Adaptive Threshold Methods for Monitoring Rates in Public Health Surveillance

Abstract

We examine one of the methods implemented by the U.S. Centers for Disease Control and Prevention’s (CDC) BioSense program. The program uses data from hospitals and public health departments to detect outbreaks using the Early Aberration Reporting System (EARS). The EARS W2r method allows one to monitor the proportion of counts of a particular syndrome at a facility relative to the total number of visits. We investigate the performance of the W2r method with negative binomial inputs designed using an empirical recurrence interval (RI). An adaptive threshold monitoring method is studied based on estimating the underlying negative binomial distributions, then converting the current counts to a Z-score through a p-value. We study the effect of the input distributions on the upper thresholds required for both Shewhart and exponentially weighted moving average (EWMA) versions of the W2r and adaptive threshold methods. We simulate 1-week outbreaks and compare the outbreak detection properties of the methods.