Embed Size (px)
Citation preview
ORIGINAL RESEARCH ARTICLE
The Integration of Allometry and Virtual Populations to PredictClearance and Clearance Variability in Pediatric Populationsover the Age of 6 Years
Andrea N. Edginton • Bhavank Shah •
Michael Sevestre • Jeremiah D. Momper
Published online: 16 April 2013
� Springer International Publishing Switzerland 2013
Abstract
Background and Objectives Pharmacokinetics play an
integral role in the pediatric drug development process.
The determination of pharmacokinetic parameters, partic-
ularly clearance, in different age groups directly informs
dosing strategies for subsequent efficacy trials. Allometric
scaling for prediction of pediatric clearance from the
observed clearance in adults has been used in this effort.
Clinical trial simulation, a powerful tool used to inform
clinical trial design, requires both an estimate of clearance
along with an estimate of the expected pharmacokinetic
variability. The standard deviations (SD) of individual
clearance values for adults are typically used and may lead
to inaccurate predictions by not taking into account the
more widespread distribution of factors such as body
weight in the pediatric population. The objective of this
study was to assess the accuracy of allometric prediction of
drug clearance as well as methods for predicting clearance
variability in children 6 years of age and older.
Methods US Food and Drug Administration (FDA) clini-
cal pharmacology reviews of pediatric studies conducted
from 2002 onwards were reviewed to collate adult and
pediatric clearance and clearance variability for studies
including children 6 years of age and older. A set of 1,000
virtual adults {A} and a set of 5,000 virtual children (aged
2–17) {P} were generated on the basis of the White Ameri-
can NHANES database. Pediatric clearances were predicted
in method 1 by using the geometric mean adult clearance
from the in vivo study and calculating pediatric clearance for
each virtual child within a subset {P0} of {P} that contained
only those children that were within the age range of the
in vivo pediatric study. In method 2, adult clearance values
were randomly generated from the geometric mean adult
clearance and standard deviation and assigned to each virtual
adult in {A}. For each adult, allometric clearance scaling was
completed with each virtual child within {P0}. The prediction
error for the predicted and observed clearance and the
clearance variability metric, coefficient of variation (CV),
was calculated. The prediction accuracy as a function of the
lowest age range (2 years and older) included in the study
was also assessed.
Results Thirty-nine unique drugs were included in the
study. For both method 1 and method 2, 100 % of predicted
pediatric mean clearances were within 2-fold of the observed
values and approximately 82 % were within a 30 % pre-
diction error. There was a significant increase in the pre-
diction accuracy of CV using method 2 vs. method 1. There
was a major bias towards underprediction of pediatric CV in
method 1 whereas method 2 was precise and not biased.
Clearance and CV prediction accuracy were not a function of
the age range included in the in vivo studies. The observed
CV between the adult and pediatric study groups was not
significantly different although, on average, the observed
pediatric CV was 32 % greater than that from adult studies.
Conclusions Allometric scaling may be a useful tool
during pediatric drug development to predict drug
The MatLab code is available to those who would like it upon
emailing the corresponding author.
A. N. Edginton (&) � B. Shah
School of Pharmacy, University of Waterloo, 200 University
St W, Waterloo, ON N2L 3G1, Canada
e-mail: [email protected]
M. Sevestre
Design2Code Inc., Waterloo, ON, Canada
J. D. Momper
Office of Clinical Pharmacology, Center for Drug Evaluation
and Research, US Food and Drug Administration,
Silver Spring, MD, USA
Clin Pharmacokinet (2013) 52:693–703
DOI 10.1007/s40262-013-0065-6
clearance and dosing requirements in children 6 years of
age and older. A novel methodology is reported that
employs virtual adult and pediatric populations and adult
pharmacokinetic data to accurately predict clearance vari-
ability in specific pediatric subpopulations. This approach
has important implications for both clinical trial simula-
tions and sample size determination for pediatric pharma-
cokinetic studies.
1 Introduction
Pharmacokinetics play an integral role in the pediatric drug
development process. The determination of pharmacoki-
netic parameters, particularly clearance, in different age
groups directly informs dosing strategies for subsequent
efficacy trials. Adequate knowledge of clearance in the
targeted clinical population enables investigators to make
rational dosing decisions that have a high probability of
achieving the intended drug exposure. In some cases, the
US Food and Drug Administration (FDA) have accepted
alternative methods, such as allometric scaling, to predict
drug clearance in older children in place of a traditional
dedicated pharmacokinetic study. Allometric scaling is
used to predict drug clearance from one population to
another on the basis of the size differential along with
allometric principles.
Clinical trial simulation (CTS), a powerful tool used in
clinical trial design and to de-risk the phases of pediatric
drug development, requires both an estimate of clearance
as well as an estimate of the expected pharmacokinetic
variability. Sample size justification for pediatric pharma-
cokinetic studies must also be derived using an estimate of
clearance variability [1]. Although the standard deviations
(SD) of individual clearance values for adults are typically
used to provide the estimate, this approach may lead to
inaccurate predictions by not taking into account the more
widespread distribution of factors such as body weight in
the pediatric population. Therefore, the objective of the
current study was to assess the accuracy of allometric
prediction of drug clearance as well as methods for pre-
dicting clearance variability in children 6 years of age and
older using adult pharmacokinetic data.
2 Methods
2.1 Database Development
Adult and pediatric drug clearance information was
extracted from the FDA clinical pharmacology reviews of
pediatric studies conducted in accordance with sec-
tion 505A(k)(1) of the Federal Food, Drug, and Cosmetic
Act (Best Pharmaceuticals for Children Act of 2007
(BPCA)) and section 505B(h)(1) of the Federal Food,
Drug, and Cosmetic Act (Pediatric Research Equity Act of
2007 (PREA)), as amended by FDAAA (Pub. L. No.
110–85) [2]. For information prior to 2007, the FDA
database of clinical pharmacology reviews from 2002
onwards was searched [3]. All eligible studies included
children who were 6 years of age or older. When the age
range included younger children, only those studies where
the lower limit of the range was 2 years old and above were
included. Extracted data for drugs administered intrave-
nously or orally included the following data for adults and
pediatrics: number of participants for which the clearance
was estimated, age range, mean body weight (BW) or body
surface area (BSA), clearance values, and clearance vari-
ability. If relevant information was missing, data were
obtained from FDA drug labels and/or primary literature
sources.
2.2 Body Weight and Body Surface Area Estimates
A set {A} of 1,000 virtual adult individuals (50 % male)
with a body mass index less than 25 kg/m2 and an age
range of 18 to 55 years were created on the basis of the
white American NHANES database [4]. A set {P} of
5,000 virtual children (50 % male) under a uniform age
distribution between 2 and 17 years were created on the
basis of the white American NHANES database [4]. PK-
Sim� (Bayer Technology Services GmbH, Leverkusen,
Germany) was used to generate all virtual populations
using the method as described in Willmann et al. [5].
