2.1 Air quality simulations

We use simulations of air quality over the CONUS from 1990 to 2010 by Gan et al. (2015, 2016). These simulations used the WRF-CMAQ model which coupled the Weather Research and Forecasting (WRF, v3.4) model and the Community Multiscale Air Quality (CMAQ, v5.02) model (Wong et al., 2012), covering the CONUS at 36 km × 36 km. These simulations used an internally consistent emission inventory from 1990 to 2010 for the US developed by Xing et al. (2013), three-dimensional meteorological fields constrained by reanalysis of available surface and aloft measurements of the atmospheric state, and time-varying lateral boundary conditions provided by the hemispheric CMAQ (Mathur et al., 2017; Xing et al., 2015). The model system used the CB05 chemical mechanism with the AERO6 module for aerosols.

Gan et al. (2015, 2016) showed that the modeled trends of total PM_2.5 and its components generally matched both the CASTNET (Clean Air Status and Trend Network) and IMPROVE (Interagency Monitoring of Protection Visual) observational data from 1995 to 2010, with correlation coefficients usually larger than 0.87 for total PM_2.5 and its components. Moreover, the trends for PM_2.5 and its species were similar in direction and magnitude (Gan et al., 2015). There was a small or nearly no trend for PM_2.5 in the western US for both the model and observations, but a dramatic decreasing trend in the eastern US, with a larger decreasing trend from the model (−0.44 µg m⁻³ year⁻¹) than from the IMPROVE observations (−0.30 µg m⁻³ year⁻¹).

For O₃, Astitha et al. (2017) used dynamical evaluation methods, and showed that the simulated O₃ trends generally agreed very well with the observed downward trends, especially for the period from 2000 to 2010, albeit underestimating trends over some regions, for both the May to September average of daily maximum 8 h (MDA8) and annual 4th highest O₃. From 2000 to 2010, the regional trends for the 4th highest O₃ from the model (observations) were −0.80 ppbv year⁻¹ (−0.73 ppbv year⁻¹) for Southwest, −1.14 (−1.53) for Southcentral, −1.31 (−1.66) for Southeast, −1.46 (−1.61) for Midwest, −1.35 (−1.79) for Northeast, and −1.11 (−1.40) for CONUS.

2.2 Mortality burden attributable to ambient air pollution

The mortality burdens attributable to ambient PM_2.5 and O₃ (ΔMort) are estimated using the health impact function (HIF) following Eq. (1):

where y₀ is the baseline mortality rate for specific diseases, AF is the attributable fraction calculated as $\mathrm{1}-\mathrm{1}/\mathrm{RR}$, with RR as the relative risk of death from a specific disease, and Pop is exposed population age 25 years and greater.

For PM_2.5, RR is calculated using the integrated exposure–response (IER) model (Burnett et al., 2014), which has been extensively used by recent studies, including Liu et al. (2017), Silva et al. (2016a, b), Wang et al. (2017), and the World Health Organization (2016). The RR is calculated as a function of PM_2.5 concentration following Eq. (2):

where C is the annual average ambient PM_2.5 concentration, C₀ is the PM_2.5 threshold concentration (5.8–8.0 µg m⁻³), below which no additional risk is assumed, and the parameter values of α, γ, and δ are given by distributions (Burnett et al., 2014). For this study, the RRs are downloaded from the GBD website (Global Health Data Exchange (GHDx), 2013).

Figure 1Annual mean PM_2.5(µg m⁻³) in 1990 (a) and 2010 (b), and the 21-year trends (c, µg m⁻³ decade⁻¹ (µg m⁻³ decade⁻¹)) and summertime average of 1 h daily maximum O₃ in 1990 (ppbv) (d) and 2010 (e), and the trend (f, ppbv decade⁻¹ (ppbv decade⁻¹)). The grey shaded areas in panels c and f indicate trends that are insignificant with p-values for the standard Student's t test larger than 0.05.

