NONMEM Users Network Archive

Hosted by Cognigen

Re: SEED status with SAEM method

From: Jakob Ribbing <jakob.ribbing>
Date: Thu, 5 Nov 2020 13:05:23 +0100

I see.

In that case this becomes kind of a theoretical exercise that can not be =
applied on a real dataset.
Estimation of both IIV and sigma may be “feasible" based on =
distributional assumptions.
But in reality, the random part of the parameter variability (IIV) does =
not perfectly follow a log-normal distribution.
And the residual error will not be perfectly normal (additive).
From that perspective I think it is healthy that FOCE/FOCEI does not =
converge, and it is not surprising that other estimation methods are ill =
behaved.

Maybe you should consider at least a percentage of subjects with =
multiple observations?

Best regards

Jakob



Jakob Ribbing, Ph.D.

Senior Consultant, Pharmetheus AB


Cell/Mobile: +46 (0)70 514 33 77

Jakob.Ribbing

www.pharmetheus.com

Phone, Office: +46 (0)18 513 328

Uppsala Science Park, Dag Hammarskjölds väg 36B
SE-752 37 Uppsala, Sweden



This communication is confidential and is only intended for the use of =
the individual or entity to which it is directed. It may contain =
information that is privileged and exempt from disclosure under =
applicable law. If you are not the intended recipient please notify us =
immediately. Please do not copy it or disclose its contents to any other =
person.




