Simple epidemic models with segmentation can be better than complex ones.

Area of Science:

• Epidemiology
• Mathematical Modeling
• Computational Biology

Background:

• Infectious diseases pose significant threats, necessitating accurate understanding and forecasting of epidemic dynamics.
• Ordinary differential equation (ODE) based epidemic models are widely used but often fail to capture long-term dynamics influenced by external factors.
• External factors like lockdowns and testing capabilities complicate epidemic modeling, challenging the efficacy of single, complex models.

Purpose of the Study:

• To investigate whether segmenting epidemic event sequences improves model fit compared to using a single model for the entire period.
• To develop a methodology for optimizing the balance between the number of segments and epidemic model complexity.
• To demonstrate an effective approach for describing and forecasting infectious disease spread.

Main Methods:

• Dividing COVID-19 case, recovery, and death data into multiple segments.
• Fitting simple epidemic models to individual segments rather than a single complex model to the entire dataset.
• Applying the Minimum Description Length (MDL) principle to balance segmentation and model complexity.

Main Results:

• Fitting simple epidemic models to segmented data provides a better fit with fewer parameters than a single complex model.
• The proposed methodology is automatic, requiring no user-defined parameters.
• The approach is model-agnostic, applicable to various ODE-based epidemic models.

Conclusions:

• Segmenting epidemic data and employing multiple simple models is superior to using a single complex model for long-term dynamics.
• The developed methodology offers an effective, automatic, and versatile approach for epidemic analysis and forecasting.
• The method was successfully applied to describe and forecast COVID-19 spread across 70 countries.