Population pharmacokinetics of valproate in Mexican children with epilepsy

BIOPHARMACEUTICS & DRUG DISPOSITIONBiopharm. Drug Dispos. 29: 511–520 (2008)

Published online 8 December 2008 in Wiley InterScience

(www.interscience.wiley.com) DOI: 10.1002/bdd.636

Population Pharmacokinetics of Valproate in MexicanChildren with Epilepsy

Tania Correaa, Ildefonso Rodrıguezb and Silvia Romanoa,�aPharmacy Department, Facultad de Ciencias Quımicas, Universidad Autonoma de San Luis Potosı, San Luis Potosı, S.L.P, MexicobClinic of Epilepsy, Hospital Central ‘Dr Ignacio Morones Prieto’, San Luis Potosı, S.L.P. Mexico

ABSTRACT: Background. The aim of this study was to determine the factors that influencevalproate clearance (CL) in Mexican epileptic pediatric patients using a mixed-effect model andsparse data of serum concentrations of valproic acid (VPA) collected during routine clinical care ofpatients.

Methods. The number of patients included in the study was 110. The population CL wascalculated by using the NONMEM program. The following covariates were tested by theirinfluence on CL: total body weight (TBW), height, age, body surface area, daily dose (DD), sex of thepatient and comedication with phenobarbital (PB) or carbamazepine.

Results. The final regression model for valproic CL found best to describe the data was:CL/F 5 (0.046610.00363 TBW10.000282 DD) * (110.236 PB). This model allows a reduction of 50%of the interindividual variability and of 31% of the residual variability described by the basic modelthat does not include covariables.

Conclusions. Total body weight, daily dose of valproate and concomitant therapy with PB arefactors that significantly influence VPA kinetic disposition and they should be considered inprogramming dosage regimens for this antiepileptic drug in the pediatric population. Thevalidation of the model supports its acceptability for clinical purposes. Copyright r 2008 JohnWiley & Sons, Ltd.

Key words: valproate; epilepsy; children; pharmacokinetics; NONMEM

Introduction

Despite the introduction of new drugs withantiepileptic activity, valproic acid (VPA) con-tinues to be one of the drugs most often used inthe treatment of all types of epileptic crisis [1].VPA is commercially available as sodium salt,but due to its high hygroscopicity, sodium hasbeen replaced by magnesium. Magnesiumvalproate is pharmacologically equivalent to the

sodium salt and VPA itself. However, uponabsorption in the bloodstream, it decomposesinto VPA and magnesium ions, which hassome advantages. It is a highly stable andeffective drug due to its added antiepilepticactivity, provided by the magnesium ions,and a slower and more regular absorption rate,which prevents variations in the serum levels ofVPA [2,3]; in addition to its pharmaceuticalfeatures, magnesium valproate costs less thanthe sodium salt formulations and is used widelyin Mexico.

In the pediatric population, the degree ofcorrelation between the dose administered andthe serum concentrations of VPA is low, because

*Correspondence to: Facultad de Ciencias Quımicas, Uni-versidad Autonoma de San Luis Potosı, Av. Manuel Nava.No 6, Zona Universitaria, 78210 San Luis Potosı, San LuisPotosı, Mexico. E-mail: [email protected]

Received 18 June 2008;Revised 15 September 2008;

Accepted 23 October 2008Copyright r 2008 John Wiley & Sons, Ltd.

of the high interindividual variability in theclearance (CL) of the drug in this population,which may be related to its dose-dependentproperty [4]. This variability in the eliminationof VPA can be attributed to metabolic changesduring childhood and adolescence in the patient,to demographic and physiopathological charac-teristics of the patient and it can also be due tointeractions with other drugs administered con-comitantly [5]. This makes it difficult to establishguidelines for determining a priori doses thatwould provide an optimal response in thepatient, preventing unnecessary risks of over-and under-dose.

Furthermore, several different factors, includ-ing pharmacogenetics, contribute to the inter-individual variability in drug response. Likemost other agents, many antiepileptic drugs aremetabolized by a variety of enzymatic reactions,and the cytochrome P450 (CYP) superfamily hasattracted considerable attention [6]. Some ofthose CYPs exist in the form of genetic variants,which may also affect the plasma concentrationsand the exposure to antiepileptic drugs. Thevariability of kinetic parameters of antiepilepticdrugs has been reported in different populations[7–12]. There are studies on the kinetic behaviorof VPA and the factors that can modify it [13–17].It has been suggested that certain variables suchas age and total body weight exert a significantinfluence on VPA pharmacokinetics, while thereis still some controversy over the influence ofother factors such as sex and combination withother antiepileptics such as phenytoin, pheno-barbital or carbamazepine. To date, no study hasbeen reported that has elucidated the populationpharmacokinetics of valproate in the Mexicanpopulation.

