From: Samtani, Mahesh [JRDUS] <*MSamtani*>

Date: Wed, 13 Apr 2016 02:13:01 +0000

Hello,

I don't think both Omega and Sigma can be estimated. We’ve used this code below for modeling the endpoint which was a change from baseline measure [3 post baseline measurements and 3 doses]. We tried using either BLOCK in Sigma or BLOCK in OMEGA. Using a BLOCK with omega gave the same results as SAS. We shared the ETA for a given time point because we noticed that as time went on the variability increased.

Thanks,

Mahesh

$PRED

Q4 = 0

Q5 = 0

Q6 = 0

Q7 = 0

Q8 = 0

Q9 = 0

Q10 = 0

Q11 = 0

Q12 = 0

IF (TIME.EQ.1.AND.MG.EQ.0) Q4 = 1

IF (TIME.EQ.1.AND.MG.EQ.5) Q5 = 1

IF (TIME.EQ.1.AND.MG.GT.5) Q6 = 1

IF (TIME.EQ.2.AND.MG.EQ.0) Q7 = 1

IF (TIME.EQ.2.AND.MG.EQ.5) Q8 = 1

IF (TIME.EQ.2.AND.MG.GT.5) Q9 = 1

IF (TIME.EQ.3.AND.MG.EQ.0) Q10 = 1

IF (TIME.EQ.3.AND.MG.EQ.5) Q11 = 1

IF (TIME.EQ.3.AND.MG.GT.5) Q12 = 1

MUX=THETA(1)*Q4 +THETA(2)*Q5 +THETA(3)*Q6 +ETA(1)

IF (TIME.EQ.26) MUX=THETA(4)*Q7 +THETA(5)*Q8 +THETA(6)*Q9 +ETA(2)

IF (TIME.EQ.39) MUX=THETA(7)*Q10 +THETA(8)*Q11 +THETA(9)*Q12+ETA(3)

Y = MUX+EPS(1)

$THETA

1 1 1 2 2 2 3 3 3

$OMEGA BLOCK(3)

2.95E+01

1.80E+01 3.49E+01

1.97E+01 2.84E+01 4.81E+01

$SIGMA 1.00E-07 FIX

$EST MAX=9999 NOABORT PRINT=5 POSTHOC

$COV PRINT=E

-----Original Message-----

From: owner-nmusers nm.com [mailto:owner-nmusers

Sent: Tuesday, April 12, 2016 7:32 PM

To: Matts Kågedal; nmusers

Subject: Re: [NMusers] MMRM implementation in nonmem?

Hi Matts,

In my experience, a typical MMRM model is made of many unconnected model parameters usually with an unstructured variance/covariance matrix of residual error and a single inter-individual parameter. It would be relatively painful to copy and paste everything, but the general model could look like this (this was typed directly into email, so it may not work precisely).

Thanks,

Bill

$PRED

USE11 = 0 ; Trial arm 1, time point 1

USE12 = 0 ; Trial arm 1, time point 2

USE21 = 0 ; Trial arm 2, time point 1

USE22 = 0 ; Trial arm 2, time point 2

IF (TIME .EQ. 0 .AND. ARM .EQ. 1) USE11 = 1 IF (TIME .EQ. 1 .AND. ARM .EQ. 1) USE12 = 1 IF (TIME .EQ. 0 .AND. ARM .EQ. 2) USE21 = 1 IF (TIME .EQ. 1 .AND. ARM .EQ. 2) USE22 = 1

EFFECT = USE11 * THETA(1) + USE12 * THETA(2) + USE21 * THETA(3) + USE22

* THETA(4) + ETA(1)

$SIGMA BLOCK(4)

0.1

0.01 0.1

0.01 0.01 0.1

0.01 0.01 0.01 0.1

$THETA

1

1

1

1

$OMEGA

0.1

$ERROR

Y = EFFECT + USE11 * EPS(1) + USE12 * EPS(2) + USE21 * EPS(3) + USE22 *

EPS(4)

On 4/12/2016 7:10 PM, Matts Kågedal wrote:

*> Hi all,
*

*> Would any one be willing to share a nonmem control stream where mixed
*

*> model repeated measures (MMRM) is implemented.
*

*>
*

*> Best,
*

*> Matts Kagedal
*

*>
*

Received on Tue Apr 12 2016 - 22:13:01 EDT

Date: Wed, 13 Apr 2016 02:13:01 +0000

Hello,

I don't think both Omega and Sigma can be estimated. We’ve used this code below for modeling the endpoint which was a change from baseline measure [3 post baseline measurements and 3 doses]. We tried using either BLOCK in Sigma or BLOCK in OMEGA. Using a BLOCK with omega gave the same results as SAS. We shared the ETA for a given time point because we noticed that as time went on the variability increased.

