Dinesh de Alwis and Leon Aarons
Eli Lilly & Company and the University of Manchester
Model based approaches such as nonlinear mixed effect modelling (NONMEM) can be a time consuming and expensive process as it requires a skilled NONMEM user. Model independent approaches have been investigated as a means of looking at systemic exposure based on plasma drug concentrations in phase III clinical trials (Nedelman et al., 1995). Our objective was to contribute to the evaluation of this approach with phase III data and compare these results to the model based NONMEM approach. The data arose from a phase III study carried out in patients with mild/moderate dementia, in which patients were given 2mg PO ondansetron twice a day for 52 weeks. Data from 145 subjects providing 188 plasma concentrations were available for analysis. Demographics available were age (61 to 95 yrs) and gender (99 females, 46 males). Weight was not recorded in the study. Other covariates such as smoking status, alcohol consumption together with coexisting disease states and concomitant medication were recorded.
Population based modelling was carried out using NONMEM. The model independent method involved the following steps 1) The concentrations were divided into one-hour time intervals from 0 to11 hours post-dose. 2) Within each one-hour interval, the quartiles of the concentrations were determined. 3)Patients were then partitioned into patient quartiles depending on which observation quartile their concentrations fell into. If all the concentrations of a selected patient fell into one observation quartile, this patients would fall into the ‘all in one’ quartile. If a patient’s observed concentrations fell into two adjacent observation quartiles, then they were allocated according to the distance from the quartile. For example, suppose of n observations for the patient, u where in the upper of the two adjacent quartiles and n-u in the lower. Therefore, if d1……du are the distances from the points in the upper quartile to the common boundary, and du+1…..dn the distances for the po ints in the lower quartile and if d1+ … +du > du+1 + … +dn then the patient was assigned to the patient quartile corresponding to the upper observation quartile or vice-versa. If a patient’s observed concentrations fell into three adjacent observation quartiles, the patient would be assigned to the patient quartile corresponding to the middle of the three observation quartile. If however, a patients concentrations fell into all four quartiles the patient was left unclassified. Once patients were allocated into their respective quartiles then graphical analysis was carried out against the covariates.
A comparison of the two methods will be presented. The model independent method was found to be simpler and quicker than NONMEM. It can be used as a screening process for important covariates. The NONMEM analysis provided estimates of individual CL values whereas the model independent approach provides only a crude approximation to the AUC. When subjects are not undergoing a common dosing regimen the model independent method cannot be used. In addition the model independent approach cannot estimate parameters such as volume of distribution.
Nedelman JR, Karara AH, Chang CT, Gibiansky E, McDonald S, Robinson WT. Inferring systemic exposure from a pharmacokinetic screen: model free and model-based approaches. Stat Med., 14, 955-968. 1995.
Reference: PAGE 8 () Abstr 148 [www.page-meeting.org/?abstract=148]
Poster: poster