Population pharmacokinetic model with time-varying clearance for lorlatinib using pooled data from patients with non-small cell lung cancer and healthy participants.

Lorlatinib, a selective inhibitor of anaplastic lymphoma kinase (ALK) and c-ROS oncogene 1 (ROS1) tyrosine kinase, is indicated for the treatment of ALK-positive metastatic non-small cell lung cancer (NSCLC) following progression on crizotinib and at least one other ALK inhibitor, or alectinib/ceritinib as the first ALK inhibitor therapy for metastatic disease. The population pharmacokinetics (PopPK) of lorlatinib was conducted by nonlinear mixed effects modeling of data from 330 patients with ALK-positive or ROS1-positive NSCLC and 95 healthy participants from six phase I studies in healthy volunteers; demographic, metabolizer phenotype, and patient prognostic factors were evaluated as covariates. Lorlatinib plasma PK was well-characterized by a two-compartment model with sequential zero-order and first-order absorption and a time-varying induction of clearance. Single dose clearance was estimated to be 9.04 L/h. Assuming that the metabolic auto-induction of lorlatinib reaches saturation in ~ 5 half-lives, clearance was estimated to approach a maximum of 14.5 L/h at steady-state after a period of ~ 7.25 days. The volume of distribution of the central compartment was estimated to be 121 L and the first-order absorption rate constant was estimated to be 3.1 h-1 . Baseline albumin and lorlatinib total daily dose were significant covariates on lorlatinib clearance. Use of proton pump inhibitors was found to be a significant covariate on the lorlatinib absorption rate constant. These factors were assessed to have no clinically meaningful impact on lorlatinib plasma exposure, and no dose adjustments are considered necessary based on the examined covariates.

WHAT IS THE CURRENT KNOWLEDGE ON THE TOPIC?

Lorlatinib is an anaplastic lymphoma kinase (ALK)/c‐ROS oncogene 1 (ROS1) receptor tyrosine kinase inhibitor (TKI) indicated for ALK‐positive metastatic non‐small cell lung cancer (NSCLC) after progression on crizotinib and at least one other ALK inhibitor, or alectinib/ceritinib as the first ALK inhibitor therapy for metastatic disease.

WHAT QUESTION DID THIS STUDY ADDRESS?

This analysis characterizes the pharmacokinetics (PK) of lorlatinib in adult patients with ALK/ROS1‐positive NSCLC and healthy participants after single or multiple doses and evaluated whether dose adjustments are necessary based on demographic factors or disease characteristics.

WHAT DOES THIS STUDY ADD TO OUR KNOWLEDGE?

This analysis provides a population PK (PopPK) model that describes the PK of lorlatinib, a drug that undergoes metabolic auto‐induction, and allows for subsequent exposure‐response analyses.

HOW MIGHT THIS CHANGE DRUG DISCOVERY, DEVELOPMENT, AND/OR THERAPEUTICS?

This PopPK model allowed for subsequent exposure‐response analysis of lorlatinib, which, in turn, helped to justify lorlatinib clinical dose selection. Furthermore, the characterization of covariate effects supports lorlatinib dosing guidelines.

INTRODUCTION

Oncogenic fusions of anaplastic lymphoma kinase (ALK) are present in ~ 5% of patients with non‐small cell lung cancer (NSCLC). , Targeted receptor tyrosine kinase inhibitor (TKI) treatment is effective in this population ; however, acquired resistance to first and second generation ALK inhibitors is common. Lorlatinib is a third generation, adenosine triphosphate (ATP)‐competitive inhibitor of ALK, indicated for the treatment of patients with ALK‐positive NSCLC following prior ALK inhibitor therapy. It is active across a range of ALK mutations, including those associated with acquired resistance to first and second generation ALK inhibitors.

The lorlatinib first‐in‐patient phase I/II dose escalation and expansion study B7461001 (NCT01970865) enrolled patients with advanced ALK‐positive or c‐ROS oncogene 1 (ROS1)‐positive NSCLC, with or without asymptomatic central nervous system (CNS) metastases. Based on safety, efficacy, and clinical pharmacology data from the phase I portion of the study, 100 mg once daily (q.d.) lorlatinib was selected as the recommended phase II dose. The efficacy of 100 mg q.d. lorlatinib was established in the phase II portion of the study and was the basis of regulatory approval. ,

