Peter J Diggle (Lancaster University) will present this seminar on Thursday‚ 19th January 2012 4:00pm – Building 54 Room 4011 (4A)
To reserve your place‚ please email [email protected]
Abstract:
A fundamental problem in many epidemiological studies of endemic disease is to understand the spatial variation in disease prevalence. In developed country settings, so-called disease mapping typically uses registry data, consisting of small-area case-counts and population denominators throughout the geographical region of interest.
In resource-poor, developing country settings registry data are typically not available and disease mapping instead utilises data from prevalence surveys at scattered locations. The term \”geostatistics\” refers to the branch of spatial statistics that uses spatially discrete sampled data to make inferences about a spatially continuous phenomenon. It is therefore well-suited to disease mapping problems in resource-poor settings. Traditionally, geostatistical models and methods have been presented as a self-contained toolbox for spatial prediction problems. The prefix \”model-based\” was coined by Diggle, Moyeed and Tawn (1998) to mean the embedding of geostatistical methods within the mainstream of statistical modelling and inference. The principle advantage of the model-based approach is that it shifts the focus from a collection of techniques to a class of scientific problems. In this talk, I will briefly review the main elements of the model-based approach to geostatistical problems.
The bulk of the talk will then use two case-studies in tropical disease epidemiology to demonstrate the flexibility of the approach: prevalence mapping of the endemic disease Loa loa in equatorial Africa; and short-term forecasting of localised meningitis epidemics in sub-Saharan Africa. Diggle, P.J., Moyeed, R.A. and Tawn, J.A. (1998). Model-based geostatistics (with Discussion). Applied Statistics, 47, 299-350.