Clinical Therapeutics/Volume 39, Number 6, 2017

Population Pharmacokinetics of Valproic Acid in

Patients with Mania: Implication for Individualized
Dosing Regimens
Janthima Methaneethorn, PhD
Pharmacokinetic Research Unit, Department of Pharmacy Practice, Faculty of Pharmaceutical Sciences,
and Center of Excellence for Environmental Health and Toxicology, Naresuan University, Phitsanulok,
Thailand

ABSTRACT can subsequently be used for individualization of

Purpose: This study characterized the population optimum valproic acid maintenance dose. (Clin Ther.
pharmacokinetic properties of valproic acid in pa- 2017;39:1171–1181) & 2017 Elsevier HS Journals,
tients with mania and determined potential factors Inc. All rights reserved.
that affect the pharmacokinetic properties of valproic Key words: bipolar disorder, mania, pharmacoki-
acid in this population. netic properties, population pharmacokinetic model,
Methods: Routine therapeutic drug monitoring of valproate, valproic acid.
valproic acid concentrations, demographic data, and
concomitant medications from 206 hospitalized pa-
tients with mania were retrospectively collected from
Somdet Chaopraya Institute of Psychiatry and
INTRODUCTION
Srithanya Hospital, Thailand. Nonlinear mixed-
Although valproic acid (VPA) is widely used as an
effect modeling was used for data analysis. Covariate
antiepileptic drug in the treatment of generalized and
model building was conducted using stepwise for-
partial seizures, it also exerts activity for treatment of
ward addition and stepwise backward elimination.
bipolar disorder and prevention of migraine attacks.1,2
The final model was evaluated using bootstrap
The drug is characterized as having a narrow therapeutic
analysis and normalized prediction distribution
window and large intersubject pharmacokinetic varia-
error. The results were compared with those previ-
bility. The initial VPA dose of 750 mg/d is generally
ously reported in patients with epilepsy given that
recommended for the treatment of bipolar disorder. The
there is an evidence of a difference in valproic acid
dose is subsequently titrated upward as tolerated to a
clearance between patients with mania and those
maximum dose of 1000 to 2000 mg/d.3 The accepted
with epilepsy.
therapeutic range of VPA for seizure control is 50 to 100
Findings: Valproic acid data were adequately de-
mg/L. However, Hiemke et al4 reported that VPA levels
scribed by a 1-compartment model. Significant pre-
above 120 mg/L are tolerated in acute mania. The
dictors for valproic acid clearance included valproic
therapeutic range of 50 to 125 mg/L is usually
acid dose and weight. The population estimates for
suggested for the treatment of bipolar disorder.1,3,5 Most
valproic acid CL/F and V/F were 0.464 L/h and 23.3
of the drug (95%) is eliminated through hepatic metab-
L, respectively. Valproic acid clearance obtained from
olism. There is a high variability of volume of distribu-
this study did not seem to be significantly different
tion (Vd) of VPA, ranging from 0.1 to 0.5 L/kg.1 Given
from that of patients with epilepsy.
its narrow therapeutic window and large intersubject
Implications: A qualified population pharmacoki-
pharmacokinetic variability, therapeutic drug
netic model for valproic acid in patients with mania
was developed. This model could be used to optimize
Accepted for publication April 12, 2017.
valproic acid therapy in patients with mania. Valproic http://dx.doi.org/10.1016/j.clinthera.2017.04.005
acid clearance could be predicted from valproic acid 0149-2918/$ - see front matter
dose and weight of patients. This predicted clearance & 2017 Elsevier HS Journals, Inc. All rights reserved.

