Recently published studies evaluate statistical alerting methods for disease surveillance based on detection of modeled signals in a data background of either authentic historical data or randomized samples. Differences in regional and jurisdictional data, collection and filtering methods, investigation resources, monitoring objectives, and systemrequirements have hindered acceptance of standard monitoring methodology. The signature of a disease outbreak and the baseline data behavior depend on various factors, including population coverage, quality and timeliness of data, symptomatology, and the careseeking behavior of the monitored population. For this reason, statistical process control methods based on standard data distributions or stylized signals may not alert as desired. Practical algorithm evaluation and adjustment may be possible by judging algorithmperformance according to the preferences of experienced human monitors.
Objective
This presentation gives a method of monitoring surveillance time series on the basis of the human expert preference. The method does not require detailed history for the current series, modeling expertize, or a well-defined data signal. It is designed for application to many data types and without need for a sophisticated environment or historical data analysis.