Body surface area was estimated using Du Bois and Du
Bois [6] for adults and children over 15 kg and the
Haycock et al. method [7] for children at most B15 kg.
2.3 Allometric Clearance Scaling
Clearance in a child (CLchild) was predicted from adult
clearance (CLadult), both given in flow units (e.g., mL/min),
using Eq. (1):
CLchild¼CLadult �BWchild
BWadult
� �3=4
ð1Þ
where BWchild is the body weight of the child and BWadult
is the body weight of the adult.
2.4 Clearance Scaling Methods
2.4.1 Method 1
Allometric clearance scaling was based on only one virtual
adult. CLadult was the geometric median of the observed
694 A. N. Edginton et al.
adult clearance value as reported in the in vivo study or as
calculated from the arithmetic mean as described in the
‘‘Statistical Methods’’ section. BWadult was set at 70 kg. A
subset {P0} of {P} was created that contained only those
virtual children that were within the age range of the
in vivo pediatric study. Therefore, BWchild was the BW of
each virtual child within {P0}. As a result, for each child a
clearance prediction was calculated and subsequently
reported in the same unit and under the same distribution
(i.e., geometric or arithmetic) as the observed pediatric
clearance.
2.4.2 Method 2
Allometric clearance scaling was based on a virtual adult
population {A}. First, one value of CLadult (either in, e.g.,
L/h/kg or L/h/m2) was randomly generated and assigned to
each virtual adult in {A}. The distribution from which the
CLadult was generated followed a log-normal distribution
with a geometric median and standard deviation derived
from the in vivo adult study or derived from the arithmetic
mean and standard deviation as described in the ‘‘Statistical
Methods’’ section. If the observed adult value was not
weight-normalized (e.g., L/h), a body weight of 70 kg was
used to derive a weight-normalized clearance (e.g., L/h/
kg). BWchild was the same as in method 1. Allometric
clearance scaling was performed for each combination of
adult {A} and child {P0}. Matlab (ver. 2011b, Mathworks,
Natick, MA, USA) was used for generating the adult
clearance distribution and for performing all clearance
scaling combinations and summary statistics.
2.5 Statistical Methods
Studies in the database generally reported clearance and
clearance variability in terms of the arithmetic mean ð�xÞ and
standard deviation (r). In this case, geometric median (mg)
values of the observed data were calculated using the same
method as in Johnson et al. [8] where the comparable stan-
dard deviation on the log scale, rg, is calculated using
Eqs. (2) and (3):
rg ¼ exp ð2Þ
where _x is
_x¼ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiln 1þ r=�x
� �2� �r
ð3Þ
The geometric median was calculated using Eqs. (4) and
(5):
mg ¼ expl ð4Þ
where l is
l ¼ lnð�xÞ � ð0:5� r2Þ ð5Þ
These values were assigned to the virtual adult as described
in the ‘‘Clearance Scaling Methods’’ section.
Following either method 1 or method 2 clearance scal-
ing, the array of predicted CLchild values were used to
generate a predicted mean and variability in the original
units assuming the same distribution (arithmetic vs. geo-
metric) as reported in the original pediatric study. The
variability metric used to directly compare observed vs.
predicted variability was the coefficient of variation (CV).
An assessment of the goodness of predictions for
clearance and CV was based on the percent of clearance
and CV values within 2-fold and 30 % of the observed
values. The accuracy and bias of the predictions were
based on calculations of the mean and 95 % confidence
intervals of the prediction error (Eq. 6).
Prediction error ð%Þ ¼ Predicted� Observed
Observed� 100 ð6Þ
Linear regression (SigmaPlot ver. 10.0, Systat Software
Inc., Chicago, IL, USA) was used to examine the
relationship between predicted and observed drug
clearance. When mean differences were compared,
unpaired t tests were used (GraphPad Software Inc., Lo
Jolla, CA, USA).
To provide an indication of the accuracy of prediction at
the top, bottom, and middle of the CV range, the method of
Graves [9] was used. This method partitions the top 16 %
of predicted clearance CV values, the bottom 16 %, and the
middle 32 % of values and assesses the prediction error
and 90 % CI within those ranges.
One goal of the present study was to determine the
accuracy of clearance and clearance variability prediction
as a function of age. As a result, studies (all included
children over the age of 6 years) were grouped according
to the lower age limit of the in vivo study and assessed
separately. Three groups were created: studies that inclu-
ded children under the age of 6 (lowest age was 2), studies
that had a lower age range that was between 6 and 11 years
of age, and studies that had a lower age range that was
greater than or equal to 12 years of age. The prediction
error as a function of the lower age range of the study was
subject to multiple comparison using a one-way analysis of
variance.
3 Results
A total of 38 unique drug products had sufficient data
available for inclusion into the database. One product had
two active components, for a total of 39 unique drugs. Of
these, 8 products were administered intravenously and 30
Prediction of Clearance and Variability in Children 695
were administered orally. The observed clearances and
associated variabilities for adult and pediatric in vivo study
groups are displayed in Table 1. Predicted pediatric
clearances and variabilities are also presented in Table 1
and are based on using either the standard adult of 70 kg
(method 1) or the virtual adult population (method 2) for
allometric clearance scaling.
The prediction accuracy for clearance and CV is pre-
sented in Table 2. Overall, 100 % of all predicted pediatric
mean clearances were within 2-fold of the observed values
and approximately 82 % were within a 30 % prediction
error. There was no significant difference in clearance
prediction accuracy (unpaired t test: t = 0.42, df = 114,
p = 0.81) between the clearance calculation using method
1 vs. method 2. For both method 1 and method 2, using all
data, the 95 % CIs did not include 0 and were biased
towards underprediction of clearance. The correlation of
observed to predicted clearances for intravenously and
orally administered drugs is presented in Figs. 1 and 2,
respectively.
There was a significant increase in the prediction
accuracy of the CV (t = 10.4, df = 114, p \ 0.001) using
clearance calculation method 2 vs. method 1. There was a
major bias towards underprediction of pediatric CV in
method 1 whereas the use of method 2 was precise and not
biased as the 95 % CI includes 0 (Table 2). For method 2,
just over half of the predicted CVs were within 30 % of the
observed CVs. The prediction accuracy of the oral clear-
ance variabilities was greater than that for intravenous
administration. Observed vs. predicted CVs are presented
in Fig. 3.
For method 2, the method of Graves [9] was used to
assess the prediction error for the lowest (e.g., 20 %), mid-
range (e.g., 35 %), and highest (e.g., 70 %) predicted CVs.
