Journal of Clinical Pharmacy and Therapeutics (2002) 27, 419–425
Population pharmacokinetics of intravenous valproic
acid in Korean patients
H.-M. Park* MD , S.-S. Kang* MD , Y.-B. Lee* MD , D.-J. Shin* MD , O.-N. Kim
PhD ,
S.-B. Lee
MD PhD and D.-S. Yim
MD PhD
*Department of Neurology, Ghil Medical Center, Gachon Medical School and
Department of Pharma-
cology, College of Medicine, The Catholic University of Korea, Inchon & Seoul, Korea
SUMMARY
Objective: To determine population-based phar-
macokinetic parameters for intravenous valproic
acid, and the factors influencing these parameters,
in Korean adults.
Methods: Valproic acid concentrations were
obtained using a peak and trough sampling
scheme for 102 Korean epileptic patients who
were not taking concurrent antiepileptic medi-
cation. Three hundred and fifty-four serum
concentrations were analysed according to a one-
compartment model with a mixed effect model-
ling method (NONMEM Ver 5Æ0). The influence
of body-weight (kg), height, daily valproic acid
dose (mg ⁄ day), body mass index (kg ⁄ m2), sex, and
age on volume of distribution (Vd) and clearance
(CL) was assessed in the course of analysis.
Results: Vd and CL of valproic acid increased
with body-weight. No significant influence of the
other screened covariates was observed.
The final regression model was:
CL(L/h) = 0Æ849 3 (body
weight/60)0Æ702 :
Vd(L) = 15Æ1 3 body
weight/60)0Æ604 :
Interindividual variabilities (coefficient of vari-
ation) for CL and Vd were 32 and 18%, respect-
ively. Residual error including intraindividual
variability was 26Æ7%.
Received 7 September 2002, Accepted 10 October 2002
Correspondence: Dong-Seok Yim, Department of Pharmacology,
College of Medicine, The Catholic University of Korea, 505 Banpo-
dong, Socho-gu, Seoul, Korea 137–701. Tel.: +82 2 590 1200;
fax: +82 2 536 2485; e-mail: yimds@catholic.ac.kr
 2002 Blackwell Science Ltd
Conclusion: The current results may be used as a
basic reference to optimize drug therapy with
intravenous valproic acid. Further research on the
paediatric population is necessary to confirm the
non-linearity of the relation between body-
weight and Vd.
Keywords: mixed effect modelling, pharmaco-
kinetics, population, valproic acid
INTRODUCTION
Valproic acid has long been used for various seiz-
ure disorders and its therapeutic range is about 50–
100 lg ⁄ mL (1, 2). Although valproic acid is an
older drug, it is still an important antiepileptic with
a broad spectrum covering absence, myoclonic,
tonic–clonic and partial seizures.
As epilepsy is a disease requiring life-long con-
trol, an oral route of medication is essential for
good antiepileptic agents. Nevertheless, intraven-
ous valproic acid has been tried for status epilept-
icus (3) and for those who cannot take oral
medications. Its intravenous formulation has some
advantages, when compared with phenobarbital
and phenytoin, in that it causes no respiratory
depression or local pain (4, 5). This makes rapid
infusion and loading possible.
Population pharmacokinetic studies give sub-
stantial information about factors influencing
pharmacokinetic parameters in various groups of
patients. The pharmacokinetic parameters of valp-
roic acid have been studied in healthy subjects and
patients to find appropriate dosing regimens (6–8).
Such studies generally involve fewer than 20
subjects and the influence of various physical or
pathological conditions on the pharmacokinetic
parameters is hard to assess. To confirm the
influence of various factors on pharmacokinetic
parameters in patients, population studies with
419
420 H.-M. Park et al.
mixed effect modelling software (NONMEM) have
been performed for a variety of drugs. There are
some reports of the investigation of valproic acid
population pharmacokinetics for oral formulations
(9, 10). The serum concentration data, mostly as
trough concentrations, was collected retrospec-
tively after oral intake and CL was the only phar-
macokinetic parameter estimated.
To evaluate the population pharmacokinetic
parameters of i.v. valproic acid, an aspect not pre-
viously investigated, we performed a prospective
blood sampling study in otherwise healthy epi-
leptic patients undergoing valproic acid mono-
therapy.
METHODS
Subjects
Serum valproic acid concentrations were measured
for 102 Korean patients, prospectively recruited to
the study. All patients were previously healthy and
were admitted for the work-up of their newly
appeared seizure symptoms. All had normal liver
and kidney function and were not taking any other
antiepileptics or drugs that could influence valp-
roic acid metabolism. Twenty of them had a history
of cerebral infarction without serious sequelae on
daily life activities. Their demographic character-
istics are summarized in Table 1. Valproic acid was
given intravenously to manage seizure symptoms.
A loading infusion was given as a total daily dose
over 30–60 min. A maintenance dose was given as
a divided dose every 6–12 h. The duration of the
maintenance infusion was <1 min. The time nee-
ded for each loading and maintenance infusion
was recorded to the second.
Peak sampling was performed 10–60 min after
the end of the loading infusion and trough sam-
Table 1. Demographic profile of the patients undergoing
intravenous valproic acid therapy
Characteristic Mean ± SD (range)
Age (years) 45Æ4 ± 19Æ2 (16–81)
Sex (male ⁄ female) 66 ⁄ 36
Daily Dose (mg ⁄ day) 17Æ8 ± 3Æ8 (5.5–33.3)
Body-weight (kg) 60Æ4 ± 10Æ8 (40–91)
Body mass index (kg ⁄ m2) 22Æ1 ± 3Æ2 (15.6–31.6)
 2002 Blackwell Science Ltd, Journal of Clinical Pharmacy and Therapeutics, 27, 419–425
plings at exactly 60 min before the maintenance
infusions in every patient. The sampling time points
were recorded to the minute. Serum valproic acid
levels were measured by fluorescence polarization
immunoassay (TDx analyser; Abbott diagnostics
division, Abbott Laboratories Co., USA). The study
protocol was approved by the institutional review
board of Ghil Medical Center.
Population pharmacokinetic analysis
Three hundred and fifty-two serum concentra-
tions were obtained for 102 patients. Analysis
assuming a one-compartment model based on
a mixed effect method was performed using
NONMEM Ver 5Æ0.
Error models. Pharmacokinetic parameters for this
model were Vd and CL (PREDPP subroutines
ADVAN1 TRANS2) and the following models
were used to describe the interindividual variabil-
ity in pharmacokinetic parameters.
Error models to measure the interindividual
variability are as following equations:
CLj ¼ CLpop  egCL :
Vdj ¼ Vdpop  egVd :
where CLj and Vdj represent parameters for
the jth individual, and CLpop and Vdpop represent
those expected as functions of the individual
covariates such as weight and age. gCL and gVd are
independent random-error variables with mean
values of zero and variances of x2CL and x2Vd,
respectively.
Residual error, describing the difference
between observed concentrations and those pre-
dicted with CLj and Vdj, was modelled as follows:
Cij ¼ CijðpredÞ  eij
where Cij represents observed ith concentration in
the jth individual and Cij(pred) that predicted with
the jth individuals, pharmacokinetic parameters
(CLj and Vdj), dose history and sampling time. eij,
the residual error, is a random variable with a
mean of zero and variance r2 that accounts for all
discrepancies caused by intraindividual variation
and other errors.
Covariate model building. In the first step of the
NONMEM analysis, a basic structural model with
 Valproic acid in Korean population 421

