Hello,
I am using an existing model to describe new study data. The covariate model
needs to be updated. The model is quite complicated: time-varying PK and many
IIVs. It takes a long time to converge. I want to remove some of the IIVs to
shorten the run time for covariate search. Unexpectedly, after removing the IIV
(one at a time), PRED=0 for all records. I got the same issue for multiple
IIVs. It's a proportional residual model on the original scale, IPRED was fine,
so dose was given. The model converged and parameter estimates were close to
values before removal, so it didn't seem like an estimation issue. Before
removal, PRED was fine and not a small number, so it couldn't be a digit issue
(I also tried to change format for $Table but it did not work).
I am having a hard time understanding why. Apology that I cannot post the full
control stream here. It's NONMEM 7.5 and intel fortran. Hope you can point me
to the right direction. Really appreciate your guidance!
Thank you,
Dongfen
All PRED=0 after removing IIV
Hi Dongfen,
Do you have IPRED computed correctly? Then there could be an error during the PRED compute. Do you have NOABORT on EST step? You can try to use NOHABORT and see what happen, or use $ABBREV PROTECT (in case this would mask the error of PRED computations).
Another option is to get PRED from IPRED, see manual, like:
PREDU=1
If(COMACT==1) PREDU=IPRED
$TABLE ... PREDU ...
If PREDU=1 in the table, then there was an error on the step that computes PRED values. If PREDU is reasonable but PRED is still 0, then the problem is somewhere else.
You may also try FNLETA=0 (thus avoiding recompute of ETAs on the table step): not sure whether it can help, but it is easy to try.
ADVAN13 is more stable than ADVAN6, so if you are using 6, you may try 13.
It looks like a numerical issue so anything that changes numerical computations is worthwhile to try (NSIG, SIGL, TOL, etc.)
Also, check the .lst file and the dos windows in case of helpful error messages.
Regards,
Leonid
Quoted reply history
On 2/22/2025 7:07 PM, Yuan, Dongfen [JRDUS] wrote:
> Hello,
>
> I am using an existing model to describe new study data. The covariate model needs to be updated. The model is quite complicated: time-varying PK and many IIVs. It takes a long time to converge. I want to remove some of the IIVs to shorten the run time for covariate search. Unexpectedly, after removing the IIV (one at a time), PRED=0 for all records. I got the same issue for multiple IIVs. It's a proportional residual model on the original scale, IPRED was fine, so dose was given. The model converged and parameter estimates were close to values before removal, so it didn't seem like an estimation issue. Before removal, PRED was fine and not a small number, so it couldn't be a digit issue (I also tried to change format for $Table but it did not work).
>
> I am having a hard time understanding why. Apology that I cannot post the full control stream here. It's NONMEM 7.5 and intel fortran. Hope you can point me to the right direction. Really appreciate your guidance!
>
> Thank you,
> Dongfen
Hi Dongfen,
I agree with Leonid, and it is likely that $ABBREV PROTECT will do the job. I however would encourage you to make an anonymous version of your model so you can share it with the group if it does not solve the problem.
It might be informative to run the table step without estimation (i.e. just with the initial values - only remove $EST and $COV) to see if PRED is still zero. If not, there may be some numerical issue propagated from EST/COV.
Anyhow, the problem is intriguing. Please let us know of your progress.
Hope it helps,
Jeroen
http://pd-value.com
[email protected]
@PD_value
+31 6 23118438
-- More value out of your data!
Quoted reply history
On 23-02-2025 02:49, Leonid Gibiansky wrote:
> Hi Dongfen,
>
> Do you have IPRED computed correctly? Then there could be an error during the PRED compute. Do you have NOABORT on EST step? You can try to use NOHABORT and see what happen, or use $ABBREV PROTECT (in case this would mask the error of PRED computations).
>
> Another option is to get PRED from IPRED, see manual, like:
> PREDU=1
> If(COMACT==1) PREDU=IPRED
>
> $TABLE ... PREDU ...
>
> If PREDU=1 in the table, then there was an error on the step that computes PRED values. If PREDU is reasonable but PRED is still 0, then the problem is somewhere else.
>
> You may also try FNLETA=0 (thus avoiding recompute of ETAs on the table step): not sure whether it can help, but it is easy to try.
>
> ADVAN13 is more stable than ADVAN6, so if you are using 6, you may try 13.
>
> It looks like a numerical issue so anything that changes numerical computations is worthwhile to try (NSIG, SIGL, TOL, etc.)
>
> Also, check the .lst file and the dos windows in case of helpful error messages.