In the phase I portion of B7461001, the accumulation after multiple dosing was less than expected based on the plasma half‐life, suggesting that lorlatinib undergoes metabolic auto‐induction. Lorlatinib is metabolized primarily by cytochrome P450 (CYP)3A and UGT1A4, and to a lesser extent by CYP2C8, CYP2C19, CYP3A5, and UGT1A3. In the phase I portion of B7461001, levels of CYP3A4 activity initially increased with time following multiple dosing, as measured by urine 6‐beta‐hydroxy‐cortisol/cortisol and plasma 4‐beta‐hydroxy‐cholesterol/cholesterol ratios, but then gradually reached a plateau. Together, these observations support time‐dependent auto‐induction in lorlatinib clearance. Additional information on lorlatinib pharmacokinetics (PK) came from an open‐label crossover drug interaction study (B7461008) in healthy volunteers, which indicated that co‐administration of the proton‐pump inhibitor (PPI) rabeprazole reduced lorlatinib maximum concentration (Cmax) by ~ 30% but had no effect on lorlatinib area under the curve to infinity (AUCinf). Furthermore, administration of lorlatinib with a high‐fat meal was found not to have a meaningful effect on lorlatinib exposure.

Here, we describe the development of a population PK (PopPK) model characterizing lorlatinib plasma PK based on data from the phase I/II clinical study and six studies in healthy participants. The potential effects of various covariates were assessed, including demographic factors, CYP metabolizer phenotype, measures of hepatic and renal function, formulation, PPI, and food effect.

METHODS

Analysis dataset

Data were pooled from patients with advanced ALK‐positive or ROS1‐positive NSCLC enrolled in Study B7461001, and from healthy subjects enrolled in six phase I studies in healthy participants (Table 1). In Study B7461001, serial PK sampling was conducted following single dosing, and after 15 days of continuous dosing (cycle 1, day 15; patients only); sparse sampling was performed in a limited number of patients after 8 days of continuous dosing (cycle 1, day 8) and on day 1 of cycles 2–5. Primary efficacy and safety data from Study B7461001 have been previously published. , , All trials were conducted in accordance with Good Clinical Practice Guidelines and the ethical principles that have their origin in the Declaration of Helsinki, and all patients provided written informed consent.

Table 1

Summary of study populations included in the PopPK analysis

Study design	Lorlatinib dosing regimen	Number of subjects included in PopPK analysis	Time points of PK sampling
B7461001 (NCT01970865) Phase I/II open‐label multicenter, multiple dose escalation/expansion study in patients with advanced ALK+ or ROS1+ NSCLC	Phase I: 10, 25, 50, 75, 100, 150, 200 mg orally q.d., or 35, 75, or 100 mg orally b.i.d.	54	Day −7: predose, 0.5, 1, 2, 3, 4, 6, 8, 9, 24, 48, 72, 96, and 120 h postdose. Cycle 1 day 1 and cycle 1 day 8: predose, 1 and 4 h postdose. Cycle 1 day 15: predose, 0.5, 1, 2, 3, 4, 6, 8, 9, and 24 h (24‐h sample not required for b.i.d. dosing). Day 1 of cycles 2–5: predose and 1 h postdose
	Phase II: 100 mg q.d.	276	Full PK: Day ­7: predose, 0.5, 1, 2, 3, 4, 6, 8, 9, 24, 48, 72, 96, and 120 h postdose. Cycle 1 day 1 and cycle 1 day 8: predose, 1, 2, and 4 h postdose. Cycle 1 day 15: predose, 0.5, 1, 2, 3, 4, 6, 8, 9, and 24 h. Day 1 of cycles 2–5: predose, 1 and 2 h postdose. Day 1 of cycle 6 and day 1 of every other cycle thereafter: predose Sparse PK: predose on day 1 of cycles 1–5 and predose on cycle 7 day 1, cycle 8 day 1, and cycle 10 day 1
B7461004 (NCT03184168) Open‐label, single dose, single center mass balance study in healthy male participants	Single oral 100 mg dose	6	Predose, 0.5, 1, 2, 4, 6, 8, 12, 24, 48, 72, 96 h, and beyond that, every 24 h postdose until amount of radioactivity recovered in excreta was at least 90% of administered radioactivity or < 1% had been recovered from excreta from 2 consecutive days
B7461005 Phase I, randomized open‐label crossover study to assess relative bioavailability in healthy participants	Lorlatinib 100 mg tablets	20	Predose, 0.25, 0.5, 1, 1.5, 2, 3, 4, 6, and 10 h postdose
B7461007 Phase I, single dose, randomized, open‐label, two‐period, two‐treatment, two‐sequence, crossover study of lorlatinib in healthy participants to assess absolute bioavailability	Treatment A: 50 mg i.v. Treatment B: 100 mg oral tablet	11	Predose, 0.5, 1, 1.5, 2, 4, 6, 12, 24, 48, 72, 96, 120, and 144 h postdose
B7461008 (NCT02569554) Phase I randomized crossover, open‐label, 4‐period study in healthy participants to evaluate effect of rabeprazole and food on lorlatinib PK and to assess bioavailability of oral solution vs. tablet formation	Single oral dose: 100 mg	26	Predose, 0.5, 1, 1.5, 2, 4, 6, 12, 24, 48, 72, and 96 h postdose
B7461011 Phase I, open‐label, two‐period, two‐treatment, fixed sequence, crossover study to estimate the effect of multiple dose rifampin on the single dose PK of lorlatinib in healthy participants	Single oral dose: 100 mg	12	Predose, 0.5, 1, 1.5, 2, 4, 6, 12, 24, 36, 48, 60, 72, 96, and 120 h postdose
B7461016 Phase I, randomized, single dose, open‐label, study to assess lorlatinib bioequivalence in healthy participants under fasted conditions	Single oral dose: 100 mg	20	Predose, 0.5, 1, 1.5, 2, 4, 6, 12, 24, 36, 48, 60, 72, 96, and 120 h postdose