June 2017 1171

 Clinical Therapeutics
monitoring (TDM) and individualized dosing regimens
are important. In clinical settings where only one
blood sample per patient is usually obtained, it is
difficult to determine individual pharmacokinetic
parameters, and the mean parameters from the
reference population will be used for dosing.
However, these mean parameters were conducted in
white populations, which might be different from
Asian populations.6
Several studies have proposed that a population
pharmacokinetic model with Bayesian prediction offers
benefits in guiding dosage adjustment. Even though
several population pharmacokinetic modeling studies of
VPA have been reported, all of them were conducted in
patients with epilepsy. However, there is an evidence
that VPA clearances in patients with acute mania are
significantly higher than those in patients with epilepsy
(10.35 vs 7.70 mL/kg/h), which may be accounted for
by membrane transport system abnormalities that affect
cellular uptake of the drug and its Vd.7 Moreover, most
population pharmacokinetic modeling studies of VPA
conducted using routine TDM data did not address the
differences in absorption rate constant (Ka) between
extended-release and controlled-release dosage forms.
On the basis of these considerations with the lack of
pharmacokinetic studies describing factors that affect
pharmacokinetic characteristics of VPA in patients with
mania, the objectives of this study were to conduct a
population pharmacokinetic modeling study of VPA in
patients with mania using routine TDM data and to
identify potential covariates that affect pharmacokinetic
properties of VPA in this population. The results of this
study can help guiding clinicians in VPA dosage adjust-
ment in patients with mania.
METHODS
Patient Data
Data on steady-state serum concentrations of VPA
were collected retrospectively from Thai patients with
mania exposed to routine monitoring of VPA at
Somdet Chaopraya Institute of Psychiatry and Sritha-
nya Hospital, Thailand during January 2009 to
December 2016. Patients with mania were identified
and recorded in medical records by physicians accord-
ing to the International Classification of Diseases,
Tenth Revision (ICD-10) criteria. Patients of all ages
were eligible to be recruited in this study. The protocol
of this study was approved by ethics committee of
1172
Naresuan University, Somdet Chaopraya Institute of
Psychiatry and Srithanya Hospital. All patients re-
ceived VPA orally as enteric coated or controlled-
release dosage forms 1 to 3 times a day with the dose
range of 250 to 2000 mg/d. According to a routine
policy of drug sampling in the hospitals, data on
trough concentrations at the end of the dosing
interval, before the morning dose, were collected for
analysis to minimize diurnal variations in plasma
protein binding. All patients had normal kidney and
liver functions as well as normal serum albumin levels,
which are defined as serum creatinine o1.5 mg/dL,
alanine aminotransferase o45 U/L, aspartate amino-
transferase o45 U/L, and serum albumin levels within
the range of 3.5 to 5.0 g/dL.
Drug Analysis
Serum VPA concentrations were measured using
the Homogeneous Enzyme Immunoassay Technique
(Cobas Mira; Roche Diagnostics, Basel, Switzerland)
with a lower limit of quantitation of 1 mg/L. The
inter-assay and intra-assay %CV was o10%.
Population Pharmacokinetic Analysis
Population pharmacokinetic modeling of VPA was
conducted using a nonlinear mixed-effect modeling
approach via NONMEM software version 7.3 (ICON
Development Solutions, San Antonio, Texas). PDx-Pop
version 5.1 (ICON Development Solutions) and Xpose
version 4.3.0 (Uppsala University, Uppsala, Sweden)
were used to generate NONMEM output. Graphical
plots were performed using R 2.10.1 (The R Founda-
tion for Statistical Computing, Vienna, Austria). VPA
pharmacokinetic models were built using a first-order
conditional estimation method with interaction. Crite-
ria for model selection included (1) minimum objective
function value (MOFV), equivalent to minus twice the
log likelihood function; (2) Akaike information criteria,
equivalent to MOFV plus 2 times the number of
parameters; (3) a condition number, defined as the
ratio of the largest Eigen value to the smallest Eigen
value; (4) goodness-of-fit plots; and (5) plausibility and
precision of parameter estimates.
Given the sparse sampling strategy of VPA concen-
trations, basic model development was conducted
using a 1-compartment pharmacokinetic model with
first-order absorption and elimination in PREDPP
subroutines ADVAN2, TRANS2. Parameterization
of the model was used with Ka, CL/F, and V/F.
Volume 39 Number 6

