SEARCH

SEARCH BY CITATION

Keywords:

  • population pharmacokinetics;
  • renal impairment;
  • smoking cessation;
  • varenicline

Abstract

  1. Top of page
  2. Abstract
  3. Introduction
  4. Methods
  5. Results
  6. Discussion
  7. Competing interests
  8. REFERENCES

WHAT IS ALREADY KNOWN ABOUT THIS SUBJECT?

• Several clinical pharmacology studies have characterized the pharmacokinetics of varenicline in young adult and elderly smokers and subjects with impaired renal function.

• Varenicline pharmacokinetics is linear over the recommended dose range.

• Varenicline total clearance is linearly related to its renal clearance.

• Both are progressively reduced as renal function declines, which results in a progressive increase in varenicline systemic exposure and prolonged half-life.

WHAT THIS STUDY ADDS?

• This work provides an integrated model-based analysis of varenicline pharmacokinetics across multiple studies in the target patient population.

• The model describes the impact of patient-specific covariates, such as renal function, and provides a rationale for dose adjustment.

• The resulting model also provides a means to predict individual-specific drug exposures to clinical responses in subsequent analyses.

AIMS

To characterize the population pharmacokinetics of varenicline and identify factors leading to its exposure variability in adult smokers.

METHODS

Data were pooled from nine clinical studies consisting of 1878 subjects. Models were developed to describe concentration–time profiles across individuals. Covariates were assessed using a full model approach; parameters and bootstrap 95% confidence intervals (CI) were estimated using nonlinear mixed effects modelling.

RESULTS

A two-compartment model with first-order absorption and elimination best described varenicline pharmacokinetics. The final population parameter estimates (95% CI) were: CL/F, 10.4 l h−1 (10.2, 10.6); V2/F, 337 l (309, 364); V3/F, 78.1 l (61.9, 98.9); Q/F, 2.08 l h−1 (1.39, 3.79); Ka, 1.69 h−1 (1.27, 2.00); and Alag, 0.43 h (0.37, 0.46). Random interindividual variances were estimated for Ka[70% coefficient of variation (CV)], CL/F (25% CV), and V2/F (50% CV) using a block covariance matrix. Fixed effect parameters were precisely estimated [most with % relative standard error < 10 and all with % relative standard error < 25], and a visual predictive check indicated adequate model performance. CL/F decreased from 10.4 l h−1 for a typical subject with normal renal function (CLcr = 100 ml min−1) to 4.4 l h−1 for a typical subject with severe renal impairment (CLcr = 20 ml min−1), which corresponds to a 2.4-fold increase in daily steady-state exposure. Bodyweight was the primary predictor of variability in volume of distribution. After accounting for renal function, there was no apparent effect of age, gender or race on varenicline pharmacokinetics.

CONCLUSIONS

Renal function is the clinically important factor leading to interindividual variability in varenicline exposure. A dose reduction to 1 mg day−1, which is half the recommended dose, is indicated for subjects with severe renal impairment.


Introduction

  1. Top of page
  2. Abstract
  3. Introduction
  4. Methods
  5. Results
  6. Discussion
  7. Competing interests
  8. REFERENCES

Varenicline, which is approved in over 80 countries worldwide to date as an aid to smoking cessation in adults, is a highly selective partial agonist of the α4β2 nicotinic acetylcholine receptor [1, 2]. Found mainly in the brain, this receptor mediates the reinforcing properties of nicotine. Because of its mixed agonist–antagonist effects, varenicline should reduce the severity of nicotine withdrawal symptoms and nicotine craving, while also reducing the satisfaction associated with smoking. The efficacy and safety of varenicline 1 mg b.i.d. have been established in large clinical trials [3–5]. The approved dosing regimen in adults is 1 mg twice daily (b.i.d.) for 12 weeks, starting with a 1-week titration.

In humans, varenicline exhibits linear kinetics following oral administration [6, 7]. Maximum plasma concentration typically occurs within 3–4 h post dose. After attaining Cmax, plasma varenicline concentrations decline in a multiphasic manner with an elimination half-life of about 24 h. Varenicline is almost exclusively excreted unchanged in the urine, primarily via glomerular filtration with an additional component of active tubular secretion via the human organic cation transport system [8, 9]. As expected, decrease in total clearance of varenicline correlated well with its decreasing renal clearance, and increases in varenicline systemic exposure showed an evident dependence on the degree of renal impairment [10]. Plasma protein binding of varenicline is low (<20%) and independent of age or renal function [10, 11].

The objective of this pooled analysis was to characterize the population pharmacokinetics (PK) of varenicline in adult smokers using nonlinear mixed effects and to identify the factors leading to its variability in exposure in this target patient population. This investigation primarily examined the effects of selected demographic or physiological factors (age, weight, gender, race and renal function) on interindividual differences in varenicline pharmacokinetics. Additionally, pharmacokinetic information generated in this analysis was utilized in subsequent population pharmacokinetic–pharmacodynamic (PK–PD) analyses of tolerability and efficacy measures.

Methods

  1. Top of page
  2. Abstract
  3. Introduction
  4. Methods
  5. Results
  6. Discussion
  7. Competing interests
  8. REFERENCES

Clinical studies and study populations

Data from 1878 adult smokers were pooled from nine varenicline clinical trials for the population PK analysis. All studies were approved by an Ethics Committee or Institutional Review Board and were conducted in accordance with the current revision of the Declaration of Helsinki. All subjects provided written informed consent after full explanation of study details and before any study-related procedures were carried out.

