An evaluation and comparison of three commonly used statistical models for automatic detection of outbreaks in epidemiological data of communicable diseases

Abstract

We evaluated three established statistical models for automated ‘early warnings’ of disease outbreaks; counted data Poisson CuSums (used in New Zealand), the England and Wales model (used in England and Wales) and SPOTv2 (used in Australia). In the evaluation we used national Swedish notification data from 1992 to 2003 on campylobacteriosis, hepatitis A and tularemia. The average sensitivity and positive predictive value for CuSums were 71 and 53%, for the England and Wales model 87 and 82% and for SPOTv2 95 and 49% respectively. The England and Wales model and the SPOTv2 model were superior to CuSums in our setting. Although, it was more difficult to rank the former two, we recommend the SPOTv2 model over the England and Wales model, mainly because of a better sensitivity. However, the impact of previous outbreaks on baseline levels was less in the England and Wales model. The CuSums model did not adjust for previous outbreaks.