These prediction errors and 90 % confidence intervals were
-6.0 (-14.2, 2.2), -3.9 (-10.3, 2.5), and 11.0 (-8.7,
31.3) for the lowest, mid-range, and highest predicted CVs,
respectively. This means, for example, that for a predicted
CV that is low (e.g., 20 %), the in vivo CV will be, on
average, 6.0 % greater than predicted (e.g., 26 %) with a
90 % CI between 17.8 and 34.2 %. Further, for a predicted
CV that is midrange (e.g., 35 %), the in vivo CV will be,
on average, 3.9 % greater than predicted (e.g., 38.9 %)
with a 90 % CI between 32.5 and 45.3 % and, for a pre-
dicted CV that is high (e.g., 70 %), the in vivo CV will be,
on average, 11 % less than predicted (e.g., 59 %) with a
90 % CI between 38.7 and 78.7 %.
Prediction error as a function of the lowest age range of
the study was assessed. For method 2, there was no sig-
nificant difference in prediction error for either clearance or
CV amongst the three age groupings (Fig. 4). Neither
clearance nor CV prediction accuracy was a function of the
age range included in the in vivo studies.
The observed CV between the adult and pediatric study
groups was not significantly different (t = 1.51, df = 114,
p = 0.13), although, on average, the observed pediatric CV
was 32.3 % greater than that from adult studies. The 95 %
CI suggests that pediatric CV was almost always greater
than in adults (Table 2).
4 Discussion
Estimation of drug clearance is an important component of
designing a pediatric drug development program. Drug
exposure is indirectly proportional to clearance (CL) or
oral clearance (CL/F) and is a primary determinant of dose
unless concentration maximums or minimums within a
dosing regimen are important. Prediction accuracy of
allometry using either method 1 or method 2 was high with
100 % of predicted mean clearances within 2-fold of
observed values and approximately 82 % within 30 % of
observed values (Table 2). There is currently no generic
definition of acceptable prediction accuracy for clearance
scaling. Although a 2-fold error is usually accepted by
modelers, this is likely unacceptable for all dosing deci-
sions, especially for drugs with a narrow therapeutic index
or in instances when clearance is highly variable within a
population. Acceptable prediction accuracy will be
dependent on a comparison of the unpredictable inter- and
intraindividual clearance variability with the safe and
effective variability (SEV) which is defined as the maxi-
mum acceptable variability in clearance around a popula-
tion average and is related to the therapeutic index of a
drug [10]. Regardless of the therapeutic index, it is pref-
erable to use an approach found to have the greatest pre-
cision for clearance prediction.
Clearance scaling using allometry was biased towards
underprediction of pediatric mean clearances as the 95 %
CIs did not include zero. The magnitude of the underpre-
diction was around 12 % and this appeared to be driven by
the underprediction of the apparent oral clearances (CL/F)
considering the lack of bias in the prediction of clearance
derived following intravenous administration (Table 2).
The bias towards underprediction of oral CL/F values may
be due to an overestimation of bioavailability (F) in chil-
dren, although a previous review of the absorption
parameters affected by age demonstrated that for children
who were not pre-term or young infants, F remained sim-
ilar across age [11]. The underprediction in oral clearance
may be the result of observed oral clearances being greater
than true values. Owing to ethical and practical consider-
ations, there tends to be sparser sampling in pediatric
studies as compared to adult studies. The derived plasma
concentration vs. time profiles from these studies were
primarily analyzed using non-compartmental techniques.
696 A. N. Edginton et al.
Ta
ble
1O
bse
rved
and
pre
dic
ted
clea
ran
ce(C
L)
and
CL
var
iab
ilit
y
Dru
g—
trad
en
ame
Un
its
Ob
serv
edad
ult
CL
Ped
iatr
icag
ean
d/o
rw
eig
ht
ban
dra
ng
e
Ob
serv
edp
edia
tric
CL
Pre
dic
ted
ped
iatr
icC
L
(met
ho
d1
)
Pre
dic
ted
ped
iatr
icC
L
(met
ho
d2
)
Intr
aven
ou
sad
min
istr
atio
n
Ace
tam
ino
ph
enL
/h/k
g0
.27
(0.0
8)a
12
–1
60
.28
4(0
.04
61
)b0
.27
0(0
.01
48
)0
.27
7(0
.08
42
)
2–
11
0.3
28
(0.0
84
1)b
0.3
34
(0.0
33
6)
0.3
39
(0.1
07
)
Cas
po
fun
gin
acet
ate—
Can
cid
asm
L/m
in1
0.6
(3.8
)1
2–
17
12
.6(5
.5)
9.0
6(1
.52
)9
.42
(3.9
6)
2–
11
6.4
(2.6
)4
.93
(1.5
2)
5.0
9(2
.41
)
Gad
ob
enat
ed
imeg
lum
ine—
Mu
ltiH
ance
L/h
/kg
0.1
58
(0.0
12
)5
–1
5c
0.1
99
(0.0
16
)0
.10
8(0
.01
11
)0
.10
6(0
.01
61
)
Gad
ob
utr
ol—
Gad
avis
tL
/h/k
g0
.11
2
(20
.8%
)de
12
–1
70
.09
(0.0
5,
0.1
0)f
0.1
16
(0.1
04
,0
.12
9)f
0.1
14
(0.0
75
3,
0.1
67
)f
7–
11
0.1
(0.0
5,
0.1
7)f
0.1
33
(0.1
17
,0
.15
1)f
0.1
32
(0.0
85
2,
0.1
