doi: 10.1111/j.1472-8206.2005.00315.

ORIGINAL Population pharmacokinetic analysis of

ARTICLE
indinavir in HIV-infected patient treated
with a stable antiretroviral therapy
Karl Brendela*, Mayeule Legrandb, Anne-Marie Taburetc, Gabriel Barona,
Cécile Goujardd, France Mentréa and the Cophar 1-ANRS 102 Trial Group
a
INSERM E0357, Department of Epidemiology, Biostatistics and Clinical research, AP-HP, Bichat University Hospital,
Paris, France
b
Clinical Pharmacology Department, AP-HP, Pitié-Salpétrière University Hospital, Paris, France
c
Clinical Pharmacy, AP-HP, Bicêtre University Hospital, Bicêtre, France
d
Internal Medicine, AP-HP, Bicêtre University Hospital, Bicêtre, France

Keywords
indinavir,
population
pharmacokinetics,
ritonavir,
therapeutic drug
monitoring
Received 26 March 2004;
revised 19 November 2004;
accepted 30 November 2004
*Correspondence and reprints:
karl.brendel@bch.ap-
hop-paris.fr
ABSTRACT
The objectives of this study were to build a population pharmacokinetic model that
describe plasma concentrations of indinavir in human immunodeficiency virus (HIV)-
infected patients with sustained virological response under a stable antiretroviral
combination, and to characterize the effect of covariates and comedications on
indinavir pharmacokinetics. Data were obtained from 45 patients who received
different dosages of indinavir: either indinavir alone t.i.d. (mostly 800 mg), either
indinavir b.i.d. (mostly 800 mg) with a booster dose of 100 mg of ritonavir. Patients
were required to have a baseline plasma HIV RNA <200 copies/mL and to have
unchanged antiretroviral treatment for 6 months. Indinavir concentrations were
measured at a first visit (one sample before drug administration and five after) and at
a second visit 3 months later (before and 1 or 3 h after drug administration). A one-
compartment model with first-order absorption and first-order elimination best
described indinavir pharmacokinetics. For patients treated with indinavir alone,
absorption rate constant was estimated to be 0.43/h, and oral clearance Cl/F was
33 L/h. For patients treated with indinavir plus ritonavir these estimates were 0.25/h
and 19 L/h, respectively. Cl/F was found to increase by 1.45-fold in men and by
1.18-fold in patients also receiving zidovudine. Oral volume of distribution (V/F) was
24 L. The inter-individual and intra-individual variability were 117 and 205% for
V/F, 42 and 58% for Cl/F, respectively. This population analysis in patients with
sustained virological response, quantified the effect of ritonavir on the absorption rate
constant and on the clearance of indinavir, showed an increase of Cl/F in men and
can be used to draw reference curve for therapeutic drug monitoring.

inhibits the viral protease enzyme, one of the key

INTRODUCTION
Since the introduction of highly active antiretroviral
therapy (HAART), life expectancy of human immuno-
deficiency virus (HIV)-infected patients has been largely
improved by potent antiretroviral regimens including
protease inhibitors (PIs) and/or non-nucleoside reverse
transcriptase inhibitors (NNRTIs) [1,2]. Indinavir is a
PI prescribed in the treatment of HIV infection which
enzymes that HIV needs in order to replicate. Results
from a number of studies suggest that indinavir should
be used as part of a combination therapy regimen with
other anti-HIV drugs [3]. Indinavir plasma concentra-
tions have been shown to be correlated with antiviral
efficacy [4–9]. Furthermore, several studies showed
a link between indinavir plasma concentration and
toxicity [8,10,11].