ALK, anaplastic lymphoma kinase; NSCLC, non‐small cell lung cancer; PK, pharmacokinetics; PopPK, population pharmacokinetic; ROS1, c‐ROS oncogene 1.

Modeling: Software and strategy

The analysis was performed using nonlinear mixed effects modeling methodology as implemented in NONMEM version 7.3 (ICON, Dublin, Ireland) using the first order conditional estimation method with interaction (FOCEI). Inspection of the $COV step output at every stage of model development as well as the value of the condition number (ratio of the largest to smallest eigenvalues obtained from the PRINT = E option on $COV) of the correlation matrix was performed to rule out extreme pairwise correlations (p > 0.95) of the parameter estimates.

Potential covariates were screened using visual examination of diagnostic plots and generalized additive modeling (GAM), then underwent stepwise covariate model (SCM) building to obtain a stable final model, including covariates that significantly improved the adequacy of the model. SCM was implemented using Perl‐speaks‐NONMEM (PsN) 4.2.0 followed by a separate NONMEM run, including covariance estimation.

Additional graphical and statistical analysis was done with R version 3.2.2 (R Foundation for Statistical Computing, Vienna, Austria).

Lorlatinib clearance estimation

As PK data corresponded to samples collected after single dose administration, or at steady‐state following auto‐induction, lorlatinib clearance (CL) was estimated using an initial clearance after single dosing (CLI), and subsequently a time‐varying clearance that achieves its maximum value at steady‐state (CLMX) according to the following expression:Where IND is the induction rate constant and t is time postdose.

Random effects model development

Between‐subject variability for CL, volume of distribution of the central compartment (V2), volume of distribution of the peripheral compartment (V3), inter‐compartmental clearance (Q), first order absorption rate constant (ka), and bioavailability (F) were modeled using multiplicative exponential random effects of the form:where θ is the typical individual (population mean) parameter value and θ denotes the between‐subject random effect accounting for the i th patient’s deviation from the typical value having zero mean and variance ω 2. The approximate percent coefficient of variation (%CV) was reported as:

The multivariate vector of between‐subject random effects has a variance‐covariance (Ω) matrix. The base structural model was developed using a diagonal Ω matrix first and a full block (unstructured) ω was explored before and after covariates were included. The most stable Ω structure with successful $COV estimates was selected for the base model.

Residual variability was initially modeled using the combined additive and proportional error model. Simplifications of this model were explored (e.g., proportional error only) when estimates of additive error were small.

Inclusion of covariates and full model development

The covariates assessed in the PopPK analysis are shown in Table 2. These were selected based on scientific rationale and clinical interest. As lorlatinib is metabolized extensively in the liver, metabolizer phenotypes for CYP3A5 and CYP2C19, and baseline bilirubin, baseline alanine aminotransferase (ALT), and baseline hepatic function were investigated as covariates on clearance. As severe renal impairment can also influence the PK of drugs metabolized in the liver, and chronic renal failure can reduce CYP activity, the impact of renal impairment was assessed by the inclusion of baseline standardized creatinine clearance (WNCL) as a covariate on V2, CL, and F. To confirm previously reported findings on the impact of PPI use and food effect on lorlatinib PK, PPI use and food effect were investigated as covariates on ka and F.

