Population pharmacokinetics of naloxegol in a population of 1247 healthy subjects and patients

Abstract

Aims

Naloxegol, a polyethylene glycol conjugated derivative of the opioid antagonist naloxone, is in clinical development for treatment of opioid‐induced constipation (OIC).

The aim of the study was to develop a population pharmacokinetic model describing the concentration vs. time profile of orally administered naloxegol, and determine the impact of pre‐specified demographic and clinical factors and concomitant medication on population estimates of apparent clearance (CL/F) and apparent central compartment volume of distribution (V _c/F).

Methods

Analysis included 12 844 naloxegol plasma concentrations obtained from 1247 healthy subjects, patients with non‐OIC and patients with OIC in 14 phase 1, 2b and 3 clinical studies. Pharmacokinetic analysis used the non‐linear mixed effects modelling program. Goodness of fit plots and posterior predictive checks were conducted to confirm concordance with observed data.

Results

The final model was a two compartment disposition model with dual absorptions, comprising one first order absorption (k _a1 4.56 h⁻¹) and one more complex absorption with a transit compartment (k _tr 2.78 h⁻¹). Mean (SE) parameter estimates for CL/F and V _c/F, the parameters assessed for covariate effects, were 115 (3.41) l h^–1 and 160 (27.4) l, respectively. Inter‐individual variability was 48% and 51%, respectively. Phase of study, gender, race, concomitant strong or moderate CYP3A4 inhibitors, strong CYP3A4 inducers, P‐glycoprotein inhibitors or inducers, naloxegol formulation, baseline creatinine clearance and baseline opioid dose had a significant effect on at least one pharmacokinetic parameter. Simulations indicated concomitant strong CYP3A4 inhibitors or inducers had relevant effects on naloxegol exposure.

Conclusions

Administration of strong CYP3A4 inhibitors or inducers had a clinically relevant influence on naloxegol pharmacokinetics.

What is Already Known About this Subject

• Naloxegol is a polyethylene glycol conjugated derivative of naloxone in clinical development for the treatment of opioid‐induced constipation.

• Phase 2/3 studies have demonstrated the clinical efficacy of naloxegol in patients receiving opioids for non‐cancer pain.

• The pharmacokinetic profile of naloxegol is well characterized.

What this Study Adds

• Population pharmacokinetic modelling of naloxegol has been conducted to describe the observed clinical data and to evaluate the potential impact of demographic characteristics, clinical characteristics and concomitant medication on the pharmacokinetics of naloxegol.

• The final model was adequate for deriving exposure metrics for subsequent exposure–response modelling.

• Concomitant administration of strong CYP3A4 inhibitors or inducers was a significant covariate that affected systemic exposure to naloxegol.

Introduction

Opioid‐induced constipation (OIC) is a common adverse event (AE) associated with opioid treatment of pain. OIC is a consequence of opioid receptor agonist effects on the gastrointestinal (GI) tract, largely mediated by agonist binding of μ‐opioid receptors 1 in the enteric nervous system 2, 3, 4. These effects include reduced GI motility and mucosal secretion, as well as increased resorption of fluid from the gut. The estimated prevalence of OIC in patients receiving opioids for non‐cancer pain ranges from 15 to 90% 2. Most patients consider OIC as the most bothersome AE associated with opioid pain medications and report at least one moderately negative impact on their quality of life 5.

Treatment for OIC, including laxatives (e.g. stool softeners, stimulants and osmotic agents) and non‐pharmacologic strategies (e.g. increased dietary fibre, fluid intake and exercise) is often suboptimal, with many patients not achieving adequate symptom relief 6, 7, 8. Laxatives do not directly address the range of underlying agonist effects of opioids on the GI tract or the symptoms associated with delayed gastric emptying (e.g. bloating, nausea and satiety). Moreover, chronic laxative use has been associated with electrolyte imbalance and abdominal cramp‐like pain 7, 8.

Naloxegol (NKTR‐118) is a polyethylene glycol (PEG) conjugated derivative of the opioid antagonist naloxone that is in clinical development as an oral, once daily treatment for OIC in patients receiving opioid therapy for pain 9. PEGylation reduces the passive permeability of naloxegol and renders the compound a substrate for the P‐glycoprotein (P‐gp) transporter, restricting its access to the brain but maintaining its peripheral opioid receptor antagonist properties 10.

Several randomized, double‐blind, placebo‐controlled phase 2/3 studies have demonstrated the efficacy of naloxegol in patients with OIC receiving opioids for non‐cancer pain. In a phase 2 study (n = 207) with a 4 week treatment period, naloxegol 25 mg once daily was significantly more effective than placebo in increasing the number of spontaneous bowel movements per week over baseline 10. In two phase 3, double‐blind studies (KODIAC‐04 and KODIAC‐05; n > 1350) with 12 week treatment periods, naloxegol 25 mg once daily significantly increased the responder rate compared with placebo (naloxegol 40–44%; placebo 29%) 11. Naloxegol was generally well tolerated, with mild GI symptoms as the most common AEs 10, 11.