no covariates was fitted to the data with the
POSTHOC option, which allows Bayesian esti-
mation of Vd and CL. The contribution of cova-
riates such as patients’ body-weight, age, sex,
height, BMI [body mass index ¼ body-weight
(kg) ⁄ height2 (m2)], and daily dose on the Bayesian
estimates of Vd and CL was screened by Pear-
son’s correlation analysis with the P-value set at
0Æ05. However, as the covariates were few, all of
them, regardless of the correlation analysis result,
were tested in preliminary screening for structural
models (Table 2). In this process, each covariate
was added to the basic model sequentially with
cut-off criteria set at P < 0Æ05 (decrease in MVO
>3Æ84 in comparison with basic model) (11).
Covariates that screening showed to be significant
were incorporated into the full model and the
contribution of each covariate was assessed by
eliminating them one by one from the full model.
The P-value was set at 0Æ01 (increase in MVO
>8Æ67 in comparison with the model before elim-
ination of the covariate) (11) for this confirmation
process.
Functions describing the relationship between
covariates and pharmacokinetic parameters were
assumed to be non-linear (h1 · covariateh2), rather
than linear (h1 · covariate + h2), except for sex, a
discrete variable. This was based on the empiricism
that a relationship without an intercept is physio-
logically preferable to those with an intercept.
Scatter plots of observed vs. predicted concentra-
tions and weighted residuals vs. predicted con-
centrations were also considered in the selection of
models.