Table 2

Evaluated covariates

PK parameters	Covariates
CL	AGE, SEX, PTST, CYP2C19, CYP3A5, CYP2C9, TDOSE, RACE, BALB, BALK, BBIL, BTG, BHGRADE, BRGRADE, WNCL, BALT
V2	AGE, SEX, BRGRADE, BTG, RACE, WNCL
ka	FOOD, PSCI, PPI
F	FOOD, PSCI, PPI, TDOSE, BHGRADE, BRGRADE, WNCL, BALT, CYP2C19, CYP3A5, CYP2C9

BALB, baseline albumin; BALK, baseline alkaline phosphatase; BALT, baseline alanine aminotransferase; BBIL, baseline total bilirubin; BHGRADE, baseline hepatic impairment as assessed by NCI criteria method (normal [A], mild [B1], mild [B2], and moderate [C]); BRGRADE, baseline renal impairment as assessed by Kidney Disease Outcomes Quality Initiative (KDOQI) staging (normal [A], mild [B], moderate [C], severe [D]); BTG, baseline triglycerides; BWT, baseline body weight; CL, clearance (consists of an initial clearance after single dose and a time‐varying induced clearance following multiple dosing); CYP2C19, cytochrome P450 C19 phenotype (poor, intermediate, extensive, or ultra‐metabolizer); CYP3A5, cytochrome P450 3A5 phenotype (poor, intermediate, extensive, or ultra‐metabolizer); F, absolute bioavailability; FOOD, fasted or fed; k a, Rate constant of absorption; PK, pharmacokinetic; PPI, proton pump inhibitor (rabeprazole) co‐administration; PSCI, formulation (acetic acid solvate, free base, or intravenous solution); PTST, healthy participant or patient; RACE, Race (White, Black, Asian, or Other); TDOSE, total daily lorlatinib dose; V2, volume of distribution of central compartment; WNCL, baseline standardized creatinine clearance.

These covariates were subsequently tested for significance in a stepwise manner using the SCM application in PsN with statistical significance criterion of α = 0.05 for the forward inclusion step, which corresponded to a change in objective function value (ΔOFV) of 3.84 based on χ 2 with degrees of freedom (df) = 1. The full model was then subjected to a backward elimination step with a statistical significance criterion of α = 0.01, which corresponded to a ΔOFV of 6.65 based on χ 2 with df = 1.

Categorical covariates were included using a linear function. The covariate parameter structures for continuous covariates (linear and power functions) were tested for all the included covariates in parallel whenever a new covariate was incorporated and was determined based on OFV.

In addition to ΔOFV, the final model was selected after careful review of ω and σ values. Furthermore, prediction‐based, residual‐based, and simulation‐based diagnostics were considered for covariate inclusion in the final model. Empirical Bayes estimate‐based diagnostics were not used due to high parameter shrinkage.

Derivation of standardized creatinine clearance

Baseline creatinine clearance (BCCL) was determined using the Cockcroft‐Gault equation. Because allometric weight scaling was used a priori on the base model, CL, V2, and BCCL all had a dependence on baseline body weight (BWT); to properly assess the relation between BCCL and CL, BCCL was transformed to WNCL as follows :

Final model development

To obtain the most parsimonious and stable final model, the candidate covariate model resulting from the backward elimination step in SCM was subjected to a separate NONMEM run, including covariance estimation excluded in PsN when running SCM. Signs of model overparameterization and poorly estimated parameters were examined.

Outliers

Outliers were identified in both the base and final models using the absolute value of conditional weighted residuals (CWRES) and individual weighted residuals (IWRES) following the criteria |CWRES|>6 and |IWRES|>6. The influence of the set of outliers was evaluated by comparing estimates of the key PK parameters (i.e., the structural and covariate fixed effects on CL), with and without the outliers removed. In addition to outliers identified by CWRES or IWRES, inspection of the dataset for illogical values (e.g., very high predose concentrations, or very low concentrations noted near typical time to maximum concentration [Tmax]) was performed.

Assessment of model adequacy and predictive performance

At all stages of model development (base and final), assessment of model adequacy was conducted through multiple approaches, including ΔOFV, visual inspection of different diagnostic plots, precision of the parameter estimates, as well as decreases in both between‐subject variability and residual variability.

A battery of diagnostic testing was conducted to evaluate the goodness of fit and detect any violation of assumptions. Plots of observed concentrations (OBS) versus population predictions (PRED) and OBS versus individual predictions (IPRED) were evaluated for randomness around the line of unity. Evaluation was also performed on longitudinal profiles of PK concentration to compare observations and predictions. Plots of CWRES versus time and IWRES versus time were evaluated for randomness around the zero line. For parameters with reasonable empirical Bayes estimate for interindividual variability (ETA) shrinkage, distribution of ETAs was checked to ensure normal distribution. In addition, plots of ETAs versus covariate for parameters with reasonable ETA shrinkage in the final model were compared with similar plots for the base model to demonstrate that the final model accounted for trends observed with the base model. Relative standard error (RSE) of the parameter estimates were generated based on asymptotic standard errors generated from the NONMEM covariance step. The 95% confidence interval (CI) around the parameters was generated from bootstrap estimates from 1,000 resampled datasets.

A comparison of the OFV statistics and parameter estimates for the base and final models were used to assess the degree of parsimony of the final model, and parameter magnitude was used to determine the clinical relevance of the covariate effects. A comparison of ω 2 between the models was made to assess the reduction in variance by inclusion of covariate effects. ETA shrinkage estimated as (1‐Stdev(ηEBE)/ω) was also evaluated.

