Statistics Seminars: Robust Threshold Estimation for Outbreak Detection in Multiple Surveillance Systems
21 October 2013 14:00 in CM221
We revisit the quasi-Poisson regression-based surveillance algorithm used in England and Wales for
infectious disease outbreak detection with a view to improving its performance. Using extensive simulations, we study the sensitivity of the false reports to various modelling choices including treatment of trend, seasonality, error structure and the influence of past outbreaks. We improve the existing model by making use of much more data so that the trend and variance are better estimated. In addition, we recommend that the trend should always be fitted even when non-significant, decreasing the discrepancies in the results. We studied several alternative re-weighting schemes and found that the method based on scaled Anscombe residuals but with much higher threshold suitably reduces the false reports. Investigations also suggest that the negative binomial model is a reasonable one for defining the error structure, although not ideal in all circumstances. We find that the new system greatly reduces the number of false alarms while maintaining good overall performance and in some instances increasing the sensitivity. To improve the algorithm even further, we model reporting delays using splines and incorporate them in the system.
Contact email@example.com for more information