An adaptive prediction and detection algorithm for multistream syndromic surveillance.

Area of Science:

• Public Health Surveillance
• Biostatistics
• Signal Processing

Background:

• Over-the-Counter (OTC) pharmaceutical sales data have been proposed as an early indicator for public health conditions and biosurveillance.
• This study extends previous work on estimating clinical data from OTC sales using linear and Finite Impulse Response (FIR) filters.

Purpose of the Study:

• To extend previous findings by predicting clinical data multiple steps ahead.
• To utilize both OTC sales and historical clinical data for enhanced prediction accuracy.

Main Methods:

• Employed a multichannel FIR filter incorporating past OTC categories and clinical data.
• Utilized recursive least squares method for adaptability to nonstationary data.
• Injected simulated events into data streams to evaluate prediction performance via Receiver Operating Characteristic (ROC) curves.

Main Results:

• Demonstrated the effectiveness of the combined filtering approach in predicting clinical data.
• Presented performance metrics of a detector based on the prediction outputs.
• Evaluated prediction accuracy using ROC curves with simulated transient events.

Conclusions:

• Multichannel adaptive FIR least squares filtering is a viable method for predicting public health conditions from OTC sales and clinical data.
• The potential value for biosurveillance requires further study, particularly with transient events.
• The proposed method, under specific conditions, can provide early warnings of clinical events using ancillary data streams like OTC sales.