Abstract
Observations on infectious diseases often consist of a sample of cases, distinguished by symptoms, and other characteristics, such as onset dates, spatial locations, genetic sequence of the pathogen and/or physiological and clinical data. Cases are often clustered, in space and time, suggesting that they are connected. By defining kernel functions for pairwise analysis of cases, a matrix of transmission probabilities can be estimated. We set up a Bayesian framework to integrate various sources of information to estimate the transmission network. The method is illustrated by analysing data from a multi-year study (2002-2007) of nosocomial outbreaks of norovirus in a large university hospital in the Netherlands. The study included 264 cases, the norovirus genotype was known in approximately 60 per cent of the patients. Combining all the available data allowed likely identification of individual transmission links between most of the cases (72%). This illustrates that the proposed method can be used to accurately reconstruct transmission networks, enhancing our understanding of outbreak dynamics and possibly leading to new insights into how to prevent outbreaks.