RE: OMEGA selection
Hi Ethan,
I think you have given too little info to diagnose your problem properly. We
don't even know if ETAs come in additively, proportionally, in logit
expressions or what (so values of 2 or 3 doesn't give the scale). Also, I
think that you mentioned 10-90% as values for correlations, whereas Bill
interpreted it as CVs for IIV. It was just not enough info to make the
distinction. If the model is so simple, why not show the whole model.
Mats
Mats Karlsson, PhD
Professor of Pharmacometrics
Dept of Pharmaceutical Biosciences
Uppsala University
Box 591
751 24 Uppsala Sweden
phone: +46 18 4714105
fax: +46 18 471 4003
Quoted reply history
From: [email protected] [mailto:[email protected]] On
Behalf Of Bill Bachman
Sent: Wednesday, April 15, 2009 7:46 PM
To: 'Ethan Wu'; 'Bachman, William'; [email protected]
Subject: RE: [NMusers] OMEGA selection
In my opinion, I would not remove those in the 10-90% range. I would be
suspect of anything over 100%, even with noisy data, they are being poorly
estimated.
_____
From: [email protected] [mailto:[email protected]] On
Behalf Of Ethan Wu
Sent: Wednesday, April 15, 2009 1:15 PM
To: Bachman, William; [email protected]
Subject: Re: [NMusers] OMEGA selection
some Etas estimated to be around 2 or 3, but since I am fitting a quite
noisy PD data, I think they are actually reasonable
no Etas close to 0
cov% esimated in the range of 10-90%.should those small ones like 10% be
taken out?
_____
From: "Bachman, William" <[email protected]>
To: Ethan Wu <[email protected]>; [email protected]
Sent: Wednesday, April 15, 2009 12:12:56 PM
Subject: RE: [NMusers] OMEGA selection
Well, the first thing that I would do is look at the magnitude of the
estimates of the etas. I would eliminate those etas that are poorly
estimated (essentially the very large values or those approaching zero).
_____
From: [email protected] [mailto:[email protected]] On
Behalf Of Ethan Wu
Sent: Wednesday, April 15, 2009 11:47 AM
To: [email protected]
Subject: [NMusers] OMEGA selection
Dear all,
I am fitting a PD response, and the equation goes like this:
total response = baseline+f(placebo response) +f(drug response)
first, I tried full omega block, and model was able to converge, but $COV
stop failed.
To me, this indicates that too many parameters in the model. The structure
model is rather simple one, so I think probably too many Etas.
I wonder is there a good principle of Eta reduction that I could implement
here. Any good reference?