The Role of Temperature in Modifying the Risk of Ozone-Attributable Mortality under Future Changes in Climate: A Proof-of-Concept Analysis

Abstract

Air pollution risk assessments typically estimate ozone-attributable mortality counts using concentration-response (C-R) parameters from epidemiologic studies that treat temperature as a potential confounder. However, some recent epidemiologic studies have indicated temperature modifies the relationship between short-term ozone exposure and mortality, which has potentially important implications when considering the impacts of climate change on public health. This proof-of-concept analysis quantifies counts of temperature-modified ozone-attributable mortality using temperature-stratified C-R parameters from a multi-city study in which the pooled ozone-mortality effect coefficients increase monotonically with temperature. Meteorology downscaled from two global climate models is used with a photochemical transport model to simulate ozone concentrations over the 21^st century using two emission inventories: one holding air pollutant emissions constant at 2011 levels and another accounting for reduced emissions through the year 2040. The late century climate models project increased summer season temperatures, which in turn yields larger total counts of ozone-attributable deaths in analyses using temperature-stratified C-R parameters compared to the traditional temperature confounder approach. This analysis reveals substantial heterogeneity in the magnitude and distribution of the temperature-stratified ozone-attributable mortality results, which is a function of regional variability in both the C-R relationship and the model-predicted temperature and ozone.

Graphical Abstract

1. Introduction

The epidemiologic literature consistently reports associations between short-term (i.e., hours to days) human exposure to ozone and a host of adverse health effects.¹ This literature has provided evidence of a relationship between ozone exposure and asthma attacks among asthmatic individuals, increased risk of respiratory-related emergency department and hospital visits, and an increased risk of respiratory death, among other outcomes.¹ In its recently published Integrated Science Assessment for Ozone, the U.S. Environmental Protection Agency concluded that there exists a causal relationship between short-term ozone exposure and respiratory outcomes, a conclusion supported in part by “…strong evidence from epidemiologic studies.”¹

Epidemiologic studies examining the association between ozone and adverse effects within and across cities often find that these effects vary by location. Several recent multi-city studies in the U.S. observing a high degree of spatial heterogeneity (regional or city-to-city) in the ozone-mortality association have attempted to explain this variability by evaluating factors including weather conditions (i.e. season, temperature or weather patterns) and indicate temperature may account for some of the observed differences in estimated risks.^2,3 Solar radiation, which strongly influences the rate at which ozone forms in the atmosphere, coincides with periods of elevated temperatures. As noted by Schwartz (2005), modeling the relationship between ozone exposure and effects including mortality is made difficult by the fact that “…results may be sensitive to [control for] weather….because high ozone days are generally quite hot.”⁴ Temperature may modify the ozone-mortality relationship if, on hot days, the effect estimate for ozone mortality is greater than it is on cooler days.

Time-series epidemiologic studies examining the relationship between short-term ozone exposure and mortality often treat temperature as a potential confounder and thus include temperature as a covariate in the model.^5–11 Epidemiologic studies conducted in the U.S. and Canada have also examined the combined effect of ozone and temperature by either including a joint term for ozone and temperature in the statistical model, or conducting a stratified analysis by temperature (long-term average or daily mean) with the aim of elucidating the role of temperature on the ozone-mortality relationship.^2,3,12–15 In addition, a smaller number of epidemiologic studies have employed complex statistical analyses, such as bivariate response surfaces, and examined alternative statistical models (i.e., additive linear, additive nonlinear, and a monotone spatial risk surface model) in the process of exploring the joint effect of temperature and ozone on mortality.^3,13

Findings from these epidemiologic studies suggest that temperature may modify the ozone-mortality relationship. These studies indicate that the association between ozone and mortality may vary depending on the temporal specification of the temperature variable. Specifically, mortality risk associated with short-term exposure to ozone was increased at lower long-term average temperatures^2–3, while increased at higher daily temperatures^6–7. While these results may appear contradictory, locations with lower long-term average temperatures may experience greater daily temperature variability and also have a lower prevalence of central air conditioning, resulting in an increase in exposure to higher daily temperatures. ⁷ In addition to studies demonstrating a synergistic relationship between daily temperature and ozone, more recent studies provide evidence of a U-shaped modifying effect, where ozone-related mortality risk is higher on low and high temperature days compared to moderate temperature days. ^8–9 Several studies conducted outside of North America also suggest similar patterns of effect modification by temperature.^16–20

