F. Mesnil(1) , F. Mentre(1) , C. Dubrue(2) , A. Mallet(1)
INSERM U 436, 91 boulevard de l'Hopital, 75634 Paris Cedex 13, France(1)
SYNTHELABO Recherche, 1 avenue Pierre Brossolette, 91380 Chilly-Mazarin,
France(2)
A population analysis of the kinetics of a new antihistamine, mizolastine, and the validation of this study were performed using the Non Parametric Maximum Likelihood approach (NPML) from data of 696 allergic patients, included in ten Phase III clinical trials. Mizolastine is proposed for the treatment of allergic rhinitis and urticaria. Its absorption is fast and its elimination is mainly hepatic. The 696 patients analysed received a single dose daily (either a tablet or a capsule). During the treatment, one or several mizolastine plasma levels were measured, and clinical and biological data were collected. Among the latter, drug presentation (tablets or capsules), sex, age, weight, creatinine and hepatic transaminases plasma levels were introduced in the analysis as covariates which could contribute to the kinetic variability. The subjects were randomly divided into two groups. the first one including two thirds of the patients (449 subjects, 619 concentrations) was used for the population analysis, and the second one including one third of the patients (247 subjects, 341 concentrations) was used to validate the results.
Mizolastine kinetics after oral administration is best described by a two compartment open model with zero order absorption, which involves five parameters : the apparent volume of the central compartment, V/F, the elimination rate constant from the central compartment k, the apparent distribution rate constant a, the apparent elimination rate constant b, and the duration of absorption T. The measurement error, knowing the kinetic parameters, is assumed to be additive and normally distributed with a zero mean and a variance depending on the square of the predicted concentration. The two coefficients of the variance model were computed from analytical error which is 20% for low measures and 10% for high concentrations (these percentages were established from quality control samples). Population analysis was performed by NPML (1). In this approach, covariates are assumed to be measured with a measurement error normally distributed. NPML provides an estimate of the joint distribution of kinetic parameters and covariates (2). Relationships between discrete covariates and parameters were established by comparing the distribution of parameters in each subgroup of covariates. Relationships between continuous covariates and parameters were determined by comparing distributions of parameters conditionnal to several levels of covariates (5th, 25th, 50th, 75th, 95th percentiles).
The mean values of the parameters were consistent with the results of earlier studies, and an important kinetic variability was found. For instance, the mean oral clearance was 8.43 l/h, with a coefficient of variation of 89%. Differences were observed between tablets and capsules concerning the distributions of V/F, k, a ; between males and females differences were found concerning the mean V/F (respectively 43.8 l and 36.0 l) and T (respectively 2.3 and 1.6 h). No relationships were observed between kinetic parameters and continuous covariates.
A validation of this study partly based on the methodology proposed by Vozeh et al was performed (3). We evaluated the prediction of the model (the relevance of the joint distribution estimated by NPML) in the validation group. For each observed concentration, we computed the distribution of the measured concentration using the covariates of the corresponding subject, the joint distribution estimated by NPML and the distribution of the measurement error. The expectation of this measured concentration was used as an estimate of the actually observed concentration, the so called predicted concentration. Observed and predicted values were compared, and standardized residuals were computed using the standard deviation of the measured concentration. Under the assumption of a good prediction of the model, the mean and the variance of standardized residuals are respectively zero and one. The values that we found were respectively 0.130 and 1.693. Another approach to evaluate the prediction of the model is to compute a 80% confidence interval for each measured concentration of the validation group, and to calculate the coverage of this interval. The proportion of observed concentrations which were found to be in the predicted confidence interval was 80.9%. These results show the good adequacy of the model.
REFERENCES
1. MALLET A: A maximum likelihood estimation method for random coefficient regression models. Biometrika 1986;73:645-656.
2. MENTRE F, MALLET A: Handling covariates in population pharmacokinetics. Int J Biomed Comput 1994;36(1,2):25-33.
3. VOZEH S, MAITRE PO, STANSKI DR: Evaluation of population (NONMEM) pharmacokinetic parameter estimates. J Pharmacokin Biopharm 1990;18;161-173.