Currently, guidelines for recommended dosesof VPA and its salts, only consider the agegroup to which the patient belongs, adult orchild, and only the total body weight in thelatter group, ignoring the changes that occurduring childhood and puberty in the metabolismof this drug, its dose-dependent kinetics andthe influence of treatments associated withother drugs, antiepileptic and otherwise. Thissituation reveals an unquestionable interest in

characterizing in the Mexican population thepharmacokinetics of VPA, which is includedin the basic medicines list of the MexicanMinistry of Health and is considered the first-line treatment of epilepsy due to its wide use inthe clinic.

As previously mentioned, the aim of this studywas to determine the population model thatdescribes the pharmacokinetic behavior of VPAin epileptic children monitored in the HospitalCentral ‘Dr Ignacio Morones Prieto’ of San LuisPotosi (Mexico).

Patients and Methods

Patients

Retrospective data were collected from 110pediatric epileptic patients receiving chronictreatment with magnesium valproate who wereexposed to routine monitoring through themeasurement of VPA steady-state serum concen-tration in Hospital Central ‘Dr Ignacio MoronesPrieto’ of San Luis Potosi between January 2004and April 2006. A total of 119 measured serumVPA concentrations were available. Table 1shows the demographic and clinical character-istics of the population study.

The medical charts of these patients recordedsteady-state serum levels of VPA and completeinformation regarding the dose administered,dosing interval, time of blood sampling andtime of last administration. In addition, thefollowing demographic and clinical informationwas recorded on standardized data collectionforms to determine the population model:age, total body weight, height, gender, bodysurface area and concomitant administrationof phenobarbital or carbamazepine. Patientswith incomplete data were excluded from thestudy.

Magnesium valproate was given orally astablets or oral solution, two or three times aday for more than 1 month at the same dose.Serum trough levels of VPA at the end of adosing interval were measured as a routinepolicy of drug sampling in the hospital. Sampleswere obtained at steady state before the morning

T. CORREA ET AL.512

Copyright r 2008 John Wiley & Sons, Ltd. Biopharm. Drug Dispos. 29: 511–520 (2008)DOI: 10.1002/bdd

dose, to minimize diurnal variations in serumprotein binding [13].

Drug analysis

Measurement of VPA serum concentration wascarried out using the cloned enzyme donorimmunoassay (CEDIA) with a Roche ModularSWAs analyser. The sensitivity of the analyticaltechnique for this drug was 3.0 mg/mL withan inter- and intra-assay variability of lessthan 10%.

Population analysis

The data were analysed by nonlinear mixedeffects modeling using NONMEM software(version V, level 1.1: GloboMax LLC, Hanover,MD, USA). The data analysis was carried outusing a stepwise approach: (i) determination of abasic pharmacokinetic model using NONMEM;(ii) selection of covariates using generalizedadditive modelling (GAM) and (iii) finalNONMEM modelling to determine the popula-tion pharmacokinetic model [19]. The first-orderapproximation (FO) was used to derive popula-tion pharmacokinetic parameters, the intersub-ject variability (Z) in these parameters, andthe residual variability between observed andpredicted concentrations (e). This model accountsfor population pharmacokinetic parametervariability (between and within subjects),residual variability (random effects) andparameter differences predicted by covariates(fixed effects).

All data from subjects were fitted simulta-neously using ADVAN2 and TRANS2 programsubroutines to develop the pharmacokineticand statistical models. These subroutines wereselected to define a one-compartment openkinetic model with first-order absorption andelimination.

The absorption constant (Ka) of VPA andbioavailability of the drug were fixed at 1.2 h�1

and 1, respectively, in agreement with previouslypublished values [20,21]. Because collected VPAsteady state trough serum concentrations datadid not provide information about the volume ofdistribution (Vd), it was fixed to 0.24 l/kg [22].Thus, only the CL of VPA, the most usefulparameter for the evaluation of the eliminationrate, was determined in the model.