Table 2. Preliminary screening to find covariates influencing CL or Vd of intravenous valproic acid

Model structure for population

Hypothesis tested CL and Vd MVO P-value* Conclusion

Basic model without covariate CL(L ⁄ h) ¼ h1 2344Æ541 –

Vd(L) ¼ h2
Does body-weight influence CL? CL(L ⁄ h) ¼ h1 · (Wt ⁄ 60)h3 2335Æ425 <0Æ01 Yes
Vd(L) ¼ h2
Does age influence CL? CL(L ⁄ h) ¼ h1 · (Age ⁄ 45)h3 2343Æ799 No
Vd(L) ¼ h2
Does daily dose influence CL? CL(L ⁄ h) ¼ h1 · (Daily dose ⁄ 17Æ91)h3 2344Æ536 No
Vd(L) ¼ h2
Does BMI influence CL? CL(L ⁄ h) ¼ h1 · (BMI ⁄ 21Æ384)h3 2340Æ284 <0Æ05 Yes
Vd(L) ¼ h2
Does height influence CL? CL(L ⁄ h) ¼ h1 · (Height ⁄ 163Æ5)h3 2340Æ545 No
Vd(L) ¼ h2
Does sex influence CL? CL(L ⁄ h) ¼ h1 · (1 + Sex · h3) 2341Æ040 No
Vd(L) ¼ h2
Does body-weight influence Vd? CL(L ⁄ h) ¼ h1 2334Æ702 <0Æ01 Yes
Vd(L) ¼ h2 · (Wt ⁄ 60)h3
Does age influence Vd? CL(L ⁄ h) ¼ h1 2342Æ391 No
Vd(L) ¼ h2 · (Age ⁄ 45)h3
Does daily dose influence Vd? CL(L ⁄ h) ¼ h1 2344Æ234 No
Vd(L) ¼ h2 · (Daily Dose ⁄ 17Æ91)h3
Does BMI influence Vd? CL(L ⁄ h) ¼ h1 2341Æ176 No
Vd(L) ¼ h2 · (BMI ⁄ 21Æ384)h3
Does height influence Vd? CL(L ⁄ h) ¼ h1 2337Æ945 <0Æ01 Yes
Vd(L) ¼ h2 · (Height ⁄ 163Æ5)h3
Does sex influence Vd? CL(L ⁄ h) ¼ h1 2334Æ306 <0Æ01 Yes
Vd(L) ¼ h2 · (1 + Sex · h3)

*P-value for the MVO difference between the basic model and tested model.

