National Morbidity, Mortality, and Air Pollution Study. Part IV: Hierarchical Bivariate Time-Series Models—A Combined Analysis of PM₁₀ Effects on Hospitalization and Mortality

In Part IV of the Morbidity, Mortality, and Air Pollution Study (NMMAPS), Dr Francesca Dominici and colleagues at Johns Hopkins Bloomberg School of Public Health addressed an important question resulting from the combined analysis of air pollution effects on mortality and on hospital admissions. Is the underlying true effect per unit PM₁₀ on mortality (the mortality slope) of the same magnitude as the effect per unit PM₁₀ on hospitalizations (the hospitalization slope) in a given city? Using a standardized analytic approach, the investigators conducted this study by using data from 10 cities with daily PM₁₀ monitoring and daily mortality and hospitalization data. They restricted analyses to deaths and hospitalizations due to cardiovascular diseases in residents 65 years of age and older. For each city, the investigators used methods they had developed earlier to evaluate the association between PM₁₀ concentration with mortality and with hospitalizations, separately. They then developed and applied a new method to estimate the correlation between the associations of PM₁₀ concentration with mortality and with hospitalization in each city. In a second stage of analysis, the investigators applied previously developed Bayesian hierarchical methods to estimate the correlation between the associations of PM₁₀ concentration with mortality and with hospitalization across all cities while accounting for variability due to sampling error.