Re: “Is the Relation Between Ozone And Mortality Confounded by Chemical Components of Particulate Matter? Analysis of 7 Components in 57 us Communities”

We read with interest the article by Anderson et al. (1) in which the authors investigated whether the ozone-mortality relationship is confounded by 7 components (sulfate, nitrate, silicon, elemental carbon, organic carbon matter, sodium, and ammonium) of particulate matter with an aerodynamic diameter less than or equal to 2.5 μm (PM_2.5) across 56–57 communities in the United States. They concluded that the ozone-mortality relationship was robust to control for these common PM_2.5 components, which make up most of the PM_2.5 mass. However, there are several study limitations that should be addressed to determine how robust the study findings are.

First, Anderson et al. (1) pooled estimates across all communities and seasons with available data (34 out of 56 or 57) even though, as they noted, there is both spatial and temporal heterogeneity in ozone concentrations and PM_2.5 composition. This heterogeneity was demonstrated by Bell et al. (2), who reported stronger correlations between PM_2.5 and ozone in certain regions of the United States (e.g., the Midwest and Northeast) and during certain seasons (e.g., spring and summer), as well as differences in mortality effect estimates across regions and seasons. Similarly, Katsouyanni et al. (3) found evidence of confounding by particulate matter (PM) in summer-only analyses of US cities that was much less evident in the year-round analyses. Franklin and Schwartz (4) limited their analysis to the summer months, when PM_2.5 and ozone levels are higher, and demonstrated that the confounding effects of sulfates on ozone mortality effect estimates occurred differentially across communities. This might be due to sulfate-enriched PM_2.5 in some areas in the Midwest and Northeast, where the correlations between PM and ozone are also strongest (2). Because pooling estimates across all communities can mask heterogeneity, the relative contributions of PM components in each community and how correlations differ depending on PM composition are unknown. Stronger correlations in some communities and during certain seasons may be why confounding is not observed for pooled estimates.

Second, Anderson et al. (1) did not show whether the PM_2.5 components were independently associated with mortality. If the PM_2.5 components are not associated with mortality, it is not likely that they confounded the ozone-mortality association.

Third, Katsouyanni et al. (3) found that ozone-mortality associations in the United States were sensitive to time and weather model specifications, but Anderson et al. (1) did not conduct sensitivity analyses using alternative model specifications with respect to these factors. Thus, the impact on results of varying these different model specifications is unknown.

Lastly, with regard to uncertainty, Anderson et al. (1) acknowledged that their results were subject to exposure measurement error, but they did not indicate the impact of this on their results. Studies have shown that ambient ozone concentrations are poorly correlated with personal exposures (5, 6). Each PM component also has specific ambient-personal correlations, adding a level of complexity that was not addressed in the analysis by Anderson et al.

Overall, although the article contributes much-needed research, the analysis should be expanded to evaluate confounding by region and season, include correlations between PM components and mortality, conduct sensitivity analyses for different model specifications, and address measurement error.