Active learning to understand infectious disease models and improve policy making.

Area of Science:

• Computational epidemiology
• Health policy modeling
• Machine learning applications in public health

Background:

• Infectious disease models are crucial for policy but often complex and computationally demanding.
• Systematic exploration is necessary for a comprehensive understanding of these complex systems.
• Existing methods may struggle with high dimensionality and correlated inputs in policy models.

Purpose of the Study:

• To present an active learning approach using machine learning for systematic analysis of complex model behaviors.
• To demonstrate the application of iterative surrogate modeling and model-guided experimentation.
• To improve understanding, reduce uncertainty, and enhance policy-making for infectious disease interventions.

Main Methods:

• Utilized active learning integrating machine learning techniques, specifically iterative surrogate modeling and model-guided experimentation.
• Employed symbolic regression for nonlinear response surface modeling with automatic feature selection.
• Applied the approach to an individual-based influenza vaccination model and a deterministic dynamic model for varicella-zoster virus vaccination cost-effectiveness.

Main Results:

• Demonstrated an inverse relationship between vaccination coverage and cumulative attack rate, reinforced by herd immunity, in influenza models.
• Identified influential variables and handled high dimensionality/correlated inputs using symbolic regression in varicella-zoster virus models.
• Showcased the ability of surrogate models to be explored at no computational expense and act as emulators for policy decisions.

Conclusions:

• Active learning is essential for fully understanding complex systems behavior in infectious disease modeling.
• Surrogate models offer a computationally inexpensive method for exploring complex model outputs.
• This approach can significantly improve rapid policy-making by reducing dimensionality and decision uncertainty.