PHARMACOKINETICS AND DISPOSITION

Population pharmacokinetics analysis of olanzapine for Chinesepsychotic patients based on clinical therapeutic drug monitoringdata with assistance of meta-analysis

Anyue Yin1,2,3& Dewei Shang4 & Yuguan Wen4

& Liang Li1,2,3 & Tianyan Zhou1,2,3&

Wei Lu1,2,3,5

Received: 24 December 2015 /Accepted: 2 March 2016 /Published online: 27 April 2016# Springer-Verlag Berlin Heidelberg 2016

AbstractPurpose The aim of this study was to build an eligible popu-lation pharmacokinetic (PK) model for olanzapine in Chinesepsychotic patients based on therapeutic drug monitoring(TDM) data, with assistance of meta-analysis, to facilitate in-dividualized therapy.Methods Population PK analysis for olanzapine was per-formed using NONMEM software (version 7.3.0). TDM datawere collected from Guangzhou Brain Hospital (China).Because of the limitations of TDM data, model-based meta-analysis was performed to construct a structural model to as-sist the modeling of TDM data as prior estimates. After ana-lyzing related covariates, a simulation was performed to pre-dict concentrations for different types of patients under com-mon dose regimens.

Results A two-compartment model with first-order absorptionand elimination was developed for olanzapine oral tablets,based on 23 articles with 390 data points. The model was thenapplied to the TDM data. Gender and smoking habits werefound to be significant covariates that influence the clearanceof olanzapine. To achieve a blood concentration of 20 ng/mL(the lower boundary of the recommended therapeutic range),simulation results indicated that the dose regimen ofolanzapine should be 5 mg BID (twice a day), ≥ 5 mg QD(every day) plus 10 mg QN (every night), or >10 mg BID forfemale nonsmokers, male nonsmokers and male smokers,respectively.Conclusion The population PK model, built using meta-anal-ysis, could facilitate the modeling of TDM data collected fromChinese psychotic patients. The factors that significantly in-fluence olanzapine disposition were determined and the finalmodel could be used for individualized treatment.

Keywords Olanzapine . Psychotic . Populationpharmacokinetic model . Model-basedmeta-analysis .

Therapeutic drugmonitoring

Introduction

Olanzapine, an atypical antipsychotic that is effective in dif-ferent types of psychotic disorders, has been widely used inclinical treatment since it was first launched. Olanzapine is aserotonin-dopamine-receptor antagonist that has affinity notonly for dopamine receptors (D1, D2, and D4), but also forserotonin receptors (5-HT2c and 5-HT2a), histamine receptors(H1), α- adrenergic receptors (α1 and α2), and muscarinicreceptors (M1). This breadth of activity delivers efficacyagainst both the negative and positive symptoms of schizo-phrenia and leads to relatively infrequent side effects [1].

Anyue Yin and Dewei Shang contributed equally to this work.

Electronic supplementary material The online version of this article(doi:10.1007/s00228-016-2040-2) contains supplementary material,which is available to authorized users.

* Wei Luluwei_pk@bjmu.edu.cn

1 School of Pharmaceutical Sciences, Peking University,Beijing 100191, China

2 Peking University/Pfizer Pharmacometrics Education Center, PekingUniversity Health Science Center, Beijing 100191, China

3 State Key Laboratory of Natural and Biomimetic Drugs, PekingUniversity, Beijing 100191, China

4 Guangzhou Brain Hospital (Guangzhou Huiai Hospital, theAffiliated Brain Hospital of Guangzhou Medical University),Guangzhou 510370, China

5 Beijing Institute for Brain Disorders, Beijing 100088, China

Eur J Clin Pharmacol (2016) 72:933–944DOI 10.1007/s00228-016-2040-2

A population pharmacokinetic (PK) modeling approachcan identify sources and correlates of pharmacokinetic vari-ability in a target patient population [2] and provide a reliablemodel to optimize drug therapy for individual patients [2].Among plenty of olanzapine PK studies, only a small numberof them have built population PK models [3–7], but thesemodels have not been evaluated as to whether they are suitablefor Chinese patients.

In order to achieve target response and minimize side ef-fects, therapeutic drug monitoring (TDM) is usually per-formed to ensure that olanzapine concentrations in psychoticpatients are within the recommended therapeutic range [8].Unlike dense PK sampling, TDM samples are only collecteda few hours after the last drug administration or before the nextdose, but these data are readily available and are representativeof the target patient population. Population PK analysis andmodeling is, therefore, an ideal way to utilize these data toestablish a quantitative model that can be used to guide indi-vidualized treatment [2].

Since such sparse data are incapable of capturing the com-plete PK profile, a structural model with reasonable PK pa-rameters is needed before investigating the TDM data. Model-based meta-analysis can be used to make up the deficiency.This approach facilitates integration and utilization ofsummary-level data, such as dense concentration data fromvarious independent studies, and provides a quantitativeframework to incorporate external data into the decision-making process [9]. Constructing a PK model using this ap-proach is a reliable way to obtain the Bayesian priors formodeling local TDM data, and then establish a complete PKmodel for the given population.

In the present study, a nonlinear mixed-effects model wasbuilt based on mean olanzapine concentrations collected fromthe literature using a model-based meta-analysis approach.Structural model and parameter estimates were then used aspriors to assist population PK analysis of routine TDM datacollected from Chinese psychotic patients. After evaluatingthe model and analyzing the related covariates, a simulationwas performed. The ultimate aim of this study was tomake thebest use of retrospective TDM data to build an eligible popu-lation PK model for olanzapine in Chinese psychotic patientsthat would facilitate individualized therapy. This type of studyhas not previously been described.