Different types of interindividual variability (IIV; η) of
all pharmacokinetic parameters and residual variabil-
ity (RV; ε), including additive, proportional, and
exponential models, were tested. Model equations
for IIV and RV are described as follows:
Additive model:
IIV : Pi ¼ Ppop þ ηi
RV : Cij ¼Cpred;ij þεij
Exponential model:
IIV : Pi ¼ Ppop  expðηi Þ
RV : Cij ¼ Cpred;ij  expðεij Þ
Proportional model:
IIV : Pi ¼ Ppop  ð1 þηi Þ
IIV : Cij ¼Cpred;ij  ð1 þ εij Þ
where Pi is the estimated parameter for individual i, Ppop
is the typical population estimate for the parameter, ηi is
the deviation of Pi from Ppop, Cij is the jth observed VPA
concentration, Cpred,ij is the model predicted VPA con-
centration, and εij is the residual random error for
individual i and observation j. The η and ε random
effects were assumed to be independent and normally
distributed with mean of zero and variance ω2 and σ2 for
η and ε, respectively.
After the proper basic model was obtained, influences
of covariates on pharmacokinetic parameters were
evaluated. Correlations among potential covariates were
explored through correlation matrix, implemented in the
R program. Potential covariates to be tested during
covariate model building included weight, body mass
index, age, sex, co-medication, and VPA dose. Only co-
prescribed medication found in 410% of total patients
were included in the analysis. All the continuous
covariates were centered at the median values of the
population and modeled using linear centered, power, or
exponential functions, and categorical covariate (sex)
was modeled using additive function as follows:
Linear centered function : P¼θ1 þ θ2
 ðCOV–COVmedian Þ
Power function : P ¼ θ1  ðCOV=COVmedian Þθ2
Exponential function : P ¼ θ1  expðθ2  ðCOV
COVmedian Þ
Additive function : P ¼ θ1 þθ2  COV
June 2017
J. Methaneethorn
where P is the pharmacokinetic parameter, COV is the
tested covariate, COVmedian is the median value of
covariate, θ1 is the parameter estimate where the
covariate is equal to the median value, and θ2 is the
factor describing the association between the covariate
and the parameter.
Stepwise forward addition and stepwise backward
elimination approaches were used for statistical anal-
ysis. The significance levels of 0.05 and 0.001, which
correspond to the difference in MOFV of 3.84 and
10.83, respectively, were used as criteria for statistical
significance for stepwise forward addition and step-
wise backward elimination, respectively. Importance
of covariates as predictors was determined using a
reduction in IIV, statistical significance of MOFV, and
precision of parameter estimates as indicated by
relative SE.
Model Evaluation
The final population pharmacokinetic model of
VPA was evaluated for its stability, robustness, and
predictive ability using goodness-of-fit plots, bootstrap
analysis, normalized prediction distribution error
(NPDE), and a condition number. Plots of population
predicted versus observed VPA concentrations (PRED
vs DV), individual predicted versus observed VPA
concentrations (IPRED vs DV), conditional weighted
residual versus predicted VPA concentrations
(CWRES vs PRED), and conditional weighted residual
versus time (CWRES vs TIME) were evaluated for
their goodness of fit. Robustness of the model was
evaluated by a nonparametric bootstrap analysis
conducted using PDx-Pop. In brief, 1000 bootstrap
data sets were generated by resampling with replace-
ment from the original data set. Subsequently, the
final model was fitted to the generated bootstrap data
sets to obtain the median and 95% CIs of the
parameter estimates. These values were then com-
pared with those obtained from the original data set.
Predictability of the model was assessed using NPDE.8
One thousand Monte Carlo simulations from the final
model were used to compute NPDEs. Under the null
hypothesis that the final model adequately describes
the simulated dataset, NPDE should follow a normal
distribution with a mean of 0 and a variance of 1.
Three tests were used to evaluate the final model: (1) a
Wilcoxon signed rank test for the mean; (2) Fisher test
for the variance; and (3) Shapiro-Wilks test for a
normal distribution. Finally, the stability of the final
1173
 Clinical Therapeutics
model was assessed based on a condition number
generated during the covariance step of model run
using the PRINT¼E command. The value of o1000
indicates that the model is stable.
Model Simulation
One thousand simulated steady-state trough con-
centrations of VPA were conducted using the final
model for 5 dosing regimens (ie, 125 mg BID, 250 mg
BID, 500 mg BID, 750 mg BID, and 1000 mg BID) to
provide information on the effects of these different
dosing regimens.
RESULTS
Demographic Data
A total of 309 VPA concentrations from 206 Thai
patients with mania were included in this population
pharmacokinetic analysis. The mean weight, height,
and body mass index of the study population were
63.2 kg, 162.6 cm, and 24 kg/m2, respectively. The
mean age of the patients was 39.3 years (197 patients
aged between 18 and 65 years and 9 patients aged
465 years). Of all the samples, 48.5% were obtained
from females and 51.5% from males. Co-prescribed
medications in this population included perphenazine,
trihexyphenidyl, clonazepam, risperidone, haloperi-
dol, chlorpromazine, and lithium. The characteristics
of study population are summarized in Table I.
Population Pharmacokinetic Model
A 1-compartment model with first-order absorp-
tion and elimination best described VPA data. At-
tempt for a 2-compartment model building was not
made given that the available data are sparse. Because
most of the collected data were obtained at the end of
the dosing interval, they provided little information
pertaining to absorption process. Therefore, the Ka
value was fixed at the previous published values of
0.78 h1 and 0.38 h1 for enteric coated and
sustained-release dosage forms, respectively.9 The
estimated V/F and CL/F of VPA for a basic model
were 27.8 L and 0.467 L/h, respectively. The IIV was
best described by an exponential association, whereas
the RV was best explained by a proportional
association.
For the covariate model, the influence of weight,
sex, and dose of VPA was found to be significant on
CL/F during the first step of stepwise forward
1174
addition, as indicated by the reduction in MOFV of
43.84. The reduction in MOFV was greatest for the
effect of VPA dose on CL/F. Moreover, weight was
also found to be a significant predictor on V/F, with a
decrease in MOFV of 6.97. However, inclusion of the
effect of weight on V/F resulted in a relative SE of IIV
estimate of 110%, representing a substantially low
precision of the parameter estimate. Thus, the effect of
weight on V/F was not carried forward to the next
step. This noninformative IIV of V/F could be because
of the nature of data set containing steady-state VPA
levels. Therefore, the IIV for V/F was fixed to zero.
The effects of co-prescribed medications were tested
on CL/F. None of them was significant during
this step.
During the second step of covariate model building,
weight or sex was added to the model that contained
the effect of VPA dose on CL/F. Both weight and sex
were statistically significant predictors. However, in-
clusion of sex on CL/F resulted in a 95% CI of the
estimate containing zero and no reduction in %CV of
IIV for CL/F. Therefore, sex was not included as a
predictor on CL/F, and the full covariate model
consisted of the effect of VPA dose and weight on
CL/F. For backward elimination, both VPA dose and
weight were still significant predictors as indicated by
the increase in MOFV of 410.84. The effect of
protein-binding saturation was also tested using the
following equations10:
Bm  Cu
Cb ¼
Kd þ Cu
Ct ¼ Cb þCu
where Cb, Cu, and Ct are the bound, unbound, and
total valproic acid concentration (in milligrams per
liter), respectively, Bm is the apparent maximum
concentration of the binding site for valproic acid (in
milligrams per liter), and Kd is the apparent
dissociation constant of valproic acid (in milligrams
per liter). Because unbound drug concentrations were
not measured, the Bm and Kd were fixed at previously
published values of 130 mg/L and 7.8 mg/L,
respectively.11 The model with protein-binding satu-
ration resulted in higher MOFV (2241.018) compared
with the model incorporating the effect of dose
(MOFV ¼ 2202.699). Moreover, the relative SE of
estimated clearance and the IIV of clearance from a
protein-binding saturation model are higher (8.28%
Volume 39 Number 6
 J. Methaneethorn

