Authors: Jiajun Luo, Andrew Craver, Zhihao Jin, Liang Zheng, Karen Kim, Tamar Polonsky, Christopher O. Olopade, Jayant M. Pinto, Habibul Ahsan, Briseis Aschebrook-Kilfoy
Categories: Original Investigation, Research, Online Only, Public Health
Source: JAMA Network Open
Compared with individual-level factors, do contextual disadvantages modify the risk of fine particulate matter (PM2.5) exposure with cardiovascular health?
In this cohort study, using data on 210 554 individuals from the All of Us Research Program, disparities in associations of PM2.5 exposure with myocardial infarction and stroke were most pronounced between subpopulations characterized by contextual deprivation. The disparities were less with individual race and ethnicity or income.
While individual race and ethnicity and income remain crucial when considering the adverse association between PM2.5 and cardiovascular outcomes, the findings of this study suggest that contextual deprivation is a more robust socioeconomic characteristic modifying the association between cardiovascular health and PM2.5.
A large body of evidence supports the association between exposure to air pollution, especially fine particulate matter (PM2.5), and cardiovascular disease (CVD).^1,2,3,4^ The PM2.5 indicator refers to airborne particles with an aerodynamic diameter of less than 2.5 μm. Due to their small size, these particles can penetrate deep into the respiratory tract, posing substantial health risks. To protect the general population, both the US Environmental Protection Agency and World Health Organization set guidelines for ambient PM2.5 exposure.^5,6^ As a result of these efforts, a decreasing trend in the average ambient PM2.5 level has been observed across the contiguous US over the past decades, roughly from 15 μg/m^3^ in 2000 to 8 μg/m^3^ in 2016.^7,8,9^ Despite the improvement in air quality, exposure disparities persist, with the socioeconomically disadvantaged subpopulations remaining the most exposed over time.^7,9,10,11^
Within this context, several recent studies have assessed whether health benefits of the decreasing PM2.5 level are distributed equitably across different subpopulations in the US.^12,13,14,15^ Race and ethnicity is the most frequently investigated characteristic, as studies consistently observed that the non-Hispanic Black population bore the highest health burdens associated with PM2.5 exposure, reflected as a higher PM2.5-attributable mortality risk.^12,13^ Research on the role of income also concluded that low-income individuals would benefit more from lower PM2.5 levels compared with high-income individuals.^13^
While individual characteristics, such as race and ethnicity and income, have been identified as key factors affecting vulnerability to PM2.5, there is a need to prioritize our understanding of contextual deprivation when examining PM2.5 exposure and its health implications.^16,17^ Contextual deprivation encompasses the collective challenges faced by a community or population, including, but not limited to, poverty levels, quality of housing, and employment opportunities.^16,17^ In 2008, the US National Institutes of Health proposed a multilevel conceptual model with an emphasis on contextual deprivation as a guideline for future investigation.^18^ A prior study suggested that individual socioeconomic advantages were not sufficient to protect individuals against the adverse influence of contextual deprivation on health.^19^ Focusing on individual characteristics alone can inadvertently overlook the intertwined nature of individuals with their surrounding context,^20,21,22^ thus preventing us from adopting a more holistic perspective to address the root causes of health disparities.
To better assess the PM2.5-related health disparities in the US, we analyzed the electronic health record (EHR) data of more than 210 000 participants aged 36 years or older in the All of Us Research Program.^23^ Our analysis examined ambient PM2.5 exposure and incident CVD. Specifically, we considered 2 CVD emergencies in this myocardial infarction (MI) and stroke, because of their substantial role in CVD mortality. We estimated their nonlinear exposure-response curves in association with PM2.5 exposure. We stratified the study population by race and ethnicity, household income, and contextual deprivation, aiming to understand the outcomes associated with individual and contextual socioeconomic characteristics.