The pharmacokinetic (PK) profile of naloxegol is well characterized in healthy subjects 12, 13, in patients with hepatic 14 or renal impairment 15 and in patients with OIC 10. Naloxegol exhibits dose proportional PK in healthy subjects and patients with OIC 10, 12, 13 and is rapidly absorbed following oral administration, with a similar half‐life across all dose levels. The role of the hepatobiliary system in the metabolism of naloxegol is well established, with biliary excretion the primary route of elimination. However, surprisingly, mild or moderate hepatic impairment had minimal effect on the PK of naloxegol 14, whereas increased exposures were observed in a small number of patients with moderate or severe renal impairment 15.

In vitro studies showed that naloxegol is metabolized by cytochrome P450 (CYP)3A4 and is a substrate of P‐gp 16. However, naloxegol was found to be metabolically stable in the presence of human CYP isoforms and, further, there was low or no inhibition or induction of CYP isoforms in the presence of naloxegol 16. A double‐blind, crossover study in healthy subjects showed no effect of concomitant naloxegol on the PK of morphine and its glucuronide metabolites 16. These data suggest a low risk for naloxegol to cause clinically significant drug–drug interactions (DDIs). However, there is a high risk that the PK of naloxegol may be affected by administration of other medications.

The aim of this analysis was to develop a population PK model for characterization of the concentration vs. time profile of orally administered naloxegol which could be used to predict systemic exposure for exposure–response analyses. In addition, the model was used to determine the impact of pre‐specified covariates on apparent clearance (CL/F) and on apparent volume of distribution of the central compartment (V _c/F) of naloxegol in healthy subjects, patients with non‐OIC, and patients with OIC. Identifying covariates with a significant impact on the PK profile of naloxegol will provide key information on optimal dosing of naloxegol relative to the identified covariates.

Methods

Studies

Development of the population PK model was based on data extracted from 11 phase 1 studies in healthy subjects and patients with renal or hepatic impairment, one PK sub‐study from a phase 2b dose escalation study in patients with OIC, and two phase 3 studies in patients with OIC 17. The 14 studies included in the population PK analysis are outlined in Table 1.

Table 1.

Studies included in the naloxegol population PK analysis

Protocol number/NCT	Phase	Study description	Subjects/patients	Samples	Number of naloxegol doses (dose, mg)	Population
05‐IN‐OX001	1	Single ascending dose safety, tolerability, PK and PD	48	569	8 (8, 15, 30, 60, 125, 250, 500 and 1000)	Healthy adults
07‐IN‐NX002	1	Multiple ascending dose safety, tolerability and PK study	24	839	4 (25, 60, 125 and 250)	Healthy adults
08‐PNL‐04	1	Relative bioavailability (tablet vs. solution)	20	752	1 (100)	Healthy adults
NCT01372826	1	Renal function impairment	32	518	1 (25)	Renal impairment
NCT01392807	1	Hepatic function impairment (excluding subjects with severe hepatic impairment)	24	312	1 (25)	Hepatic impairment
NCT01325415	1	Thorough QTc	51	995	2 (25 and 150)	Healthy adults
NCT01318655	1	Single and multiple ascending dose safety, tolerability and PK	40	1175	4 (12.5, 25, 50 and 100)	Healthy Japanese adults and elderly
NCT01365000	1	Bioavailability of different formulations	24	1582	1 (25)	Healthy adults
NCT01520896	1	DDI with ketoconazole (strong CYP3A4 inhibitor/P‐gp inhibitor)	22	687	1 (25)	Healthy adults
NCT01533870	1	DDI with rifampicin (strong CYP3A4 inducer/P‐gp inducer)	22	546	1 (25)	Healthy adults
NCT01594619	1	DDI with diltiazem (moderate CYP3A4 inhibitor/P‐gp inhibitor)	43	1298	1 (25)	Healthy adults
NCT00600119	2b	Double‐blind, multiple‐dose escalation in patients with OIC (PK sub‐study)	21	479	3 (5, 25 and 50)	Patients with OIC
NCT01309841	3	Efficacy and safety in patients with non‐cancer‐related pain and OIC	422	1983	2 (12.5 and 25)	Patients with OIC
NCT01323790	3	Efficacy and safety in patients with non‐cancer‐related pain and OIC	454	2057	2 (12.5 and 25)	Patients with OIC
Total			1247	13 792

CYP3A4, cytochrome P450 3A4; DDI, drug–drug interaction; NCT, ClinicalTrials.gov identifier; OIC, opioid‐induced constipation; PD, pharmacodynamics; P‐gp, P‐glycoprotein; PK, pharmacokinetics.

All subjects provided written informed consent prior to entry into the studies, which were all approved by independent Institutional Review Boards/Ethics Committees (studies 05‐IN‐OX001 and 07‐IN‐NX002 were approved by approved by The Stichting Therapeutishce Evaluatie Geneesmiddelen–Medisch Ethische Toetsingscommissie (STEG/METC), an independent ethics committee (IEC). Study 08‐PNL‐04 was approved by Independent Investigational Review Board Inc., Plantation, Florida, USA. Other studies were registered on www.clinical trials.gov: NCT01372826, NCT01392807, NCT01325415, NCT01318655, NCT01365000, NCT01520896, NCT01533870, NCT01594619, NCT00600119, NCT01309841, NCT01323790). Each study was performed in accordance with the ethical principles that have their origin in the Declaration of Helsinki and that were consistent with the International Conference on Harmonization/Good Clinical Practice and applicable regulatory requirements.