Quantitative risk assessments characterizing ozone-attributable mortality have applied concentration-response (C-R) parameters from epidemiologic studies that rely on the traditional approach of controlling for the potential confounding effects of temperature in order to quantify counts of ozone-attributable effects.^21,22 Very few analyses have drawn upon evidence from epidemiologic studies modeling temperature as an effect-modifier and reported parameters that can be used directly in a health impact assessment.²³ Such an approach might give insight to the role of temperature in affecting the magnitude of ozone-attributable risk reflecting changes in the future climate, which is expected to increase temperatures and otherwise create conditions more conducive to ozone formation.^24–26

Jhun et al. (2014) used weather, air pollution, and mortality data to examine the potential modification of the relationship between short-term ozone exposure and total (nonaccidental) mortality by temperature across 97 U.S. cities.¹⁵ In contrast with other literature modeling temperature-stratified ozone-related mortality, Jhun and co-authors: (1) specify an epidemiologic model that can be parametrized in an air pollution risk assessment; (2) report the input parameters—including city-specific temperature-modified concentration-response parameters—needed to quantify temperature-stratified ozone-related mortality; (3) separately model non-temperature-modified ozone-mortality risk including temperature as a covariate, the results of which can be used to quantify ozone-only attributable death and then compared against the results of the temperature-stratified analysis.²⁷ Building on our previous work in Juhn et al. 2014, in this proof-of-concept analysis, twenty-first century climate scenarios from two alternative global climate models are dynamically downscaled and used with a photochemical transport model to simulate ozone concentrations. A developmental version of the environmental Benefits Mapping and Analysis Program—Community Edition (BenMAP-CE) modified to incorporate the temperature-stratified concentration-response parameters from Jhun et al. (2014) is then used to estimate changes in ozone-related mortality attributable to climate change.

This analysis aims to help answer two questions, one methodological and one practical:

1. How do estimated counts of ozone-attributable mortality quantified using concentration-response parameters from an epidemiologic study examining the modification of ozone effects by temperature compare to those estimated using concentration-response parameters that control for temperature as a potential confounder?

2. To what extent is the additional computational complexity of a temperature-stratified ozone mortality analysis warranted by the resulting number of attributable deaths, as compared to the traditional approach?

2. Materials and Methods

2.1. Climate and Air Quality Modeling

This analysis employs simulated future changes in meteorology and ozone concentrations using methods and models described in detail elsewhere.^28,29 In the interest of brevity, we summarize these here, though additional information may be found in the Supplement. We simulate future changes in climate using two global climate models: the Community Earth System Model (CESM – CCSM4) and the Geophysical Fluid Dynamics Laboratory (GFDL) Coupled Model version 3 (CM3). We ran each model using the Representative Concentration Pathway (RCP) 8.5, a high warming scenario reaching 8.5 W/m² in the year 2100; computational constraints prevented us from simulating additional warming scenarios.³⁰ Using the Weather Research and Forecasting (WRF) model, we dynamically downscale these global scenarios to a 36-km resolution over North America.³¹

The above downscaled meteorology serves as an input to the Community Multiscale Air Quality (CMAQ) chemical transport model, and simulates ozone concentrations over the conterminous U.S. for three 11-year periods centered on 2000, 2030, and 2095.²⁸ In each period, the model simulated ozone concentrations resulting from anthropogenic air pollutant emissions reported in the National Emissions Inventory (NEI) for the years 2011 and those projected to the year 2040; this latter inventory accounts for planned federal, state and local emission controls.³² Relative to the year 2011, the 2040 inventory reduces ozone precursor emissions by a large margin: nitrogen oxides (NO_x), sulfur dioxide (SO₂), and volatile organic compound (VOC) decrease by 44%, 57%, and 12%, respectively (Supplemental Information). Emissions of biogenic VOCs are simulated using the downscaled meteorology, and thus respond to the future climate scenarios.

