Synergistic Effects of Particle Radioactivity (Gross β Activity) and Particulate Matter ≤2.5 μm Aerodynamic Diameter on Cardiovascular Disease Mortality

Abstract

Background

Although the effects of fine particulate matter (particulate matter ≤2.5 μm aerodynamic diameter [PM_2.5]) on cardiovascular disease (CVD) morbidity and mortality are well established, little is known about the CVD health effects of particle radioactivity. In addition, there are still questions about which of the PM_2.5 physical, chemical, or biological properties are mostly responsible for its toxicity.

Methods and Results

We investigated the association between particle radioactivity, measured as gross β activity from highly resolved spatiotemporal predictions, and mortality for CVD, myocardial infarction, stroke, and all‐cause nonaccidental mortality in Massachusetts (2001–2015). Within both difference‐in‐differences model and generalized linear mixed model frameworks, we fit both single‐exposure and 2‐exposure models adjusting for PM_2.5 and examined the interaction between PM_2.5 and gross β activity. We found significant associations between gross β activity and PM_2.5 and each mortality cause. Using difference‐in‐differences and adjusting for PM_2.5, we found the highest associations with myocardial infarction (rate ratio, 1.16 [95% CI, 1.08–1.24]) and stroke (rate ratio, 1.11 [95% CI, 1.04–1.18]) for an interquartile range increase (0.055 millibecquerels per cubic meter) in gross β activity. We found a significant positive interaction between PM_2.5 and gross β activity, with higher associations between PM_2.5 and mortality at a higher level of gross β activity. We also observed that the associations varied across age groups. The results were comparable between the 2 statistical methods also with and without adjusting for PM_2.5.

Conclusions

This is the first study that, using highly spatiotemporal predictions of gross β‐activity, provides evidence that particle radioactivity increases CVD mortality and enhances PM_2.5 CVD mortality. Therefore, particle radioactivity can be an important property of PM_2.5 that must be further investigated. Addressing this important question can lead to cost‐effective air‐quality regulations.

Nonstandard Abbreviations and Acronyms

DID

difference‐in‐differences

EPA

Environmental Protection Agency

GLMM

generalized linear mixed model

mBq/m³

millibecquerels per cubic meter

PM_2.5

particulate matter ≤2.5 μm aerodynamic diameter

PR

particle radioactivity

RadNet

Radiation Network

TOT

all‐cause non‐accidental total deaths

ZCTA

ZIP code tabulation area

Clinical Perspective.

What Is New?

• This is the first study that provides evidence of the chronic effects of long‐term exposure to spatiotemporal predictions of gross βactivity, a radioactive property of fine particulate matter, and total cardiovascular, myocardial infarction, and stroke mortality.

• We found a positive interaction between particulate matter ≤2.5 μm aerodynamic diameter (PM_2.5) and gross β activity, with higher associations between PM_2.5 and mortality at a higher level of gross β activity.

• We also found that the effects varied between different age groups and were comparable using 2 statistical methods with different assumptions, with and without the adjustment for PM_2.5.

What Are the Clinical Implications?

• This is the first study to apply a causal inference method to demonstrate that gross β‐activity contributes significantly to cardiovascular disease mortality, especially for myocardial infarction and stroke, which is similar in magnitude to that of fine particulate matter.

• We also demonstrate that gross β‐activity, an important radioactive property of fine particulate matter, enhances PM_2.5 cardiovascular disease mortality.

• A better understanding of the physical, chemical, or biological properties of PM_2.5 responsible for its toxicity could lead to more targeted and cost‐effective air‐quality regulations.

Cardiovascular disease (CVD) is the most common cause of death in the United States. ¹ Despite the decline in CVD mortality in the late 20th century related to advances in health care and public health, CVD mortality is no longer falling. ² This may be attributed, at least partially, to environmental exposures. Air pollution is a well‐established risk factor for CVD morbidity and mortality. ³ , ⁴ , ⁵ Although many studies have linked CVD mortality to short‐ and long‐term exposures to ambient particulate matter, ⁴ , ⁶ , ⁷ few have investigated the cardiovascular health effects of particle radioactivity (PR).

