Problem with handling BQL with M3 - additive & proportional error
Dear NONMEM Users
I'm working (with NM7) on a data set in which 24% of the data is BLQ. The data
set comprises 39 individuals, all of whom received a single bolus dose of an IV
drug, with samples collected for 12 hours afterwards. To deal with the BLQ
samples, I'm trying to implement the M3 method using a version with a combined
additive and proportional model. I took the code for this from Andreas
Lindauer's post to NMUsers on 20th April 2010 (thank you Andreas :-).
My models are minimising successfully and the $COV step is running, however,
something is clearly wrong :-S Here is the output for the first subject from a
3 comp, naive pooled (no ETAs) model:
ID EVID AMT TIME IPRED CWRES DV PRED RES
WRES
1 1 6840000 0 0 0
0 0 0 0
1 0 0 1 719.28 0 897.7
719.28 0 0
1 0 0 2 598.54 0 699.5
598.54 0 0
1 0 0 3 508.11 0 569
508.11 0 0
1 0 0 5 387.57 0 425.1
387.57 0 0
1 0 0 10 252.56 0 279
252.56 0 0
1 0 0 15 197.01 0 216.7
197.01 0 0
1 0 0 30 119.59 0 131.7
119.59 0 0
1 0 0 45 81.439 0 79.05
81.439 0 0
1 0 0 60 59.581 0 64.43
59.581 0 0
1 0 0 90 36.048 0 33.71
36.048 0 0
1 0 0 120 23.365 0 21.41
23.365 0 0
1 0 0 150 15.457 0 15.39
15.457 0 0
1 0 0 180 10.285 0 9.557
10.285 0 0
1 0 0 240 4.5695 0 6.273
4.5695 0 0
1 0 0 300 2.0317 0 0
0.60983 0 0
1 0 0 360 0.90337 0 0
0.65014 0 0
1 0 0 480 0.1786 0 0
0.67515 0 0
1 0 0 600 0.0353 0 0
0.67999 0 0
1 0 0 720 0.0069 0 0
0.68095 0 0
Firstly, I'm unclear as to why IPRED and PRED differ (for the BLQ samples, 300
minutes onwards) in the absence of ETAs. Secondly, you can see how the IPRED
predictions for the BQL points decrease over time as would be expected, while
the PRED concentrations rise? A model including ETA parameters shows the same
result i.e. IPRED predictions decrease over time, PRED for BLQ samples rise.
My control stream is pasted below, grateful as always for any ideas :-)
With thanks and best wishes
Ann
_______________________________________________________________________
Ann Rigby-Jones PhD MRSC
Research Fellow in Pharmacokinetics & Pharmacodynamics
Peninsula College of Medicine & Dentistry
N31, ITTC Phase 1
Tamar Science Park
1 Davy Road
Derriford
Plymouth
PL6 8BX
_______________________________________________________________________
$PROB PATIENTS AND NORMAL CONTROLS
$INPUT ID TIME DV FLG BRN EVID AMT RATE SEX AGE HGT WGT
$DATA alldata.CSV IGNORE=#
$SUBROUTINES ADVAN11 TRANS4
$PK
TVCL=THETA(1) ;*WGT**0.75
TVQ2=THETA(2) ;*WGT**0.75
TVQ3=THETA(3) ;*WGT**0.75
TVV1=THETA(4) ;*WGT**1
TVV2=THETA(5) ;*WGT**1
TVV3=THETA(6) ;*WGT**1
SDADD = THETA(7) ; the standard deviation of the additive part
CVPROP = THETA(8) ; the CV of the proportional part
CL=TVCL*EXP(ETA(1))
Q2=TVQ2*EXP(ETA(2))
Q3=TVQ3*EXP(ETA(3))
V1=TVV1*EXP(ETA(4))
V2=TVV2*EXP(ETA(5))
V3=TVV3*EXP(ETA(6))
S1=V1
$THETA
(0, 482) ; CL
(0, 644) ; Q2
(0, 87) ; Q3
(0, 9747) ; V1
(0, 9356) ; V2
(0, 20300) ; V3
(0, 3) ;SDADD
(0, 0.2) ;CVPROP
$OMEGA
(0 FIX) ;CL
(0 FIX) ;Q2
(0 FIX) ;Q3
(0 FIX) ;V1
(0 FIX) ;V2
(0 FIX) ;V3
$ERROR
;M3-Method
LOQ=5
IPRED=F
IRES = DV-IPRED
W=SQRT(SDADD**2+CVPROP**2*IPRED**2)
DEL=0
IF(W.EQ.0) DEL=1
IWRES=IRES/(W+DEL)
DUM=(LOQ-IPRED)/(W+DEL)
CUMD=PHI(DUM)
IF (FLG.EQ.0) THEN
F_FLAG=0
Y = IPRED+ERR(1)*W
ELSE
F_FLAG=1
Y=CUMD
ENDIF
$SIGMA 1 FIX
$EST SIG=4 METHOD=1 INTER LAPLACIAN NUMERICAL SLOW NOABORT MAXEVAL=99999
PRINT=5
$COV
$TABLE ID EVID AMT TIME IPRED CWRES
NOPRINT FILE=AllRecords.txt
$TABLE ID CL Q2 Q3 V1 V2 V3
ETA1 ETA2 ETA3 ETA4 ETA5 ETA6
FIRSTONLY NOPRINT NOAPPEND FILE=FirstRecords.txt