MODELING DOSE-RESPONSE USING GENERALIZED LINEAR-MODELS

This paper describes a method to determine and model dose-response rel ationships from binomial response data using generalized linear models (GLM). The main advantage of this technique is that it allows LC(p) o r LD(p) to be determined without an initial linearizing transformation . (LC(p) and LD(p) are the lethal concentration or dose that causes p proportion of test animals to die at a specified time period.) Thus, t he method of GLM is an appropriate way to analyze a dose-response rela tionship because it utilizes the inherent S-shaped feature of the toxi cologic response and incorporates the sample size of each trial in par ameter estimation. This method is also much better behaved when the ex tremes of the response probability are considered because responses of 0% and 100% are included in the model. Another advantageous feature o f this method is that confidence intervals (C.I.s) for both the dose e stimate and response probabilities can be computed with GLM, which pro vides a more complete description of the estimates and their inherent uncertainty. Because C.I.s for both the dose estimate and response pro babilities can be constructed, the lowest observed effect concentratio n (LOEC) can also be determined.