A Framework for Prioritizing and Characterizing Statistical Anomalies Identified in Biosurveillance Systems

Background
CDC BioIntelligence Center personnel analyze and interpret national BioSense data daily.  Priority is given to diseases and geographic areas with known recent public health events.  Statistical anomalies are assessed according to criteria developed by the BioSense Monitoring Protocols Working Group.  Anomalies are categorized as confirmed (attributable to a known event after communication with state/local personnel) or unconfirmed (not attributable to known event; most represent random variation).  We summarize the criteria met by 9 confirmed anomalies during January 2007-April 2008.

Methods

The criteria used to assess data anomalies are: A) involves >1 patient class (outpatients, emergency department, inpatients) within the past week, B) increase is noted for ≥2 of the past 3 days, C) rate or count is the maximum within the past 6 months, D) exceeds rate or count during the same time period the previous year, and E) ≥1 patient also maps to the severe illness/death syndrome.

Results
The 9 anomalies involved gastrointestinal outbreaks, influenza and pneumonia related febrile and respiratory illness, and exposures to meningitis and rabies.  The 9 anomalies occurred on 38 days: one or more criteria was met on 33 (87%) days; 22 (53%) days met criteria B, 18 (47%) met criteria C, 15 (39%) met criteria D; and none met criteria A or E (total >100% since 15 days met multiple criteria).

Conclusions
Criteria B, C, and D, but not A and E, often occur in data anomalies attributable to a known event and may be valuable in prioritizing data anomalies.  Continued work and collaboration with state and local public health personnel is needed to improve methods for identifying potentially important data anomalies.