Dynamical footprints enable detection of disease emergence

Area of Science:

• Epidemiology
• Computational Biology
• Machine Learning

Background:

• Predicting infectious disease emergence is critical but challenging due to uncertainties in disease drivers.
• Traditional model-based approaches often struggle with complex epidemiological dynamics.

Purpose of the Study:

• To develop and validate a mechanism-agnostic early warning system (EWS) for detecting disease (re-)emergence.
• To leverage critical slowing down theory and machine learning for enhanced outbreak prediction.

Main Methods:

• Trained a supervised learning algorithm on computer simulations to identify dynamical footprints of disease (re-)emergence.
• Applied the algorithm to epidemiological data, including mumps, pertussis, dengue, and plague case studies.
• Utilized early warning signals (EWSs) derived from critical slowing down theory.

Main Results:

• Successfully forecasted mumps reemergence in England four years in advance.
• Identified resurgent pertussis states in the US with increasing accuracy from 1980, achieving reliable classification by 1992.
• Demonstrated efficacy in vector-transmitted diseases, including dengue outbreaks in Puerto Rico and a plague outbreak in Madagascar.

Conclusions:

• Theoretically informed machine learning provides a powerful tool for developing robust early warning systems for infectious diseases.
• This approach offers a promising strategy for proactive public health management and intervention planning.
• The mechanism-agnostic nature of the algorithm enhances its applicability across diverse disease contexts.