Syndromic surveillance involves the analysis of time series of health indicators to identify changes in disease patterns. To this end, statistical modeling is used to reduce systematic data variation. Still, there is variation that cannot be accounted for in this approach, e.g. mass gatherings, extreme weather and other high-profile events. To filter sporadic events, data transformation can be applied, e.g. proportion data from correlated data streams (Peter, Najmi and Burkom, 2011; Reis, Kohane and Mandl, 2007). However, we lack systematic criteria for applying data transformations, e.g. ratios versus geometric means. To develop guidelines, we conducted a power analysis and compared the results with empirical findings (Andersson et al, 2013).
Objective
For the purpose of optimizing baselines for point-source outbreak detection, we carried out a power analysis of the effects of data transformations. More specifically, the aim was to develop statistical criteria for using composite baselines, i.e. ratios and geometric means of data streams. The results were validated by outbreak data on acute gastroenteritis (The Swedish National Telephone Health Service 1177).