Low-Concentration Air Pollution and Mortality in American Older Adults: A National Cohort Analysis (2001–2017)

Abstract

Mounting epidemiological evidence has documented the associations between long-term exposure to multiple air pollutants and increased mortality. There is a pressing need to determine whether risks persist at low concentrations including below current national standards. Air pollution levels have decreased in the United States, and better understanding of the health effects of low-level air pollution is essential for the amendment of National Ambient Air Quality Standards (NAAQS). A nationwide, population-based, open cohort study was conducted to estimate the association between long-term exposure to low-level PM_2.5, NO₂, O₃, and all-cause mortality. The study population included all Medicare enrollees (ages 65 years or older) in the contiguous U.S. from 2001 to 2017. We further defined three low-exposure subcohorts comprised of Medicare enrollees who were always exposed to low-level PM_2.5 (annual mean ≤12μg/m³), NO₂ (annual mean ≤53-ppb), and O₃ (warm-season mean ≤50-ppb), respectively, over the study period. Of the 68.7-million Medicare enrollees, 33.1% (22.8-million, mean age 75.9 years), 93.8% (64.5-million, mean age 76.2 years), and 65.0% (44.7-million, mean age 75.6 years) were always exposed to low-level annual PM_2.5, annual NO₂, and warm-season O₃ over the study period, respectively. Among the low-exposure cohorts, a 10-μg/m³ increase in PM_2.5, 10-ppb increase in NO₂, and 10-ppb increase in warm-season O₃, were, respectively, associated with an increase in mortality rate ranging between 10 and 13%, 2 and 4%, and 12 and 14% in single-pollutant models, and between 6 and 8%, 1 and 3%, and 9 and 11% in tripollutant models, using three statistical approaches. There was strong evidence of linearity in concentration–response relationships for PM_2.5 and NO₂ at levels below the current NAAQS, suggesting that no safe threshold exists for health-harmful pollution levels. For O₃, the concentration–response relationship shows an increasingly positive association at levels above 40-ppb. In conclusion, exposure to low levels of PM_2.5, NO₂, and warm-season O₃ was associated with an increased risk of all-cause mortality.

Graphical Abstract

INTRODUCTION

Air pollution is a major risk to global health.¹ It is well established that exposure to ambient air pollution is associated with increased mortality.² Past epidemiological evidence has well-documented the associations between long-term exposure to fine particulate matter (particles with a mass aerodynamic diameter below 2.5 μm; PM_2.5) and reduced life expectancy among adults.^3–7 More recently, studies have investigated the link between long-term exposure to multiple pollutants, such as nitrogen dioxide (NO₂) and ozone (O₃), and premature mortality.^8–13 However, few studies have simultaneously assessed the impact of PM_2.5, NO₂, and O₃, and even fewer have characterized the association of long-term exposure to air pollution and mortality at low concentrations.^3,14–16

The National Ambient Air Quality Standards (NAAQS), regulated by the United States Environmental Protection Agency (EPA) under the Clean Air Act,^2,17,18 have contributed to the steady decline in air pollution over the past few decades. NAAQS for long-term exposure to PM2.5 and NO2 are set at an annual average of 12 μg/m³ and 53 ppb, respectively, and an 8-h average of 70 ppb for O₃. There is no standard for long-term O₃ exposure. However, it is unclear whether current standards for PM_2.5 and NO₂ are sufficient to adequately protect public health. The role of NO₂ on mortality risk is less clear, although existing evidence suggests that long-term NO₂ exposure may be linked to premature mortality.^19,20 In fact, two recent reviews^21,22 found that existing research does support an association between long-term NO₂ exposure and all-cause mortality. However, previous studies are limited to relatively coarse measurements of exposure over a small geographical extent.^23–25 Both reviews concluded that additional research is needed using large-scale cohort studies to better determine the certainty in associations.

