The improved quality of hospital admissions data makes them a valuable resource for researchers. However, we show that, when multiple providers of health care are considered, the provider (in this case hospital) may act like other confounders such as socio-economic status, and hence must be controlled for. Such control is not as straightforward as for conventional confounders, however, but we describe a method that is appropriate under certain assumptions. We also describe a number of other statistical issues, such as the modelling of spatial and non-spatial overdispersion, that arose during the use of hospital data in a study to investigate the possible adverse health effects of living in proximity to six cokeworks groups in England and Wales. The outcome data that we consider consist of hospital admissions for all respiratory disease in the under-5s. The ecological level of the analysis is the census-defined enumeration district, and the main (proxy) exposure measure utilised is the spatial location of the enumeration district population-weighted centroid in relation to the cokeworks. We focus on the Teesside cokeworks group, for which we also had sulphur dioxide measurements from dispersion modelling as an alternative exposure measure. The major local providers varied appreciably in their standardized admission ratios for respiratory disease, and when provider was controlled for, the size of the observed excess risk found close to the cokeworks was decreased, making control for the provider of health care vital. However, the presence of multiple pollution sources, in addition to the usual shortcomings of ecological studies, makes interpretation difficult.