94
)f
Lin
ezo
lid
—Z
yv
ox
mL
/min
/
kg
1.7
(34
%)
12
–1
72
.1(5
3%
)1
.67
(5.5
%)
1.7
1(3
5.5
%)
Pem
etre
xed
dis
od
ium
—A
lim
taL
/h/m
22
.75
(1.6
6–
5.0
1)g
4–
18
2.5
9(3
4%
)d2
.46
(6.4
%)
2.6
5(1
2.3
%)
Ro
curo
niu
mb
rom
ide—
Zem
uro
nL
/h/k
g0
.25
(0.0
8)
11
–1
70
.29
2(0
.08
46
)h0
.25
0(0
.01
51
)0
.26
3(0
.08
94
)
2–
11
0.2
93
(0.1
37
)h0
.30
7(0
.03
09
)0
.31
6(0
.10
3)
Zo
led
ron
icac
id—
Zo
met
aL
/h9
.8(2
.16
)3
–1
7i
8.4
5(3
.86
)6
.41
(2.3
2)
6.5
(2.8
3)
Ora
lad
min
istr
atio
n
Ad
efo
vir
—H
epse
raL
/h4
3.4
(11
.5)
12
–1
74
2.1
(10
.1)
38
.1(6
.40
)3
9.1
(12
.7)
7–
11
1.0
(0.2
4)
L/h
/kg
0.7
15
(0.0
47
)L
/h/k
g0
.73
1(0
.20
1)
L/h
/kg
Alm
otr
ipta
n—
Ax
ert
L/h
36
.5(5
.8)
12
–1
74
1.2
(10
.2)
32
.7(5
.47
)3
2.4
(7.4
4)
Am
ox
icil
lin
/cla
vu
lan
ate
po
tass
ium
—
Au
gm
enti
nX
R
L/h
AM
OX
:2
9.9
(7.6
)
7–
15
(wei
gh
ing
C4
0k
g)
AM
OX
:3
4.6
(9.3
4)
AM
OX
:2
4.6
(4.1
4)
AM
OX
:2
4.9
(7.5
1)
CL
AV
:2
3.6
(5.3
3)
CL
AV
:3
9.3
(11
.8)
CL
AV
:2
2.0
(3.7
)C
LA
V:
21
.9(6
.41
)
An
astr
ozo
le—
Ari
mid
exL
/h1
.24
(0.2
4)j
11
–1
81
.54
(37
%)
1.0
5(1
8.4
%)
1.0
6(2
7%
)
Ari
pip
razo
le—
Ab
ilif
yL
/h3
.45
(1.0
7)k
10
–1
73
.44
(45
.6%
)(3
.26
,
3.6
3)l
2.6
3(2
0.1
%)
(2.6
0,
2.6
6)
2.7
4(3
7.2
%)
(2.6
8,
2.7
9)
Azi
thro
my
cin
—Z
max
(ER
/SR
)L
/h1
00
(33
.3)
13
–1
69
3.9
(41
.3)
87
.4(1
3.8
)9
0.9
(34
.0)
9–
12
66
.4(1
7.4
)m6
6.1
(12
.9)
68
.9(2
7.5
)
Cel
eco
xib
—C
eleb
rex
L/h
44
.9(4
4.4
%)
12
–1
64
6(4
2.7
%)
36
.7(1
6.8
%)
40
.0(4
9.3
%)
6–
11
33
.4(3
5.3
%)
23
.8(2
2.2
%)
26
.0(5
1.5
%)
Did
ano
sin
e—V
idex
EC
(DR
)L
/h1
74
.5(6
9.7
)[
60
kg
n1
96
(55
.8)
16
5.5
(18
.2)
17
0.7
(69
.7)
25
–6
0k
gn
11
6.2
(38
.6)
10
8.1
(20
.5)
11
7.3
(51
.2)
Em
tric
itab
ine—
Em
triv
am
L/m
in3
39
(20
%)
13
–1
76
.4(4
5%
)m
L/m
in/k
g4
.9(5
.1%
)m
L/m
in/k
g4
.9(2
0.2
%)
mL
/min
/kg
7–
12
8.4
(54
%)
mL
/min
/kg
5.5
8(7
.3%
)
mL
/min
/kg
5.5
3(2
1.8
%)
mL
/min
/kg
Prediction of Clearance and Variability in Children 697
Ta
ble
1co
nti
nu
ed
Dru
g—
trad
en
ame
Un
its
Ob
serv
edad
ult
CL
Ped
iatr
icag
ean
d/o
rw
eig
ht
ban
dra
ng
e
Ob
serv
edp
edia
tric
CL
Pre
dic
ted
ped
iatr
icC
L
(met
ho
d1
)
Pre
dic
ted
ped
iatr
icC
L
(met
ho
d2
)
Ep
lere
no
ne—
Insp
raL
/h/
70
kg
7.8
3(3
.76
)o1
2–
16
6.6
9(7
5%
)dL
/h6
.24
(16
.5%
)L
/h6
.17
(50
.8%
)L
/h
6–
11
4.9
(11
0.7
%)d
L/h
4.0
(21
.8%
)L
/h4
.0(5
1.2
%)
L/h
Esc
ital
op
ram
—L
exap
roL
/h3
0.3
(13
.4)
12
–1
73
6.2
(14
.3)
25
.1(4
.21
)2
7.0
(12
.57
)
Gli
mep
irid
e—A
mar
yl
L/h
3.1
7(0
.96
6)
10
–1
72
.95
(1.7
7)
2.5
8(0
.52
)2
.63
(0.9
88
)
Lefl
un
om
ide—
Ara
va
mL
/h2
3.7
(7.8
1)
[4
0–
75
kg
p2
6(1
6)
19
.1(2
.4)
19
.6(6
.8)
[2
0–
40
kg
p1
8(9
.5)
11
.7(1
.8)
12
.0(4
.5)
Lev
etir
acet
am—
Kep
pra
mL
/min
/
kg
0.9
(0.0
8)q
7–
16
2.7
4(1
.67
,4
.13
)r
(23
.3%
)L
/h
2.8
2(2
.7,2
.94
)
(26
.6%
)L
/h
2.7
8(2
.65
,2
.91
)
(28
.2%
)L
/h
Lo
pin
avir
/rit
on
avir
—K
alet
ras
L/h
5.9
8(5
.75
)6
–1
13
.49
(3.1
2)
2.5
(0.5
6)
3.2
7(3
.39
)
12
–1
74
.31
(2.3
1)
3.9
1(0
.65
4)
5.2
4(4
.81
)
Mel
ox
icam
—M
ob
icm
L/m
in8
.8(2
9%
)7
–1
63
.59
(45
.3%
)dt
6.1
(27
.3%
)5
.97
(40
.0%
)
Nel
fin
avir
—V
irac
ept
L/h
/kg
0.7
2(0
.43
)3
–1
11
.32
(0.6
5)
0.7
84
(0.0
71
7)
0.9
09
(0.5
44
)
Ose
ltam
ivir
—T
amifl
um
L/m
in/
kg
6.6
(19
.8%
)1
3–
18
5.3
(2.0
)6
.7(0
.34
)6
.7(1
.34
)
9–
12
9.4
(1.3
)7
.5(0
.40
8)
7.5
(1.5
5)
5–
81
1.1
(1.6
)8
.3(0
.49
)8
.3(1
.65
)
Pal
iper
ido
ne—
Inv
ega
(ER
)m
L/m
in2
68
(44
.4%
)1
0–
17
20
9(5
6.3
%)
20
8.7
(20
.1%
)2
33
.7(4
9.1
%)
Pan
top
razo
leso
diu
m—
Pro
ton
ix(D
R)
L/h
6.9
3(1
01
%)d
u1
2–
16
6.8
(57
.9%
)dv
6.1
2(1
6.7
%)
6.1
8(9
8%
)
6–
11
6.6
(66
.3%
)dv
3.9
2(2
1.8
%)
3.8
6(9
9.3
%)
Qu
etia
pin
efu
mar
ate—
Ser
oq
uel
L/h
10
5(2
2)w
13
–1
71
15
.8(5
5.4
%)d
94
.4(1
4.5
%)
92
.6(2
4.7
%)
10
–1
28
0.7
(24
.8%
)d7
3.8
(12
.9%
)7
2.6
(21
.2%
)
Rab
epra
zole
sod
ium
—A
cip
hex
(DR
)m
L/m
in/
kg
9.5
6(5
.86
)1
2–
16
10
.1(6
.91
)8
.51
(0.4
65
)9
.52
(5.6
1)
Ro
feco
xib
—V
iox
xm
L/m
in6
5(2
0)
12
–1
78
7(2
1)
56
.3(9
.4)
58
.7(2
0.2
)
6–
11
52
(13
)3
6(8
.0)
36
.3(1
4)
Ro
sig
lita
zon
e—A
van
dia
L/h
3.0
3(0
.87
)1
0–
17
3.1
5(9
5%
CI
2.1
,4
.87
)x2
.48
(20
.1%
)(9
5%
CI
2.4
,2
.58
)
2.5
1(3
5.7
%)
(2.3
3,
2.6
9)
Ro
suv
asta
tin
—C
rest
or
L/h
/kg
3.1
5
(29
.8%
)dy
10
–1
72
21
.6(3
2.9
%)y
L/h
18
7.8
(20
.1%
)L
/h1
92
.4(3
4.4
%)
L/h
Sir
oli
mu
s—R
apam
un
em
L/h
/kg
13
9(6
3)
12
–1
81
36
(57
)1
34
(7.3
)1
42
(60
.7)
6–
11
21
4(1
29
)1
56
(11
.2)
16
3(6
7.6
)
Tam
sulo
sin
hy
dro
chlo
rid
e—F
lom
axL
/h2
.01
(0.9
5)z
2–
16
1.0
8(4
8.1
%)d
1.0
5(4
1.5
%)
1.0
4(6
5.6
%)
Ten
ofo
vir
dis
op
rox
ilfu
mar
ate—
Vir
ead
L/h
91
.0(2
8)*
12
–1
78
8.4
(31
.8)
78
.7(1
3.2
)8
0.6
(27
.1)
Zid
ov
ud
ine—
Ret
rov
irL
/h/k
g2
.28
(0.9
)#2
–1
32
.68
(90
%C
I1
.97
,
3.6
4)#
2.6
6(1
1.7
%)
2.7
8(4
1.7
%)
(90
%C
I2
.45
,2
.87
)(9
0%
CI
2.0
,3
.56
)