parameters were described as follows:

Table I. Summary of patient demographic
characteristics. CL=FðL=hÞ¼ ½ð0:464  ðdose of VPA=17Þ0:402
Characteristic Finding  ðweight=60Þ0:779   expðηÞ

Weight, mean (SD) 63.2 (13.0) [40.0–107.2] To see the effect of fixed Ka on the final model, the
[range], kg Ka values for both enteric coated and sustain-released
Height, mean (SD) 162.6 (7.4) [147.0–183.0] formulations were varied in the range of 0.1 to
[range], cm 1.0 h1. By changing the Ka values within this range,
Body mass index, 24.0 (4.7) [14.9–37.8] the selection of covariates during forward addition
mean (SD) and backward elimination was not affected. Likewise,
[range], kg/m2 the parameter estimates obtained from the final model
Age, mean (SD), 39.3 (13.1) [18.2–73.1] were not affected by the change of Ka values, as
[range] y indicated by a slightly change in their estimates. The
Age group, no. (%) estimated CL/F, V/F, exponent of dose on CL/F, and
18-65 y 197 (95.6) exponent of weight on CL/F varied from 0.454 to
465 y 9 (4.4) 0.468 L/h, 21.8 to 25.4 L, 0.398 to 0.405, and 0.767
Sex, no. (%) to 0.784, respectively.
Female 100 (48.5)
Male 106 (51.5)
Valproic dose, mean Model Evaluation
(SD) [range] Figure 1 shows goodness-of-fit plots of VPA of
Per day, mg/d 1035.9 (382.7) [250–2000] the final model. Both population- and individual-
Per weight, 16.9 (6.8) [3.3–40.5] predicted VPA concentrations were randomly scat-
mg/kg/d tered along the identity lines, suggesting that the
Concomitant medication, no. (%) model optimally predicts the observed concentra-
Perphenazine 58 (28.2) tions. Moreover, the CWRES were randomly scat-
Trihexyphenidyl 143 (69.4) tered along the zero lines without significant trend,
Clonazepam 58 (28.2) indicating that statistical model is appropriate. The
Risperidone 70 (34.0) final model was repeatedly fitted to 1000 bootstrap
Haloperidol 39 (18.9) datasets. One hundred percent of the bootstrap
Chlorpromazine 59 (28.6) runs were successfully converged. The median para-
Lithium 17 (8.3) meter estimates and 95% CIs obtained from boot-
strap are summarized in Table III. The bootstrap
estimates were reasonably close to those obtained
and 42.8%, respectively), reflecting the imprecision of from the final model with o10% difference, and the
the estimated IIV. In addition, goodness-of-fit plots estimates obtained from the final model were
are subject to a higher bias of model prediction contained within the 95% CIs obtained from
compared with the model with the exponent of weight bootstrap, suggesting that the model is robust. The
on clearance (data not shown). Therefore, the model results of NPDE analysis are presented in Figure 2.
with the effect of dose on clearance was considered The mean and variance of NPDE were 0.08 and
superior to the model with protein-binding saturation 1.03, respectively. The Wilcoxon signed rank test
and was chosen as a final model. Goodness-of-fit plots for a mean, Fisher test for a variance, and Shapiro-
of the final model had no major bias in model Wilks test for a normal distribution were 0.105,
prediction (Figure 1). The results of covariate model 0.330, and 0.202, respectively. The results indicated
building are presented in Table II. The final that the final model has a satisfactory predictive
pharmacokinetic parameter and the IIV and RV ability. Finally, the condition number of the final
estimates are presented in Table III. The associations model was 8.0, indicating that the final model was
between significant covariates and pharmacokinetic stable.