>
> Regards,
> Leonid
>
> On 2/22/2025 7:07 PM, Yuan, Dongfen [JRDUS] wrote:
>
> > Hello,
> >
> > I am using an existing model to describe new study data. The covariate model needs to be updated. The model is quite complicated: time-varying PK and many IIVs. It takes a long time to converge. I want to remove some of the IIVs to shorten the run time for covariate search. Unexpectedly, after removing the IIV (one at a time), PRED=0 for all records. I got the same issue for multiple IIVs. It's a proportional residual model on the original scale, IPRED was fine, so dose was given. The model converged and parameter estimates were close to values before removal, so it didn't seem like an estimation issue. Before removal, PRED was fine and not a small number, so it couldn't be a digit issue (I also tried to change format for $Table but it did not work).
> >
> > I am having a hard time understanding why. Apology that I cannot post the full control stream here. It's NONMEM 7.5 and intel fortran. Hope you can point me to the right direction. Really appreciate your guidance!
> >
> > Thank you,
> > Dongfen
Hi Leonid and Jeroen,
The problem is solved. I really appreciate your timely guidance!
Diagnosis:
*
Remove $EST and $COV: PRED !=0 -->numerical issue propagated from $EST/$COV
What worked:
*
Changing from NOABORT to NOHABORT in $EST
*
Adding $ABBREV PROTECT
What didn't work:
*
FNLETA=0
Thank you,
Dongfen
Quoted reply history
________________________________
From: Jeroen Elassaiss-Schaap (PD-value B.V.) <[email protected]>
Sent: Sunday, February 23, 2025 2:56 AM
To: Leonid Gibiansky <[email protected]>; Yuan, Dongfen [JRDUS]
<[email protected]>; [email protected] <[email protected]>
Subject: [EXTERNAL] Re: [NMusers] All PRED=0 after removing IIV
Hi Dongfen,
I agree with Leonid, and it is likely that $ABBREV PROTECT will do the
job. I however would encourage you to make an anonymous version of your
model so you can share it with the group if it does not solve the problem.
It might be informative to run the table step without estimation (i.e.
just with the initial values - only remove $EST and $COV) to see if PRED
is still zero. If not, there may be some numerical issue propagated from
EST/COV.
Anyhow, the problem is intriguing. Please let us know of your progress.
Hope it helps,
Jeroen
http://pd-value.com/
[email protected]
@PD_value
+31 6 23118438
-- More value out of your data!
On 23-02-2025 02:49, Leonid Gibiansky wrote:
> Hi Dongfen,
>
> Do you have IPRED computed correctly? Then there could be an error
> during the PRED compute. Do you have NOABORT on EST step? You can try
> to use NOHABORT and see what happen, or use $ABBREV PROTECT (in case
> this would mask the error of PRED computations).
>
> Another option is to get PRED from IPRED, see manual, like:
> PREDU=1
> If(COMACT==1) PREDU=IPRED
>
> $TABLE ... PREDU ...
>
> If PREDU=1 in the table, then there was an error on the step that
> computes PRED values. If PREDU is reasonable but PRED is still 0, then
> the problem is somewhere else.
>
> You may also try FNLETA=0 (thus avoiding recompute of ETAs on the
> table step): not sure whether it can help, but it is easy to try.
>
> ADVAN13 is more stable than ADVAN6, so if you are using 6, you may try
> 13.
>
> It looks like a numerical issue so anything that changes numerical
> computations is worthwhile to try (NSIG, SIGL, TOL, etc.)
>
> Also, check the .lst file and the dos windows in case of helpful error
> messages.
>
> Regards,
> Leonid
>
>
>
>
> On 2/22/2025 7:07 PM, Yuan, Dongfen [JRDUS] wrote:
>> Hello,
>>
>> I am using an existing model to describe new study data. The
>> covariate model needs to be updated. The model is quite complicated:
>> time-varying PK and many IIVs. It takes a long time to converge. I
>> want to remove some of the IIVs to shorten the run time for covariate
>> search. Unexpectedly, after removing the IIV (one at a time), PRED=0
>> for all records. I got the same issue for multiple IIVs. It's a
>> proportional residual model on the original scale, IPRED was fine, so
>> dose was given. The model converged and parameter estimates were
>> close to values before removal, so it didn't seem like an estimation
>> issue. Before removal, PRED was fine and not a small number, so it
>> couldn't be a digit issue (I also tried to change format for $Table
>> but it did not work).
>>
>> I am having a hard time understanding why. Apology that I cannot post
>> the full control stream here. It's NONMEM 7.5 and intel fortran. Hope
>> you can point me to the right direction. Really appreciate your
>> guidance!
>>
>> Thank you,
>> Dongfen
>
>