PRE-CISE: A PRE-calibration Coverage, Identifiability, and SEnsitivity analysis workflow to streamline model...

Area of Science:

• Mathematical modeling in health sciences
• Computational epidemiology
• Biostatistics

Background:

• Model calibration is crucial for health policy analysis.
• Nonidentifiability issues can compromise model reliability.
• Pre-calibration analysis is needed to streamline complex modeling workflows.

Purpose of the Study:

• Introduce PRE-CISE, a pre-calibration workflow.
• Integrate coverage analysis, local sensitivity, and collinearity diagnostics.
• Streamline model calibration and address nonidentifiability.

Main Methods:

• PRE-CISE involves coverage analysis, local sensitivity analysis, and collinearity diagnostics.
• Prior distribution bounds are resized based on sensitivity analyses to improve coverage.
• Identifiability is assessed using collinearity analysis.

Main Results:

• Coverage analyses identified initial misfits, which were improved by resizing prior distribution bounds.
• Local sensitivity analysis prioritized key parameters in a COVID-19 model, enhancing calibration efficiency.
• Daily calibration targets in the COVID-19 model resulted in collinearity indices below practical thresholds.

Conclusions:

• PRE-CISE offers a practical and transparent method for refining model parameters and calibration targets.
• This workflow improves uncertainty reporting and strengthens health policy analysis reliability.
• The approach is demonstrated on a Sick-Sicker Markov model and a COVID-19 case study.