Ideal anomaly detection algorithms should detect both sudden and gradual changes, while keeping the background false positive alert rate at a tolerable level. Further, the algorithm needs to perform well when the need is to detect small outbreaks in low-incidence diseases. For example, when surveillance is done based on the specific ICD9 diagnosis of flu rather than a larger syndromic grouping, the baseline counts will generally be low, in the range of 0 or 1 per day even in a large sample of EDs.
Objective
Our goal was to determine the sensitivity of detection of various inserted outbreak sizes and shapes using a modified Holt-Winters detection algorithm applied to daily flu count data before the flu season and after its peak. We compare our results to C3 of EARS.