Download

For the O₃-related mortality burden, RR =exp^βΔX, where β is the concentration response factor, and ΔX is the difference in O₃ concentration (summertime 1 h daily maximum O₃) between the current year (1990 to 2010) and the low-concentration threshold. For RR, we use the value of 1.040 (with 95 % CI: 1.013, 1.067) from Jerrett et al. (2009) following recent studies (e.g., Cohen et al., 2017; GBD, 2015, 2016; Lim et al., 2012). Turner et al. (2016) found a larger RR for respiratory mortality (RR, 1.12; 95 % CI, 1.08–1.16) associated with the annual average of MDA8 O₃, and using these results would likely lead to a larger O₃ mortality burden (Malley et al., 2017). We account for all chronic respiratory disease (RESP), to be consistent with Jerrett et al. (2009). The counterfactual concentration of 37.6 ppbv (Lim et al., 2012; Lelieveld et al., 2015) is used in our study, to be comparable with Cohen et al. (2017). We use adults above 25 years, to be comparable with other calculations of PM_2.5 mortality burden following Silva et al. (2016a, b), even though the estimated RRs from Jerrett et al. (2009) were for adults above 30 years old only. Uncertainties in air pollution-related mortality burden calculations are based on the uncertainty in RRs only, ignoring those in modeled air pollutant concentrations, and population and baseline mortality rates, which may be larger than that from the RRs but are not in the scope of our study.

We use baseline mortality rates from each year to calculate deaths from air pollution in each year, as changes in baseline mortality rates from other socioeconomic determinants are likely more important than changes in deaths from air pollution. To estimate the annual baseline mortality rates (y₀) for each disease associated with PM_2.5 (chronic obstructive pulmonary disease, COPD; ischemic heart disease, IHD; lung cancer, LC; cerebrovascular disease and ischemic stroke, STROKE) and O₃ (RESP), we acquire US county-level specific causes of mortality data for each year from the National Center for Health Statistics (NCHS) (CDC, 2017). We then aggregate the county-level mortality data to each model grid cell at 36 km × 36 km. The specific causes of mortality data for some counties are sometimes suppressed when the total deaths are lower than 10 year⁻¹ to protect privacy (Jian et al., 2016), missing or considered “unreliable” when the total deaths are less than 20 year⁻¹, and are corrected following established procedures (BenMAP, 2017; Fann et al., 2017; also see the Supplement).

Definitions of each disease follow the GBD study (Lim et al., 2012; Table S1 in the Supplement). Note that the CDC changed the disease codes from the International Classification of Diseases 9th Revision (ICD 9) in 1998 to ICD10 in 1999, and there were discontinuities in the death counts of specific diseases (Anderson et al., 2001; Anderson and Rosenberg, 2003). To account for the discontinuities, we group the total deaths for each of the five diseases for ICD9 and ICD10 using the results of Anderson et al. (2001) and Anderson and Rosenberg (2003), who reported deaths for 135 specific causes in 1996 for both the ICD9 and ICD10 codes, and calculate comparability ratios (Table S1). We then recalculate comparability ratios for the five diseases (RESP, COPD, IHD, LC and STROKE) as the ratios of deaths for ICD9 and ICD10 (Table S2). Finally, we apply these ratios to the ICD9 baseline mortality rates from 1990 to 1998.

Annual population in the US at county level was taken from the US Bureau of Census, which reported populations associated with the 1990, 2000, and 2010 censuses and estimated population for each year in between (CDC 2017; https://www.cdc.gov/nchs/data/nvss/bridged_race/Documentation_bridge_postcenv2017.pdf, last access: 5 September 2018). The adult population above 25 years in the US has steadily grown between 1990 and 2010, with an average 1.23 % year⁻¹ rate of increase (Fig. S1 in the Supplement).

2.3 The contribution of different factors to mortality trends

The overall trends in PM_2.5- and O₃-related mortality between 1990 and 2010 are a combination of contributions from trends in population, baseline mortality rates, and concentration. Here we separate the contributions of each factor by assuming that only a single factor was changing from 1990 to 2010, with the other two constant at 1990 levels. For example, the mortality burden change associated with air pollution changes in year y ($\mathrm{\Delta }{\mathrm{Mort}}_{\mathrm{P}}^y$), relative to 1990, is calculated following Eq. (4):

Similarly, we also calculate the mortality burden change without accounting for ambient air pollution changes ($\mathrm{\Delta }{\mathrm{Mort}}_{\mathrm{noP}}^y$) following Eq. (5):

3.1 Air quality trends

From 1990 to 2010, annual average PM_2.5 in the model decreases significantly in the eastern US (Fig. 1c), but slightly decreases or even increases in the northwest, southwest and west (Fig. S2 and Table S3; also see Fig. S3 for the US 9 regions defined by the National Oceanic and Atmospheric Administration, Zhang et al., 2016). The dramatic decreasing trends of PM_2.5 in the eastern US were also reported in previous studies (Gan et al., 2015; Xing et al., 2015) due to emission reductions. The increasing trend in the western central area is due in part to frequent wildfires (Dennison et al., 2014; Hand et al., 2013, 2014; Jaffe et al., 2008; Murphy et al., 2011; Spracklen et al., 2007). In general, the decadal decreasing trends in the east are larger than 2 µg m⁻³ decade⁻¹ from 1990 to 2010, especially in the central area (−3.48 µg m⁻³ decade⁻¹) and northeast (−3.14 µg m⁻³ decade⁻¹). The summertime average of 1 h daily maximum O₃ decreases significantly in the central and eastern US, generally at a rate greater than 4 ppbv decade⁻¹. It also decreases in the western US, but at a much smaller rate than in the east, generally less than 1 ppbv decade⁻¹ (Fig. 1f; Table S3).

