A meta-analysis regression model for quantitative prediction of the impact of polymorphism on drug exposure
Michel Tod (1), Sylvain Goutelle (1), Marie Claude Gagnieu (2), and the Genophar II Working Group
(1) Hospices Civils de Lyon and University Lyon I, France. (2) Dpt of pharmacotoxicology, HEH, Hospices Civils de Lyon, France
Objectives: A framework to predict quantitatively the impact of CYP2D6 polymorphism on drug exposure is proposed. The metrics of interest is the ratio of drug AUC in mutant to wild-type patients. A model was derived to rely the AUC ratio with two characteristic parameters, one for the drug (the fraction metabolized by CYP2D6 in vivo, CR), the other for the genotype (the fraction of activity with respect to the homozygous wild type, FA). Any combination of alleles, as well as duplications, may be accommodated. The model allows to combine all available data arising from all drugs and genotypes.
Methods: The primary goal of the analysis was to estimate the CRs and FAs for 40 drugs and 5 classes of genotypes respectively, including poor, intermediate, and ultra-metabolizers. Data were available for 73 (drug, genotype) couples. A three-step approach was applied. First, initial estimates of CRs and FAs were obtained by several methods, using data from the literature. Second, an external validation of these initial estimates was carried out, by comparing the AUC ratios predicted by the model equation to the observed values, using a second set of published data. Third, refined estimates of CRs and FAs were obtained by a bayesian orthogonal regression, using all the data and initial estimates of CRs and FAs from step 1. The posterior distributions of the AUC ratios, CRs and FAs were obtained by Monte Carlo Markov chain simulation by using WinBugs 1.4.
Results: With the refined estimates, the mean prediction error of AUC ratios was -0.05, while the mean prediction absolute error was 0.20. The model may be used to predict the variations of exposure for all 200 combinations between drugs and genotypes. An application to a rare combination of alleles (*4*10) is described.
Conclusion: The predictive performances of the model were good. The method is very easy to use, once the characteristic parameters (CRs and FAs) have been established. This framework may be easily applied to other polymorphisms.