Bioanalytical methods

The concentration–time data extracted from the 14 studies were acquired through either rich or sparse sampling schemes. For the phase 1 studies and the phase 2b PK sub‐study, a rich plasma sampling scheme was applied, which, although variable across the studies, resulted in the collection of a minimum of 10 samples after each dosing at times between 0.08 and 120 h post‐dosing, in order to define adequately the PK profile of naloxegol. For the phase 3 studies, a sparse sampling scheme was used, comprising a baseline sample and on‐treatment samples collected 2 h after dosing at 1, 2, 4, 8 and 12 weeks after initiation of naloxegol.

Plasma concentrations of naloxegol were determined from a 0.1 ml sample using a validated bioanalytical method using solid phase extraction and liquid chromatography followed by tandem mass spectrometric detection (LC‐MS/MS) 12. Calibration curves extended over the concentration range of 0.1 to 50 ng ml^–1. Naloxegol concentrations that were below the analytical assay quantification limit (0.1 ng ml^–1), otherwise missing or did not have corresponding dosing times were excluded from the analysis.

Population pharmacokinetic analysis

Software

The software package Non‐linear Mixed Effects Modeling (nonmem™) version VII (ICON, Hanover, MD, USA) was used for modelling and simulation. Pre‐ or post‐processing of nonmem runs was performed using Statistical Analysis System (SAS Institute Inc., Cary, NC, USA) and S‐Plus version 8.2 (MathSoft, Inc., Seattle, WA, USA) software packages. The first order conditional estimation with inter‐individual variability (IIV; η) – residual variability (ε) interaction (FOCEI) method in nonmem was employed for model runs.

Datasets

SAS software was used to extract the relevant data files from clinical study databases. Model development utilized three nonmem compatible datasets. Set 1, which was used for initial development of the base model, incorporated data from eight of the 11 phase 1 studies (excluding DDI studies) and the phase 2b PK sub‐study. Set 2 consisted of data from all phase 1 studies, the phase 2b PK sub‐study and both phase 3 studies. Set 3 was composed of the same data as set 2, with additional phase 3 subjects and updated concomitant medication codes.

A number of potentially clinically relevant covariates were screened to determine if they contributed to IIV in PK parameter estimates. Selection of the covariates to be tested for each parameter was based on clinical judgment and mechanistic plausibility. Covariates that were considered included age, weight, gender, race, study phase, naloxegol formulation, hepatic function, renal function, baseline opioid dose, and concomitant CYP3A4 and P‐gp inhibitors and inducers. In each patient, alanine aminotransferase (ALT) was used for evaluation of hepatic function and creatinine clearance (CL_cr) was used for evaluation of renal function. Since ALT and aspartate transaminase were highly correlated, only ALT was tested in the model based on initial model selection using phase 1 and 2 data. Missing continuous covariate data were replaced by the median value. Missing categorical covariate data were modelled as a separate category or combined with one of the other categories for that covariate.

Base model development

Initially, a two compartment model with first order absorption and first order elimination was developed to describe the time course of naloxegol exposure. Subsequently, a more complex structural model with dual, parallel first order absorptions and first order elimination was developed to take into account the presence of a double peak in the naloxegol plasma concentration vs. time profile observed in a phase 1 study 12.

The structural model was developed using the FOCEI method in nonmem. Model development and selection were based on various goodness‐of‐fit indicators. Data were classified as outliers using the population conditional weighted residuals (CWRES) or individual weighted residuals (IWRES). Observations with |CWRES| > 6 or |IWRES| > 6 were considered potential outliers. The outliers were considered influential if key parameter estimates differed by more than 15% 18.

Close attention was paid to the stability of the models throughout their development process. To avoid ill‐conditioning, inspection of the covariance matrix of estimates at every stage of model development was performed to verify that extreme pair‐wise correlations (P > 0.95) of the parameters were not encountered. The condition number of the correlation matrix of the parameter estimates (i.e. ratio of the largest to smallest eigenvalues) was assessed to ensure values less than 1000. Values greater than 1000 are indicative of an ill‐conditioned model. When convergence or covariance estimation problems occurred, ad hoc nonmem runs were performed to evaluate the nature of the ill‐conditioning.

In order to incorporate IIV of the PK parameters, a log normal random effects model was used:

$${\theta }_i={\theta }_{\mathit{TV}}\cdot exp\left({\eta }_i\right)$$

where θ _i is the individual value of the parameter $\left(\text{e.g.},,\text{CL},/,F,,\text{or},,V_c/F\right),{\theta }_{TV}$ is the typical value model parameter, and η _i denotes the inter‐individual random effect accounting for the ith individual's deviation from the typical value. The η _i was assumed to have a normal distribution with a mean of zero and variance of ω². The approximate coefficient of variation (%CV) is reported as:

$$\%\mathrm{CV}\left(\mathrm{IIV}\right)=\sqrt{{\omega }^2}\cdot 100$$

Residual variability was modelled using the combined proportional plus additive error structure:

$$\begin{array}{l}{Y}_{\mathit{ij}}=C_{\mathit{ij}}+w_{\mathit{ij}}{\epsilon }_{\mathit{ij}}\\ w_{\mathit{ij}}=\sqrt{C_{\mathit{ij}}^2{\sigma }_1^2+{\sigma }_2^2}\end{array}$$

where Y_ij denotes the observed concentration for the ith individual at time t _j; C _ij denotes the corresponding predicted concentration based on the PK model, ε _ij denotes the intra‐individual (residual) random effect (zero mean and unit variance), and w_ij denotes the residual standard deviation (SD) with corresponding proportional and additive variance compartments, ${\sigma }_1^2$ and ${\sigma }_2^2$, respectively.

