Reassessing the relationship between ozone and short-term mortality in U.S. urban communities.

Time-series studies that use daily mortality and ambient ozone concentrations exhibit estimates of ozone effects that are variable across cities. We investigate this intercity variability, as well as the sensitivity of the ozone- mortality associations to modeling assumptions and choice of daily ozone metric, based on reanalysis of data from the National Morbidity, Mortality and Air Pollution Study (NMMAPS). Previous work from NMMAPS reported a statistically significant association between ambient 24-h ozone and short-term mortality when averaged across 98 U.S. cities. Separation of ozone health associations from effects due to weather and co-pollutants is central to their interpretation. We examined the sensitivity of city-specific ozone-mortality estimates to adjustments for confounders and effect modifiers, showing substantial sensitivity. We examined ozone-mortality associations in different concentration ranges, finding a larger incremental effect in higher ranges, but also larger uncertainty. Alternative ozone exposure metrics defined by maximum 8-h averages or 1-h maxima show different ozone-mortality associations that cannot be explained by simple scaling relationships. The emphasis in earlier studies based on NMMAPS has been on the reporting of "national" effects, together with prediction intervals that suggest that these national values are precisely estimated. Our view is that ozone-mortality associations, based on time-series epidemiologic analyses of daily data from multiple cities, reveal still-unexplained inconsistencies and show sensitivity to modeling choices and data selection that contribute to serious uncertainties when epidemiological results are used to discern the nature and magnitude of possible ozone-mortality relationships or are applied to risk assessment.