Early detection of disease outbreaks and non-outbreaks using incidence data: A framework using feature-based time...

Area of Science:

• Epidemiology
• Computational Biology
• Data Science

Background:

• Forecasting novel disease outbreaks is critical for public health management.
• Current methods often lack generalizability, require extensive preparation, and neglect non-outbreak prediction.
• A need exists for robust, proactive systems to anticipate both outbreaks and periods of low disease activity.

Purpose of the Study:

• To develop and validate a novel framework for forecasting both disease outbreaks and non-outbreaks.
• To utilize feature-based time series classification (TSC) for early detection of disease dynamics.
• To identify predictive statistical features and early warning signals in time series data.

Main Methods:

• A feature-based time series classification (TSC) framework was developed.
• Methods were tested on synthetic data from a Susceptible-Infected-Recovered (SIR) model.
• Performance was evaluated using area under the receiver-operating curve (AUC) on synthetic and empirical datasets (COVID-19, SARS).

Main Results:

• Incipient differences in time series data, captured by 22 statistical features and 5 early warning signal indicators, distinguish outbreak from non-outbreak sequences.
• Classifier performance (AUC) ranged from 0.7 to 0.99, depending on data window size.
• The framework demonstrated consistently high accuracy on real-world COVID-19 and SARS datasets.

Conclusions:

• Detectable statistical features can distinguish between sequences leading to outbreaks and non-outbreaks well in advance.
• The proposed TSC framework offers a promising approach for proactive disease surveillance and management.
• This method enhances the prediction of both disease emergence and absence, addressing a critical gap in current epidemiological tools.