Our study adhered to the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) reporting guideline for cohort studies. The All of Us Research Program is a prospective cohort that currently includes more than 544 000 adults living in the US and its territories initiated in 2017. The goals, recruitment methods and sites, and scientific rationale for All of Us have been described.^23^ All of Us data include participants’ responses to a series of questionnaires, physical measurements collected by study staff at enrollment, and information from participants’ EHRs. The data are made available to researchers via the Researcher Workbench. The study was approved and overseen by the All of Us institutional review board. Informed consent was waived because of the use of deidentified archival data. This study used the All of Us data at the controlled tier released as of February 15, 2023. The present study used data from calendar years 2016 to 2022, and statistical analysis was performed from September 25, 2023, through February 23, 2024 We restricted the study population to those with valid EHR and residential data. We restricted our analysis to participants aged 36 years or older because of the low risk of MI or stroke among the younger population.^24^
Electronic health record–derived diagnoses were determined using Systematized Nomenclature of Medicine–Clinical Terms codes and mapped to Observational Health and Medicines Outcomes Partnership concept ID by the All of Us Data and Research Center. The full list of concepts used to determine MI and stroke can be found in the eMethods 1 and eMethods 2 in Supplement 1.
We identified primary diagnoses or conditions of MI and stroke after enrollment (ie, incident MI and stroke) from EHRs. For participants with incident MI and/or stroke, the follow-up time was calculated as the difference between enrollment and the initial diagnosis. For participants without an outcome of interest, the follow-up time was calculated as the difference between enrollment and December 31, 2022, or death, whichever came earlier. We also retrieved atherosclerotic CVD history, hypertension status, and type 2 diabetes status from EHRs. Hypertension was defined as 2 or more hypertension diagnoses and/or description and at least 1 hypertension medication prescription in EHRs.^25^
All of Us participants answered the Basics and Lifestyle questionnaires when they were enrolled. The Basics questionnaire elicits demographic information including age, race and ethnicity, educational level, marital status, household income, and geography. The Lifestyle questionnaire collects data on the use of tobacco and alcohol.^26^
Based on these survey questionnaires, we retrieved data on age, gender (male, female, and other), self-reported race and ethnicity (Hispanic, non-Hispanic Black [hereinafter, Black], non-Hispanic White [hereinafter, White], and Other [American Indian, Asian, >1 race and ethnicity, and groups in addition to those listed]), household income (<35 000-<50 000-75 000-150 000 per year), smoking status (never, former, and current), and health insurance status (no or yes). Body mass index was calculated as weight in kilograms divided by height in meters squared, using the height and weight measured at enrollment and grouped into underweight (<18.5), normal weight (18.5-25), overweight (25-30), and obese (>30) categories.
The PM2.5 exposure data were obtained from the Atmospheric Composition Analysis Group at Washington University at St Louis. Annual surface PM2.5 levels were estimated using a satellite-derived model.^27,28,29^ The resultant values showed great cross-validated agreement (R^2^ = 0.99). The datasets at a 0.01° × 0.01° scale (approximately 1.1 km^2^) for surface PM2.5 levels were used in this study.
The All of Us data contain the 3-digit residential zip code for each participant as of the 2023 release due to privacy concerns. We therefore averaged PM2.5 levels during the 5 years preceding the end of follow-up on all 0.01° × 0.01° grids within the 3-digit zip code to address the temporal variability of PM2.5. This approach captures the cumulative exposure effect, offering a comprehensive assessment of long-term exposures and commonly used in prior studies.^1,2,3^ eFigure 1 in Supplement 1 shows the distribution of All of Us participants.
Contextual deprivation was represented using the deprivation index in this study. The deprivation index is a composite score based on 6 different socioeconomic variables at the community level, including poverty prevalence, household income, educational level, insurance prevalence, reliance on public assistance, and vacant house prevalence.^30^ The deprivation index was normalized to a range from 0 to 1, with a higher index indicating more deprivation. In this study, we stratified the study population into 2 groups based on the less deprived (lower than the median) and more deprived (higher than the median).
