ORIGINAL RESEARCH ARTICLE Pooled Population Pharmacokinetic Analysis of Phase I, II and III Studies of Linifanib in Cancer Patients Ahmed Hamed Salem • Denise Koenig • Dawn Carlson Published online: 4 December 2013 Ó Springer International Publishing Switzerland 2013 Abstract Background and Objective Linifanib is a multi-targeted receptor tyrosine kinase inhibitor, which can inhibit members of the vascular endothelial growth factor and platelet-derived growth factor receptor families. The objective of this analysis was to characterize the population pharmacokinetics of linifanib in cancer patients. Methods We pooled 7,351 linifanib plasma concentra- tions from 1,010 cancer patients enrolled in 13 clinical studies. Population pharmacokinetic modelling was per- formed using NONMEM version 7.2. The covariates that were screened included the cancer type, co-medications, creatinine clearance, formulation, fed status, liver function markers (bilirubin, blood urea nitrogen [BUN], aspartate aminotransferase [AST], alanine aminotransferase [ALT]), albumin, age, sex, race, body weight, surface area and body mass index. Results A two-compartment model with first-order absorption and disposition best described linifanib phar- macokinetics. An increase in body weight was associated with less than proportional increases in volumes of distri- bution. Subjects with hepatocellular carcinoma and renal cell carcinoma were estimated to have 63 and 86 % larger volumes of distribution, respectively, than subjects with the other cancer types. Females had 25 % slower oral clear- ance (CL/F) than males, while subjects with colorectal cancer had 41 % faster CL/F than other subjects. For linifanib bioavailability, subjects with refractory acute myeloid leukaemia or myelodysplastic syndrome had 43 % lower bioavailability, evening doses were associated with 27 % lower bioavailability than morning doses, and administration of linifanib under fed conditions decreased the bioavailability by 14 %. Finally, the oral solution for- mulation showed two-fold faster absorption than the tablet formulations. Conclusion The use of mixed-effects modelling allowed robust assessment of the impact of the concomitant effects of body size, different cancer types, formulation, diurnal variation, sex and food on linifanib pharmacokinetics. The developed population pharmacokinetic model describes linifanib concentrations adequately and can be used to conduct simulations or to evaluate the linifanib exposure– response relationship. 1 Introduction Angiogenesis is essential for tumour growth and metasta- sis. Lack of adequate vasculature results in tumours becoming necrotic or apoptotic, and restrict the tumour size [1, 2]. Angiogenesis involves an imbalance between anti- angiogenic and proangiogenic molecules such as vascular endothelial growth factor (VEGF) secreted from tumour cells [3]. In addition, platelet-derived growth factor (PDGF) stimulates tumour growth and enhances angio- genesis by facilitating pericyte coverage of new micro- vessels [4, 5]. Thus, inhibition of VEGF and PDGF receptors (VEGFR and PDGFR) has been a compelling target for cancer therapy [6, 7]. Results from clinical studies of bevacizumab, a selective monoclonal antibody against VEGFR, validated the A. H. Salem (&) D. Koenig D. Carlson Clinical Development, AbbVie Inc., 1 North Waukegan Road, AP13A-3, Dept. R4PK, North Chicago, IL 60064, USA e-mail: [email protected]A. H. Salem Department of Clinical Pharmacy, Faculty of Pharmacy, Ain Shams University, Cairo, Egypt Clin Pharmacokinet (2014) 53:347–359 DOI 10.1007/s40262-013-0121-2
Transcript
ORIGINAL RESEARCH ARTICLE
Pooled Population Pharmacokinetic Analysis of Phase I, II and IIIStudies of Linifanib in Cancer Patients
Ahmed Hamed Salem • Denise Koenig • Dawn Carlson
Published online: 4 December 2013
� Springer International Publishing Switzerland 2013
Abstract
Background and Objective Linifanib is a multi-targeted
receptor tyrosine kinase inhibitor, which can inhibit
members of the vascular endothelial growth factor and
platelet-derived growth factor receptor families. The
objective of this analysis was to characterize the population
pharmacokinetics of linifanib in cancer patients.