For the development of the covariate model ofCL, univariate selection was followed by back-ward deletion. The incorporation of continuouscovariates was both linear and non-linear:

CL=F ¼ ðy1 þ y2 covcont þ y3 covy4contÞ

where cont is continuous and y represents thevalues to be estimated.

The continuous covariates analysed were: totalbody weight, age, height, body surface area(calculated by the Mosteller’s formula: BSA ¼ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiðHeight�Weight=3600

p[18] and the daily dose

of magnesium valproate.The categorical covariates analysed were:

gender and concomitant medication with carba-mazepine and/or phenobarbital, because thesedrugs were administered to more than 10%

Table 1. Demographic and clinical characteristics of the pediatric patients included in the study

Characteristic Population study Mean7SD (range) Validation analysis Mean7SD (range)

No. of patients 110 40Sex (male/female) 63/47 26/14Age (years) 7.074.5 (0.5–17) 7.7574.4 (1–15)Total body weight (kg) 27.1716.5 (8–68.5) 30.7717.3 (13–73)Height (cm) 118.6726.8 (66–174) 125.2725.1 (82–173)Body surface area (m2)a 0.9370.37 (0.37–1.79) 1.0170.38 (0.56–1.83)VPA concentration (mg/l) 63.51724.6 (12.0–128.3) 63.1719.8 (21.0–107.0)Dose (mg/day) 700.77370.8 (100–1800) 7207394 (200–1800)Dose (mg/kg/day) 29.3714.2 (9.1–70.0) 25.5712.03 (9.5–55.0)Phenobarbital (Y/N) 13/97 2/38Carbamazepine (Y/N) 17/93 3/37

aCalculated by the Mosteller’s formula: BSA ¼ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiðHeight�Weight=3600

p[18].

VALPROATE PHARMACOKINETICS IN MEXICAN CHILDREN 513


of the study subjects. The categorical covariateswere included in the model in a multiplicativeway.

CL=F ¼ ðy1 þ y2 covcont þ y3 covy4contÞ

� ð1þ y5 covcatÞ ð1þ y6 covcatÞ

where cont is continuous and cat is category.Between-subject variability was modeled using

additive, proportional or exponential errormodels. Residual error was modeled usingadditive, proportional or combined error models.A maximum likelihood objective function wasestimated using an extended least squarenon-linear regression method and the NONMEMcomputer program. The combination of errormodels more appropriate for the constructionof the pharmacokinetic model was determinedby a preliminary analysis with the basicmodel and it was later confirmed with thefinal model.

With the fixed and random effects modelschosen, the empirical Bayesian estimate of CLwas obtained using the POSTHOC option withinthe NONMEM program.

Stepwise GAM implemented in Xpose3 (ver-sion 3.11) running within the S-Plus (version2000, Mathsoft, Inc., Cambridge, MA, USA),program was performed on the empirical Baye-sian parameter estimates for the selection ofcovariates that influence the pharmacokineticparameter.

Bootstrap in conjunction with GAM wasrun to assess the importance of the selectedcovariates and to give information on howcovariates interact with respect to inclusion/exclusion from the covariate model. The numberof bootstrap iterations made in this studywas 150.

Each covariate screened as influential by thegraphic explorations was entered into the basicpharmacokinetic model. Depending on the nat-ure of the covariate, a linear and non-linearmodel for continuous variables or a step modelfor categorical variables was tested by stepwiseaddition to the basic model.

In the comparisons between alternative mod-els, a likelihood ratio test was used [23]. Thechange in objective function (DOBJF) produced

by the inclusion of a covariate is proportional totwice the negative log likelihood of the data andapproximate a w2 distribution with degrees offreedom equal to the difference in the number ofparameters between the two models. A DOBJF of43.8 (po0.05) is chosen to represent the statis-tical significance for the addition of one fixedeffect. When the effect of a covariate wasstatistically significant, it was kept in the modeland a new covariate was added and tested.Otherwise, the covariate was dropped from themodel and the effect of another covariate wasevaluated (step-up approach). The tentative finalmodel was further tested by eliminating eachcovariate one at a time to evaluate the DOBJF(step-down approach). Because of multiple com-parisons, the level of significance for putting thecovariate back in the model was set atDOBJF410.8 (po0.001, d.f. 5 1). Only covariatesthat showed a significant contribution werepreserved in the model.

