Using combined diagnostic test results to hindcast trends of infection from cross-sectional data

We have developed a Bayesian approach that can estimate the historic trend of incidence from cross-sectional samples, without relying on ongoing surveillance. This could be used to evaluate changing disease trends, or to inform outbreak responses. We combine two or more diagnostic tests to estimate the time since infection for the individual, and the historic incidence trend in the population as a whole. We evaluate this procedure by applying it to simulated data from synthetic epidemics. Further, we evaluate its real-world applicability by applying it to two scenarios modelled after the UK 2007 bluetongue epidemic, and a small outbreak of whooping cough in Wisconsin, USA. We were able to recover the epidemic trends under a range of conditions using sample sizes of 30–100 individuals. In the scenarios modelled after real-world epidemics, the hindcasted epidemic curves would have provided valuable information about the distribution of infections. The described approach is generic, and applicable to a wide range of human, livestock and wildlife diseases. It can estimate trends in settings for which this is not possible using current methods, including for diseases or regions lacking in surveillance; recover the pattern of spread during the initial “silent” phase once an outbreak is detected; and can be used track emerging infections. Being able to estimate the past trends of diseases from single cross-sectional studies has far-reaching consequences for the design and practice of disease surveillance in all contexts.