Our study complements recent work that utilizes similar data to estimate the mortality risk associated with low-concentration pollution exposure among Medicare enrollees.^3,7 However, this study aims to contribute to improved understanding in this area by better clarifying the independent mortality effects of long-term exposure to low-level pollutant concentrations with the inclusion of NO₂ exposure. The use of multipollutant models allows for the adjustment of copollutants that may be correlated with each other and may provide more precise estimates of each individual pollutant’s long-term effect on mortality. Further studies are critical to reevaluate current air quality standards and better inform future regulations in protecting public health. Likewise, identifying population subgroups that are most vulnerable to air pollution is also a key research priority to provide a scientific basis for maximally effective policies.

To address these knowledge gaps, we conducted a nationwide, population-based, open cohort study by leveraging massive data sets of high-resolution environmental exposure data (PM_2.5, NO₂, and O₃) and the largest elderly cohort to date (nationwide Medicare population 2001–2017). Focusing on the simultaneous health effects of low-level PM_2.5, NO₂, and O₃, this study captures the complex spatiotemporal patterns in long-term pollution exposure, provides more information on subpopulations exposed to low-level air pollution, controls for temperature, humidity, and pre-existing conditions in addition to the usual socioeconomic status (SES) variable, and compares causal modeling methods to standard methods. We applied multiple statistical methods to evaluate the robustness of our main findings, in an effort to provide strong epidemiologic evidence to better frame environmental policy.

MATERIALS AND METHODS

Study Population.

Health data were obtained from the Centers for Medicare and Medicaid Services (CMS), including all Medicare beneficiaries, aged 65 years or older, in the contiguous U.S. from 2001 to 2017. We extracted data including age and year of Medicare entry, sex, race, Medicaid eligibility (a proxy for SES), the date of death, and ZIP code of residence for each beneficiary. Medicaid eligibility and ZIP code were updated annually. We constructed an open, full cohort containing all Medicare beneficiaries who were alive on January 1 of the year following enrollment into Medicare, through each calendar year of follow-up, with all-cause mortality as the outcome of interest. We further restricted the entire population cohort to define three subsets (low-exposure cohorts) comprised of Medicare beneficiaries who were always exposed to low-level PM_2.5 (annual mean ≤12 μg/m³), NO₂ (annual mean ≤53 ppb), and O₃ (warm-season mean ≤50 ppb), respectively, over the study period. In this study, warm-season O₃ is defined as the period beginning on May 1 to October 31. As there is currently no seasonal or annual standard for O₃ under the NAAQS, warm-season O₃ ≤ 50 ppb was specified as the threshold of exposure to more closely reflect the current global interim target 1 for peak-season average O₃ concentration (i.e., 100 μg/m³, ~50 ppb) established by WHO global air quality guidelines.²⁶

Our research is approved by Emory’s IRB (#STUDY00000316) and the Centers for Medicare & Medicaid Services (CMS) under the data use agreement (#RSCH-2020–55733). The Medicare data set was stored and analyzed in Emory Rollins School secure cluster environment (HPC), with Health Insurance Portability and Accountability Act (HIPAA) compliance. Restricted by our Data Use Agreement with the U.S. Centers for Medicare & Medicaid Services, the Medicare data that support the findings of this study are neither sharable nor publicly available. Academic and nonprofit researchers who are interested in using Medicare data should contact the U.S. Centers for Medicare & Medicaid Services directly to obtain their own data sets upon completion of a Data Use Agreement.

Exposure Assessment.

Using a high-resolution spatiotemporal ensemble model, we fit a generalized additive model integrating three separate machine learning models, including neural networks, random forest, and gradient boosting, and allowing for spatially varying weights for the three different models. The ensemble-based model was calibrated using satellite-retrieved data, chemical transport model simulations, land-use variables, meteorological variables, and EPA monitoring measurements. We predicted daily ambient PM_2.5 (24-h average), NO₂ (1-h maximum), and O₃ (8-h maximum) concentrations at 1-km resolution across the U.S. between 2000 and 2016. The ensemble-based machine learning approach yielded strong model performance, with an average cross-validated R² of 0.89, 0.84, and 0.86 for annual predictions of PM_2.5, NO₂, and O₃, respectively.^27,28 We then aggregated predictions for each pollutant to ZIP codes using all covered 1 km² grid cells and calculated ZIP code annual averages for PM_2.5 and NO₂. For O₃, we restricted daily predictions to the warm-seasonal period (May 1 to October 31) to obtain peak-seasonal O₃ averages. We then assigned annual and seasonal averages to each Medicare beneficiary by the ZIP code of residence. All-cause mortality events were linked to exposures in the preceding calendar year (i.e., 1-year lag exposure), to account for temporality in the exposure-outcome relationship. Any residential mobility changes by ZIP code were accounted for annually.