2.2. Quantifying ozone-attributable deaths

We quantify ozone-attributable deaths using two health impact functions developed from concentration-response parameters reported in Jhun et al. (2014). Performing a new epidemiologic study was beyond the scope of this manuscript, instead we parametrized concentration-response relationships from the Jhun et al. (2014) study, following an approach well established in the literature.³³ We constructed the first health impact function using a concentration-response parameter for ozone-attributable mortality from models that specified temperature as a confounder. We constructed the second health impact function using concentration-response parameters from models that specified temperature as an effect modifier and stratified ozone-attributable mortality according to daily mean temperature. We describe each below. In both cases we use a developmental version of the open-source environmental Benefits Mapping and Analysis Program—Community Edition (BenMAP-CE).³⁴ The BenMAP-CE source code is available via GitHub.³⁵

2.2.1. Parameterizing concentration-response relationships: ozone-mortality estimates including temperature as a confounder

The first approach to quantifying ozone-attributable deaths follows techniques well established in the literature.^28,36–38 We selected from Jhun et al. (2014) a concentration-response parameter (β₁) from a model accounting for temperature as a potential confounder through the incorporation of a smoothing function as:

$$\log \left[E\left(Y_t\right)]=\alpha +{\beta }_1{ozone}_t+ns\left({temp}_t,df=4\right)+ns({dptemp}_t,df=4\right)+ns\left({season}_t,df=3perseason\right)+{\gamma dow}_t+{\epsilon }_t$$

Jhun and co-authors define subscript t to denote the time of the observation and β₁ as the main effect of ozone; E(Y_t) is the expected value of the daily count of deaths (Y_t); ns is the natural smoothing spline function; α is the intercept. Ozone_t, temp_t, and dptemp_t are daily 24-h ozone, temperature, and dew point temperature, respectively. Season_t is the day of the calendar year and controls for seasonal variation of mortality. Dow_t is the day of week at time t and γ is a vector of coefficients; ε_t is the residual.¹⁵

Next, we parameterize a health impact function using the results of the above model. We calculated counts of O₃-attributable non-accidental deaths (y_ij) during two periods i (i= an 11-year period centered on 2030 and 2095) for population stratum a in each county j (j=1,…,J where J is the total number of counties) as

$${\text{y}}_{\text{ii}}={\mathrm{\Sigma }}_{\text{a}}{\text{y}}_{\text{iia}}$$

$${\text{y}}_{\text{ija}}={\text{mo}}_{\text{ija}}\times \left({\text{e}}^{\text{β}\cdot {\text{C}}_{\text{ij}}}-1\right)\times {\text{P}}_{\text{ija}},$$

where mo_ija is the baseline all-cause mortality rate for populations aged a=0–99 in county j in year i stratified in 10-year age groups; β is the risk coefficient for non-accidental mortality associated with O₃ exposure; C_ij is daily mean warm season (i.e., May-September) O₃ concentration in county j in year i for each projected 11-year average period compared to an 11-year average centered on the year 2000; and P_ija is the number of individuals aged a=0–99 in county j in year i stratified into 5-year age groups. Ozone concentrations were assigned from the 36km by 36km model grids to each county using a population-weighting procedure described elsewhere.³⁹ In the absence of evidence regarding a threshold in the relationship between daily ozone and mortality, we did not specify a threshold in the concentration-response parameter.¹

2.2.2. Parametrizing concentration-response relationships: ozone-mortality estimates including interaction with temperature strata

Jhun et al. (2014)⁹ also examined the potential modification of O₃-mortality risk at low, moderate, and high temperature strata using an interaction term:

$$\log \left[E\left(Y_t\right)\right]={\alpha }^{\text{*}}+{\beta }_1^{\text{*}}ozon{e}_t+{\beta }_2^{\text{*}}\left(ozon{e}_t:tem{p}_{kt}\right)+{\beta }_3^{\text{*}}\left(tem{p}_{kt}\right)+ns\left(dptem{p}_t,df=4\right)+ns\left(seaso{n}_t,df=3perseason\right)+{\gamma }^{\text{*}}do{w}_t+{\epsilon }_t^{\text{*}}$$

Jhun et al. define temp_kt according to levels of the same-day 24-h temperature at time t and temperature category k, where k indicates the low, moderate, and high temperature categories. β₁^⁎ is the main effect of ozone per 10ppb. β₂^⁎ is a vector of coefficients of the interaction between ozone and temperature that is also adjusted for temperature (temp_kt) within each temperature strata. β₃^⁎ is the vector of coefficients of temperature levels.¹⁵ Jhun et al. did not separately report effect estimates according to alternative lags.