PR is a radiometric characteristic of particulate matter that mainly reflects radon ⁸ gas through its progeny. The primary source of PR in the United States is radon gas, through its decay products. ⁹ Radon is a naturally occurring, radioactive, colorless, and odorless noble gas, and it is a product of the natural radioactive decay of uranium found in soil and rocks. Once radon is formed, it migrates upwards into the atmosphere. ²²² Radon, the most common radon isotope, decays to α, β, and γ radiation–emitting isotopes of elements such as polonium (²¹⁸Po, ²¹⁴Po, and ²¹⁰Po), and lead (²¹⁴Pb and ²¹⁰Pb), among others, which are referred to as radon progeny. ¹⁰ Radon progeny initially forms unattached respirable ultrafine clusters (ranging from 0.5 – 5 nm), which then rapidly attach to larger local airborne particles and make them radioactive. ¹¹ , ¹² Therefore, the fine particulate matter (particulate matter ≤2.5 μm aerodynamic diameter [PM_2.5]) can act as a PR vector, penetrate deeply into the lung, and enter the circulation. ¹³ , ¹⁴ These radioactive particles then continue to decay and emit α, β, and γ radiation into the bronchial passages and alveoli or become systemic and cause adverse effects. Because of its relatively large mass and 2 positive charges (2 protons and 2 neutrons), α radiation causes more biological damage than β radiation. ¹⁵

Focusing on the characteristics of gross β activity (a measurement of all particle‐bound ambient β activity representing the total PR) in our study area, in a previous study ¹⁶ we identified 6 PM_2.5 sources and found that regional pollution had the highest impact on both α and β activities, followed by motor vehicles. ¹⁶ In another study, ¹⁷ we investigated the temporal and spatial variability of gross β‐activity measured by 7 Environmental Protection Agency (EPA) Radiation Network (RadNet) monitors located in and around Massachusetts, and we found that gross β‐activity exhibited little variation across space. The variation in β activity was driven by seasonal trends and meteorology. In general, particle β‐activity has more temporal than spatial variability when compared with PM_2.5 in Massachusetts.

Many studies have comprehensively demonstrated the adverse health effects of high‐level radiation exposures from atomic bombs, nuclear accidents, and occupational hazards on both cancer and noncancer health outcomes. ¹⁸ , ¹⁹ Studies have highlighted the relationship between increased risks to cardiovascular health and high‐level radiation exposures, ²⁰ , ²¹ such as radiation therapy. ²² , ²³ Conflicting results were found among studies on the association between low levels of radon exposure and leukemia, ²⁴ cancer, ²⁵ , ²⁶ and mortality. ²⁷ , ²⁸ Therefore, the potential risks associated with exposure to low‐level internal radiation are less well understood. At lower doses, many inflammatory markers are upregulated after exposure to radiation. ²⁹ , ³⁰ There is also evidence that radon exposure causes oxidative damage; in rats, radon exposure resulted in a dose‐dependent increase in urinary 8‐hydroxy‐2′‐deoxyguanosine and reactive oxygen species in bronchoalveolar lavage fluid. ³¹ , ³² Similarly, PM_2.5 has the potential to exacerbate CVD through pulmonary and systemic oxidative stress and inflammation. ³³ , ³⁴ However, few studies have investigated low‐level radiation exposure's health effects, and no study has examined the effects of gross β‐activity on CVD mortality.

Our group was the first to examine the association between ambient PR and multiple health outcomes. We found significant associations between gross β‐activity and both diastolic and systolic blood pressure (BP), ³⁵ biomarkers of oxidative stress and inflammation among nonsmokers, ³⁶ and ventricular arrhythmias. ³⁷ We have also investigated the effects of PM_2.5 on daily total and CVD mortality and observed greater effects in cities with higher mean city‐level radon concentrations. ³⁸ These and other studies that used gross β‐activity as a surrogate for PR were based primarily on monitor‐based observations of PR from the US EPA RadNet. RadNet is made up of 200 fixed and mobile radiation monitors, which measure gross β activity approximately twice a week. ³⁹ Using RadNet data in epidemiological studies can lead to 2 limitations: first, linking subjects to the nearest RadNet monitor exposure data can introduce differential and nondifferential exposure misclassification; second, it can limit the generalizability of population‐based studies because of less coverage of rural areas. Thanks to a recently published spatiotemporal ensemble model from our group that predicted ambient gross β‐activity, ⁴⁰ we can, for the first time, investigate the adverse effect of long‐term exposure to PR on CVDs to address these gaps.

