SB Duffull (1), JJ Deely (2). NHG Holford (3).
(1) Dept of Clinical Pharmacology, Christchurch Hospital, Christchurch. (2) Dept of Mathematics and Statistics, University of Canterbury. Christchurch. (3) Dept of Pharmacology, University of Auckland, Auckland.
Introduction: A semiparametric method of population analysis is described. This method differs from traditional methods of population analysis (eg NONMEM) as it gives the frequency histogram of the population parameters.
Aim: To validate and compare this method using simulated data.
Methods: The values of clearance (Cl) and volume of distribution (Vd) for gentamicin were simulated for a set of 200 subjects assuming a log normal distribution of the parameters. Each subject ‘received’ a 400 mg dose of gentamicin as a 30 min intravenous infusion and concentrations were simulated at 1, 4 and 8 hours after the start of the dose. The CV of the residual error component was 15% (limit of quantitation = 0.25 mg/L). In order to test the effects of a small numbers of subjects and sparse data, two sets of analyses were performed: 1) the set was divided into n=5, 10, 25 and 100 subjects, 2) the 4 hour concentration was discarded and the set was divided into n=10 and 25 subjects. Both methods were used to determine the geometric mean and CV% of the population parameters.
Results: 1) Datasets with three concentrations. Both methods provided accurate estimates of the population mean (within 10% of the true value), irrespective of the number of subjects (n). When n was small (≤25) NONMEM significantly underpredicted the CV% for Vd. The semiparametric method gave accurate estimates of the CV% when n 5 25, and was similar to NONMEM at greater n (100 or 200). The estimate of the CV% of the parameters given by NONMEM improved as the n became larger. NONMEM gave better estimates of the residual error CV% than the semiparametric method. 2) Datasets with two concentrations. Both methods gave accurate estimates of the population mean. The semiparametric method was superior to NONMEM at estimating the CV% when n=10 for both parameters and also when n=25 for Vd.
Discussion: A new method of population analysis is presented. It performed well when the number of subjects was small and with both sparse datasets. It may have advantages over established techniques. It is, however, computationally intensive taking up to an hour to run.
Reference: PAGE 7 (1998) Abstr 676 [www.page-meeting.org/?abstract=676]
Poster: oral presentation