The performance of the final model was evaluated by simulating data using the parameter estimates from the final model (fixed and random effects) and by conducting a prediction‐corrected visual predictive check (pcVPC), which normalizes observed and simulated dependent variables based on the typical population to remove variability associated with binning across independent variables. Simulations were performed using patient status variable (PTST) as well as the dosing and sampling history from the original dataset. From these simulations, concentration‐time data were summarized using median, low, and high percentiles. The concordance between individual observations and simulated values as well as the distribution of observed and simulated data were evaluated.

Sensitivity analysis of covariate effects on exposure

A clinically meaningful change in lorlatinib plasma exposure was considered as one that would necessitate lorlatinib dose adjustment. Based on observations from the dose escalation portion of Study B7461001, a clinical no‐effect boundary of 70–142.9% (equivalent to a 30% change on the log scale) was defined for lorlatinib exposure, as compared with 100 mg q.d.

In order to assess whether the effect of the covariates included in the final model resulted in a meaningful change in lorlatinib plasma exposure warranting a dose adjustment, 500 individuals were simulated from the final model parameters for each of the covariates, at various extremes, and compared with the simulations for a typical individual; defined as a 70 kg individual with no PPI use, WNCL of 100 mL/min, baseline albumin (BALB) of 4 mg/dL, and dosed at 100 mg with steady‐state Cmax of 606 ng/mL and a steady‐state AUCtau of 5180 ng.h/mL. Forest plots were constructed using the conditions reported on the Y‐axis as fixed effects and the simulations as random effects.

RESULTS

The final analysis dataset was composed of data from 425 study participants, including 330 patients with advanced ALK‐positive or ROS1‐positive NSCLC and 95 healthy participants (Table 1). Baseline covariate data from the PopPK analysis set are summarized in Table 3.

Table 3

Summary of baseline covariates

Covariate	PopPK analysis dataset (n = 425)
Continuous, mean (SD)
Age, years	49.86 (13.20)
Albumin, g/dL	3.92 (0.58)
BMI, kg/m2	24.62 (4.77)
Creatinine clearance, mL/min	98.31 (32.13)
Weight, kg	70.53 (16.89)
Categorical, n (%)
Sex, female	191 (45)
Race
White	220 (52)
Black	32 (8)
Asian	113 (27)
Other	29 (7)
No PPI use	402 (95)
Baseline renal impairment
A, normal	243 (57)
B, mild	130 (31)
C, moderate	51 (12)
D, severe	1 (0)
Baseline hepatic impairment
A, normal	365 (86)
B1, mild	50 (12)
B2, mild	10 (2)
C–D, moderate–severe	0 (0)
CYP2C19 phenotype
Poor	18 (4)
Intermediate	100 (24)
Extensive	153 (36)
Ultra‐rapid	7 (2)
CYP2C9 phenotype
Poor	5 (1)
Intermediate	62 (15)
Extensive	211 (50)
Ultra‐rapid	0 (0)
CYP3A5 phenotype
Poor	195 (46)
Intermediate	66 (16)
Extensive	17 (4)
Ultra‐rapid	0 (0)

BMI, body mass index; CYP, cytochrome P450; PopPK, population pharmacokinetic; PPI, proton pump inhibitor.

Model development

Lorlatinib PK was initially described using a one‐compartment model defined in terms of CL, V, ka, and F; however, addition of a second compartment significantly improved model fit. The initial base model failed to adequately characterize Cmax in many of the subject PK profiles; therefore, a sequential zero‐order and first‐order (ka) absorption (with D1 parameter) was investigated. Allometric BWT correction was included a priori in the base model on CL and V2 by using a scaling factor exponent of 0.75 and 1, respectively, to remove a confounding effect observed between BWT and sex. A final two‐compartment base model, with auto‐induction of CL, and sequential zero‐first order absorption, defined in terms of CL, V2, V3, Q, zero‐order input duration (D1), ka, and F was found to best describe the data and was carried forward to the next stage of modeling.

Of the evaluated covariates, BALB, WNCL, and lorlatinib total daily dose (TDOSE) were found to be significant and were retained as covariates on CL in the final model. PPI use was retained as a covariate on ka.