Full model development

A full model was constructed to examine the effects of pre‐specified covariates on CL/F and V _c/F using a stepwise forward selection procedure.

Power models were used to describe the relationship between continuous covariates and the typical value of PK parameters:

$${\theta }_{\mathit{TV}}={\theta }_{\mathit{REF}}\cdot {\left(\frac{X_i}{X_{\mathit{REF}}}\right)}^{{\theta }_x}$$

θ_REF and θ_x are the fixed effect parameters and X_REF is a reference value of the covariate X_i. Reference values represent the approximate median of the population.

The relationship between binary categorical covariates (X_i) and the typical value of PK parameters was modelled as:

$${\theta }_{\mathit{TV}}={\theta }_{\mathit{REF}}\cdot \left(1+{\theta }_x{X}_i\right)$$

θ_REF and θ_x are fixed effect parameters and X_i is the indicator variable, which is equal to 1 or 0, dependent on the category of the covariates. The lower bound values for θ_x were constrained to be > −1, such that PK parameters were always positive. For covariates such as moderate or strong inhibitors, liver function, and study phase, a separate value for CL was estimated for each covariate condition.

Once a stable full model was established, diagnostic plots of the individual random effect values vs. covariate values were generated in order to identify any covariate effects that had not been properly accounted for. Box plots of the inter‐individual random effects for each parameter were evaluated for different patient populations to determine whether the selected covariates adequately accounted for all observed differences.

Dose invariance and adequacy of pooling studies for this analysis were also assessed by testing a variety of models to explore dosage dependence on PK parameters.

Final model development and evaluation

A stepwise forward selection procedure was implemented to evaluate reduced models relative to the full model in order to identify a parsimonious final model containing comparable information to the full model despite a reduced number of covariates. For a covariate to be included in the model, it had to meet statistical significance criteria, pre‐specified as P ≤ 0.001 or 10.8 point reduction or larger in the objective function value for one parameter (see Table S1 in Supporting Information).

The final model was tested using goodness of fit diagnostic plots. In addition, η plots vs. each covariate were compared with similar plots for the base model to verify that the final model accounts for trends observed with the base model. The predictive performance of the final model was evaluated using the posterior predictive check (PPC) 19, 20. Parameter estimates were assumed to have a multivariate normal distribution, with the mean vector set to the population mean parameter estimates and the covariance matrix set to the covariance matrix of estimates from the final model. The multivariate normal distribution was used as an approximate posterior distribution to generate 1000 sets of population parameter values, and each of these was used to simulate 1000 datasets replicating the design, dose regimen, and covariates of the final model dataset; 90% prediction intervals (PI) were calculated and compared for observed and simulated data.

Model simulation

Simulations were conducted to investigate the clinical relevance of covariate effects, using the final model, at the 25 mg naloxegol once daily dose. Naloxegol exposure (AUC) and maximum plasma concentration (C _max) were calculated. A total of 100 subjects were re‐sampled from the observed data to provide a plausible combination of covariate values. Two periods were simulated per subject. For continuous demographic variables, the 25^th and 75^th percentiles were calculated from the entire population PK dataset and these values were used. For categorical variables, 0 was set as the reference value and 1 was set as the investigational value. Simulations were conducted including parameter uncertainty, IIV and residual variability (n = 10 constructing 1000 subjects). The software program R was used to sample randomly 10 values for each parameter from their respective inverse‐gamma uncertainty distributions and 100 values from their respective multivariate normal variability distributions. These were then combined to create 1000 full sets of simulation parameter values for the population PK model. AUC and C _max ratios were then calculated to produce 1000 ratio values with mean and 90% PI. The impact of each covariate was estimated using Forest plots.

Results

Baseline characteristics

The population PK dataset consisted of 1247 subjects (healthy subjects, patients with non‐OIC and patients with OIC) contributing a total of 13 792 naloxegol plasma concentrations (Table 1). A total of 9273 samples (67.2%) from 349 subjects (28.0%) were collected during the phase 1 studies, 479 (3.5%) from 21 patients (1.7%) in the phase 2b sub‐study and 4040 (29.3%) from 876 patients (70.3%) in the phase 3 studies. Overall, 948 samples (6.9%) were excluded from the analysis. Thus 12 844 valid plasma concentrations were used to develop the model. All plasma concentrations below the limit of quantification (BLQ) were included in the dataset as missing values except those prior to the first dose (n = 350) which were commented out (removed) from the analysis. There were no outliers in the final model run. Only one continuous covariate value was imputed with the population median (one missing CL_cr value). All subjects were adults. Mean (SD) age and body weight were 47.6 (14.1) years and 86.2 (21.5) kg, respectively (Table 2). Approximately 51% were male, 74% were Caucasian and 20% were Black. Most subjects had normal renal function (98%) and normal hepatic function (99%).

Table 2.

