Clinical Implications of the Pharmacokinetics of Crizotinib in Populations of Patients with Non–Small Cell Lung Cancer

Drug pharmacokinetics can be affected by many factors, such as age, sex, race, and hepatic or renal impairment. As a result of variations in drug pharmacokinetics and exposure, the recommended drug dose determined in clinical trials may need to be adjusted for specific populations of patients. Crizotinib (XALKORI) is a selective small-molecule inhibitor of the receptor tyrosine kinases anaplastic lymphoma kinase (ALK), hepatocyte growth factor receptor (HGFR/MET), ROS1, and Recepteur d'Origine Nantais (RON; refs. 1–3). In 2011 crizotinib was granted accelerated approval by the FDA for the treatment of ALK-positive advanced non–small cell lung cancer (NSCLC), with full approval granted once the results were obtained from a randomized controlled phase III trial (PROFILE 1007, NCT00932893; ref. 4). Crizotinib has since been approved for the treatment of ALK-positive advanced NSCLC in many countries and regions around the world.

As a targeted therapy, crizotinib has marked antitumor activity in patients with ALK-positive advanced NSCLC. Objective response rates of 53% and 61% were achieved in the multinational phase I (PROFILE 1001, NCT00585195) and phase II (PROFILE 1005, NCT00932451) trials, respectively (5, 6). Progression-free survival (PFS) was significantly longer with crizotinib than standard chemotherapy, both in PROFILE 1007 in previously treated patients with ALK-positive advanced NSCLC (median 7.7 vs. 3.0 months; P < 0.001; ref. 4) and in the phase III PROFILE 1014 trial in patients with untreated ALK-positive advanced NSCLC (median 10.9 vs. 7.0 months; P < 0.001; ref. 7).

The recommended starting dose of crizotinib for patients with ALK-positive advanced NSCLC is 250 mg twice daily. Pharmacokinetic analyses have shown crizotinib to be orally bioavailable, with a median time to maximal plasma concentration (T_max) of approximately 4 hours and mean apparent terminal half-life of 42 hours in patients with cancer after a single 250-mg oral dose (8, 9). Crizotinib can be taken with or without food, as crizotinib administered with a high-fat meal reduced the area under the concentration–time curve (AUC) extrapolated to infinity (AUC_inf) and maximal plasma concentration (C_max) by 14% (9, 10).

Patient characteristics such as body weight have the potential to influence drug metabolism (11). In a phase I study of crizotinib, higher crizotinib exposure was observed in Asian patients than non-Asian patients, and an evaluation of body weight–adjusted pharmacokinetic data suggested that body weight may have contributed to these results (8). However, based on the safety and efficacy of crizotinib observed in this study, no adjustment of the approved crizotinib dose was considered necessary for Asian patients (8). Hepatic and renal functions also have the potential to influence drug metabolism. Crizotinib is extensively metabolized in the liver by cytochrome P450 (CYP) 3A4/5 (12), and, as such, crizotinib metabolism and pharmacokinetics may be affected in patients with impaired hepatic function. While mild and moderate renal impairment were not found to have clinically relevant effects on crizotinib exposure, crizotinib AUC_inf and C_max were found to increase by 79% and 34%, respectively, in patients with severe renal impairment compared with those with normal renal function (10).

We developed and validated a robust population pharmacokinetics model (mathematical and statistical) to assess whether patient characteristics such as age, body weight, race, sex, renal function, or liver function influence crizotinib pharmacokinetics parameters such as oral clearance (CL/F) and AUC at steady state (AUC_ss) following multiple doses of crizotinib administered at 250 mg twice daily. We also considered the clinical implications of the data on this dosing regimen for these specific patient populations.

Pharmacokinetic data were combined from the PROFILE 1001 (5), PROFILE 1005 (6), and PROFILE 1007 (4) trials (Supplementary Table) of crizotinib in patients with ALK-positive advanced NSCLC and other solid tumors. The patients in PROFILE 1005 had received one or more prior therapies for metastatic disease and in PROFILE 1007 had received one prior therapy for metastatic disease that was platinum-based. Only patients with a crizotinib starting dose of 250 mg twice daily and available plasma crizotinib concentration data were included in these analyses.

