Pharmaceutica Acta Helvetiae 73 1998.

135144

Effect of variability in hepatic clearance on the bioequivalence

parameters of a drug and its metabolite: simulations using a
pharmacostatistical model
)
Sara E. Rosenbaum
Department of Applied Pharmaceutical Science, College of Pharmacy Uniersity of Rhode Island, Rhode Island, USA
Received 16 October 1997; revised 4 February 1998; accepted 4 February 1998

Abstract

A semi-physiological pharmacostatistical model was developed and used to study the manner in which intra-individual variability in
hepatic clearance is transmitted to the area-under the plasma concentration-time curve from zero to infinity AUC. of a drug and its
metabolite. In order to clarify the effects, the model contained no other sources of variability. As the drugs hepatic extraction ratio
increased, the coefficient of variation CV. of the AUC of the drug increased, whereas that of the metabolite decreased. The AUC of the
metabolite increased as the drugs non-hepatic clearance increased. In contrast, at low extraction ratios, the CV of the AUC of the drug
decreased as the drugs non-hepatic clearance increased. At high extraction ratios, the CV of the AUC of the drug was insensitive to the
contributions of other forms of clearance. The results suggest that under certain circumstances, smaller samples sizes could be used in
bioequivalence studies for highly variable drugs if the test were based on the metabolite rather than the parent drug. q 1998 Elsevier
Science B.V. All rights reserved.

Keywords: Bioequivalence; Bioavailability; Pharmacokinetics; Population pharmacokinetics

1. Introduction the bioequivalence parameters AUC and Cmax . of the

generic and those of the innovator product must fall within
The evaluation of bioequivalence for the approval of
the limits of 80 and 125%. The economic advantages of
generic versions of brand-name products is based on the
the lower priced generics are estimated to save consumers
demonstration of equivalency with respect to the rate and
and managed health care organizations billions of dollars
extent of absorption of the drug. The extent of drug
each year.
absorption is usually assessed using the area under the
The sample size used by drug sponsors is based on
plasma concentration time curve AUC., while the rate of
power analysis and must be sufficiently large to permit the
absorption is usually assessed using the peak plasma con-
demonstration of bioequivalency if indeed it exists. Intra-
centration Cmax . and sometimes the time of peak Tmax ..
individual variability of the AUC is a critical factor in
Bioequivalence is established if the sponsor can demon-
determining sample size Diletti et al., 1991; Hauschke et
strate that the bioequivalence parameters of the drug from
al., 1992.. In the case of highly variable drugs, usually
the dosage form are similar to those of the innovator
defined as those which exhibit intra-subject variability of
product. In the U.S., the 90% confidence interval of the
bioequivalence parameters of greater than 2530%, the
difference between the logarithmic transformed mean of
demonstration of bioequivalency can be very difficult and
require the use of excessively large cohorts of subjects
)
Corresponding author. Tel.: q1 401 874 5021; fax: q1 401 874 Boddy et al., 1995.. In many cases large intra-subject
2181; e-mail: sararos@uriacc.uri.edu variability in the AUC occurs for drugs subject to exten-

0031-6865r98r$19.00 q 1998 Elsevier Science B.V. All rights reserved.

PII S 0 0 3 1 - 6 8 6 5 9 8 . 0 0 0 0 8 - 9
136 S.E. Rosenbaumr Pharmaceutica Acta Heletiae 73 (1998) 135144

Fig. 1. Pharmacokinetic model. The rate of all processes except absorption were based on clearance. Where, Cl nh is the drugs non-hepatic clearance, Cl m
is the metabolites clearance, E is the hepatic extraction ratio w Es Cl int = f u .r QqCl int = f u .x, Cl int is intrinsic unbound hepatic clearance of the drug,
f u is the fraction of the drug unbound in the plasma, Q is hepatic blood flow. The drugs hepatic clearance, Cl h is equal to EQ. The rate of absorption was
expressed using the first order rate constant k a .. ) An error model for intra-individual variability in hepatic clearance was added as described in Section
2.

