Calibration of complex models through Bayesian evidence synthesis: a demonstration and tutorial.

Area of Science:

• Biostatistics
• Mathematical Modeling
• Epidemiology

Background:

• Decision-analytic models often rely on indirectly related data.
• Calibrating model parameters with diverse evidence is challenging.

Purpose of the Study:

• To outline a Bayesian synthesis method for calibrating complex model parameters using indirect evidence.
• To demonstrate uncertainty propagation in decision-making outputs.

Main Methods:

• Constructing a graphical model to link observed data, statistical models, and decision-making parameters.
• Developing an algorithm to estimate posterior probability distributions for updated evidence.
• Rebuilding a Markov model for human papillomavirus (HPV-16) progression within a Bayesian framework.

Main Results:

• The Bayesian framework allows indirect inference of model parameters from various data sources.
• Posterior distributions implicitly weight model scenarios based on data support.
• Uncertainties in model parameters are propagated to decision-making results.

Conclusions:

• Bayesian synthesis provides a robust method for parameter calibration and uncertainty quantification.
• This approach enhances the reliability of decision-analytic models informed by indirect evidence.
• The framework offers a more nuanced assessment of model plausibility compared to discrete scenario analysis.