Addressing uncertainty in epidemiologically based risk assessment.
Stayner-LT; Vrijheid-M; Stram-D
Society for Risk Analysis 2003 Annual Meeting. Bridging Risk Divides, Baltimore, Maryland, December 7-10, 2002. McLean, VA: Society for Risk Analysis, 2003 Dec; :108
The use of epidemiologic data for quantitative risk assessments (QRA) is becoming increasingly common. Part of the reason for this increase is that using human studies avoids the uncertainties associated with extrapolating from animals to humans. However, using epidemiologic data in QRA introduces a whole other set of uncertainties, which are largely related to the observational nature of epidemiologic data. Of particular concern are issues related to inadequate sample size, unresolved confounding or other biases, inadequate length of follow-up, and errors in the ascertainment of exposures. The issue of errors in the estimation of exposures has been of large concern particularly in occupaional studies. We are evalkuating methods for estimating confidence intervals that reflect both uncertainties related to random error, and potential errors in exposure using Monte Carlo maximum likelihood methods. These methods will be illustrated using an epidemiological study of nuclear power plant workers.
Risk-analysis; Epidemiology; Risk-factors; Quantitative-analysis; Exposure-assessment; Statistical-analysis
Abstract; Conference/Symposia Proceedings
Society for Risk Analysis 2003 Annual Meeting Bridging Risk Divides, Baltimore, Maryland, December 7-10, 2002