trendbreaker

trendbreaker implements the ASMODEE method in an R package to detect sudden changes in COVID-19 incidence by automatic time-series model selection and outlier detection for epidemiological surveillance.

Key Features:

• Model selection: Automatically selects the best-fitting or predictive model from a predefined set of time series models while excluding recent data points to characterize historical incidence trends.
• Outlier detection via prediction intervals: Derives prediction intervals and flags data points falling outside these intervals as outliers indicating deviations from established trends.
• Optimal breakpoint selection: Determines optimal breakpoints that define the number of recent data points to exclude during trend fitting for robust analysis.
• Recent-data exclusion: Excludes the most recent data points to avoid bias from short-term fluctuations when fitting historical trends.
• Validation: Validated through simulations of COVID-19 outbreaks and compared to existing outbreak-detection algorithms.

Scientific Applications:

• Epidemiological surveillance: Automated analysis of COVID-19 incidence time series to detect emerging localized outbreaks and changes in transmission dynamics.
• Public health interventions: Provides objective indicators of sudden trend changes to inform timely public health responses to emerging hotspots.

Methodology:

Excluding recent data points; automatic selection from a predefined set of time series models; calculation of prediction intervals to identify outliers; determination of optimal breakpoints; validation using simulations of COVID-19 outbreaks and application to NHS Pathways data in England.