Other criteria used in evaluating alternativemodels were: inspection of weighted resi-dual plots, minimization of interindividualvariances and improvement in their precision,and a reduction in the magnitude of residualvariability.

Model validation

To evaluate the results of the population phar-macokinetics analysis, a second group of 40epileptic patients was studied who had the sameinclusion criteria, but they were not included inthe original analyses. The demographic andtreatment characteristics of this group are alsoshown in Table 1. The VPA concentrationsmeasured in these patients were compared withthe corresponding concentrations predicted bythe basic and the final model.

The prediction error was calculated as sug-gested by Sheiner and Beal [24]. Bias wasassessed through the mean prediction error(MPE) and its 95% confidence interval (CI) andprecision was assessed through the absoluteprediction error (APE) and the root meansquared error (RMSE) and 95% CI. The valuesof the two models were compared.

T. CORREA ET AL.514


Results

Preliminary analysis of the basic pharmacoki-netic model [CL/F 5 y1] combining a differenterror model pointed to the superiority of theproportional error model for estimating inter-individual variability and the additive errormodel for estimating residual variability. Thiserror model was used in all subsequent analyses.

The mean value obtained for VPA clearancefrom the simplest basic model was 0.365 l/h(OBJF 5 961.089). The interindividual variabilityexpressed as the coefficient of variation was30.9% and the residual variability for this basicmodel was 25 mg/l.

Stepwise incorporation of patient covariatesand the use of the criteria defined to evaluate thegoodness of fit led to the full population model.Subsequently covariates were excluded one byone from the full model to obtain the final model,by the use of a more restrictive criterion thatensures the inclusion, within the CL model, ofonly the covariates with more influence on thisparameter. The model-building process for char-acterizing VPA pharmacokinetics in the Mexicanpediatric population is shown in Table 2.

These results show that VPA clearance in-creases with either the daily dose of valproate(DD), the total body weight (TBW) and withconcomitant phenobarbital (PB) administration.

Table 3 shows the population parameterestimates and associated percentage relativestandard errors of the final model. The finalmodel allows the explanation of 50% of theinterindividual variability determined for CLwith the basic model to obtain a CV of 14.1%for this parameter. The residual variability of theconcentrations from the basic to the final modelwas reduced by 31%, to a value of 17.3 mg/l for aVPA concentration of 75 mg/l, the middle of thetherapeutic range for VPA (50–100 mg/l). Allfixed effects were estimated precisely; none of95% CI included the value zero.

Figure 1 depicts the scatterplot of thepredicted vs observed VPA serum concentrationsfrom the basic to the final model, showing theimprovement in the goodness of fit.

The results of prediction error are summari-zed in Table 4. The final model showedbetter prediction performance than the basicmodel in terms of bias and precision as shownin Figure 2.

Discussion

The determination of anticonvulsive drugconcentrations in serum may provide importantinformation in monitoring patients withepilepsy.

Table 2. Structural evolution related to the model-building process

Model Pharmacokinetic parameter Objective function (OBJF)

Basic CL/F 5 0.365 961.089Intermediate CL/F 5 0.58710.00355 TBW10.000277 DD 844.110Full 5 Final CL/F 5 (0.046610.00363 TBW10.000282 DD) (110.236 PB) 838.108

CL, clearance (l/h); TBW, total body weight (kg); DD, daily dose of valproate (mg/day); PB, phenobarbital.

Table 3. Final parameter estimates

Parameter Meaning Estimate SE (%) 95% CI

y1 Clearance 0.0466 33.4 0.0154, 0.0778y2 Proportionality constant for TBW 0.00363 22.2 0.0020, 0.0052y3 Proportionality constant for DD 0.000282 12.0 0.0002, 0.0003y4 Proportionality constant for PB 0.236 33.3 0.065, 0.406o2

CL/F Interindividual variability of CL/F 0.0200 45e (mg/l) Residual variability 17.3 28

y, population specific pharmacokinetic parameter; SE, standard error; CI, confidence interval.



The principal aim of population pharmacoki-netic analysis is to account for inherent kineticvariability in a population of patients, due tophysiological and pathological factors and treat-ment characteristics (e.g. age, sex, renal or hepaticfunction). This information can be used to designrational dosage guidelines that should result in

therapeutic concentrations. For valproate, theprediction of the appropriate dose and dosinginterval for a child or adolescent is complicatedbecause of the unusual pharmacokinetics of thisdrug and the limited knowledge of the influenceof age and body size upon its disposition.

