Estimating upper confidence limits for extra risk in quantal multistage models.
Risk Analysis 1994 Dec; 14(6):1001-1010
Models for constructing upper confidence limit estimates of carcinogenic risk were compared. Upper confidence limits on risk were determined in a simulation study of carcinogenicity experiments using two bootstrap methods and compared with those obtained using commonly used likelihood based models. The models examined in the study were described. Parametric bootstrap and likelihood based methods gave better coverage probabilities compared with a nonparametric bootstrap model. The likelihood procedure tended to provide estimates higher than that of true extra risk and was more variable than the bootstrap procedures. The distance from the true extra risk was found to decrease with increasing background tumor rates and the amount of curvature in the actual underlying dose response function was found to affect the relative bias of all estimators. The authors conclude that although the likelihood based method had the best coverage probability properties, the parametric bootstrap procedure was less biased and less variable; however, neither method is considered sufficient for highly curved dose response patterns.
NIOSH-Author; Statistical-analysis; Mathematical-models; Dose-response; Risk-factors; Analytical-models; Carcinogenesis; Risk-analysis