Environmental benzene and risk of childhood leaukaemia: a case study in applied Bayesian modelling

 

The field of spatial epidemiology has benefited from a number of methodological advances in Bayesian modelling, starting with the paper of Besag et al (1991) on Markov random field priors for disease mapping, through to more recent work on models for multivariate spatial lattice processes (e.g. Gelfand and Vounatsou, 2003; Best et al, 2005), spatial random coefficient models (e.g. Assuncao, 2003), geoadditive regression models (e.g. Fahrmeir and Lang, 2001) and ecological regression models combining individual and areal data (e.g. Wakefield and Salway, 2001; Jackson et al, 2006; Haneuse and Wakefield, 2006). Development of techniques for criticising and comparing hierarchical Bayesian models (Spiegelhalter et al, 2002; Marshall and Spiegelhalter, 2003) have also facilitated the application of these methods to real-world problems.

In this talk, I will present a case study in spatial epidemiology to illustrate many of the above methodological developments. The case study concerns an investigation of the risk of childhood leukaemia associated with exposure to environmental benzene. Leukaemia is one of the most prevalent forms of cancer in children, but its aetiology remains largely unexplained. In adults, occupational exposure to benzene (e.g. in the petrochemical industry) is one of the few established risk factors. Recent concern has focused on whether continuous exposure to low concentrations of atmospheric benzene (arising primarily from petrol vehicle exhaust emissions) also presents a leukaemogenic hazard, particularly in vulnerable groups such as children. Data used for this analysis include postcoded incident cases of leukaemia between 1985 and 2001 from the Thames Cancer Registry (covering the London region) , census population counts, and modelled benzene concentrations for each London postcode for three time periods spanning 1985-1990, 1991-1996 and 1997-2001.

Joint work with Chris Jackson, Kees de Hoogh