Detection of Outbreaks in Syndromic Surveillance Data Using Monotonic Regression

Due to nonstationarity and substantial variability in outbreak profiles, early detection of disease outbreaks is challenging. In this paper we propose a method to detect outbreaks in syndromic surveillance data using a generalized likelihood ratio test in which both the null and alternative hypotheses are normally distributed. The data is daily counts of interactions between patients and the National Bioterrorism Syndromic Surveillance Demonstration Program System in the Boston area. Using Poisson regression, we estimate the daily means and variances of the data as well as day of the week variations. The estimated means serve as the means under the null hypothesis. To determine the means under the alternative hypothesis we use a generalized form of the Pool-Adjacent-Violators algorithm on five-day windows of data. For each window a test statistic is computed and an outbreak is indicated if it exceeds a threshold.