Methods We pooled 7,351 linifanib plasma concentra-
tions from 1,010 cancer patients enrolled in 13 clinical
studies. Population pharmacokinetic modelling was per-
formed using NONMEM version 7.2. The covariates that
were screened included the cancer type, co-medications,
creatinine clearance, formulation, fed status, liver function
relative standard error, Vc/F apparent volume of the central compartment, Vp/F apparent volume of the peripheral compartmenta Based on 854/1,000 successful runs
352 A. H. Salem et al.
with unphysiological estimates of Vp/F and Q/F and hence
were not included in the model. Estimating the full
covariance matrix between CL/F, Vc/F and ka was
attempted; however, the covariance parameters were esti-
mated with poor precision. Therefore, covariance between
only CL/F and Vc/F was included throughout model
development, with their correlation estimated to be 0.52 in
the base model.
3.2 Covariate Model
Covariates identified by the univariate stepwise forward-
inclusion procedure as significant determinants of linifanib
pharmacokinetic parameters were as follows: sex, cancer
type (colorectal cancer [CRC]) and cytarabine co-admin-
istration as covariates on CL/F, WT, cancer type (HCC and
RCC) as covariates on Vc/F and Vp/F, formulation effect on
ka and cancer type (refractory acute myeloid leukaemia
[AML] or myelodysplastic syndrome [MDS]), time of dose
administration and fed status on bioavailability. No other
evaluated covariates were found to significantly improve
the goodness of fit. Removal of the cytarabine effect on
CL/F in the stepwise backward elimination was associated
with a nonsignificant increase in OFV (p value [0.05);
hence, the cytarabine effect was removed from the model.
After reaching the final model, we tested the need for
estimating the exponent of the allometric model for WT on
Vc/F and Vp/F to ensure model parsimony. Fixing the
exponents to 1, as suggested by some researchers, resulted
in a significant increase in the OFV (p value \0.001);
hence, estimating the exponent was deemed appropriate.
3.3 Final Model
The final model parameter estimates and the precision
associated with their estimation are shown in Table 3. Both
fixed and random effects were precisely estimated, with a
relative standard error (RSE) of 30 % or less. In the final
model, an increase in WT was associated with a less than
proportional increase in Vc/F and Vp/F. Subjects with HCC
and RCC were estimated to have 63 and 86 % larger vol-
umes of distribution, respectively, than subjects with the
other cancer types. With regard to CL/F, females had 25 %
slower clearance than males. Furthermore, subjects with
colorectal cancer had 41 % faster CL/F than non-colorectal
cancer subjects.
For linifanib bioavailability, subjects with relapsed/
refractory AML/MDS had 43 % lower bioavailability, and
evening doses were associated with 27 % lower bioavail-
ability than morning doses, while administration of linifa-
nib under fed conditions decreased the bioavailability by
14 %. Finally, the oral solution formulation showed
approximately two-fold faster absorption than the tablet
formulations.
Relative to the base model, the covariates explained
27, 9 and 29 % of the variability of Vc/F, CL/F and the
correlation between Vc/F, CL/F, respectively. The
effects of some covariates on linifanib CL/F and
Vc/F are shown in Fig. 1. Figure 2 demonstrates the
effect of sex on CL/F and the lack of association
between body weight and CL/F.
The final equations for the typical values (TVs) of the
structural model parameters are presented in Eqs. 7, 8, 9
and 10:
TVCL=F ¼ 2:82� hCRC � hSEX ð7Þ
where:
hCRC ¼1:41; colorectal cancer
1; else
�;
hSEX ¼0:75; female
1; male
�
TVVc=F ¼ 50:75� WT=70ð Þ0:52�hHRC ð8Þ
where:
hHRC ¼1:63;1:86;
1;
hepatocellular carcinoma
renal cell carcinoma
else
8<:
TVka ¼ 0:46� hSOL ð9Þ
where:
hSOL ¼1:97; oral solution
1; tablets
�
TVF ¼ hAMPM � hFCOND � hAML MDS ð10Þ
where:
hAMPM ¼0:73; evening dose
1; morning dose
�
hFCOND ¼0:86; non-fasting
1; else
�
hAML MDS ¼0:57; Refractory AML/MDS cancer
1; else
�
The goodness of fit for the final model was evaluated
graphically (Figs. 3, 4). The individual and population
predicted linifanib concentrations versus the observed
concentrations were randomly distributed across the line
of unity, indicating that the model adequately described the
observations over the entire linifanib concentration range.
The conditional weighted residuals plots showed
symmetrical distribution and no time- or concentration-
related trends.