sive and variable pre-systemic metabolism by the gastro-
intestinal andror hepatic enzymes systems Shah et al.,
1996.. Large sample sizes may also be necessary for drugs
that form active metabolites, where bioequivalence usually
has to be demonstrated with respect to both parent drug
and the metabolite Hauck et al., 1995.. In such cases,
owing to the differing forms and degrees of variability
experienced by the drug and its metabolite, it is possible
for a study to demonstrate equivalency with respect to one
species but not the other.
The manner in which variability in hepatic clearance
affects the AUC of the drug and that of the metabolite is
potentially complex and a function of the magnitude and
existence of alternative elimination processes for the drug
andror the metabolite. An understanding of these factors
could be valuable to help optimize the design of bioequiva-
lence studies and avoid the need to use extremely large
sample sizes. For example, if it were possible to predict
the probable variability in the bioequivalence parameters
of the drug and metabolite, based on their population
pharmacokinetics, greater weight could be placed on the
species with the more robust, less variable bioequivalence
parameter. Thus fewer subjects would be needed for the
study.
In order to clarify these factors a semi-physiological
pharmacokinetic model with controlled error in hepatic
clearance was developed. Plasma concentration time data
were simulated under a variety of conditions and the
variability in the AUC of the drug and its metabolite
studied.

Fig. 2. a. STELLAw diagram of the pharmacokinetic model used for oral drug administration. Where k a , is the first order absorption rate constant; Q,
hepatic blood flow; Clh i , hepatic clearance of the drug in individual i; Clh i, j , hepatic clearance of the drug in individual i at time j; Cl nh is the
non-hepatic clearance of the drug; Cl m is the clearance of the metabolite; Vd d and Vd m are the volumes of distribution for the drug and metabolite
respectively and Cp d and Cp m are the plasma concentrations of the drug and metabolite respectively. Hepatic clearance was calculated using the well
stirred venous equilibrium model as described in Section 2. Details of the statistical model for intra-subject variability are also given in Section 2. b.
STELLAw diagram of the pharmacokinetic model used for intravenous drug administration. Where Clh i, j , hepatic clearance of the drug in individual i at
time j; Cl nh is the non-hepatic clearance of the drug; Cl m is the clearance of the metabolite; Vd d and Vd m are the volumes of distribution for the drug
and metabolite respectively and Cp d and Cp m are the plasma concentrations of the drug and metabolite respectively. Hepatic clearance was calculated
using the well stirred venous equilibrium model as described in Section 2. Details of the statistical model for intra-subject variability are also given in
Section 2.
 S.E. Rosenbaumr Pharmaceutica Acta Heletiae 73 (1998) 135144 137

2. Experimental procedures tion k GI s 5 hy1 . through the lumen was created in

2.1. Pharmacokinetic model
A one-compartmental semi-physiological pharmacoki-
netic model, with first order gastro-intestinal GI. absorp-
STELLAw High Performance Systems, Hanover, NH..
The general nature of the model is shown in Fig. 1 and the
STELLAw model is shown in Fig. 2a. The absorption rate
constant was maintained constant throughout the study and
138 S.E. Rosenbaumr Pharmaceutica Acta Heletiae 73 (1998) 135144

complete absorption of the dose through the GI lumen was ered separately from its function in the clearance of the
assumed for all simulations. Thus, the fraction of the dose drug from the systemic circulation.
reaching the liver Ff . was maintained at unity. The The drug was assumed to undergo metabolism in the
function of the liver in the first pass process was consid- liver by a single, linear enzymatic process. A value of 90