The study designs, study populations, and timing of blood samples are summarized in Table 1. Varenicline doses were given orally as immediate-release tablets. Doses ranged from 0.3 to 3 mg day−1 administered once (q.d.) or twice (b.i.d.) daily. Dose and sampling times for the clinical studies type I were pre-specified in the protocols. For the larger clinical trials, study types II and III, investigators were asked during the clinic visit to record dates and times of the random samples as well as prior doses.

Table 1.  Summary of all varenicline studies used in the population pharmacokinetic analysis
DesignDurationNumber of subjectsPharmacokinetic samplingDose regimen
  1. I, II and III refer to different phases of the clinical drug development of varenicline. ET, early termination, i.e. at premature exit evaluation/visit; q.d., once daily dosing; b.i.d., twice daily dosing; hpd, hour post dose.

Study I-112 days24 subjects with mild to severe renal impairmentDay 1: 0, 1, 2, 3, 4, 6, 8, 12, 16 hpd Days 2, 4: 0 and Days 7, 10: 0, 3 hpd Day 12: 0, 1, 2, 3, 4, 6, 8, 12, 16, 24, 36, 48, 72, 96, 120, 144, 168, 192 hpd0.5 mg q.d.
Study I-27 days16 elderly (≥65 years old) adult smokersQ.d. group: Day 1: 0, 1, 2, 3, 4, 6, 8, 12, 16, 24 hpd Days 4, 5, 6: 0, 3 hpd B.i.d. group: Day 1: 0, 1, 2, 3, 4, 6, 8, 12 hpd Day 4 (0, 3, 12, 15 hpd), days 5 and 6: 0, 3 hpd Q.d. and b.i.d. groups: Day 7: 0, 1, 2, 3, 4, 6, 8, 12, 16, 24, 36, 48, 72, 96, 120, 144, 168 hpd1 mg q.d., 1 mg b.i.d.
Study I-321 days119 adult smokersDays 1, 8 and 15: 0, 0.25 and 2 hpd Days 2, 9 and 16: 0, 0.5 and 3 hpd Days 3, 10 and 17: 0, 1 and 4 hpd Days 4, 11 and 18: 0, 0.75 and 6 hpd Days 6, 13 and 20: 0 Day 7: 0, 1, 3 and 8 hpd Day 14: 0, 1, 2, 3, 4, 8 and 12 hpd Day 21: 0, 1, 2, 3, 4, 8, 12, 24, 48, 72 and 96 hpd1 mg b.i.d., 1.5 mg b.i.d.
Study I-47 days39 adult smokersDay 7: 0, 1, 2, 3, 4, 8, 10, 12, 14, 15, 16, 17, 18, 22, 24, 26 and 38 hpd2 mg q.d.
Study II-17 weeks368 adult smokersWeek 1, week 2 and week 4 (or ET)0.3 mg q.d., 1 mg q.d., 1 mg b.i.d.
Study II-212 weeks490 adult smokersWeek 1, week 2, week 4 and week 12 (or ET)0.5 mg b.i.d., 1 mg b.i.d.
Study III-112 weeks301 adult smokersWeek 2 and week 12 (or ET)1 mg b.i.d.
Study III-212 weeks301 adult smokersWeek 2 and week 12 (or ET)1 mg b.i.d.
Study III-352 weeks220 adult smokersWeek 2, week 12, week 24, week 36 and week 52 (or ET)1 mg b.i.d.

Bioanalytical methods

Plasma samples were analysed for varenicline content using a validated assay employing liquid–liquid extraction followed by high-performance liquid chromatography/tandem mass spectrometry [7]. The assay had a dynamic range extending from 0.100 to 50.0 ng ml−1, and plasma varenicline concentrations below the lower limit of quantification (BLQ) were reported as < 0.100 ng ml−1. Plasma varenicline concentrations (664 observation records; 4% of total) that were reported as BLQ values were removed from the analysis dataset, a strategy advocated when the fraction of missing data points is small [12]. These included pre-first dose observations as well as other time points.

Data analysis

The population PK analysis was performed using a qualified installation of the nonlinear mixed effects modelling software, nonmem Version V, Level 1.1 (ICON Development Solutions, Ellicott City, MD, USA) and the nmtran subroutines version III level 1.1, and PREDPP model library version IV level 1.1 [13]. The first-order conditional estimation method with η-ε interaction (FOCE-INT) was employed for all model runs.

The initial modelling work investigated an open one-compartment (ADVAN2, TRANS2) vs. two-compartment (ADVAN4, TRANS4) PK model in their ability to characterize the pharmacokinetics of single and multiple doses of varenicline. Since no reference intravenous data were available, the absolute bioavailability of varenicline was not identifiable and the base model was parameterized in terms of apparent clearance (CL/F; l h−1), apparent volumes of distribution (central = V2/F; peripheral = V3/F; l), apparent intercompartmental clearance (Q/F; l h−1) and first-order absorption rate constant (Ka; h−1). An absorption lag time (Alag, h) was also estimated, given that sufficient data describing the absorption phase for varenicline were available. Thus, the base model was defined as the structural PK model with appropriate random effects (defined below) and no covariates.

Intersubject variability in the PK parameters was modelled assuming a log-normal parameter distribution (Equation 1). Pi is the estimated pharmacokinetic parameter for individual i, inline image is the typical value of the parameter, ηPi are individual-specific random effects for individual i and parameter P and are assumed to be normally distributed with mean 0 and covariance matrix Ω. The estimate of intersubject variance was converted to an approximate percent coefficient of variation (% CV). An attempt was made to define a full block covariance matrix for the interindividual random effects (Ω) when possible.

  • image(1)

Residual variability in this hierarchical mixed-effects model was described as a combined additive and proportional error model (Equation 2). Cij is the jth observed plasma varenicline concentration in individual i, Ĉij is the jth model predicted plasma varenicline concentration based on the pharmacokinetic model and individual parameters in individual, i.e. εpij and εaij are the proportional and additive error components of residual variability, respectively, for individual i and measurement j and were assumed to be normally distributed with mean 0 and variance σ12 or σ22, respectively.

