Impact of Lopinavir Limit of Quantification (LOQ)-Censored Data Replacement on Population Pharmacokinetic (PK) Plasma and Saliva Modeling in HIV-Infected Children
Michael Neely (1), Natella Rakhmanina (2), John Van Den Anker (2), Steven Soldin (2), Mike Van Guilder (1), Alan Schumitzky (1), Roger Jelliffe (1)
(1) University of Southern California, Los Angeles; (2) National Children’s Medical Center, Washington DC
Objectives: Numerous methods have been proposed to replace LOQ-censored drug concentrations.[1] We quantified the effect of 4 on a PK model of lopinavir (LPV) in plasma and saliva in HIV-infected children.
Methods: Analyses were performed on data from 23 HIV-infected children. LPV concentrations were available from blood and saliva obtained immediately before a witnessed LPV dose, and then 0.5, 1, 4, 8 and 12 hours post-dose. The LPV assay had a validated LOQ of 10 ng/mL.[2] Coefficients of variation at concentrations of 10, 100, and 500 ng/mL were reported as 1.8 to 5.2%. Data were fitted using the USC*PACK NPAG software to a 2 compartment model (VC, VP) with linear parameters, including absorption (KA) after a lag time (Tlag), elimination (KEL), and inter-compartmental exchange (KCP, KPC). Because the dosing history was unreliable, initial conditions in each compartment were set to the first trough concentration. Four sets of patient data were used, identical save for replacement of LOQ with either: 1) 0.5*LOQ (5); 2) 0; 3) omit; or 4) random number 0-10. USC*PACK was used to model a polynomial fitted to the assay calibration concentrations and their associated standard deviations (SD), and two additional points at 0 and 5 ng/mL, each with an SD of 3. All measured or random concentrations, including 0, were weighted by the reciprocal of the calculated SD2 at that concentration
Results:
Mean parameter estimates:
|
Method |
TLAG |
KA |
KEL |
KCP |
VC |
KPC |
VP |
|
0.5*LOQ |
2.07 |
3.32 |
0.06 |
34.06 |
68.10 |
1.28 |
1093.91 |
|
0 |
1.91 |
2.43 |
0.06 |
25.57 |
60.69 |
0.53 |
1055.99 |
|
Omit |
1.99 |
2.49 |
0.21 |
25.43 |
60.71 |
0.59 |
895.02 |
|
Random |
1.81 |
2.47 |
0.22 |
21.38 |
57.68 |
1.82 |
951.61 |
Log-likelihood:
|
0.5 |
0 |
Omit |
Random |
|
-360.285 |
-328.363 |
-303.349 |
-290.189 |
Regression line of predicted vs. observed:
|
0.5 |
0 |
Omit |
Random | |||||
|
Plasma |
Saliva |
P |
S |
P |
S |
P |
S | |
|
R-squared |
0.967 |
0.114 |
0.975 |
0.067 |
0.977 |
0.103 |
0.981 |
0.147 |
|
Intercept |
0.252 |
0.101 |
0.135 |
0.121 |
0.226 |
0.135 |
0.242 |
0.104 |
|
Slope |
0.969 |
0.473 |
0.980 |
0.259 |
0.974 |
0.434 |
0.992 |
0.507 |
Conclusions: All four methods resulted in similar parameter estimates and predicted vs. observed LPV concentrations in plasma, less so in saliva. Substitution of random concentrations performed the best for both compartments, particularly saliva. The population parameter estimates from the Random method were 2.8 x 1030, 3.8 x 1016 and 5.2 x 105 times as likely as the 0.5, 0 and Omit methods, respectively. Since model likelihood is strongly dependent on method of LOQ replacement, LOQ censoring should be abolished in favor of reporting all concentrations and SDs, even 0.
References:
[1] Beal SL. Ways to fit a PK model with some data below the quantification limit. J Pharmacokinet Pharmacodyn. 2001; 28(5):481-504.
[2] Volosov A, Alexander C, Ting L, Soldin SJ. Simple rapid method for quantification of antiretrovirals by liquid chromatography-tandem mass-spectrometry. Clin Biochem. 2002; 35(2):99-103.
