Peak-Detection in Online Influenza Monitoring

• Activity Number: 113
• Type: Contributed
• Date/Time: Monday, July 30, 2007 : 8:30 AM to 10:20 AM
• Sponsor: Section on Statistics in Epidemiology
• Author(s): Eva Andersson*+ and Marianne Frisén and David Bock
• Companies: Göteborg University and Göteborg University and Astra Zeneca
• Keywords: surveillance ; on-line detection ; influenza ; likelihood ratio ; peak detection

Abstract

A non-parametric likelihood-ratio method (SRnpPeak) for peak detection is applied to Swedish influenza data. These data vary a lot between the years, with regard to the peak time, the peak height and the shape of the peak. The SRnpPeak method is derived from the optimality results of Shiryaev (1963). The method is non-parametric with regard to the shape of the peak, using previous results on ordered restricted regression. The behavior of the method is evaluated, both regarding the ability to detect peaks with different shapes, but also the effect of a larger variance than the one specified in the method. The results show that an influenza season with a low peak takes longer to detect (the delay of the alarm is longer). If the variance of the observed process is larger than the specified variance, the delay of an alarm is shorter, but the predictive value of an alarm is also lower.