The present research is the first populationpharmacokinetic study of valproate performed inMexico. Given the wide variability in pharmacoki-netics, the data will help to improve dosing ofmagnesium valproate in Mexican pediatric patients.

The population pharmacokinetic model pro-posed in this work demonstrates the significanteffect on VPA clearance of factors that are relatedto the hepatic metabolism, such as total bodyweight (related to the age of the child), poly-therapy (PB co-administration, which is knownto induce liver metabolism enzymes), and byfactors that affect plasma protein binding, as thedaily dose (mg/kg). These results coincide withprevious studies [13,15,25,26].

A large part of the CL variability wasexplained by patients’ covariates. The variabilitybetween-subjects (CV%) of VPA CL in Mexicanpediatric epileptic patients decreased from 30.9%in the basic model to 14.2% corresponding tothe final model. The introduction of covariatesallowed a 50% reduction in the interindividualvariability of VPA clearance. In addition, thereduction in residual variability (expressed asstandard deviation) decreases from 25 mg/l inthe basic model to 17.3 mg/l corresponding tothe final model. This residual error was similar tothat found by Blanco-Serrano et al. [13], who alsoused an additive model. The high residualvariability would mainly be because the Ka, Fand Vd values were fixed from the literature, andalso the variability must be considered in theanalytic assay. Additionally, there is an inherentproblem in the use of retrospective data.

In the present study, the TBW yields a moreaccurate CL estimation than age from the NON-MEM analysis. Age and weight are strictlyrelated to the development and functionalityof organs responsible for drug elimination.However, weight may be a better indicator ofthe patient’s physiological conditions and meta-bolic capacities than age. The final regression

Figure 1. Scatterplot of the predicted concentrations versusobserved concentrations for the (A) basic model and (B) finalmodel (including the identity line)

T. CORREA ET AL.516


model obtained for the CL of VPA suggests thatthis pharmacokinetic parameter increases line-arly with increments in TBW of the patient,which agrees with a study reported in theliterature [13].

According to our results, the valproate dailydose showed a linear relationship with VPA CL;the inclusion of this covariate improves thepopulation model fit and shows that patientsreceiving higher doses have a higher CL rate thanpatients receiving lower doses. This has beenreported by other authors in both adult andchildren patients [25,27,28]. This relationship isconsistent with the concept of concentration-dependent binding of VPA to proteins. Theprotein binding is a saturable process withinthe therapeutic range. Therefore, an increasedVPA dose leads to an increased total concentra-

tion of the drug, a consequent increase in its freefraction and associated increased CL.

The inclusion of the dose in the populationmodel, even though justified by the dose-depen-dent kinetics of valproate, can be attributedpartially to the effect of drug monitoring, whichlimits the use of higher doses in patients whoshow a higher CL [29,30]. The model accepts orpresumes that CL is induced by the dose in allpatients, without taking into consideration thosepatients who have an intrinsically elevated CL.

Figure 3 shows the average CL values (l/h)within the weight range of 5–60 kg for pediatricpatients taking different doses of magnesiumvalproate on monotherapy, obtained by applyingthe proposed model. Low weight patients have asmall change in valproate clearance in a widerange of normalized dose by weight (mg/kg/day). This is due to these patients receiving totaldaily doses substantially less than patients with ahigher total body weight, for this reason, in lowweight patients it is not feasible to observeimportant changes in valproate clearance.

In the treatment of epilepsy, it is important toknow possible drug interactions to anticipate theirclinical effects and to reduce the risk of bothtoxicity and seizures worsening when a drug isadded to or withdrawn from the patient’s regimen[31]. NONMEM analysis is superior in druginteraction detection compared with traditionalmeans because of its flexible experimental design.