The final PopPK model parameters are shown in Table 4. Lorlatinib CL, V2, and ka for a typical subject were parameterized as follows:

Table 4

Final PopPK model parameter estimates

Parameter	Model results			Bootstrap results
	Value	CV (%)	Shrinkage (%)	Mean	95% CI
θ CLI (L/h)	9.035	–	–	9.088	(8.0115–10.0609)
θ V2 (L)	120.511	–	–	120.618	(103.3633–137.6947)
θ ka (h−1)	3.113	–	–	3.128	(2.3125–3.9145)
θ Q (L/h)	22.002	–	–	22.491	(17.6495–26.3563)
θ V3 (L)	154.905	–	–	156.640	(134.2215–175.6205)
θ IND	0.020	–	–	0.027	(−0.2136 to 0.2535)
θ D1 (h)	1.148	–	–	1.149	(1.0344–1.2611)
θ F	0.759	–	–	0.764	(0.6728–0.8462)
θCLMX (L/h)	14.472	–	–	14.584	(12.7286–16.2186)
θ Res Error for IV	0.115	–	–	0.110	(0.0811–0.1487)
θ Res Error for PO	0.438	–	–	0.437	(0.4090–0.4670)
θ BALB on CL	0.067	–	–	0.069	(0.0214–0.1122)
θ TDOSE on CL	0.001	–	–	0.001	(0.0004–0.0023)
θ WNCL on CL	0.235	–	–	0.240	(0.1457–0.3238)
θ PPI on ka	–0.675	–	–	−0.664	(−0.8508 to −0.4986)
IIV parameter
ω2 CL	0.030	17.201	23.212	0.030	(0.0159–0.0433)
ωFωCL (L/h)	−0.006	7.460	–	−0.005	(−0.0173 to 0.0061)
ω2 F	0.022	14.964	40.174	0.023	(0.0027–0.0420)
ω2 V2	0.086	29.268	52.835	0.085	(0.0430–0.1284)
ωV2ωV3	−0.017	12.881	–	−0.017	(−0.0492 to 0.0160)
ω2 V3	0.101	31.742	53.123	0.099	(0.0513–0.1502)
ω2 ka	2.329	152.626	45.113	2.345	(1.5982–3.0608)

The mean and 95% CIs were generated from a bootstrap run of 1,000 resampled datasets, including runs with successful minimization and failed $COV steps.

BALB, baseline albumin; CI, confidence interval; CLI, initial clearance; CLMX; maximum induced clearance at steady state; CV, coefficient of variation; D1, zero‐order duration of absorption; F, bioavailability; h, hour; IIV, interindividual variability; IND, rate constant of induction; k a, rate constant of absorption; PPI, proton pump inhibitor use; Q, intercompartmental clearance; RSE, relative standard error; TDOSE, total daily dose (mg); V2, central volume of distribution; V3, peripheral volume of distribution; WNCL, baseline standardized creatinine clearance.

Lorlatinib CL was estimated to be 9.04 L/h, and CLMX was estimated to be 14.5 L/h. IND was estimated to be 0.0199 h−1, or 0.478 days−1, corresponding to an induction half‐life of ~ 34.8 hours (1.45 days). This suggests that lorlatinib initial clearance will approach CLMX after 7.25 days, assuming that the metabolic auto‐induction of lorlatinib reaches steady‐state in ~ 5 half‐lives. Thus, the exposure at cycle 1, day 15 represents the lorlatinib exposure after the completion of auto‐induction. The typical value for V2 was estimated to be 121 L. For the sequential zero‐first order absorption, D1 was estimated to be 1.15 hours and the typical value for ka was estimated to be 3.11 (h−1). The PPI use covariate was equal to 1 if the subject was randomized to receive PPI with lorlatinib and 0 otherwise. In addition, V3 was estimated to be 155 L, Q was estimated to be 22.0 L/h, and F was estimated to be 0.759.

Plots of observed concentrations versus PRED and IPRED are presented in Figure 1. Several data points were identified as potential outliers; however, after exclusion of these data points, the parameter estimates did not change by greater than 10%. This suggested that the outliers were not influential, and they were therefore retained in the final dataset.

Figure 1

Plots of (a) observed vs. predicted population and individual lorlatinib plasma concentrations and (b) conditional residuals versus time after first dose and population predicted values from the final lorlatinib population pharmacokinetic (PopPK) model. In Panel a, the red dashed line represents unity and black line represents linear smooth. Dose is in milligrams. CWRES, conditional weighted residuals; log, natural log; IWRES, individual weighted residuals; TAFD, time after first dose.

As shrinkage was relatively high (>30%) for all parameters except CL and F, ETA plots were considered unreliable and are not presented.

Final model predictive performance

The pcVPC plots showed that the final model had good predictive performance, with the 5th, 50th, and 95th percentiles of the observed data lying within the 90% CIs of the simulated 5th, 50th, and 95th percentiles (Figure 2).