Baseline demographic characteristics, laboratory values and concomitant medications

Study (Dataset)	Phase 1 (2, 3)	Phase 2b (2, 3)	Phase 3 (2)	Overall (2)	Phase 3 (3)	Overall (3)
	n = 349	n = 21	n = 896	n = 1266	n = 898	n = 1268
Mean age (years) (SD)	36.3 (15.4)	46.9 (11.7)	51.9 (10.9)	47.5 (14.1)	52 (10.9)	47.6 (14.1)
Mean weight (kg) (SD)	77.4 (14.2)	86.5 (17.3)	89.6 (22.9)	86.2 (21.5)	89.6 (22.9)	86.2 (21.5)
Gender, male, n (%)	291 (83.4)	10 (47.6)	345 (38.5)	646 (51.0)	345 (38.4)	646 (50.9)
Race, n (%)
Caucasian	203 (58.2)	19 (90.5)	713 (79.6)	935 (73.9)	714 (79.5)	936 (73.8)
Black	89 (25.5)	2 (9.5)	166 (18.5)	257 (20.3)	167 (18.6)	258 (20.3)
Asian	46 (13.2)	0 (0)	7 (0.8)	53 (4.2)	7 (0.8)	53 (4.2)
Other	11 (3.2)	0 (0)	10 (1.1)	21 (1.7)	10 (1.1)	21 (1.7)
Patients with OIC, n (%)	0 (0)	21 (100)	896 (100)	917 (72.4)	898 (100)	919 (72.5)
Mean SBM (SD)	NA	1.4 (1.0)	1.4 (1.0)	1.4 (1.0)1	1.4 (1.0)	1.4 (1.0)2
Mean opioid dose (mg) (SD)	0 (0)	NA	135.8 (145.2)3	97.7 (137.5)4	135.9 (145.2)	97.8 (137.5)5
Mean CLcr (ml min–1) (SD)	115.1 (35.4)	114.3 (27.0)	111.1 (38.2)	112.2 (37.3)7	111.1 (38.4)	112.3 (37.5)8
Mean ALT (IU l–1) (SD)	22.7 (13.0)	19.2 (8.8)	22.4 (15.5)	22.5 (14.8)	22.4 (15.5)	22.5 (14.7)
Mean ALP (IU l–1) (SD)	90.2 (57.5)6	NA	79.7 (25.9)	82.2 (36.2)	79.7 (25.9)	82.2 (36.2)
Concomitant strong CYP3A4 inhibitor, n (%)	0 (0)	0 (0)	4 (0.4)	4 (0.3)	4 (0.4)	4 (0.3)
Concomitant moderate CYP3A4 inhibitor, n (%)	0 (0)	0 (0)	21 (2.3)	21 (1.7)	27 (3)	28 (2.2)
Concomitant weak CYP3A4 inhibitor, n (%)	0 (0)	10 (47.6)	206 (23.0)	216 (17.1)	274 (30.5)	284 (22.4)
Concomitant strong CYP3A4 inducer, n (%)	0 (0)	0 (0)	10 (1.1)	10 (0.8)	10 (1.1)	10 (0.8)
Concomitant moderate CYP3A4 inducer, n (%)	0 (0)	0 (0)	5 (0.6)	5 (0.4)	10 (1.1)	10 (0.8)
Concomitant weak CYP3A4 inducer, n (%)	0 (0)	0 (0)	14 (1.6)	14 (1.1)	27 (3)	27 (2.1)
Concomitant P‐gp inhibitor, n (%)	0 (0)	0 (0)	52 (5.8)	52 (4.1)	57 (6.3)	58 (4.6)
Concomitant P‐gp inducer, n (%)	0 (0)	0 (0)	11 (1.2)	11 (0.9)	11 (1.2)	11 (0.9)

¹ n = 917; ² n = 919; ³ n = 895; ⁴ n = 1244; ⁵ n = 1246; ⁶ n = 277; ⁷ n = 1173; ⁸ n = 1175; ALP, alkaline phosphatase; ALT, alanine aminotransferase; CL_cr, creatinine clearance; CYP3A4, cytochrome P450 3A4; NA, not available; OIC, opioid‐induced constipation; P‐gp, P‐glycoprotein; SBM, spontaneous bowel movement; SD, standard deviation

Base model

Based on exploratory analysis and the mechanism of elimination of naloxegol, development of the structural model temporarily excluded data from subjects taking concomitant moderate/strong CYP3A4 inhibitors or strong CYP3A4 inducers, and subjects with impaired liver function. Thus, development of the base model was started on a subset of the dataset (eight phase 1 studies and one phase 2 study), with additional data (three phase 1 drug interaction studies and two phase 3 studies) added during the modelling process with the final base model based on the whole dataset. The initial model building investigated the effects of covariates, including CYP3A4 inhibitors and inducers, on absorption and bioavailability parameters. However, incorporation of additional data into the model suggested the absorption of naloxegol was apparently more complex than first assumed. Because of the relative paucity of phase 3 data it was decided to focus the covariate analysis on key PK parameters such as CL and V _c. This modelling approach helped achieve a good initial estimate for the PK parameters and contributed to model stability and convergence.