Fig. 3. Typical plasma concentration time data generated from the model. Graph of plasma concentration against time for the parent drug a. and the
metabolite b.. The 100 data sets were generated from the computer model described Section 2. The data in the graph were generated from the oral
administration model with an intrinsic hepatic clearance of 90 lrh Es 0.5. and a non-hepatic clearance of 1.2 lrh.
 S.E. Rosenbaumr Pharmaceutica Acta Heletiae 73 (1998) 135144
lrh was used for hepatic blood flow Gibaldi and Perrier,
1982. and 1.34 l for the volume of the liver Bourne,
1995.. It was assumed that the drug did not bind to the
plasma proteins. Hepatic clearance was calculated using
the well-stirred venous equilibrium model Wilkinson and
Shand, 1975; Rowland et al., 1973. using values of intrin-
sic hepatic clearance from 360 to 10.0 lrh. Thus, the
hepatic extraction ratio E . was varied from 0.8 to 0.1
respectively, with corresponding values of the maximum
oral bioavailability Fh . varying from 0.2 to 0.9 respec-
tively. During the simulations, the drugs clearance by
other routes andror pathways non-hepatic clearance. was
set either at a low or high value; 1.2 lrh or 50 lrh
respectively. The drugs volume of distribution was varied
from 54 l to 206 l, in direct proportion to the changes in
total body clearance in order to maintain a constant half-life
of 1.95 h. The clearance of the metabolite was varied from
2.25 to 200 lrh and its volume of distribution from 19.4 to
1729 l in order to maintain the half-life at 5.99 h.
2.2. Intraenous model
The basic model was converted to an intravenous model
by eliminating the gastro-intestinal compartment and hep-
Fig. 4. The coefficient of variation CV. of the drug as a function of the hepatic extraction ratio. Plasma concentration time data were generated from a
pharmacostatistical model with oral solid lines. and intravenous dashed lines. drug input. Error was added to hepatic clearance. Simulations were
performed when non-hepatic clearance was high 50 lrh. B. and low 1.2 lrh. '..
139
atic first pass extraction Fig. 2b.. This model was used to
study the effect of the drugs extraction ratio on the
variability of the AUC of the drug.
2.3. Variability
Intra-subject variability in hepatic clearance was added
to the pharmacokinetic model as described below. A con-
stant coefficient of variation model was used. Thus, intra-
individual variability was modeled to cause the hepatic
clearance to deviate from the mean value in an individual,
by an amount that was proportion to the mean value in that
individual. Thus:
Clh i , j s Clh i = 1 q l .
where: Clh i, j represents hepatic clearance in individual i at
time j. Clh i represents the mean hepatic clearance in
individual i. l is a normally distributed random variable
with an average value of 0. The standard deviation of l
represents the coefficient of variation for intra-individual
variability.
The coefficient of variation for intra-individual variabil-
ity was set at 30%, which is within the range reported in
140 S.E. Rosenbaumr Pharmaceutica Acta Heletiae 73 (1998) 135144

the literature Sheiner and Ludden, 1992.. During a simu-
lation, hepatic clearance was varied according to the model
above every hour so that the effects of intra-individual
variability would be magnified. Assay error was not in-
cluded in the simulations.
2.4. Simulations
Simulations were carried out using a fixed dose of 50
mg. For each set of model parameters 100 sets of plasma
concentration time data were simultaneously generated for
the drug and the metabolite. Plasma concentrations were
obtained from zero to 24 h after the dose, and were
generated by numerical integration every 0.01 h. All data
were expressed to at least three significant figures.
2.5. Bioequialence parameters
The bioequivalence parameters for the drug and the
metabolite were determined from an abbreviated data set,
designed to mimic the timed samples of a bioequivalence
study. Thus, assessments were based on the simulated
concentrations at: 0.5, 1, 1.5, 2, 2.5, 3, 4, 5, 6, 8, 10, 12,
18 and 24 h. The elimination rate constant was determined
from the slope of the regression line that best fit the
terminal last five data points. portion of the loglinear
concentration-time curve. The area under the curve AUC.
from zero to the last measured sample was determined
using the trapezoidal rule. The AUC from the last mea-
sured sample to infinity was determined by dividing the
last plasma concentration by the elimination rate constant.
2.6. Model alidation
The operation of the structural component of the model
was assessed by generating output in the absence of an
error model. Pharmacokinetic analysis was then performed
on the simulated data using commercially available soft-
ware MKMODELw , Biosoft, Cambridge, UK.. The fol-
lowing pharmacokinetic parameters were determined:
VdrF, ClrF, the overall absorption rate constant k a ..
Data simulated from the intravenous bolus injection model
was also used in order to permit the estimation of F
following the oral dose. This procedure was carried out for
the whole spectrum of pharmacokinetic parameters stud-
ied. A similar procedure was performed for the metabolite.
In all cases the modeled values of Cl, Vd and F for the
drug and Cl and Vd for the metabolite were within 99.5%
of their values assigned in the model. The overall absorp-
tion rate constant, which was a function of GI absorption