In Fig. 2, both the spatial average and population-weighted average (PWA) annual PM_2.5 exhibit smooth decreasing trends (Fig. 2, top): the spatial average of annual PM_2.5 has decreased by 29 %, from 9.07 µg m⁻³ in 1990 to 6.45 µg m⁻³ in 2010, with a decadal rate of decrease of 1.1 µg m⁻³ decade⁻¹. The corresponding PWA PM_2.5 decreases by 39 %, from 17.61 µg m⁻³ in 1990 to 10.73 µg m⁻³ in 2010, with a decadal decreasing rate of 3.2 µg m⁻³ decade⁻¹. Years with high PM_2.5, such as in 1994, 1996, and 2000, are mainly caused by increases in organic carbon due to large wildfires in the western US (Spracklen et al., 2007). Both the spatial average and PWA O₃ also exhibit decreasing trends over the past 2 decades, with greater inter-annual variability resulting from meteorological variability (Porter et al., 2017). The spatial average O₃ concentration decreases by 9 %, from 55.02 ppbv in 1990 to 49.99 ppbv in 2010, decreasing at a rate of 2.4 ppbv decade⁻¹. The PWA O₃ also decreased by 9 %, from 58.96 ppbv in 1990 to 53.57 ppbv in 2010, decreasing at a rate of 3.0 ppbv decade⁻¹. We also calculate the air quality and mortality burden trends separately for two 11-year periods, 1990 to 2000 and 2000 to 2010, following Astitha et al. (2017). Both PM_2.5 and O₃ decrease more strongly in the second decade than in the first decade for both spatial average and PWA (Table S5), consistent with previous findings (Astitha et al., 2017; Gan et al., 2015; Porter et al., 2017; Xing et al., 2015).

Figure 2Population-weighted average (Popweighted avg) and spatial average over CONUS land areas of annual average PM_2.5 (a) and summertime average of 1 h daily maximum O₃ (b) concentration from 1990 to 2010. Population-weighted average concentrations are based on population in each year. Using the same population in each year yields estimates of population-weighted concentrations that are only slightly different (not shown).

Download

We then calculate trends in the number of days annually that exceed the daily PM_2.5 standard (35 µg m⁻³), and the daily MDA8 O₃ standard (70 ppbv) (Fig. S4). The exceedance days decrease for both PM_2.5 and O₃, especially in the eastern US. In 2010, fewer than 5 days exceed the air quality standard for the majority of the US (Fig. S4b, e). We also calculate the population exposure exceedances by multiplying the population (adults > 25 years old) by the number of air quality exceedance days in each grid cell. The PM_2.5 population exposure exceedances have decreased from 5340 million people-days in 1990 to 1042 million people-days in 2010, and the O₃ population exposure exceedances has decreased from 4691 million people-days in 1990 to 2236 million people-days in 2010 (Fig. S1). These decreases in population exposure exceedances occur despite population growth over this period.

3.2 Mortality burden trends and contributing factors

The mortality burdens associated with exposure to ambient PM_2.5 in the US steadily decreased by 54 %, from 123 700 (95 % confidence interval considering the uncertainty in relative risk only, 70 800–178 100) deaths year⁻¹ in 1990 to 58 600 (24 900–98 500) deaths year⁻¹ in 2010 (Fig. 3). The leading cause of PM_2.5-related mortality is IHD, which decreases by 55 %, from 96 500 (62 600–132 500) deaths year⁻¹ in 1990 to 43 600 (21 500–68 700) deaths year⁻¹, followed by LC, which has decreased by 44 %, from 12 500 (2500–21 000) deaths year⁻¹ in 1990 to 7000 (900–13 400) deaths year⁻¹ in 2010 (Table S4). The PM_2.5 mortality burden per 100 000 adults is much higher in the east than the west for both 1990 and 2010 (Fig. 4), due to the higher PM_2.5 concentrations (Fig. 1).

