Spatial targeting of infectious disease control: identifying multiple, unknown sources

Summary Geographic profiling (GP) was originally developed as an analytical tool in criminology, where it uses the spatial locations of linked crimes (for example murder, rape or arson) to identify areas that are most likely to include the offender's residence. The technique has been extremely successful in this field and is now widely used by police forces and investigative agencies around the world. More recently, the same method has been applied to biological data, notably in spatial epidemiology, where it uses the locations of disease cases to identify infection sources: the identification of these sources is critical to control efforts of diseases such as malaria, since targeted intervention is more efficient and cost‐effective than untargeted intervention. Here, we solve the problem of identifying multiple sources, even when the number of sources is unknown – a requirement for many biological studies. We present a new, rigorous mathematical and computational method and show why previous Bayesian methods were often outperformed by the empirically developed criminal geographic targeting (CGT) algorithm used in criminology. We use simulations and real‐world examples to compare our model to both the CGT algorithm and to an existing Bayesian model. We demonstrate that our method combines the advantages of both previous methods, particularly in cases featuring large data sets and multiple sources. Our approach provides an increase in search efficiency over other methods and is likely to lead to improved targeting of interventions and more efficient use of resources. We suggest that the Dirichlet process mixture (DPM) model provides a useful and practical tool for conservation biologists and epidemiologists that can be used to inform management decisions and public health policy.