The semi-individual study in air pollution epidemiology: a valid design as compared to ecologic studies

The assessment of long-term efects of air pollution in humans relies on epidemiologic studies. A widely used design consists of cros-sectional or cohort studies in which ecologic a n of exposure, based on a fixed-site ambient monitor, is employed. Alhough health outcome'an usu- ally a large number of covarates are measured in individuals, these studies are ofien called ecolog- ical. We will inrdu the term semi-individual design for these studies. We review the major properties and limittiowithregad xto caus inrence of tr ecologic stuies, in Which out- come, exposure, and c iatesaavailable on an egate level only. Mislassification problems and issues related to unding ant model spedflcatio n truly ecologic studies limit etiologic inference to individuals. In contrast, the semi-individual study shares its methodological and inferential propeties th typical individual-level study designs. The major caveat relates to the case where too few study area, eg o or three, are used, which render conl ofaggregate kvel confounding imposble. The issue of exposure sifion is of genera concern in epidemi- ology and not an exdusive problem of the semi-individual design. In a multicenter setting, the