A Bayesian network model for analysis of detection performance in surveillance systems.

Area of Science:

• Public Health
• Epidemiology
• Biostatistics

Background:

• Global infectious disease and bioterrorism threats necessitate enhanced public health surveillance.
• Aberrancy detection algorithm performance is typically measured by sensitivity, specificity, and timeliness.
• These performance metrics are probabilistically dependent, creating a trade-off that is difficult to quantify.

Purpose of the Study:

• To develop and evaluate a Bayesian network framework for analyzing aberrancy detection algorithm performance measures.
• To provide a principled method for comparing different algorithms.
• To aid in identifying suitable algorithms for specific public health surveillance contexts.

Main Methods:

• Development of a Bayesian network framework.
• Evaluation of the framework using performance measures of aberrancy detection algorithms.
• Analysis of probabilistic dependencies between sensitivity, specificity, and timeliness.

Main Results:

• The Bayesian network framework allows for a quantitative analysis of the trade-offs between key performance metrics.
• The framework facilitates a principled comparison of diverse aberrancy detection algorithms.
• Demonstrated suitability of the framework for informing algorithm selection in public health surveillance.

Conclusions:

• A Bayesian network framework offers a robust approach to evaluating and comparing aberrancy detection algorithms.
• This quantitative framework addresses the limitations of fragmented and qualitative evidence on algorithm performance.
• The developed framework supports informed decision-making for public health surveillance system design and implementation.