1997 年 57 巻 1 号 p. 25-32
The accuracy with which the M estimation (ME), a robust analysis, can estimate pharmacokinetic parameters was investigated and compared with that of the nonlinear leastsquared method (NLS). A one-compartment open model in which first-order absorption, with ka=2.0 (h-1) or 0.3 (h-1), ke=0.15 (h-1) and Vd=0.2 (ι/kg), was assumed. Data obtained at various sampling times (t) were used; 6-16 points were yielded in a geometric ratio within 24 h. Group A (normal distribution errors added to each data point) and Group B (ten times the normal distribution error added to one point selected randomly and errors added to other points were the same as for Group A data points) data points were generated. The data points in each of these groups were generated from 1,500 data sets generated using Monte Carlo simulation. These data were analyzed by NLS and ME using the adapted Biweight method. Using an absorption rate constant of 2.0 (h-1), the ke and Vd estimated using Group A-ME data points were consistent with those estimated using Group A-NLS data points. For Group B-ME data points, ke and Vd estimated using 7 or more data points and ka estimated using 10 or more data points were the same as those estimated using Group A-NLS data points. For Group B-NLS data points, Vd and ka estimated using 14 or more data points were consistent with those estimated using the data points in Group A-NLS, but the ke estimated, even when a large number of data points was used, was different from that estimated using the data points in Group A-NLS. Using an absorption rate constant of 0.3 (h-1), the estimatability of ka using the data points in Group B-ME was improved compared to using the absorption rate constant of 2.0 (h-1). The application of ME to data collected under nonideal conditions giving rise to non-normal distribution errors was shown to be useful in the estimation of pharmacokinetic parameters.
