CP Jewell, MJ Keeling and GO Roberts
Predicting infected but undetected infections during the 2007 Foot-and-Mouth Disease outbreak
Date: April 2008
Abstract: Active disease surveillance during epidemics is of utmost importance in detecting and eliminating new cases quickly, and targetting such surveillance to high-risk individuals is generally more efficient than applying a random strategy7;12. Contact tracing has been used as a form of at-risk targetting, and a variety of mathematical models have indicated that it is likely to be highly efficient4;14. However for fast-moving epidemics, resource constraints limit the ability of the authorities to perform, and follow up, contact-tracing effectively. Here, as an alternative, we present a real-time Bayesian statistical methodology to determine currently undetected (occult) infections10. For the UK Foot-and-Mouth epidemic of 2007, we use real-time epidemic data synthesised with previous knowledge of FMD outbreaks in the UK to predict which premises might have been infected, but remained undetected, at any point during the outbreak. This provides both a framework for targetting surveillance in the face of limited resources, and also an indicator of the current severity and spatial extent of the epidemic. Our results demonstrate how statistical approaches to real-time prediction can yield meaningful results in the face of small amounts of data and, therefore, a high degree of statistical uncertainty. This methodology thus provides a compromise between targetted manual surveillance and random or spatially targetted strategies. We anticipate such methods to be of substantial benefit in future outbreaks of livestock diseases, providing a rigorously quantitative base for the assessment of control measures, targetting of disease surveillance, and assessment of the current extent of the outbreak.