  • image(2)

Model selection was guided by various goodness-of-fit criteria [14–16], including diagnostic scatter plots, convergence with at least three significant digits, plausibility of parameter estimates, precision of parameter estimates, parameter estimation error correlation <0.95, and the Akaike's information criterion, given the minimum objective function value and number of estimated parameters. Final model parameter estimates were reported with a measure of estimation uncertainty based on stratified nonparametric bootstrap 95% confidence intervals (CI) [17, 18].

Covariate modelling

A covariate modelling approach emphasizing parameter estimation rather than stepwise hypothesis testing was implemented for this population PK analysis. The analysis of the effect of various (demographic or physiological) covariates on varenicline PK parameters was conducted by developing a full covariate model, specified a priori. First, covariate–parameter relationships were identified based on scientific interest, mechanistic plausibility or prior knowledge; then the full model was constructed with care to avoid inclusion of collinear or correlated covariates. This full model approach is a simplification of the global model approach described by Burnham and Anderson [19], and is preferable when the objective is to develop models for effect estimation [20]. An attempt was made to incorporate known physiological relationships into the covariate–parameter models. In general, continuous covariates were normalized on a typical reference value (approximately the population median) and included in the model using a power function (Equation 3). The effects of categorical covariates were similarly described, as shown below.

  • image(3)

The typical value of a model parameter (TVP) was described as a function of m individual continuous covariates (covmi) and p individual categorical (0 or 1) covariates (covpi) such that qn is an estimated parameter describing the typical PK parameter value for an individual with covariates equal to the reference covariate values (covmi = refm, covpi = 0), while q(m+n) and q(p+m+n) are estimated parameters describing the magnitude of the covariate-parameter relationship.

Covariate effects were added to the model in a multiplicative fashion. Inferences about the clinical importance of covariate effects were made based on the magnitude and precision of covariate parameter estimates. The predefined covariates of interest included the effects of race, body size and renal function on CL/F, body size, age, and race on V2/F, and body size on both V3/F and Q/F. Any remaining trends related to correlated covariates (e.g. gender and weight) were also explored with the full model by fixing all structural and covariate model parameters to their final estimates with the exception of the correlated covariate of interest. Effects of concomitant medications were not assessed in this population PK analysis as drug–drug interactions of particular interest with varenicline were investigated in specific PK studies. Because of the simple metabolic and dispositional properties of varenicline, no clinically meaningful pharmacokinetic drug–drug interactions have been identified [8–10].

Model evaluation

The final population PK model was evaluated using a stratified nonparametric bootstrap and a predictive check. For the nonparametric bootstrap procedure [17], 1000 replicate datasets were generated by random sampling with replacement using the individual as the sampling unit, stratified by race (White, Black and Other) and renal function (CLcr <30 ml min−1, 30 ≤ CLcr < 50 ml min−1, 50 ≤ CLcr < 80 ml min−1, 80 ≤ CLcr ≤ 120 ml min−1, and 120 < CLcr ≤ 150 ml min−1). Population parameters for each dataset were subsequently estimated using the final model; parameter uncertainty was then expressed as 95% CI about the estimate, by observing the 0.025th and 0.975th quantiles of the resulting bootstrap parameter distributions from runs with successful convergence [18]. For the predictive check, 500 Monte-Carlo simulation replicates of a subset of the population PK dataset were generated using the final model and point-estimates of the population fixed and random parameters [21]. For the observed data and for each of the simulation replicates, a summary metric of interest, the average of concentration within each individual (Cavg), was calculated [22]. The entire distribution of simulated Cavg values for each replicate was then compared with the distribution of observed Cavg values using overlaid quantile–quantile plots. This method of predictive check assessment is a rigorous evaluation of the model's ability to describe adequately the entire population distribution of the metric of interest, and is not sensitive to imbalance in sampling across individuals; both of these issues have been identified as problematic characteristics of more simplistic concentration–time type visual predictive checks [23].

Results

  1. Top of page
  2. Abstract
  3. Introduction
  4. Methods
  5. Results
  6. Discussion
  7. Competing interests
  8. REFERENCES

A total of 1878 adult smokers [954 (50.8%) male; 924 (49.2%) female] contributing 11 935 plasma varenicline concentrations, ranging from 0.100 to 28.3 ng ml−1 (about 50% within the concentration range observed at 1 mg b.i.d.), were included in the population PK analysis. The characteristics of the population studied are summarized in Table 2. About 15% of the subjects in the database had mild to severe renal impairment (as defined by creatinine clearance estimated using the Cockroft–Gault formula [24]). Estimated creatinine clearance values were truncated at an upper value of 150 ml min−1, thus removing very high and physiologically improbable values generated by the Cockroft–Gault (Equation 24).

Table 2.  Summary of subject demographics
Baseline characteristic (units)MeanMedianRange
  • *

    Estimated from a serum creatinine measurement, total body weight, age and gender using the Cockroft–Gault formula [24].

Age (years)44.044.218–76
Height (cm)171170135–202
Total body weight (kg)78.077.041.0–129
Body mass index (kg m−2)26.626.016.0–44.8
Estimated creatinine clearance* (ml min−1)11210715.6–268
SexMale = 954; female = 924
RaceWhite = 81.0%; Black = 12.6%; Asian = 1.22%; Other = 5.22%

Total body weight (WT), ideal body weight (IBW), height (HT) and body mass index (BMI) were considered as potential size descriptors and were strongly correlated with correlation coefficients (r) of at least 0.55; BMI was, however, not correlated with HT or IBW (r < 0.06). As expected, estimated creatinine clearance (CLcr) as an indicator of renal function was correlated with observed weight (r = 0.65). Age was not strongly correlated with any other covariates. Continuous covariate distributions, when viewed across the sex and race categories, appeared similar, although women were generally shorter and had lower median body weight.

