Statistical variability: what does it mean in risk assessment?

An essential part of toxicologic research involves assessing the risk from exposure to chemicals, drugs, or other potentially harmful substances. Epidemiologic information establishes an association but not a causal relationship between risk factors and adverse health outcomes. Controlled experiments in laboratory animals, however, provide insight into the agent-outcome relationship and aid in establishing low-risk exposure levels for humans. This is especially true in studies of reproduction and development since epidemiologic data are often limited. While controlled experiments reduce certain difficulties inherent in interpreting epidemiologic data, there still remains the problem of careful evaluation of statistical information. This note addresses issues in the interpretation of statistical variability in fitting dose-response models and estimating exposures associated with a low (or at least specified) level of risk.