Ó 2005 Blackwell Publishing Fundamental & Clinical Pharmacology 19 (2005) 373–383 373
374
Indinavir is metabolized in seven compounds: a
glucuronide conjugate and six oxidative metabolites
(CRIXIVANÒ; Merck & Co., Imc, Whitehouse station,
New Jersey, USA, US Package insert Merk RCO, 2002). It
is mainly metabolized by the liver into oxidative metab-
olites and in vitro studies indicate that cytochrome P450
(CYP 3A4) is the major isoenzyme for oxidative meta-
bolism. As ritonavir is a very potent CYP 3A4 inhibitor,
indinavir plasma concentrations increase when ritonavir
is co-administrated [12–14]. With a booster dose of
ritonavir, 100 mg b.i.d., patients usually receive 800 mg
b.i.d. of indinavir instead of the standard regimen of
800 mg t.i.d. when indinavir is given without ritonavir.
Although the increase in indinavir peak concentration
with ritonavir enhancement is relatively low, the overall
exposure enhancement appears to increase the risk of
renal toxicity significantly in boosted regimens [15].
The population approach is well adapted to describe
the pharmacokinetics of indinavir in HIV patients. This
method can analyze sparse individual data with few
concentrations by patient. In HIV-infected patients, four
population pharmacokinetic analyses of indinavir are
published [16–19]. They involved data on 25, 22, 11,
and 171 HIV-infected patients, respectively. The purpose
of the three first studies was to analyze cerebral
penetration of indinavir from the ratio of the area under
the cerebrospinal fluid and under the plasma concentra-
tion curves. The last study analyzed the effect of the
coadministration of several PIs on the pharmacokinetic
of efavirenz, and did not report a detailed population
pharmacokinetic analysis of indinavir.
The purpose of this paper were to build a population
pharmacokinetic model that describe plasma concentra-
tions of indinavir in HIV-infected patients with sustained
virological response under a stable antiretroviral combi-
nation, and to characterize the effect of covariates and
comedications on indinavir pharmacokinetics. The other
aim was to estimate inter-individual and intra-individual
variability of the pharmacokinetic parameters.
MATERIALS AND METHODS
Study population
Data were obtained from patients enrolled in the
COPHAR 1-ANRS 102 study which was an open,
multicentric (18 centers in France) and prospective trial
whose aim was to describe indinavir and nelfinavir
plasma concentrations in HIV-infected patients. This trial
complies with the current laws of France: all subjects
gave a written consent and the protocol was approved
Ó 2005 Blackwell Publishing Fundamental & Clinical Pharmacology 19 (2005) 373–383
K. Brendel et al.
by the Ethic committee of Bicêtre Hospital (France). One
arm of this study was composed of patients who took
indinavir and the other arm of patients with nelfinavir.
Patients from the indinavir arm were analyzed in this
paper. Patients were required to have a baseline plasma
HIV RNA value below 200 copies/mL for at least
4 months and to have the same antiretroviral treatment,
including indinavir, for at least 6 months. There was no
indication given to physician on the indinavir dosage
and indinavir could be associated with ritonavir booster
(100 mg b.i.d.).
Exclusion criteria were: co-administration of a treat-
ment known for its interaction with indinavir (such as
rifampicin or itraconazole), renal deficiency (creatinemia
>180 lmol/L), hepatic deficiency (prothrombin time
<50% or cirrhosis), ongoing opportunistic infection,
previous or ongoing chemotherapy.
Inclusion criteria were verified at visit 0 (V0). A first
indinavir pharmacokinetic profile was performed at visit
1 (V1) within a maximum of 30 days after V0. Four
months after V0, two other plasma samples were collected
at visit 2 (V2). The duration of the trial was 8 months
from V0, to the last visit V3. Routine clinical data, drug-
related adverse event were recorded at each visit.
Adherence was evaluated at V1 and V2 using a validated
auto-questionnaire and applying the algorithm proposed
by Carrieri et al. [20] to compute a score from 0 to 100%
adherence. Patients were classified as highly adherent if
they reported taking 100% of their prescribed regimen in
the last five days, and non-highly adherent otherwise.
Plasma collection and analysis
For the pharmacokinetic study at V1, five blood samples
were collected: a trough sample before indinavir admin-
istration and samples at 0.5, 1, 3 and 6 h after indinavir
administration. The two samples collected at V2 are a
trough concentration and a peak plasma concentration
(Cmax). Because of the reported impact of ritonavir on
indinavir absorption, the peak concentration at V2 was
collected 1 h after administration for indinavir alone and
3 h after administration for indinavir associated with a
booster dose of ritonavir.
Indinavir concentrations were, when possible, assayed
in the laboratory of pharmacology or toxicology of the
clinical site having included the patient or sent to a close
one. A specific quality control was set up before the trial
to validate the laboratories which were allowed to take
part in the study to analyze indinavir concentrations.
Twelve laboratories participated to indinavir assay in
plasma, after validation by a blind external quality
Population pharmacokinetics of indinavir
control. Plasma concentrations of indinavir were deter-
mined by specific high-performance liquid chromato-
graphy assays with ultraviolet–photodiode array or
spectrofluorometric detection, according to previously
published assays [21–25]. In brief, indinavir was isolated
from alkaline plasma samples by double-step liquid–liquid
extraction or by solid–liquid extraction. Dry extracts were
injected onto C18 reversed phase column. Results of the
blind interlaboratory quality control at three concentra-
tions were within 15% of the target values for medium
and high values and within 20% for low values [26].
Lower limit of quantification (LOQ) were 0.005–
0.1 mg/L for indinavir, depending on each method.
Population pharmacokinetic modeling
A one-compartment model with first-order absorption
and first-order elimination was used to fit the concen-
tration–time data for indinavir and for ritonavir. This
model was parameterized with the oral volume of
distribution (V/F), the first-order absorption rate con-
stant (Ka), and the oral clearance (Cl/F).
The statistical model for the observed plasma concen-
tration of one drug, Ci,j in patient i at time ti,j, is given by:
Ci;j ¼ f ðti;j ; hi Þ þ eij
where hi is the pharmacokinetic parameter vector of
patient i, eij the measurement error and f the pharma-
cokinetic model.
An exponential random-effect model was chosen to
describe inter-individual variability:
hi ¼ h expðgi Þ
where h is the population mean vector of the phar-
macokinetic parameters, and gi represents the random
effect vector. Random effects were assumed to follow a
normal distribution with zero mean and variance matrix
X which was supposed to be diagonal.
Residual variability was modeled using a combined
proportional and additive error model. The measurement
error eij is assumed to follow a normal distribution with
zero mean and an heteroscedastic variance r2i;j given by:
r2i;j ¼ r2 ða þ f ðti;j; hi ÞÞ2 :
This model is often used in population pharmacokinet-
ics. For high concentrations, variance becomes propor-
tional to the square of the concentration. For the lowest
concentrations, the variance becomes proportional to a2.
When indinavir concentrations were below the LOQ of
Ó 2005 Blackwell Publishing Fundamental & Clinical Pharmacology 19 (2005) 373–383
375
the center where it was analyzed, we fixed them to LOQ/2
using the LOQ of the corresponding assay [27].
Computational methods
The models were fitted to the data by use of WinNonMix
version 2.0.1 (Pharsight Corporation, Mountain View,
CA, USA) with the first-order method (FO). Simulations
were also performed using WinNonMix version 2.0.1.
Analysis of concentrations at V1
First the indinavir model was built with the data at V1.
Using the best error model, selected using the Akaike
criterion (AIC), several models were tested to determine
the basic model that best fit the data. The basic model was
constructed by testing the inclusion of random effects on
Ka, Cl/F and V/F, and by testing the influence of taking or
not ritonavir on the three parameters. Goodness-of-fit
plots (observed vs. predicted population and individual
concentrations, weighted residuals vs. predicted concen-
trations and vs. time) were examined for each model.
Models were also compared using the AIC. The correlation
between the pharmacokinetic parameters was also tested.
Secondly, the effects of the following covariates were
evaluated from the basic model: age, weight, gender,
other antiretroviral drugs and adherence at V1 (as a
categorical variable). The effects of covariates on the
empirical Bayes estimates of each individual random
effects from the basic model with ritonavir, were tested
by using a Mann–Whitney non-parametric test for
categorical variables and a correlation test for continu-
ous variables. The covariate model was then built with
the covariates which were found to have a significant
effect in this first step (P < 0.05). All population models
with all the combinations of these covariates were then
evaluated. The combination with the smallest AIC was
chosen in order to built the population covariate model.
Analysis of concentrations at V1 and V2
From the model with covariates obtained with the data
at V1, the inter-occasion variability (IOV) was studied by
analyzing together the data at V1 and V2. The following
model was used where k ¼ 1 at V1 and k ¼ 2 at V2. Let
bki be the vector of random effects for total variability at
occasion k, gi for inter-individual variability (IIV) and jki
for inter-occasion variability at occasion k. We have:
bki ¼ gi þ jki
Varðbki Þ ¼ Varðgi Þ þ Varðjki Þ
because gi and jki are assumed to be independent.
376
However, with the software WinNonMix, this more
complex random effects model cannot be implemented
directly. Because there are only two occasions, we found
a way to implement it using IOV as a covariate. Two
random effects were modeled in WinNonMix: g0 for
occasion 1 only and g1 added to g0 for occasion 2. The
expressions for this implementation of the IOV are given
in the Appendix.
The addition of random effects for IOV on one or
several parameters was evaluated and goodness-of-fit
plots were again used to assess the reliability of the
models and AIC in order to compare them.
From the best model including IOV, a backward
elimination procedure was then used to test whether
all covariates selected from V1 concentrations should
remain in the final model. The likelihood ratio test (LRT)
was used to compare the corresponding nested models.
More precisely, if the elimination of a covariate signifi-
cantly (P < 0.05) decreased the log-likelihood, that
covariate was kept in the model.
Model evaluation
Simulations of indinavir steady-state concentrations
profile were performed and compared with the observed
ones to evaluate the predictive performance of the
models. We simulated 1000 patients with 800 mg t.i.d.
without ritonavir and 1000 patients with 800 mg b.i.d.
of indinavir and 100 mg b.i.d. of ritonavir. The simula-
tions were done using the final model of indinavir. We
used, in the simulations the total variability (inter plus
intra) of the random effects and the residual error
variability in order to mimic the observations.
We kept in each group of simulation the same
distribution of the significant covariates as in the study
population.
The 10th and 90th predicted percentiles for the 1000
simulations in each group were compared with the
observed concentrations at V1 and V2 for the patients
receiving the corresponding dosage. It should be noted
that in order to evaluate the non-parametric 10th and
90th percentiles, a rather large sample of simulated
patients is needed.
Simulation of trough concentrations
with the final model
From the final population model of indinavir, we
simulated the range of individual steady-state trough
concentrations for the two usual dosage regimen:
800 mg t.i.d. for indinavir alone and 800 mg b.i.d. for
indinavir with 100 mg b.i.d. of ritonavir. We simulated
Ó 2005 Blackwell Publishing Fundamental & Clinical Pharmacology 19 (2005) 373–383
K. Brendel et al.
200 patients for each group of combination of significant
covariates given the total variability of the random
effects without measurement error. This was performed
in order to illustrate the clinical relevance of the
significant covariates found in the population analysis
on the real-through concentrations (without measure-
ment error) which are thought to be related both to
efficacy and toxicity [1,6].
RESULTS
Patients
Concentration–time data were obtained from 45 patients
who received indinavir. Among these patients, 14 also
received a booster dose of ritonavir (100 mg b.i.d.).
Indinavir dosage regimens were as follows: 600 mg t.i.d.
(one patient), 800 mg t.i.d. (27 patients), 1000 mg t.i.d.
(two patients) and 1200 mg b.i.d. (one patient) in
patients receiving indinavir alone and 400 mg b.i.d.
(five patients), 600 mg b.i.d. (one patient) and 800 mg
b.i.d. (eight patients) in patients receiving also ritonavir.
Table I describes the main characteristics of these
patients. The antiretroviral drugs associated with indi-
navir were lamivudine (3TC, n ¼ 36), stavudine (D4T,
n ¼ 22), zidovudine (AZT, n ¼ 13), didanosine (DDI,
n ¼ 9), nevirapine (NVP, n ¼ 5), abacavir (ABC, n ¼ 3),
and efavirenz (EFV, n ¼ 2).
Among the 45 patients included in the trial who
received indinavir, three patients experienced virological
failure during the trial and all received indinavir alone.
One patient experienced nephrolithiasis (at V1 but not at
V2) which led to treatment discontinuation after V1.
Liver function remained unchanged during the trial. No
clinically relevant grade III or IV laboratory abnormal-
ities were observed.
Basic model
Figure 1 displays indinavir plasma concentrations at V1
and V2 vs. time for patients with or without ritonavir.
Table I Patient characteristics at baseline.
Patients (n ¼ 45) Median (range)
Age (years) 40.3 (21–65)
Weight (kg) 72 (41–110)
Sex (male/female) 35/10
Time since first ARV treatment (years) 4.9 (0.9–10.7)
CD4 cell count (/mm3) 599.6 (150–1425)
Adherence (%) 100 (44–100)
Population pharmacokinetics of indinavir 377

