From: Ahn, Jae Eun <*jaeeun.ahn*>

Date: Thu, 11 Feb 2010 12:06:49 -0500

Dear Ethan,

The denominator of DUM should be in the standard deviation level, not

variance, so it should be

W = =

SQRT(THETA(7)**2*FLAG+THETA(8)**2*(1-FLAG)+THETA(9)**2*IPRED**2)

IF (TYPE.EQ.2.AND.CMT.EQ.2) DUM=(LOQ-IPRED)/W

Note that the way you coded DUM is equivalent to DUM=(LOQ-IPRED)/W**2

In your code for log-transformed data, I see that the denominator is

specified as a standard deviation.

However, it is not clear to me whether you want to use a different

magnitude of proportional error or additive error for studies whose

number is greater than 3. In the code for natural DV, THETA(8) is for

the different additive error of such studies, whereas THETA(8) in the

log DV corresponds to the different proportional error.

Best regards,

Jae Eun

________________________________

From: owner-nmusers

On Behalf Of Ethan Wu

Sent: Thursday, February 11, 2010 10:53 AM

To: nmusers

Subject: [NMusers] help on diagnostic of a problem -- about M3 method

for LLOQ:

Dear users,

I am working on a dataset with not small percentage below LLOQ.

when I implemented M3 method to account for this, I always getting

this error message:

0PROGRAM TERMINATED BY OBJ

ERROR IN CLIK WITH INDIVIDUAL 2 ID=0.19000000E+02

CONDITIONAL LIKELIHOOD FOR SOME OBSERVATION ESTIMATED TO BE NEGLIGIBLE

note that ID 19 is the first subject in the dataset.

this is how I specify the $ERR (a combinative erro) and $EST:

__________

$ERROR

IPRED = F

IRES = DV-IPRED

FLAG=1

IF(STDY.GT.3) FLAG=0

DEL=0

W =

SQRT(THETA(7)**2*FLAG+THETA(8)**2*(1-FLAG)+THETA(9)**2*IPRED**2)

IF(W.EQ.0)DEL=1

IWRES = IRES/(W+DEL)

IF (TYPE.EQ.2.AND.CMT.EQ.2)

DUM=(LOQ-IPRED)/(THETA(7)**2*FLAG+THETA(8)**2*(1-FLAG)+THETA(9)**2*IPRE=

D

**2)

CUMD=PHI(DUM)

IF (TYPE.EQ.2) THEN

F_FLAG=1

Y=CUMD

ENDIF

IF (TYPE.EQ.1) THEN

F_FLAG=0

Y = IPRED+W*EPS(1)

ENDIF

$ESTIMATION METHOD=1 LAPLACIAN MAXEVAL=9999 PRINT=10 NOABORT

or this:

$ESTIMATION METHOD=1 INT LAPLACIAN MAXEVAL=9999 PRINT=10 NOABORT

_____

Interstingly, when I use Log-transformed data, the model would run fine.

this is how I specify $ERR (an additive error) and $EST, all the other

aspects of model are the same between two tries.

Could someone help me understand what is happening? Thanks.

_____

$ERROR

DEL = 0

IF(F.LE.0) DEL = 1

IPRED = LOG(F+DEL)

IRES = DV-IPRED

FLAG=1

IF(STDY.GT.3) FLAG=0

W = SQRT(THETA(7)**2*FLAG+THETA(8)**2*(1-FLAG))

DEL1=0

IF(W.EQ.0)DEL1=1

IWRES = IRES/(W+DEL1)

IF (TYPE.EQ.2.AND.CMT.EQ.2) DUM =

(LOG(LOQ)-IPRED)/(THETA(7)*FLAG+THETA(8)*(1-FLAG))

CUMD=PHI(DUM)

IF (TYPE.EQ.2) THEN

F_FLAG=1

Y=CUMD

ENDIF

IF (TYPE.EQ.1) THEN

F_FLAG=0

Y=IPRED +EPS(1)*W

ENDIF

$ESTIMATION METHOD=1 LAPLACIAN MAXEVAL=9999 PRINT=10 NOABORT

Received on Thu Feb 11 2010 - 12:06:49 EST

Date: Thu, 11 Feb 2010 12:06:49 -0500

Dear Ethan,

The denominator of DUM should be in the standard deviation level, not

variance, so it should be

W = =

SQRT(THETA(7)**2*FLAG+THETA(8)**2*(1-FLAG)+THETA(9)**2*IPRED**2)