A mathematical and statistical model was developed prospectively in three stages: (i) initial development of a base model, (ii) inclusion of selected covariates into a full model, and (iii) removal of covariates that were not statistically significantly associated with pharmacokinetics to establish a final population pharmacokinetic model (Fig. 1). In a previous noncompartmental analysis, reduced elimination of crizotinib was observed over time in patients given multiple daily doses of crizotinib (8), which may have been related to the known autoinhibition of CYP3A by crizotinib. An approach was taken to characterize this nonlinear time-dependent elimination in the base model. Additional details are provided in the Supplementary Material.

In the base model, interpatient variability in the pharmacokinetic parameters was modeled using multiplicative exponential random effects of the form |${\theta _i} = \theta \cdot {e^{{\eta _i}}}$|⁠, where θ was a reference (typical) individual (population mean) value of the parameter and η_i denoted the between-patient random effect accounting for the ith individual's deviation from the reference value, having zero mean and variance ω². A two-compartment model with a first-order absorption and time-dependent CL/F element using a nonlinear function produced a robust fit to the data. This model was selected as the base model on which to build the full model to assess the impact of covariates on pharmacokinetic parameters.

Covariates assessed in the full model were patient characteristics (at baseline) that were predefined on the basis of clinical and regulatory considerations (11). Covariates tested for impact on CL/F were age, body weight, sex, race, Eastern Cooperative Oncology Group (ECOG) performance status (0 or 1 vs. ≥ 2), smoking status (smoker vs. never-smoker and former smoker), creatinine clearance (CLcr), and serum laboratory parameters, that is, total bilirubin, aspartate aminotransferase (AST), and albumin. The impact of age, body weight, sex, race, and albumin on the apparent volume of distribution in the central compartment (V2/F) was also assessed. In both the CL/F and V2/F analyses, race was tested as a dichotomous covariate [Asian vs. non-Asian (which included white, black, Hispanic, or other non-Asian race)].

The final model was developed by applying a backward elimination algorithm to the full model and using likelihood ratio tests based on the change in the minimum objective function (ΔMOF) to assess the significance of covariate(s) removed from the full model. Using a stepwise approach (Fig. 1), covariate(s) giving the smallest nonsignificant (P > 0.001) increase in objective function values (OFV; the sum of squared deviations between the predictions and the observations) were removed from the model, and the process was repeated until all remaining covariates had a statistically significant effect on pharmacokinetic parameters. The final model was obtained when all of the remaining covariates resulted in significant (P ≤ 0.001) likelihood ratios.

Covariates with statistically significant relationships with CL/F were assessed for their impact on crizotinib AUC_ss. Probability density plots were generated using a stratified, nonparametric bootstrapping approach to evaluate the magnitude and precision of statistically significant covariate effects on AUC_ss. The nonlinear mixed effects modeling program NONMEM 7.1.2 was used for the estimation of all models. Effects of a covariate on CL/F or AUC_ss that were between 80% and 125% in magnitude relative to that of the reference patient were not considered to be clinically relevant (consistent with commonly used criteria for bioequivalence; ref. 13). Additional details of the population pharmacokinetic model are provided in the Supplementary Information.

Throughout model building, the "goodness of fit" of different models to the data was evaluated using the following criteria: (i) change in OFV, (ii) inspection of scatter plots and precision of the parameter estimates, and (iii) decreases in both between-patient and residual variability. At all stages of model development, diagnostic plots of observed versus predicted and individual predicted values, weighted residuals versus time, and conditional weighted residuals versus time were examined to assess model adequacy, possible lack of fit, or violation of assumptions.

Box plots of the empirical Bayes predictions of the interpatient random effects were generated to evaluate dose invariance and adequacy of combining studies for the stable final model. ΔMOF statistics and parameter estimates were used to determine the clinical relevance of the covariate effects, and ω² between the models was compared to evaluate the reduction resulting from inclusion of covariate effects.

The final model was evaluated by simulating data with the model (fixed and random effects) and conducting visual predictive checks (VPC). Simulations for VPC were performed using patient characteristics as well as dosing and pharmacokinetic sampling history from the original dataset to summarize concentration–time data using median, low, and high percentiles. The concordance between individual patient observations and simulated values, as well as the distribution of observed and simulated data, were evaluated.