Covariates.

To adjust for potential confounding, we acquired ZIP code-scale and county-scale covariates in our study, including health care capacity variables (number of hospitals and active medical doctors), a land-use variable (residential greenness), meteorological variables (temperature and relative humidity), demographic and SES variables (population density, education, median household income, % owner-occupied housing units, and % black population), behavioral risk factors (smoking rate and body mass index), and geographical region. To assess potential effect modifications, we additionally included individual-level variables for pre-existing health conditions among Medicare fee-for-service beneficiaries, including ischemic heart disease (IHD), stroke, and chronic obstructive pulmonary disease (COPD). Land-use and meteorological variables were acquired at gridded resolution and aggregated to ZIP code-level to align with residential ZIP codes. SES and demographic characteristics were obtained at ZIP code-level. Behavioral risk factors were obtained at county-level. Individual-level data included age, sex, race/ethnic category, Medicaid eligibility status, and pre-existing health conditions among Medicare fee-for-service beneficiaries. All covariates were included in the models as linear terms unless otherwise noted. Supporting Information (SI) Section 1 provides more details on the data sources and how these covariates were treated in the models.

Statistical Analysis.

We fit a series of stratified Cox proportional-hazards models with comprehensive confounding adjustment, restricting person-time to three low-exposure cohorts comprised of Medicare beneficiaries who were exposed only to low-levels of PM_2.5 (annual mean ≤12 μg/m³), NO₂ (annual mean ≤53 ppb), and O₃ (warm-season mean ≤50 ppb), respectively, over the study period.

For each of the three low-exposure cohorts, we fit both single-pollutant and tripollutant models and estimated hazard ratios (HRs) per 10-unit increase in annual PM_2.5, NO₂, and warm-season O₃ concentrations in the preceding calendar year. A 10-unit scale was utilized for consistency with previous studies.^3,7 To allow for strata-specific baseline hazard functions, we stratified the models by four individual-level covariates: sex, race (white, black, other), Medicaid eligibility, and age at study entry (5-year categories). We adjusted for 12 ZIP code-level and county-level time-varying covariates, including healthcare capacity, SES, behavioral risk factors, land-use, and meteorological variables. Potential residual spatial and temporal trends were considered by including an indicator variable for geographical region and a linear term for calendar-year. For comparison, we fit these models for the entire cohort as well.

To assess any deviation from linearity in the concentration–response (C-R) relationship between mortality and pollution exposure below current U.S. EPA standards, we fit penalized regression splines in the restricted analyses for PM_2.5, NO₂, and O₃, adjusting for all covariates included in the main analysis.

To identify subpopulations that might be particularly susceptible, we assessed potential effect modification by demographic (sex, race, age), environmental (greenness), SES variables (Medicaid eligibility), and pre-existing diagnoses by conducting subgroup analyses for each of the three low-exposure cohorts.

To account for spatial dependence present in the previously applied model estimates, we used the m-out-n bootstrap method by randomly sampling ZIP codes for each bootstrap replicate, to calculate statistically robust confidence intervals.²⁹ Therefore, it is less likely that our findings are influenced by spatial correlation.

We conducted a series of sensitivity analyses to assess the robustness of the main findings to exposure time window, model selection, and confounding. First, we restricted analyses to a subpopulation with at least 10 years of follow-up and assessed different exposure time windows, including 10-year, 5-year, 1-year lag, and no lag effect, as well as cumulative lag periods (including lag 0–1 years, lag 0–3 years, and lag 0–5 years) to account for potentially large annual variations in exposure (SI Table S3). Second, we assessed sensitivity to alternative cut-points for low exposure warm-season O₃ concentrations (SI Table S4). In addition to the three low-exposure subcohorts, we also examined the effect on mortality of those exposed to low levels of all three pollutants (SI Table S5). We additionally conducted a stratified analysis by geographic region by categorizing residential ZIP codes into five distinct regions of the U.S. (SI Figure S1) and assessed differential mortality risk and spatial disparities in exposure resulting from region-specific background ambient concentrations (SI Table S6). Lastly, we employed a stratified Poisson model and further applied inverse probability weighting (IPW, a causal modeling approach that renders exposure independent of measured confounders) to the Poisson model to further account for confounding bias. SI Section 2 describes each approach in further detail.

