Empirical Bayes and Adjusted Estimates Approach to Estimating the Relation of Mortality to Exposure of PM10

Forward Links to Citing Articles Retraction. Risk Analysis 25: 6, 1683 In the framework of the APHEIS program (Air Pollution and Health: A European Information System), a health impact assessment of air pollution in 26 European cities was performed for particles of an aerodynamic diameter less than or equal to 10 μm (PM10). For short‐term effects, it was based on overall estimates from the APHEA‐2 project (Air Pollution and Health: A European Approach). These city‐specific risk assessments require city‐specific concentration‐response functions, raising the question of which concentration‐response is most appropriate. Estimates from city‐specific models are more specific, but have greater uncertainty than those provided from multicity analyses. We compared several estimates derived from the city‐specific analyses in cities that were part of the APHEA‐2 project, as well as in a city that was not included in APHEA‐2 but was part of the APHEIS project. These estimates were: the estimates from a local regression model, the adjusted estimates based on two significant effect modifiers identified through meta‐regression models, and the city‐specific empirical Bayes (shrunken) estimates and their underlying distribution. The shrunken and adjusted estimates were used to improve the estimation of city‐specific concentration‐response function. From these different estimates, attributable numbers of deaths per year were calculated. The advantages and limits of the different approaches are discussed through real data and in a simulation study.