Individual varenicline concentration–time data are presented by study in Figure 1. Full pharmacokinetic profiles could be obtained for all clinical studies type I as a result of intensive blood sampling, whereas in the larger type II and III studies data points were observed within approximately 200 h (week 1) to 400 h (week 2) post time of first dose with other large groupings of sampling times between 600 h (week 4) and 2000 h (week 12) after the first dose administered due to random sampling throughout the 12-week treatment period. Distributions in the varenicline plasma concentrations were comparable from week 1 or week 2 onwards (following an initial titration week) suggesting that steady-state conditions, consistent with a terminal elimination half-life of approximately 24 h, have been attained. Moreover, varenicline pharmacokinetics appeared to be stationary over time, as indicated by the constant ranges in the varenicline plasma concentrations measured over 52 weeks of 1 mg b.i.d. dosing (Figure 1).

image

Figure 1. Observed varenicline plasma concentrations vs. time, by study

Download figure to PowerPoint

Initial modelling efforts focused on the data obtained from studies I and II. Based on a preliminary structural model evaluation and graphical investigation of varenicline concentration–time profiles, a two-compartment PK model with first-order absorption (and a lag time) and elimination was chosen as the best structural model for varenicline pharmacokinetics. Variations on the PK structural model and interindividual and residual variance–covariance structures were explored, as driven by model selection diagnostics. In vitro kinetic studies showed that passive diffusion primarily contributes to the renal transport of varenicline along with some minor component of active transport via the organic cation transport system [8]. Varenicline was characterized as a low-to-moderate substrate for hOCT2; and although data spanned varenicline doses from 0.3 to 3 mg day−1, there was no evidence of apparent dose-dependence in any of the model diagnostics, when viewed as a function of dose. Predefined covariates were added to the base model simultaneously to form the full model, and the parameters and precision of the parameters were estimated.

A summary of key model-building steps for varenicline pharmacokinetics is presented in Table 3. The dependence of varenicline clearance on renal function (CRCL) was modelled by means of an estimated power model, as function of estimated CLcr normalized on a typical reference value of 100 ml min−1. This model was consistent with the absorption, distribution, metabolism, and excretion information, which indicated that varenicline undergoes minimal metabolism with >90% excreted unchanged in the urine [9]. The effect of body size was investigated as a potential predictor for V2/F and described as a function of WT, normalized by the typical reference weight of 70 kg. The observed range of individual body weights in the analysis database was deemed sufficient to allow estimation of this covariate effect; results showed that the average population estimate for the power model effect of weight was 0.77 [19% relative standard error (RSE)] with a bootstrap 95% CI including the allometric value of 1 (Table 4). The model also contained a power function for the effect of body weight on V3/F and Q/F using fixed allometric exponents of 1 on V3/F and 0.75 on Q/F[25]. More stable models were obtained without interindividual variability on both these parameters during the model-building process, thus suggesting that the data did not provide sufficient information to fully support estimation of the variability in the peripheral distribution process. Alternative predictors of body size were investigated as substitutes for weight in the full model, but neither IBW nor BMI resulted in any improvements in goodness-of-fit. Race entered the model as power functions with a separate dichotomous (0, 1) covariate serving as an on–off switch for each category. Because the number of subjects in the Hispanic, Asian and Other races was small (≤5% of total population studied), these categories were grouped together. Age was included as a power function, normalized by the reference age of 45 years. Gender did not add further to the model. The final forms of the equation for the model are given below:

Table 3.  Summary of covariate model building steps for varenicline pharmacokinetics
ModelDescription (reference model)MOFΔMOFAIC
  1. MOF, minimum objective function; ΔMOF, change in MOF; AIC, Akaike's information criterion.

Base model
013Base model – reduced omega block (Ka, CL, V2) and V3/Q (0, fixed)21016.921 044.94
Full model – selection of demographic and physiological covariates
016Full model with primary covariates (CRCL, WT, race)20563.020 599.03
042Full model no. 016 with two distinct error structure models19740.5−82319 780.52
043Full model no. 042 – reduced omega block (Ka, CL, I2) and V3/Q (0, fixed)19719.5−2119 763.53
Final model
059Final model with (CRCL/race on CL/F, and WT/race/age on V2/F)19715.9−3.619 761.89
Table 4.  Final model parameter estimates
Pharmacokinetic parameterEstimate%RSE*Bootstrap 95% CI (lower, upper)
  • *

    %RSE, percent relative standard error of the estimate = SE/|parameter estimate| × 100.

  • †95% confidence interval (CI) of the parameter estimate derived from a nonparametric bootstrap analysis.

