The Authors Reply

We appreciate Sax and Goodman's interest (1) in our study (2). Although controlled experimental studies have shown that ozone pollution can harm human health (3, 4), observational studies are critical to quantify the impact of real-world ambient ozone on human health and to provide guidance to decision-makers who are establishing health-based ozone regulations. Observational ozone-health studies are, like all observational studies, subject to potential confounding. Our study adds to research (5–7) that investigated whether measured ozone-mortality associations are sensitive to particulate matter pollution. As in previous studies in which investigators found no sensitivity of the ozone-mortality relationship to particulate matter with an aerodynamic diameter less than or equal to 10 μm or less than or equal to 2.5 μm (PM_2.5) measured by mass (5, 6), we found no evidence that 7 PM_2.5 components confounded the ozone-mortality associations observed across numerous US communities (2).

Sax and Goodman raise concerns about estimating the ozone-mortality association (1) that we welcome the opportunity to address. First, although there is regional and temporal heterogeneity across the United States in concentrations of ozone and PM_2.5 and in the composition of PM_2.5 (8), this heterogeneity does not prevent combining of community-level estimates of ozone-mortality associations or affect the interpretation of such results. Additionally, our study presented community-level results, including histograms of community-level correlations between ozone and PM_2.5 components (Figure 1 in our original article (2)) and community-level measurements of confounding (Figure 2 in our original article). These results show, at most, low to moderate correlations between each PM_2.5 component and ozone and indicate little or no evidence that our overall estimates mask substantial community-level confounding within certain communities.

As Sax and Goodman note, a smaller previous study in which analysis was limited to summer months showed some evidence of confounding by sulfate (7), but this result was less relevant to our analysis, in which we used year-round data. We chose to conduct our analysis using year-round data to correspond with the modeling process used to estimate the ozone-mortality association in several national-scale US studies (5, 9, 10). Using models, communities, and times of year similar to those considered in other studies, we found no evidence that results from previous studies were sensitive to model control for 7 components of PM_2.5.

Second, it is true that if certain PM_2.5 components are not associated with mortality, they would not confound the ozone-mortality relationship. However, this question does not affect the validity of the analysis performed here. Regardless of why these PM_2.5 components do not confound the ozone-mortality relationship, we found no evidence of confounding. Although quantifying the associations between PM_2.5 components and mortality risk is very important, it was beyond the scope of our study and was not our focus.

Third, in an extensive study of 95 US communities, models almost identical to those used in our study were found to be robust to a variety of changes in model control for long-term and seasonal mortality trends and weather (5). Finally, measurements of ambient pollution may not be perfectly correlated with personal exposure (11, 12). Although exposure measurement error can bias estimated associations between pollutants and health, the only cases in which the true effect of one pollutant is likely to be attributed to a second pollutant because of exposure measurement error are 1) when concentrations of the 2 pollutants are strongly positively correlated; 2) when exposure measurement errors of the 2 pollutants are strongly negatively correlated; and 3) when the pollutant with a true effect is measured with much less certainty than the pollutant with no effect (12). It is unlikely that these conditions were met in our study, and so our findings were likely not qualitatively affected by exposure measurement error. A more likely repercussion is that the ozone-mortality associations estimated in our and other studies that use ambient ozone data underestimate the true association between ozone and mortality risk (12). Future research could explore these questions more thoroughly.