Daily mean temperature observations for May-September of 1987–2000 were used to calculate 5^th and 95^th percentile temperatures for each of 97 cities across the U.S. Days were then categorized as low (lower than the 5^th percentile), moderate (greater than the 5^th and less than the 95^th percentile) and high (greater than the 95^th percentile) temperature days. The authors provided separate effect coefficients by temperature strata (β₁ + β₂) for each city as a means of accounting for regional heterogeneity on the influence of temperature-modified ozone-related mortality risk (Supplemental Information (SI) Figure 1); this stands in contrast to the model specifying temperature as a confounder, where a single national effect coefficient is estimated. For each city included in Jhun et al. (2014), we specified three health impact functions that incorporated logic enabling BenMAP-CE to select the appropriate effect coefficient based on whether that day’s simulated 24-hour temperature fell into the low, medium, or high category for that city.

Jhun et al. (2014)⁹ did not report temperature-stratified effect coefficients for locations outside of these 97 cities. Because we wanted to quantify temperature-stratified ozone mortality nationwide, we developed temperature-specific health impact functions for each of the seven U.S. National Mortality and Morbidity Air Pollution Study (NMMAPS) regions: Midwest, Northeast, Northern Great Plains, Northwest, Southeast, Southern Great Plains, Southwest. For each region, we calculated a U.S. census population-weighted average of the temperature cutoffs, effect coefficients, and standard errors from the cities within each region; we used these functions to quantify temperature-modified mortality within each of seven regions, not including the 97 cities for which Jhun et al. (2014) supplied effect coefficients. As with each city, we specified regional functions that selected between three effect coefficients that represented the low, medium, or high 24-hour temperature category.

2.2.3. Demographic and baseline health parameters

The BenMAP-CE tool contains projected cause-specific and age-stratified death rates through the year 2060 at the U.S. county level. We start first by calculating a three-year average of the baseline rates of all-cause mortality from the U.S. Centers for Disease Control and Prevention WONDER database for the years 2012–2014.^40,39 We next project this rate to future years by drawing upon U.S. Census Bureau projected mortality rates, available in 5-year increments through the year 2060.⁴¹ When estimating impacts in the year 2095 we applied projected 2060 death rates (SI).

The Integrated Climate and Land Use Scenarios (version 2) (ICLUS) supplied county-level age-stratified projected population counts to the year 2095.⁴² ICLUSv2 was harmonized with the median variant projection of the United Nation’s 2015 World Population Prospects dataset, a mid-range population projection similar to Shared Socioeconomic Pathway 2 (SSP2).^43,44

3. Results

3.1. Predicted temperature and ozone

The CESM-CCSM4 and CM3 models each predict substantially higher daily temperatures throughout the U.S. by late century (an 11-year average centered around 2095), as compared to early century (an 11-year average centered on 2000) (SI Figure 1 & 2). Both models project increased maximum daily temperature in late century to be as much as 10 degrees higher, as compared to early century. Across the 97 cities, both models predict over 30% of maximum daily temperatures in the late century to be in the high temperature strata, compared to just 4.7% (CM3) and 10.5% (CESM) in the early century (SI Table 1). Where CESM predicts temperatures to be highest in the Industrial Midwest, CM3 projects higher temperatures predominantly in the West. Neither model projects temperatures to decrease between early and late century.

