R Jelliffe(1), A Schumitzky(1), D Bayard(1), M Van Guilder(1), X Wang(1), M Milman(1), S Leonov(1), R Leary(2), P Maire(3), X Barbaut(3), and S Lecoq(3)
1-Laboratory of Applied Pharmacokinetics, USC School of Medicine, Los Angeles, USA; 2- San Diego Supercomputer Center, UCSD, San Diego, USA; 3-ADCAPT, Hopital Geriatrique Antoine Charial, Francheville, France.
Population PK/PD modeling describes variability between and within subjects. It employs the stated assay variability. Parametric (P) population modeling obtains PK/PD parameter means and variances, and intrasubject variability. A specific shape of the parameter distributions is assumed. P methods can separate inter- from intra- individual variability, and from assay variability. Since only single parameter values are obtained, there is only one population model. It makes only one prediction of future responses, and cannot predict the precision of the results of any dosage regimen based on it.
Nonparametric (NP) population modeling finds instead the N most likely sets of model parameter values given the raw data from N subjects. There are thus N models to use for predicting responses. Parameter means, variances, etc are also obtained.
“Multiple model” (MM) dosage design employs results of assay and intraindividual variability, and the N models available with the NP population model. N predicted responses can be compared with the target goal(s), permitting prediction of the precision with which a desired target will be achieved. The optimal MM dosage regimen can be found which achieves the target with minimum weighted squared prediction error.
Thus the sequence of 1) explicit determination of the assay error pattern over its working range, 2) P population modeling to estimate intraindividual variability and parameter ranges, 3) NP modeling to obtain the full discrete joint density of N different PK/PD models, and 4) MM dosage design, yields dosage regimens which achieve target goals optimally. This is the final and most useful action to be taken based on population modeling. A clinical Windows MM program is in development.
Supported by NIH LM05401 and RR11526.
Reference: PAGE 8 (1999) Abstr 132 [www.page-meeting.org/?abstract=132]
Poster: oral presentation