Commonly used syndromic surveillance methods based on the spatial scan statistic first classify disease cases into broad, pre-existing symptom categories ("prodromes") such as respiratory or fever, then detect spatial clusters where the recent case count of some prodrome is unexpectedly high. Novel emerging infections may have very specific and anomalous symptoms which should be easy to detect even if the number of cases is small. However, typical spatial scan approaches may fail to detect a novel outbreak if the resulting cases are not classified to any known prodrome.

Disease surveillance data often has an underlying network structure (e.g. for outbreaks which spread by person-to-person contact). If the underlying graph structure is known, detection methods such as GraphScan (1) can be used to identify an anomalous subgraph which might be indicative of an emerging event. Typically, however, the network structure is unknown, and must be learned from unlabeled data, given only the time series of observed counts (e.g. daily hospital visits for each zip code).

Objective

The status of each Intensive Care Unit (ICU) patient is routinely monitored and a number of vital signs are recorded at sub-second frequencies which results in large amounts of data. We propose an approach to transform this stream of raw vital measurements into a sparse sequence of discrete events. Each such event represents significant departure of an observed vital sequence from the null distribution learned from reference data. Any substantial departure may be indicative of an upcoming adverse health episode.