Abstract
This article develops a detection framework using Bayesian philosophy by adaptation of Shiryaev's and Roberts' methodology. We propose two unifying versions directly applicable in industrial process control and easily extendable to public health infectious disease surveillance via some data detrending and/or demodulation. The root idea uses the sum of likelihood ratios upon which an optimal stopping criterion is based. It sets a prior on the epoch of a change, allows the flexibility to elicit a prior distribution on other process parameters, and attempts to minimize an expected loss function. A sensitivity analysis is conducted for validation and performance assessment and analytical formulas are derived. The methods are successfully applied to the European Union Centre for Disease Control (ECDC) open-source global COVID-19 incidence data. We further lay out scenarios where interest may switch to the detection of separate outbreaks with similar syndromes during an already evolving epidemic. We view our approach as a toolkit with a potential to augment early reports to sentinels in syndromic surveillance and in biosurveillance.
DISCLOSURE
The authors have no conflicts of interest to report.