June 2017 1175

 Clinical Therapeutics

200 200

Observed concentration (mg/L)

Observed concentration (mg/L)

150 150

100 100

50 50

0 0
0 50 100 150 200 0 50 100 150 200

Individual Predicted concentration (mg/L) Predicted concentration (mg/L)

4 4

3 3
Conditional Weighted Residuals

Conditional Weighted Residuals

2 2

1 1

0 0

-1 -1

-2 -2

-3 -3

-4 -4
0 50 100 150 0 50 100 150 200
Population predicted concentration (mg/L) Time (hours)

Figure 1. Goodness of fit plots of valproic acid for the final model. The upper plots are plots of observed
valproic acid concentration versus individual (left) and population (right) predicted valproic acid
concentration for the final model. The lower plots are plots of conditional weighted residuals versus
population predicted valproic acid concentration (left) and versus time (right).

Model Simulation DISCUSSION

The 5th, 50th, and 95th percentiles of simulated
steady-state trough concentrations of VPA are pre-
sented in Table IV. The results indicated that dosage
regimen of r500 mg BID may result in the trough
concentration below the therapeutic window.
TDM of VPA concentration levels for an individual’s
dose titration during treatment of bipolar disorder is
recommended by several guidelines given the narrow
therapeutic range and the high IIV of the drug. The
accepted therapeutic range of VPA required to treat

1176 Volume 39 Number 6

 J. Methaneethorn

Table II. Summary of covariate model development.

Step Covariate MOFV Difference in MOFV

Stepwise forward addition

0 Base model 2327.306 –
1 Weight on CL/F 2304.882 22.42
Sex on CL/F 2322.364 4.94
VPA dose on CL/F* 2269.239 58.07
Perphenazine on CL/F 2330.395 3.089
Trihexyphenidyl on CL/F 2328.035 0.729
Clonazepam on CL/F 2324.255 3.051
Risperidone on CL/F 2328.03 0.724
Haloperidol on CL/F 2327.285 0.021
Chlorpromazine on CL/F 2327.811 0.505
Lithium on CL/F 2328.146 0.84
Weight on V/F 2320.341 6.97
2 VPA dose on CL/F 2269.239 –
VPA dose on CL/F, weight on CL/F† 2202.699 66.54
VPA dose on CL/F, sex on CL/F 2262.096 7.143
Stepwise backward elimination
1 VPA dose on CL/F, weight on CL/F 2202.699 –
VPA dose on CL/F 2308.02 105.30
Weight on CL/F 2269.239 66.54

MOFV ¼ minimum objective function value; VPA ¼ valproic acid.

*
Initial model for second step.
†
Full covariate model.

patients with bipolar disorder is 50 to 125 mg/L.1,5,12
Model-based simulations for prediction of optimal
dosing schemes for VPA in both adult and pediatric
patients have been proposed.13–19 However, those were
conducted in patients with epilepsy in whom pharma-
cokinetic parameters of VPA might be different from
those in patients with bipolar disorder. On the basis of
this consideration, an effort to elaborate the population
characteristic that might affect VPA pharmacokinetic
properties in this population was made.
Nonlinear mixed-effect modeling was used for
population pharmacokinetic analysis. Patients in-
cluded in this study cover a large weight range
(40–107 kg), which might be of benefit in terms of
explaining weight-related pharmacokinetic changes.
Moreover, data collected from 2 hospitals could
potentially offset any system deviations between 2
centers. In this analysis, VPA pharmacokinetic proper-
ties were best described by a 1-compartment model
with first-order absorption and elimination. The sig-
nificant predictors for CL/F included VPA dose and
weight. The power relationship was used to describe
the VPA pharmacokinetic profile. The effect of weight
with an exponent of 0.779 indicates that the CL/F of
VPA increases with an increase in weight, which is
expected because there is an association between
weight and organ development and functionality
responsible for elimination.
Mohammadpour et al7 reported significantly higher
VPA clearances in patients with acute mania compared
with patients with epilepsy (0.621 vs 0.462 L/h for
patients with a mean weight of 60 kg). They proposed
that this could be because of membrane transport
system abnormalities that affect cellular uptake of the
drug and its Vd. Even though, direct comparison could
not be made given the difference in study design, the
result of this study revealed that the estimated VPA
CL/F in patients with mania was 0.464 L/h, which falls

June 2017 1177

 Clinical Therapeutics

Table III. Final population pharmacokinetic properties and bootstrap results for VPA.

