Pharmacokinetics of single-agent axitinib across ... - Springer Link

6 downloads 24 Views 726KB Size Report
Oct 22, 2014 - metastatic renal cell carcinoma (mRCC) to compare axi- tinib pharmacokinetics across different tumor types. Results Axitinib disposition based ...
Cancer Chemother Pharmacol (2014) 74:1279–1289 DOI 10.1007/s00280-014-2606-6

ORIGINAL ARTICLE

Pharmacokinetics of single-agent axitinib across multiple solid tumor types Michael A. Tortorici · Ezra E. W. Cohen · Yazdi K. Pithavala · May Garrett · Ana Ruiz-Garcia · Sinil Kim · John P. Fruehauf 

Received: 18 February 2014 / Accepted: 7 October 2014 / Published online: 22 October 2014 © Springer-Verlag Berlin Heidelberg 2014

Abstract  Purpose  Axitinib, a potent and selective inhibitor of vascular endothelial growth factor receptors, showed antitumor activity as a single agent against several solid tumor types in Phase II and III trials. This study was conducted to evaluate axitinib pharmacokinetics across a variety of solid tumors. Methods  The current study analyzed the pharmacokinetics of axitinib in 110 patients with non-small cell lung cancer (NSCLC), thyroid cancer, or melanoma from three Phase II trials plus 127 healthy volunteers, using nonlinear mixed-effects modeling. Boxplots of maximum observed plasma concentration (Cmax) and area under the plasma concentration–time curve (AUC) of data from these tumor populations was compared to Cmax and AUC from the final population pharmacokinetic model developed for

Michael A. Tortorici, May Garrett, and Sinil Kim were employed at Pfizer Inc during the time of this study and development of the manuscript.

metastatic renal cell carcinoma (mRCC) to compare axitinib pharmacokinetics across different tumor types. Results  Axitinib disposition based on data from 237 subjects was best described using a two-compartment model with first-order absorption and lag time. Population estimates for systemic clearance, central volume of distribution, absorption rate constant, absolute bioavailability, and lag time were 20.1 L/h, 56.2 L, 1.26/h−1, 0.663, and 0.448 h, respectively. Statistically significant covariates included gender on clearance, and body weight on central volume of distribution. However, predicted changes due to gender and body weight were found not clinically meaningful. The final analysis indicated that the pharmacokinetic model for mRCC was able to successfully describe axitinib pharmacokinetics in patients with NSCLC, thyroid cancer, and melanoma. Conclusion  The pharmacokinetics of axitinib appears to be similar across a variety of tumor types. Keywords  Axitinib · Population pharmacokinetics · Non-small cell lung cancer · Thyroid cancer · Melanoma

Electronic supplementary material  The online version of this article (doi:10.1007/s00280-014-2606-6) contains supplementary material, which is available to authorized users. M. A. Tortorici · Y. K. Pithavala · A. Ruiz-Garcia  Department of Clinical Pharmacology, Pfizer Inc, San Diego, CA, USA M. A. Tortorici (*)  Clinical Pharmacology and Pharmacometrics, CSL Behring Biotherapies for Life™, 1020 First Avenue, King of Prussia, PA 19406, USA e-mail: [email protected]

M. Garrett  Pfizer Global Pharmacometrics, Pfizer Inc, San Diego, CA, USA S. Kim  Clinical Oncology, Pfizer Inc, San Diego, CA, USA J. P. Fruehauf  Department of Medicine, Pharmaceutical Sciences, Biological Chemistry, and Biomedical Engineering, University of California-Irvine, Orange, CA, USA

E. E. W. Cohen  Moores Cancer Center, University of California, San Diego, La Jolla, CA, USA

13

1280

Introduction Axitinib is a potent and selective inhibitor of vascular endothelial growth factor (VEGF) receptors 1, 2, and 3 [1–3]. Axitinib blocked VEGF-mediated endothelial cell survival, tube formation, and downstream signaling pathways in vitro and inhibited angiogenesis and tumor growth in preclinical animal models [1]. In Phase II clinical trials, this antiangiogenic agent showed antitumor activity with acceptable safety profile as a single agent against several different solid tumor types, including metastatic renal cell carcinoma (mRCC), non-small cell lung cancer (NSCLC), thyroid cancer, and melanoma [4–8]. In a randomized pivotal Phase III clinical trial (AXIS) [9] of axitinib versus sorafenib in patients with previously treated mRCC, axitinib demonstrated significantly improved progressionfree survival compared with sorafenib, leading to approval of axitinib as second-line therapy for mRCC in the USA, Europe, Japan, and elsewhere. Pharmacokinetic evaluation of axitinib in a Phase I dose-finding study in patients with advanced solid tumors showed that axitinib plasma concentrations peaked within 2–6 h after oral dosing in the fed state and declined with an effective plasma half-life [10] of 2–5 h [11]. Maximum plasma concentration (Cmax) and area under the plasma concentration–time curve were linear at doses tested [up to 20 mg twice daily (BID)], and there was a negligible accumulation following multiple dosing [11]. The maximumtolerated dose was determined to be 5 mg BID, which was recommended as a starting dose for subsequent clinical trials of axitinib. Individualized stepwise dose increases to 7 mg BID, and then to a maximum of 10 mg BID, or decreases to 3 mg BID, and then to 2 mg BID were permitted based on patient tolerability to the drug [12]. Axitinib pharmacokinetics has been studied extensively in over 500 healthy volunteers of different ethnic groups [13–16]. Inter-individual variability (IIV) in axitinib plasma exposure (area under the plasma concentration–time curve) has been observed in both healthy subjects and cancer patients (i.e., 9–62 % coefficient of variation [CV] in healthy volunteers after a single 5-mg dose of axitinib [14, 16] and 39–94 % in patients following multiple doses of axitinib 5 mg BID [11]). In the Phase I dose-finding study [11], dose-limiting toxicities, such as hypertension, were observed at higher doses of axitinib, indicating a potential exposure– toxicity relationship. There have been some reports of an association between pharmacokinetics and clinical outcome and/or toxicities for other tyrosine kinase inhibitors [17, 18]. Therefore, it is important to elucidate factors that may contribute to the IIV of axitinib pharmacokinetics to minimize its toxicities while maximizing the clinical benefits. A base model (defined as the structural model prior to covariate analysis, including the effect of food and