All computations of this study’s analyses were run on the Rollins High-Performance Computing (HPC) Cluster of Emory University. R software, version 4.0.2, was used for all analyses.

RESULTS

The Medicare cohort included all beneficiaries, ages 65 years and older, in the contiguous U.S. from 2001 through 2017. Table 1 presents descriptive statistics for the full cohort and each of the subset low-exposure cohorts. The full cohort included 68.7-million Medicare enrollees, of which 43.9% were males, 99.4% were between the ages of 65 to 74 at time of enrollment, 84.3% were white, and 11.5% were eligible for Medicaid. Regarding previously diagnosed comorbidities, 16.4% had a prior stroke, 42.2% had IHD, and 25.1% had COPD. There were 27.2-million deaths (39.7%) in the full cohort, with approximately 599.8-million person-years of follow-up.

Table 1.

Descriptive Statistics for the Study Population

	full cohort		low-PM2.5 cohorta (≤12 μg/m3)		low-NO2 cohorta (≤53 ppb)		low-O3 cohorta (≤50 ppb)
variables	number	%	number	%	number	%	number	%
total population	68 721 015	100	22 780 797	100	64454304	100	44684756	100
total deaths	27 265 708	39.7	8 168 363	35.9	26 473 703	41.0	16 507 164	36.9
total person-years	599 815 815		190 557 320		583 354 510		377 148 242
median follow-up year	8		8		8		8
mean age (years)	76.2		75.9		76.2		75.6
Age at Entry (Years)
65–74	68 321 191	99.4	22 670 987	99.5	64091 112	99.4	44409158	99.4
75–115 Sex	399 824	0.6	109 810	0.5	363 192	0.6	275 598	0.6
male	30 142 520	43.9	10 269 690	45.1	28 200 817	43.8	19 746 752	44.2
female Race	38 578 495	56.1	12 511 107	54.9	36 253 487	56.2	24 938 004	55.8
white	57 964 174	84.3	20 456 488	89.8	54673 287	84.8	37474287	83.9
black	5 627 149	8.2	674744	3.0	5 269 023	8.2	3 778 693	8.5
other Medicaid Eligibility	5 129 692	7.5	1 649 565	7.2	4511 994	7.0	3 431776	7.7
eligible	7 935 403	11.5	2 032 401	8.9	6 601 344	10.2	4 642 870	10.4
ineligible Comorbidity	60 785 612	88.5	20 748 396	91.1	57 852 960	89.8	40 041 886	89.6
stroke	11286 541	16.4	3 188 847	14.0	11006 871	17.1	6 800 211	15.2
ischemic heart disease	29 007 076	42.2	8 669 482	38.1	28 160 279	43.7	18 331 767	41.0
COPDb	17 267 809	25.1	5 248 594	23.0	16 798 924	26.1	10 681098	23.9

^aEach low-exposure cohort was restricted to populations who were always exposed to levels below the respective thresholds (PM_2.5: annual mean ≤12 μ/m³; NO2: annual mean ≤53 ppb; O3: warm-season mean ≤50 ppb) over the study period, i.e., exposures in both the preceding calendar year (lag 1) and in the current year (lag 0) were always below the threshold of the pollutant of interest.

^bCOPD, chronic obstructive pulmonary disease.

