The Bayesian microbial subtyping attribution model: robustness to prior information and a proposition.

Area of Science:

• Food safety
• Epidemiology
• Statistical modeling

Background:

• Attributing foodborne illnesses to sources is crucial for public health policy.
• The Hald Bayesian microbial subtyping attribution model is advanced but faces overparameterization challenges.
• Previous methods addressed overparameterization by setting parameters to constant values.

Purpose of the Study:

• To investigate the impact of parameterization choices on the Hald model's robustness.
• To propose and evaluate an alternative parameterization for the Hald model.
• To improve the accuracy and reliability of foodborne illness source attribution.

Main Methods:

• Analysis of parameterization choices in the Hald model.
• Development and application of an alternative parameterization using data-based values.
• Comparison with Mullner's modified Hald model and a simple deterministic model.
• Illustration using the 2005 French non-typhi Salmonella dataset.

Main Results:

• Setting bacterial type parameters specific to unique sources and using data-based values improved model convergence and adequacy.
• Attribution estimates became consistent with results from other models.
• Type and source parameter estimates were coherent with Mullner's model.
• The Hald model demonstrated high sensitivity to parameterization.

Conclusions:

• The proposed parameterization enhances the robustness and reliability of the Hald attribution model.
• This improved approach enables successful application of the model with real-world data.
• Optimized parameterization is key for accurate foodborne illness source attribution and effective public health interventions.