PopPK Analysis of Linifanib in Cancer Patients 353
Fig. 1 Effects of cancer type,
body weight and sex on
linifanib pharmacokinetic
parameters. The boxes represent
the 25th, 50th and 75th
percentiles; the whiskers
represent the lowest datum still
within 1.5 interquartile ranges
(IQRs) of the lower quartile, and
the highest datum still within
1.5 IQRs of the upper quartile
range; the bullets represents
outliers. AML acute
myelogenous leukaemia, CL/F
apparent clearance from the
central compartment, CRC
colorectal cancer, HCC
hepatocellular carcinoma, MDS
myelodysplastic syndrome,
NSCLC non-small cell lung
cancer, RCC renal cell
carcinoma, Vc/F apparent
volume of the central
compartment
Fig. 2 Relationships between
sex and body weight versus
ETA on apparent clearance
from the central compartment
(CL/F) in the base and final
models. The boxes represent the
25th, 50th and 75th percentiles;
the whiskers represent the
lowest datum still within 1.5
interquartile ranges (IQRs) of
the lower quartile and the
highest datum still within 1.5
IQRs of the upper quartile
range; the bullets represents
outliers
354 A. H. Salem et al.
Fig. 3 Diagnostic plots of the
final model
Fig. 4 Population predicted
linifanib concentration–time
after last dose profiles for all
studies. The solid lines ? error
bars denote the medians and
90 % percentiles of the
population predicted linifanib
concentrations, and the circles
denote the observed linifanib
concentrations
PopPK Analysis of Linifanib in Cancer Patients 355
In order to confirm the stability of the model preci-
sion of the estimated pharmacokinetic parameters, a
non-parametric bootstrap analysis was performed, and
85 % of the bootstrap replicates converged successfully.
In accordance with the estimated standard error of
estimate (SEE) values for pharmacokinetic parameters in
the linifanib pharmacokinetic model, the bootstrap
showed narrow confidence intervals for all parameters.
The median and the 5th and 95th percentiles of the
parameter estimates from the fit of the final model to the
bootstrap samples are shown in Table 3. The asymptotic
estimates obtained from the original dataset showed
close agreement with the median and were all included
within the 5th to the 95th percentiles of the bootstrap-
ping values, indicating model stability. None of the
95 % confidence intervals for the parameters from the
bootstrap datasets included zero, confirming the robust-
ness of the parameters.
In the VPC plots shown in Fig. 5, the 5th, 50th and 95th
percentiles of the prediction-corrected observations were in
close agreement with the confidence intervals of the 5th,
50th and 95th percentiles of the prediction-corrected sim-
ulated data, indicating a robust predictive ability of the
model to describe linifanib concentrations. Only 4 % of the
data were below the 5th percentile, and 5 % of the data
were above the 95th percentile of the predictions.
4 Discussion
Using non-linear mixed-effects analysis, linifanib concen-
tration–time data from 13 clinical studies were integrated
across the different development phases in order to char-
acterize the effect of covariates on linifanib pharmacoki-
netics. The developed model showed high parameter
estimate precision, as well as good predictability and hence
will be of immense value in predicting linifanib pharma-
cokinetics in various cancer patient populations.
The final structural pharmacokinetic model was a two-
compartment model with first-order absorption and elimi-
nation. Compared with the one-compartment model, the
use of a two-compartment model in our analysis was
associated with a better fit to individual profiles and a
larger reduction in the objective function value. A previous
analysis based on a phase II study used a one-compartment
model for linifanib in subjects with NSCLC [23]. The
sparse data included in the previous analysis may not have
allowed characterization of the biphasic disposition of
linifanib.
The population estimate of linifanib CL/F in the final
model was 2.8 L/h. Previous pharmacokinetic assessments
of individual studies have shown similar estimates. In a
phase I study in refractory solid tumours, linifanib
CL/F was 2.7 (±1.2) L/h [17]. In phase II studies, linifanib
Fig. 5 Prediction and variance-
corrected visual predictive
check for the final model. The
circles denote the prediction-
corrected observations; the lines
denote the 5th, 50th, and 95th
percentiles of the prediction-
corrected observed data; the
shaded areas denote the
confidence intervals of the 5th,
50th, and 95th percentiles of the
prediction-corrected simulated
data
356 A. H. Salem et al.
CL/F ranged from 3 to 3.9 L/h [23, 24]. The steady-state
volume of distribution (Vss/F) of linifanib was estimated to
be 61.1 L. This may indicate extensive distribution of
linifanib in the body, which is consistent with its high
lipophilicity (Log D of 4.2 at pH of 7.4) [12]. The beta-
phase half-life of 17.2 h is supportive of the once-daily
dosing used throughout development.