CMAQ predicts summer season mean ozone concentrations to be up to 7 ppb higher in late century as compared to early century; this is true when simulating ozone concentrations using downscaled meteorology from both the CESM and CM3 climate models for the 2011 emission inventory. The CESM-CMAQ predicted increases in late-century summer season ozone concentrations are highest in the upper Midwest and Northeast (Figure S1). Regions including the Southeast and West are projected to experience a decrease in summer season ozone by late century as compared to early century (Figure S1). The CM3-CMAQ predicted summer season ozone concentrations are highest in the Industrial Midwest and portions of the South (Figure S2). Conversely, CM3-CMAQ predicts lower ozone concentrations in the Southeast and West (Figure S2). Irrespective of climate model, the predicted summer season ozone concentrations decline significantly when modeled using a 2040 emission inventory. Predicted changes in ozone concentrations correlate well with the CESM-predicted increases in late-century temperature but are not as well correlated with CM3-predicted increases in temperature.

3.2. Ozone-attributable deaths: temperature as a confounder

We estimate between 620 additional, and 340 fewer, ozone-attributable deaths in the early century depending on the climate model and emission inventory modeled; and between 1,600 additional, and 1,000 fewer, ozone-attributable deaths are estimated to occur in late-century (Table 1). When modeling ozone concentrations using an emission inventory that accounts for reduced anthropogenic emissions through the year 2040, we quantify over a hundred fewer cases of attributable deaths compared to ozone-attributable deaths quantified using an inventory that reflects 2011 anthropogenic emissions. The greatest change in the estimated number of ozone-attributable deaths (per 100k) occur in the Northeast and Industrial Midwest, followed by the Upper Midwest; a comparatively small number of attributable deaths occur in the Southeast, Southwest, Northwest, and Southern California (SI Table 2).

Table 1:

Estimated number of additional ozone-attributable deaths, with temperature treated as a confounder, in the early and late century quantified using two alternate climate models and anthropogenic emission inventories (ei) (95% confidence intervals)^A

Model	Early-Century (2030)		Late-Century (2095)
	2011ei	2040ei	2011ei	2040ei
CESM	620 (260 to 980)	360 (150 to 580)	1,600 (650 to 2,500)	340 (140 to 530)
CM3	24 (9.8 to 39)	−340 (−530 to −140)	930 (390 to 1,500)	−1,000 (−1,600 to −420)

^AResults calculated relative to year 2000, where all periods represent 11-year averages.

3.3. Ozone-attributable deaths: using a categorical interaction term

The estimated size of the temperature-modified ozone-attributable deaths is consistently larger than those estimated using concentration-response parameters treating temperature as a confounder; the difference between the two estimates ranges from as few as about 60 to as many as 2,600 (Table 2). We estimate the greatest number of attributable deaths in the Northeast and Industrial Midwest, followed by the Upper Midwest, in both the CM3 and CESM scenarios (Figures 1 and 2). The estimated number of attributable deaths declines when simulating ozone concentrations using a 2040 emission inventory. In both the CM3 and CESM scenarios we estimate avoided ozone-attributable deaths in the Southeast using the 2040 inventory; and the CM3 model with the 2040 inventory yields more avoided ozone-attributable deaths compared to the CESM model, specifically in regions including Southern California, the Southwest and Northwest (Figures 1 and 2).

Table 2:

Estimated number of additional temperature-stratified ozone-attributable deaths in the early and late century quantified using two alternate climate models and anthropogenic emission inventories (ei)^A

Model	Early-Century (2030)		Late-Century (2095)
	2011ei	2040ei	2011ei	2040ei
CESM	1,200 (1,100 to 1,400)	740 (600 to 900)	4,200 (4,000 to 4,300)	1,800 (1,600 to 2,000)
CM3	100 (−100 to 300)	400 (200 to 600)	2,800 (2,600 to 3,000)	870 (700 to 1,000)

^AResults calculated relative to year 2000, where all periods represent 11-year averages.

Figure 1.

Estimated number of temperature-stratified ozone-attributable premature deaths in early and late century, quantified using meteorological inputs from the CESM climate model and a 2011 and 2040 emission inventory, relative to the year 2000 (per 100k people)

Figure 2.

Estimated number of temperature-stratified ozone-attributable premature deaths in early and late century, quantified using meteorological inputs from the CM3 climate model and a 2011 and 2040 emission inventory, relative to the year 2000 (per 100k people)

The city-level results exhibit a high degree of heterogeneity. We estimate as many as 360 additional ozone-attributable deaths in Denver, 780 additional deaths in Chicago and 73 additional deaths in Memphis, depending on the climate scenario and emission inventory (SI Table 2). By contrast, we estimate up to 210 fewer ozone-attributable deaths in Phoenix, up to 77 fewer deaths in Las Vegas and up to 70 fewer deaths in El Paso.