Parameter Estimate (95% CI) %RSE* %CV† Bootstrap Estimate (95% CI)‡

CL/F, L/h 0.464 (0.449–0.479) 1.65 0.463 (0.448–0.478)

V/F, L 23.3 (6.52–40.1) 36.7 27.5 (12.6–57.6)
Ka, h1
Enteric coated 0.78 (fixed) – 0.78
Sustained release 0.38 (fixed) – 0.38
VPA dose on CL/F (power function) 0.402 (0.305–0.499) 12.3 0.405 (0.319–0.506)
Weight on CL/F (power function) 0.779 (0.605–0.953) 11.4 0.783 (0.603–0.963)
IIV-CL/F 0.037 (0.018–0.056) 26.4 19.1 0.035 (0.018–0.054)
RV (proportional error) 0.027 (0.017–0.036) 18.0 0.163 0.027 (0.019–0.038)

F ¼ bioavailability; IIV ¼ interindividual variability; Ka ¼ first-order absorption rate constant; %RSE ¼ relative SE;
RV ¼ residual variability.
*
Computed by (SE/mean)  100%.
†
Computed by (SD/estimate)  100%.
‡
The 95% CIs of bootstrap estimate are the 2.5 and 97.5 percentiles of the bootstrap estimates.

within the range of previous studies conducted in
patients with epilepsy.13,16,18,20,21 The discrepancy
between the results from this study and those proposed
by Mohammadpour et al7 could be because of their
small sample size of 19 patients with mania with a high
variation of CL/F. The estimated exponent of VPA dose
was 0.402, indicating that the CL/F of VPA increases
with dose. This finding is also consistent with previous
results and could be explained by the concentration-
dependent protein-binding characteristic of VPA.19
VPA is highly bound to albumin (90%–95%).1,22
Binding of VPA to albumin appears to become satu-
rated when the concentrations exceed 50 mg/L.1
Therefore, as the VPA dose increases, the CL/F
increases because more free fraction of VPA is
available to hepatic metabolism, resulting in a
disproportional increase in steady-state concentra-
tions.22 In addition, the increase in CL/F of VPA with
respect to the increase VPA dose could be attributed to
TDM effect in which patients with high CL/F tend to
receive a higher VPA dose. To better understand the
extent of nonlinearity, the effect of protein-binding
saturation was tested. However, on the basis of the
standard model evaluation criteria, the model with
protein-binding saturation was considered inferior to
the model with the exponent of VPA dose on CL/F.
Takinawa et al23 reported a significant effect of age
on ke of VPA. In that study, ke is linearly increased up
to the age of 12 years and is constant thereafter. This
could be because of the differentiation of elimination
organ functionality in children. Age was also evaluated
as a predictor for CL/F in this study. However, a
significant effect of age on VPA clearance was not
observed, which is expected given that the youngest
patient included in this analysis was 18 years old in
which the elimination organ should be fully developed.
Sex was a significant covariate during the first and
second steps of covariate model building. Neverthe-
less, there was no reduction in %CV of IIV for VPA
clearance, indicating that this predictor is not clin-
ically significant. Moreover, inclusion of sex in the
model yielded the 95% CI that contained zero,
confirming the insignificant effect of sex on VPA
clearance. Although Birnbaum et al24 reported that
VPA CL/F was 27% lower in female residents, they
proposed that this could be because of difference in
weight between males and females given that their
model did not include the effect of weight on CL/F.
The effects of several commonly co-prescribed
medications with VPA, including perphenazine, trihex-
yphenidyl, clonazepam, risperidone, haloperidol, chlor-
promazine, and lithium, were assessed. None of them
was significant as a predictor for VPA CL/F. Discrep-
ancies of the effect of risperidone on VPA concentrations
have been reported. According to a case report by van
Wattum,25 risperidone could potentially increase VPA

1178 Volume 39 Number 6

 J. Methaneethorn

3
2 40

30
1
Sample Quantiles

Frequency
0

20
-1

10
-2
-3

0
-3 -2 -1 0 1 2 3 -3 -2 -1 0 1 2 3
Theoretical Quantiles NPDE

3
3

2
2
1

1
NPDE

NPDE
0

0
-1

-1
-2

-2
-3

-3

0 50 100 150 200 0 50 100 150 200

Time (hour) PRED (mg/L)

Figure 2. Normalized prediction distribution errors. The upper plots are quantile-quantile plot of NPDE
versus the expected standard normal distribution (left) and histogram of NPDE (right). The lower
plots are plots of NPDE versus time (left) and versus predicted valproic acid concentration (right).

concentration in a 10-year-old boy with mood swings.26
In contrast, Bertoldo27 reported a single case of a 15-
year-old girl with VPA concentrations that decreased
from 80 to 57 mg/L when risperidone was added to the
treatment regimen. However, Sund et al28 reported that
no change in VPA concentration was observed when co-
administered with risperidone. As previously mentioned
because there is a high IIV in the pharmacokinetic
properties of VPA, the effect of risperidone on VPA
pharmacokinetic properties should not be concluded
based on data from a small number of patients. The
result of this study of a large number of patients
confirmed the result reported by Sund et al28 that
risperidone does not affect the pharmacokinetic
properties of VPA. As for the effect of haloperidol, the
result of this study is consistent with that reported by
Ishizaki et al29 in that there was no significant effect of
haloperidol on VPA pharmacokinetic properties. The
authors investigated the effect of haloperidol on VPA
pharmacokinetic properties in 6 patients with
schizophrenia. All patients received VPA 200 mg twice
daily with and without haloperidol 6 to 10 mg/d. The
results revealed no significant interaction between VPA
and haloperidol. The influence of chlorpromazine on
VPA pharmacokinetic properties was also reported by
Ishizaki et al29 in the same study. They found a