Figure 2

Prediction‐corrected visual predictive check of the final population pharmacokinetic (PopPK) model for (a) all patient data. (b) First 120 hours (both patients and healthy participants). (c) Day 15 of cycle 1 (patients only). TAFD was reset on period 1 day 1 and thus the first 120 hours represents a pooling of day −7 and period 1 day 1. Shaded areas represent a simulation based 90% prediction interval of the 5th, 50th, and 95th percentile of the simulated data. Red lines represent the 5th, 50th, and 95th percentile of the observed data. h, hour; PTST, patient status: 0 for healthy participant, 1 for patients; TAFD, time after first dose.

Covariate effects on lorlatinib steady‐state exposure

A sensitivity analysis was conducted to assess whether the covariates included in the final model affected lorlatinib exposure to an extent that may warrant dose adjustment.

In the final PopPK model, an individual with low BWT (50 kg; 10th percentile of the analysis population) was predicted to have a 22.3% reduction in CL and 28.6% reduction in V2 relative to a typical patient. For an individual with high BWT (91.3 kg; 90th percentile), CL and V2 were increased 22% and 30.4%, respectively, relative to the typical patient.

According to the final PopPK model, individuals with low BALB (3.2 mg/dL; 10th percentile) had a 5.3% reduction in CL, whereas individuals with high BALB (4.6 mg/dL; 90th percentile) had a 4% increase in CL compared with the typical patient (BALB 4 mg/dL).

Individuals with PPI use had a 67.5% decrease in ka relative to a typical patient without PPI use. This change in absorption rate constant did not impact meaningfully on overall lorlatinib exposure (represented by Cmax and AUC) as shown in Figure 3.

Figure 3

Lorlatinib simulated steady‐state Cmax and AUCtau. For the BALB and BWT sensitivity analysis, the exposure ratios were determined based on the 10th and 90th percentile BWT and BALB of the population in the PopPK dataset. For the WNCL sensitivity analysis, the exposure ratios were determined based on the median WNCL values for patients in the normal, mild, moderate, and severe renal impairment groups based on KDOQI criteria. AUCtau, area under the curve over the dosing interval tau; BALB, baseline albumin; BWT, baseline body weight; CI, confidence interval; Cmax, maximum observed concentration; KDOQI, Kidney Disease Outcome Quality Initiative; PopPK, population pharmacokinetic; PPI, proton pump inhibitor; WNCL, weight normalized creatinine clearance. Vertical red dotted lines represent the 80–125% bioequivalence boundary, and the vertical blue dotted lines represent the 70–142.9% no‐effect boundary established for lorlatinib.

TDOSE was also found to be a significant predictor of CL. Individuals who received 10 mg q.d. lorlatinib had a 12.4% lower CL (both initial and time‐varying induced CL) compared with individuals receiving 100 mg q.d. lorlatinib. Similarly, individuals receiving the 200 mg q.d. dose had a 13.8% higher CL compared with individuals receiving 100 mg q.d. lorlatinib. At the lowest possible lorlatinib exposure at 100 mg q.d. (based on an individual with BWT of 91.3 kg, BALB of 4.6 mg/dL and PPI use), the final PopPK model predicted an increase in CL of 26.9%, an increase in V2 of 30.4%, and a decrease in ka of 67.5%, compared with a typical patient. At the highest possible lorlatinib exposure at 100 mg q.d. (based on an individual with BWT of 50 kg, BALB of 3.2 mg/dL and no PPI use), the final PopPK model predicted a decrease in CL of 26.5%, a decrease in V2 of 28.6%, and unchanged ka compared with a typical patient.

Sensitivity analysis conducted using 500 simulated profiles from the final PopPK model supported the conclusion that the identified significant covariates on lorlatinib CL did not have a clinically meaningful impact on lorlatinib plasma exposure, with all model‐simulated values falling within the predefined clinical no‐effect boundary, suggesting that that no dose adjustments were required (Figure 3).

Effect of hepatic and renal impairment on lorlatinib PK

Although the hepatic function indicators BALT and BBIL were not statistically significant covariates and were not included in the final model, the effect of hepatic function on lorlatinib CL was assessed by summarizing individual estimates of lorlatinib CL in each baseline NCI hepatic impairment group. Within the range of hepatic impairment in the pooled dataset, no changes in lorlatinib CL were predicted with worsening hepatic impairment after either single or multiple doses (Table S1).

To assess possible relationships between renal impairment and lorlatinib CL, individual estimates of lorlatinib CL were summarized by renal impairment group as defined by Kidney Disease Outcome Quality Initiative (KDOQI) staging. There was a trend of decreasing median and mean individual estimates of single dose and steady‐state lorlatinib CL with worsening renal function, although ranges overlapped in patients with mild or moderate renal impairment and patients with normal renal function (Table S2).