Figure 3Trends in the total mortality burden (black) for PM_2.5 (a, as a total of ischemic heart disease (IHD) + stroke (STROKE) + chronic obstructive pulmonary disease (COPD) + lung cancer (LC)) and O₃ (b, chronic respiratory disease (RESP)), and mortality burdens considering the air quality change only (blue), and with air quality changes excluded (red). Units are deaths year⁻¹. The error bars are the 95 % CI for the total mortality burden (black).

Download

Table 1The total mortality burdens in 2010 and the burdens in 2010 due to changes since 1990 in each of three factors (concentration, baseline mortality rates and population) and where the concentration change is excluded, for PM_2.5 and O₃, and the relative changes between 2010 and 1990. The relative changes are calculated as (2010–1990) / 1990. The mortality burdens in the US for PM_2.5 and O₃ in 1990 are 123 700 deaths year⁻¹ (70 800–178 100) and 10 900 deaths year⁻¹ (3700–17 500).

Download Print Version | Download XLSX

Table 1 shows the mortality burdens for PM_2.5 and O₃ in 2010, and also the burden changes since 1990 from different contributing factors. From the table, we see that the PM_2.5-related mortality burden in 2010 would have decreased by only 24 % (94 400 deaths year⁻¹ in 2010, 95 %CI, 50 300–139 800) compared with that in 1990, if the PM_2.5 concentrations had stayed constant over the period 1990–2010, due to decreases in the baseline mortality rates for the specific causes of death that PM_2.5 influences (Fig. 3), especially IHD (Fig. 5), despite the population increase. Therefore, the reduction in PM_2.5 concentrations from 1990 to 2010 significantly accelerates the decrease in the mortality burden. The decreased PM_2.5 concentration avoided roughly 35 800 (38 %) PM_2.5-related deaths in 2010, compared to the case if current air quality stays at level in 1990 (estimated as the 2010 mortality burden minus the “concentration change excluded” case in 2010). The benefit of the decreased PM_2.5 concentration could also be estimated as the “concentration change only” case in Fig. 3, yielding 78 900 (35 700–129 200) deaths year⁻¹ in 2010, decreasing by 36 % (−44 800 deaths year⁻¹) compared with 1990. The population increases from 1991 to 2010 would lead to increases in the PM_2.5 mortality burden, but that increase is smaller than the combined reduction from decreasing PM_2.5 concentrations and baseline mortality rates (Figs. S5 and S6).

When separating the two 11-year periods, the PM_2.5-related mortality burden decreased by 45 % from 2000 to 2010 (decreasing trend of −4400 deaths year⁻¹), much higher than the 15 % decrease from 1990 to 2000 (decreasing trend of −2100 deaths year⁻¹) (Table S5). The detrended annual PM_2.5-related mortality burden has a coefficient of variation (CV, standard deviation divided by the average) of 4 %, mainly caused by inter-annual variation in PM_2.5 concentrations (Table S6 and Fig. S6).

We also calculate burdens and trends for each state individually (Table 2). The three states with the highest PM_2.5 mortality burden in 1990 are New York (NY, 13 700 deaths year⁻¹), California (CA, 9500 deaths year⁻¹) and Pennsylvania (PA, 9200 deaths year⁻¹); and in 2010, NY (5100 deaths year⁻¹), Texas (TX, 4200 deaths year⁻¹) and Ohio (OH, 3900 deaths year⁻¹). NY has seen the largest benefits of mortality burden decreases (−8500 deaths year⁻¹), followed by CA (−6100 deaths year⁻¹) and PA (−5500 deaths year⁻¹). For the relative mortality burden changes, generally large percent decreases in PM_2.5-related mortality are seen in western, northern, and northeastern states (including Nevada, Utah, Colorado, Montana, Maine and Vermont) (Fig. 6), because the PM_2.5 concentrations in 2010 are very low or even fall below the low-concentration threshold in these states (Fig. 1), as confirmed by the mortality burden changes from concentration changes alone (Table S7). For other states in the eastern US with large relative mortality burden changes, the contributing factors are different. For example, for Connecticut, the relative mortality burden changes from the decrease in PM_2.5 concentration are larger than that from the decrease in the baseline mortality rates. However, for Massachusetts, NY and PA, the decreases in baseline mortality rates have a slightly larger effect than that from the decrease in PM_2.5 concentration. For CA, the effects from the decrease in baseline mortality rates and PM_2.5 concentration are comparable (Table S7).

