Statistical Challenges in BioSurveillance

One goal in biosurveillance is to detect patterns in disease rates, such as temporal and/or geographic clustering. Traditionally, disease rates are available by geographic unit over weekly, monthly, or yearly time bins, and covariates such as age, gender, and socio-economic status can be used to adjust predicted rates prior to testing for clustering. Recently, more timely pre-diagnostic data including emergency department visits have been used in “syndromic surveillance” in order to more rapidly detect either natural or bioterrorist-related outbreaks. Typically, such data are categorized by chief complaint into one of several syndromes such as gastro-intestinal or respiratory. This chapter describes outbreak detection using either traditional diagnosed case rates or syndromic surveillance data. Outbreak detection involves many issues; our focus is the associated statistical challenges, including: (1) approaches to characterizing the natural background; (2) algorithms for detecting abnormal increases above background disease rates, (3) methods for adjusting for covariates such as gender, age, etc.; (4) detecting spatial-temporal clusters, and (5) methods for protecting data confidentiality.