CL/F (l h−1)
θCL10.40.9(10.2, 10.6)
θCRCL0.545.6(0.48, 0.59)
θBlack1.162.1(1.11, 1.21)
θOther1.113.3(1.04, 1.18)
V2/F (l)
θV23374.4(309, 364)
θWT0.7718.7(0.50, 1.05)
θAGE0.1354.1(−0.01, 0.30)
θBlack0.924.6(0.83, 1.00)
θOther0.7110.4(0.58, 0.89)
V3/F (l)
θV378.112.7(61.9, 98.9)
θWT1 (Fixed)
Q/F (l h−1)
θQ2.0822.2(1.39, 3.79)
θWT0.75 (Fixed)
Ka (h−1)
θKa1.699.2(1.27, 2.00)
Alag (h)
θAlag0.434.6(0.37, 0.46)
Interindividual variance
ω2CL0.061 (24.7% CV)6.7(0.054, 0.069)
ω2V20.25 (50.0% CV)25.3(0.15, 0.40)
ω2Ka0.49 (70.1% CV)38.9(0.23, 0.97)
covKa-V20.24 (r = 0.67)54.0(−0.05, 0.53)
covCL-V20.006 (r = 0.05)133(−0.013, 0.021)
covKa-CL−0.009 (r = −0.05)186(−0.045, 0.040)
ω2V30 (Fixed)
ω2Q0 (Fixed)
Residual variance
σ2add, 10.28 (SD = 0.5)23.5(0.155, 0.358)
σ2prop, 10.030 (17.2% CV)10.7(0.024, 0.036)
σ2add, 24.38 (SD = 2.1)21.0(0.047, 5.89)
σ2prop, 20.046 (21.5% CV)29.3(0.025, 0.128)
  • image
  • image
  • image
  • image
  • image
  • image

Interindividual variability was described for CL/F, V2/F and Ka with a full BLOCK structure. The model utilized separate combined additive and proportional error models to describe distinct random residual variability for the data arising from different studies with volunteers vs. large patient trials. This was to reflect greater uncertainty about process noise, compliance, and the accuracy of the timing of blood samples and drug administration in outpatient setting typical of large clinical trials compared with the more controlled volunteer study settings. Furthermore, differences were present in the density of sampling as well as the dosing history for the data obtained from subjects in these two settings.

Parameter estimates, including measures of parameter estimation uncertainty, from the final model are presented in Table 4. Diagnostic plots revealed that the model was consistent with the observed data, although there may be a slight underprediction of varenicline high concentrations (Figure 2). Conditional weighted residuals were generally well scattered across the range of predicted concentrations (Figure 3) and time after dose (Figure 4) indicating no overall systematic bias in the model diagnostics. The data points with positive bias for study II-2 were associated with PK samples for which the exact dosing times during the initial titration week were unknown. The typical population PK parameter estimates (bootstrap 95% CI) given the reference covariates effects (White, 70 kg, 45 years old) were 10.4 l h−1 (10.2, 10.6) for CL/F, 337 l (309, 364) for V2/F, 78.1 l (61.9, 98.9) for V3/F, 2.08 l h−1 (1.39, 3.79) for Q/F, 1.69 h−1 (1.27, 2.00) for Ka and 0.43 h (0.37, 0.46) for Alag. Typical value parameters for CL/F, V2/F, V3/F, Q/F and Ka were estimated with good precision (most with RSE <10% and all with RSE <25%). Unexplained random interindividual variability (% CV) was reduced for CL/F (25%), V2/F (50%) and Ka (70%) in the final model when compared with the base model CL/F (33%), V2/F (68%) and Ka (84%) variance estimates. The interindividual variance on CL/F was estimated with the greatest precision (6.7% RSE). Also, covariance terms describing the correlations between these intersubject random effects were greatly reduced in the final model, suggesting that large fractions of the total interindividual variability in CL/F, V2/F and Ka have been explained; hence correlations were not as well estimated, reflecting their reduced significance in the model. The estimated η-shrinkage, calculated as 1 – SD(η)/ω, where η are the interindividual random effects and ω is the square-root of the estimated interindividual variance, was 21% for CL/F and 55% for V2/F[26]. The estimated shrinkage may be attributed to the substantial contribution of concentration–time data obtained from the large trials with sparse sampling collection; two to four samples per subject were collected over the 12-week course of administration (Table 1).

image

Figure 2. Observed vs. population predicted varenicline plasma concentrations, by study. Observed plasma concentration vs. population predictions are indicated by open circles with a dotted line connecting individual data points. The line of identity (solid) is included as a reference

Download figure to PowerPoint

image

Figure 3. Conditional weighted residuals vs. population predicted varenicline plasma concentration, by study. Residuals are indicated by open circles with a dotted line as a smoothing spline trend line through the data. A solid line at Y = 0 is included as a reference

Download figure to PowerPoint

image

Figure 4. Conditional weighted residuals vs. time after dose by study. Residuals are indicated by open circles with a dotted line as a smoothing spline trend line through the data. A solid line at Y = 0 is included as a reference

Download figure to PowerPoint

The model evaluation results demonstrated, overall, good estimation and predictive performance of the final model. Typical structural model parameters and random variance terms were estimated precisely, while some covariate effects were less precisely estimated (Table 4). A total of 11 observations from the final model analysis were identified as potential outliers based on conditional weighted residuals, i.e. |CWRES| > 5. The subjects contributing these observations were not identified as having extreme age, weight or CLcr values and the range in varenicline concentrations was from 4.36 to 12.7 ng ml−1. Since the influence of these few outlying observations on the fixed effects is unlikely to be significant, they were allowed to remain in the dataset. Predictive check of studies III-1 and III-2 demonstrated that the distribution of the simulated Cavg values was consistent with the distribution of the same metric derived from the observed data (Figure 5a), although some evidence of slight negative bias was observed. Additional checking using steady-state exposure over the dosing interval, AUC0–τ, as the metric of interest across the pool of all four Studies I (Figure 5b) showed a comparable performance of the final PK model to that seen in Figure 5a. Although there were too few subjects with mild, moderate or severe renal impairment to support a comparison of AUC distributions, model performance was assessed at the expected exposure range for these patients. The observed geometric mean AUC0–τ values, shown as vertical lines on the plot, were 58.7, 84.4 and 114.9 ng*h ml−1 for patients with mild, moderate and severe renal impairment, respectively (Figure 5b). It can be seen that the model performance is reasonable throughout the observed range of the AUC distribution. Overall, these results illustrate the reliability of the final population fixed and random effect components of the model for descriptive and Monte Carlo simulation purposes.