The crizotinib population pharmacokinetic model was developed from a dataset comprising 8,973 pharmacokinetic samples obtained from 1,214 patients who received crizotinib at 250 mg twice daily. Of these patients, 165 (14%) were from the phase I PROFILE 1001 trial, 898 (74%) from the phase II PROFILE 1005 trial, and 151 (12%) from the phase III PROFILE 1007 trial. In this population 1,182 patients (97%) were diagnosed with advanced NSCLC and 32 patients (3%), with other solid tumors. Demographic and major baseline characteristics of this population are summarized in Table 1. Females comprised the majority of the population (56%), and most patients (84%) were aged <65 years. A total of 57% of patients were non-Asian [656 white (including 15 Hispanic), 22 black, and 13 patients of other races] and 43% were Asian (216 Chinese, 124 Japanese, 160 Korean, and 23 patients from other Asian countries). Eighty-five percent of patients had an ECOG performance status of 0 or 1. The reference patient was defined on the basis of the population median as a 65-kg male, white (non-Asian) cancer patient with a baseline CLcr of 91.6 mL/minute and total bilirubin of 0.41 mg/dL.

Moderate interpatient variability was observed in the base model, estimated to be 39.7% for CL/F and 51.9% for V2/F. In general, the predicted values agreed well with those observed in patients following multiple oral doses of crizotinib 250 mg twice daily (Fig. 2), indicating that the model adequately described crizotinib pharmacokinetic data. The final model included the statistically significant covariates for CL/F (Asian race, female sex, body weight, CLcr, and total bilirubin), which reduced CL/F interpatient variability to 34.6% from 39.7% in the base model (Table 2). The final model also included the statistically significant covariates for V2/F (Asian race and female sex), which reduced V2/F interpatient variability to 48.6% from 51.9% in the base model. All parameters were estimated with acceptable precision (relative standard error <30%), with the exception of body weight and total bilirubin for CL/F, for which the relative SE was approximately 33%.

The final estimated equation for the parameter CL/F was as follows:

where sex = 0 for male or 1 for female and race = 0 for non-Asian or 1 for Asian. Using this model, the initial CL/F for the reference patient was estimated to be 136 L/hour on day 1 after a single 250-mg dose of crizotinib. After multiple daily 250-mg doses, CL/F was calculated to decrease to 76 L/hour at steady state on day 28.

The relationships between CL/F and ECOG performance status, AST, albumin, smoking status, and age (19–83 years) were found to be nonsignificant (P > 0.001; Table 3), and these covariates were removed from the final population pharmacokinetic model (Fig. 1). In the population pharmacokinetic model, CL/F was predicted to be 23% lower in Asian patients than non-Asian patients and 11% lower in females than males, irrespective of body weight [both statistically significant (P ≤ 0.001); Table 2]. Body weight (range 33–160 kg) was a statistically significant covariate for CL/F, with patients of lower weight (≤40 kg) having a lower CL/F than patients of higher weight (≥100 kg). As an example, CL/F in a 40-kg patient was 9% lower than in a 65-kg (median-weight) patient; CL/F in a 100-kg patient was 9% higher than in a 65-kg patient (Table 3). CLcr also had a statistically significant effect on crizotinib CL/F: CL/F was 16% lower in patients with a CLcr of 30 mL/minute than in the reference patient with a CLcr of 91.6 mL/minute (Table 2). There was a nonlinear relationship between total bilirubin and CL/F, with a small covariate effect (Table 2).

The final estimated equation for the parameter V2/F was as follows:

Applying this model, the estimated V2/F for the reference patient was 3,520 L. The relationships between V2/F and body weight, albumin, and age were found to be nonsignificant (P > 0.001; Table 3), and these covariates were removed from the model (Fig. 1). In the population pharmacokinetic model, V2/F was 23% smaller in Asian patients than non-Asian patients, and 23% lower in female than male patients (Tables 2 and 3). The estimated absorption rate constant (Ka) for the reference patient was 0.73 L/hour (Table 2).

To evaluate model stability and the CIs of the final parameter estimates, a nonparametric bootstrapping approach was used. The median values from 1,000 bootstrapping analysis runs were similar to the parameter estimates of the original dataset, and the bootstrapped 95% CIs overlapped with those of the original dataset in Table 2, suggesting that the final model was stable.