IF (TYPE.EQ.2.AND.CMT.EQ.2) DUM=(LOQ-IPRED)/W

Note that the way you coded DUM is equivalent to DUM=(LOQ-IPRED)/W**2

In your code for log-transformed data, I see that the denominator is

specified as a standard deviation.

However, it is not clear to me whether you want to use a different

magnitude of proportional error or additive error for studies whose

number is greater than 3. In the code for natural DV, THETA(8) is for

the different additive error of such studies, whereas THETA(8) in the

log DV corresponds to the different proportional error.

Best regards,

Jae Eun

________________________________

From: owner-nmusers

On Behalf Of Ethan Wu

Sent: Thursday, February 11, 2010 10:53 AM

To: nmusers

Subject: [NMusers] help on diagnostic of a problem -- about M3 method

for LLOQ:

Dear users,

I am working on a dataset with not small percentage below LLOQ.

when I implemented M3 method to account for this, I always getting

this error message:

0PROGRAM TERMINATED BY OBJ

ERROR IN CLIK WITH INDIVIDUAL 2 ID=0.19000000E+02

CONDITIONAL LIKELIHOOD FOR SOME OBSERVATION ESTIMATED TO BE NEGLIGIBLE

note that ID 19 is the first subject in the dataset.

this is how I specify the $ERR (a combinative erro) and $EST:

__________

$ERROR

IPRED = F

IRES = DV-IPRED

FLAG=1

IF(STDY.GT.3) FLAG=0

DEL=0

W =

SQRT(THETA(7)**2*FLAG+THETA(8)**2*(1-FLAG)+THETA(9)**2*IPRED**2)

IF(W.EQ.0)DEL=1

IWRES = IRES/(W+DEL)

IF (TYPE.EQ.2.AND.CMT.EQ.2)

DUM=(LOQ-IPRED)/(THETA(7)**2*FLAG+THETA(8)**2*(1-FLAG)+THETA(9)**2*IPRE=

D

**2)

CUMD=PHI(DUM)

IF (TYPE.EQ.2) THEN

F_FLAG=1

Y=CUMD

ENDIF

IF (TYPE.EQ.1) THEN

F_FLAG=0

Y = IPRED+W*EPS(1)

ENDIF

$ESTIMATION METHOD=1 LAPLACIAN MAXEVAL=9999 PRINT=10 NOABORT

or this:

$ESTIMATION METHOD=1 INT LAPLACIAN MAXEVAL=9999 PRINT=10 NOABORT

_____

Interstingly, when I use Log-transformed data, the model would run fine.

this is how I specify $ERR (an additive error) and $EST, all the other

aspects of model are the same between two tries.

Could someone help me understand what is happening? Thanks.

_____

$ERROR

DEL = 0

IF(F.LE.0) DEL = 1

IPRED = LOG(F+DEL)

IRES = DV-IPRED

FLAG=1

IF(STDY.GT.3) FLAG=0

W = SQRT(THETA(7)**2*FLAG+THETA(8)**2*(1-FLAG))

DEL1=0

IF(W.EQ.0)DEL1=1

IWRES = IRES/(W+DEL1)

IF (TYPE.EQ.2.AND.CMT.EQ.2) DUM =

(LOG(LOQ)-IPRED)/(THETA(7)*FLAG+THETA(8)*(1-FLAG))

CUMD=PHI(DUM)

IF (TYPE.EQ.2) THEN

F_FLAG=1

Y=CUMD

ENDIF

IF (TYPE.EQ.1) THEN

F_FLAG=0

Y=IPRED +EPS(1)*W

ENDIF

$ESTIMATION METHOD=1 LAPLACIAN MAXEVAL=9999 PRINT=10 NOABORT

Received on Thu Feb 11 2010 - 12:06:49 EST