Materials and methods

Model-based meta-analysis

Literature resources

The key words ‘olanzapine AND pharmacoki* NOT longacting’, occurring in the title and/or abstract, were used to

retrieve records from the PubMed database, and the key words‘olanzapine’ and ‘pharmacokinetics’ were used to retrieve re-cords from the SciFinder database. Literature reports pub-lished from 1996 toMarch 2015 that met the following criteriawere included in the analysis: a) any bioequivalence study orclinical trial with full olanzapine monotherapy concentrationprofiles; b) studies that reported olanzapine concentrations aspopulation mean values; c) studies in which olanzapine wasadministered as an oral tablet.

Exclusion criteria were: a) methodology studies in whichonly individual concentrations were reported; b) animal stud-ies; c) studies in which olanzapine was not the target drug; d)studies of extended (long-term) drug delivery systems that didnot include a concentration profile for normal tablets; e) com-bination therapy studies without an olanzapine single arm; f)studies that used an excessive or unclear dose; g) studies inwhich the drug formulation was a disintegrating tablet; h)studies with a special target population, e.g., children; i) stud-ies for which the full text was not available.

Records from the SciFinder database that had already beenretrieved from the PubMed database were excluded from theanalysis in the first place, to avoid duplication.

Data extraction

Basic information, including year of publication, author,study design, dose regimen, dose formulation, adminis-tration route, number of participants (sample size), typeof subjects (healthy volunteers or schizophrenia pa-tients), subject ethnicity, drug manufacturer, and linearrange of the measurements, were extracted from theeligible studies. PK profiles and detailed informationabout each study arm, including olanzapine concentra-tions, sampling times, percentage of female subjects,percentage of smoking subjects, mean age, mean bodyweight, and mean height, were also obtained fromFigures and Tables.

Olanzapine concentration data were extracted from theFigures in each literature using WebPlotDigitizer (version3.8, http://www.arohatgi.info/WebPlotDigitizer/app/). Allreported olanzapine concentrations were converted to ng/mL.

The extracted concentrations were evaluated using a dia-gram of normalized blood concentrations versus time. Studywhose results were deviated from others while the measure-ment method was uncommon compared with other studieswould be excluded.

Model development

A nonlinear mixed-effects model, based on the meta-analysisdata set, was developed using the first order conditional esti-mation method with interaction (FOCEI), implemented inNONMEM software (version VII, level 3.0; ICON

934 Eur J Clin Pharmacol (2016) 72:933–944

Development Solutions, Ellicott City, Maryland, USA). One-,two- and three-compartment models, with first-order absorp-tion (with or without a time lag) and first-order elimination,were evaluated as the structural model. Inter-arm variability(IAV) and residual errors were included in the models.

IAV was assumed to be log-normally distributed and wasdescribed by the following equation (Eq. 1).

Pi ¼ Ptv⋅eηi ð1Þ

where Pi is the parameter for arm i; Ptv is the typical value ofthe parameter, and η represents the random variability of theparameter, which was assumed to be normally distributed witha mean of zero and a variance of ω2.

The residual error was weighted by the sample size and acombined proportional and additive model was selected(Eq. 2):

Yobs ¼ Ypred⋅ 1þ ε1ffiffiffiffiN

p� �

þ ε2ffiffiffiffiN

p ð2Þ

where Yobs represents the observation, Ypred is the correspond-ing model-predicted value, N is the sample size of each study,and ε1 and ε2 represent the proportional and additive residualerrors, respectively, which were assumed to follow a normaldistribution with a mean of zero and variances of σ1

2 and σ22,

respectively.During the modeling process, potential parameter covari-

ates, includingmean age (Eq. 3), proportion of female subjectsin each arm (Eq. 4), proportion of smoking subjects in eacharm (Eq. 4), and type of subjects (Eq. 4, where COV i =1 if thesubjects were healthy and 0 if subjects were patients), wereinvestigated using the following equations with a forward in-clusion and a backward elimination process in a stepwisemanner [10]:

Pi ¼ Ptv⋅ 1� θCOV ⋅ COVi−COVmð Þð Þ⋅eηi ð3Þ

Pi ¼ Ptv⋅ θCOV ⋅COV i þ 1−COV ið Þð Þ⋅eηi ð4Þwhere θCOV is the estimate of covariate effect, COVi is thecovariate value in the i-th arm, and COVm is the mean ormedian value of the covariate. In the forward inclusion pro-cess, each covariate was incorporated stepwise into the basicmodel. A covariate was considered statistically significantwhen its addition led to a decrease in the objective functionvalue (OFV) > 3.84 (p<0.05, degree of freedom=1). The fullmodel was established when no more covariates could beincluded. Then the covariates were deleted from the modelstepwise. A covariate was retained in the model when itselimination led to an increase in the OFV > 10.83 (p<0.001,degree of freedom=1). The final model was established whenno more covariates could be excluded [10].

Model evaluation

Predictability and stability of the meta-analysis model wereevaluated using goodness-of-fit plots, the visual predictivecheck (VPC) method, and the nonparametric bootstrap meth-od. VPC was performed by 1000 times of simulation and thebootstrap was conducted based on 1000 times of resampling.Diagrams were created by R software (Version 3.2.0, The RFoundation for Statistical Computing).

TDM data analysis

Data resources

The TDM data were collected from psychotic patients whowere hospitalized in the Guangzhou Brain Hospital (theAffiliated Brain Hospital of Guangzhou Medical University,Guangzhou, China) during January 2014 to January 2015.Olanzapine tablets (Zyprexa, Eli Lilly or Ou Lanning,Stockhausen, Jiangsu) were administrated with complex regi-mens (dose ranged from 2.5 to 30 mg) that were constantlyadjusted throughout the treatment. Samples were collected be-fore the next or 10–10.5 h after the last dose, once or twice amonth after the start of treatment. Demographic informationand physiological index for each individual, including age,gender, smoking habit, red blood cell (RBC), hemoglobin(HGB), hematocrit (HCT), platelet (PLT), aspartate amino-transferase (AST), alanine aminotransferase (ALT), total pro-tein (TP), albumin (ALB), blood glucose (GLU), blood ureanitrogen (BUN), serum creatinine (SCr), uric acid (UA), andsuperoxide dismutase (SOD), were also collected. For this typeof retrospective data analysis, formal consent is not required.