(a) 18 ritonavir. From both the Akaike criterion and the LRT,
the best model was the model with a ritonavir effect
Indinavir concentration (mg/L)

16

14 on Ka and on Cl/F. With that model (AIC ¼ 800.8),

12
the variability was low and poorly estimated for gKa. A
10

8
model with random effect only on Cl/F and V/F was
6 then tested (AIC ¼ 800.5) and kept as the basic
4
model.
2

0
The equations of this model are:
0 2 4 6 8 10 12 14 16
Time (h) Ka;i ¼ ðKa
ð1  ritonaviri Þ þ Ka;rito
ritonaviri Þ
(b) 18
Indinavir concentration (mg/L)

16
Cl=Fi ¼ ðCl=F
ð1  ritonaviri Þ þ Cl=Frito
ritonaviri Þ
14

12
expðgCl;i Þ
10

8 V=Fi ¼ V=F
expðgV;i Þ

6

4
where ritonaviri ¼ 0 if patient took indinavir alone and
2

0
0 2 4 6 8 10
Time (h)
Figure 1 Observed plasma indinavir concentration vs. time for
patients without ritonavir (a) and with ritonavir (b) at the two visits
V1 and V2.
It shows that the decrease of concentrations was
slower when ritonavir was co-administered with indina-
vir and that indinavir Cmax was obtained later. That is
why at V2 the first sample was collected at 1 h for
indinavir alone and at 3 h for indinavir with ritonavir
(except for three patients where it was collected at 1 h).
Among the 305 samples, two indinavir plasma concen-
trations in two different patients, were below the limit of
quantification at 10 and 10.6 h and were fixed to the
LOQ/2 of the assay of the corresponding center, i.e.
0.025 mg/L in both cases.
First, the combined error model with random effects
on Ka, Cl/F and V/F was used. The initial value for a2
was the average (LOQ/2)2 ¼ 0.0004 (mg/L)2 and the
model converged to a2 ¼ 0.029. This model (AIC ¼
867.8) was better than a model with only a proportional
error model (AIC ¼ 876.6). Because of the high gain in
the AIC when a combined error variance was used on
the data at V1, and because the subsequent models with
combined error variance did not converge during the
tests of covariates and of the inter-occasion variability,
this value of a2, 0.029, was fixed and kept in the
following analyses.
Significant effects of ritonavir on empirical Bayes
estimates of Ka (P < 0.001) and Cl/F (P < 0.003) were
found. We then implemented and tested in the
population model the effect of co-administration of
ritonaviri ¼ 1 if it was coadministrated.
12 14 16 The population estimates for Ka and Cl/F were
0.43/h and 34.22 L/h, respectively, for patients who
received indinavir without ritonavir, 0.26/h and
28.88 L/h for patients with ritonavir. The estimated
volume of distribution was 21.45 L. Large inter-indi-
vidual variabilities were observed for Cl/F, 47.5%, and
for V/F, 134% and r in the residual error model was
estimated to be 55%. The goodness-of-fit plots of that
model were satisfactory.
The standard error of estimation expressed as
coefficient of variations (CV) were lower than 25%
for all population parameters. The model with a
correlation (estimated to be 0.22) between Cl/F and
V/F had the same likelihood than the model without
correlation but a higher AIC. This correlation was
therefore not kept in the model and the random effects
were assumed to be independent. The inter-individual
variability of Cl/F and V/F was estimated globally from
all patients with and without ritonavir. When we
considered different variances for inter-individual vari-
ability in patients with and without ritonavir, that
model had a higher AIC than the model with only 1
variance.
So the basic model for indinavir concentrations was
described by a one-compartment model with first-order
absorption and first-order elimination with random
effects on Cl/F and V/F and with a different Ka and
Cl/F in patients who took ritonavir.
Covariates model building
From that basic model, we first tested the effects of the
covariates on the individual estimates of the random
effects. Significant effects of AZT, D4T, and gender on gCl

Ó 2005 Blackwell Publishing Fundamental & Clinical Pharmacology 19 (2005) 373–383

378 K. Brendel et al.

(P ¼ 0.011, 0.011 and 0.049, respectively) and of AZT estimated to be 19.53 L and increased by 1.60 with
on gV (P ¼ 0.038) were found. AZT. The inter-individual variability was 39% for
The model with the lowest AIC (732.1) had AZT, D4T clearance and 113% for volume of distribution. They
and gender effects on Cl/F and AZT effect on V/F. The were slightly decreased from the basic model by the
equations are: incorporation of the covariates. Residual variability
was 47%.
Ka;i ¼ ðKa
ð1  ritonaviri Þ þ Ka;rito
ritonaviri Þ
Inter-occasion variability
Cl=Fi ¼ ðCl=F
ð1  ritonaviri Þ þ Cl=Frito
ritonaviri Þ
We then built the model of inter-occasion variability

ðbAZT
bD4T
bmale Þ
expðgCl;i Þ based on the model at V1 with covariates and with the
data of the second visit V2. The best model had an IOV on
V=Fi ¼ ðV=F
aAZT Þ
expðgV;i Þ:
both Cl/F and V/F (AIC ¼ 1035.6). Performing back-
When a patient receives zidovudine (AZTi ¼ 1), his ward elimination of the covariates from that model, the
clearance is multiplied by a factor bAZT and his volume best AIC was obtained when the effects of only AZT
by a factor aAZT; when a patient receives stavudine and gender on Cl/F were kept. Elimination of one of
(D4Ti ¼ 1), his clearance is multiplied by a factor bD4T; these covariates decreased significantly the likelihood
when a patient is a male (genderi ¼ 1), his clearance is (P < 0.001 for gender, P ¼ 0.01 for AZT).
multiplied by a factor bmale. The equations of that final model are:
The population parameters of this model and their
Ka;i ¼ ðKa
ð1  ritonaviri Þ þ Ka;rito
ritonaviri Þ
standard error are given in the first column of Table II.
Absorption rate constant was estimated to be 0.27/h
Cl=Fi ¼ ðCl=F
ð1  ritonaviri Þ þ Cl=Frito
ritonaviri Þ
for patients who received ritonavir and 0.49/h for
patients without ritonavir. Oral clearance was 23.43
ðbAZT
bmale Þ
expðg0Cl;i þ IOV
g1Cl;i Þ
and 30.33 L/h in women with and without ritonavir,
respectively. It increased by 1.47-fold in men, V=Fi ¼ V=F
expðg0V;i þ IOV
g1V;i Þ
decreased by 1.22-fold in patients who received AZT
where IOV ¼ 0 for data at V1 and IOV ¼ 1 for data at
and increased by 0.68-fold in patients who received
V 2.
D4T. Oral volume of distribution without AZT was
Table II shows the parameters estimated for this final
model with IOV. The same estimations were approxi-
Table II Population pharmacokinetic parameters of indinavir mately observed for the three fixed effects. Using equa-
(estimate and CV of estimation) for the intermediate model with tions (1) and (2) in the appendix, the inter-individual
data at V1 only and for the final model with data at V1 and V2. and inter-occasion variability were derived from the
V1 only V1 and V2
estimated variances of g0i and g1i provided by WinNon-
Mix. They were 42 and 58% for Cl/F, 110 and 205% for
Parameters Estimate CV (%) Estimate CV (%) V/F. The total variability was 72% for Cl/F and 232% for
Ka (/h) 0.49 6 0.43 5 V/F. r in the error model was estimated to be 47%. We
Ka,rito (/h) 0.27 5 0.25 9 tried to perform the estimation using the FOCE method of
Cl/F (L/h) 30.33 15 33.41 15 WinNonMix for this final model but the run did not
Cl/Frito (L/h) 23.43 17 19.35 18 converge.
V/F (L) 19.53 18 24.21 15
The goodness-of-fit plots of this final model with IOV
bmale 1.47 11 1.45 15
bD4T 0.68 14 – –
and covariates was satisfactory. The evaluation proce-
bAZT 1.22 12 1.18 11 dure was applied to the 27 patients receiving 800 mg
aAZT 1.60 20 – – t.i.d. alone and the eight patients receiving 800 mg b.i.d.
x0Cl (%) 39 30a 36 31a with ritonavir. The simulated 10th, 90th percentiles
x1Cl (%) – – 82 53a and the median are displayed in Figure 2, together with
a
x0V (%) 113 2 110 34a all the observed concentrations of these patients. The
x1V (%) – – 285 72a
evaluation results of this indinavir population model
r (%) 47 22a 47 24a
were very satisfactory both when indinavir is adminis-
a
CV for x2V , x2Cl and r2. trated alone or with ritonavir.

