Biol. Pharm. Bull. 38, 996–1004 (2015)

996

Vol. 38, No. 7

Regular Article

Population Pharmacokinetics in China: The Dynamics of Intravenous Voriconazole in Critically Ill Patients with Pulmonary Disease Wenying Chen,a,# Hui Xie,a,# Fenghua Liang,a Dongmei Meng,a Jianzhong Rui,b Xueyan Yin,c Tiantian Zhang,c Xianglin Xiao,a Shaohui Cai,d Xiaoqing Liu,*,a and Yimin Li*,a a

 The First Affiliated Hospital of Guangzhou Medical University; Guangzhou 510120, P. R. China: b Jinling Hospital; Nanjing 210002, P. R. China: c School of Pharmaceutical Science, Sun Yat-Sen Univeristy; Guangzhou 510006, P. R. China: and d Department of Clinical Pharmacy, College of Pharmacy, Jinan University; Guangzhou 510632, P. R. China. Received November 9, 2014; accepted April 27, 2015 Pharmacokinetic research in China on the use of voriconazole in critically ill adult patients with different pulmonary diseases remains to be explored. This study evaluated the population pharmacokinetics of the use of voriconazole (VRC) in critically ill patients to determine covariate effects on VRC pharmacokinetics by NONMEM, which could further optimize VRC dosing in this population. A one-compartment model with first-order absorption and elimination best fit the data, giving 4.28 L/h clearance and 93.4 L volume of distribution of VRC. The model variability, described as an approximate percentage coefficient of interindividual variability in clearance and volume of distribution, was 72.94% and 26.50%, respectively. A significant association between Cmin and drug response or grade 2 hepatotoxicity was observed ( p=0.002, 18 years old) who  received  VRC  therapy  from  March  2012  to  May  2013  to build population pharmacokinetics of VRC in this special group. For these patients, VRC therapeutic drug monitoring (TDM)  was  a  standard  clinical  care  and  determined  the  VRC 

* To whom correspondence should be addressed.  e-mail: [email protected]; [email protected] © 2015 The Pharmaceutical Society of Japan

Biol. Pharm. Bull. Vol. 38, No. 7 (2015)997

trough concentration under steady state after 6 maintenance doses (72 h). The  blood  samples  were  taken  at  0.5,  1,  1.5,  2,  4,  6,  9,  and  12 h,  or  any  two  points  of  that  in  every  study  occasion.  VRC  concentrations  were  determined  according  to  a  reported  high-performance  liquid  chromatography  (HPLC)-ultraviolet  detector method.16)  In  our  lab,  the  calibration  curves  of  VRC  were  linear  from  208  to  20800 ng/mL.  The  lower  limit  of  quantification  (LLOQ)  was  70 ng/mL.  Coefficients  of  variation  (CVs)  for  intraday  and  interday  precision  were  0.5–0.6%  and  1.5–5.0%  respectively  at  VRC  concentrations  of  0.208,  4.16, and 20.8 µg/mL. All  enrolled  patients  had  the  following  information  recorded:  age,  sex,  weight,  VRC  dose,  concomitant  drugs,  renal  function, hepatic function, invasive fungal infection (IFI) progression, and adverse events. The IFI progression was assessed  by clinical assessment (fever, signs, symptoms of infection, and  inflammatory  markers),  radiological  assessment  [computed  tomography  (CT)  or  magnetic  resonance  imaging  (MRI)]  and  microbiological  European  Organization  for  Research  and  Treatment  of  Cancer-Mycoses  Study  Group  (EORTC-MSG)  criteria.17)  Hepatotoxicity  was  classified  according  to  absolute  liver enzyme values according to the National Cancer Institute  grade  [for  T-Bil  grade  0,  none;  grade  1,  >ULN  (upper  limit  of normal)−1.5×ULN;  grade  2,  >1.5–3.0×ULN;  grade  3,  >3.0–10.0×ULN;  grade  4,  >10.0×ULN;  and  for  ALP,  GGT,  AST and ALT grade 0, none; grade 1, >ULN–2.5×ULN;  grade 2, >2.5–5.0×ULN;  grade  3,  >5.0–20.0×ULN;  grade  4,  >20.0×ULN].18) Population Pharmacokinetic Analysis This analysis was  performed  by  NONMEM  (double  precision,  Version  VI,  Level  2.0,  GloboMax  LLC,  Hanover,  MD,  U.S.A.)  with  G95  FORTRAN  complier  (gcc  3.3.2;  Free  Software  Foundation,  Boston,  MA,  U.S.A.)  using  mixed-effects  regression.  One-  and  two-compartment  models  with  first-order  elimination  were  adapted  to  evaluate  the  proper  basic  structural  pharmacokinetic model. The interindividual variability, residual variability,  and  model  misspecification  were  analyzed  based  on  additive-error,  exponential  error,  and  slope/intercept-error  model.  The  full  model  was  developed  with  the  forward  inclu-