image

Figure 5. (a) Results of simulation predictive check of a subset (Studies III-1 and III-2) of the population pharmacokinetic (PK) database. The predictive check results are presented as quantile–quantile plots of simulated vs. observed average varenicline concentrations within each individual (Cavg), with results overlaid from each of 500 simulated replicates of the original population PK data. The solid line is the reference line of identity. (b) Results of simulation predictive check of the four Studies I of the population PK database. The predictive check results are presented as quantile–quantile plots of simulated vs. observed AUC0–τ values, with results overlaid from each of 500 simulated replicates of the original population PK data. The solid line is the reference line of identity. Vertical lines (left to right) represent the geometric mean AUC0–τ values following repeat administration of varenicline 0.5 mg q.d. to patients with mild, moderate or severe renal impairment, respectively

Download figure to PowerPoint

Discussion

  1. Top of page
  2. Abstract
  3. Introduction
  4. Methods
  5. Results
  6. Discussion
  7. Competing interests
  8. REFERENCES

This integrated analysis of pooled studies across the varenicline clinical programme provided an understanding of the pharmacokinetic properties of varenicline in the target adult smoking population and its sources of interindividual variability in exposure. An open two-compartment model with first-order absorption and elimination adequately described individual concentration–time data following single and multiple dosing. Diagnostic plots revealed no apparent time- or concentration-dependencies in varenicline pharmacokinetics. The final model included estimates of covariate effects for renal function and race on apparent clearance, and for weight, age and race on central volume of distribution (Table 4, Figure 6). The addition of these covariate factors as predictors of variability in varenicline exposure did result in a reduction of unexplained interindividual variability in the parameters CL/F (25% CV) and V2/F (50% CV) when compared with the base model. In all, observed covariate factors in the final PK model described a large fraction of the total observed interindividual variability in both apparent clearance {44.5% [calculated as (ω2base model – ω2full model)/ω2base model %]} and central volume of distribution (45.6%) of varenicline. It is possible, however, that other covariate factors, not included in this database, might explain some of the remaining variability.

image

Figure 6. Covariate effects on varenicline pharmacokinetic parameters. Magnitude and precision of covariate effects are illustrated relative to the typical individual. Points and boxes represent effects given the maximum likelihood parameter estimates and observed covariate range, while whiskers represent precision (95% CI) of covariate effects

Download figure to PowerPoint

Apparent clearance of varenicline, CL/F, tended to increase with increasing body weight and to decrease with age. These effects are expected, based on the physiological principles that clearance increases with the size of clearing organs (the kidney in the case of varenicline), and that ageing typically results in a decrease in organ function [27–29]. It was also evident that renal function alters substantially the pharmacokinetics of varenicline; with declining renal function, CL/F progressively decreased, which correlated well with the progressive increase in varenicline systemic exposure. Since varenicline is almost entirely eliminated by the kidney (>90%), the effect of renal function on the clearance is not unexpected. In this database, estimated CLcr was highly correlated with weight and age, two other covariates of interest for predicting CL/F. This was expected since the Cockroft–Gault formula, used to estimate the subject's creatinine clearance in this study, utilizes weight and age in the formula [24]. Creatinine clearance was chosen because of its physiological significance, and as a result of its inclusion as a covariate for varenicline clearance the relationships between CL/F parameter random effects and weight, renal function and age were no longer evident. Hence, weight and age do not appear to have separate effects on varenicline pharmacokinetics, beyond what is described by the covariate CLcr. The lack of an effect of age (apart from the age-related loss of renal function) is indeed consistent with noncompartmental analysis results indicating that varenicline pharmacokinetic parameters and systemic exposure in elderly subjects with normal renal function were similar to those observed in a non-elderly healthy population [11], and that no dose adjustment is necessary on the grounds of age alone. It should also be noted that sex is among the components of the Cockroft–Gault formula, and a correction resulting in a 15% reduction in estimated glomerular filtration rate is applied to account for different relative amounts of fat and muscle in women; hence, female subjects have a slightly lower CL/F than men, due to differences in glomerular filtration rate [24, 30]. Finally, race had a minimal effect on varenicline apparent clearance (Figure 6); mean increases of 16% and 11% were detected in Blacks and other races, respectively. The effect size was deemed clinically irrelevant based on an upper bound of the 95% CIs predicting a maximum mean change on CL/F of about 20%.

Similarly, body weight appeared to be the most important predictor of variability in the central volume of distribution (Figure 6). V2/F was estimated to decrease, on average, approximately 35% between a typical individual weighing 70 kg (V2/F = 337 l) and one weighing 40 kg (V2/F = 219 l). Based upon individual varenicline plasma concentrations predicted at 2, 3 and 4 h post dose using the final pharmacokinetic parameters (including the interindividual and residual random effects), this would predict an average increase of about 30% in the steady-state peak plasma concentrations in individuals with small body size. Other surrogates for body size, including BMI and IBW, did not improve goodness of fit, relative to the weight-based model. The estimated effect of age on V2/F was poorly defined and no conclusive trends were identified (95% CI included 0). Given the age range in the PK database (18–76 years), this effect should be viewed within the context of the data and model, and any extrapolation should be done with extreme caution. Some race-related changes in volume of distribution were also detected (Figure 6). Blacks were estimated to have an 8% lower V2/F than Whites with an upper bound of the 95% CI including the no effect value of 1. The effect of other (Asian, Hispanic, Other) races was somewhat larger, but the interpretation of this remains unclear. In all, the clinical relevance of these covariate-related changes in the volume of distribution of varenicline is probably minimal, since systemic clearance remains the determinant for drug exposure and, hence, the basis for dose adjustment.