June 2017 1179

 Clinical Therapeutics
Table IV. Simulated valproic acid trough concentrations at different dosing regimens.
5th Percentile of 1000 50th Percentile of 1000
Dosing Simulated Valproic Acid Simulated Valproic Acid
Regimen Trough Concentrations, mg/L Trough Concentrations, mg/L
125 mg BID 8.90
250 mg BID 17.81
500 mg BID 35.61
750 mg BID 53.42
1000 mg BID 71.22
significant increase in VPA concentration when co-
administered with chlorpromazine. In contrast, the effect
of chlorpromazine on VPA CL/F was not significant in
this population pharmacokinetic modeling study. A
study by Yukawa et al21 in 250 Japanese patients aged
0.3 to 32.6 years found that concomitant administration
of clonazepam with VPA resulted in a 17.9% decrease
in VPA clearance with an unknown mechanism of
interaction. In this analysis, the effect of clonazepam
on VPA clearance was almost significant with a
reduction in MOFV of 3.051. The nonsignificant effect
could be attributed to the small number of patients
(30%) receiving both drugs concomitantly in this study.
This study has some limitations in that the data were
collected retrospectively and the number of samples per
patients was limited. Moreover, given that only trough
concentrations were available for population pharma-
cokinetic analysis, the effect of formulations (ie, enteric
coated and sustained-release dosage forms) could not
be investigated in this study. Despite these limitations,
this study has many strengths, such as the large number
of patients with a wide range of population character-
istics and various concomitant medications that had
not been investigated. In addition, this is the first study
that provides evidence on population pharmacokinetic
properties of VPA in patients with mania, which would
be valuable for clinical utility.
CONCLUSIONS
In conclusion, a population pharmacokinetic model
for VPA in patients with mania was developed. The
model has stability, robustness, and good predictive
ability, as indicated by a condition number, bootstrap
analysis, and normalized prediction distribution error.
1180
95th Percentile of 1000
Simulated Valproic Acid
Trough Concentrations, mg/L
11.86 15.83
23.73 31.67
47.46 63.34
71.19 95.00
94.92 126.67
Given these factors, this model together with Bayesian
estimation, an approach that relies on prior knowl-
edge, could be efficiently used to optimize individu-
alized VPA dosage adjustment.
ACKNOWLEDGMENTS
The author thanks Dr. Orabhorn Suanchang for
accommodating data acquisition and medical record
officers at Somdet Chaopraya Institute of Psychiatry
and Srithanya Hospital, Thailand, for provision of
relevant medical records.
CONFLICTS OF INTEREST
The author has indicated that she has no conflicts of
interest regarding the content of this article.
FUNDING SOURCES
This study received financial support from Coordinat-
ing Center for Thai Government Science and Tech-
nology Scholarship Students (CSTS), National Science
and Technology Development Agency (NSTDA).
REFERENCES
1. Winter ME. Basic Clinical Pharmacokinetics. 5th ed. Phila-
delphia: Lippincott Williams & Wilkins; 2010.
2. Peterson GM, Naunton M. Valproate: a simple chemical
with so much to offer. J Clin Pharm Ther. 2005;30:417–421.
3. Leo RJ, Narendran R. Anticonvulsant use in the treatment
of bipolar disorder: a primer for primary care physicians.
Prim Care Companion J Clin Psychiatry. 1999;1:74–84.
4. Hiemke C, Baumann P, Bergemann N, et al. AGNP
consensus guidelines for therapeutic drug monitoring in
psychiatry: update 2011. Pharmacopsychiatry. 2011;44:
195–235.
Volume 39 Number 6