sion–backward  elimination  technique.19)  A  stepwise  method  was  utilized  to  analyze  the  covariate.  The  final  regression  model  was  evaluated  by  plotting  predicted  concentration  versus  observed  concentrations,  concentration–time  profiles,  and  weighted  residuals  versus predicted concentrations. The accuracy  and  robustness  of  the  selected  final  model  were  evaluated by nonparametric bootstrap method and visual predictive check (VPC) method. The  bootstrap  procedure  was  performed  using  the  Wings  for  NONMEM  program  (WFN;  http://wfn.sourceforge.net/).  VPC  was  executed  using  Perl-speak-NONMEM  (PsN  3.1.0,  Uppsala  University,  Uppsala,  Sweden)  under  the  environment  of  Perl  language  (ActiveState  Software  Inc.).  X-pose  4.1.0  (Uppsala  University)  was  used  for  data  visualization  under  the environment of the R 2.9.0 (The R Development Core Team).20)  Using  1000  data  samples  simulated  by  NONMEN  based  on  the  parameters  of  the  final  model,  VPC  calculated  the 95% confidence interval (95% CI) of simulated concentrations at the corresponding time points and validated the model by comparing the observed values to the 95% CI. Concentration–Effect Relationship and the Optimal VRC Dose Simulation   Logistic  regression  analysis  [SPSS  (version  18.0,  SPSS  Inc,  Chicago,  IL,  U.S.A.)]  was  performed  to  evaluate  the  relationship  between  trough  concentration  (Cmin)  and  drug  response  as  well  as  grade  2  hepatotoxicity  (absence:  0;  presence:  1)  with  Cmin,  recorded  IFI  progression,  and  adverse  reaction.  Statistical  significance  was  assigned  at  2-sided p values of 0.05. The lower and upper acceptable VRC  concentrations  were  defined  to  be  more  than  80%  probability  of  efficacy  and  less  than  15%  probability  of  grade  2  hepatotoxicity based on logistic regression result. Based  on  the  established  population  pharmacokinetics  model,  we  further  adopted  the  predicted  CL and volume of distribution (Vd)  to  simulate  the  VRC  concentration  fluctuation under steady state in 1000 replicates with Monte Carlo  simulation. Four different maintenance dosages (100, 150, 200, 250 mg twice daily) were evaluated in the simulation by Crystal  Ball  12.1.2.2.0  (Oracle,  California,  CA,  U.S.A.)  to  verify  the optimal regimen for Chinese ICU patients with pulmonary  diseases.

Fig.  1.  VRC Individual Plasma Concentration-versus-Time Profiles (A) and Mean Plasma Concentration-versus-Time Profiles (B) in 62 Patients

998

Biol. Pharm. Bull.

RESULTS Analytical Quantifications of VRC After six times maintenance  therapies,  we  assumed  that  the  VRC  had  achieved its steady state. Then the VRC concentrations monitoring  was  started  since  the  sixth  maintenance  therapy  as  the  zero  points.  During  the  quantification  process,  no  measurements  were  less  than  the  LLOQ  and  the  individual  and  mean  VRC  concentration-versus-time  profile  of  62  eligible  patients  were listed in Fig. 1. The median minimum plasma concentration (Cmin)  was  3.26 µg/mL at time 0 and 3.76 µg/mL at 12 h, indicating that VRC did achieve the steady state since the sixth maintenance therapy.

Vol. 38, No. 7 (2015)

Demographics According to the inclusion criteria, 62 adult  (19–90  years  old)  ICU  patients  with  VRC  therapy  were  enrolled from March 2012 to May 2013. Patient demographics and other corresponding information are listed in Table 1. The 62 eligible patients all had clinical indication for fungal infection and 52 of them with positive microbiology results. Population Pharmacokinetic Analysis The pharmacokinetic  analysis  was  based  on  240  VRC  plasma  concentration  measurements. Since VRC can only be given intravenously, absorption rate constant (Ka)  and  bioavailability  of  VRC  were  excluded  in  this  study.  One-compartment  pharmacokinetic  model  with  first-order  elimination  adequately  described  the  data  with  significant  difference  of  objective  function  value 