A two compartment disposition model with dual absorptions, comprising one first order absorption (k _a1 4.56 h⁻¹) and one more complex absorption with a transit compartment (k _tr 2.78 h⁻¹), was selected as the structural model (see Figure S1). Goodness of fit plots showed that the structural model adequately described the naloxegol plasma concentration vs. time profiles observed in the two phase 3 studies (excluding patients with impaired hepatic function and patients receiving concomitant strong/moderate CYP3A4 inhibitors or strong CYP3A4 inducers; see Figures S2 and S3), and subjects who received naloxegol concomitant with ketoconazole (strong CYP3A4 inhibitor), rifampicin (strong CYP3A4 inducer) or diltiazem (moderate CYP3A4 inhibitor) (see Figures S4, S5 and S6). At this stage, concomitant administration of moderate/strong CYP3A4 inhibitors or strong CYP3A4 inducers, liver function status and phase of clinical study were included in the model as structural covariates.

A variety of models were run to explore dose dependency of PK parameters and revealed that the shapes of the PK profiles of naloxegol showed apparent complex features during the absorption, but not disposition, period. Dosage influenced the shape of the profile but not the AUC. Indeed, in the single ascending dose study, the AUC values determined via a non‐compartmental model were found to increase proportionally with dose. Dose dependence on the first order absorption rate constants, central and peripheral volumes of distribution and inter‐compartment clearance parameters improved the model fit but naloxegol clearance was not dose dependent. No outliers were observed in the base model run.

Effect of covariates on PK parameters (final model)

Following identification of the base model, the effects of pre‐specified covariates on principal PK parameters (CL/F and V _c/F) were investigated. Mean (standard error) population parameter estimates for CL/F and V _c/F were 115 (3.41) l h^–1 (IIV 48%) and 160 (27.4) l (IIV 51%), respectively (Table 3). Residual error was 44% in phase 1 and 2b studies and 56% in phase 3 studies. The final model identified race (Black) and concomitant administration of P‐gp inducers or inhibitors to have statistically significant effects on naloxegol CL/F. Covariates determined to have a statistically significant effect on naloxegol V _c/F were age, gender, race (Asian), baseline opioid dose, naloxegol formulation, renal function (CL_cr) and concomitant administration of P‐gp inducers. Figure 1 presents several diagnostic plots for the final model of naloxegol.The effect of the concurrent administration of proton‐pump inhibitors (PPI) was evaluated in a post hoc model run, which demonstrated a non‐significant decrease of ˂0.3 points in objective function value (see Figure S7).

Table 3.

Covariate parameter estimates in the final PK model of naloxegol

Parameter	Estimate (SE)
CL/F (l h–1)	115 (3.41); IIV = 48%
Strong CYP3A4 inducer – CL (l h–1)	317 (117)
Moderate CYP3A4 inhibitor – CL (l h–1)	74.7 (5.88)
Mild hepatic impairment – CL (l h–1)	110 (11.9)
Moderate hepatic impairment – CL (l h–1)	126 (17.1)
Phase 3 – CL (l h–1)	82.4 (2.21)
Race – Black on CL/F	0.265 (0.0573)
P‐gp inhibitor on CL/F	−0.343 (0.0548)
P‐gp inducer on CL/F	2.14 (1.13)
V c/F (l)	160 (27.4); IIV = 51%
C3HS – V c/F	0.124 (0.0122)
Phase 3 – V c (l)	277 (52.4)
Age on V c/F	−0.209 (0.0848)
Gender on V c/F	−0.169 (0.0507)
Race – Asian on V c/F	−0.519 (0.0581)
Baseline opioid dose on V c/F	−0.107 (0.0385)
Baseline opioid maintenance drug (weak type) on V c/F	0.267 (0.139)
Naloxegol formulation on V c/F	1.37 (0.478)
CLcr on V c/F	0.109 (0.0485)
P‐gp inducer on V c/F	−0.237 (0.075)
Absorption rates (h−1)
k a1	4.56 (0.468)
k a2	0.416 (0.0117)
k tr	2.78 (0.41)

CL/F, apparent clearance; CL_cr, creatinine clearance; k _a1, first order absorption rate constant for first absorption compartment; k _a2, first order absorption rate constant from the transit to central compartment; k _tr, first order absorption rate constant for the delay compartment; IIV, inter‐individual variability; P‐gp, P‐glycoprotein; PK, pharmacokinetics; SE, standard error; V _c/F, apparent volume of distribution of the central compartment

Figure 1.

Diagnostic plots for the final model of naloxegol: (A) observed vs. individual predicted, (B) observed vs. population predicted, (C) individual predicted vs. population conditional weighted residuals, (D) population predicted vs. individual weighted residuals, (E) subject identification vs. population conditional weighted residuals and (F) time since last dose vs. individual weighted residuals. CWRES, population conditional weighted residuals; IPRED, individual predicted; IWRES, individual weighted residuals; PRED, population predicted; ID, identification

Model evaluation

An internal PPC check was employed to evaluate the predictive capacity of the final model. Generated PPC plots showed predicted median naloxegol concentrations to be consistent with observed geometric mean concentrations, regardless of study phase. PPC plots for the phase 1, 2b and phase 3 studies at the 25 mg once daily dose are presented in Figure 2. Observed plasma concentrations were generally contained within the 90% PI, including those patients receiving concomitant strong/moderate CYP3A4 inhibitors/inducers. Results from the full stable model showed an under prediction for t _max was most obvious for the phase 2 25 mg data (Figure 2B). The diagnostic plots for other doses in this study showed a slight over‐prediction for the 5 mg data and a reasonable prediction for the 50 mg data (data not shown). Overall, no general trends or bias were seen in the predictions around t _max when the PPC plots for all studies and dosages were evaluated. Based on the PPC results, the population PK model developed for naloxegol was considered adequate for subsequent exposure–response modelling.