4. Discussion

In this study we calculated attributable cases of ozone-related mortality using dynamically downscaled meteorology from two alternative climate models to quantify effects using both temperature-controlled and temperature-stratified concentration-response parameters from Jhun et al. (2014). We were particularly interested in whether the estimated ozone-attributable mortality counts would differ when performing the analysis by temperature strata (with the higher temperature showing higher impacts) as opposed to ozone-attributable mortality counts calculated using a concentration-response parameter controlling for temperature as a potential confounder.

Increasing temperatures over the 21^st century yield thousands of additional ozone-attributable deaths in late century as compared to temperature-stratified results estimated for early century. A traditional approach using a concentration-response relationship from an epidemiologic study specifying temperature as a covariate would not fully capture this potential increase in premature mortality.²⁸ For instance, our estimated number of deaths attributable to a 10 ppb increase in ozone in the high temperature strata is consistently two to three times larger than that for a 10 ppb increase in ozone exposure in models where temperature was considered a confounder. This suggests that the location-specific influence of temperature is a substantial contributor to ozone-related mortality impacts that is not captured in epidemiologic studies that treat temperature solely as a potential confounder. The estimated number of additional ozone-attributable deaths occurring in 2030 and 2095 quantified using the traditional approach is commensurate with those reported by Fann et al. (2021), which used the same predicted air quality changes but a different concentration-response parameter.²⁸

In addition, simulating future ozone concentrations with an emissions inventory that accounts for reduced levels of anthropogenic emissions of ozone precursors yield hundreds of fewer ozone-attributable deaths; this is true when using models that treat temperature as a confounder or as an effect-modifier. This projected decline in the ozone-attributable burden is consistent with previous work indicating other things being equal, reduced anthropogenic emissions will dampen the climate penalty.²⁸

As noted above, this proof-of-concept analysis quantifying temperature-modified ozone-attributable mortality risks using concentration-response relationships from epidemiologic studies specifying a categorical interaction term is computationally intensive. This additional complexity comes from: (1) simulating temperature and ozone concentrations for each hour of each day of the year; (2) quantifying ozone-attributable deaths for individual cities and regions; and (3) selecting a temperature-conditional concentration-response parameter for each day. By contrast, analyses that examine mortality impacts attributed to ozone frequently quantify counts of premature deaths using a seasonal average ozone concentration and a single concentration-response parameter for the entire nation.⁴⁵

Temperature-modified ozone-mortality impacts are between approximately 13% and 430% larger than those quantified using the traditional approach. Differences of this magnitude suggest that the additional computational burden associated with quantifying temperature-modified ozone mortality is worth bearing. However, these results should be interpreted within the context of the analytical choices above—namely, using two alternative emission inventories and projecting ozone and temperature to late century—which each tend to increase the difference in temperature levels and ozone concentrations between the reference case (the year 2000) and each climate case. More modest changes in ozone or temperature may yield estimates of temperature-modified ozone mortality that are more similar to the ozone-only approach.

The estimated temperature-modified ozone deaths display a high degree of temporal and spatial heterogeneity. The city-level results are highly variable across urban areas, as future climate changes could result in some cities experiencing an estimated increase in temperature-modified ozone-related attributable deaths while some are expected to see a decrease. Although a clear geographic trend in the city-level results is difficult to discern, these results suggest that some larger cities located in the Upper Midwest, Industrial Midwest and Northwest may experience an increase in ozone-attributable deaths while other cities located in the Southwest may experience a decrease. Additional research is warranted to further elucidate reasons why this heterogeneity in results is observed. Plausible explanations include regional variability in population time-activity patterns; differences in how populations respond, and adapt, to heat; the prevalence of air conditioning; relative levels of NOx and VOC in each location; and the baseline health status of the populations exposed jointly to ozone and temperature, as reflected by the baseline death rate.