 2002 Blackwell Science Ltd, Journal of Clinical Pharmacy and Therapeutics, 27, 419–425
422 H.-M. Park et al.
RESULTS
Results of preliminary screening of covariates are
shown in Table 2. Further analysis to find the final
model, based on these screening results, produced
the following:
CLðL/hÞ ¼ 0849  ðBody-weight=60Þ0702 :
VdðLÞ ¼ 151  ðBody-weight=60Þ0604 :
CL and Vd of valproic acid increased to the
power of the parameter theta. Although the influ-
ence of sex was the greatest on Vd in preliminary
screening, subsequent steps showed that only
body-weight was a significant influencing factor.
None of the other covariates significantly influ-
enced Vd or CL.
In a patient of body-weight 60 kg, predicted
CL and Vd were 0Æ849 L ⁄ h (14Æ2 mL ⁄ min,
0Æ24 mL ⁄ min per kg) and 15Æ1 L (0Æ25 L ⁄ kg),
respectively. The analysis results obtained by seri-
ally eliminating the covariates are in Table 3.
Interindividual variabilities (coefficients of vari-
ation) for CL and Vd were 32 and 18%, respect-
ively. Residual error, including intraindividual
variability, was 26Æ7%. The correlation between Vd
Table 3. Differences in the MVO between the full model and the models from which one of the covariates was deleted in
the CL or Vd model
Full* and reduced models
Full model
CL(L ⁄ h) ¼ h1 · (Wt ⁄ 60)h3 · (BMI ⁄ 21Æ384)h5
Vd(L) ¼ h2 · (Wt ⁄ 60)h4 · (Height ⁄ 163Æ5)h6 · (1 + Sex · h7)
Reduced models
CL(L ⁄ h) ¼ h1 · (Wt ⁄ 60)h3
Vd(L) ¼ h2 · (Wt ⁄ 60)h4 · (Height ⁄ 163Æ5)h5 · (1 + Sex · h6)
CL(L ⁄ h) ¼ h1 · (Wt ⁄ 60)h3
Vd(L) ¼ h2 · (Wt ⁄ 60)h4 · (Height ⁄ 163Æ5)h5
CL(L ⁄ h) ¼ h1 · (Wt ⁄ 60)h3
Vd(L) ¼ h2 · (Wt ⁄ 60)h4
CL(L ⁄ h) ¼ h1 · (Wt ⁄ 60)h3
Vd(L) ¼ h2
CL(L ⁄ h) ¼ h1
Vd(L) ¼ h2 · (Wt ⁄ 60)h3
CL(L ⁄ h) ¼ h1
Vd(L) ¼ h2
*Covariates of full model are based on the screening results of Table 2.
**from the comparison of the MVO calculated before and after the elimination of covariate.
NS: Not significant.
 2002 Blackwell Science Ltd, Journal of Clinical Pharmacy and Therapeutics, 27, 419–425
and CL was acknowledged when evaluated
with $OMEGA BLOCK in the control file for
NONMEM. The mean and variability estimates of
parameters are shown in Table 4. Scatter plots
of observed vs. predicted concentrations and
weighted residuals vs. predicted concentrations are
shown in Fig. 1.
DISCUSSION
Previous articles on oral administration of valproic
acid have shown a negative correlation between
patients’ age and CL in children (10), but not in
adults (9). Both reports also showed increasing
total CL with increasing body-weight. How-
ever, Yukawa et al. (10) showed that clearance per
body-weight decreased with body-weight (body-
weight)0Æ252), in a population aged 0Æ3–54Æ8 years.
This appears to correspond with our result, as their
total clearance equals body-weight(1–0Æ252) [body-
weight(0Æ748)]. In the study on adult patients,
Blanco-Serrano et al. (9) reported a direct propor-
tional relation with body-weight; however, the
same author has reported a non-linear relation
[body-weight(0Æ715)] in a paediatric population (12).
MVO P-value**
2317Æ968 –
2318Æ379 NS
2320Æ980 NS
2321Æ722 NS
2335Æ425 <0Æ01
2334Æ702 <0Æ01
2344Æ541 <0Æ01
 Valproic acid in Korean population 423

Table 4. Fixed effect parameters, interindividual variability and residual error of the finally chosen model

Estimated Standard 95% Confidence

Parameters Meaning value error interval

Structural
h1 Clearance (L ⁄ h) 0Æ850 0Æ034 0Æ782–0Æ918
h3 ¼ h1 · (Body-weight ⁄ 60)h3 0Æ702 0Æ182 0Æ338–1Æ066
h2 Vd(L) 15Æ1 0Æ497 14Æ106–16Æ094
h4 ¼ h2 · (Body-weight ⁄ 60)h4 0Æ604 0Æ192 0Æ22–0Æ988
Variance
2
xCL Interindividual variability in CL 0Æ101 0Æ021 0Æ059–0Æ143
2
xVd Interindividual variability in Vd 0Æ0514 0Æ0219 0Æ008–0Æ0952
2
r Residual variability 0Æ0712 0Æ0165 0Æ0382–0Æ1042
Covariance Covariance between gCL and gVd 0Æ0306 0Æ0127 0Æ005–0Æ0560