> On 5 Nov 2020, at 11:58, hai le ba <lebahai01
>
> Sorry, I've copied a wrong code of model.
> The right is:
> "$PK
>
> MU_1= LOG(THETA(1))
>
> KA=EXP(MU_1+ETA(1))
>
> MU_2=LOG(THETA(2))+ THETA(4)*LOG(AGE/60.13)
> CL= EXP(MU_2+ETA(2))
>
> MU_3=LOG(THETA(3))+THETA(5)*LOG(WEI/66.49)
> V=EXP(MU_3+ETA(3))
>
> S2=V
> IF (TIME.LT.12) TAD=TIME
> IF (TIME.GE.12.AND.TIME.LT.24) TAD=TIME-12
> IF (TIME.GE.12.AND.TIME.LT.36) TAD=TIME-24
> IF (TIME.GE.36.AND.TIME.LT.48) TAD=TIME-36
>
> $ERROR
> EP1=EPS(1)
> IPRED=F
> IRES=DV-IPRED
> W=F
> IWRES=IRES/W
> Y=IPRED+EPS(1)*W
>
> Many thanks for your support!
> Best regards,
>
> Le jeu. 5 nov. 2020 à 11:15, hai le ba <lebahai01
<mailto:lebahai01
> Dear Sir Jakob and Jeroen,
> Thanks for your reply.
> In fact, i've a one compartment model, with two levels of random =
effect (IIV, EPS) and covariate effect on CL, V.
> $PK
>
> MU_1= LOG(THETA(1))
>
> KA=EXP(MU_1+ETA(1))
>
> MU_2=LOG(THETA(2)+ THETA(4)*LOG(AGE/60.13))
> CL= EXP(MU_2+ETA(2))
>
> MU_3=LOG(THETA(3)+THETA(5)*LOG(WEI/66.49))
> V=EXP(MU_3+ETA(3))
>
> S2=V
> IF (TIME.LT.12) TAD=TIME
> IF (TIME.GE.12.AND.TIME.LT.24) TAD=TIME-12
> IF (TIME.GE.12.AND.TIME.LT.36) TAD=TIME-24
> IF (TIME.GE.36.AND.TIME.LT.48) TAD=TIME-36
>
> $ERROR
> EP1=EPS(1)
> IPRED=F
> IRES=DV-IPRED
> W=F
> IWRES=IRES/W
> Y=IPRED+EPS(1)*W
>
> I've used this model to simulate the complete database (10 000 IDs). =
From this database, the smaller dataset was randomly generated with =
respect to the rule of one PK sample per each ID. After that, I want to =
check how many IDs are necessary to get a good estimation for each =
method ( FOCEI, SAEM, BAYES, composite (ITS/SAEM/IMP/BAYES))?
>
> For FOCEI, I can't get convergence.
>
> For SAEM, I've got some troubles with the stability for EPS when I =
tried to change the SEED status?
> Otherwise, the good estimations were obtained with RSE (10-20%) for =
the other parameters.
>
> <>
>
convergence for this setting of NBURN. <>
> "
> Convergence achieved
>
> <>
> STOCHASTIC PORTION WAS COMPLETED
> REDUCED STOCHASTIC PORTION WAS COMPLETED"
>
> Could you give me some instructions?
>
> Many thanks in advance for your responses.
> Best regards,
> Hai, LE Ba
> Ph.D Student
> La faculté de Pharmacie - Aix Marseille université
> Hanoi university of Pharmacy
> Email: lebahai01
> hailb
bahai.le
>
>
> On Thu, Nov 5, 2020 at 10:43 AM Jakob Ribbing =
<jakob.ribbing
wrote:
> Dear LE Ba,
>
> You did not share much information on your model, except that you have =
one observations per subject and that EPS (or rather sigma) is estimated =
(I assume then a continuous endpoint).
> Is thisyour only level of random effect, so that you have no omega/eta =
for IOV or IIV? With one observations per subject it will be difficult =
to find support for more than one level.
> And for a model with only residual error (no IIV), would not the FO =
Estimation method be sufficient, so why would you need to use SAEM for =
this data/model?
>
> Best regards
>
> Jakob
>
>
>
> Jakob Ribbing, Ph.D.
>
> Senior Consultant, Pharmetheus AB
>
>
> Cell/Mobile: +46 (0)70 514 33 77
>
> Jakob.Ribbing
>
> www.pharmetheus.com <http://www.pharmetheus.com/>
>
> Phone, Office: +46 (0)18 513 328
>
> Uppsala Science Park, Dag Hammarskjölds väg 36B
> SE-752 37 Uppsala, Sweden
>
>
>
> This communication is confidential and is only intended for the use of =
the individual or entity to which it is directed. It may contain =
information that is privileged and exempt from disclosure under =
applicable law. If you are not the intended recipient please notify us =
immediately. Please do not copy it or disclose its contents to any other =
person.
>
>
>
>
>> On 5 Nov 2020, at 09:41, hai le ba <lebahai01
<mailto:lebahai01
>>
>> Hello nmusers,
>> I've tried to test the SAEM method on my dataset (1 sample for each =
ID) with different SEED status. Sometimes, I've got good results (with =
RSE <50%), but sometimes I can't get it.
>>
>> Example:
>> $EST METHOD=SAEM INTERACTION NBURN=3000 NITER=500 AUTO=1 =
PRINT=100 SEED=123456789
>> $EST METHOD=IMP EONLY=1 PRINT=1 NITER=5 ISAMPLE=1000 =
MAPITER=0
>> I've got EPS : 0.00064 (RSE=63%)
>>
>> but
>> $EST METHOD=SAEM INTERACTION NBURN=3000 NITER=500 AUTO=1 =
PRINT=100 SEED=123235489
>> $EST METHOD=IMP EONLY=1 PRINT=1 NITER=5 ISAMPLE=1000 =
MAPITER=0
>> I've got EPS : 0.0223 (RSE=18%)
>>
>> Could i trust on the good results (EPS=0.0223)? How can we control =
this problem?
>>
>> Thank you very much,
>> Hai, LE Ba
>> Ph.D Student
>> La faculté de Pharmacie - Aix Marseille université
>> Hanoi university of Pharmacy
>> Email: lebahai01
>> hailb
bahai.le
>


Received on Thu Nov 05 2020 - 07:05:23 EST

The NONMEM Users Network is maintained by ICON plc. Requests to subscribe to the network should be sent to: nmusers-request@iconplc.com.

Once subscribed, you may contribute to the discussion by emailing: nmusers@globomaxnm.com.