Specifically, in this article, we examined for the first time the link between PR, measured as gross β‐activity from these highly resolved spatiotemporal predictions, and mortality for CVD, myocardial infarction, stroke, and all‐cause nonaccidental mortality in Massachusetts from 2001 to 2015. We applied 2 statistical methods: a causal inference difference‐in‐differences (DID) approach, and a generalized linear mixed model approach. In both models, we fit both single‐exposure (gross β‐activity/PM_2.5) and 2‐exposure models (PM_2.5+gross β activity) and examined the interaction between the 2 exposures.

METHODS

Study Population and Health Outcomes

We obtained mortality data from the Massachusetts Department of Public Health for the years 2001 to 2015. The data include the coordinates of the residential address, date of death, place of death, and underlying cause of death based on the International Classification of Diseases, Tenth Revision (ICD‐10). We identified causes of death as follows: CVDs (ICD‐10: I00–I51), myocardial infarction (MI; ICD‐10: I21–I22), stroke (ICD‐10: I60–I69), and all‐cause nonaccidental total death (TOT; ICD‐10: A00‐R99). Because of the sensitive nature of the data collected for this study, requests to access the data set from qualified researchers trained in human subject confidentiality protocols may be sent to the Massachusetts Department of Public Health at https://www.mass.gov/orgs/department‐of‐public‐health. This study was approved by IRB‐20222 at Harvard University. Informed consent was not required because all subjects are deceased.

We excluded mortality of those aged <18 years, and we also stratified the population into different age groups: ≥18 years, 18 to 65 years, 65 to 85 years, and ≥85 years.

We converted the residential address coordinates to the corresponding ZIP code tabulation areas (ZCTAs), and we considered the ZCTA as the spatial unit of analysis, given that the confounders in our models are census variables that are tabulated at the ZCTA level. We then aggregated deaths for each of the above causes by year and by ZCTA. In the analysis by age groups, we calculated the annual death count for each mortality cause, for each age group, each ZCTA, and each calendar year.

This study was approved by the Human Subjects Committee at Harvard T.H. Chan School of Public Health and by the Massachusetts Department of Public Health.

Exposure: Gross β‐Activity and PM_{2 .5}

Levels of gross β‐activity are a good qualitative indicator of the total radiation activity for particles collected on air sampling. Therefore, in our study, we used gross β‐activity (a measurement of all particle‐bound ambient β‐activity) as a proxy to represent PR. Gross β‐activity data come from a published, spatiotemporal model. ⁴⁰ Briefly, the model provided monthly estimates of gross β activity across the contiguous United States with a spatial resolution of a 32‐km² grid applying a multistage ensemble‐based model. ⁴⁰ We obtained daily PM_2.5 predictions at the 1‐km² grid from a previously published and validated spatiotemporal model ⁴¹ based on satellite data, land‐use information, meteorological variables, and output from chemical transport model simulations. The model had a cross‐validated R ² of 0.86 for daily PM_2.5 predictions and 0.89 for annual predictions.

We aggregated the daily PM_2.5 data from the 1‐km² grid to ZCTA and the monthly gross β‐activity from the 32‐km² grid to ZCTA by averaging PM_2.5 and gross β‐activity over all grid cells that centroids located within each ZCTA. We then averaged predictions for each year to generate the annual averages.

Temperature and Other Potential Confounders

We obtained daily minimum and maximum temperature from Daymet ⁴² daily surface weather data available at the 1‐km² grid for North America. ⁴² To calculate seasonal temperature, we first aggregated minimum and maximum temperature data spatially from the 1‐km² grid to ZCTA, then we averaged the minimum and maximum temperature by season, and subsequently, we estimated the average seasonal temperature by averaging the minimum and maximum seasonal temperature. We defined the season as winter (December, January, and February) and summer (June, July, and August).

We collected information on socioeconomic status and general life habits. Specifically, we included (1) 2 county‐level variables from the Behavioral Risk Factor Surveillance System ⁴³ : average body mass index and smoking rate; and (2) 7 ZCTA‐level variables from the US Census Bureau 2000 and 2010 Census ⁴⁴ , ⁴⁵ : median household income, median value of housing units, proportion of residents below the poverty line, proportion of residents with a high‐school diploma, proportion of Hispanic residents, proportion of Black residents, and population density. We merged gross β activity, PM_2.5, seasonal temperatures, and other potential confounders into the aggregated count data set by ZCTA and calendar year.

Statistical Analysis

First, we summarized the distribution of PR and PM_2.5, death counts from different causes, and other covariates at the ZCTA level, and we performed Spearman correlation tests between variables. We applied a causal inference DID approach and, as secondary analysis, a generalized linear mixed model (GLMM) approach to investigate the link between gross β‐activity and cause‐specific mortality. We fitted single‐exposure and 2‐exposure models including PM_2.5 in the model. Furthermore, we examined the interaction between gross β activity and PM_2.5 and performed the analysis stratified by age group.