Model development

The TDM data were analyzed using NONMEM software withFOCEI. The model structure was set as the one selected inmodel-based meta-analysis, and typical values of most PK pa-rameters were fixed to the corresponding obtained estimates,since those values could not be estimated based on our TDMdata, except for apparent systematic clearance (CL/F). Inter-individual variabilities (IIV) were set to be the correspondingIAVobtained in the meta-analysis, but multiplied by a factor (θf)(Eq. 5). The IIVof the parameters that related to drug absorptionprocess were fixed to zero. A combined proportional and addi-tive model, without weight, was selected to describe the residualerror of this model (Eq. 6).

P j ¼ Ptv⋅eη j η j ¼ ηi⋅θ f ð5ÞYobs ¼ Ypred⋅ 1þ ε1ð Þ þ ε2 ð6Þ

The demographic information and physiological indexwere both considered as potential covariates of CL/F and were

Eur J Clin Pharmacol (2016) 72:933–944 935

investigated during the modeling process as described abovewhile other parameters were fixed. The effects of continuouscovariates (Eqs. 7 and 8) and categorical covariates (Eq. 9,θCOV was 1 for the basic group and estimated for othergroups) were tested using different equations:

Pj ¼ Ptv⋅ 1� θCOV ⋅ COV j−COVm

� �� �⋅eη j ð7Þ

Pj ¼ Ptv⋅ COV j=COVm

� �θCOV ⋅eη j ð8ÞPj ¼ Ptv⋅θCOV ⋅eη j ð9Þ

where Pj is the parameter for individual j and COVj is thecovariate value for individual j. In the process of stepwiseforward inclusion, each continuous covariate was first triedin one of the two relations to CL/F [e.g. linear relation(Eq. 7)] according to the regression pattern identified fromthe plot where individual CL/Fs obtained from the basic mod-el were plotted versus the covariate. Once a covariate has beenincluded in the model, the other relation [e.g. nonlinear rela-tion (Eq. 8)] was also tried in the next step. The covariaterelationship that gave the largest decrease to OFV wasretained in the model. As for categorical covariates (i.e. genderand smoking habit), if both of them were included, consider-ing the characteristic of our TDM data, the covariate effectswere estimated separately for different categories of patients.

Model evaluation

The population PK model for TDM concentrations was eval-uated by goodness-of-fit plots, nonparametric bootstrap meth-od, and the normalized prediction distribution error (NPDE)method, which is suitable for evaluating models developedfrom disordered data [11]. The NPDE for each observationwas calculated using R software with an attached package,NPDE (version 2.0), based on 1000 simulations. R softwarewas also used to plot the diagrams.

Simulation

According to the finalized model of the TDM data set, thetime course of olanzapine serum concentrations at steady statewas simulated using typical parameter estimates under regularoral administration of commonly used dose regimens [20 mgevery day (QD), 15 mg QD, 10 mg QD, 10 mg twice a day(BID), 5 mgBID, and 5mgQDplus 10mg every night (QN)].The simulated concentrations, represented by population pre-dicted values, were compared with the minimum target clini-cal treatment concentration (20 ng/mL) [8] to serve as a toolfor treatment regimen selection for different clinicalpopulations.

Results

Model-based meta-analysis

Data set characteristics

After screening and evaluating the retrieved studies, 390olanzapine serum concentrations acquired from 23 differentstudies, including 27 arms which were later treated as virtualindividuals, constituted the data set for this meta-analysis. Theprocess of screening literature is shown in Fig. 1. Theolanzapine used in most studies was manufactured byEli Lilly and Company. One study used two differentbrands of olanzapine, but details of the manufacturerswere not provided [31].

Among these studies, seven of them involved female sub-jects and four studies included smokers among the subjects.And only three studies were carried out in schizophrenia pa-tients. Two studies separated the volunteers into groups, basedon different smoking habits and metabolic types, respectively.Two studies collected samples after multiple dosing whileothers were single-dose studies. The mean age and bodyweight of the 27 virtual individuals were 30.2 (range 20.0–44.4) years and 67.9 (range 59.0–85.0) kg, respectively. Forthose studies with missing records, the records were set as thecorresponding median values of all virtual individuals.Detailed information is summarized in Table 1.

Model development

The data set for olanzapine in this meta-analysis was ade-quately described by a two-compartment model with first-order elimination (Online Resource 1). The absorption ofolanzapine tablets was well described by a first-order processincorporating a time lag. Estimates of population PK parame-ters for the meta-analysis are summarized in Table 2. Therelative standard errors (RSE) for all parameter estimates wererelatively small and the parameter estimates were in goodagreement with the bootstrap results (Table 2). The IAV ofthe apparent distribution clearance (Q/F) and the apparent vol-ume of distribution in the peripheral compartment (V3/F)were found to be too small during the modeling process, andthese values were eventually set at zero. CL/F and apparentvolume of distribution for the central compartment (V2/F)were found be correlated and the correlation was as high as98.2 %. As for the covariates analysis, it showed that none ofthe considered covariates had a significant influence on thePK parameters in this model.

