Bayesian Spatial Health Surveillance

1 Two important problems Clustering of disease: PART 1  Development of Space-time models  Modelling vs Testing  Hidden process models  Example from Scotland  Future directions Relative risk change detection: PART 2  What is meant by surveillance  Statistical aspects of surveillance  Bayesian models space-time risk estimation  discussion 2 Clustering of disease What is a cluster ?  No universally accepted definition  A working (spatial) definition was given by Knox (1989) 'A geographically bounded group of occurrences of sufficient size and concentration to be unlikely to have occurred by chance' 3  In space-time slightly more complicated since we can have 3 different types of clusters.  Temporal cluster: occurs in the whole study region, but for a limited time.  Spatial cluster: occurs during the whole study period, but in a small area.  Spatio-temporal cluster: should exist in a small area and for a limited time. All have different properties, however some common concerns are  adjustment for multiple testing results in loss of power  how to deal with covariates, stratification is wasteful. Better to fit a model  no adjustment needed for multiple testing  deals with covariates in 'traditional' way  flexibility 5 Hidden Process Model development  From definition of clustering we expect clustering to be a localized phenomenon and occur around some 'centre'.  'Centre' can be any shape – most commonly a point or a line  Point may reflect the location of a factory or a waste processing site  Line may reflect the location of a highway or a river  Centres not directly observed, they form a hidden process 6  Observed cases form a realisation of a hetrogeneous Poisson process with intensity given by  x, t  space and time coordinates  gx, t  background rate due to 'population at risk'   T z  is the linear predictor of covariates  c 1  spatial cluster centres  c 2  temporal cluster centres  c 3  spatio-temporal cluster centres 7  Different choices of the function m result in different models. We normally use mc 1 , c 2 , c 3 , x, t  1   1  i1 n s Kx  c 1i    2  i1 n t Kt  c 2i    3  …