Related Experiment Videos
Use of multiple data streams to conduct Bayesian biologic surveillance
Weng-Keen Wong1, G Cooper, D Dash
1RODS Laboratory, University of Pittsburgh, Pittsburgh, Pennsylvania 15213, USA.
MMWR Supplements
|September 24, 2005
Summary
This study introduces a new causal Bayesian model to combine emergency department (ED) and over-the-counter (OTC) sales data for improved outbreak detection. The model demonstrates efficient processing times for real-time public health surveillance.
Area of Science:
- Public Health Surveillance
- Epidemiology
- Biostatistics
Background:
- Syndromic surveillance commonly uses emergency department (ED) and over-the-counter (OTC) sales data separately.
- Existing algorithms often fail to integrate these data sources effectively for outbreak detection.
Purpose of the Study:
- To develop a novel causal Bayesian network model for coherent integration of ED and OTC data.
- To extend the Population-wide Anomaly Detection and Assessment (PANDA) algorithm for multi-stream data surveillance.
- To ensure the scalability of the model for real-time monitoring of large populations.
Main Methods:
- A causal Bayesian network was extended to incorporate daily OTC sales data.
- Individual-level actions leading to ED visits and OTC purchases were modeled.
- The approach was designed for scalability in real-time surveillance.
Main Results:
- The integrated model demonstrated a tractable running time: 209 seconds for initialization.
- Processing approximately 4 seconds per hour of ED data was achieved on standard hardware.
- Preliminary results indicate efficient performance for large-scale surveillance.
Conclusions:
- The new Bayesian algorithm effectively models the interaction between ED and OTC data.
- The algorithm shows promising results for efficient and accurate outbreak detection.
- The developed model is suitable for real-time public health surveillance applications.