PM health effects: responses to source-related exposures.

We review literature providing insights on health-related effects caused by inhalation of ambient air particulate matter. Many studies report associations between ambient PM concentrations and mortality/morbidity, but PM may originate from various emission source types and its toxicity may vary depending on its source and chemical composition. Pollution source apportionment has to do with the quantification of the contribution of one or more pollution emitting sources in determining the chemical composition of fine particulate matter at a given monitoring station. It is based on considering replicated measurements of the amounts of different chemical compounds in the atmosphere at a receptor and possibly at some previously defined sources. The former can be approximated by the sum of the PM composition at the sources weighted by the importance of the sources themselves, as stated by the chemical mass balance equation. In statistical terms this equation lends itself to different interpretations if knowledge of the amounts of chemical compounds at the sources is available or not. In the first case estimates of the source contributions are obtained by regression techniques and the problem is referred to as a chemical mass balance problem in the specialized literature. In the case that sources are unknown or uncertain, the so called multivariate receptor models rely on statistical factor analytic techniques to estimate the source-specific contributions from a large number of observed chemical concentrations. Examining the associations between source-apportioned PM and health outcomes may lead to more efficient and effective control strategies, rather than with PM mass overall. The representation of unobservable sources of pollution and the estimation of source-specific contributions in health effects analysis were addressed in two ways. The simplest characterization selects a set of tracer species to represent the known sources, and the measured concentrations of the tracers are taken as surrogates for the source contributions. As an alternative a 2-stage strategy uses source contributions from a previously estimated receptor model to assess the impact of pollution sources on health effects within a Poisson regression model. In both cases the statistical properties of health effects estimates are not well understood. In particular, neglecting the uncertainty associated with estimated contributions can lead to biased and unreliable estimates of regression coefficients. Structural equation models (SEM) were also recently proposed to assess source specific health effects using speciated air pollution data within designed laboratory experiments. The approach corresponds to jointly fitting a multivariate receptor model and a Poisson regression model for the health outcome given source contributions. Inferences on the health effects resulting from this joint modelling strategy account for the uncertainty associated with the exposure. In this framework switching from experimental to epidemiological data seems a key issue which is going to be addressed in the talk in connection with the possible use of SEM’s for public health surveillance