Structural and Practical Identifiability of Phenomenological Growth Models for Epidemic Forecasting

Area of Science:

• Epidemiology
• Mathematical Modeling
• Computational Biology

Background:

• Phenomenological models are crucial for disease dynamics forecasting, especially with limited mechanistic data.
• Model reliability hinges on the structural and practical identifiability of parameters.
• Existing models often face challenges with non-integer exponents, impacting analysis.

Purpose of the Study:

• To systematically analyze the identifiability of six prevalent epidemiological growth models.
• To address challenges in models with non-integer exponents through reformulation.
• To validate identifiability and robustness using real-world epidemiological data.

Main Methods:

• Reformulated six growth models (GGM, GLM, Richards, GRM, Gompertz, modified SEIR) by introducing state variables.
• Conducted structural identifiability analysis using the StructuralIdentifiability.jl package in JULIA.
• Validated results via parameter estimation and forecasting with the GrowthPredict MATLAB Toolbox on monkeypox, COVID-19, and Ebola data.
• Assessed practical identifiability using Monte Carlo simulations to evaluate robustness against observational noise.

Main Results:

• All six reformulated models were confirmed as structurally identifiable.
• Parameter estimates demonstrated practical identifiability and robustness across varying noise levels.
• Model and dataset specific sensitivities in parameter estimation were observed.

Conclusions:

• The study validates the structural identifiability and practical robustness of key phenomenological growth models for epidemic forecasting.
• These models are adaptable to real-world data complexities and noise.
• Findings support the use of these models for informing public health strategies and interventions.