Ó 2005 Blackwell Publishing Fundamental & Clinical Pharmacology 19 (2005) 373–383

Population pharmacokinetics of indinavir 379

(a) (a)
18 3
16
2.5
14

Trough concentration (mg/L)

12
2
10

8 1.5
6
1
4

2
0.5
0
0 2 4 6 8 10 12 14 16
0
Time (h)
Women Women Men Men
(b)
18 without AZT with AZT without AZT with AZT
16
(b) 3
14

12
2.5

Trough concentration (mg/L)

10

8
2
6

4 1.5
2

0
1
0 2 4 6 8 10 12 14 16
Time (h) 0.5

Figure 2 Indinavir model evaluation: comparison between the 0

10th, 90th (dashed lines) and 50th (continued line) predicted Women Women Men Men
percentiles for the 1000 simulations and the observed concentra- without AZT with AZT without AZT with AZT

tions at V1 and V2 for the patients receiving the corresponding

dosage in each group for patients with 800 mg t.i.d. of indinavir
alone (a) and for patient with 800 mg b.i.d. of indinavir associated
with a booster dose of ritonavir (b).
Simulation of indinavir trough concentrations
Figure 3 displays the steady-state trough concentrations,
simulated with this final model and the total variability
for gCl and gV, in the four groups of patients as defined
by the covariates included in the final model: male or
female with or without AZT.
The simulations were done with the two most usual
indinavir dosages: 800 mg t.i.d. alone or 800 mg b.i.d.
with 100 mg b.i.d. of ritonavir. The trough concentra-
tions were the highest for women who did not take AZT
and the lowest for men who took AZT, both with and
without ritonavir. When the dosage regimen of indinavir
is 800 mg b.i.d. with ritonavir, trough concentrations
were higher than for 800 mg t.i.d. of indinavir alone.
However, in both cases the variability was very large.
Median trough concentrations obtained with 800 mg
of indinavir with 100 mg ritonavir b.i.d., showed
an increase with AZT compared to without AZT by
1.05-fold in women and by 1.36-fold in men. AZT had
Figure 3 Box plots of trough indinavir concentrations simulated
from the population model for 800 mg of indinavir t.i.d. without
ritonavir (a) and for 800 mg b.i.d. of indinavir with 100 mg of
ritonavir (b). The influence of gender and of the co-administration
of zidovudine (AZT) which were the two significant covariates in
the population analysis is illustrated. The boxes represent 25th to
75th percentiles, with the 50th percentile shown within the boxes;
the 10th and 90th percentiles are shown as capped bars.
therefore an effect with a rather small clinical rele-
vance. With respect to gender effect, median trough
concentrations increased by 1.47-fold in women com-
pared with men without AZT and by 1.83-fold with
AZT. Gender effect on trough concentrations had
therefore a clear clinical relevance. For instance,
median trough concentrations (25th and 75th percen-
tiles) was 1.22 mg/L (0.81, 1.75) in women receiving
800 mg indinavir plus 100 mg ritonavir with AZT,
which is very high.
DISCUSSION
The pharmacokinetics of indinavir was described by a
one-compartment model with first-order absorption and

Ó 2005 Blackwell Publishing Fundamental & Clinical Pharmacology 19 (2005) 373–383

