The in uence of di erent weighting schemes

Pharmacokinetic parameters are usually determined by noncompartmental methods or by assuming a multicompartment model in order to describe the behaviour of the drug in the patient's body. With a noncompartmental analysis the terminal half-life and the volume of distribution are derived from log-linear regression of the terminal elimination phase, whereas the estimation of these parameters is based on nonlinear least squares (NLS) regression when assuming a multicompartment model. The concentration of paclitaxel, a novel antitumor agent, is usually determined by a high-performance liquid chromatography (HPLC), a method that may produce higher variability with increasing concentration. Many applications are based on the so called constant coeecient of variation model, leading to weighted NLS estimation with xed weights 1/concentration 2 or to unweighted NLS for the log-transformed data. However, examples from the literature show that there are various strategies. Using data of a phase I clinical trial, where paclitaxel (Taxol g R) was administered as 1-hour infusion to cancer patients, we considered in this paper log-linear regression of the last 3 and 4 observations, respectively, as well as a two-compartment model. It is shown that the values of half-lives and the volume of distribution depend on the model and the chosen weighting scheme. In this context we comment also on the variability of the estimated pharmacokinetic parameters. We suggest how to use information on the variability of the used HPLC-method in order to nd the most appropriate statistical method.