Distributions of individual steady-state AUC0–24ss derived from empirical Bayes estimates of CL/F are illustrated in Figure 7. Varenicline pharmacokinetics is linear across the dose range of 0.5–3 mg day−1. At 1 mg b.i.d., the overall mean AUC0–24ss and average Cav,ss were estimated to be 186 ng*h ml−1 and 7.73 ng ml−1, respectively; means and ranges were consistent across Studies II and III. These daily exposures were comparable to the results of noncompartmental analyses [6], which reported a mean of AUC0–24ss of 194 ng*h ml−1 (range 101–312) in adult smokers dosed at 1 mg b.i.d.

image

Figure 7. Final model distribution of individual predicted steady-state 24-h varenicline plasma AUC by daily dose. The box represents the interquartile distance with the median indicated by a solid line in the centre of the box; whiskers represent data ≤1.5 times the interquartile range and outliers are represented by single solid lines outside of the whiskers

Download figure to PowerPoint

From the population PK modelling, varenicline CL/F is predicted to decrease from 10.4 l h−1 for a typical subject with normal renal function (CLcr = 100 ml min−1) to 8.3 l h−1 (CLcr = 65 ml min−1), 6.4 l h−1 (CLcr = 40 ml min−1) and 4.4 l h−1 (CLcr = 20 ml min−1) for a typical subject with mild, moderate and severe renal impairment, respectively. The cut-off values used for these predictions reflect the degree of renal impairment observed in each subgroup of renally impaired subjects. Based on this, a dose reduction to 1 mg day−1, which is half the recommended dose, is indicated for subjects with severe renal insufficiency. Using the final PK model, simulations scenarios were performed with nonmem, using 500 replicates of six subjects with severe renal impairment and six subjects with normal renal function randomly sampled from the original dataset. The concentrations that encompassed the 2.5th and 97.5th percentiles at each time point were retrieved to construct the 95% prediction interval for the population distribution. As depicted in Figure 8, the 0.5 mg b.i.d. regimen provides predicted steady-state exposures in severely renally impaired subjects that fall within the 95% population prediction interval of exposures in subjects with normal renal function administered varenicline 1 mg b.i.d. Additionally, the underlying renal pathophysiology is expected to affect CL/F while resulting in minimal effects on volume of distribution, thus predicting a prolongation in half-life for individuals with severe renal impairment. Consistent with this expectation, varenicline elimination half-life was significantly prolonged in these patients, and 1 mg administered once daily also provides comparable steady-state exposure levels of varenicline when compared with those of healthy subjects (Figure 8).

image

Figure 8. Predicted steady-state pharmacokinetic profiles of varenicline in subjects with severe renal impairment. Bold lines (black and grey) are the median steady-state pharmacokinetic profiles of 1 mg q.d. and 0.5 mg b.i.d., respectively; the shaded areas are the corresponding 95% prediction intervals for the population distribution. The upper and lower dotted lines are the 95% population prediction interval from Cmax to Cmin in the population with normal renal function at the 1 mg b.i.d. dose regimen. 0.5BID Severe (inline image); 1 QD Severe (—)

Download figure to PowerPoint

In summary, this work provides an integrated model-based analysis of varenicline pharmacokinetics across multiple studies in the target smoking population. Renal function is the clinically important factor leading to interindividual variability in systemic exposure of varenicline. This robust and predictive model also provides a means to relate individual-specific drug exposures to clinical responses in subsequent PK–PD analyses.

Competing interests

  1. Top of page
  2. Abstract
  3. Introduction
  4. Methods
  5. Results
  6. Discussion
  7. Competing interests
  8. REFERENCES

The data reported in this study were obtained from clinical studies funded by Pfizer, Inc. H.M.F., P.R. and T.G.T. are full-time employees of Pfizer, Inc., and M.R.G. is a paid consultant for Pfizer, Inc.

The authors would like to thank Kevin Rohrbacher for coordination of varenicline sample analysis, as well as Pamela Melch, Regina Coleman and Stephen Faulkner for data management support.