Thanks,

Mahesh

$PRED

Q4 = 0

Q5 = 0

Q6 = 0

Q7 = 0

Q8 = 0

Q9 = 0

Q10 = 0

Q11 = 0

Q12 = 0

IF (TIME.EQ.1.AND.MG.EQ.0) Q4 = 1

IF (TIME.EQ.1.AND.MG.EQ.5) Q5 = 1

IF (TIME.EQ.1.AND.MG.GT.5) Q6 = 1

IF (TIME.EQ.2.AND.MG.EQ.0) Q7 = 1

IF (TIME.EQ.2.AND.MG.EQ.5) Q8 = 1

IF (TIME.EQ.2.AND.MG.GT.5) Q9 = 1

IF (TIME.EQ.3.AND.MG.EQ.0) Q10 = 1

IF (TIME.EQ.3.AND.MG.EQ.5) Q11 = 1

IF (TIME.EQ.3.AND.MG.GT.5) Q12 = 1

MUX=THETA(1)*Q4 +THETA(2)*Q5 +THETA(3)*Q6 +ETA(1)

IF (TIME.EQ.26) MUX=THETA(4)*Q7 +THETA(5)*Q8 +THETA(6)*Q9 +ETA(2)

IF (TIME.EQ.39) MUX=THETA(7)*Q10 +THETA(8)*Q11 +THETA(9)*Q12+ETA(3)

Y = MUX+EPS(1)

$THETA

1 1 1 2 2 2 3 3 3

$OMEGA BLOCK(3)

2.95E+01

1.80E+01 3.49E+01

1.97E+01 2.84E+01 4.81E+01

$SIGMA 1.00E-07 FIX

$EST MAX=9999 NOABORT PRINT=5 POSTHOC

$COV PRINT=E

-----Original Message-----

From: owner-nmusers nm.com [mailto:owner-nmusers

Sent: Tuesday, April 12, 2016 7:32 PM

To: Matts Kågedal; nmusers

Subject: Re: [NMusers] MMRM implementation in nonmem?

Hi Matts,

In my experience, a typical MMRM model is made of many unconnected model parameters usually with an unstructured variance/covariance matrix of residual error and a single inter-individual parameter. It would be relatively painful to copy and paste everything, but the general model could look like this (this was typed directly into email, so it may not work precisely).

Thanks,

Bill

$PRED

USE11 = 0 ; Trial arm 1, time point 1

USE12 = 0 ; Trial arm 1, time point 2

USE21 = 0 ; Trial arm 2, time point 1

USE22 = 0 ; Trial arm 2, time point 2

IF (TIME .EQ. 0 .AND. ARM .EQ. 1) USE11 = 1 IF (TIME .EQ. 1 .AND. ARM .EQ. 1) USE12 = 1 IF (TIME .EQ. 0 .AND. ARM .EQ. 2) USE21 = 1 IF (TIME .EQ. 1 .AND. ARM .EQ. 2) USE22 = 1

EFFECT = USE11 * THETA(1) + USE12 * THETA(2) + USE21 * THETA(3) + USE22

* THETA(4) + ETA(1)

$SIGMA BLOCK(4)

0.1

0.01 0.1

0.01 0.01 0.1

0.01 0.01 0.01 0.1

$THETA

1

1

1

1

$OMEGA

0.1

$ERROR

Y = EFFECT + USE11 * EPS(1) + USE12 * EPS(2) + USE21 * EPS(3) + USE22 *

EPS(4)

On 4/12/2016 7:10 PM, Matts Kågedal wrote:

Received on Tue Apr 12 2016 - 22:13:01 EDT