698 A. N. Edginton et al.
Ta
ble
1co
nti
nu
ed
Dru
g—
trad
en
ame
Un
its
Ob
serv
edad
ult
CL
Ped
iatr
icag
ean
d/o
rw
eig
ht
ban
dra
ng
e
Ob
serv
edp
edia
tric
CL
Pre
dic
ted
ped
iatr
icC
L
(met
ho
d1
)
Pre
dic
ted
ped
iatr
icC
L
(met
ho
d2
)
Zip
rasi
do
ne—
Geo
do
nL
/h4
2.6
(21
%)$
11
–1
61
1.5
(21
%)
mL
/min
/kg
10
.5(6
.0%
)m
L/m
in/k
g1
0.6
(22
.3%
)m
L/m
in/k
g
7–
91
3.1
(27
%)
mL
/min
/kg
12
.2(5
.4%
)m
L/m
in/k
g1
2.1
(22
.2%
)m
L/m
in/k
g
Pre
dic
tio
ns
eith
erem
plo
yed
on
e7
0k
gad
ult
(met
ho
d1
)o
ra
vir
tual
po
pu
lati
on
(n=
1,0
00
)o
fad
ult
s(m
eth
od
2).
Val
ues
are
exp
ress
edas
arit
hm
etic
mea
n(v
aria
bil
ity
)u
nle
sssp
ecifi
ed
oth
erw
ise.
Var
iab
ilit
yis
pre
sen
ted
asst
and
ard
dev
iati
on
or
%co
effi
cien
to
fv
aria
tio
n(C
V%
)u
nle
sssp
ecifi
edo
ther
wis
e.P
red
icte
dp
edia
tric
val
ues
are
inth
esa
me
un
its
and
foll
ow
the
sam
e
dis
trib
uti
on
(no
rmal
vs.
log
-no
rmal
)as
the
ob
serv
edp
edia
tric
val
ues
aT
aken
fro
mJo
nes
20
11
[16
]b
Cle
aran
cefo
llo
win
gin
trav
eno
us
adm
inis
trat
ion
of
15
mg
/kg
acet
amin
op
hen
c1
of
the
25
ped
iatr
icp
atie
nts
wer
e3
.2y
ears
of
age;
all
oth
ers
wer
eb
etw
een
the
ages
of
5an
d1
5.6
yea
rsd
Geo
met
ric
mea
nan
dco
effi
cien
to
fv
aria
tio
ne
Val
ues
fro
mn
on
-eld
erly
men
fG
eom
etri
cm
ean
and
95
%co
nfi
den
cein
terv
alg
Geo
met
ric
mea
nan
dra
ng
e;u
sed
a1
0%
geo
met
ric
coef
fici
ent
of
var
iati
on
wh
ich
lead
sto
asi
mil
arad
ult
ran
ge
hV
aria
bil
ity
calc
ula
ted
fro
mcl
eara
nce
vs.
age
gra
ph
incl
inic
alp
har
mac
olo
gy
rev
iew
[2]
iN
oef
fect
of
age
on
syst
emic
exp
osu
rej
Tak
enfr
om
Yu
etal
.[1
7]
kT
aken
fro
mst
ud
y1
of
Mal
lik
aarj
un
etal
.[1
8]
lA
rith
met
icm
ean
and
95
%co
nfi
den
cein
terv
al,
con
fid
ence
inte
rval
sta
ken
fro
m1
,00
0b
oo
tstr
aps
for
bo
tho
bse
rved
and
pre
dic
ted
mA
ssu
min
ga
do
seo
f2
gZ
max
per
chil
do
fag
e9
–1
2n
Vir
tual
po
pu
lati
on
for[
60
kg
incl
ud
edal
lch
ild
ren
inth
ed
atab
ase
wit
ha
wei
gh
tC
60
kg
.V
irtu
alp
op
ula
tio
nfo
r2
5–
60
kg
incl
ud
edal
lch
ild
ren
wit
ha
low
erw
eig
ht
cuto
ffo
f2
5k
gan
dan
up
per
wei
gh
tcu
toff
of
60
kg
oT
aken
fro
mR
avis
etal
.[1
9]
fro
mn
orm
als
that
wer
em
atch
edto
mil
dre
nal
imp
airm
ent
pat
ien
tsg
iven
asi
ng
led
ose
pA
ge
ran
ge
inst
ud
yw
as3
–1
7y
ears
;th
eref
ore
,th
ev
irtu
alp
op
ula
tio
nin
clu
ded
ages
3–
17
and
was
spli
tan
db
ou
nd
edb
yth
ew
eig
ht
cuto
ffs
qT
aken
fro
mS
tro
lin
etal
.[2
0]
rM
edia
nan
d9
0%
con
fid
ence
inte
rval
sF
or
lop
inav
iro
nly
tT
aken
fro
mB
urg
os-
Var
gas
etal
.[2
1]
uC
alcu
late
dfr
om
area
un
der
the
pla
sma
con
cen
trat
ion
vs.
tim
e(A
UC
)cu
rves
foll
ow
ing
ped
iatr
icg
ran
ule
adm
inis
trat
ion
vC
alcu
late
dfr
om
AU
Cs
inF
ig.