Fig. 5. The coefficient of variation CV. of the drug B. and the metabolite '. as a function of the ratio of the drugs hepatic clearance to its total
clearance. Plasma concentration-time data were derived from a pharmacostatistical model for oral drug input with error added to hepatic clearance. The
drugs hepatic extraction ratio was set to 0.2.
 S.E. Rosenbaumr Pharmaceutica Acta Heletiae 73 (1998) 135144 141

k GI . and intake from the liver was modeled to a value of
4.6 hy1 , compared to the assigned value of 5 hy1 for k GI .
Additionally, all the AUCs estimated during the course of
the study were compared to their theoretical values calcu-
lated using basic pharmacokinetic equations see Section
4.. The estimated values "s.d.. were 102.6% "6.2. and
95.5% "1.5. of their theoretical values for the drug and
metabolite respectively.
2.7. Effect of the drugs clearance by other pathways
In order to study the effect of the drugs clearance by
other pathways, the drugs non-hepatic clearance was var-
ied from 1 to 100 lrh. This effect was studied when the
drugs extraction ratio was equal to 0.2 and 0.8.
2.8. Effect of the metabolites clearance
The effect of the metabolites clearance was investi-
gated by varying its value from 5 to 200 lrh. The metabo-
lites volume of distribution was also varied from 43.25 l
to 1730 l. in order to maintain the half-life at an mean
value of 5.99 h.
3. Results
Typical output from the computer simulations for the
drug and the metabolite are shown in Fig. 3a and b
respectively. The pharmacostatistical model demonstrated
that after oral administration, the variability in the AUC of
the drug increased as the drugs extraction ratio increased
Fig. 4.. After intravenous input, however, less variability
in the AUC of the drug was apparent Fig. 4.. At high
extraction ratios E s 0.8., the CV of the AUC was 5.5
and 3.5 fold higher after oral input compared to intra-
venous, which suggests that pre-systemic extraction is the
main source for the of variability in the AUC observed
after oral input.
Only the model for oral input was used to study the
effect of non-hepatic clearance. At low extraction ratios, as
non-hepatic clearance increased and hepatic clearance be-
came a lower fraction of overall clearance, the CV of the

Fig. 6. The coefficient of variation CV. of the drug B. and the metabolite '. as a function of the ratio of the drugs hepatic clearance to its total
clearance. Plasma concentration-time data were derived from a pharmacostatistical model for oral drug input with error added to hepatic clearance. The
drugs hepatic extraction ratio was set to 0.8.
142 S.E. Rosenbaumr Pharmaceutica Acta Heletiae 73 (1998) 135144

Fig. 7. The coefficient of variation CV. of the drug B. and the metabolite '. as a function of the hepatic extraction ratio. Plasma concentration-time
data were derived from a pharmacostatistical model for oral drug input with error added to hepatic clearance. Simulations were performed when
non-hepatic clearance was high 50 lrh. dashed lines. and low 1.2 lrh. solid lines..