To estimate the nonlinear exposure-response curve, we fit stratified Cox proportional hazards regression models with penalized splines for PM2.5 exposure to estimate the pointwise hazard ratios (HRs) and corresponding 95% CIs. The stratified terms included age at enrollment, race and ethnicity, sex, household income, and contextual deprivation. The models were also adjusted for potential confounders, including body mass index, health insurance status, smoking status, hypertension status, atherosclerotic CVD history, type 2 diabetes, penalized splines for the average temperature over the same period,^31^ and average nitrogen dioxide concentration between 2017 and 2019, the latest data for other air pollutant we could find during the study period.^32^ A full description of the model can be found in eMethods 3 in Supplement 1. Missing data were imputed using the random forest imputation algorithm.^33^ We used 6 μg/m^3^ as the reference value and present the results for exposure level from 6 to 12 μg/m^3^.
We ran the regression models in the subpopulations of race and ethnicity (Black vs White), household income (<50 000 per year), and contextual deprivation (less deprived vs more deprived). For race and ethnicity, we present the comparison between the Black and White cohorts. The exposure-response results for the Hispanic population are available in eFigure 4 in Supplement 1.
To statistically compare the exposure-estimate curves between subpopulations, we calculated the pointwise ratio of HRs (RHR) and corresponding 95% CI for each socioeconomic characteristic (ie, race and ethnicity, household income, and contextual deprivation). The RHR was calculated as the HR in disadvantaged groups divided by the HR in advantaged groups. The 95% CI was generated using bootstrapping.
We conducted several sensitivity analyses to examine the robustness of our (1) using exposure 3 years preceding the end of follow-up, (2) excluding patients with a history of MI or stroke, (3) excluding individuals living at the current address for less than 3 years, and (4) specifying the degree of freedom as 3 and placing knots at the tertiles for the penalized splines for PM2.5. The statistical analysis was performed using the survival package in R, version 4.4.0 (R Foundation for Statistical Computing).
The study population included 210 554 participants across the US and 690 311 person-years through December 31, 2022 (Table; eFigure 1 in Supplement 1). Among them, 40% were older than 60 years, 59.4% female, 38.4% male, 19.4% Black, 16.7% Hispanic, and 56.1% White. A total of 954 incident MIs (incidence rate, 0.14) and 1407 incident strokes (incidence rate, 0.20) were identified from the EHRs. The incidence rates of MI and stroke were higher in the less-deprived group than the more-deprived group (MI: 0.16 vs 0.12 per 100 person-years; 0.23 vs 0.18 per 100 person-years). In comparison, the incidence rates were higher in the disadvantaged groups of race and ethnicity and household income.
The mean (SD) 5-year PM2.5 exposure level during the study period was 7.7 (1.4) μg/m^3^ across the overall study population (Figure 1; eTable 1 in Supplement 1). White participants were exposed to lower PM2.5 levels than Black participants (mean [SD], 7.5 [1.3] vs 8.2 [1.1] μg/m^3^). The exposure levels were similar between household income and contextual deprivation groups. Crude Kaplan-Meier analysis showed that the group of both higher PM2.5 exposure (higher than the median) and socioeconomic disadvantages had the greatest risk for MI and stroke (eFigure 2 in Supplement 1).
Figure 1. Distributions of 5-Year Mean Fine Particulate Matter (PM
2.5) Exposure Levels in All of Us Participants
When the PM2.5 exposure level increased from 6 to 12 μg/m^3^, we observed increasing risks of incident MI and stroke along the continuum (eFigure 3 in Supplement 1). When stratified by socioeconomic characteristics, we observed differing exposure-response curves between the subpopulations. Overall, disadvantaged groups (ie, more deprived, household income<$50 000 per year, and Black) showed steeper exposure-response curves compared with advantaged groups, suggesting that the disadvantaged groups were more vulnerable to the detrimental association of high PM2.5 exposures (Figure 2 and Figure 3; original values are presented in eTable 2 and eTable 3 in Supplement 1).