Model evaluation

The goodness-of-fit plots are shown in Fig. 2. Both populationpredictions and individual predictions were in accordance

936 Eur J Clin Pharmacol (2016) 72:933–944

with observations and the conditional weighted residual errors(CWRES) were randomly distributed around zero. The VPCplot for the final model is shown in Online Resource 1. Thearea between the 5th and 95th percentiles of the predictionsperfectly covers most of the observed concentrations. The 5th,50th, and 95th percentiles of the observations and predictionsare consistent with each other, and the 95 % confidence inter-vals (95 % CI) around percentiles of predictions could coverthe corresponding observed percentiles, suggesting that thepredictability of the model is acceptable. The 95 % CI aroundpercentiles of predictions for the data collected from multiple-dose studies were not plotted because they were considered tobe meaningless, since only two multiple-dosing studies wereincluded.

TDM data analysis

TDM data set characteristics

The TDM data set comprised 543 olanzapine TDM se-rum concentrations, collected from 128 psychotic pa-tients (68 males and 60 females) with a median age of43.5 (range 18–85) years. The sample number of eachpatient was around 1 to 24, with a median of 3. Thedata set included 187 trough concentrations and 356concentrations measured 10–10.5 h after olanzapine

administration. Among the participants, there were 23male smokers (Ms), 33 male non-smokers (Mns), onefemale smoker (Fs), 51 female non-smokers (Fns), and20 patients (12 males and eight females) whosesmoking habit was unknown (Munk and Funk). Detailedinformation that could be used in covariate analysis aresummarized in Online Resource 2.

Model development

The two-compartment model developed during the meta-analysis was applied to the olanzapine TDM data set. Theparameter estimates of the final population PK model of theTDM data set are listed in Table 3, and are in good agreementwith the bootstrap results (Table 3). The estimated standarddeviation (SD) of additive residual error was 1.03 ng/mL,lower than the lower limit of quantification (2 ng/mL).

The covariate analysis showed that both gender andsmoking habit had a significant influence on CL/F ofolanzapine (p < 0.001). Because there was only one Fs, shecould not represent the group. Study subjects were thus divid-ed into five categories: Ms, Mns, Fns, Munk, and Funk. After theestimation, the coefficient multiplied on CL/F for Funk andMunk was close to that for Fns and Mns (0.734 versus 1.00and 1.20 versus 1.48) when the coefficient for Ms was higherthan that for Mns (2.13 versus 1.48). Therefore, the smoking

Fig. 1 Flow diagram for theinclusion and exclusion ofliterature selected for meta-analysis

Eur J Clin Pharmacol (2016) 72:933–944 937

Tab

le1

Summarized

inform

ationof

theselected

studies

Publishyear

Author

Volunteertype

Dose

Studydesign

Samplesize

Age

Weight

Female

(n/N)

Smoking

(n/N)

2013

P.Zakeri-Milani

[12]

Health

y10

mg

Singledose,bioequivalencestudy

2424.6

71.7

0/24

–2014

BirgitS

.Jacobsa

[13]

Health

y10

mg

Open-label,random

ized,two-period,

cross-over,single-center

study

2042

–13/24

4/20

2004

Chuan-Yue

Wang[14]

Health

y10

mg

Randomized,open-labelstudy

1226.7

63.1

0/12

0/12

2008

Tzu-H

uaWu[15]

Schizophrenic

10mg

Singleoraldose

forthreekindsof

patientsstudy

9(non-smokers)

34.9

590/9

0/9

9(light

smokers)

40.9

63.2

0/9

9/9

9(heavy

smokers)

44.4

65.2

0/9

9/9

2006

Jagdev

Sidhu

[16]

Health

y15

mg

Randomized,partially

blind,placebo-

controlled,parallel-groupstudy

16–

–0/16

0/16

2004

Chih-ChiangChiu[17]

Schizophrenic

10mg

–10

42.2

59.8

0/10

10/10

2002

ScottR

.Penzak[18]

Health

y10

mg

Open-labeld

ruginteractionstudy

1425

–1/14

0/14

2006

John

S.Markowitz

[19]

Health

y5mg

3-way,open-label,random

ized

crossoverstudy

1031.4

850/10

0/10

2012

Q.C

hen[20]

Health

y10

mg

Single-dose,open-label,crossover,

bioequivalence

study

2424.3

59.9

0/24

0/24

2009

Ahm

edH.E

ishafeey

[21]

Health

y10

mg

Single-dose,randomized,two-way

crossover,bioequivalence

study

2424.7

73.4

0/24

0/24

2002

DenisGossen[22]

Health

y5mg

Open-label,single-sequencecrossover

study

1532

714/15

0/15

2002

John

S.Markowitz

[23]

Health

y5mg

Single-dose,random

ized,4-period,

double-blin

d,crossoverstudy

1228

–2/12

0/15

2001

S.H

agg[24]

Health

y7.5mg

Singledose

study

5(CYP2D

6poor

metabolisers)

27.4

810/5

0/5

12(CYP2

D6extensive

metabolisers)

24.7

70.6

0/12

0/12

1998

William

L.M

acias[25]

Health

y10

mg

Two-way,randomized,crossover

study

1235

79.2

0/12

11/12

1997

John

T.Callaghan

[26]

Health

y5mg

Open-label,three-way

random

ized

crossoverstudy

940.9

68.8

0/9

0/9

2015

CavalcantiB

edor

NC[27]

Health

y10

mg

Openlabel,random

ized,two-period,

two-treatm

ent,two-sequence,2×2

crossovertrial

28–

–0/28

–

1998

R.A

.Lucas

[28]

Health

y10

mg

Singlecenter,openstudywith

two

periods.

