NIOSHTIC-2 Publications Search

Statistical properties of carcinogen threshold estimates using log-linear regression.

Dankovic-DA; Wheeler-MT
Toxicologist 2005 Mar; 84(Suppl 1):78
The objective of this study is to examine the statistical properties of a method that has been proposed for estimation of dose-response thresholds for carcinogens. Simulation studies have been conducted to examine the statistical properties of a method proposed by Dr. W. J. Waddell, based on log-linear regression. The simulations were based on published data sets that have been analyzed by this method. For this presentation we focus on the simulation results for an NTP study of the food additive 2, 4-hexadienal. The simulation study was done by fitting 1-stage, logistic, piece-wise linear ("hockey-stick"), and log linear regression models to the data. These models all fit the data adequately, based on a chi-square goodness of fit test using a p-value >/= 0.1 as the acceptance criterion. Monte Carlo methods were used to generate sets of 2000 stochastically-varying simulated data sets. Dose-response thresholds estimated using log-linear regression were compared with the known properties of the underlying dose-response relationship, and the performance of the log-linear regression method was characterized. The primary result was that simulations based on no-threshold models (1-stage and logistic) resulted in an above-zero threshold estimate in every case, using log-linear regression. Simulations based on threshold models (log-linear regression and piece-wise linear regression) returned similar threshold estimates. However, the estimates using piece-wise linear regression exhibited considerably better precision than estimates using log-linear regression. These results suggest that the log-linear regression procedure is unreliable for determining the presence of a dose-response threshold, and sub-optimal for determining the location of the threshold in cases where the threshold is assumed to exist.
Carcinogens; Carcinogenesis; Carcinogenicity; Dose-response; Simulation-methods; Statistical-analysis; Models; Computer-models
Publication Date
Document Type
Fiscal Year
NIOSH Division
Priority Area
Research Tools and Approaches: Risk Assessment Methods
Source Name
The Toxicologist. Society of Toxicology 44th Annual Meeting and ToxExpo, March 6-10, 2005, New Orleans, Louisiana
Page last reviewed: March 11, 2019
Content source: National Institute for Occupational Safety and Health Education and Information Division