DID Approach

We fit a DID model that has been previously described. ⁴⁶ This is a causal inference method that by design addresses the potential for both measured and residual confounding. In the model, we included indicator variables for each ZCTA to control for geographic differences in measured or unmeasured covariates; indicator variables for each year to control for statewide time trend ²⁵ ; summer and winter average temperatures, which vary differentially over time and space and are correlated with PR; and an offset term defined as the natural log of the population in each ZCTA. This method approximates a random assignment of exposure across the population and therefore can estimate a causal effect if the appropriate assumptions are met. Specifically, we assumed that changes in outcome over time in each ZCTA that are not captured by the statewide trend are nondifferential concerning for exposure. ⁴⁶ The model is:

$$\begin{array}{c}\mathrm{lnE}\left({\mathrm{Y}}_{\mathrm{tz}}\right)={\mathrm{\beta }}_0+{\mathrm{\beta }}_{\mathrm{i}}{\mathrm{A}}_{\mathrm{tz}}+{\sum }_{\mathrm{z}}{\mathrm{\beta }}_{\mathrm{z}}{\mathrm{I}}_{\mathrm{z}}+{\sum }_{\mathrm{t}}{\mathrm{\beta }}_{\mathrm{t}}{\mathrm{I}}_{\mathrm{t}}+\mathrm{s}\left({\mathrm{Ts}}_{\mathrm{tz}},;,{\mathrm{\beta }}_{\mathrm{Ts}}\right)\hfill \\ +\mathrm{s}\left({\mathrm{Tw}}_{\mathrm{tz}},;,{\mathrm{\beta }}_{\mathrm{Tw}}\right)+\ln {\mathrm{ME}}_{\mathrm{tz}}\hfill \end{array}$$

where, $Y_{\mathit{tz}}$ is the death count in year $t$ at ZCTA $z$; $A_{\mathit{tz}}$ is the vector of exposures in year $t$ and ZCTA $z$, with $i$ representing the number of pollutants included in the model, with $i=1$ for a single exposure, and $i=1,2$ for the 2‐pollutant model; $I_z$ and $I_t$ are indicator variables for each ZCTA $z$ and year $t$, and ${\mathit{Ts}}_{\mathit{tz}}$ and ${\mathit{Tw}}_{\mathit{tz}}$ are the mean summer and winter temperatures, models as splines (function $s$); ${\mathit{ME}}_{\mathit{tz}}$ is the total population in year $t$ and ZCTA $z$ such that $\mathit{ln}{\mathit{ME}}_{\mathit{tz}}$ is the offset term corresponding to examining death rates.

Generalized Linear Mixed Models

As a secondary analysis, we applied GLMM with a random intercept for each ZCTA using a quasi‐Poisson assumption for the outcome to allow for overdispersion. The model is:

$$\mathrm{lnE}\left({\mathrm{Y}}_{\mathrm{tz}}\right)=\left({\mathrm{\beta }}_0+{\mathrm{\gamma }}_{\mathrm{z}}\right)+{\mathrm{\beta }}_{\mathrm{i}}{\mathrm{A}}_{\mathrm{tz}}+{\mathrm{\beta }}_{\mathrm{j}}{\mathrm{C}}_{\mathrm{tz}}+\ln {\mathrm{ME}}_{\mathrm{tz}}$$

where ${\mathrm{\gamma }}_z$ is the random intercept for each ZCTA; $A_{\mathit{tz}}$ is the vector of exposures; and $C_{\mathit{tz}}$ is the vector of all the potential $j$ confounders such as average seasonal temperatures, Census variables, Behavioral Risk Factor Surveillance System variables, and indicator variables for each calendar year to capture the time trend.

In both approaches, we used natural splines to determine whether the relationships between the outcome and temperatures/exposures were nonlinear. We also included an interaction between continuous gross β‐activity and PM_2.5 and computed the effect of PM_2.5 at the 10^th, 50^th, and 90^th percentiles of the gross β‐activity distribution, with the corresponding CIs.