Figure 2.

Naloxegol plasma concentration vs. time profiles (posterior predictive check from final model) for phase 1 studies (A), phase 2b sub‐study (B) and two phase 3 studies (C), at the 25 mg once daily dose. Grey open circles represent observed individual data; black filled circles represent observed geometric mean data. LLOQ, lower limit of quantitation. ‐‐‐‐ Observed median and 90% CI, — Predicted median and 90% PI

Clinical relevance

A panel of clinical development experts who had a thorough understanding of naloxegol and OIC predefined clinical relevant covariates based on the drug properties and pharmacological activity. The panel included amongst its members a physician and a pharmacologist. In order to assess the clinical relevance of covariate effects on the PK profile of naloxegol, simulations using the 25 mg daily dosing regimen were conducted based on the final model. Also Forest plots were generated presenting the effects of covariates on the exposure (AUC and C _max) of naloxegol (Figure 3). Concomitant administration of naloxegol with strong CYP3A4 inducers decreased systemic exposure by 90%, while CYP3A4 strong inhibitors resulted in an eight‐fold increase in naloxegol systemic exposure. Moderate CYP3A4 inhibitors increased AUC and C _max by 60% and 30%, respectively. Concomitant administration of naloxegol with P‐gp inducers resulted in reduced naloxegol systemic exposure (by 60–70%), and P‐gp inhibitors increased systemic exposure by 40–50%. Compared with the phase 3 studies, the phase 1 or 2b studies had 30% lower AUC and 10% lower C _max values. Black subjects showed a 20% decrease in AUC and a 10% decrease in C _max, and Asian subjects had a 30% increase in C _max, compared with other races.

Figure 3.

Forest plots showing the effect of covariates on (A) AUC and (B) C _max of naloxegol (point estimate and 90% PI). Dashed lines represent (80%, 125%) interval. ALT, alanine aminotransferase; AUC, area under the plasma concentration vs. time curve; C _max, maximum plasma concentration; CL_cr, creatinine clearance; CYP3A4, cytochrome P450 3A4; P‐gp, P‐glycoprotein; PI, prediction interval; SD, standard deviation

Discussion

This analysis aimed to predict systemic exposure to naloxegol for exposure–response analyses and to identify covariates that impact on the PK profile of naloxegol through the development of a population PK model. Data were obtained from phase 1, 2b and 3 studies in healthy subjects or patients with non‐OIC and patients with OIC.

Various PK models, including an enterohepatic recycling model, were explored to describe the naloxegol plasma concentration vs. time profile. The initial structural model was a two compartment model with first order absorption and first order elimination. Based on the presence of a double peak in the naloxegol absorption profile observed in a phase 1 study 12, more complex absorption models were subsequently investigated. A two compartment disposition PK model, including two sites of absorption (one of which had a delay compartment), was found to describe adequately the naloxegol plasma concentration vs. time curves following oral administration. Goodness of fit plots confirmed that the model was consistent with PK data observed in the phase 1 DDI and phase 3 studies. Furthermore, PPCs based on simulations of 1000 model datasets indicated concordance with PK observations in the phase 1 DDI, 2b and 3 studies.

Examination of 23 pre‐specified covariates using the final model identified phase of clinical study, gender, race (Asian or Black), baseline opioid dose, naloxegol formulation, renal function (baseline CL_cr), strong and moderate CYP3A4 inhibitors, strong CYP3A4 inducers and P‐gp inhibitors and inducers to have a statistically significant effect on at least one PK parameter. The effect of these factors on the PK profile of naloxegol was typically small, with the exception of strong CYP3A4 inducers or inhibitors, whose concomitant administration with naloxegol resulted in up to 90% changes in systemic exposure (AUC and C _max). Of note, weak CYP3A4 inhibitors and inducers, and moderate CYP3A4 inducers, did not significantly impact the exposure profile of naloxegol. Further, although moderate CYP3A4 inhibitors were identified as a significant covariate, there was a high level of inter‐subject variability making it difficult to determine the clinical impact of the observed changes in exposure following concomitant administration of naloxegol with moderate CYP3A4 inhibitors.

P‐gp inducers and inhibitors were also identified as covariates with clinically significant effects on naloxegol exposure. However, P‐gp inducers and inhibitors also have activity as CYP3A4 inducers or inhibitors 21, somewhat confounding interpretation of these results as it is difficult to determine whether the significant impact on exposure results from activity on P‐gp or CYP3A4. Given the inter‐individual variability, a five‐fold or higher change in AUC or C _max was considered to be clinical meaningful. There was a strong correlation between the P‐gp and CYP3A4 inhibitor and inducer covariates.

Analyses of the clinical relevance of covariate effects also found no differences in systemic exposure (AUC and C _max) to naloxegol in patients with mild or moderate hepatic impairment, compared with patients with normal hepatic function. These data are perhaps surprising, given the extensive involvement of hepatic pathways in eliminating naloxegol. However, findings are consistent with those from a recent PK study showing that hepatically impaired patients had only a small decrease in naloxegol AUC (17–18%) and similar C _max, compared with healthy participants 14.

