Assessing parameter identifiability in compartmental dynamic models using a computational approach: application to...

Area of Science:

• Epidemiology
• Mathematical Biology
• Computational Science

Background:

• Mathematical models are vital for understanding infectious disease dynamics and forecasting epidemics.
• Reliable parameter estimation and uncertainty quantification are essential for applying these models to public health interventions.
• Parameter identifiability is a key challenge in developing robust epidemic models.

Purpose of the Study:

• To present a simple computational method for assessing parameter identifiability in compartmental epidemic models.
• To quantify parameter uncertainty and identifiability using a parametric bootstrap approach.
• To demonstrate the method's utility across various compartmental models.

Main Methods:

• Utilized a parametric bootstrap approach to generate simulated data from dynamical systems.
• Calculated confidence intervals and mean squared error of estimated parameter distributions.
• Applied the method to SEIR models of increasing complexity, including those for influenza, Ebola, and Zika.

Main Results:

• Parameter identifiability issues increase with model complexity and the number of estimated parameters.
• The basic reproduction number (R0) often remains robust despite identifiability issues with other parameters.
• R0 can be accurately and precisely estimated even when other parameters exhibit high uncertainty.

Conclusions:

• Parameter identifiability analyses should precede model fitting to ensure reliable public health policy recommendations.
• Quantifying parameter uncertainty is crucial for model-based inferences in epidemiology.
• The presented method enhances the toolkit for model-based inference in compartmental dynamic models.