Biosurveillance systems commonly depend on free-text chief complaints (CC)s for timely situational awareness. However, diagnosis codes may not be available soon enough and may have uncertain value because they are assigned for billing purposes rather than for population monitoring. Existing systems use syndrome categories to classify records based on these free-text fields. A syndromic cluster determination method (TOA) based on patient arrival times has been implemented in versions of ESSENCE and in NCDETECT [1]. While effective for finding case clusters whose CC terms are classifiable into syndromes, TOA implementations do not find clusters whose CC terms share only uncategorized terms.
Objective
Explain and demonstrate the performance of a statistical method for detection of anomalous terms in pooled, contiguous blocks of freetext chief complaints from a health facility with emergent or urgent care capability.