We report all the results as rate ratios (RRs) for an interquartile range (IQR) increase in gross β‐activity (0.055 millibecquerels per cubic meter [mBq/m³]) and PM_2.5 (2.82 μg/m³), reported with 95% CIs. All statistical analyses were performed using R 3.5.1 (R Foundation for Statistical Computing, Vienna, Austria). ⁴⁷

RESULTS

Based on the census data, we studied 538 ZCTAs in Massachusetts from 2001 to 2015 with a total of 743 873 nonaccidental deaths. After the exclusion of missing data attributable to incomplete information on location, ICD codes, death date, and age, we had a total of 716 653 nonaccidental deaths in the final data set, among which 186 371 (26.0%), 36 692 (5.1%), and 39 069 (5.5%) deaths were attributable to CVD, MI, or stroke, respectively; and 129 043 (18.0%), 317 728 (44.3%), and 269 882 (37.6%) deaths were in the 18 to 65, 65 to 85, >85 years old age groups, respectively.

Table 1 summarizes the distribution of mortality, exposure, and other covariates. The variation was calculated across each year and each ZCTA. In general, the mean annual death rate per 1000 person‐years per ZCTA was 7.24 for TOT, 1.88 for CVD, and <1 for MI (0.37) and stroke (0.40). The annual average of gross β activity was 0.36 mBq/m³. The annual average of PM_2.5 was 8.30 μg/m³, which is below the annual US EPA National Ambient Air Quality Standard. ⁴⁸ There is no current standard for gross β activity. PM_2.5 concentrations were not much different for low versus high levels of gross β activity, as demonstrated in Table S1. The seasonal average temperature did not vary much across ZCTAs, with a summer average of 20.8 °C (SD, 1.0 °C) and a winter average of −1.9 °C (SD, 2.1 °C). Spearman correlation tests between variables are shown in Figure S1. The exposure‐response relationship estimated using a spline was linear (results not shown). We also summarized the distribution of mortality by age groups as shown in Table S2.

Table 1.

Distribution of Annual Cause‐Specific Death Counts and Rates for Ages 18+, Exposure, Summer and Winter Average Temperature, and Census Variables, Across Years and ZIP Code Tabulation Areas

	Mean	SD	10th percentile	25th percentile	Median	75th percentile	90th percentile
Death count, n
Cardiovascular disease	26.4	27.5	2	6	18	38	63
Myocardial infarction	5.2	6.3	0	1	3	7	13
Stroke	5.5	6.3	0	1	4	8	14
Nonaccidental causes	101.6	101.2	8	25	71	146	237
Death rate (per 1000 person‐years)
Cardiovascular disease	1.88	1.13	0.74	1.24	1.76	2.38	3.04
Myocardial infarction	0.37	0.41	0.00	0.14	0.31	0.51	0.77
Stroke	0.40	0.40	0.00	0.17	0.34	0.55	0.82
Nonaccidental all causes	7.24	3.24	3.84	5.22	6.93	8.81	10.80
Exposure
Gross β‐activity, mBq/m3	0.36	0.04	0.31	0.33	0.35	0.38	0.41
PM2.5, μg/m3	8.30	1.90	5.84	6.84	8.21	9.66	10.80
Covariates
Summer average temperature, °C	20.8	1.0	19.4	20.2	21.0	21.4	21.8
Winter average temperature, °C	−1.9	2.1	−4.6	−3.3	−1.9	−0.5	0.9
Body mass index	26.8	0.6	25.9	26.4	26.8	27.3	27.6
Smoking rate	0.50	0.04	0.45	0.47	0.49	0.52	0.56
Median household income (×$1000)	69.6	26.6	40.4	51.6	65.4	82.5	104.5
Median value of housing units (×$1000)	309.8	154.7	148.5	201.7	280.2	374.7	502.4
Proportion of residents below poverty line, %	8.1	6.8	1.9	4.0	6.5	10.3	16.0
Proportion of residents with a high school diploma, %	20.6	13.8	5.2	10.2	18	28.6	39.7
Proportion of Hispanic residents, %	5.8	10.4	0.7	1.2	2.3	5.4	14.5
Proportion of Black residents, %	3.8	8.4	0.1	0.5	1.3	3.3	8.8
Population density (per sq. mile)	2816.6	5108.2	84.7	274.0	765.5	2646.8	8281.5

mBq/m³ indicates millibecquerels per cubic meter.

Table 2 and Figure 1A and 1B show the results for gross β activity and PM_2.5 using both GLMM and the DID approach. The results using both methods were comparable. We found significant associations (P<0.05) between PM_2.5 and gross β‐activity and all the mortality causes examined.

Table 2.