The PK profile of naloxegol showed apparent complex features during the absorption period. A variety of models were run to explore dosage dependence on PK parameters. The shapes of the PK profiles were changed at different dosages around absorption rather than disposition. Dose was introduced as a covariate for PK parameters k _a1, k _tr, V _c, V _p and Q that was centred at 50 mg using a power function with estimated exponents of −0.143, −0.35, −0.36, −0.167 and −0.129, respectively.

Exposure to naloxegol was found to be approximately 30% higher in phase 3 studies compared with phase 1 and 2b studies. However, there are several confounding factors that might influence interpretation of these data. First, there were a limited number of samples available from phase 3 studies. Further, the differences could be related to the phase 3 trials themselves rather than the study population as a whole, with potential errors/uncertainty in dosing and/or sampling times, different patterns of food consumption or variation in underlying medical conditions being observed across different trial populations. Food causes a 45% increase in the exposure of naloxegol 22 and may partially contribute to the high AUC seen in the phase 3 studies if patients were not fully compliant with the dosing instructions. Consequently, the validity of phase 3 studies as a significant covariate should be treated with caution.

Black subjects showed a slightly lower exposure to naloxegol with AUC_ss decreased by approximately 20% compared with Asian and other populations (Figure 3C). Naloxegol is metabolized by CYP3A4 and CYP3A5 and it is known that CYP3A5 is polymorphically expressed at a higher frequency in African Americans 23. The observed racial effect is probably caused by this CYP3A5 polymorphism.

In conclusion, the population PK model adequately described naloxegol phase 1, 2b and 3 data. The model provided individual estimates of AUC, C _max and average plasma concentrations for the duration of the phase 3 studies, which were further used in exposure–response analyses. Of all covariates tested using the model, strong CYP3A4 inhibitors and inducers showed marked effects on the exposure of naloxegol. This model represents a tool for providing guidance on potential covariates with a clinically meaningful impact on the PK profile of naloxegol and may be useful in directing optimal usage of naloxegol.

Competing Interests

All authors have completed the Unified Competing Interest form at http://www.icmje.org/coi_disclosure.pdf (available on request from the corresponding author) and declare NAH, HX, KHB and MS are employees of the study sponsor, AstraZeneca, and hold stock/shares with AstraZeneca. SC is an employee of Ann Arbor Pharmacometrics Group, Ann Arbor, MI, USA and reports personal fees from AstraZeneca, outside the submitted work.

AstraZeneca LP, the manufacturer of naloxegol, sponsored this study.

The authors thank Ioana Edgeworth, PhD (QXV Communications, Macclesfield, UK) for providing writing support which was funded by AstraZeneca.

Author contributions

Conception and design of the work: NAH, KHB, MS. Analysis and interpretation of data: NAH, SC, HX, KHB, MS. Drafting or revising the manuscript: NAH, SC, HX, KHB, MS. Final approval of the manuscript: NAH, SC, HX, KHB, MS.

Supporting information

Table S1. Summary of stepwise covariate selection for forward addition of covariates to the base model

Figure S1. Diagram of the structural model

Figure S2. Goodness of fit plots for 12.5 mg naloxegol in phase 3 studies NCT01309841 (left panel) and NCT01323790 (right panel) excluding subjects with impaired liver function or on concomitant strong or moderate cytochrome P450 3A4 inhibitors or strong cytochrome P450 3A4 inducers: observed, individual predicted (IPRED) and population predicted (PRED) concentrations vs. time

Figure S3. Goodness of fit plots for 25 mg naloxegol in phase 3 studies NCT01309841 (left panel) and NCT01323790 (right panel) excluding subjects with impaired liver function or on concomitant strong or moderate cytochrome P450 3A4 inhibitors or strong cytochrome P450 3A4 inducers: observed, individual predicted (IPRED) and population predicted (PRED) concentrations vs time

Figure S4. Goodness of fit plots in phase 1 drug–drug interaction study (NCT01520896) for subjects given 25 mg naloxegol without (left panel) or with (right panel) concomitant ketoconazole: observed, individual predicted (IPRED) and population predicted (PRED) concentrations vs. time

Figure S5. Goodness of fit plots in phase 1 drug–drug interaction study (NCT01533870) for subjects given 25 mg naloxegol without (left panel) or with (right panel) concomitant rifampicin: observed, individual predicted (IPRED) and population predicted (PRED) concentrations vs. time

Figure S6. Goodness of fit plots in phase 1 drug–drug interaction study (NCT01594619) for subjects given 25 mg naloxegol without (left panel) or with (right panel) concomitant diltiazem: observed, individual predicted (IPRED) and population predicted (PRED) concentrations vs. time

Figure S7. Box plots of apparent clearance of naloxegol by concomitant use of proton pump inhibitors (PPI)

supporting info item

Al‐Huniti, N. , Chapel, S. , Xu, H. , Bui, K. H. , and Sostek, M. (2016) Population pharmacokinetics of naloxegol in a population of 1247 healthy subjects and patients. Br J Clin Pharmacol, 81: 89–100. doi: 10.1111/bcp.12756.