REFERENCES

  1. Top of page
  2. Abstract
  3. Introduction
  4. Methods
  5. Results
  6. Discussion
  7. Competing interests
  8. REFERENCES
  • 1
    Rollema H, Chambers LK, Coe JW, Glowa J, Hurst RS, Lebel LA, Lu Y, Mansbach RS, Mather RJ, Rovetti CC, Sands SB, Schaeffer E, Shulz DW, Tingley FD, Williams KE. Pharmacological profile of the α4β2 nicotinic acetylcholine receptor partial agonist varenicline, an effective smoking cessation aid. Neuropharmacology 2007; 52: 98594.
  • 2
    Coe JW, Brooks PR, Vetelino MG, Wirtz MC, Arnold EP, Huang J, Sands SB, Davis TI, Lebel LA, Fox CB, Shrikhande A, Heym JH, Schaeffer E, Rollema H, Lu Y, Mansbach RS, Chambers LK, Rovetti CC, Schulz DW, Tingley FD 3rd, O'Neill BT. Varenicline: a novel, potent, and α4β2 selective nicotinic receptor partial agonist as an aid for smoking cessation. J Med Chem 2005; 48: 34747.
  • 3
    Jorenby DE, Hays JT, Rigotti NA, Azoulay S, Watsky EJ, Williams KE, Billing CB, Gong J, Reeves KR. Efficacy of varenicline, an α4β2 nicotinic acetylcholine partial receptor agonist, vs. placebo or sustained-release bupropion for smoking cessation. A randomized controlled trial. JAMA 2006; 296: 5663.
  • 4
    Gonzales D, Rennard SI, Nides M, Oncken C, Azoulay S, Billing CB, Watsky EJ, Gong J, Williams KE, Reeves KR. Varenicline, an α4β2 nicotinic acetylcholine partial receptor agonist, vs. sustained-release bupropion and placebo for smoking cessation. A randomized controlled trial. JAMA 2006; 296: 4755.
  • 5
    Williams KE, Reeves KR, Billing CB Jr, Pennington AM, Gong J. A double-blind study evaluating the long-term safety of varenicline for smoking. Curr Med Res Opin 2007; 23: 793801.
  • 6
    Faessel HM, Gibbs MA, Clark DJ, Rohrbacher K, Stolar M, Burstein AH. Multiple-dose pharmacokinetics of the selective nicotinic receptor partial agonist, varenicline, in healthy smokers. J Clin Pharmacol 2006; 46: 143948.
  • 7
    Faessel HM, Smith BJ, Gibbs MA, Gobey JS, Clark DJ, Burstein AH. Single-dose pharmacokinetics of varenicline, a selective nicotinic receptor partial agonist, in healthy smokers and nonsmokers. J Clin Pharmacol 2006; 46: 9918.
  • 8
    Feng B, Obach RS, Burstein AH, Clark DJ, Morais SMD, Faessel HM. Effect of human renal cationic transporter inhibition on the pharmacokinetics of varenicline, a new therapy for smoking cessation: an in vitroin vivo study. Clin Pharmacol Ther 2008; 83: 56776.
  • 9
    Obach RS, Reed-Hagen AE, Krueger SS, Obach BJ, O'Connell TN, Zandi KS, Miller S, Coe JW. Metabolism and disposition of varenicline, a selective alpha4beta2 acetylcholine receptor partial agonist, in vivo and in vitro. Drug Metab Dispos 2005; 4: 12130.
  • 10
    Chantix™[Package Insert]. New York: Pfizer Inc, 2009.
  • 11
    Burstein AH, Fullerton T, Clark DJ, Faessel HM. Pharmacokinetics, safety and tolerability following single and multiple oral doses of varenicline in elderly smokers. J Clin Pharmacol 2006; 46: 123440.
  • 12
    Beal SL. Ways to fit a pharmacokinetic model with some data below the quantification limit. J Pharmacokinet Pharmacodyn 2001; 28: 481504.
  • 13
    Beal SL, Sheiner LB, Boeckmann AJ. NONMEM User's Guide – Part I–VII. Ellicott City, MD: Icon Development Solutions; 19892006.
  • 14
    Ette EI. Stability and performance of a population pharmacokinetic model. J Clin Pharmacol 1997; 37: 48695.
  • 15
    Ette EI, Ludden TM. Population pharmacokinetic modeling: the importance of informative graphics. Pharmaceut Res 1995; 12: 184555.
  • 16
    Hooker AC, Staatz CE, Karlsson MO. Conditional Weighted Residuals (CWRES): a model diagnostic for the FOCE method. Pharmaceutical Res 2007; 24: 218797.
  • 17
    Parke J, Holford NH, Charles BG. A procedure for generating bootstrap samples for the validation of nonlinear mixed-effects population models. Comput Methods Programs Biomed 1999; 59: 1929.
  • 18
    Yafune A, Ishiguro M. Bootstrap approach for constructing confidence intervals for population pharmacokinetic parameters. I: a use of bootstrap standard error. Stat Med 1999; 18: 58199.
  • 19
    Burnham KP, Anderson DR. Model Selection and Multimodel Inference: A Practical Information-Theoretic Approach. New York: Springer-Verlag, 2002.
  • 20
    Harrell FE. Regression Modeling Strategies: With Applications to Linear Models, Logistic Regression, and Survival Analysis. New York: Springer-Verlag, 2001.
  • 21
    Yano I, Beal SL, Sheiner LB. Evaluating pharmacokinetic/pharmacodynamic models using the posterior predictive check. J Pharmacokinet Biopharm 2001; 28: 17192.
  • 22
    Gelman A, Carlin JB, Stern HS, Rubin DB. Bayesian Data Analysis, 2nd edn. Boca Raton, FL: CRC Press LLC, 2004.
  • 23
    Post TM, Freijer JI, Ploeger BA, Danhof M. Extensions to the Visual Predictive Check to facilitate model performance evaluation. J Pharmacokinet Pharmacodyn 2008; 35: 185202.
  • 24
    Cockcroft DW, Gault MH. Prediction of creatinine clearance from serum creatinine. Nephron 1976; 16: 3141.
  • 25
    Holford NH. A size standard for pharmacokinetics. Clin Pharmacokinet 1996; 30: 32932.
  • 26
    Karlsson MO, Savic RM. Diagnosing model diagnostics. Clin Pharmacol Ther 2007; 82: 1720.
  • 27
    Darmady EM, Offer J, Woodhouse AM. The parameters of the ageing kidney. J Pathol 1972; 109: 195207.
  • 28
    Hollenberg NK, Adams DF, Solomon HS, Rashid A, Abrams HL, Merrill JP. Senescence and the renal vasculature in normal man. Circ Res 1974; 34: 30916.
  • 29
    Davies DF, Shock NW. Age changes in glomerular filtration rate, effective renal plasma flow, and tubular excretory capacity in adult males. J Clin Invest 1950; 29: 496507.
  • 30
    Matzke GR, Millikin SP. Influence of renal function and dialysis on drug disposition. In: Applied Pharmacokinetics. Principles of Therapeutic Drug Monitoring, 3rd edn, eds EvansWE, SchentagJJ, JuskoWJ. Vancouver, WA: Applied Therapeutics, Inc., 1992; 149.