13

Cancer Chemother Pharmacol (2014) 74:1279–1289

formulation) to describe the pharmacokinetics of axitinib has previously been developed using a nonlinear mixedeffects modeling approach based on pharmacokinetic data from 337 healthy volunteers from 10 Phase I studies [19]. The model was then applied to analyze data from patients with mRCC across three Phase II studies [20]. Although single-agent axitinib has also showed potential clinical benefits [objective response rate (defined as the percent of patients with confirmed complete or partial response according to Response Evaluation Criteria In Solid Tumors) 9–30 %; stable disease lasting longer than 16 weeks 19–38 %] in the treatment of NSCLC [7], thyroid cancer [6], and melanoma [8], the pharmacokinetics of axitinib has not been previously reported in these patient populations. The objectives of the current analysis were to (1) develop a model that describes the pharmacokinetics of axitinib following multiple oral dose administrations in patients with advanced NSCLC, advanced thyroid cancer, and metastatic melanoma, (2) identify covariates that are important determinants of pharmacokinetic variability in these patients, and (3) evaluate axitinib pharmacokinetics across different tumor types by comparing the results of the current population pharmacokinetic analysis with that reported previously for the mRCC population [20].

Patients and methods Study design and subjects This pooled population pharmacokinetic analysis was based on data from three Phase II clinical trials of singleagent axitinib in patients with NSCLC, thyroid cancer, or melanoma combined with four Phase I studies in healthy volunteers (Online Resource 1). The four Phase I studies in healthy volunteers included in this analysis were a subset of those used in the previous population pharmacokinetic analysis [19, 20] and were selected since these studies were conducted using the axitinib form IV, the same crystal polymorph used in the three Phase II patient studies. Clinical details of these studies in NSCLC, thyroid cancer, or melanoma patients have been described previously [6–8]. The results of axitinib population pharmacokinetic analysis obtained using data from three Phase II studies in patients with mRCC have been reported previously [20]. Study treatment and pharmacokinetic sampling In all studies (studies 1–7, summarized in Online Resource 1), axitinib was administered as crystal polymorph form IV film-coated immediate-release tablets. In healthy volunteers (studies 1–4), blood samples were collected, typically at 0 (pre-dose), 1, 1.5, 2, 3, 4, 6, 8, 12, 16, 24, 36, and 48 h

Cancer Chemother Pharmacol (2014) 74:1279–1289

after oral administration of a single 5-mg dose of axitinib in the fasted state only (studies 1, 4) or in both the fasted and fed state (studies 2, 3). In one study (study 3), subjects also received 1 mg axitinib intravenously in a cross-over fashion to determine absolute oral bioavailability. In cancer patients (study 5 in NSCLC, study 6 in thyroid cancer, and study 7 in melanoma), axitinib was administered at a starting dose of 5 mg BID in the fasted state (with food in the NSCLC study). In the NSCLC study, the dose could be increased in 2-mg increments up to a maximum of 10 mg BID in patients who had no treatment-related grade >2 adverse events (AE) according to Common Terminology Criteria for Adverse Events version 3.0 and no hypertension during any 2-week period in cycle 1 or 2. Hypertension was defined as at least 2 in-clinic readings of systolic blood pressure >150 mmHg or diastolic blood pressure >90 mmHg, separated by at least 1 h. In the thyroid cancer and melanoma studies, dose could be increased by 20 % if patients did not have treatment-related grade >1 AE (the thyroid cancer study) or grade ≥2 AE (the melanoma study) for 8 weeks, unless the patient was responding to the treatment. Patients experiencing treatment-related grade ≥2 AEs that could not be controlled with supportive treatment had their dose temporarily interrupted and restarted at either the same dose (for grade 2 AE) or 20 % lower (for grade ≥3 AE) dose after resolution. Blood samples for pharmacokinetic analysis were collected from patients within 15 min prior to the morning dose and 1–2 h after dosing on days 1 (not in the NSCLC study), 29, and 57 and every 8 weeks thereafter. Plasma samples were analyzed for axitinib concentrations using a validated high-performance liquid chromatography with tandem mass spectrometric detection method (Charles River Laboratories; Shrewsbury, MA, USA) [11]. All axitinib-treated subjects with drug concentrations and information on dosing and collection time available from at least one post-dose visit were included in the analysis (Online Resource 1). Axitinib pharmacokinetic model development A strategy used for building an axitinib population pharmacokinetic model involved development of a base model and assessment of random effects, followed by evaluation and identification of covariates to be retained in the final model [19]. The first-order estimation method was used for initial model development to gain insight into initial estimates. The first-order conditional estimation method with interaction was used for base model development and covariate testing. All analyses were performed using a nonlinear mixed-effects modeling (NONMEM® VI, level 1.2; ICON Development Solutions: Hanover, MD, USA). The base model was developed using a two-compartment model defined in terms of systemic clearance (CL), central