The three low-level PM_2.5, NO₂, and O₃ cohorts included 22.8 million (33.1%), 64.5 million (93.8%), and 44.7 million (65.0%) Medicare enrollees, respectively, who were exposed only to low-level PM_2.5 (annual mean ≤12 μg/m³), NO₂ (annual mean ≤53 ppb), and O₃ (warm-season mean ≤50 ppb) over the study period (see Table 1 for details). About 16 million (23.4%) Medicare enrollees were exposed to low levels of all three pollutants. Among the low-level exposure cohorts, age, sex, race, Medicaid eligibility, and comorbidities were about the same across the three cohorts. There were 8.2-million deaths (35.9%), with approximately 190.6-million person-years of follow-up in the low-level PM_2.5 cohort; 26.5-million deaths (41.0%), with approximately 583.4-million person-years of follow-up in the low-level NO₂ cohort; and 16.5-million deaths (36.9%), with approximately 377.1-million person-years of follow-up in the low-level O₃ cohort. Across all cohorts, median follow-up was 8 years.

For the period 2000–2016, the long-term mean PM_2.5 concentration was 9.7 ± 3.3 μg/m³; mean NO₂ was 16.4 ± 9.4 ppb; and mean warm-season O₃ was 43.5 ± 5.6 ppb across the contiguous U.S. Long-term PM_2.5 concentrations were highest in California and the southeastern and eastern U.S. The highest NO₂ levels were in New York, New Jersey, and Colorado. Long-term O₃ concentrations were also highest in California and the southwestern U.S. (Figure 1). In the restricted analyses of low-exposure PM_2.5, NO₂, and O₃, the long-term mean PM_2.5 concentration was 7.4 ± 0.87 μg/m³, mean NO₂ was 19.1 ± 9.45 ppb, and mean warm-season O₃ was 40.2 ± 4.79 ppb (SI Table S1). The temporal distribution of annual average PM_2.5, NO₂, and warm-season O₃ is presented in SI Figure S2. Overall, from 2000 to 2016, national annual average concentrations showed a downward trend for each pollution, with more variability observed in seasonal O₃ from year to year, which is likely attributable to seasonal weather conditions.

Figure 1.

Seventeen-year mean concentrations of annual PM_2.5 (μg/m³), annual NO₂ (ppb), and warm-season O₃ (ppb) across the contiguous United States.

Our findings indicate that among older adults who were always exposed to low levels of air pollutants over the study period, PM_2.5, NO₂, and O₃ each independently increased the risk of all-cause mortality (Figure 2). Overall, the single-pollutant models yielded larger effect estimates than the tripollutant models, and all results were consistent across all three statistical approaches. Assessing each pollutant individually in the full cohort analysis, a 10-μg/m³ increase in PM_2.5, 10-ppb increase in NO₂, and 10-ppb increase in warm-season O₃ was associated with an increase in mortality rate (i.e., HR-1) ranging between 5 and 7%, 2 and 3%, and 1 and 3%, respectively. The low exposure analysis yielded larger effect estimates, with corresponding increases in mortality rate ranging between 10 and 13%, 2 and 4%, and 12 and 14%, respectively. After adjusting for copollutants, the effect estimates were attenuated with observed increases in mortality rate ranging between 6 and 8%, 1 and 3%, and 9 and 11%, respectively. All corresponding HR estimates and 95% CIs are shown in SI Table S2.

Figure 2.

Hazard ratios of mortality associated with an increase of 10-μg/m³ in annual PM_2.5 (a), or an increase of 10-ppb in annual NO₂ (b), or an increase of 10-ppb in warm-season O₃ (c) concentration. The estimated hazard ratios were obtained using single-pollutant and tripollutant models under three statistical approaches among the low-exposure cohort and full cohort, respectively.

Figure 3a–c presents the estimated C-R relationship for low-level exposure to mean annual PM_2.5, NO₂, and warm-season O₃. Based on the shape of the C-R curve, we found a linear association for annual mean PM_2.5 below the current annual standard of 12 μg/m³. The C-R relationship for annual mean NO₂ appears to be linear around the majority of data (below the 97th percentile of NO2: 40 ppb). Examining the shape of the C-R curve for mean warm-season O₃ (<50-ppb), we found a nonlinear relationship. A relatively flat association was observed as O₃ levels increased below 40-ppb, with a steep rise in the slope as O₃ levels increased above 45-ppb. Notably, the C-R analyses did not indicate a threshold for mortality at low concentrations of PM_2.5 and NO₂.

Figure 3.

