ESTIMATING THE MAXIMUM EFFECTIVE DOSE IN A QUANTITATIVE DOSE-RESPONSEEXPERIMENT

A simulation study was conducted to compare several procedures for est imating the maximum effective dose in a quantitative dose-response exp eriment. Using four equally spaced dose levels, data were generated fr om four different model types: the quadratic growth curve, the Mitcher lich growth curve, the linear-linear plateau spline model, and the qua dratic-linear plateau spline model. Each model type was parameterized to create three different model ranges, and for each range, data were generated from populations with three different standard deviations. T he existence of unique dose-response curves is assumed; thus, all the procedures compared in this paper require that the data have been mode led by a polynomial or nonlinear regression model. An attempt was made to fit each generated data set with each of the four model types. Max imum effective dose estimation procedures were applied to a data set o nly when the data were adequately described by a given model. The simu lation indicated that the estimate of the maximum effective dose is in fluenced more by the choice of model than by the method of estimation. Because of the consistently low estimates produced when the data were modeled by the linear-linear plateau spline, this model is not recomm ended for use in maximum effective dose estimation experiments. The si mulation also demonstrated that the design failed to provide sufficien t information about the form of the dose-response curve. Designs with more than four dose levels should be considered.