AUC of the drug decreased Fig. 5.. At high extraction
ratios, where the CV was already high, the non-hepatic
clearance had no discernible effect on the CV of the AUC
of the drug Fig. 6..
In contrast, the CV of the AUC of the metabolite
increased as non-hepatic clearance increased Figs. 5 and
6. and decreased as the drugs extraction ratio increased
Fig. 7.. Additionally, although the variability of the AUC
of the metabolite was low when the extraction ratio was
large, a further reduction in the CV was apparent as the
value of non-hepatic clearance decreased.
The value of the metabolites clearance had no impact
on either the variability of the AUC of the drug nor the
metabolite data not shown..
4. Discussion
Bioequivalence studies for drugs that undergo pre-sys-
temic extraction are often complicated by the high intra-in-
dividual variability in hepatic clearance, which can de-
mand an unreasonably large sample size. A pharmacosta-
tistical model was developed and used to study the condi-
tions that control the manner in which this source of
intra-individual variability is translated to the AUC of a
drug and its metabolite. In order to clarify these relation-
ships, a simple linear one compartmental pharmacokinetic
model was used and intra-individual variability was re-
stricted to the single metabolic process that produced the
metabolite under investigation.
The model demonstrated that the variability experienced
in the AUC of the drug and the metabolite were affected in
different ways by the pharmacokinetic parameters consid-
ered in this study. For example, as the drugs extraction
ratio increased, the CV of AUC of the drug increased
while that of the metabolite decreased. Also as non-hepatic
clearance increased, the CV of the AUC of the drug
decreased while that of the metabolite increased. All of the
effects found in the study are in keeping with predications
based on the fundamental pharmacokinetic equations.
The AUC of an orally administered drug is given by the
equation:
Ff = Fh = Dd
AUC d s
Cl h ,d q Cl nh ,d
where: Ff s fraction of the dose reaching the liver, set to 1
 S.E. Rosenbaumr Pharmaceutica Acta Heletiae 73 (1998) 135144
in this study. Fh s fraction of the drug entering the liver
that is not extracted, or the maximum oral bioavailability;
Fh s 1 y E .. Dd s dose of the drug. Cl h,d s hepatic clear-
ance of the drug. Cl nh,d s non-hepatic clearance of the
drug.
Thus, variability in hepatic clearance will impact upon
the AUC of the drug directly through systemic clearance
and indirectly through the fraction remaining after variable
pre-systemic clearance. At high extraction ratios, the over-
all variability in the AUC of the drug was reduced by
several orders of magnitude when pre-systemic clearance
was removed in the model with intravenous input. This
suggests that at high extraction ratios, pre-systemic
metabolism may contribute proportionally more to the
overall variability in the AUC of the drug than systemic
clearance.
Based on the formula above, as non-hepatic clearance
of the drug increases, the impact of the variability in
hepatic clearance on overall systemic clearance of the drug
will decrease, leading to an overall reduction in the CV of
the AUC of the drug. Non-hepatic clearance however
would not be expected to affect pre-systemic clearance and
the variability in the AUC of the drug that arises here. At
low extraction ratios, the CV of the AUC of the drug
decreased as the non-hepatic clearance increased. How-
ever, when the extraction ratio was large, the value of
non-hepatic clearance had little effect on the CV of the
AUC of the drug. For example, when E s 0.8, the overall
variability on the AUC of the drug was the same when
non-hepatic clearance was 40% of total clearance as it was
when it was 1.6%. This insensitivity to non-hepatic clear-
ance at high extraction ratios also supports the theory that
pre-systemic extraction is the main source of variability in
the AUC of a drug subject to extensive pre-systemic
extraction.
The AUC of the metabolite is given by the formula:
Dm
AUC m s
Cl m
where: Cl m s total body clearance of the metabolite. Dm
s amount of metabolite available systemically.
The amount of metabolite available will be the sum of
that produced during the first pass extraction of the drug,
and that produced during the systemic clearance of the
drug. Thus:
Dm s E = Ff = Dd . q 1 y E . = Ff = Dd = f m .
and
Dm s Ff = Dd E q 1 y E . f m
where: f m s fraction of the systemically available drug
subject to metabolism and
f m s Cl h ,d Cl d s Cl h ,d Cl h ,d q Cl nh ,d
143
Thus, in this study:
Dd E q 1 y E . = f m
AUC m s .
Cl m
When non-hepatic clearance is at a minimum:
Clh Clh
fm s f s1
Clh q Clnh Clh
and
Dd
AUC m s .
Cl m
Thus, the AUC of the metabolite is independent of
hepatic clearance and by extension variability in this pro-
cess. Drug is cleared only by metabolism and since the
whole dose of the drug is ultimately converted to the
metabolite, the amount of metabolite formed is insensitive
to variability in the rate of the process. This was confirmed
in the study, which illustrated negligible variability in the
AUC of the metabolite when non-hepatic clearance was
low Fig. 5..
As the drugs non-hepatic clearance increased, variabil-
ity in hepatic clearance of the drug exerted an increasingly
large effect on the AUC of the metabolite; firstly through
an effect on the amount of metabolite produced during first
pass extraction and secondly through an effect on the
fraction of the systemically available drug that is metabo-
lized. Thus, the CV of AUC of the metabolite increased as
non-hepatic clearance increased. Under conditions of high
non-hepatic clearance of the drug, when significant vari-
ability in the AUC of the metabolite existed, the CV of the
AUC decreased as the hepatic extraction ratio increased.
5. Conclusions
In summary, these simulations have demonstrated that
the CV of the AUC of the drug is greatest when the drugs
hepatic extraction ratio is large and suggest that variability
in pre-systemic rather than systemic clearance is the major
source of this variability. At high extraction ratios, the
amount of non-hepatic clearance had little influence on the
variability translated to the drug. In contrast, the AUC of
the metabolite was least variable at high extraction ratios
and became less variable as non-hepatic clearance de-
creased.
Clearly, intra-individual variability associated with the
other elimination pathways. of the drug and metabolite
will control the complete variability experienced in the
AUC of both species. However, providing the metabolite
does not in itself experience extensive first pass extraction,
the results of this study suggest it is unlikely that the CV
of the AUC of the metabolite of a highly extracted drug
would ever reach that of the parent drug. Under these
144 S.E. Rosenbaumr Pharmaceutica Acta Heletiae 73 (1998) 135144
circumstances, bioequivalence studies based on the
metabolite may require a much smaller sample size.
The pharmacostatistical model was valuable in facilitat-
ing an understanding of the influence of factors that con-
trol the translation of intra-individual variability in hepatic
clearance to the AUC of a drug and its metabolite. These
models should also prove helpful for assessing the impact
of other sources of variability on the bioequivalence pa-
rameters. Additionally, models with an added pharmacody-
namic component could be used to relate variability in the
either pharmacokinetic parameters andror bioequivalence
parameters to response, which may provide information
about circumstances in which existing bioequivalency lim-
its could be loosened without any added consumer risk.
References
Boddy, A.W., Snikeris, F.C., Kringle, R.O., Wei, G.C.G., Opperman,
J.A., Midha, K.K., 1995. An approach for widening the bioequiva-
lence limits in the case of highly variable drugs. Pharm. Res. 12,
18651868.
Bourne, D.W., 1995. In: Mathematical Modeling of Pharmacokinetic
Data. Technomic Lancaster, pp. 32.
Diletti, E., Hauschke, D., Steinijans, V.W., 1991. Sample size determina-
tion for bioequivalence assessment by means of confidence intervals.
Int. J. Clin. Pharmacol. Ther. Toxicol. 29, 18.
Gibaldi, M. Perrier, D., 1982. In: Pharmacokinetics, 2nd edn. Marcel
Dekker, New York, pp. 364.
Hauck, W.W., Hyslop, T., Anderson, S., Bois, F.Y., Tozer, T.N., 1995.
Statistical and regulatory considerations for multiple measures in
bioequivalence testing. Clin. Res. Regulatory Affairs 12, 249265.
Hauschke, D., Steinijans, V., Diletti, E., Burke, M., 1992. Sample size
determination for bioequivalence assessment using a multiplicative
model. J. Pharmacokin. Biopharm. 20, 557561.
Rowland, M., Benet, L.Z., Graham, G.G., 1973. Clearance concepts in
pharmacokinetics. J. Pharmacokinet. Biopharm. 1, 123136.
Shah, V.P., Yachobi, A., Barr, W.H., 1996. Evaluation of orally adminis-
tered highly variable drugs and drug formulations. Pharm. Res. 13,
15901594.
Sheiner, L.B., Ludden, T.M., 1992. Population pharmacokineticsrdy-
namics. Annu. Rev. Pharmacol. Toxicol. 32, 185209.
Wilkinson, G.R., Shand, D.G., 1975. A physiologic approach to hepatic
drug clearance. Clin. Pharmacol. Ther. 18, 377390.