5in
the
clin
ical
ph
arm
aco
log
yre
vie
w[2
]w
Mea
no
ral
clea
ran
ce(C
L/F
)v
alu
efr
om
hea
lth
yco
ntr
ol
stu
die
sas
list
edin
DeV
ane
and
Nem
ero
ff[2
2].
Val
ue
isfr
om
the
stu
dy
of
Th
yru
met
al.
[23
]x
Co
effi
cien
to
fv
aria
tio
no
f3
4%
was
use
dy
Cal
cula
ted
fro
mA
UC
so
fa
80
-mg
ora
ld
ose
;ad
ult
dat
afr
om
Mar
tin
etal
.[2
4]
zC
alcu
late
dfr
om
AU
Cin
the
fast
edst
ate
*C
alcu
late
dfr
om
AU
Cfo
llo
win
gta
ble
tad
min
istr
atio
n#
Ad
ult
val
ue
fro
mh
ealt
hy
vo
lun
teer
sin
Tab
ure
tet
al.
[25]
and
ped
iatr
icv
alu
esca
lcu
late
dfr
om
clin
ical
ph
arm
aco
log
yre
vie
w[2
]an
dB
erg
sho
eff
etal
.[2
6]
usi
ng
TID
dat
a.A
coef
fici
ent
of
var
iati
on
of
40
%u
sed
for
the
ob
serv
edp
edia
tric
CV
$A
du
ltv
alu
esta
ken
fro
mW
iln
eret
al.
[27
]
Prediction of Clearance and Variability in Children 699
As sampling number in the absorption phase declines, there
is a lower probability of sampling at the true Cmax. The
result of missing Cmax in a non-compartmental analysis is
an underprediction of area under the concentration–time
curve and a resulting overestimation of oral clearance. This
overestimation would be more likely in pediatric studies.
Our method does not address this because allometric
scaling is based on an observed adult oral clearance value
that has a greater probability of being close the true value.
The accuracy of clearance prediction was not a function of
the lower age range (minimum age of 2 years) of the
observed studies. It has been previously demonstrated that
clearance prediction using allometry is reasonable for chil-
dren once the ontogeny of the enzymatic [e.g., cytochrome
P450 (CYP)] or physiological (e.g., glomerular filtration)
processes responsible for clearance reaches adult levels [12,
13]. These processes tend to be fully mature by the age of
5 years, except for CYP1A2 activity which has a relatively
CLobs
0.01 0.1 1 10 100
CL
pred
_met
hod_
2
0.01
0.1
1
10
100
Slope = 1.09; r2=0.99
Fig. 1 Correlation between predicted (using method 2) (CLpred_-
method_2) and observed (CLobs) pediatric clearances for drug products
administered via the intravenous route. Clearance units are in the
original form as presented in Table 1. The solid line is the regression
line and dashed lines depict the 2-fold error lines
CVobs
0 20 40 60 80 100 120
CV
pred
_met
hod_
2
0
20
40
60
80
100
120
Fig. 3 Correlation between predicted (using method 2) (CVpred_-
method_2) and observed (CVobs) pediatric clearance variability, as
defined by the coefficient of variation (CV), for drug products
administered via the intravenous (open circles) or oral (squares)
route. The solid line is the line of unity and dashed lines depict the
2-fold error lines
CLobs
0.1 1 10 100 1000
CL
pred
_met
hod_
2
0.1
1
10
100
1000
Slope = 0.99; r2=0.98
Fig. 2 Correlation between predicted (using method 2) (CLpred_-
method_2) and observed (CLobs) pediatric clearances for drug products
administered via the oral route. Clearance units are in the original
form as presented in Table 1. The solid line is the regression line and
dashed lines depict the 2-fold error lines
Table 2 Comparison of mean clearances (CL or CL/F) and coefficients of variation (CV) for intravenously (IV) administered and orally (PO)
administered drug products
xa ya Mean percent errorb (%)
(95 % CI)—all data
% within
2-foldc—all data
(%)
% ± 30 %—
all data (%)
Mean percent error (%)
(95 % CI)—IV only
Mean percent error (%)
(95 % CI)—PO only
CLpred_method_1 CLped_obs -13.6 (-18.7, -8.60) 100 81 -4.96 (-16.1, 6.22) -15.9 (-21.3, -10.4)
CLpred_method_2 CLped_obs -10.7 (-15.7, -5.73) 100 83 -3.09 (-13.6, 7.46) -12.7 (-18.3, -7.14)
CVpred_method_1 CVped_obs -58.4 (-64.9, -51.8) 29 12 -56.3 (-75.6, -37.0) -58.9 (-65.6, -52.1)
CVpred_method_2 CVped_obs 3.36 (-6.18, 12.9) 91 52 5.95 (-18.3, 30.2) 2.69 (-7.70, 13.1)
CVped_obs CVadult_obs 32.3 (16.4, 48.2) 86 52 36.9 (-5.89, 79.8) 29.4 (12.6, 46.2)
a CL for IV and CL/F for oral administrationb % error = (CLx or CVx - CLy or CVy)/CLy or CVy 9 100 and 95 % CI = 95 % confidence intervalc Percent within the range of percent error of -50 to 100 %
700 A. N. Edginton et al.
long maturation half-life. Mahmood [14] observed that a
power function of 3/4 was accurate for clearance scaling over
the age of 5 years of age and unreliable in the less than
1-year-old category. Although all studies included children
who were 6 years of age and older, the lack of age depen-
dence (i.e., lower age range of study) of prediction accuracy
in the present study was either the result of lower numbers of
children less than 6 years of age within some of the studies,
or the children at the lower end of the age range had fully
matured clearance pathways. Assessment of the accuracy of
predictions for younger children will require individual
clearance data as opposed to a mean clearance within an age
range, as was used in this study. This would allow for char-
acterization of clearance process maturation that differs from
the effects of size for which allometry accounts [15]. This
uncertainty due to a lack of individual data for the youngest
individuals is sufficient enough to disallow interpolation/
extrapolation of our results to children under the age of
6 years. For studies using only children over the age of
6 years, 100 % of mean clearance values were within 2-fold
of observed values and 82 % were within 30 % error. The
greatest deviation in this group was a CL/F prediction error
of 66 %. This was for meloxicam, a nonsteroidal anti-
inflammatory drug (NSAID) that is eliminated primarily by
CYP-mediated hepatic metabolism. Mean clearance was
predicted to be 5.97 mL/min but was observed to be
3.59 mL/min (Table 1).
In order to also predict variability, a virtual pediatric
population was required and, when coupled to a
randomization of adult clearance and a virtual adult pop-
ulation ({A}; n = 1,000), clearance variability was pre-
dicted with good accuracy and no bias (Table 2). Method 2
assumes that the same degree of variation in the bio-
chemical and physiological parameters that drive clearance
variability in adults is equivalent to that in children (i.e.,
clearance randomization in adults) with a scalar for growth
(i.e., body weight ratios to the power of 3/4). This is not
mechanistic but it does produce reasonably accurate mea-
sures of pediatric clearance variability.