1230.4

–0/12

–

2012

YULing-Yan

[29]

Health

y5mg

Randomized

cross-over

study

2024.7

–0/2 0

0/20

2011

Somruedee

Chatsiricharoenkul[30]

Health

y10

mg

Singledose,randomized,2-period,

2-sequence,crossover

study

2425.8

61.6

12/24

0/24

2011

MercedesCanovas

[31]

Health

y10

mg

Open-label,random

ized,two-period,

two-sequence,crossover

study

2026.9

6410/20

0/20

2005

Wenbiao

Li[32]

Health

y10

mg

Randomized

andcrossoverclinicaltrial

2224.91

65.32

0/22

0/22

2002

Huifang

Ji[33]

Health

y10

mg

–20

20.03

65.05

0/20

0/20

2005

AdrienneC.L

ahti[34]

Schizophrenic

15mg

Singledose

638

–2/6

–

938 Eur J Clin Pharmacol (2016) 72:933–944

status of Funk and Munk was eventually treated as non-smoking. The typical values of CL/F for Ms and Mns were,respectively, 2.25-fold and 1.47-fold than that for the basicgroup Fns (16.6 L/h), eventually (Table 3). Other covariatesdid not significantly affect the PK parameters. The multipliedfactor θf is 1.10.

Model evaluation

The goodness-of-fit plots for the final population PK modelfor the TDM data set are shown in Fig. 3 and the results ofNPDE are shown in Fig. 4. The NPDE was distributed arounda mean of 0.086 with a variance of 1.078 and showed no

Table 2 Population PKparameters estimates andbootstrap results of the finalmodel for meta-analysis

Final model Bootstrap

Parameter Estimate RSE (%) IAV (CV%) Median 95 % CI RSE (%)

Ka (/h) 0.868 14.4 51.2 0.863 0.602–1.15 16.0

CL/F (L/h) 18.9 7.4 36.7 19.0 16.4–22.5 8.07

V2/F (L) 599 9.6 42.7 (98.2 %correlated)

599 484–744 10.9

Q/F (L/h) 21.3 20.3 0 fixed 21.4 12.7–34.9 24.9

V3/F (L) 174 12.5 0 fixed 173 127–234 15.5

Tlag (h) 0.706 6.2 17.9 0.706 0.565–0.787 8.07

Sample size weightedresidual error

PRO (CV%) 26.1 22 – – – –

ADD (SD, ng/ml) 1.47 42 – – – –

Ka first-order absorption rate constant, CL/F apparent systematic clearance, V2/F apparent volume of distributionfor the central compartment, Q/F apparent distribution clearance, V3/F apparent volume of distribution for theperipheral compartment, Tlag time lag of absorption, IAV inter-arm variability, CV coefficient of variation, CIconfident interval, RSE relative standard errors of parameters, PRO proportional residual error, ADD additiveresidual error, SD standard deviation of the error

Fig. 2 Goodness-of-fit plots ofthe population PK model formeta-analysis

Eur J Clin Pharmacol (2016) 72:933–944 939

obvious trend versus time (Fig. 4c) or versus the predictions(Fig. 4d), suggesting that the population PKmodel establishedhere could properly characterize the TDM data.

Simulation

The typical time courses of steady state olanzapine concentra-tions in psychotic patients were simulated for Ms, Mns, and Fns

after using different dose regimens according to the finalmodel (Fig. 5). The population predicted trough concen-tration of the Fns group could achieve 20 ng/mL, the lowerboundary of the recommended therapeutic range forolanzapine [8], with an olanzapine dose of 5 mg BID,while it needed to be ≥ 5 mg QD plus 10 mg QN for Mns

group and > 10 mg BID for Ms group to achieve sametrough concentration.

Table 3 Population PK parameters estimates and bootstrap results of the final PK model for TDM data set

Parametera Final model Bootstrap

Estimate RSE (%) IIV (CV%) Median 95 % CI RSE (%)

CL/F female non-smokers (L/h) 16.6 6.4 36.7 × θf 16.7 14.6–18.7 6.13

θ male smokers 2.25 10 – 2.23 1.85–2.69 10.1

θ male non-smokers 1.47 8.8 – 1.46 1.25–1.74 8.73

V2/F (L) 599 fixed – 42.7 × θf (98.2 %correlation fixed)

599 fixed – –

θf 1.10 8.4 – 1.07 0.876–1.24 8.66

Residual error

PRO (CV%) 31.1 5.3 – – – –

ADD (SD, ng/ml) 1.03 120.8 – – – –

CL/F apparent systematic clearance, θ covariate effect of gender and smoking habit on CL/F, V2/F apparent distribution volume for the centralcompartment, IIV the inter-individual variability, θf correction factor on IIV, CV coefficient of variation, CI confident interval, RSE relative standarderrors of parameters, PRO proportional residual error, ADD additive residual error, SD standard deviation of the errora Other parameters were fixed to the corresponding values obtained in meta-analysis

Fig. 3 Goodness-of-fit plots ofthe final population PK model forTDM data set

940 Eur J Clin Pharmacol (2016) 72:933–944

Fig. 4 The NPDE plots of thepopulation PK model based onTDM data set, including thequantile–quantile plot (a), thedistribution histogram of NPDE(b), and the NPDE versus time (c)and predictions (d)

Fig. 5 The simulated time courses of olanzapine serum concentrations of patients at steady state for female non-smokers (grey lines), male non-smokers(black lines), and male smokers (black dash lines) after using different dose regimens

Eur J Clin Pharmacol (2016) 72:933–944 941

Discussion

Because of the limitations of TDMdata, it is not usually possibleto construct a complete populationPKmodel basedonlyon thesedata.Severalmethodshavebeenused toovercomethedifficultiesassociatedwith themodeling of TDMdata. These include fixingsomeof theparameters in linewithprevious reports [35], poolingTDM data together with dense individual concentrations [36],and model-based meta-analysis [37]. As in most cases, denseindividual concentrations were not available in this study.Model-based meta-analysis can integrate various studies, whereChinese subjectsmight be contained, rather than refer to onlyonestudy.We therefore carried out a model-basedmeta-analysis andused the resulting populationPKmodel to assist in the analysis ofTDMdata collected fromChinese psychotic patients.