Figure 2. Exposure-Response Curves for the Association Between Fine Particulate Matter (PM
2.5) Exposure and Incident Myocardial Infarction
Figure 3. Exposure-Response Curves for the Association Between Fine Particulate Matter (PM
2.5) Exposure and Incident Stroke
For MI (Figure 2), increasing exposure levels from 6 to 10 μg/m^3^ was associated with an HR of 1.20 (95% CI, 0.90-1.60) in the less-deprived group vs 3.39 (95% CI, 2.51-4.56) in the more-deprived group, corresponding to an RHR of 2.83 (95% CI, 1.93-4.13) (Figure 4; original values are presented in eTable 4 in Supplement 1). When the exposure level was further increased, the RHR for contextual deprivation was strengthened to 4.75 (95% CI, 2.69-8.39) for 12 μg/m^3^. A similar pattern was observed in subpopulations defined by household income and racial and ethnic groups. Specifically, the RHR for 12 μg/m^3^ was 4.18 (95% CI, 2.20-7.92) for household income groups and 3.33 (95% CI, 1.36-8.12) for racial and ethnic groups.
Figure 4. Ratio of Hazard Ratios for Myocardial Infarction or Stroke Between Different Subpopulations
We observed similar increasing risks for stroke that the disparities increased in subpopulations along with PM2.5 exposure increase (Figure 3). Briefly, with increasing PM2.5 from 6 to 10 μg/m^3^, the HR for stroke was 1.13 (95% CI, 0.85-1.51) in the less-deprived vs 2.57 (95% CI, 2.06-3.21) in the more-deprived cohort; 1.46 (95% CI, 1.07-2.01) in the 50 000 per year cohort; and 1.70 (95% CI, 1.35-2.16) in White individuals vs 2.76 (95% CI, 1.89-4.02) in Black individuals. The RHR was highest for contextual deprivation (2.27; 95% CI, 1.59-3.24), compared with income (1.55; 95% CI, 1.05-2.29) and race and ethnicity (1.62; 95% CI, 1.02-2.58). Increasing the exposure level from 6 to 12 μg/m^3^ corresponded to RHRs of 3.42 (95% CI, 2.00-5.83) for contextual deprivation groups, 1.93 (95% CI, 1.07-3.46) for household income groups, and 2.06 (95% CI, 1.02-4.14) for racial and ethnic groups. Results for Hispanic people can be found in eFigure 4 in Supplement 1.
In summary, we observed disparities in risks associated with PM2.5 exposure in all subpopulations defined by the 3 socioeconomic characteristics for both MI and stroke. The difference was most evident between subpopulations characterized by contextual deprivation instead of race and ethnicity or household income. All sensitivity analyses corroborated that the difference in MI and stroke risks was most pronounced between subpopulations characterized by contextual deprivation (eFigures 5-9 in Supplement 1).
Findings from this study add evidence that lower PM2.5 exposure levels would benefit all US residents, as well as across subpopulations defined by different socioeconomic characteristics. Moreover, the results align with prior studies^12,13,14,15^ concluding that socioeconomically disadvantaged groups are more vulnerable to high PM2.5 exposure in the US. The present study further illustrates that contextual deprivation, rather than individual race and ethnicity or household income, shows the strongest potential for modifying the association of PM2.5 exposure, as disparities in CVD risks are most pronounced between subpopulations defined by contextual deprivation. Our findings suggest a paradigm shift may be warranted, one that pivots from an emphasis on individual-level factors to the wider lens of contextual deprivation. This shift would involve designing policies that not only regulate emissions and reduce ambient pollution levels, but also enhance community resilience and access to clean environments, particularly in underprivileged areas, rather than a focus on individuals.