For the most part, observed clearance variability was
greater in the pediatric populations investigated in com-
parison to their adult counterparts. On average, observed
pediatric clearance variability was 32 % greater than the
observed adult variability with little probability of being
less than adult variability (Table 2). Therefore, it should be
an expectation that exposure to a given dose will vary to a
greater degree in children than it will in the adult popula-
tion. This greater clearance variability was not adequately
captured when only one adult was used in the allometric
clearance scaling exercise (method 1) such that variability
in clearance was due only to the BW differences amongst
the virtual pediatric population. This led to gross under-
estimation of pediatric clearance variability (Table 2).
However, by including a random selection of clearances
based on the mean and standard deviation of clearances in
adults along with varying adult BWs (method 2), the
clearance variability in pediatric populations was well
predicted. Using method 2, just over half of the predicted
CVs were within 30 % of the observed CV and 91 % were
less than 2-fold off. On the basis of the 95 % CIs, there was
no bias towards over or underestimation. Using a mecha-
nistic approach to clearance scaling based on the SimCyp�
software, Johnson et al. [8] demonstrated that integration of
anatomical (e.g., BW and height), physiological (e.g., liver
blood flow), and biochemical (e.g., Vmax as a function of
age) information could be used to generate virtual pediatric
populations with clearance variability that was reasonable
(63 % of all studies from 11 drug products were within
2-fold of the observed variability). Similar to our case,
there appeared to be no bias towards over- or underpre-
diction of clearance variability. Clearly, when no adult
pharmacokinetic information is available, a mechanistic
means of predicting clearance and clearance variability is
the only biologically rational method. However, when
adult information is available, clearance variability is well
predicted using allometric clearance scaling inclusive of
both an adult and pediatric virtual population.
Drug clearance is an important component of dose
selection for pediatric clinical trials as doses are chosen to
target a specific exposure. Traditionally, pediatric clear-
ance is obtained from dedicated pharmacokinetic studies
prior to the initiation of efficacy trials. Alternative
2 to<6 6 to 11 12 to <18 2 to <6 6 to 11 12 to <18
% e
rror
in p
redi
ctio
n
-80
-60
-40
-20
0
20
40
60
80
100ClearanceCoefficient of variation (%)
Fig. 4 Clearance (CL) and clearance variability (coefficient of
variation, CV) prediction accuracy as a function of the lowest age
range included in the in vivo studies, between 2 and \6 years
(n = 13), between 6 and 11 years (n = 26), and between 12 and
\18 years (n = 19). The line within the box marks the median (solid
line) or mean (dotted line) with box boundaries as the 25th and 75th
percentiles. Error bars above and below the box indicate the 90th and
10th percentiles and dots represent outliers. On the basis of an
analysis of variance (ANOVA), there was no significant difference in
the prediction error for CL (F = 0.36, df = 57, p = 0.70) or CV
(F = 0.07, df = 57, p = 0.93) amongst the three groups of studies
Prediction of Clearance and Variability in Children 701
approaches that can predict pediatric drug clearance
potentially offer the ability to streamline the drug devel-
opment process and avoid unnecessary pediatric pharma-
cokinetic studies. The current study shows that allometry is
a reasonable method for predicting clearance in children
older than 6 years of age. However, quantitatively identi-
fying optimal dosing schedule, sample size, and the prob-
ability of a trials success through CTS requires knowledge
of the expected pharmacokinetic variability. When linked
to a response model, CTS is a powerful tool used to inform
clinical trial design and de-risk the clinical phases of drug
development. Variability in clearance and distribution
amongst virtual patients defines the probability of reaching
an exposure endpoint that is linked to response. The use of
method 2 in the current study adequately predicts clearance
variability and can be applied in CTS.
There are limitations to the proposed workflow for
clearance and clearance variability prediction. This study did
not match, outside of the age range, observed study partici-
pants to virtual study participants owing primarily to a lack of
information in most cases on the weight and/or heights of the
individual in vivo study participants. It was assumed that the
adult and pediatric patients in the in vivo studies were a
random selection of the white American population (50 %
male) as defined by NHANES without consideration of the
actual race proportions within the in vivo studies. Mixed
races would alter the body weight and heights within the
populations. On the basis of the NHANES database, for male
and female children between the ages of 2 and about
12 years, mean weights in the black, Mexican–American,
and white databases are similar with comparable 5th per-
centiles. The 95th percentiles, however, are greatest in the
black population and lowest in the white population. For
adult males from 18 years of age onwards, the three races
follow similar mean weights and 5th percentiles, but the 95th
percentiles are highest in the black population and lowest in
the Mexican–American population. For adult females from
18 years of age onwards, the mean weights are higher in the
black population with greater variability than in the Mexi-
can–American and white populations, which are similar.
Construction of a mixed race population would impart
greater adult and pediatric variability with a tendency
towards increased weights. This would increase both the
mean predicted clearances as well as the predicted CVs with
greater increases being seen with increased inclusion of other
races; however, on the basis of similar weight means of the
races, this increase would be expected to be small. Without
knowledge of the race mix within the studies, however,
inclusion of different races was not possible. We did not
assess the accuracy of clearance prediction when the average
body weight of the in vivo pediatric study was used, which
would produce one clearance value for the entire age range.
This method has been used previously [12, 14]. Further, this
method has not been evaluated for non-oral extravascular
administration such as intramuscular or subcutaneous
administration. This was due to a relatively small number of
studies using these administration routes in children and
greater uncertainty in the understanding of how age-depen-
dent physiology (e.g., blood flow, lymph flow) affects bio-
availability in these situations. Another limitation is in the
use of clearance values for adults that were, for the most part,
generated in healthy volunteers or in adults with disease
states not necessarily the same as that of the pediatric pop-
ulation. Pediatric clearance values are derived only from
patients. Therefore, if the disease has an effect on clearance
in either the adult or pediatric population, or the same disease
is not being compared, this may have an effect on the accu-
racy of prediction.
In March 2012, the FDA’s Pharmaceutical Science and
Clinical Pharmacology Advisory Committee voted 12 to 1
in support of the following question: Can dose(s) for the
adolescent (12 years and older) population be derived
using adult data without the need for a dedicated pharma-
cokinetic (PK) study? The meeting minutes state:
‘‘Some committee members were in favor of the
collection of sparse PK sampling from the adolescent
population during safety/efficacy studies if there is no
dedicated PK study. In addition, some members
opined that allometric scaling to determine doses for
the adolescent population is a reasonable approach,
but drug-specific PK or PD properties and the thera-
peutic benefit versus risks would need to be taken
into consideration’’.