To illustrate the effects of the significant covariates on CL/F, an alternative representation of the results was produced in which the effects of these covariates on the AUC_ss following multiple dosing of crizotinib 250 mg twice daily were plotted (Figure 3). This analysis showed that AUC_ss values for patients with statistically significant covariates (except Asian race) were predicted to be between 80% and 125% of the AUC_ss of the reference patient. AUC_ss values for Asian patients were predicted to be in the range of 125% to 140% of the reference value.

We have developed a population pharmacokinetic model to explore whether baseline patient and disease characteristics including body weight, age, race, sex, or renal or hepatic function are likely to affect crizotinib pharmacokinetics. Crizotinib CL/F and V2/F were examined as pharmacokinetic parameters with the potential to influence drug exposure (AUC_ss) to provide insight into any potential need for starting-dose modification. The two-compartment model with first-order absorption rate constant and time-dependent decrease in CL/F used in our analysis was found to effectively characterize the disposition of crizotinib. Goodness-of-fit criteria showed that the selected base model was consistent with the observed data with no evidence of systematic bias. In general, individual predicted values agreed well with observed values across the range of observations, and the model showed acceptable descriptive and predictive performance providing confidence in its validity.

The model describing crizotinib pharmacokinetics across a range of populations was developed from a large number of pharmacokinetic samples collected from 1,214 patients enrolled in three crizotinib clinical trials. The dataset included repeated pharmacokinetic samples gathered over several months and samples from many patients with ALK-positive cancer, predominantly advanced NSCLC (97% of patients). The inclusion of these samples provides confidence that the results will be applicable to the broad crizotinib patient population with ALK-positive NSCLC in clinical practice. The model was characterized by moderate interpatient variability in pharmacokinetic parameter estimates, which was sometimes greater than the observed differences between the specific populations and the reference population, as discussed in more detail below.

Age (range, 19–83 years; 16% of patients ≥65 years of age) and ECOG performance status did not statistically significantly influence crizotinib clearance, suggesting that no adjustments are needed to the crizotinib starting dose based on these characteristics. Similarly, smoking status was not found to have a statistically significantly impact on crizotinib clearance.

Mild or moderate renal impairment had a minimal effect on crizotinib CL/F, with crizotinib clearance 16% lower in patients with a CLcr of 30 mL/minute than in the reference patient with a CLcr of 91.6 mL/minute, the population median value. This latter value was consistent with normal renal function (CLcr of ≥90 mL/minute). A decreasing baseline CLcr was associated with increasing values for AUC_ss relative to the reference value. However, for patients with CLcr 30 mL/minute, the AUC_ss was predicted to increase to <125% of the reference value in all circumstances (Fig. 3). On the basis of these results, mild (CLcr ≥60–<90 mL/minute) or moderate (CLcr ≥30–<60 mL/minute) renal impairment is not expected to result in a clinically relevant accumulation of crizotinib, and no adjustment to the starting dose of crizotinib is warranted for these patients (10). Of note, the analysis did not include patients with severe renal impairment (CLcr <30 mL/minute) based on clinical trial enrolment eligibility criteria.

Three laboratory test covariates indicative of hepatic function were assessed: baseline albumin, AST, and total bilirubin levels. AST had no impact on CL/F, and albumin did not affect V2/F. The impact of AST or total bilirubin on V2/F was not explored in this analysis. Total bilirubin had only a minimal effect on crizotinib CL/F. Consistent with these small changes in CL/F, the AUC_ss increased with increasing total bilirubin levels relative to that of the reference patient, but the magnitude was minimal (Fig. 3). Thus, differences in baseline laboratory test values of hepatic function did not appear to have a clinically relevant effect on crizotinib pharmacokinetics, indicating that no adjustment of the crizotinib starting dose is needed for patients with baseline hepatic function within the range observed in the population studied (0.1–2.1 mg/dL for total bilirubin and 7–124 U/L for AST). However, these findings should be considered with caution and must not be extrapolated beyond the population analyzed. The analysis population was limited by the eligibility criteria of the clinical trials, which restricted patients to those with AST levels ≤2.5 × the upper limit of normal (ULN; or AST levels ≤5 × ULN if liver function abnormalities were due to the underlying malignancy) and total bilirubin levels ≤1.5 × ULN (with an exception for patients with documented Gilbert's syndrome in PROFILE 1001). Severe baseline hepatic impairment has not yet been studied formally.