380
first-order elimination with random effects on Cl/F and
on V/F but not on Ka when the effect of ritonavir was
incorporated. Although the mechanism of interaction
between ritonavir and indinavir is well known [28–30],
the effect of ritonavir was tested for the first time in a
pharmacokinetic model of indinavir. We found that
when patients receive ritonavir, they have an absorption
rate constant lower than patients without ritonavir
(0.25 and 0.43/h, respectively). With respect to elimin-
ation, the oral clearance of indinavir was found to
decrease when patients take ritonavir; this was expected
as ritonavir is a CYP3A4 inhibitor. The absence of
significant effect on V/F found here suggests that there is
no effect of ritonavir on the bioavailability of the drug
which is also supported by the absence of correlation
found between Cl/F and V/F.
In this study, a gender effect was found on the oral
clearance of indinavir which was estimated to increase
by 1.45-fold in men. To further study whether this
gender difference was due to a weight effect or was a real
gender difference in metabolism, we also tested the
weight effect on the individual indinavir clearances. We
studied both weight as a continuous covariate or as a
categorical covariate with a cut off of 67.5 kg (the mean
between median weight in men, 75 kg, and in women,
60 kg). By using a Mann–Whitney non-parametric test
or a correlation test, no significant weight effect was
found on Cl/F. This strongly suggests that the gender
effect is more the result of the difference of metabolism
between men and women than a weight effect. Few
other studies also reported similar results. Csaika et al.
[31], in a study with 239 HIV-infected patients, found
that indinavir clearance increased by 1.64-fold in men
very close to the increase of 1.45 found here. In a recent
in vivo and in vitro animal study [32], indinavir
clearance was found to increase by 2.17-fold in male
rats. Consistent with the in vivo observations, hepatic
microsomes from male rats had a substantially higher
metabolizing activity toward indinavir than that from
females. Cytochrome 3A isoforms may contribute to the
observed gender-related differences in indinavir metabo-
lism in rat [32]. These results are in contradiction with
the US package insert, where AUC was found to be lower
in 10 HIV seropositive women who received 800 mg
indinavir t.i.d. in addition to zidovudine and lamivudine
compared with a pool of historical control data in men.
But we think that the results found here and by Csajka
et al. [31] in a prospective study are more reliable. The
clinical relevance of this gender effect was illustrated in
the simulation of the trough levels for the two most
Ó 2005 Blackwell Publishing Fundamental & Clinical Pharmacology 19 (2005) 373–383
K. Brendel et al.
frequent regimens in this trial. In these simulations, the
impact of gender appeared especially important when
patients also took zidovudine. However, it should be
noticed that in this trial only three women received
zidovudine and this result should be further confirmed.
The high trough levels predicted with 800 mg b.i.d. and
ritonavir might explain the high level of nephrolithiasis
found in the BEST study [15]. This result is important
and suggests to decrease indinavir initial dosage in
women compared with men.
With respect to the effect of co-administration of
zidovudine, Cl/F was estimated to increase by 1.18-fold
in patients who received AZT. The P-value was signifi-
cant, but the clinical relevance of this effect was low in
the simulations of trough concentrations. A pharmaco-
kinetic interaction between indinavir and AZT has been
reported in the US package insert of indinavir because
in a study, after 1 week of co-administration of AZT,
indinavir clearance was found to decrease. In this paper,
we found an opposite interaction with a slight increase of
indinavir clearance.
The individual pharmacokinetic parameters of a lot of
drugs can change between two periods of time [33–35].
Some of these variations are due to the fact that
pharmacokinetic parameters are linked with covariates,
which can vary physiologically with time, as serum
creatinine. However, the biggest variability is not fore-
seeable and might be due to changes in bioavailability.
Modeling the inter-occasion variability (also called intra-
individual variability) is very important in population
pharmacokinetics [36], and this is why patients were
sampled at two occasions (V1 and V2) in the COPHAR
1-ANRS 102 trial. A population model including inter-
occasion variability has never been implemented in the
software WinNonMix before this study, and we had to
develop a method to do it. The quality control procedure
implemented before the trial was very important to
minimize the inter-laboratory variability (below 20%).
We did not test the assay as a covariate in the model,
because assays differed mostly in their LOQ and only two
concentrations below the same LOQ (0.05 mg/L) were
found.
In this study, we found both large inter and intra-
patient variabilities of indinavir pharmacokinetic param-
eters. These variabilities estimated here confirm the need
of drug monitoring of indinavir in patients [37]. The
potential benefit of therapeutic drug monitoring (TDM)
of PIs has been shown by the ATHENA study based
on concentration ratio [38]. This ratio is obtained by
dividing the observed concentration by the correspond-
Population pharmacokinetics of indinavir
ing time-adjusted value of a reference pharmacokinetic
curve. More precisely, this reference curve is based on
the fit of a pharmacokinetic model to the median values
of concentrations grouped by 30-min time intervals. In
the ATHENA study, after 1 year of follow up of 55 HIV-
infected patients, significantly fewer patients in the TDM