Previous studies have shown that concomitanttherapy of VPA with other antiepileptic drugsincreases VPA clearance in both adults andchildren as a consequence of their enzymeinduction capacity [27,32–34]. In agreement withthese studies, the final CL regression modelshowed that PB, a known enzyme inducer thatenhances the biotransformation of VPA [34,35],

Figure 2. Prediction error of concentrations estimated withthe basic and final model in the validation data set

Table 4. Prediction errors established in the validation set by the basic and final models

Error Basic model Final model

Mean prediction error (95% CI) �12.68 (1.29, �26.65) �6.79 (�11.65, 1.93)Mean absolute error (95% CI) 35.625 (26.50, 44.74) 13.47 (10.27, 16.67)Mean squared prediction error (95% CI) 2127.65 (898.95, 3356.35) 285.01 (145.0, 425.0)Root mean squared prediction error (95% CI) 46.12 (26.61, 65.62) 16.88 (10.3, 23.46)

95% CI, 95% confidence interval.



produces an increase of 28.9% in this parameter,which would imply the administration of highervalproate doses to achieve therapeutic concen-trations. This increase in the CL of VPA is lessthan that reported in a previous study in healthyadults (39%) [5].

Unlike the results reported by other authors[13,15] in our study the combined use of CBZ intreatments with magnesium valproate did notshow a significant influence on the CL of thisdrug. In any case, the low proportion of patients(16%) in our population treated concomitantlywith valproate and CBZ should be pointed out,which along with this influence being moderatewhen observed, could have been the reason forthis interaction not being evident in our study.

In the present study, inclusion of gender in thebasic model did not appear to improve the datafitting. This finding coincides with other studies[5,13,21], which did not observe statisticallysignificant differences in the CL of this drugbetween males and females. By contrast, it hasbeen reported that girls have a lower CL of VPAthan boys probably because estrogen secretion ingirls increases with maturity, and it is known theestrogens are able to inhibit microsomal enzymes[25,36].

Model validation is recognized as an importantstep in model establishment, and external valida-tion provides the most stringent method fortesting a developed model [37]. The accuracy ofthe obtained population model was evaluated byforecasting the serum VPA concentrations in avalidation population, whose data were not usedto calculate that model. The model predictions inthe validation group were found to have satis-factory precision and bias since the values of theMPE (bias) and RMSE (precision) between thepredicted and observed concentrations weresmall, and the MPE 95% CI includes zero. Thissatisfactory predictive performance of the popu-lation model may prove valuable as a means ofestimating a priori individual patient dosageschedules for VPA.

The final proposed model can be used toestimate the individual CL of a patient receivingmagnesium valproate to establish the initial dose.Figure 4 shows the dosage recommendations inchildren of different weights treated on a mono-therapy regimen to obtain a desired trough VPAsteady-state concentration between 60 and65 mg/l obtained by incorporating our popula-tion final into the Abbottbase pharmacokineticprogram (ABBOTTBASE; Pharmacokinetic Sys-tem, IL, USA). Because of changes in CL, therecommended doses decreased as weight in-creased. In addition, the interindividual andresidual variabilities still persisting in the model,together with the magnitude of the predictionerrors, suggest the need for using therapeuticdrug monitoring to guarantee VPA concentra-tions within the desired therapeutic range. Onelimitation of the model is that the data used inconstructing it do not include combination withdrugs other than PB and CBZ. As a result,because only the effect of the former on the CLof VPA can be demonstrated, the model cannotbe used in patients receiving other types ofcomedication. Despite this, because most patientscan be completely or optimally controlled onmonotherapy regimens, combination therapyshould be avoided whenever possible.

Estimation of pharmacokinetic parameters inthe target population rather than implementationof parameters established in other populations to

Figure 3. Relationship of VPA clearance and magnesiumvalproate daily dose for pediatric patients with a total bodyweight ranging between 5 and 60 kg

T. CORREA ET AL.518


improve individual estimations is advisablewhenever possible, and allows a more accurateand precise dosing regimen design based onlimited feedback concentrations. The investiga-tion of the VPA CL has been the subject ofmany publications worldwide [5,13–15,25,26].However, the difference in age, weight, dosageand analysis method of each study make itcomplicated to compare the absolute CL indifferent ethnic groups. The predictive perfor-mance of the valproate pharmacokinetic modeldeveloped in Spanish epileptic children [13] wasevaluated in our validation population (mono-therapy patients, taking into account the effectsof body weight and daily dosage). This analysisshowed that a priori predictions based on theSpanish model were less accurate and precise(MPE:34.098, APE: 34.573, RMSE: 39.24) thanpredictions based on our model.

The results of our study support the idea thatthe Mexican pharmacokinetic parameters ofantiepileptics should be determined to improvethe rationale for and cost effectiveness of antic-onvulsant therapy.

