Achieving credibility in risk assessment models.

Abstract

Validation of a mathematical model requires demonstrating that a model is free of mathematical errors (internal consistency), is sensitive to large but not small errors or uncertainties in parameter values (verifiability and robustness), reproduces experimental observations on the system being modeled (external consistency), and leads to testable predictions of the system's biological properties. To be heuristically valid, a model also must be a realistic representation of the actual biological system. Only then would the model's predictions be credible to the wider community of biological scientists who would use the model for risk assessment and dose or species extrapolation. Owing to incomplete data, most current dosimetric models are insufficiently realistic to pass this test of credibility. Enhancements to such models that would help achieve credibility are presented, and suggestions are offered for institutionalizing realistic modeling practices in risk assessment.