An increase in CL with daily dose has been a

Fig. 1. Upper: Scatter-plot of observed valproic acid
concentration vs. concentration predicted by final cova-
riate model. Bottom: Plot of weighted residuals vs. pre-
dicted concentrations.
common finding reported by several authors (10,
13, 14), in both adults and children. In a pharma-
cokinetic study on 52 young epileptics (14), valp-
roic acid clearance was higher in patients taking
more than 20 mg ⁄ kg per day than that in lower
dose groups. It is not clear whether the increased
valproic acid dose caused a reduction in absorption
or autoinduction of metabolism. Patients in the
present study were drug-naı̈ve; and their thera-
peutic drug level monitoring data were from the
first week of their antiepileptic therapy, which
makes the chance of metabolic autoinduction
unlikely. Additionally, the daily doses in our
patients (5Æ5–33Æ3 mg ⁄ kg per day) did not vary
enough to detect a dose–CL relationship. The typ-
ical valproic acid CL predicted in a Korean patient
of 60 kg with the current model was 0Æ24 mL ⁄ min
per kg, slightly higher than the CL predicted in
Japanese and Spanish patients (9, 10).
As there are no reports on the population phar-
macokinetics of Vd of valproic acid, the current
model for Vd was compared with results of con-
ventional pharmacokinetic studies. Bryson et al. (7)
reported increased Vd in patients older than
75 years in comparison with those of 20–35-year-
old patients (0Æ19 L ⁄ kg vs. 0Æ13 L ⁄ kg). We did not
observe this, but only three of our subjects were
older than 75 years.
In the preliminary screening of covariates, sex
difference was the most outstanding factor for Vd,
although body-weight was found to be more

 2002 Blackwell Science Ltd, Journal of Clinical Pharmacy and Therapeutics, 27, 419–425
424 H.-M. Park et al.

Table 5. Comparison of several non-linear and linear models suggested for the relation between Vd and body-weight.
The structural model for clearance was the same among models: CL ¼ h1 · (Body-Weight ⁄ 60)h3

Structural model Estimated Standard error

for Vd (L) Parameters value (CV) x2Vd r2 MVO

Non-linear h2 15Æ1 0Æ497 (3.3) 0Æ0514 0Æ0712 2321Æ722

h2 · (Body-weight ⁄ 60)h4 h4 0Æ604 0Æ192 (30)
Linear h2 9Æ06 2Æ86 (32) 0Æ0516 0Æ0711 2321Æ619
h2 · (Body-weight ⁄ 60)h4 h4 5Æ95 2Æ87 (48)
Linear without Intercept h2 15Æ1 0Æ517 (3.4) 0Æ0566 0Æ0710 2327Æ897
h2 · (Body-weight ⁄ 60)

CV: Coefficient of variation, %.

