Assessing Psychiatric Morbidity from Community Registers: Methods for Bayesian Adjustment

Abstract

This paper considers Bayesian approaches to adjusting small area prevalence estimates derived from a community register of the seriously mentally ill, by taking account of underlying variability in latent prevalence between areas. Adjustment of individual prevalence rates to take account of the entire spatial distribution has implications both for epidemiological inference and resource rankings for localities. The more commonly adopted empirical Bayes approaches are here compared with fully Bayes approaches. Fully Bayes methods allow for uncertainty in the prevalence parameters and also permit the inclusion of information from other sources (for example, hospital admissions for mental illness) or from previous studies. The impact of socioeconomic indices on morbidity and its relevance for adjusted prevalence estimates is also considered. A case study is applied to an East London Health Authority, Barking and Havering.

Socio-economic status and geographies of psychiatric inpatient service use. Places, provision, power and wellbeing

“…There is now a large international literature that makes use of ecological statistical modelling to test the population level associations, at the scale of local areas, between rates of use of psychiatric services and the socio-economic characteristics of the population. (This brief review does not allow space for a comprehensive review, so the reader is invited to see more detailed overviews by, for example, Tansella et al, 1993;Congdon et al, 1998;Curtis et al, 2006a). This literature is an important source of information to address the questions posed in this journal issue, so it is interesting here to consider its potential and limitations.…”

“…Alternatives may be available in the application of Bayesian models used in the 'new generation' of ecological analyses (e.g. Congdon et al, 1998). These permit modelling of ecological data based on small and variable numbers of cases of service use at the local level (and can deal with zero counts in a proportion of the geographical units).…”

Quantitative methods: Bayesian inference, Bayesian thinking

“…Empirical Bayes estimation treats all parameters as fixed but unknown and uses the data to estimate them (Carlin and Lewis, 1996). Empirical Bayes methods are particularly effective in depicting spatial demography (Marshall, 1991;Devine et al, 1996;Pringle, 1996;Congdon, 1997;Congdon et al, 1998;Cressie et al, 2000;Carlin and Lewis, 2000). Still, maps generated by empirical Bayes methods have important epidemiological and political consequences.…”