Concentration–response curves of the association between long-term exposure to low-level PM_2.5, NO₂, O₃, and mortality. The concentration–response curves, derived from the tripollutant models, are shown for the concentration ranges between 0.5th and 99.5th percentiles of the PM_2.5, NO₂, and O₃ concentrations among the three corresponding low-exposure cohorts, respectively, that is, with 1% poorly constrained extreme values excluded.

In subgroup analyses, a higher estimated risk of mortality was observed across several covariates for each pollutant. Effect estimates for PM_2.5 were higher amongblack individuals and residents in locations with low residential greenness. Effect estimates for NO₂ were higher among males, relatively younger individuals, white individuals, those ineligible for Medicaid, and IHD patients. Furthermore, we found similar patterns for O₃ with higher effect estimates among residents in locations with lower residential greenness, relatively younger individuals, and those eligible for Medicaid. All estimates are shown in Figure 4. Our sensitivity analyses, including alterative lag specification and stratification by geographical region, indicating that our results are robust to exposure window, model selection, and measured confounding bias (SI Section 3).

Figure 4.

Hazard ratios of mortality associated with an increase of 10 μg/m³ in annual PM_2.5 (a), or an increase of 10 ppb in annual NO₂ (b), or an increase of 10 ppb in warm-season O₃ (c) concentration in low-exposure cohorts by study subgroups. All results were derived from tripollutant models. The shading represents the estimated main effects for the overall population. Dual indicates Medicaid and Medicare dual eligibility. *Significant modification (at α = 0.05 level).

DISCUSSION

We used multiple statistical methods and high-resolution exposure data to estimate the effects of long-term exposure to low levels of PM_2.5, NO₂, and O₃ on all-cause mortality among a nationwide Medicare cohort between 2001 and 2017. We found that long-term exposure to low levels of PM_2.5, NO₂, and O₃ was significantly and independently associated with an increased risk of mortality among the elderly population. Our subgroup analyses indicated that the mortality risk of exposure to each pollutant differed across age, gender, racial/ethnic group, insurance and pre-existing health status, and geographic regions of the U.S. We observed that mortality linearly increased with increasing PM_2.5 and NO₂ concentrations below the NAAQS, with no evidence of a threshold value. For O₃, the C-R relationship shows an increasingly positive association at levels above 40-ppb. These findings suggest that reevaluating the current national standards may yield substantial public health benefits.

Our results are consistent with the findings of previous Medicare cohort studies that long-term exposure to PM_2.5 is significantly associated with increased mortality, particularly at low levels, and contributes further evidence by quantifying the independent effects of both NO₂ and O₃.^3,7,15 Our effect estimates are also similar to those reported in other cohort studies on low-level PM_2.5 exposure.^9,14,30,31 Few cohort studies have assessed low-level NO₂ or O₃. Di et al.³ observed similar effect estimates at low levels when comparing the entire O₃ exposure distribution. By contrast, our results for O₃ are much higher at low levels. One methodological difference is that Di et al.³ did not adjust for time trends in their analyses, which might confound the exposure–outcome relationship. Additionally, they defined the low-exposure cohorts differently by removing person-years with high exposure levels. Previous literature suggests that increased measurement error at higher exposures and depletion of susceptible with higher exposure are possible reasons for the attenuated risk estimates at higher exposures and the larger effect estimates observed in low-exposure analyses, which also indicates that the full cohort analyses likely underestimate the effect estimates.^32,33 The saturation of biological pathways at higher levels is another possibility.³⁴ In addition, as with ambient PM_2.5, it is plausible that compositional differences in fine particles and the surface area to mass ratio may impact toxicity.^35,36

The differential mortality risk observed in the subgroup analyses is generally comparable to estimates reported by Di et al.;³ however, we found a greater magnitude of risk associated with low-level O₃ exposure. Comparatively, a recent study by Yazdi et al.³⁷ found inconsistent results for NO₂-associated mortality by race/ethnicity and Medicaid eligibility status. However, statistical and methodological differences challenge the ability to make direct comparisons in the magnitude and heterogeneity of mortality estimates across subgroups. Overall, previous studies have reported evidence of heterogeneity in air pollution effects among nonwhite racial/ethnic groups, dually eligible beneficiaries or low socioeconomic status, and in urban settings.^9,38–40 The additional mortality risk reported among these subgroups may be explained by inequities stemming from social and environmental determinants of health, including neighborhood deprivation, built environment, or access to health care services. Further, individual-level risk factors such as obesity, smoking status, indoor air quality and second-hand smoke exposure, diet, physical activity, psychosocial stressors, or baseline health status may modify mortality risk. However, Medicare claims data do not include information on these individual-level covariates, limiting our ability to measure important risk factors below an ecological scale. The role of both individual-level and area-level risk factors in modifying individual pollutant-associated mortality warrants further investigation.