The pharmacostatistical model can be incorpo-rated in clinical pharmacokinetics computer

programs employed in hospitals and institutionsthat monitor VPA in blood samples. Its applica-tion will permit individual pharmacokineticsparameter determination and would allow theoptimization of antiepileptic therapy with thisdrug to minimize the recurrence of epilepticcrisis due to under-dosing or to prevent unto-ward secondary reactions caused by the admin-istration of toxic doses of magnesium valproatein the pediatric population.

References

1. Perucca E. Pharmacological and therapeuticproperties of valproate. Drugs 2002; 16: 695–714.

2. Rabasseda X. Magnesium valproate: A broad-spectrum anticonvulsant agent. Clinical Review.Drugs Today 1995; 31: 185–190.

3. Rabasseda X. Indicaciones para el uso deValproato de magnesio. Drugs Today 2000;36(Suppl 1): 1–2.

4. Sanchez-Alcaraz A, Quintana MB, Rodrıguez I.Valproic acid clearance in children with epilepsy.J Clin Pharm Ther 1998; 23: 31–34.

5. Blanco-Serrano B, Otero MJ, Santos Buelga D,Garcıa-Sanchez MJ, Serrano J, Domınguez-Gil A.Population estimation of valproic acid clearancein adult patients using routine clinical pharma-cokinetic data. Biopharm Drug Dispos 1999; 20:233–240.

6. Klotz U. The role of pharmacogenetics in themetabolism of antiepileptic drugs: pharmacoki-netic and therapeutic implications. Clin Pharma-cokinet 2007; 46: 271–279.

7. Allabi AC, Gala JL, Horsmans Y. CYP2C9,CYP2C19, ABCB1 (MDR1) genetic polymorph-isms and phenytoin metabolism in a BlackBeninese population. Pharmacogenet Genomics2005; 15: 779–786.

8. Almeida AM, Falcao AC, Sales F et al. Kineticprofile of carbamazepine in an adult Portugueseoutpatient population. J Clin Pharm Ther 1998; 23:315–321.

9. Chan E, Ti TY, Lee HS. Population pharmacoki-netics of phenytoin in Singapore Chinese. Eur JClin Pharmacol 1990; 39: 177–181.

10. Goto S, Seo T, Murata T et al. Populationestimation of the effects of cytochrome P450 2C9and 2C19 polymorphisms on phenobarbital clear-ance in Japanese. Ther Drug Monit 2007; 29:118–121.

11. Valodia P, Seymour MA, Miller R, McFadyen ML,Folb PI. Factors influencing the population phar-macokinetic parameters of phenytoin in adult

Figure 4. Magnesium valproate dosage recommendationsin patients of different weight (desired VPA Css troughE60 mg/l) treated on a monotherapy regimen



epileptic patients in South Africa. Ther Drug Monit1999; 21: 57–62.

12. Yukawa E, Mamiya K. Effect of CYP2C19 geneticpolymorphism on pharmacokinetics of phenytoinand phenobarbital in Japanese epileptic patientsusing Non-linear Mixed Effects Model approach.J Clin Pharm Ther 2006; 31: 275–282. DOI: 10.1111/j.1365–2710.2006.00712.x.

13. Blanco-Serrano B, Otero MJ, Santos Buelga D,Garcıa-Sanchez MJ, Serrano J, Domınguez-Gil A.Valproate population pharmacokinetics in chil-dren. J Clin Pharm Ther 1999; 24: 73–80.

14. Park HM, Kang SS, Lee YB et al. Populationpharmacokinetics of intravenous valproic acid inKorean patients. J Clin Pharm Ther 2002; 27:419–425.

15. Desoky E, Fuseau E, Din Amry S, Cosson V.Pharmacokinetic modeling of valproic acid fromroutine clinical data in Egyptian epileptic pa-tients. Eur J Clin Pharmacol 2004; 59: 783–790.

16. Jiang D, Wang L. Population pharmacokineticmodel of valproate and prediction of valproateserum concentrations in children with epilepsy.Acta Pharmacol Sin 2004; 25: 1576–1583.

17. Jiang D, Wang L, Wang Y, Li L, Lu W, Bai X.Population pharmacokinetics of valproate inChinese children with epilepsy. Acta PharmacolSin 2007; 28: 1677–1684. DOI: 10.1111/j.1745–7254.2007.00704.x.