5. Hirschfeld RM, Bowden CL, Gitlin
MJ, et al. Practice guideline for the
treatment of patients with bipolar
disorder. 2nd ed: American Psychiat-
ric Associaiton Web site, April 2002.
http://www.psychiatryonline.com/
pracGuide/pracGuideTopic_8.aspx.
Accessed March 8, 2017.
6. Saruwatari J, Ishitsu T, Nakagawa K.
Update on the genetic polymor-
phisms of drug-metabolizing en-
zymes in antiepileptic drug therapy.
Pharmaceuticals. 2010;3:2709–2732.
7. Mohammadpour AH, Foroughipour
M, Azarpazhooh MR, et al. Com-
parison of Valproic acid Clearance
between Epileptic Patients and Pa-
tients with Acute Mania. Iran J Basic
Med Sci. 2011;14:546–550.
8. Comets E, Brendel K, Mentré F.
Computing normalised prediction
distribution errors to evaluate non-
linear mixed-effect models: the npde
add-on package for R. Comput Meth-
ods Programs Biomed. 2008;90:154–
166.
9. Bondareva IB, Jelliffe RW, Sokolov
AV, Tischenkova IF. Nonparametric
population modeling of valproate
pharmacokinetics in epileptic pa-
tients using routine serum monitor-
ing data: implications for dosage. J
Clin Pharm Ther. 2004;29:105–120.
10. Ding J, Wang Y, Lin W, et al. A
population pharmacokinetic model
of valproic acid in pediatric patients
with epilepsy: a non-linear pharmaco-
kinetic model based on protein-
binding saturation. Clin Pharmacokinet.
2015;54:305–317.
11. Ueshima S, Aiba T, Makita T, et al.
Characterization of non‐linear rela-
tionship between total and unbound
serum concentrations of valproic
acid in epileptic children. J Clin
Pharm Ther. 2008;33:31–38.
12. Yatham LN, Kennedy SH, O'dono-
van C, et al. Canadian Network for
Mood and Anxiety Treatments
(CANMAT) guidelines for the man-
agement of patients with bipolar
disorder: consensus and controver-
sies. Bipolar Disord. 2005;7:5–69.
June 2017
13. EL Desoky ES, Fuseau E, EL Din
Amry S, Cosson V. Pharmacokinetic
modelling of valproic acid from
routine clinical data in Egyptian
epileptic patients. Eur J Clin Pharma-
col. 2004;59:783–790.
14. Blanco‐Serrano B, Otero MS, San-
tos‐Buelga D, et al. Population esti-
mation of valproic acid clearance in
adult patients using routine clinical
pharmacokinetic data. Biopharm
Drug Dispos. 1999;20:233–240.
15. Jiang D-C, Wang L. Population phar-
macokinetic model of valproate and
prediction of valproate serum con-
centrations in children with epilepsy.
Acta Pharmacol Sin. 2004;25:1576–
1583.
16. Park HM, Kang SS, Lee YB, et al.
Population pharmacokinetics of in-
travenous valproic acid in Korean
patients. J Clin Pharm Ther. 2002;27:
419–425.
17. Jiang Dc, Wang L, Li L, Lu W.
Population pharmacokinetics of val-
proate in Chinese children with epi-
lepsy. Acta Pharmacol Sin. 2007;28:
1677–1684.
18. Correa T, Rodríguez I, Romano S.
Population pharmacokinetics of val-
proate in Mexican children with
epilepsy. Biopharm Drug Dispos.
2008;29:511–520.
19. Lin W-W, Jiao Z, Wang C-L, et al.
Population pharmacokinetics of val-
proic acid in adult Chinese epileptic
patients and its application in an
individualized dosage regimen. Ther
Drug Monit. 2015;37:76–83.
20. Botha J, Gray A, Miller R. A model for
estimating individualized valproate
Address correspondence to: Janthima Methaneethorn, PhD, Faculty of
Pharmaceutical Sciences, Naresuan University, Phitsanulok 65000, Thailand.
E-mail: janthima.methaneethorn@gmail.com
J. Methaneethorn
clearance values in children. J Clin
Pharmacol. 1995;35:1020–1024.
21. Yukawa E, Nonaka T, Yukawa M,
et al. Pharmacoepidemiologic inves-
tigation of a clonazepam‐valproic
acid interaction by mixed effect
modeling using routine clinical phar-
macokinetic data in Japanese pa-
tients. J Clin Pharm Ther. 2003;28:
497–504.
22. Bauer L. Applied Clinical Pharmacoki-
netics. 3rd ed. New York: McGraw-
Hill Education; 2014.
23. Takinawa K, Matsumoto Y, Matsu-
moto M, et al. Population pharma-
cokinetic parameters of valproic
acid; conventional and slow release
formulation. Jpn J Clin Pharmacol
Ther. 1998;29:489–494.
24. Birnbaum AK, Ahn JE, Brundage RC,
et al. Population pharmacokinetics
of valproic acid concentrations in
elderly nursing home residents. Ther
Drug Monit. 2007;29:571–575.
25. van Wattum PJ. Valproic acid and
risperidone. J Am Acad Child Adolesc
Psychiatry. 2001;40:866–867.
26. Vitiello B. Valproic acid and risper-
idone: Commentary. J Am Acad Child
Adolesc Psychiatry. 2001;40:867.
27. Bertoldo M. Valproic acid and ris-
peridone. J Am Acad Child Adolesc
Psychiatry. 2002;41:632.
28. Sund JK, Aamo T, Spigset O. Val-
proic acid and risperidone: a drug
interaction? J Am Acad Child Adolesc
Psychiatry. 2003;42:1–2.
29. Besag FM, Berry D. Interactions
between antiepileptic and antipsy-
chotic drugs. Drug Saf. 2006;29:
95–118.
1181