Outbreak detection algorithms for seasonal disease data: a case study using Ross River virus disease

Area of Science:

• Epidemiology
• Public Health Surveillance
• Infectious Disease Dynamics

Background:

• Disease surveillance relies on effective outbreak detection algorithms.
• Vector-borne diseases, such as Ross River virus, often exhibit distinct seasonal patterns.
• Assessing algorithm performance against seasonal disease data is crucial but underexplored.

Purpose of the Study:

• To evaluate the performance of five outbreak detection algorithms using Ross River virus disease data.
• To compare the accuracy and timeliness of different algorithms, including seasonally-adjusted variants.
• To identify the most effective algorithms for detecting outbreaks in seasonal disease surveillance.

Main Methods:

• Applied five algorithms (EARS, NBC, HLM, POD, SaTScan) to Ross River virus data (1991-2007) from four Queensland regions.
• Developed and tested seasonally-adjusted Negative Binomial Cusum (NBC) and SaTScan methods.
• Utilized 17 algorithm variants by adjusting parameter values for comprehensive assessment.

Main Results:

• Marked seasonality in Ross River virus data impacted some algorithm performances.
• Negative Binomial Cusum (NBC), Poisson Outbreak Detection (POD), and temporal SaTScan demonstrated high true positive rates and low false positive/negative rates.
• Timeliness of outbreak detection varied inconsistently across methods, outbreaks, and regions.

Conclusions:

• Seasonal patterns in disease data pose challenges for standard outbreak detection algorithms.
• NBC, POD, and temporal SaTScan methods are recommended for detecting outbreaks in seasonal disease surveillance.
• Caution is advised when interpreting algorithm performance metrics (true/false positives, sensitivity, specificity) without a definitive gold standard.