Association of short-term exposure to fine particulate air pollution and mortality: effect modification by oxidant gases

Short term changes in exposure to outdoor fine particulate matter (PM2.5) concentrations are associated with an increased risk of mortality. However, less is known about how oxidant gases may modify the acute health effects of PM2.5. Our objective was to investigate whether associations between acute exposure to PM2.5 and mortality were modified by the oxidant gases O3 and NO2 using their redox-weighted average (Ox). We conducted a multi-city case-crossover study in 24 cities across Canada between 1998–2011 including 1,179,491 nonaccidental mortality events. Interquartile increases in lag-0 and 3-day mean PM2.5 and Ox concentrations were each associated with small increases in nonaccidental and cardiovascular mortality. In stratified analyses, associations between PM2.5 and nonaccidental and cardiovascular mortality tended to be greatest in the highest tertile of Ox with a significant interaction observed between lag 0 PM2.5 and 3-day mean Ox (interaction p-value = 0.04). There was no evidence of effect modification by Ox in the relationship between PM2.5 and respiratory mortality. Overall, the relationship between short-term changes in outdoor PM2.5 and nonaccidental mortality may be greater when oxidant gas concentrations are also elevated. In some regions, reductions in oxidant gas concentrations may also reduce the acute health impacts of PM2.5.

Short-term increases in outdoor fine particulate air pollution (PM 2.5 ) are known to be associated with increased mortality [1][2][3] . Other pollutants including nitrogen dioxide (NO 2 ) and ozone (O 3 ) have also been associated with daily mortality events 4,5 , but it is not clear how these oxidant gases may modify the acute health effects of PM 2.5. This is an important public health issue as populations are simultaneously exposed to both PM 2.5 and oxidant gases (e.g. O 3 and NO 2 ). Moreover, understanding interactions between these pollutants may help to inform preventative measures aimed at reducing the public health impacts of outdoor air pollution.
A recent study conducted in London, England found that O 3 , NO 2 , and their combined oxidant capacity (O x ) were each associated with daily mortality with the strongest associations observed for O x 6 . However, this study did not specifically evaluate how O x may modify the acute health effects of PM 2.5 . Recently, we reported that the strength of associations between long-term exposures to outdoor PM 2.5 and nonaccidental, cardiovascular, and respiratory mortality were greater in regions with higher O x concentrations 7 . Biological mechanisms explaining this observation may include the fact that oxidant gases are known to deplete anti-oxidants in the lung lining fluid 8 and increase the permeability of the lung epithelium [9][10][11][12] . Alternatively, photochemical aging of PM 2.5 may increase particle toxicity 13,14 and this process may be accelerated in regions with higher oxidant gas concentrations. In either case, this evidence suggests that PM 2.5 may be more harmful on days with increased concentrations of oxidant gases.
In this study, we examined how O x may modify the acute health effects of PM 2.5 using a multi-city case-crossover study of non-accidental, cardiovascular, and respiratory mortality.  Table S1 shows Pearson correlation coefficients among the air pollutants and weather variables. PM 2.5 was moderately correlated with NO 2 and weakly correlated with the other pollutants and weather variables. Figure 1 and Table S2 show associations between ambient air pollutants and nonaccidental, cardiovascular and respiratory mortality during the time period of 1998 to 2011. For PM 2.5 , lag-0 and 3-day mean concentrations were each associated with small increases in nonaccidental and cardiovascular mortality. Short term changes in lag-0 and 3-day mean O x concentrations were positively associated with all three mortality outcomes, but 95% confidence intervals for respiratory mortality included the null. In general, risk estimates for O x tended to be slightly larger than for O 3 or NO 2 individually with the exception of lag-0 respiratory mortality which was similar for O 3 and O x .
The results of stratified analyses examining the relationship between PM 2.5 and mortality across tertiles of O x are presented in Figs 2-4 and Table S5. For lag-0 PM 2.5 , increased risks of nonaccidental (interaction p-value = 0.04) and cardiovascular mortality (interaction p-value = 0.19) were limited to the highest tertiles of O x . This trend was less clear for 3-day PM 2.5 concentrations and evidence of effect modification by O x was not observed for the relationship between PM 2.5 and respiratory mortality. In sensitivity analyses, PM 2.5 -mortality associations were not modified by NO 2 or O 3 individually (data not shown). As well, findings when restricted to the warm season only were similar to the whole year analyses (data not shown).

