, and smoking status to control for any confounding variables. Future studies could also use a multilevel analysis approach that combines individual-level and community-level factors to help understand the complex relationships between the variables.

5. Conclusions

This is the first study to assess place-based associations between age-standardized BCM rates and PM_2.5. This was a geospatial analysis to determine the impact of zip code-level determinants on BCM, including air pollution and sociodemographic factors. While many studies have analyzed social determinants of BCM, previous studies were limited to certain metropolitan areas or were performed at the county level. This research offers a unique insight into a geospatial relationship between BCM and PM_2.5 in California. However, it is important to note that this study’s associations are based on an ecological design, which limits causal inferences at the individual level, so key individual-level risk factors could not be considered as confounders. Future research with a multilevel analysis approach in a cohort study is crucial to understanding the relationship between air pollution and breast cancer mortality.