Cause‐Specific (CVD; MI; TOT) RRs and 95% CIs for an Interquartile Range (Gross β‐Activity: 0.055 mBq/m³, PM_2.5: 2.82 μg/m³) Increase Using Both GLMM and DID Approaches for the 18+ Age Group With 3 Exposure Sets: Gross β‐Activity+PM_2.5, Gross β‐Activity Alone, and PM_2.5 Alone

Exposure	Cause of death	GLMM	DID
		RR (95% CI)	RR (95% CI)
PM2.5
PM2.5	CVD	1.08 (1.05–1.11)	1.11 (1.08–1.14)
	MI	1.06 (1.00–1.12)	1.04 (0.97–1.11)
	Stroke	1.08 (1.02–1.14)	1.09 (1.03–1.16)
	TOT	1.07 (1.05–1.08)	1.09 (1.07–1.11)
PM2.5+gross β‐activity	CVD	1.10 (1.07–1.13)	1.12 (1.09–1.16)
	MI	1.08 (1.02–1.14)	1.06 (0.99–1.13)
	Stroke	1.09 (1.04–1.16)	1.11 (1.04–1.18)
	TOT	1.07 (1.05–1.09)	1.10 (1.08–1.11)
Gross β‐activity
Gross β‐activity	CVD	1.04 (1.01–1.08)	1.05 (1.02–1.08)
	MI	1.09 (1.03–1.16)	1.15 (1.07–1.23)
	Stroke	1.05 (0.99–1.11)	1.09 (1.02–1.16)
	TOT	1.03 (1.01–1.04)	1.03 (1.01–1.04)
Gross β activity+PM2.5	CVD	1.07 (1.03–1.10)	1.07 (1.04–1.10)
	MI	1.11 (1.14–1.19)	1.16 (1.08–1.24)
	Stroke	1.07 (1.01–1.14)	1.11 (1.04–1.18)
	TOT	1.03 (1.01–1.05)	1.04 (1.02–1.06)

CVD indicates cardiovascular disease; DID, difference‐in differences; GLMM, generalized linear mixed model; mBq/m³, millibecquerels per cubic meter; MI, myocardial infarction; PM2.5, particulate matter ≤2.5 μm aerodynamic diameter; RR, rate ratio; and TOT, all‐cause nonaccidental, total deaths.

Figure 1. Cause‐specific rate ratios with 95% CIs for an interquartile range increase (gross β‐activity: 0.055 mBq/m³, PM_2.5: 2.82 μg/m³) in both gross β‐activity (subfigure A) and particulate matter ≤2.5 μm aerodynamic diameter (PM_2.5; subfigure B) using generalized linear mixed model (GLMM) and difference‐in‐differences (DID) approaches for the 18+ years age group with different exposure sets.

 

Using the causal inference DID approach, we found the highest effects of gross β‐activity on MI with an RR of 1.16 (95% CI, 1.08–1.24) and stroke with an RR of 1.11 (95% CI, 1.04–1.18) for an IQR increase (0.055 mBq/m³) after adjusting for PM_2.5. The results using GLMM were comparable with the DID results (Figure 1A). The results adjusted for PM_2.5 were similar or slightly higher compared with the single‐exposure model (Table 2).

Figure 1B presents the results for PM_2.5 from the models of PM_2.5 alone and PM_2.5 adjusting for gross β‐activity for the 18+ age group. Using the DID model, for an IQR increase of 2.82 μg/m³ in PM_2.5, the estimated RR in a model with gross β activity were 1.10 (95% CI, 1.08–1.11) for TOT, 1.12 (95% CI, 1.09–1.16) for CVD, 1.06 (95% CI, 0.99‐1.13) for MI, and 1.11 (95% CI, 1.04–1.18) for stroke. The results were similar when we did not adjust for gross β‐activity and when using GLMM; in addition, the effect sizes were also comparable with those found for gross β‐activity.

We found a significant positive interaction between PM_2.5 and gross β‐activity for each of the specific causes of mortality. Figure 2 presents the RR increase in mortality for an IQR increase in PM_2.5, calculated at the 10th, 50th, and 90th percentiles of the gross β‐activity distribution. The figure shows that the association between PM_2.5 and mortality is stronger at higher levels of gross β‐activity. Using the DID model, for an IQR increase in PM_2.5, the RR at the 90th percentile of gross β‐activity (0.41 mBq/m³) were 1.13 (95% CI, 1.11–1.15) for TOT, 1.16 (95% CI, 1.12–1.20) for CVD, 1.09 (95% CI, 1.01–1.16) for MI, and 1.14 (95% CI, 1.07–1.22) for stroke.