Based on our inclusion and exclusion criteria, 24 articleswereinitially included. However, dose-normalized blood concentra-tions of one studyweremuch higher than those reported in otherstudies. The olanzapine concentrations in this study were mea-suredbyHLPC/UVwith a linear rangeof5–320ng/mL,which isdifferent fromwhat was used in the other studies. Models devel-oped using a data set with or without this study were evaluated.The plot where population predictions were drawn versus obser-vations of the former one showed that the predicted results devi-ated from observed results. Therefore, this study was finally ex-cluded (Fig. 1).

A two-compartment model with first-order elimination wasdeveloped for olanzapine using meta-analysis. The absorptionof olanzapine oral tablets was well described by a first-orderprocess (Ka=0.868/h) with a time lag (0.706 h). The estimat-ed elimination half-life and CL/F were 29.9 h and 18.9 L/h,respectively, which were in agreement with previously pub-lished results [1, 38]. In addition, a close correlation (up to98.2 %) between CL/F and V2/F was observed for olanzapineand was involved in the model, which optimized the popula-tion PK model of the target drug. All estimates of parameterswere relatively precise (RSE < 30 %).

Several covariates were tested through forward inclusionand backward elimination methods, in order to identify thosefactors that significantly affect drug disposition (p<0.001). Inour meta-analysis, age, gender, and smoking habit did notsignificantly affect drug disposition, although analyses of in-dividual data showed that these factors could have a significantimpact on olanzapine elimination [6, 39]. A possible explana-tion for this discrepancy is that the covariate values werederived from a group of people. Since only averages andpercentages were available, the impact of these factors on thePK parameters in meta-analysis is likely to be less conspicu-ous. Patient type was also investigated and no substantial effectwas observed. Parameters obtained based on meta-analysisdata set, wheremost individuals were healthy volunteers, couldthus be applied on our TDM data, which were collected onlyfrom patients. The potential impact of ethnicity was not

analyzed during the meta-analysis, since not all selected stud-ies recorded this information and studies where the informationwas recorded were carried out in ethnic groups where the PKcharacteristics of olanzapine were similar [6, 40]. The fit andprediction accuracy and stability of the meta-analysis modelwere shown to be acceptable by the goodness-of-fit plots(Fig. 2), VPC plot (Online Resource 1), and bootstrap results(Table 2).

When the model was applied to the TDM data, the IIVvalue was adjusted by multiplying with a coefficient θf onthe basis of IAV. The aim of this adjustment was to ensurethe degree of correlation between CL/F and V2/F in the finalmodel is similar to that obtained in themeta-analysis when IIVis not appropriate to be equal to IAV. The result showed thatIIV was slightly larger than IAV. During the covariate analy-sis, after treating the smoking status of Funk and Munk as non-smoking, although the OFVof the model increased by 6.963,the predictive effect was still acceptable and the results wereconsidered to be more suitable for clinical application. TheCL/F of olanzapine for Ms was found to be 1.53-fold higherthan that for Mns, suggesting that smoking accelerates theclearance of olanzapine. This agrees with the fact that cigarettesmoke contains a potent inducer of CYP1A2, an importantenzyme in the metabolism of olanzapine [38]. It was alsofound that Fns cleared olanzapine only 68.0 % as fast as Mns,which was in agreement with the results in previous studies [6,39]. This could be explained by physiological differences be-tween males and females and the fact that estrogen is an in-hibitor of CYP1A2 [6]. Additionally, age did not significantlyimpact the clearance of olanzapine when gender and smokinghabits were already taken into consideration in this study, andgenotype was not considered in this research because the ge-notype of CYP1A2 is not commonly measured. The finalmodel for olanzapine was confirmed to have good predictiveability and stability (Figs. 3 and 4, Table 3). The simulatedconcentrations for Fns were higher than Mns and those for Ms

were the lowest, which is consistent with the real-life situationin patients, confirming that the final model is a useful tool forpersonalized therapy of Chinese psychotic patients.

The present study has three limitations that should be im-proved by further work. Firstly, our TDM concentrations forolanzapine were retrospectively obtained from a limited num-ber of patients; thus, patient groups of some classificationscould not be fully analyzed. Secondly, the differences in thestudy design and subject selection among reports in the meta-analysis may lead to the bias of estimations. Thirdly, informa-tion about drug co-administration was not included because ofits complexity, and most parameters obtained from the meta-analysis were fixed when the model was applied to the TDMdata. These factors may also, to some extent, result in the biasof estimations.

Despite these limitations, this study has undeniable merits.Firstly, a two-compartment model, involving the correlation

942 Eur J Clin Pharmacol (2016) 72:933–944

between CL/F and V2/F, was constructed to describeolanzapine concentrations and validated to have good predict-ability on olanzapine concentrations. In addition, since themodel was built based on TDM data from Chinese psychoticpatients, the results obtained have greater referential value forthe individualized therapy for Chinese patients. Most impor-tantly, the model built from meta-analysis was used asBayesian priors to assist the modeling of the TDM data in thisstudy, overcoming the limitation of constructing a completepopulation PK model using only TMD data. This could pro-vide an example for modeling other TDM data.

In conclusion, the two-compartment model with first-orderabsorption (with a time lag) and first-order elimination builtbased on meta-analysis and our TDM data was validated tohave good predictability on olanzapine concentrations inChinese psychotic patients. The final model could be usedas a suitable tool for designing individualized therapy forChinese psychotic patients.

Acknowledgments This work was supported by Pfizer Scholarship forPharmacometrics, National Natural Science Foundation of China (GrantNo 81403016), and Natural Science Foundation of Guangdong Province(Grant No 2015A030313808).

