ML McFadyen(1), SF Marshall (2), PA Milligan(2), DJ Nichols(2) MO Karlsson(3)
1) Dept. of Pharmacology, University of Durban-Westville, Private Bag X54001, Durban, South Africa ECRG, 2) Pfizer Central Research, Sandwich, Kent, United Kingdom 3) Dept. of Pharmacy, Division of Biopharmaceutics & Pharmacokinetics, Uppsala University, Box 580, S-751 23 Uppsala, Sweden
When carrying out population analyses on very large data sets the associated statistical power may necessitate the use of criteria in addition to statistical significance to identify only clinically meaningful covariates. These criteria need to be developed taking into account the therapeutic range of the drug, the seriousness of potential side effects and the pharmacokinetic variability.
We describe a population pharmacokinetic analysis of a large phase III data set for Dofetilide, a selective class III antiarrhythmic, which is effective in the treatment of supra-ventricular cardiac arrhythmias. The data comprised 10335 plasma concentration-time pairs from 1445 individuals after oral dosing. Forty-four covariates were preselected for testing in the population model.
A one-compartment model with first order absorption was the most appropriate structural model. Inclusion of creatinine clearance on CL/F reduced the interindividual variability from 35 to 24%. Subsequent GAM analysis identified a further 29 potential covariates. On univariate NONMEM analysis, 18 were statistically significant but only 7 covariates were included in the full model after application of the criteria below:
* Statistical significance at p<0.05 and either
* A change in the covariate at the extremes of at least 10% for CL/F, 15% for V/F and 20% for Ka or
* A reduction in interindividual variability (% change in CV) > 5% for CL/F, 10% for V/F and 15% for Ka.
In the next stage, the following more rigorous criteria were applied to determine whether a covariate should remain in the final model or not:
* Statistical significance at p<0.001 and either
* A change in the covariate at the extremes of at least 20% for CL/F, 30% for V/F, and 40% for Ka or
* Increase in interindividual variability of > 10% for CL/F, 20% for V/F and 30% for Ka.
Although all 7 covariates remained statistically significant (p<0.001) with single covariate deletion, only weight on V/F met the additional criteria and thus remained in the model. Notwithstanding that they achieved statistical significance, the other covariates explained very little of the remaining interindividual variability (no change for V/F and a change from 24 to 22 % for CL/F). Therefore in accordance with our criteria, creatinine clearance on CL/F and weight on V/F were the only clinically important covariates.
Reference: PAGE 7 () Abstr 673 [www.page-meeting.org/?abstract=673]
Poster: oral presentation