Table  1.  Demographics and Biochemical Characteristics of the All 62 Patients and the 17 Patients with Full Amount Haemospasia Characteristic Age (years) Gender (male/female) Body weight (kg) Race (Chinese/other) Blood sample (n) Observed VRC concentration (µg/mL) Duration in ICU (d) Duration in hospital (d) Observed mortality [n/N (%)] SOFA score [median (range)] APACHE II score White blood cell (1×1012 L−1) Percentage of neutrophils (%) Alanine aminotransferase (U/L) Total bilirubin (µmol/L) Direct bilirubin (µmol/L) Total bile acid (µmol/L) Creatinine (µmol/L) Pathogenic microorganism Aspergillosis Candidiasis Other Infected location Pulmonary Possible pulmonary Suspected (persistent neutropenic fever) IFI response to antifungal therapy Time to clinical assessment after VRC therapy (d) Success [n (%)] Lack of response [n (%)] ADR probably associated with VRC Occurrence after VRC therapy (d) Hepatic toxicity (cholestatic hepatopathy) [n (%)] Neurotoxicity (encephalopathy) [n (%)] Visual disturbance [n (%)] GI reaction [n (%)] Concomitant drugs Levofloxacin Reduced glutathione Methylprednisolone Omeprazole Azithromycin

62 Patients

17 Patients

Mean±S.D. (range)

Mean±S.D. (range)

59.71±16.67 (19–90) 62 (42/20) 60.13±10.03 (41–84) 62/0 240 4.27±2.73 35.54±42.38 (4–81) 86±44.54 (6–262) 9/62 (14.52%) 4 (0–20) 21.6±13.8 9.12±6.18 (0.4–40.1) 75.59±16.26 37.53±46.37 (7–344) 12.13±5.86 3.16±2.21 (0.2–10.3) 8.89±7.94 (1.1–54.4) 65.25±27.44 52 37 12 3 62 45 12 5

63.59±16.16 17 (11/6) 58.59±7.15 17/0 125 4.76±2.55 35.5±49.87 86±41.03 (9–181) 4/17 (23.53%) 5 (0–20) 22.3±11.5 10.64±5.60 82.66±10.31 28.91±21.19 8.44±6.45 4.02±2.40 14.66±7.75 67.50±24.85 17 13 4 0 17 16 1 0

3.93±1.36 (1–7) 55 (88.71%) 7 (11.29%)

3.86±0.76 (2–5) 15 (88.24%) 2 (11.76%)

9.32±7.02 (2–30) 15/62 (24.19%) 4/62 (6.45%) 1/62 (1.61%) 1/62 (1.61%)

4.5±3.32 (2–9) 5/17 (29.41) 1/17 (5.88%) 0/17 (0%) 0/17 (0%)

5 8 34 29 5

2 1 9 11 1

All values are depicted as mean±S.D. or number of patients. SOFA: sequential organ failure assessment; APACHE: acute physiology and chronic health evaluation; ADR:  adverse reaction; GI: gastrointestinal.

Biol. Pharm. Bull. Vol. 38, No. 7 (2015)999

(DOFV).  The  interindividual  variability  (IIV)  was  described  in  an  exponential  model  and  the  residual  variability  was  described  in  a  constant  coefficient  model.  The  population  estimates of total CL, Vd,  interindividual  and  residual  errors  were  summarized in Table 2. By  the  forward  inclusion-backward  elimination  technique,  18  covariates  were  identified  to  add  into  the  full  model,  including  age,  gender,  body  weight,  concomitant  drugs  which  were  known  to  be  inhibitors  or  inducers  through  any  way  of  interaction  with  voriconazole  and/or  azoles,  blood  urea  nitrogen  (BUN),  creatinine  (CR),  uric  acid  (UA),  creatinine  clearance (CLCR),  hepatic  function  tests  included  albumin  (ALB),  alanine aminotransferase (ALT), aspartate aminotransferase (AST), alkaline phosphatase (ALP), γ-glutamyltranspeptidase  (γGT),  total  bilirubin  (TBIL),  direct  bilirubin  (DBIL),  triglyceride (TG), total cholesterol (CHO) and total bile acid (TBA). The  backward  method  demonstrated  the  direct  bilirubin  (DBIL)  as  a  significant  covariate  on  CL  (DOFV=17.45). The final  model  was  CL=θCL·(DBIL/2.6)θDBIL , Vd=θvd. And the

CL  was  estimated  to  be  4.28 L/h,  and  Vd  were  93.4 L.  The  inter-variability,  being  described  as  approximate  percentage  coefficient  of  variation  in  CL and Vd  were  72.94%  and  26.50%,  respectively.  The  residual  variability  was  13.0%.  The  estimates of shrinkage for CL and Vd were 4.15% and 48.32%,  respectively. The final population pharmacokinetic parameters  were summarized in Table 2. Model Evaluation Scatter plots of population and individual predictions versus observed VRC concentrations demonstrated  the  goodness-of-fit  of  the  structural  model  with  a  symmetric  distribution  (Figs.  2A,  B,  R 2  were  0.0978  and  0.9569,  respectively).  Weighted  residuals  were  in  the  acceptable  range  with  mean  and  variance  highly  close  to  zero.  A  symmetric distribution was found both in the plot of weighted  residuals versus population/individual predicted values VRC concentrations  (Figs.  2C,  D).  The  weighted  residual  versus population-predicted  concentration  plot  showed  a  slight  bias  at