1281

volume of distribution (Vc), peripheral volume of distribution (Vp), inter-compartmental clearance (Q), absorption rate constant (ka), absolute bioavailability (F), and lag time for absorption (tlag). The IIV in CL, Vc, and ka was modeled using multiplicative exponential random effect of the form,

θi = θ · eηi where θi is the individual value of the population parameter θ, and ηi is the inter-individual random effect assumed to have a mean of zero and a variance ω2. Residual variability was modeled using the log-transformed error model of the form,     ln Yij = ln Fij + εij where Yij denotes the observed concentration for the ith individual at time j and Fij denotes the corresponding predicted concentration. εij is the intra-individual residual error with a mean of zero and a variance σ2. Attempts were made to define a full variance–covariance matrix omega, Ω, for the inter-individual random effects (ω) when possible. Following validation of the base model and estimation of individual model parameters, relationships between covariates and IIV in CL and Vc were explored graphically. Covariates were selected based on scientific and/or clinical interest and prior knowledge, and included gender, race, tumor type, smoking status [a known inducer of cytochrome P450 (CYP)1A2, which is involved in axitinib metabolism], body weight, and age on CL and gender and body weight on Vc. Covariates were added to the base model in a stepwise manner. The effect of a categorical covariate x (gender, race, tumor type, and smoking status) with n groups was modeled as,

Covθ = θ0 · (1 + θx · x) where θ0 denotes the population value of the parameter and θx denotes the fractional change in θ0 for a group of the population within each category. Continuous covariates (age and body weight) were modeled as multiplicative effects of the form,

Covθ = θ0 · (x/xmedian )θx where θ0 denotes the population value of the parameter when x  =  xmedian and θ denotes the population value conditional on the value x. When θx = 1, θ is proportional to x. For testing of covariates, the stepwise covariate model building tool, Perl-speaks-NONMEM (PsN® version 2.3.1; Karlsson M, Jonsson N, and Hooker A; http://psn.sourceforge.net/) was used, which implemented the forward selection followed by the backward elimination of each covariate to the model [21]. Each relevant parameter–covariate relationship was prepared and evaluated in a univariate manner. Two hierarchical models were compared by the chi-square test of the difference in their respective objective

13

1282

function values (OFV), with the number of degrees of freedom (df) equal to the difference in the number of parameters between two models. A covariate was retained in the forward selection if OFV decreased by at least 3.84 points (P 10.83 points (P 15 %), and, therefore, they were kept in the final dataset. Covariates and final model Following development of the base model and assessment of IIV and residual variability, the effect of additional covariates (body weight, age, smoking status, gender, race, and tumor type) on parameter estimates was first assessed graphically. Visual inspection of Eta plots of parameters from the base model versus each covariate revealed

possible correlations between CL and body weight, age, smoking status, gender, race, or tumor type; and between Vc and body weight or gender (Online Resource 5). These covariates were then tested sequentially in the forward selection followed by backward elimination process, resulting in gender on CL and weight on Vc being statistically significant and retained in the final model. Estimates for CL, Vc, ka, Q, Vp, F, and tlag in the final model (20.1 L/h, 56.2 L, 1.26 h−1, 2.00 L/h, 63.3 L, 0.663, and 0.448 h, respectively) were comparable to those from the base model (Table 1). The IIV in CL, Vc, and ka for the final model were 47.9, 16.9, and 87.4 %, respectively. The η-shrinkage was found to be adequate (8.6 % for ηCL, 24 % for ηVc, and 23 % for ηka). Of the covariates examined, gender was found to significantly affect CL, whereas body weight substantially impacted Vc and was retained in the final model. Thus, the model predicted a 35 % lower CL in female versus male subjects, leading to a proportionately greater plasma exposure in females. Axitinib Vc was affected by body weight according to the following relationship,

Vc = Vc0 · (Wgt/80)0.933 where Wgt denotes body weight, and 80 is the median body weight for the populations studied. For the range of body weights observed in this pooled analysis (43.5–143 kg), the estimated lower and upper end of Vc would be 31.8 and 96.6 L, respectively. The η-shrinkage for CL, Vc, and ka (all