Figure 2. Cause‐specific rate ratios and 95% CIs for an interquartile range (2.82 μg/m³) increase in PM_2.5 at the 10th, 50th, and 90th percentile of the particulate matter gross βactivity distribution using both generalized linear mixed model and difference‐in‐differences approaches.

 

When stratified by age groups, as shown in Figure S2, the estimates for MI and stroke had large CIs because of the small number of deaths by age group. We found the strongest associations in the 18 to 65 years and 65 to 85 years age groups, although the CIs in the age group 18 to 65 years were large. We found that the estimated RR in 65 to 85 group controlling for PM_2.5 were 1.09 (95% CI, 1.04–1.14) for CVD, 1.21 (95% CI, 1.10–1.32) for MI, 1.08 (95% CI, 0.99–1.19) for stroke and 1.06 (95% CI, 1.03–1.08) for TOT.

Figure S3 presents the results for PM_2.5 from the 2‐exposure models for all ages and by age group using both approaches. When analyzing the relationship by age group (Figure S3), we found a different pattern compared with the results of gross β activity. For PM_2.5 we detected the strongest associations in the >85 years old age group for CVD, stroke, and nonaccidental mortality. There was no difference in the associations by age groups for MI mortality. The large CI for MI and stroke, mostly in the age group 18 to 65 years, can be explained by the low number of deaths in these age groups, which leads to lower power to detect effects in this subpopulation.

DISCUSSION

To our knowledge, this is the first study to analyze the chronic effects of long‐term exposure to PR and CVD mortality and the first to use fine spatiotemporal predictions of gross β‐activity. We found significant associations between gross β‐activity and each of the specific causes of death, with stronger effects for MI and stroke. We found a significant positive interaction between PM_2.5 and gross β‐activity, with higher associations between PM_2.5 and mortality at a higher level of gross β activity. We also observed that the associations varied across different age groups. The results were comparable when using 2 statistical methods with different assumptions, and with and without adjusting for PM_2.5, which provided insights into the PM_2.5 toxicity on CVD.

In the 2‐exposure models, including both PM_2.5 and gross β‐activity, we found that the association of gross β‐activity with mortality increased slightly compared with the single‐exposure model. Additionally, the magnitude of the effect estimates of PM_2.5 in the model with gross β‐activity was similar to those found for gross β‐activity. This suggests that gross β activity is independently associated with CVD mortality.

This is the first study to examine the association between gross β‐activity measured using fine spatiotemporal predictions and CVD mortality; therefore, we are unable to directly compare our results with other studies. In 2 previously published studies, we examined the association between radon and mortality, although these studies used a different exposure metric and different study designs compared with the one presented here. In one study, we examined the association between the logarithm of county‐averaged indoor radon and total mortality ⁴⁹ and found a 2.62% increase (95% CI, 2.52%–2.73%) in mortality per IQR of indoor radon, independent of PM_2.5 exposure. This is similar to our results since we obtained a 3% to 4% increase in total mortality from GLMM and DID methods, respectively. In another study ³⁸ that focused on the effect modification of indoor radon on PM_2.5‐associated daily total, cardiovascular, and respiratory mortality in 108 US cities, we found stronger mortality effects of PM_2.5 in cities with high average indoor radon concentrations.

In other studies, we examined the association of gross β‐activity measured from the RadNet monitoring network and individual markers of CVD, which may provide insight into the association with mortality reported here. In patients ³⁷ with implanted cardioverter‐defibrillators in Boston, we found that an IQR increase in the 2‐day moving average RadNet gross β‐activity was associated with 13% (95% CI, 1%–26%) higher odds of a ventricular arrhythmia event. We also examined the association between ambient gross β‐activity and BP among 852 men and found increases in diastolic and systolic BP, ³⁵ as 2.95 mm Hg increase in diastolic BP and 3.94 mm Hg increase in systolic BP, for an IQR increase in 28‐day exposure, independent of PM_2.5 exposure. ³⁵ Although these studies used different data and methods, the results from these studies showing an increase in markers of CVD are consistent with our results.