While this proof-of-concept analysis is informative in examining not only the potential public health impacts of the joint effect of temperature and ozone, along with how these impacts may change under future climate scenarios, the analysis is subject to certain limitations. First, our results are based on simulations from only two climate models following just one greenhouse gas forcing scenario. Though warm season temperatures over the U.S. in 2000 are well simulated in CESM, there is a cold bias in CM3²⁹ that likely causes an underestimation of the fraction of high temperature days during the baseline period, potentially overestimating the future increase. Expanding the analysis to include a larger ensemble of models and scenarios would help better characterize the differences between the temperature-as-confounder and temperature-as-effect-modifier approaches. Second, we are uncertain as to whether the temperature-modified ozone relationships characterized by Jhun et al. (2014) will persist into the future. That is, the slope and shape of the concentration-response parameters may change over time. By focusing this analysis on only quantifying the direct ozone and temperature-modified ozone mortality impacts, we do not quantify counts of direct temperature-attributable mortality. Quantifying the temperature-only mortality might provide greater insight into the extent to which the temperature-modified ozone mortality impacts commensurate with the sum of the individually estimated ozone mortality and temperature mortality impacts. Prior analyses suggest this value may be substantial and of the same order of magnitude as the estimated pollutant-attributable impacts.^46–49 However, Jhun et al. (2014) did not report temperature-mortality effect coefficients and as a result, we were unable to provide a direct comparison of temperature-related and ozone-related mortality.¹⁵ Lastly, although there is extensive epidemiologic evidence indicating consistent, positive associations between short-term ozone exposure and total (nonaccidental) mortality, “recent evidence from controlled human exposure studies, in combination with the lack of coherence between animal toxicological and epidemiologic studies of cardiovascular morbidity, specifically the lack of epidemiologic evidence for cardiovascular-related ED visits and hospital admissions, leads to substantial gaps in the biologically plausible pathways by which short-term ozone exposure could lead to cardiovascular mortality”, the biggest contributor to total (nonaccidental) mortality.¹ This combination of evidence, as noted in the most recent Integrated Science Assessment for Ozone and Related Photochemical Oxidants, resulted in the U.S. EPA concluding that the relationship between ozone and total (nonaccidental) mortality is “suggestive of, but not sufficient to infer, a causal relationship”.¹ Despite the uncertainty regarding the biologically plausible pathways by which ozone may lead to mortality, the attributable cases of temperature-modified ozone-related mortality quantified in this study provides important insight under future climate scenarios.

Table 3:

Regional Distribution of Additional Temperature-Stratified Ozone-Attributable Mortality Simulated Using Meteorology Predicted by the CESM and CM3 Climate Models

	Projected emissionsA	Time PeriodB	RegionC
			Midwest	Northeast	Northern Great Plains	Northwest	Southeast	Southern Great Plains	Southwest
CESM	2011ei	Early	620	340	150	−13	43	20	50
	2011ei	Late	1,900	1,700	170	−57	180	−110	420
	2040ei	Early	470	140	130	−10	−9	1	23
	2040ei	Late	1,200	610	110	0	−39	−180	70
CM3	2011ei	Early	120	−68	21	45	−7	63	−72
	2011ei	Late	2,100	1,200	160	−530	130	1	−230
	2040ei	Early	10	−230	−14	36	−53	−1	−150
	2040ei	Late	950	−250	13	−360	−130	−210	−890

^A2011ei = 2011 emission inventory; 2040ei = 2040 emission inventory

^BEarly = additional temperature-stratified ozone-attributable deaths calculated using an average of 2025 to 2035 predicted ozone and temperature levels; Late = additional temperature-stratified ozone-attributable deaths calculated using an average of 2090 to 2100 predicted ozone and temperature levels. Warmer colors indicate an increase in ozone-attributable deaths. Cooler colors indicate a decrease in ozone-attributable deaths. Values summed across each of three temperature strata.

^CThe values reported for each region represent the sum of the estimated temperature-modified ozone deaths for each city within that region and the estimated temperature modified ozone deaths occurring outside of those cities but within the region.

Synopsis:

Quantifying the combined effect of temperature and ozone on health may more completely account for the influence of a changing climate on public health.