A statistical algorithm for outbreak detection in multisite settings: an application to sick leave monitoring

Area of Science:

• Epidemiology and Public Health Surveillance
• Statistical Modeling for Outbreak Detection
• Biostatistics

Background:

• Public health surveillance relies on algorithms to detect unusual increases in health-related cases, known as aberrations, to signal potential outbreaks.
• Existing methods, like the Farrington Flexible algorithm, primarily focus on detecting aberration *times* but lack location-specific detection capabilities.
• The growing volume and diversity of epidemiological data, coupled with emerging epidemic threats, necessitate more sophisticated surveillance algorithms for both temporal and spatial aberration detection.

Purpose of the Study:

• To develop an enhanced aberration detection algorithm for multisite surveillance, extending the capabilities of the quasi-Poisson regression Farrington Flexible algorithm.
• The primary goal is to improve the accuracy of outbreak detection by identifying not only when but also where aberrations occur.
• To incorporate sick leave data monitoring across companies as a practical application for identifying company-specific aberrations.

Main Methods:

• Development of a novel algorithm based on a negative-binomial mixed effects regression model, incorporating a random effects term for different sites.
• Introduction of a new reweighting procedure designed to mitigate the impact of past aberrations on current detection sensitivity.
• Implementation and validation using simulations and real-world sick leave data during the COVID-19 pandemic, performed in R statistical software (glmmTMB package).

Main Results:

• Simulations demonstrate that the new algorithm offers improved false positive rates compared to the Farrington Flexible algorithm, while maintaining similar detection probabilities for outbreaks exceeding 3 standard deviations above baseline.
• The algorithm achieves higher detection rates for significant surges, reaching 100% detection when case counts exceed eight baseline standard deviations.
• Application to COVID-19 sick leave data successfully identified the pandemic's effect, showcasing its real-world applicability and effectiveness in multisite surveillance scenarios.

Conclusions:

• The proposed negative-binomial mixed effects model with site-specific random effects and a reweighting procedure represents a significant advancement in multisite aberration detection.
• The algorithm provides enhanced performance, particularly in reducing false positives and improving the detection of true outbreaks across diverse data scenarios.
• This new approach offers valuable perspectives and practical implementation for modern public health surveillance systems facing complex epidemiological challenges.