M.-E. Ebelin, J.-L. Steimer, R. Laplanche
Drug Safety Assessment, Sandoz Pharma, Basel, Switzerland.
Pharmaceutical industry gives more and more focus on the implementation and evaluation of data analysis methodology for population approach, during the development of new drugs. In order to evaluate the ability of NONMEM to handle sparse data as they would arise from a pharmacokinetic (PK) screen design, a retrospective population analysis was performed on data from controlled PK studies with ICS 205-930 (TropisetronR). In this context, the question of the complexity of the PK model in regard of the data has been investigated.
Methods: A “full” kinetic sample of 4390 plasma concentrations from 113 subjects (about 20 concentrations per subject and per administration) was built from seven phase I and phase II pharmacokinetic studies. A reduced data set of 560 plasma levels randomly drawn was constructed in order to mimic a pharmacokinetic screen approach. Subsets of both extensive and sparse data sample were used for the analysis, containing respectively 1151 (16.9 concentrations/volunteer/administration) and 266 concentrations (3.9 concentrations/volunteer/administration) from 68 healthy young and elderly volunteers, extensive metabolizers for debrisoquine, having received an intravenous infusion of 10 mg over different infusion durations. The NONMEM analysis was performed with bi- and three-compartment models involving a physiological parametrisation (total and intercompartmental clearances and central and peripheral volumes of distribution). A constant coefficient of variation model was used for the residual variability model. No interindividual variability was considered at this stage.
Results: The NONMEM analysis of the extensive data subset over the full time profile with the three compartment pharmacokinetic model provided a statistically significantly lower objective function (p<0.001) than with the bicompartment model. The same observation arose from the analysis of the sparse data subset. In contrast, the analysis restricted to post-infusion data showed no difference in the quality of the fit between both models. The CPU time was about twenty times longer with the three compartment model than with the two compartment one. In addition, the number of required PK parameters (six instead of four) provides much more complexity regarding introduction of pathophysiological covariates in the model and handling of random effects.
Conclusion: In this case study, considering all these points together, the NONMEM analysis (with covariates) can be achieved with the bicompartment PK model. The degree of model misspecification seems acceptable for prediction of post-infusion response. However, in a real prospective situation, clinical considerations related to the future use of the model should primarily drive the choice of the kinetic model for a population analysis.
Reference: PAGE 1 (1992) Abstr 892 [www.page-meeting.org/?abstract=892]
Poster: oral presentation