Author contribution Wei Lu, Tianyan Zhou, Liang Li designed theresearch; Yuguan Wen, Dewei Shang provided raw TDM data; AnyueYin, Dewei Shang performed the research, and Anyue Yin wrote thepaper.

References

1. Bever KA, Perry PJ (1998) Olanzapine: a serotonin–dopamine-re-ceptor antagonist for antipsychotic therapy. Am J Health-SystPharm 55:1003–1016

2. Kiang TK, Sherwin CM, Spigarelli MG, Ensom MH (2012)Fundamentals of population pharmacokinetic modelling: model-ling and software. Clin Pharmacokinet 51(8):515–525

3. Lobo ED, Robertson-Plouch C, Quinlan T, Hong Q, Bergstrom RF(2010) Oral olanzapine disposition in adolescents with schizophre-nia or bipolar I disorder: a population pharmacokinetic model.Pediatr Drugs 12(3):201–211

4. Pilla Reddy V, Kozielska M, Suleiman AA, Johnson M, VermeulenA, Liu J et al (2013) Pharmacokinetic-pharmacodynamic modelingof antipsychotic drugs in patients with schizophrenia part I: the useof PANSS total score and clinical utility. Schizophr Res 146(1–3):144–152

5. Tylutki Z, Jawień W, Ciszowski K, Wilimowska J, Anand JS(2015) Abnormal olanzapine toxicokinetic profiles – populationpharmacokinetic analysis. Toxicol Mech Methods 25(1):1–12

6. Bigos KL, Pollock BG, Coley KC, Miller DD, Marder SR,Aravagiri M et al (2008) Sex, race, and smoking impact olanzapineexposure. J Clin Pharmacol 48(2):157–165

7. Davies SJ, Mulsant BH, Flint AJ, Meyers BS, Rothschild AJ,Whyte EM et al (2015) The impact of sertraline Co-administration on the pharmacokinetics of olanzapine: a populationpharmacokinetic analysis of the STOP-PD. Clin Pharmacokinet54(11):1161–1168

8. Hiemke C, Dragicevic A, Gründer G, Hätter S, Sachse J,Vernaleken I et al (2004) Therapeutic monitoring of new antipsy-chotic drugs. Ther Drug Monit 26(2):156–160

9. Mandema JW, Gibbs M, Boyd RA, Wada DR, Pfister M (2011)Model-based meta-analysis for comparative efficacy and safety:application in drug development and beyond. Clin PharmacolTher 90(6):766–769

10. WählbyU, JonssonEN,KarlssonMO(2002)Comparison of stepwisecovariate model building strategies in population pharmacokinetic-pharmacodynamic analysis. AAPS PharmSci 4(4):E27

11. RenYP,DengCH,WangXP,ZhouTY,LuW(2011)Comparisonstudyofmodel evaluationmethods: normalized prediction distribution errorsvs. visual predictive check. YaoXueXueBao 46(9):1123–1131

12. Zakeri-Milani P, Islambulchilar Z, Ghanbarzadeh S, Valizadeh H(2013) Single dose bioequivalence study of two brands ofolanzapine 10 mg tablets in Iranian healthy volunteers. Drug Res(Stuttg) 63(7):346–350

13. Jacobs BS, Colbers AP, Velthoven-GraaflandK, Schouwenberg BJ,Burger DM (2014) Effect of fosamprenavir/ritonavir on the phar-macokinetics of single-dose olanzapine in healthy volunteers. Int JAntimicrob Agents 44(2):173–177

14. Wang CY, Zhang ZJ, Li WB, Zhai YM, Cai ZJ, Weng YZ et al(2004) The differential effects of steady-state fluvoxamine on thepharmacokinetics of olanzapine and clozapine in healthy volun-teers. J Clin Pharmacol 44(7):785–792

15. Wu TH, Chiu CC, Shen WW, Lin FW, Wang LH, Chen HY et al(2008) Pharmacokinetics of olanzapine in Chinese male schizo-phrenic patients with various smoking behaviors. ProgNeuropsychopharmacol Biol Psychiatry 32(8):1889–1893

16. Sidhu J, Job S, Bullman J, Francis E, Abbott R, Ascher J et al(2006) Pharmacokinetics and tolerability of lamotrigine andolanzapine coadministered to healthy subjects. Br J ClinPharmacol 61(4):420–426

17. Chiu CC, Lane HY, Huang MC, Liu HC, Jann MW, Hon YY et al(2004) Dose-dependent alternations in the pharmacokinetics ofolanzapine during coadministration of fluvoxamine in patients withschizophrenia. J Clin Pharmacol 44(12):1385–1390

18. Penzak SR, Hon YY, Lawhorn WD, Shirley KL, Spratlin V, JannMW (2002) Influence of ritonavir on olanzapine pharmacokineticsin healthy volunteers. J Clin Psychopharmacol 22(4):366–370

19. Markowitz JS, DeVane CL, Malcolm RJ, Gefroh HA, Wang JS,Zhu HJ et al (2006) Pharmacokinetics of olanzapine after single-dose oral administration of standard tablet versus normal and sub-lingual administration of an orally disintegrating tablet in normalvolunteers. J Clin Pharmacol 46(2):164–171

20. Chen Q, Zhang MQ, Liu Y, Liu YM, Li SJ, Lu C et al (2012)Pharmacokinetics and bioequivalence of 2 tablet formulations ofolanzapine in healthy Chinese volunteers: a randomized, open-la-bel, single-dose study. Arzneim Forsch 62(11):508–512

21. Elshafeey AH, Elsherbiny MA, Fathallah MM (2009) A single-dose, randomized, two-way crossover study comparing twoolanzapine tablet products in healthy adult male volunteers underfasting conditions. Clin Ther 31(3):600–608