Few studies have assessed the shape of the C-R curve between low-level PM_2.5, NO₂, O₃, and mortality. Our study addressed this gap and observed a linear relationship at levels below the current national standards for PM_2.5 and NO₂. There is no current seasonal or annual standard for O₃; however, the daily (8-h) standard is 70-ppb under the NAAQS and 50-ppb during peak-season under WHO global air quality guidelines²⁶ (interim target 1). We observed an increasingly positive association at mean warm-season levels above 40-ppb. To the extent that ambient O₃ production is closely related to temperature, whereby higher O₃ levels are experienced during warmer months, increased seasonal mortality risk may be explained in part by differences in human behavior, such as more time spent outdoors during warmer periods. The apparent 40-ppb threshold is close to the policy-relevant background O₃ in the absence of North American anthropogenic emissions.⁴¹

Previous studies, including Wu et al.⁷ and Di et al.,³ have explored exposure-outcome relationships using same year exposure. In the present study, we used a 1-year lag to ensure with greater certainty that exposure precedes mortality events. Our sensitivity analysis using alternative lags indicated that specifying alternative lags in our models did not meaningfully change HR estimates. Additionally, our geographic region-specific risk estimates indicate some variability by region and consistent positive associations with each pollutant despite spatial disparities in pollutant exposure. PM_2.5, NO₂, and O₃-associated mortality was significant in all regions, except PM_2.5 in the Midwest and warm-season O₃ in the Southeastern U.S.

Taken together, our study supports evidence that there is no “safe” level of PM_2.5 or NO₂ and confirms that health effects persist below the NAAQS; that previously reported associations are not confounded by temperature and humidity; and that the mortality effects of air pollution exposure (and the benefits of reducing it) occur within a few years, given the observed weaker effects at longer exposure lags. In addition, we contribute evidence in support of developing a seasonal O₃ standard.

Our study has several advantages. First, this is a nationwide, population-based U.S. open cohort study focusing on the health effects of low-level PM_2.5, NO₂, and O₃. Our nationwide data sets give us ample statistical power to characterize the complex spatiotemporal patterns inherent in long-term pollution exposure and the risk of mortality, below the current national standards, with greater accuracy. Second, our high-resolution environmental exposure data allow us to capture more low-level air pollution data (e.g., in rural and suburban areas), and thus provides more information on populations exposed to low-level air pollution. Lastly, given our comprehensive statistical approach, multiple sensitivity analyses, and validated robustness of effect estimates, our findings strengthen the evidence for the mortality effects of air pollution at levels below the current national ambient air quality standards.

Despite these advantages, some key limitations should be noted. First, the use of predicted pollutant concentrations in our exposure assessment may have led to measurement error despite showing strong predictive performance. Nevertheless, this source of error is likely nondifferential as the predicted exposure levels are probably independent of outcome status. Thus, any resulting bias would be toward the null. Second, despite having adjusted all models for multiple ecological variables, we cannot rule out the potential for residual bias or unmeasured confounding bias resulting from lack of adjustment for individual-level risk factors (such as indoor air quality and second-hand smoke exposure). Further, the generalizability of our findings may be limited by the characteristics of the Medicare population. Lastly, we were not able to examine cause-specific mortality as it is not available in the Medicare data. Further studies investigating the role of air pollution in cause-specific mortality would provide a valuable addition.

In conclusion, using a large nationwide cohort and robust epidemiological analyses, we provide evidence that long-term exposure to PM_2.5, NO₂, and O₃, at levels below the current national standards, is significantly and independently associated with increased mortality.