Discussion
In this study, we examined how oxidant gases may modify associations between short-term changes in outdoor PM 2.5 concentrations and nonaccidental, cardiovascular, and respiratory mortality. As in previous studies, we found that short-term changes in ambient air pollution concentrations were associated with small increased risks of mortality, predominantly nonaccidental and cardiovascular mortality. For PM 2.5 specifically, we noted that same day exposures were only associated with nonaccidental and cardiovascular mortality during periods with the highest O x concentrations (i.e. above 21.38 ppb). Moreover, we found that short term changes in O x were more strongly associated with nonaccidental and cardiovascular mortality than PM 2.5 in mutually adjusted models.
One previous study conducted in London, England reported an association between daily variations in O x and mortality 6 , but to our knowledge this is the first study to evaluate how oxidant gases may modify the relationship between short-term changes in outdoor PM 2.5 and mortality. However, we recently reported that O x levels modified the relationship between short-term changes in ambient PM 2.5 and the risk of myocardial infarction and our   All models are adjusted for 3-day mean ambient temperature (cubic splines) and relative humidity and daily counts of hospitalization for influenza (in respiratory mortality models only). results for cardiovascular mortality are consistent with this finding 15 . In addition, we previously reported that oxidant gases modified the relationship between long-term PM 2.5 exposures and mortality with stronger associations observed in areas with higher O x concentrations 7 . Collectively, these findings suggest that oxidant gases may modify both the acute and chronic health effects of PM 2.5 exposures with larger risk estimates observed for chronic health impacts. While our study could not directly evaluate how O x concentrations may modify PM 2.5 health impacts existing evidence suggests that such a relationship is biologically plausible. For example, one possibility is that oxidant gases deplete anti-oxidants in the lung lining fluid which in turn may lower our natural defense against reactive oxygen species generated in response to PM 2. 5 8,16 . Moreover, some findings suggest that the lung epithelium barrier is more permeable following ozone exposures and this may facilitate the absorption of particles and/or inflammatory mediators from the lungs directly into the systemic circulation 9-12 . On the other hand, increased O x concentrations may influence the toxicity of particles themselves as photochemical aging has been shown to increase particle toxicity 13,14 .
While this study had a number of important strengths including a large number of mortality cases from multiple cities across Canada it is important to recognize several limitations. First, as in all epidemiological studies, exposure measurement error likely impacted our results as mean daily air pollution concentrations were assigned to case and control periods using fixed-site monitors at the city level. This error was likely most important for NO 2 exposures as within-city spatial variations are greater for NO 2 than for O 3 or PM 2.5 and fixed-site measurements may not adequately represent spatial differences in NO 2 exposures over large geographic areas. However, assuming that measurement errors are non-differential between case and control periods, this would usually result in an underestimation of risk estimates and is not a likely explanation of increased PM 2.5 mortality association in upper tertiles of O x . Satellite based air pollutants modeling have been proved as an effective method to accurately capture the spatial variability of ambient air pollution [17][18][19] . However, in this study, we could not obtain daily satellite-based air pollutant concentrations, as these are mainly available on a long-term basis across Canada 20 . In addition, we relied on data from 1998 to 2011 and thus more recent years are excluded from our analyses.
In summary, our results suggest that oxidant gases may act to strengthen associations between same day PM 2.5 exposures and nonaccidental and cardiovascular mortality. While these risks remain small, they suggest that the health benefits of reductions in O x concentrations may be larger than expected as such reductions may also decrease the health impacts of PM 2.5 even if mass concentrations remain unchanged. Statistical analysis. Conditional logistic regression models were used to estimate the association between short-term changes in ambient air pollutant concentrations and the risk of mortality 21 . All models pooled cases across cities using a random intercept at the city level to account for potential within-city correlations. We developed models for the whole year and separately for the warm season (April-September) in order to specifically capture the portion of the year with elevated O 3 concentrations . All ambient air pollutants (i.e. PM 2.5 , NO 2 , O 3 , and O x , as defined above) were evaluated in single pollutant models and all odds ratios (and 95% confidence intervals (CI)) reflect interquartile range (IQR) changes in pollutant concentrations. We used lag-0 IQR values for all statistical analyses since interquartile ranges were similar across exposure lag periods (within 1 µg/m 3 for PM 2.5 and within 1 ppb for NO 2 and O 3 ). We evaluated two different exposure periods for ambient air pollutants: lag-0 (the same day as the mortality event or the control period) and 3-day mean concentrations (including the day of the mortality even or the control period). As sensitivity analyses we also examined the time periods lag-1 (the day prior to the event or the control period) and lag-2 (two days prior to the event or the control period); the magnitudes of these associations were similar to or less than values for the main analyses and are not discussed further. Since the case-crossover design compares cases to themselves at different points in time it adjusts for factors that do not vary within . Odds ratios (ORs) 1 and 95% CIs for associations between lag 0 and 3-day mean PM 2.5 and respiratory mortality across tertiles (I, II, III) of same day, 3-day mean, and 5-day mean O x in 24 cities across Canada (1998-2011). ORs reflect a 6.63 µg/m 3 change in PM 2.5 . All models are adjusted for 3-day mean ambient temperature (cubic splines) and relative humidity and daily counts of hospitalization for influenza. individuals over short time-periods (e.g. age, smoking status, body mass index). In this study, the case period consisted of the day of the mortality event and control periods were selected on the same day of the week in the same month and year as the case period. This time-stratified approach to referent selection has been shown to result in unbiased conditional logistic regression estimates in case-crossover studies 23 . All models were adjusted for 3-day mean ambient temperature with a quadratic B-spline with three internal knots placed at the 10th, 75th, and 90th percentiles of location-specific temperature distributions and 3-day mean relative humidity 24 . To evaluate effect modification by O x in the relationship between acute exposure to PM 2.5 and mortality we conducted stratified analyses across tertiles of O x (<16.41 ppb, 16.41-<21.38 ppb, ≥21.38 ppb) based on the distribution of O x across all cities. We evaluated the statistical significance of effect modification by including a cross-product interaction term between PM 2.5 and the categorical variable for tertiles of O x . Wald's method was used to assess the presence of interaction on the multiplicative scale. Effect modification was considered statistically significant if the p-value for the interaction term was less than 0.05.
As sensitivity analyses, we investigated two-pollutant models to evaluate the extent to which PM 2.5 -mortality associations may be confounded by O x . We also evaluated effect modification of PM 2.5 -mortality associations across tertiles of NO 2 and O 3 . Finally, we investigated effect modification of O x in the warm season only. All statistical analyses were conducted with R software (version 3.2.4) using the packages dlnm and lme4.
Institutional Approvals. The use of the data in this study was approved by the Statistics Canada Policy Committee after consultation with the Statistics Canada Confidentiality and Legislation Committee, Data Access and Control Services Division, and the Federal Privacy Commissioner. This approval is equivalent to that of standard research ethics boards.