Race and sex were found to have statistically significant effects on crizotinib pharmacokinetics as shown in this population pharmacokinetic model. In the Asian population, both crizotinib CL/F and V2/F were 23% lower than in the non-Asian population (Tables 2 and 3). After multiple 250-mg twice daily daily crizotinib dosing, the AUC_ss value in Asian patients was predicted to increase to 125% to 140% of the reference value, which is outside the 80% to 125% range considered not to be clinically relevant (Fig. 3). Given the differences in pharmacokinetic parameters between Asians and non-Asians, exploratory subgroup analyses within crizotinib clinical trials have been conducted and have found the efficacy and safety of crizotinib in Asian patients to be comparable with that of either non-Asian patients or the all-patient population (5, 14–16). On this basis, no starting dose adjustment is warranted for Asian patients (10).

In female patients, CL/F was 11% lower and V2/F, 23% lower than in male patients (Tables 2 and 3). AUC_ss was predicted to increase, but to <125% of the reference value after multiple 250-mg twice daily doses, suggesting that sex does not affect exposure to crizotinib in a clinically meaningful way (Fig. 3). The statistically significant changes in crizotinib clearance detected in this model therefore do not warrant a starting-dose adjustment based on sex.

Extremes of body weight influenced CL/F and AUC_ss: CL/F increased and the typical AUC_ss decreased with increasing body weight. As a result, crizotinib exposure was higher in patients with low body weight (<40 kg) than in patients with high body weight (>100 kg; Fig. 3). However, no relationship was detected between body weight and V2/F, and no adjustment in starting dose is necessary.

In summary, age, ECOG performance status, AST level, albumin level, and smoking status at baseline had no effect on crizotinib pharmacokinetic parameters based on our population pharmacokinetic analysis. A number of covariates (female sex, Asian race, body weight, CLcr, and total bilirubin on CL/F and Asian race and female sex on V2/F) were identified as having a statistically significant effect on crizotinib pharmacokinetics. Inclusion of these covariates into the final model had little effect on interpatient variability, reducing interpatient variability of CL/F and V2/F by only 5% and 3%, respectively, compared with the base model. Changes in exposure (AUC_ss) as a result of the statistically significant covariates were not considered to be clinically relevant: exposure in patients with any covariate except Asian race was predicted to be between 80% and 125% of that of the reference patient. Although exposure in Asian patients was predicted to be outside of this range, the efficacy and safety of crizotinib in Asian versus non-Asian patients were found to be comparable. The approved 250-mg twice-daily starting dose of crizotinib therefore appears to be appropriate for all patients, irrespective of age, sex, race, body weight (in the range evaluated: 33–160 kg), mild or moderate renal impairment, or hepatic function (in the range evaluated: total bilirubin ≤2.1 mg/dL or AST levels ≤124 U/L).

E. Wang is a director at Pfizer Inc. R. Khosravan is a senior director at Pfizer Inc and has ownership interest (including patents) in Pfizer Stocks. K. Wilner is a senior director at Pfizer. W. Tan is a director at Pfizer. D. J. Nickens is a director at Pfizer Inc. M. Amantea is a senior director at Pfizer Inc. A. Bello is a former senior director at Pfizer Inc. K. Parivar is a vice president at Pfizer Inc.

Conception and design: E. Wang, A. Bello, K. Parivar, W. Tan

Development of methodology: E. Wang, D.J. Nickens, R. Khosravan, M. Amantea, K. Parivar, W. Tan

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): E. Wang

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): E. Wang, D.J. Nickens, A. Bello, R. Khosravan, K. Wilner, K. Parivar

Writing, review, and/or revision of the manuscript: E. Wang, D.J. Nickens, A. Bello, R. Khosravan, M. Amantea, K. Wilner, K. Parivar, W. Tan

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): E. Wang, D.J. Nickens

We would like to thank all of the participating patients and their families, as well as the investigators, research nurses, study coordinators, and operations staff. This study was sponsored by Pfizer Inc. Medical writing support was provided by Susan Hasmall and Wendy Sacks at ACUMED, an Ashfield Company, part of UDG Healthcare plc, and was funded by Pfizer Inc.

This study was sponsored by Pfizer Inc.

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.