group had discontinued indinavir than in the control
group (25% vs. 48%) because of virological failure or
indinavir toxicity. Also the TDM group showed a
significantly higher proportion of patients (78% vs.
55%) with a viral load below 500 copies/mL after
12 months of treatment. The present prospective trial,
performed in patients with a sustained virological
response, could be used to define more realistic reference
concentrations. The reference curve should be derived
from the mean population PK parameters using the
model at steady state. For 800 mg of indinavir without
ritonavir and for 800 mg b.i.d. of indinavir with ritona-
vir (100 mg b.i.d.), the most frequent regimen now-
adays, the reference curve is similar to the median curve
displayed in Figure 2. Furthermore, the population
pharmacokinetic parameter estimates obtained here
allow for the adaptative adjustment of indinavir regimen
based on Bayesian therapeutic drug monitoring.
ACKNOWLEDGEMENTS
This study was financially supported by the Agence
Nationale de la Recherche sur le SIDA (ANRS), Paris,
France.
We gratefully acknowledge all the members of the
scientific committee of the Cophar 1-ANRS 102 trial: Dr V.
Calvez (Hôpital Pitié-Salpétrière, Paris), Pr G. Chêne
(INSERM U330, Bordeaux) Pr B. Diquet (Hôpital Pitié-
Salpétrière, Paris), Pr C. Katlama (Hôpital Pitié-
Salpétrière, Paris), Pr C. Leport (Hôpital Bichat, Paris),
Dr A. Metro (ANRS, Paris), Dr G. Peytavin (Hôpital Bichat,
Paris), Pr F. Raffi (Hôpital Hôtel Dieu, Nantes), Dr A. Roux
(Hôpital Ambroise Paré, Boulogne), and Pr D. Salmon-
Ceron (Hôpital Cochin, Paris). We would like to thank the
investigators in the clinical centres for including the
patients at Pr J.M. Estavoyer, Pr R. Laurent (Besançon), Pr
X. Bazin (Caen), Pr C. Perrone (Garches), Pr J.F. Delfraissy
(Kremlin Bicêtre), Pr X. Gallais, Pr J.A. Gastaut (Marseille),
Pr J. Reynes (Montpellier), Pr F. Raffi (Nantes), Pr E.
Bouvet, Pr F. Bricaire, Pr S. Herson, Pr W. Rozenbaum, Pr
D. Séréni, Pr D. Sicard, Dr A. Simon, Pr J.L. Vildé (Paris), Pr
X. Lang (Strasbourg) and the pharmacological laborator-
ies for performing drug assays at Dr B. Royer, Dr P. Muret
(Besançon), Dr M. Tod (Bobigny), Dr A.M. Taburet
Ó 2005 Blackwell Publishing Fundamental & Clinical Pharmacology 19 (2005) 373–383
381
(Kremlin Bicêtre), Dr C. Solas (Marseille), Dr D. Hillaire
(Montpellier), Dr E. Dailly (Nantes), Pr B. Diquet, Dr G.
Peytavin, Dr J.M. Poirier, Dr E. Rey, Dr H. Sauvageon
(Paris), Dr J.C. Alvarez (Versailles). We also thank the
patients for their participation.
REFERENCES
1 Palella F.J., Delaney K.M., Moorman A.C. et al. Declining
morbidity and mortality among patients with advanced human
immunodeficiency virus infection. N. Engl. J. Med. (1998) 338
853–860.
2 Jordan R., Gold L., Cummins C., Hyde C. Systematic review and
meta-analysis of evidence for increasing numbers of drugs in
antiretroviral combination therapy. Br. Med. J. (2002) 324 1–10.
3 Yeni P., Hammer S.M., Carpenter C.J. et al. Antiretroviral
treatment for adult HIV infection: updated recommendations of
the International AIDS Society-USA Panel. JAMA (2002) 288
222–235.
4 Lichterfeld M., Nischalke H.D., Bergmann F. et al. Long-term
efficacy and safety of ritonavir/indinavir at 400/400 mg twice
a day in combination with two nucleoside reverse transcriptase
inhibitors as first line antiretroviral therapy. HIV Med. (2002) 3
37–43.
5 Polis M.A., Sidorov I.A., Yoder C. et al. Correlation between
reduction in plasma HIV-1 RNA concentration one week after
start of antiretroviral treatment and long-term efficacy. Lancet
(2001) 358 1760–1765.
6 Condra J.H., Petropoulos C.J., Ziermann R., Schleif W.A.,
Shivaprakash M., Emini E.A. Drug resistance and predicted
virologic responses to human immunodeficiency virus type 1
protease inhibitor therapy. J. Infect. Dis. (2000) 182 758–765.
7 Salzberger B., Rockstroh J., Wieland U. et al. Clinical efficacy of
protease inhibitor based antiretroviral combination therapy – a
prospective cohort study. Eur. J. Med. Res. (1999) 4 449–455.
8 Launay O., Gerard L., Morand-Joubert L. et al. Nevirapine or
lamivudine plus stavudine and indinavir: examples of 2-class
vs. 3-class regimen for the treatment of human immunodefi-
ciency virus type I. Clin. Infect. Dis. (2002) 35 1096–1105.
9 Duval X., Peytavin G., Albert I. et al. Determination of indinavir
and nelfinavir trough plasma concentration efficacy thresholds
according to virological response in HIV-infected patients.
HIV Med. (2004) 5 307–313.
10 Dieleman J.P., Gyssens I.C., Van der Ende M.E., De Marie S.,
Burger D.M. Urological complaints in relation to indinavir
plasma concentrations in HIV-infected patients participating in
the ADAM study. AIDS (2000) 14 59–98.
11 Reijers M.H., Weigel H.M., Hart A.A. et al. Toxicity and drug
exposure in a quadruple drug regimen in HIV-1 infected
patients participating in the ADAM study. AIDS (2000) 14
59–68.
12 Von Moltke L.L., Greenblatt D.J., Grassi J.M., Granda B.W.,
Duan S.X., Fogelman S.M. Protease inhibitors as inhibitors of
human cytochromes P450: high risk associated with ritonavir.
J. Clin. Pharmacol. (1998) 38 106–111.
382
13 Hsu A., Granneman G.R., Cao G., Carothers L., Japour A.,
Shourbagy T. Pharmacokinetic interaction between ritonavir
and indinavir in healthy volunteers. Antimicrob. Agents
Chemother. (1998) 42 2784–2791.
14 Saah A., Winchell G., Nessly M., Seniouk E., Rhodes R.
Pharmacokinetic profile and tolerability of indinavir–ritonavir
combinations in healthy volunteers. Antimicrob. Agents
Chemoter. (2001) 45 2710–2721.
15 Arnaiz J.A., Mallolas J., Podzamczer D. et al. Continued
indinavir vs. switching to indinavir/ritonavir in HIV-infected
patients with suppressed viral load. AIDS (2001) 17 831–
840.