Based on studies from different disciplines over decades, it has been widely accepted that social structural factors, rather than innate biologic differences, are the primary factor for greater susceptibility to PM2.5 exposure.^13,34^ As structural racism stands out as a major cause of health disparities in the US,^35,36^ a large research effort has been devoted to dissecting the nuances of how individual race and ethnicity is associated with environmental health disparities. This focus on individual factors, while yielding critical insights into disparities, may inadvertently overshadow the broader and potentially more impactful role of contextual deprivation. While individual race and ethnicity and income are undeniably crucial factors, to unearth and tackle inequalities more effectively, it is essential to consider the broader concept of contextual deprivation, which indicates the state of disadvantage that arises from the broader social and environmental conditions in which individuals live.^37^ But we should also be reminded that contextual deprivation and individual characteristics are closely interlinked, creating a feedback loop where race and ethnicity can influence the level of contextual deprivation a person experiences, which in turn can affect their health outcomes. This association highlights the need for public health interventions to tackle the broader systemic issues more comprehensively.^38^
This study is the one of the first environmental studies from the All of Us Research Program. The present study reflects the latest changes and exposures, providing a more current snapshot of the association between PM2.5 exposure and CVD risks and how socioeconomic factors modify the association. Moreover, our research was strengthened by the rich longitudinal nature of the All of Us Research Program, capturing a diverse, population-wide spectrum of participants with ongoing EHRs. This allows a more detailed examination of incident CVD. Another strength of this study is that we are able to include adult participants of all ages. Previous studies have often been constrained by the reliance on ecologic data^12^ or datasets limited to populations aged 65 years or older.^13^ Such constraints potentially skew the understanding of PM2.5 impact to reflect predominantly the health outcomes of older adults or render studies prone to ecologic fallacy.
This study has limitations. First, the incidence rates of MI and stroke were 0.14 and 0.20 per 100 person-years in this study. In comparison, the incidence rate reported by the US Centers for Disease Control and Prevention was roughly 0.24 per 100 person-years for both MI and stroke in the general population.^39,40^ However, underdiagnosis appears to be nondifferential across the subpopulations defined by different socioeconomic characteristics. For example, studies reported that the risk of stroke was approximately 1.5 times as high for non-Hispanic Black people as for non-Hispanic White people,^41,42^ while our data also reported a 1.5 times risk of stroke for Black compared with White individuals. Because the present study focuses on the disparities between subpopulations, it is reasonable to conclude that the nondifferential overestimate of MI and stroke should have minimal influence on the observed disparities. Second, the All of Us Research Program has not been linked to the National Death Index yet. Mortality status in All of Us is currently reported by each health provider organization. Therefore, it is possible that some deaths, either from CVD or other causes, were not recorded. However, loss of this information should be nondifferential, because missing information on mortality is not a result of PM2.5 exposure. The nondifferential factor would bias the estimates toward the null. Third, the PM2.5 exposure was assigned at the 3-digit zip code level, which may not precisely reflect individuals’ exposure levels and thus cause bias. However, using a less fine spatial resolution typically leads to an underestimation of the true effect.^43^ In practice, findings from epidemiologic studies are relatively robust against the exposure error.^43^ Therefore, the exposure at the 3-digit zip code level in this study, although not optimal, still provided crucial evidence for the disparities between subpopulations. Fourth, data on some crucial covariates, including blood cholesterol level, medication prescriptions, physical activity, and other air pollutants during the study period, were not available or only available for a subset of the participants in the recent release. The shorter follow-up in All of Us may also underestimate the risk for MI and stroke in this study. However, our sensitivity analyses suggested that the results are robust in different scenarios.
While individual race and ethnicity and income remain important socioeconomic factors modifying the association of PM2.5 with CVD risks, the findings of this cohort study advocate for a broader approach that emphasizes the influence of contextual deprivation. Future research should incorporate contextual deprivation to develop a more comprehensive understanding of disparities in environmental exposure and health.