18. Mosteller RD. Simplified calculation of body-surface area. N Engl J Med 1987; 317: 1098.

19. Jonsson EN, Karlsson MO. Xpose: an S-PLUSbased population pharmacokinetic/pharmacody-namic model building aid for NONMEM. ComputMethods Programs Biomed 1999; 58: 51–64.

20. Cloyd JC, Fisher JH, Kriel RI, Kraus DM. Valproicacid pharmacokinetic in children. IV: Effects ofage and antiepileptic drugs on protein bindingand intrinsic clearance. Clin Pharmacol Ther 1993;53: 22–29.

21. Botha JH, Gray AL, Miller R. A model forestimating individualized valproate clearancevalues in children. J Clin Pharmacol 1995; 35:1020–1024.

22. MacKichan JJ. Valproic Acid. In A Textbook for theClinical Application of Therapeutic Drug Monitoring,Taylor WJ, Diers Caviness MH (eds). AbbottLaboratories, Diagnostic Division: Irving Texas,1986; 269–280.

23. Sheiner LB, Rosenberg B, Marathe VV. Estima-tions of population characteristics of pharmaco-kinetic parameters from routine clinical data. JPharmacokinet Biopharm 1977; 5: 445–479.

24. Sheiner LB, Beal S. Some suggestions for measur-ing predictive performance. J Pharmacokinet Bio-pharm 1981; 9: 503–512.

25. Yukawa E, To H, Ohdo S, Higuchi S, Aoyama T.Population-based investigation of valproic acidrelative clearance using nonlinear mixed effectsmodeling: influence of drug–drug interaction andpatient characteristics. J Clin Pharmacol 1997; 37:1160–1167.

26. Yukawa E, Nonaka T, Yukawa M, Higuchi S,Kuroda T, Goto Y. Pharmacoepidemiologic in-vestigation of a clonazepam–valproic acid inter-action by mixed effect modeling using routineclinical pharmacokinetic data in Japanese patient.J Clin Pharm Ther 2003; 28: 497–504.

27. Hall K, Otten N, Johnson B, Irvine-Meek J, LerouxM, Seshia S. A multivariate analysis of factorsgoverning the steady-state pharmacokinetics ofvalproic acid in 52 young epileptics. J ClinPharmacol 1985; 25: 261–268.

28. Abadin JA, Duran JA, Sanchez A, Serrano JB.Total and free valproic acid: plasma level/doseratio in monotherapy. Methods Find Exp ClinPharmacol 1991; 13: 221–225.

29. Martin ES, Crismon ML, Godley PJ. Postinductioncarbamazepine clearance in adult psychiatricpopulation. Pharmacotherapy 1991; 11: 296–302.

30. Suzuki Y, Cox S, Hayes J, Walson PD. Carbama-zepine age-dose ratio relationship in children.Ther Drug Monit 1991; 13: 201–208.

31. Patsalos PN, Froscher W, Pisani F. The importanceof drug interactions in epilepsy therapy. Epilepsia2002; 43: 365–385.

32. Chiba K, Suganuma T, Ishizake T, Iriki T, Hori M.Comparison of steady-state pharmacokinetics ofvalproic acid in children between monotherapyand multiple antiepileptic drug therapy. J Pediatr1985; 106: 653–658.

33. Cloyd JC, Kriel RI, Fisher JH. Valproic acidpharmacokinetic in children. II: Discontinuationof concomitant antiepileptic drug therapy. Neurol-ogy 1985; 35: 1623–1627.

34. Riva R, Albani F, Conti M, Baruzzi A. Pharmaco-kinetic interactions between antiepileptic drugs:clinical considerations. Clin Pharmacokinet 1996;31: 470–493.

35. Summers B, Summers RS. Carbamazepine clear-ance in paediatric epilepsy patients: influence ofbody mass, dose, sex and co-medication. ClinPharmacokinet 1989; 17: 208–216.

36. Yukawa E. A feasibility study of the multiple-peakapproach for pharmacokinetic screening: popula-tion based investigation of valproic acid relativeclearance using routine clinical pharmacokineticdata. J Pharm Pharmacol 1995; 8: 1048–1052.

37. FDA. Guidance for Industry – Population phar-macokinetics Website. http://www.fda.gov/cder/guidance/1852fnl.pdf [22 January 2006].

T. CORREA ET AL.520


Population pharmacokinetics of valproate in Mexican children with epilepsy

Documents

Transcript of Population pharmacokinetics of valproate in Mexican children with epilepsy