significant in the subsequent step. In the screening
process, male sex was predicted to increase Vd by a
factor of 0Æ24. This can be attributed to the larger
body-weight of the male subjects (54Æ7 ± 9Æ2 kg)
compared with the females (63Æ5 ± 10Æ4 kg).
In the course of NONMEM analysis to test the
linearity of body-weight and Vd (Table 5), the non-
linear model (Vd ¼ h1 · weighth2) was not differ-
ent from the linear model (Vd ¼ h1 · weight + h2)
in MVO, but the standard error of theta parameters
was smaller. The non-linear model had signifi-
cantly smaller MVO than the linear model with the
intercept (h2) fixed at zero. The use of an intercept
parameter for the linear model can improve the
goodness-of-fit, but the reason for this is not clear.
The body-weight linear model with intercept can
be applied to other populations if their demogra-
phic profile is similar to the original population
that was used in the model development. How-
ever, the model is likely to cause some overesti-
mation (intercept > 0) or underestimation (intercept
<0) when applied to patients with very low body-
weight (e.g. young children or neonates). For
further delineation of valproic acid Vd vs. body-
weight relationship, population pharmacokinetic
study in paediatric patients will be necessary.
As for antiepileptic drugs, routine serum con-
centration monitoring has been a mainstay for the
optimization of valproic acid therapy. In the pre-
sent study, we examined the relationship between
pharmacokinetic parameters and some commonly
considered factors such as body-weight and daily
dose, after intravenous infusion, isolated from the
implications of drug–drug or disease–drug inter-
actions. The widely scattered pattern in the pre-
dicted vs. observed levels (Fig. 1) reflects the large
interindividual and intraindividual variability in
Vd and CL (Table 4). This confirms the importance
of dose titration via careful clinical observation and
drug level monitoring when necessary.
ACKNOWLEDGEMENTS
This study was performed in accordance with the
regulations of the Korean Government. The au-
thors thank Ms. Eun-Shi Choi for her devotion and
excellence shown in this study.
REFERENCES
1. Chadwick DW. (1985) Concentration-effect relation-
ships of valproic acid. Clinical Pharmacokinetics, 10,
155–163.
2. Schobben F, van der Kleijn E, Gabreels FJ. (1975)
Pharmacokinetics of di-n-propylacetate in epileptic
patients. European Journal of Clinical Pharmacology, 8,
97–105.
3. Yamamoto LG, Yim GK. (2000) The role of intra-
venous valproic acid in status epilepticus. Pediatric
Emergency Care, 16, 296–298.
4. Limdi NA, Faught E. (2000) The safety of rapid
valproic acid infusion. Epilepsia, 41, 1342–1345.
5. Venkataraman V, Wheless JW. (1999) Safety of
rapid intravenous infusion of valproate loading
doses in epilepsy patients. Epilepsy Research, 35, 147–
153.
6. Perucca E, Gatti G, Frigo GM, Crema A, Calzetti S,
Visintini D. (1978) Disposition of sodium valproate
in epileptic patients. British Journal of Clinical Phar-
macology, 5, 495–499.
7. Bryson SM, Verma N, Scott PJ, Rubin PC. (1983)
Pharmacokinetics of valproic acid in young and

 2002 Blackwell Science Ltd, Journal of Clinical Pharmacy and Therapeutics, 27, 419–425

elderly subjects. British Journal of Clinical Pharmacol-
ogy, 16, 104–105.
8. Nitsche V, Mascher H. (1982) The pharmacokinetics
of valproic acid after oral and parenteral adminis-
tration in healthy volunteers. Epilepsia, 23, 153–162.
9. Blanco-Serrano B, Otero MJ, Santos-Buelga D,
Garcia-Sanchez MJ, Serrano J, Dominguez-Gil A.
(1999) Population estimation of valproic acid
clearance in adult patients using routine clinical
pharmacokinetic data. Biopharmaceutics and Drug
Disposition, 20, 233–240.
10. Yukawa E, To H, Ohdo S, Higuchi S, Aoyama T.
(1997) Population-based investigation of valproic
acid relative clearance using nonlinear mixed effects
modeling: influence of drug–drug interaction and
patient characteristics, Journal of Clinical Pharmacol-
ogy, 37, 1160–1167.
 2002 Blackwell Science Ltd, Journal of Clinical Pharmacy and Therapeutics, 27, 419–425
Valproic acid in Korean population 425
11. Boeckmann A, Sheiner L, Beal S. (1991) NONMEM
Users Guide, NONMEM Project Group, San Fran-
cisco.
12. Blanco-Serrano B, Garcia Sanchez M, Otero M,
Santos Buelga D, Serrano J, Dominguez Gil A. (1999)
Valproate population pharmacokinetics in children.
Journal of Clinical Pharmacology and Therapeutics, 24,
73–80.
13. Abadin JA, Duran JA, Sanchez A, Serrano JS. (1991)
Total and free valproic acid: plasma level ⁄ dose ratio
in monotherapy, Methods and Findings in Experimental
and Clinical Pharmacology, 13, 221–225.
14. Hall K, Otten N, Johnston B, Irvine-Meek J, Leroux
M, Seshia S. (1985) A multivariable analysis of fac-
tors governing the steady-state pharmacokinetics of
valproic acid in 52 young epileptics. Journal of Clin-
ical Pharmacology, 25, 261–268.