We explored the effect of body size measures on
linifanib pharmacokinetic parameters. No association was
found between the body size measures and linifanib CL/F,
despite the wide body weight range of the subjects included
in the analysis (Table 1). This corroborates the switch of
linifanib dosing from body weight-guided dosing in phase I
and II studies to fixed dosing in the phase III study. On the
other hand, Vc/F and Vp/F were associated with body
weight, and the data supported the estimation of the
exponent of the allometric size model. Subjects with WT
that is 10 and 20 % larger than the population typical body
weight of 70 kg are expected to have 5 and 10 % larger
Vss/F values than the population typical estimate.
The final model also estimated that morning dosing was
associated with 27 % greater exposure than evening dos-
ing. This diurnal variation has been demonstrated previ-
ously in a phase I study, which assessed the impact of the
dose time on linifanib pharmacokinetics [21]. In order to
minimize variability and improve the tolerability of
linifanib, the hepatocellular carcinoma subjects enrolled in
the phase III study were instructed to administer the doses
in the evening.
Diurnal variation in pharmacokinetics, often referred to
as chronopharmacokinetics, is exhibited by several orally
administered anti-cancer drugs; such as 6-mercaptopurine,
busulfan and tegafur/uracil [25–29]. Moreover, some
anticancer drugs, such as fluorouracil and doxorubicin, also
show diurnal variation in their plasma levels although they
are administered by constant continuous intravenous infu-
sion [30–33]. It is hypothesized that such diurnal variations
are often masked by the high inter-individual variability
observed in cancer patients [34]. Diurnal variation has been
also reported with drugs from other classes [35, 36].
The lower linifanib exposure in the evening could be
explained by the reduction in the gastric emptying rate in
the evening, due to reduced enterokinesis [37]. The slower
gastric emptying rate in the evening was deemed respon-
sible for lower peak concentrations and longer tmax values
for many lipophilic drugs [37]. Another possible mecha-
nism is the circadian rhythm in the activity of metabolizing
enzymes. Animal studies have shown several-fold greater
activity in some microsomal oxidases during the dark span
than in the light span [38–40]. Previous population phar-
macokinetic analyses reported strong associations between
microsomal oxidases activity and the clearance of another
anticancer drug, gefitinib, with 60 % of the variability of
unbound gefitinib plasma concentrations being explained
by individual cytochrome P450 3A4 activity [41, 42].
This population analysis showed that the cancer type
was an important determinant of linifanib pharmacokinet-
ics. According to the final model, subjects with colorectal
cancer have 41 % faster linifanib oral clearance. Colorectal
cancer subjects included in the analysis were co-medicated
with mFOLFOX6 (oxaliplatin [85 mg/m2], folinic acid
[400 mg/m2] and 5 fluorouracil [400 mg/m2 IV bolus]) on
the first day of each cycle. In a lead-in cohort in this
population, linifanib pharmacokinetics was evaluated with
and without mFOLFOX6. Therefore, we explored the
possibility of a drug interaction between linifanib and
mFOLFOX6 being responsible for the increase in CL/F in
this population. However, mFOLFOX6 showed no effect
on linifanib pharmacokinetics. Furthermore, our analysis
demonstrated 43 % lesser bioavailability in subjects with
relapsed or refractory AML or MDS than in those with
other cancer types. The lesser bioavailability estimated in
this patient population is consistent with the reportedly
greater CL/F estimates in this population (4.1–6.9 L/h)
[20] than in patients with other cancer types (2.7–3.9 L/h)
[17, 23, 24]. We initially attempted to include the relapsed/
refractory AML/MDS cancer type as a covariate on CL/F;
however, including it on bioavailability provided a much
larger OFV reduction. The formulation and assay were
precluded as sources of the lesser bioavailability in this
population, since the same formulation and assay were
used in studies conducted in other populations. We
hypothesize that this reduction in oral bioavailability is due
to cytotoxic therapy-induced malabsorption. The subjects
with relapsed/refractory AML or MDS who were included
in our analysis had been heavily pretreated with cytotoxic
therapy. Cytotoxic therapy affects the dividing cells of the
gastrointestinal mucosa and has been reported to cause
intestinal epithelial damage in AML subjects [43, 44].
Intestinal injury induced by cytotoxic therapy has been
linked to reduced absorption and lesser bioavailability of
several agents, such as acyclovir, ciprofloxacin and
D-xylose [45–47].
