MEASURES OF UNCERTAINTY FOR PHARMACOKINETIC AND PHARMACODYNAMIC PARAMETER ESTIMATES - A NEW COMPUTERIZED ALGORITHM

Numerous algorithms exist to fit data to nonlinear models of the type used in chemistry, pharmacology, physiology, etc. Most include modules that provide some measure of the reliability of the estimated model p arameters. The variance-covariance matrix (VCM) is the common tabulati on of information that is used to quantify the parameter uncertainty a s well as correlations between parameters. The VCM has its mathematica l foundation in the linear regression world, where the dependent varia ble is a linear function of the parameters. However, when the model is not linear in its parameters, then the VCM is no longer an absolute q uantitative measure of reliability of the parameter estimates and shou ld be interpreted with caution. If the goal is to obtain a realistic a nd quantitative rather than a qualitative measurement of the parameter reliability, then it is necessary to have an alternative approach to describe the parameter likelihood region. We present a computerized al gorithm that fills that need, and we compare its performance with the traditional VCM approach for different data sets. We also discuss crit eria that may be used to determine when the VCM approach should and sh ould not be used. (C) 1996 Academic Press, Inc.