Table 2The mortality burden for 48 US states and the District of Columbia in 1990 and 2010, and the absolute changes from 1990 to 2010. Units are deaths year⁻¹.

Download Print Version | Download XLSX

Figure 4The mortality burdens associated with PM_2.5 (a, b), O₃ (d, e) in 1990 (a, c) and 2010 (b, d), and the differences (2010 minus 1990) (c, f) for each 36 km × 36 km grid cell. Units are deaths year⁻¹ per 100 000 adults (above 25 years old).

Download

The mortality burden associated with exposure to O₃ from RESP has increased by 13 %, from 10 900 (3700–17 500) deaths year⁻¹ in 1990 to 12 300 (4100–19 800) deaths year⁻¹ in 2010 (Fig. 3). The O₃ mortality burden per 100 000 adults is highest in the midwest and southwest (Fig. 4). The O₃-related mortality burden in 2010 would have increased by 55 % (10 600 deaths year⁻¹ in 2010, 95 % CI, 3600–17 100) compared with that in 1990 if the O₃ concentration had stayed constant over the period 1990–2010 (Fig. 3), due to increases in both population and baseline mortality rates (Fig. S5). The decreased O₃ concentration would have avoided roughly 4600 (27 %) O₃-related deaths in 2010, compared to the case if ozone concentrations stay at level in 1990 (estimated as the 2010 mortality burden minus the “concentration change excluded” case in 2010). The benefit of the decreased O₃ concentration could also be estimated as the “concentration change only” case in Fig. 3, yielding 8100 (2700–13 100) deaths year⁻¹ in 2010, decreasing by 25 % (−2800 deaths year⁻¹) compared with 1990. The change in O₃ generally reduces the mortality burden relative to 1990 with some inter-annual variation (Fig. S6) due to meteorology and wildfires (Porter et al., 2017), while the increases in population and baseline mortality rates generally increase the mortality burden, with a larger contribution from the population change (Fig. S6).

When separating the O₃ mortality trends into 2 decades, we find that the burdens decrease slightly (−70 deaths year⁻¹) from 2000 to 2010, compared with the increasing trend from 1990 to 2000 (240 deaths year⁻¹) (Table S5). The increasing trend in the first decade is caused by the combined effect of increases in baseline mortality rates and population, while the decreasing trend in the second decade is dominated by decreases in O₃ concentration (Fig. S6). The inter-annual variability for the detrended annual O₃ mortality burden from 1990 to 2010 (CV of 12 %) is larger than PM_2.5 (CV of 4 %), caused mainly by variations in O₃ concentrations from 1990 to 2010 (Table S6).

The three states with the highest O₃ mortality burden in 1990 are CA (910 deaths year⁻¹), Florida (FL, 740 deaths year⁻¹) and NY (700 deaths year⁻¹); and in 2010, CA (1270 deaths year⁻¹), TX (900 deaths year⁻¹) and FL (770 deaths year⁻¹) (Table 2). CA has seen the largest O₃ mortality burden increases (360 deaths year⁻¹), followed by TX (230 deaths year⁻¹) and Arizona (AZ, 140 deaths year⁻¹), with the greatest decrease in NY (−90 deaths year⁻¹). For the relative mortality burden changes, large percent decreases in O₃-related mortality are seen in the northwestern (Washington and Oregon) and northeastern US (Fig. 6), mainly caused by significant O₃ decreases (Table S7), while the greatest percent increases occur in the southwestern US driven mainly by large population increases, and also the baseline mortality rate increases.

Previous health impact assessments have used national baseline mortality rates (Cohen et al., 2017; Silva et al., 2016a, b, etc.), but baseline mortality rates can vary strongly within individual counties (Fig. 5; Dwyer-Lindgren et al., 2016). We performed sensitivity analyses by applying the national average baseline mortality rates for each disease to every county in the mortality burden calculations. We find that the PM_2.5 mortality burden calculated from the national average baseline mortality rates is lower than those calculated from the county-level baseline mortality rates, ranging among individual years from −2.2 % to −1.3 % (Table S8). For the O₃ mortality burden, the differences between using the national average baseline mortality rates and our best estimates range from −1.1 % to 2.0 % (Table S8). However, using the national average baseline mortality rates fails to capture regional mortality burden hotspots for both PM_2.5 and O₃ (Figs. S7–S8), demonstrating the value of using county-level baseline mortality rates where possible.