The cardiovascular effects of both gross β‐activity and PM_2.5 exposure can be compared with the effects produced by PM_2.5 in previous epidemiological studies. First, we found significant effects of PM_2.5 and each cause‐specific mortality (10% increase in TOT, 12% increase in CVD, 6% increase in MI, and 11% increase in stroke). These results are similar to the associations of gross β‐activity (4% increase in TOT, 7% increase in CVD, 16% increase in MI, and 11% increase in stroke). Both results are also comparable with what has been recently published in reviews and meta‐analyses. ⁵⁰ , ⁵¹ The 10% increase in total mortality for PM_2.5 is also similar to what was previously found using the same DID method ⁴⁶ and among Medicare beneficiaries in the United States. ⁵²

Regarding biological mechanisms, the majority of PR attaching to PM_2.5 penetrate deeply into the lung and can enter the circulation and subsequently lead to systemic inflammation and oxidative stress. Studies have shown that proinflammatory markers are upregulated after exposure to radiation. ²⁹ , ³⁰ Indeed, in a recent study, we reported associations between ambient gross β activity and biomarkers of oxidative stress and inflammation among nonsmokers from the Framingham Offspring and Third Generation cohorts in the northeastern United States. ³⁶ Therefore, the literature supports the theory that PR may contribute to CVD through systemic oxidative stress and inflammation, which is the same biological mechanisms by which PM_2.5 causes adverse health effects. ⁴ , ³³

An important finding of this study is that PR enhances PM_2.5 mortality. Although the effects of PM_2.5 have been well established, to date, there are still questions about which physical, chemical, or biological properties of PM_2.5 are mostly responsible for its toxicity. This is the first study to provide evidence that PR can be an important property of particulate matter that needs to be investigated. In a previous study, ³⁸ we found stronger cardiovascular mortality effects of PM_2.5 in cities with high average indoor radon concentrations, demonstrating that particle toxicity is higher in areas with higher radon levels, suggesting that it may be more effective to develop regional rather than national particulate matter air quality standards. Addressing this important question can have important implications for environmental and public health policy. When we stratified the analyses into age groups, the numbers of deaths from MI and stroke in each year and ZCTA were <1; therefore, the results were less stable and with wider CIs because of lack of power. However, we found the strongest associations in the 18 to 65 and 65 to 85 years age groups for gross β activity, and in the 85+ years age group for PM_2.5, although there were no significant differences between the age groups. More studies are needed to understand susceptibility by age.

Our study has several strengths. We provide new evidence on the association between exposure to low‐level particle radioactivity and CVD mortality. We used geocoded registry data and used modeled exposure data providing more accurate PR exposure at a finer resolution. Specifically, we used for the first time gross β activity from a spatiotemporal prediction model; therefore, we reduced the possible measurement error. Moreover, our confidence in the results was based on the validity of our statistical methods. We applied 2 different statistical approaches relying on different assumptions to examine the same associations. GLMM is a standard statistical method in observational epidemiology, which is more susceptible to confounding bias. ⁵³ Unmeasured or residual confounding may cast doubt on the validity of causal conclusions drawn from observational studies. To address this, we applied DID, a causal inference approach robust to omitted confounders across neighborhoods. The results using both methods were comparable; therefore, our results were robust to different methods and model adjustments.

Our study also had some limitations as did other epidemiological studies. Although our data were at the individual level, we aggregated the data at the ZCTA level. This allowed us to examine the effects of chronic air pollution on mortality using registry data instead of cohort data. Residual and unmeasured confounding may exist, but we used a DID approach to adjust for both observed and unobserved confounders by design.

Air pollution is a well‐established risk factor for CVD morbidity and mortality, and recently few studies have shown associations between PR and CVD markers. For the first time, we used fine spatiotemporal predictions of gross β activity and applied a causal inference method to demonstrate that gross β activity contributed significantly to CVD mortality, especially MI and stroke. The magnitude of the estimates for PM_2.5 was similar to those found for gross β activity, and the results were comparable with and without adjusting for PM_2.5. More importantly, we found that gross β activity enhances PM_2.5 mortality. A better understanding of the physical, chemical, or biological properties of PM_2.5 responsible for its toxicity could inform future public health and environmental policy worldwide.

Sources of Funding

This publication was made possible by US EPA grant RD‐835872, National Institutes of Health grant ES000002 and National institute of Aging grant AG066793‐01. Its contents are solely the responsibility of the grantee and do not necessarily represent the official views of the US EPA. Furthermore, the US EPA does not endorse the purchase of any commercial products or services mentioned in the publication.

Disclosures

None.

Supporting information

Tables S1–S2

Figures S1–S3