Model averaging in microbial risk assessment using fractional polynomials

Area of Science:

• Microbiology
• Risk Assessment
• Statistical Modeling

Background:

• Foodborne diseases from microbial pathogens pose significant public health risks.
• Dose-response models are crucial for understanding the relationship between ingested dose and infection risk.
• Traditional methods rely on selecting a single best-fit model, which can be problematic.

Purpose of the Study:

• To propose modified fractional polynomials as competitive dose-response models for risk assessment.
• To demonstrate the utility of model averaging in circumventing the dilemma of selecting a single best model.
• To improve the accuracy and reliability of risk estimates, especially at low doses.

Main Methods:

• Utilized a modified set of fractional polynomials as candidate dose-response models.
• Selected models based on biological plausibility and rationale.
• Employed model averaging with Akaike's weights to estimate risk across multiple models.
• Applied the approach to Salmonella typhi and Campylobacter jejuni human data for low-dose risk estimation.

Main Results:

• Model averaging effectively addresses situations where a single best model is not apparent.
• This approach provides more honest estimations of standard errors and confidence intervals for risk.
• Simulation studies showed model averaging reduced bias and improved precision compared to single best-fit models.
• Coverage probabilities for risk estimates were closer to the nominal 95% level with model averaging.

Conclusions:

• Model averaging provides a superior method for dose-response modeling in risk assessment compared to traditional single-model selection.
• The proposed fractional polynomial models and averaging technique enhance the reliability of microbial risk assessments.
• This approach leads to more robust and trustworthy estimations of foodborne illness risks.