Another difference in pharmacokinetics between cancer
types was the higher Vss/F value in subjects with HCC and
RCC. This could be related to fluid retention and increased
fluid volume during hepatic and renal impairment. In
addition, linifanib is highly protein bound ([99 %), and
drugs that are strongly bound to plasma constituents (e.g.
phenytoin, diazepam) are known to demonstrate an
increased volume of distribution in patients with liver or
kidney disease, which is due to lower plasma binding [48].
In order to investigate whether the increase in linifanib
Vd/F in subjects with HCC and RCC could be explained by
change in protein binding in these populations, we explored
the use of albumin concentrations as a covariate in the
PopPK Analysis of Linifanib in Cancer Patients 357
model. However, no relationship was shown between
albumin concentrations and linifanib Vc/F or CL/F. In
addition, plasma protein binding data from 13 subjects with
HCC (eight Child–Pugh class A and five Child–Pugh
class B) showed that [99.7 % of linifanib was protein
bound in plasma, which was similar to values in subjects
with normal hepatic function [49]. Therefore, we do not
believe that the larger Vd/F in HCC and RCC subjects is
linked to a decrease in protein binding.
An oral solution formulation was used initially during
linifanib clinical development, before tablet formulations
were developed. In alignment with the results of compar-
ative bioavailability studies, our population analysis
showed no difference in linifanib bioavailability among the
formulations. Nevertheless, the analysis showed a differ-
ence in the absorption rate constant between the formula-
tions, with the liquid formulation having two-fold faster
absorption than the tablet formulation. We attempted to
estimate different inter-individual variability for ka of the
oral solution compared with the tablet formulations; how-
ever, a high shrinkage ([70 %) in the inter-individual
variability of ka of the oral solution was observed. A model
with only one random effect for ka for all formulations
decreased the shrinkage, and hence the same inter-indi-
vidual variability was estimated for ka regardless of the
formulation.
Our analysis showed a sex effect on linifanib CL/F, with
males having 25 % faster CL/F than females. This differ-
ence is similar to that observed with another tyrosine
kinase inhibitor, erlotinib, where female subjects achieved
greater exposure (25–43 %) than male subjects [50]. Faster
clearance in male subjects for some drugs is often sec-
ondary to the body weight influence on clearance. In an
attempt to investigate whether the sex effect seen on
linifanib CL/F is due to a body weight effect, we explored
the relationship between linifanib CL/F and body weight.
However, no association was found between CL/F and
body weight (Fig. 2). Numerous clinical pharmacokinetic
studies have demonstrated sex differences in drug absorp-
tion and bioavailability for certain drugs [51]. Neverthe-
less, we do not believe that this is the underlying
mechanism of the sex effect on linifanib CL/F, since there
was no difference in Vd/F between males and females. We
hypothesize that sex-based differences in drug metabolism
play a role in the faster clearance observed in males.
In several clinical studies, linifanib was co-administered
with other anti-cancer agents, such as cytarabine, paclit-
axel, carboplatin and mFolfox6. Our analysis suggests that
these drugs do not affect linifanib pharmacokinetics. This
may be because of the multiple pathways involved in
linifanib metabolism. In addition, we found no difference
in pharmacokinetic parameters between Asians, who were
well represented in our dataset, and non-Asian populations.
Finally, there was no association between linifanib
CL/F and CLCR. This was anticipated, since urinary elimina-
tion is a minor pathway in linifanib pharmacokinetics, with
less than 5 % of the linifanib dose being recovered in urine
as the unchanged drug and metabolite [20].
5 Conclusion
The use of mixed-effects modelling allowed robust
assessment of the impact of the concomitant effects of
body size, different cancer types, formulation, diurnal
variation, sex and food. The developed population phar-
macokinetic model describes linifanib concentrations ade-
quately and can be used to conduct simulations or to
evaluate the linifanib exposure–response relationship.
Conflicts of Interest This study was sponsored by AbbVie Inc.
AbbVie Inc. contributed to the study design; research; data interpre-
tation; and writing, review and approval of the manuscript for pub-
lication. Ahmed Hamed Salem, Denise Koenig and Dawn Carlson
are employees of AbbVie Inc.
References
1. Carmeliet P, Jain RK. Angiogenesis in cancer and other diseases.
Nature. 2000;407(6801):249–57.
2. Zetter P, Bruce R. Angiogenesis and tumor metastasis. Annu Rev