16 Zhou X.J., Havlir D.V., Richman D.D. et al. Plasma population
pharmacokinetics and penetration into cerebrospinal fluid of
indinavir in combination with zidovudine and lamivudine in
HIV-1-infected patients. AIDS (2000) 14 2869–2876.
17 Letendre S.L., Capparelli E.V., Ellis R.J., McCutchan J.A.
Indinavir population pharmacokinetics in plasma and cere-
brospinal fluid. Antimicrob. Agents Chemother. (2000) 44
2173–2175.
18 Martin C., Sonnerborg A., Olof Svensson J., Stahle L. Indinavir-
based treatment of HIV-1 infected patients: efficacy in the
central nervous system. AIDS (1999) 13 1227–1232.
19 Pfister M., Labbé L., Hammer S.M. et al. Population pharma-
cokinetics and pharmacodynamics of efavirenz, nelfinavir,
and indinavir: Adult AIDS Clinical Trial Group Study 398.
Antimicrob. Agents Chemoter. (2003) 47 130–137.
20 Carrieri P., Cailleton V., Le Moing V. et al. The dynamic of
adherence to highly active antiretroviral therapy: results from
the French National APROCO Cohort. J. Acq. Immun. Def.
Synd. (2001) 28 232–239.
21 Aymard G., Legrand M., Trichereau N., Diquet B. Determin-
ation of twelve antiretroviral agents in human plasma sample
using reverse-phase high-performance liquid chromatography.
J. Chromatogr. B (2000) 744 227–240.
22 Proust V., Toth K., Hulin A., Taburet A.M., Gimenez F., Singlas
E. Simultaneous high-performance liquid chromatographic
determination of the antiretroviral agents amprenavir, nelfin-
avir, ritonavir, saquinavir, delavirdine and efavirenz in human
plasma. J. Chromatogr. B. (2000) 742 453–458.
23 Dailly E., Thomas L., Kergueris M.F., Jolliet P., Bourin M. High-
performance liquid chromatographic assay to determine the
plasma levels of HIV-protease inhibitors (amprenavir, nelfina-
vir, ritonavir, saquinavir) and the non-nucleoside reverse
transcriptase inhibitor (nevirapine) after liquid. J. Chromatogr.
B. (2001) 758 129–35.
24 Poirier J.M., Robidou P., Jaillon P. Determination of indinavir in
plasma by solid-phase extraction and column liquid chromato-
graphy. Ther. Drug Monit. (1999) 21 404–410.
25 Bouley M., Briere C., Padoin C., Petitjean O., Tod M. Sensitive
and rapid method for the simultaneous quantification of the
HIV-protease inhibitors indinavir, nelfinavir, ritonavir and
saquinavir in human plasma by reversed-phase liquid chro-
matography. Ther. Drug Monit. (2001) 23 56–60.
26 The United States Pharmacopoeia XXII: validation of compen-
dial methods. USP Convention Inc., Rockville, 1990, pp. 1710.
Ó 2005 Blackwell Publishing Fundamental & Clinical Pharmacology 19 (2005) 373–383
K. Brendel et al.
27 Beal S.L. Ways to fit a PK model with some data below the
quantification limit. J. Pharmacokinet. Pharmacodyn. (2001)
28 481–504.
28 Lee C.G., Gottesman M.M., Cardarelli C.O. et al. HIV-1 protease
inhibitors are substrates for the MDR1 multidrug transporter.
Biochemistry (1998) 37 3594–3601.
29 Washington C.B., Duran G.E., Man M.C., Sikic B.I., Blaschke T.F.
Interaction of anti-HIV protease inhibitor with the multidrug
transporter P-glycoprotein (P-gp) in human cultured cells. J. Acq.
Immun. Def. Synd. Hum. Retrovirol. (1998) 19 203–209.
30 Raffi F. Resistance to protease inhibitors: the role of the MDR
gene. Presse Med. (2000) 29 2036–2037.
31 Csaika C., Marzolini C., Fattinger K. et al. Population pharma-
cokinetics and effects of indinavir in HIV infected patients. Clin.
Pharmacol. Ther. (2003) 73 69.
32 Lin J.H., Chiba M., Chen I.W., Nishime J.A., Vastag K.J. Sex
dependant pharmacokinetics of indinavir: in vivo and in vitro
evidence. Drug Metab. Dispos. (1996) 24 1298–1306.
33 De Maat M.M., Huitema A.D., Mulder J.W., Meenhorst P.L.,
Van Gorp E.C., Beijnen J.H. Population pharmacokinetics of
nevirapine in an unselected cohort of HIV-1-infected individ-
uals. Br. J. Clin. Pharmacol. (2002) 54 378–385.
34 Huitema A.D., Mathot R.A., Tibben M.M., Schellens J.H.,
Rodenhuis S., Beijnen J.H. Population pharmacokinetics of
thioTEPA and its active metabolite TEPA in patients undergoing
high-dose chemotherapy. Br. J. Clin. Pharmacol. (2001) 51
61–70.
35 Frame B., Koup J., Miller R., Lalonde R. Population pharma-
cokinetics of clinafloxacin in healthy volunteers and patients
with infections: experience with heterogeneous pharmaco-
kinetic data. Clin. Pharmacokinet. (2001) 40 307–315.
36 Karlsson M.O., Sheiner L.B. The importance of modeling
interoccasion variability in population pharmacokinetic
analyses. J. Pharmacokinet. Biopharm. (1993) 21 735–750.
37 Back D., Gatti G., Fletcher C. et al. Therapeutic drug monitoring
in HIV infection: current status and future direction. AIDS
(2002) 16 S5–S37.
38 Burger D., Hugen P., Reiss P. et al. Therapeutic drug monit-
oring of nelfinavir and indinavir in treatment-naı̈ve HIV-1-
infected individuals. AIDS (2003) 17 1157–1165.
APPENDIX
The following model was implemented for incorporation
of IOV in WinNonMix, using IOV as a covariate:
bi ¼ g0i þ IOV
g1i
where IOV ¼ 0 for occasion 1 (V1) and IOV ¼ 1 for
occasion 2 (V2).
Therefore,
g0i ¼ gi þ j1i
g0i þ g1i ¼ gi þ j2i
Population pharmacokinetics of indinavir 383

and g0i and g1i are correlated. We estimated Var(g0i ), To derive the variance of inter-individual variability
Var(g1i ) and the covariance between g0i and g1i with we used:
WinNonMix.
Then to derive the variance of inter-occasion variab- 2g0i þ g1i ¼ 2gi þ j1i þ j2i
ility from these values, we used: so that
g1i ¼ j2i  j1i 4Varðg0i Þ þ Varðg1i Þ þ 4Covðg0i ; g1i Þ
so that ¼ 4Varðgi Þ þ Varðg1i Þ:

Varðg1i Þ ¼ 2
Varðjki Þ: The inter-individual variability is therefore:

The inter-occasion variability is therefore: Varðgi Þ ¼ Varðg0i Þ þ Covðg0i ; g1i Þ: ð2Þ

Varðjki Þ ¼ 1=2 Varðg1i Þ: ð1Þ Total variance was estimated by Var(gi) + Var(jki ).

Ó 2005 Blackwell Publishing Fundamental & Clinical Pharmacology 19 (2005) 373–383