DISCUSSION

Lorlatinib plasma PK was well‐characterized by a two‐compartment model with CL estimated using an initial CL after a single dose, and a time‐varying induced CL after multiple doses. The time‐varying, induced CL approached its maximum value at steady‐state. To improve Cmax estimation, the final model employed a sequential zero‐first order absorption process. Two‐compartment PopPK models incorporating time‐varying clearance mechanisms and PopPK models utilizing sequential zero‐first order absorption have been previously described in the literature. , , , ,

Clearance was evaluated using concentration‐dependent saturable clearance, and models linking single‐dose and time‐dependent clearance. However, successful minimization and variance covariance steps were not achieved. This was likely due to the limitations in the available PK data on different days following multiple dosing, with an absence of PK data available between day 1 and day 15 to characterize the transition from single dose to steady‐state lorlatinib clearance.

Lorlatinib TDOSE was found to be a statistically significant covariate for lorlatinib CL, which is to be expected for a drug undergoing concentration‐related (and dose‐related) auto‐induction. This was also consistent with lorlatinib plasma exposures estimated by noncompartmental analysis. In Study B7461001, the steady‐state lorlatinib CL increased with increasing dose. In this analysis, initial and time‐varying induced CL was reduced by 12.4% with 10 mg q.d. lorlatinib and increased by 13.8% with 200 mg q.d. lorlatinib, compared with individuals receiving 100 mg q.d. lorlatinib. This dose nonlinearity is likely due to an increase in potency of induction with increasing dose, which was accounted for with TDOSE as a covariate on CL.

BALB was also found to be statistically significant on CL. With every unit of BALB over 4 mg/dL, there was a 6.68% increase in CL. For individuals with a BALB value of 3.2 mg/dL, the clearance was 5.3% lower compared with the typical patient with 4 mg/dL of BALB. At a BALB value of 4.6 mg/dL, the clearance was increased 4% compared with the typical patient. Although the exact reason for this relationship is unknown, it is possible that BALB acts as a predictor of a patient’s overall health, with poorer health resulting in impaired drug clearance.

The impact of PPI use was modeled only on the first‐order absorption component of the final PopPK model and was found to be associated with a 67.5% decrease in ka. This was consistent with the results of the formal drug interaction study B7461008 in healthy participants, where PPI use decreased Cmax by 30% without changing the area under the curve over infinity (AUCinf).

As lorlatinib is extensively metabolized in the liver, the effects of hepatic impairment on lorlatinib PK were also assessed. Hepatic impairment was not found to be a statistically significant covariate in the final model and no correlation was found between baseline NCI hepatic impairment stage and lorlatinib clearance.

There is evidence that chronic or severe renal impairment can also influence the PK of drugs that are exclusively metabolized by the liver. Renal impairment can change levels of serum albumin, affecting the fraction unbound to protein in plasma for a drug and thereby altering its volume of distribution. In addition, chronic renal failure has been known to reduce CYP activity. Thus, renal impairment as measured by WNCL was tested as a potential covariate on lorlatinib V2, CL, and F, and was found to be statistically significant on CL in the final model.

A clinically meaningful change in lorlatinib exposure was considered as one that would require a dose adjustment. In the dose escalation portion of Study B7461001, dose‐limiting toxicities were seen at the 150 mg q.d. dose and 100 mg q.d. was the recommended phase II dose. Furthermore, whereas the next lower dose from 100 mg q.d. (i.e., 75 mg q.d.) has been associated with plasma lorlatinib concentrations that cover in vitro efficacious concentration targets, the majority of the lorlatinib clinical experience has been at the 100 mg q.d. dose level, which has been shown to be safe and efficacious. Thus, a clinical no‐effect boundary was defined as 70–142.9% (30% change on the log scale) as compared with 100 mg q.d., with the lower bound as the exposure at 75 mg q.d. and an upper bound as the exposure at 150 mg q.d. Based on this clinical no‐effect boundary, none of the covariate effects identified as statistically significant in the PopPK model were seen to be associated with clinically meaningful changes in lorlatinib plasma exposure. Hence, these findings support current lorlatinib dosing guidelines, which do not recommend dose modifications based on PPI use, age, weight, race, renal and hepatic impairment, BALB, and CYP3A5 and CYP2C9 metabolizer phenotype. As only one patient in the analysis had severe renal impairment, conclusions cannot be made regarding dose recommendations in this patient population.

Conflict of Interest

J.C., B.H., Y.K.P., and A.R.‐G. are current or former employees of Pfizer. J.C. and Y.K.P. also own Pfizer stock.

Author Contributions

J.C., B.H., Y.K.P., and A.R.‐G. wrote the manuscript and designed the research. J.C. and Y.K.P. performed the research. J.C. and B.H. analyzed the data.

Table S1

Click here for additional data file.

Table S2

Click here for additional data file.

File S1

Click here for additional data file.

List S1

Click here for additional data file.

Video S1

Click here for additional data file.