Figure 5The baseline mortality rates for specific causes of death related to PM_2.5, including chronic obstructive pulmonary disease (a), lung cancer (b), ischemic heart disease (c) and stroke (d), and respiratory diseases related to O₃ (e). The bottom whiskers, bottom border, middle line, top border and top whiskers of the boxes indicate the 5th, 25th, 50th, 75th, and 95th percentiles, respectively, across all counties; the red circles are the national average rate. Baseline mortality rates are shown for 1990–1998 after they are corrected to ensure comparability between ICD9 and ICD10 codes. The units on the y axis are per 100 000 people.

Download

3.3 Comparison with previous studies

The mortality burden associated with PM_2.5 calculated in our study generally aligns with several previous findings (Fig. 7; also Table S9). Our PM_2.5 mortality burden is higher than that reported by Cohen et al. (2017) in 1990 (17 % higher) and 1995 (4 % higher), and lower in 2000 (−0.5 %), 2005 (−17 %) and 2010 (−30 %) (Fig. 7). The overestimation of PM_2.5 mortality burdens in the early 2000s are likely due to the higher population-weighted PM_2.5 concentration simulated by WRF-CMAQ (Fig. 2), compared with Cohen et al. (2017), in which they estimated the PM_2.5 concentration based on data fusion of air quality model outputs, satellite retrievals and ground observations. The lower mortality burdens in the second decade (from 2000 to 2010) in our study likely reflect that Cohen et al. (2017) included hemorrhagic stroke and lower respiratory infections in the PM_2.5-related mortality burden, in addition to COPD, LC, IHD and STROKE, and used an updated integrated exposure–response function. While the absolute value is similar, our results show a stronger decreasing trend (−3000 deaths year⁻¹) than Cohen et al. (2017) (−1000 deaths year⁻¹), which may result from the overestimation of PM_2.5 decreasing trends in our model relative to ground observations (Gan et al., 2016). The PM_2.5 mortality burdens estimated in our study are much lower than those from Fann et al. (2017), but the temporal patterns are similar, mainly because Fann et al. (2017) estimated the total all-cause mortality with a different HIF.

Figure 6Relative mortality burden changes from 1990 to 2010 for the 48 states and the District of Columbia for PM_2.5 (a) and O₃ (b). The relative changes are calculated as (2010–1990)/1990 × 100 %. Note the different color scales for the two plots. The values for the District of Columbia are −37 % for PM_2.5 and −23 % for O₃.

Download

Figure 7Comparisons of the US mortality burdens attributed to PM_2.5 (a) and O₃ (b) in this study, with Cohen et al. (2017), Fann et al. (2012a, 2017), Punger and West (2013), and Giannadaki et al. (2016). The black line for O₃ is the recalculated O₃ mortality burden from the COPD, and the black dashed line is the recalculated O₃ mortality burden from RESP using the pre-industrial O₃ concentration as the counterfactual risk exposure factor. The error bars show the 95 % CI from the RRs, shown for this study and Cohen et al. (2017).

Download

To compare with Cohen et al. (2017), who reported the O₃ mortality burden from the COPD, which is a subset of RESP, we recalculate the O₃ mortality burden from the COPD (Table S4). The newly calculated O₃ mortality burden from the COPD is generally lower than the estimate of Cohen et al. (2017) by 8 %–30 % (Fig. 7). This could be caused by the fact that for the O₃ changes, we use the summertime (April to September) average of the 1 h daily maximum, while Cohen et al. (2017) used the 3-month average, which will be higher. The temporal trend for the O₃ mortality burdens from our study is similar to that from Cohen et al. (2017), except that the burden decreases after 2005 in our study but increases in Cohen et al. (2017). The O₃ mortality burden from the RESP disease in 2005 estimated from our study is much lower than two previous studies (Fann et al., 2012a; Punger and West, 2013; Table S10). As discussed in the methods, the lower US background O₃ concentration used in these two studies (22 ppb in the eastern US and 30 ppb in the western US) could lead to a higher O₃ mortality burden. We then did sensitivity analysis by using the pre-industrial O₃ concentration simulated by an ensemble of model outputs from the Atmospheric Chemistry and Climate Model Intercomparison Project (Lamarque et al., 2013; see Fig. S9 and Sect. 2 in the Supplement) as the counterfactual risk exposure factor, and recalculated the O₃ mortality burden with RESP. The new calculated O₃ mortality burdens are estimated to be 64 %–100 % higher than the current estimation from RESP using the low-concentration threshold (Table S10). In Fig. 7, we see that the new estimated O₃ mortality burden from RESP in 2005 (dashed line) is now comparable with the two previous studies.