Barring therapeutic response/adverse events and focus-
ing on pharmacokinetics, dose scaling through clearance
scaling is appropriate for the over 12 age group and, as our
study has demonstrated, may be appropriate in some cir-
cumstances for the younger age group of 6–11 years. It is
prudent, however, to confirm anticipated exposure through
sparse sampling during safety/efficacy trials as suggested
by some advisory committee members.
5 Conclusion
In summary, allometric scaling may be a useful tool during
pediatric drug development to predict drug clearance and
dosing requirements in children 6 years of age or older. In
addition, we report a novel methodology employing virtual
adult and pediatric populations and adult pharmacokinetic
data to accurately predict clearance variability in specific
pediatric subpopulations. This approach has important
implications for both clinical trial simulations and sample
size determination for pediatric pharmacokinetic studies.
702 A. N. Edginton et al.
Acknowledgments The authors report no conflicts of interest. No
sources of funding were used for the conduct of this study. The
opinions and findings expressed in this paper are those of the authors
and do not necessarily represent those of the US Food and Drug
Administration.
References
1. Wang Y, Jadhav PR, Lala M, Gobburu JV. Clarification on
precision criteria to derive sample size when designing pediatric
pharmacokinetic studies. J Clin Pharmacol. 2012;52:1601–6.
2. US FDA. Medical, statistical, and clinical pharmacology reviews
of pediatric studies conducted under Section 505A and 505B of
the Federal Food, Drug, and Cosmetic Act (the Act), as amended
by the FDA Amendments Act of 2007 (FDAAA). http://www.
fda.gov/Drugs/DevelopmentApprovalProcess/
DevelopmentResources/ucm049872.htm. Accessed 1 Sept 2012.
3. US FDA. Summaries of medical and clinical pharmacology
reviews—summaries of medical and clinical pharmacology
reviews as of 15 Jan 2008. http://www.fda.gov/Drugs/Develop
mentApprovalProcess/DevelopmentResources/ucm161894.htm.
Accessed 1 Sept 2012.
4. Third National Health and Nutrition Examination Survey,
(NHANES III). 1997. National Center for Health Statistics
Hyattsville, MD 20782. http://www.cdc.gov/nchs/nhanes.htm.
5. Willmann S, Hohn K, Edginton A, Sevestre M, Solodenko J,
Weiss W, et al. Development of a physiology-based whole-body
population model for assessing the influence of individual vari-
ability on the pharmacokinetics of drugs. J Pharmacokinet
Pharmacodyn. 2007;34:401–31.
6. Du Bois D, Du Bois EF. A formula to estimate the approximate
surface area if height and weight be known. 1916. Nutrition.
1989;5:303–11.
7. Haycock GB, Schwartz GJ, Wisotsky DH. Geometric method for
measuring body surface area: a height-weight formula validated
in infants, children, and adults. J Pediatr. 1978;93:62–6.
8. Johnson TN, Rostami-Hodjegan A, Tucker GT. Prediction of the
clearance of eleven drugs and associated variability in neonates,
infants and children. Clin Pharmacokinet. 2006;45:931–56.
9. Graves RE. Accuracy of regression equation prediction across the
range of estimated premorbid IQ. J Clin Exp Neuropsychol.
2000;22:316–24.
10. Holford NH, Buclin T. Safe and effective variability—a criterion
for dose individualization. Ther Drug Monit. 2012;34:565–8.
11. Edginton AN, Fotaki N. Oral drug absorption in pediatric popu-
lations. Oral drug absorption: prediction and assessment, 2nd ed.
New York: Informa Healthcare; 2010.
12. Edginton AN, Willmann S. Physiology-based versus allometric
scaling of clearance in children; an eliminating process based
comparison. Paed Perinat Drug Ther. 2006;7:146–53.
13. Anderson BJ. Pediatric models for adult target-controlled infu-
sion pumps. Paediatr Anaesth. 2010;20:223–32.
14. Mahmood I. Prediction of drug clearance in children: impact of
allometric exponents, body weight, and age. Ther Drug Monit.
2007;29:271–8.
15. Anderson BJ, Holford NH. Tips and traps analyzing pediatric PK
data. Paediatr Anaesth. 2011;21:222–37.
16. Jones VM. Acetaminophen injection: a review of clinical infor-
mation. J Pain Palliat Care Pharmacother. 2011;25:340–9.
17. Yu J, He J, Zhang Y, Qin F, Xiong Z, Li F. An ultraperformance
liquid chromatography-tandem mass spectrometry method for
determination of anastrozole in human plasma and its application
to a pharmacokinetic study. Biomed Chromatogr. 2011;25:511–6.
18. Mallikaarjun S, Salazar DE, Bramer SL. Pharmacokinetics, tol-
erability, and safety of aripiprazole following multiple oral dos-
ing in normal healthy volunteers. J Clin Pharmacol.
2004;44:179–87.
19. Ravis WR, Reid S, Sica DA, Tolbert DS. Pharmacokinetics of
eplerenone after single and multiple dosing in subjects with and
without renal impairment. J Clin Pharmacol. 2005;45:810–21.
20. Strolin BM, Whomsley R, Nicolas JM, Young C, Baltes E.
Pharmacokinetics and metabolism of 14C-levetiracetam, a new
antiepileptic agent, in healthy volunteers. Eur J Clin Pharmacol.
2003;59:621–30.
21. Burgos-Vargas R, Foeldvari I, Thon A, Linke R, Tuerck D.
Pharmacokinetics of meloxicam in patients with juvenile rheu-
matoid arthritis. J Clin Pharmacol. 2004;44:866–72.
22. DeVane CL, Nemeroff CB. Clinical pharmacokinetics of que-
tiapine: an atypical antipsychotic. Clin Pharmacokinet.
2001;40:509–22.
23. Thyrum PT, Wong YW, Yeh C. Single-dose pharmacokinetics of
quetiapine in subjects with renal or hepatic impairment. Prog
Neuropsychopharmacol Biol Psychiatry. 2000;24:521–33.
24. Martin PD, Warwick MJ, Dane AL, Cantarini MV. A double-
blind, randomized, incomplete crossover trial to assess the dose
proportionality of rosuvastatin in healthy volunteers. Clin Ther.
2003;25:2215–24.
25. Taburet AM, Naveau S, Zorza G, Colin JN, Delfraissy JF, Chaput
JC, et al. Pharmacokinetics of zidovudine in patients with liver
cirrhosis. Clin Pharmacol Ther. 1990;47:731–9.
26. Bergshoeff AS, Fraaij PL, Verweij C, van Rossum AM, Verweel
G, Hartwig NG, et al. Plasma levels of zidovudine twice daily
compared with three times daily in six HIV-1-infected children.
J Antimicrob Chemother. 2004;54:1152–4.
27. Wilner KD, Hansen RA, Folger CJ, Geoffroy P. The pharmaco-
kinetics of ziprasidone in healthy volunteers treated with cimet-
idine or antacid. Br J Clin Pharmacol. 2000;49(Suppl 1):
57S–60S.
Prediction of Clearance and Variability in Children 703