22. Gossen D, de Suray JM, Vandenhende F, Onkelinx C, Gangji D(2002) Influence of fluoxetine on olanzapine pharmacokinetics.AAPS PharmSci 4(2):E11

23. Markowitz JS, DevaneCL, ListonHL, BoultonDW,Risch SC (2002)The effects of probenecid on the disposition of risperidone andolanzapine in healthy volunteers. Clin Pharmacol Ther 71(1):30–38

24. Hägg S, Spigset O, Lakso HA, Dahlqvist R (2001) Olanzapinedisposition in humans is unrelated to CYP1A2 and CYP2D6 phe-notypes. Eur J Clin Pharmacol 57(6–7):493–497

25. Macias WL, Bergstrom RF, Cerimele BJ, Kassahun K, Tatum DE,Callaghan JT (1998) Lack of effect of olanzapine on the pharma-cokinetics of a single aminophylline dose in healthy men.Pharmacotherapy 18(6):1237–1248

Eur J Clin Pharmacol (2016) 72:933–944 943

26. Callaghan JT, Cerimele BJ, Kassahun KJ, Nyhart EH Jr, Hoyes-Beehler PJ, Kondraske GV (1997) Olanzapine: interaction studywith imipramine. J Clin Pharmacol 37(10):971–978

27. Cavalcanti Bedor NC, Galindo Bedor DC, Miranda de Sousa CE etal (2015) The development and validation of a method for quanti-fying olanzapine in human plasma by liquid chromatography tan-dem mass spectrometry and its application in a pharmacokineticstudy. Clin Exp Pharmacol Physiol 42(3):305–313

28. Lucas RA, Gilfillan DJ, Bergstrom RF (1998) A pharmacokineticinteraction between carbamazepine and olanzapine: observationson possible mechanism. Eur J Clin Pharmacol 54(8):639–643

29. Yu LY, Lou HG, Ruan ZR, Jiang B, Chen JL (2012)Pharmacokinetics and bioequivalence of domestic olanzapine tab-lets in healthy Chinese volunteers. Zhongguo Yaoxue Zazhi(Beijing, China) 47(23):1934–1936

30. Chatsiricharoenkul S, Niyomnaitham S, Pongnarin P, Sathirakul K,Kongpatanakul S (2011) Bioequivalence study of 10mg olanzapinetablets in healthy Thai volunteers. J Bioequivalence Bioavailab3(5):082–085

31. Cánovas M, Torres F, Domenech G, Cebrecos J, Pelagio P,ManríquezM et al (2011) Bioequivalence evaluation of two dosageforms of olanzapine 10 mg formulations in healthy volunteers.Arzneimittelforschung 61(2):75–79

32. Li WB, Wang CY, Zhai YM, Weng YZ, Jiang ZN (2005)Pharmacokinetics and bioequivalence of domestic and importedolanzapine tablets in healthy volunteers. Zhongguo Xinyao Zazhi14(5):599–602

33. Ji HF, Xie L, Wang J, Zhang Y, Hou YY, Liu XD (2002)Determination of olanzapine in plasma by HPLC and its pharma-cokinetics in men. Zhongguo Yaoke Daxue Xuebao 33(5):397–400

34. Lahti AC, Weiler MA, Medoff DR, Tamminga CA, Holcomb HH(2005) Functional effects of single dose first- and second-generation antipsychotic administration in subjects with schizo-phrenia. Psychiatry Res 139(1):19–30

35. Zuo XC, Ng CM, Barrett JS, Luo AJ, Zhang BK, Deng CH et al(2013) Effects of CYP3A4 and CYP3A5 polymorphisms on tacro-limus pharmacokinetics in Chinese adult renal transplant recipients:a population pharmacokinetic analysis. Pharmacogenet Genomics23(5):251–261

36. Shang DW, Li LJ, Wang XP, Wen YG, Ren YP, GuoWet al (2014)Population pharmacokinetic/pharmacodynamic model of clozapinefor characterizing the relationship between accumulated exposureand PANSS scores in patients with schizophrenia. Ther DrugMonit36(3):378–386

37. Ji SM, Shang DW, Wang XP, Li AN, Ren YP, Li L et al (2014)Population pharmacokinetics of risperidone based on meta-analysisand its application in therapeutic drug monitoring of Chineseschizophrenic patients. J Chin Pharm Sci 23(2):75–82

38. Perera V, Gross AS, Polasek TM, Qin Y, Rao G, Forrest A et al(2013) Considering CYP1A2 phenotype and genotype for optimiz-ing the dose of olanzapine in the management of schizophrenia.Expert Opin Drug Metab Toxicol 9(9):1115–1137

39. Citrome L, Stauffer VL, Chen L, Kinon BJ, Kurtz DL, Jacobson JGet al (2009) Olanzapine plasma concentrations after treatment with10, 20, and 40 mg/d in patients with schizophrenia: an analysis ofcorrelations with efficacy, weight gain, and prolactin concentration.J Clin Psychopharmacol 29(3):278–283

40. Sathirakul K, Chan C, Teng L, Bergstrom RF, Yeo KP, Wise SD(2003) Olanzapine pharmacokinetics are similar in Chinese andCaucasian subjects. Br J Clin Pharmacol 56(2):184–187

944 Eur J Clin Pharmacol (2016) 72:933–944

本文献由“学霸图书馆-文献云下载”收集自网络，仅供学习交流使用。

学霸图书馆（www.xuebalib.com）是一个“整合众多图书馆数据库资源，

提供一站式文献检索和下载服务”的24 小时在线不限IP

图书馆。

图书馆致力于便利、促进学习与科研，提供最强文献下载服务。

图书馆导航：

图书馆首页 文献云下载 图书馆入口 外文数据库大全 疑难文献辅助工具