Jump to content


Photo

Why does Phoenix sometimes fit all etas as positive


  • Please log in to reply
5 replies to this topic

#1 DouglasFerguson

DouglasFerguson

    Newbie

  • Members
  • Pip
  • 4 posts

Posted 17 June 2016 - 07:06 PM

Now and again I encounter the situation where when applying a population model where all the etas (for a given parameter) are all determined to be positive (or all determined to be negative)?  Is there something that can cause this to happen (such that the estimated population mean is outside the range of individual values?)

 

Today, when fitting a standard 2 compartment PK model (with inter-individual variability on Cl only) to well defined bi-phasic PK data, using well defined started values (and no upper or lower bounds) I get the following etas (nCl) for the 10 individuals when applying FOCE ELS.

 

0.0875447
0.119935
0.196114
0.364142
0.308166
0.57629
-0.0774463
0.482184
0.112062
0.212903
 
Appreciate this might be difficult to answer without further details - but I have observed the same thing on several occasions.  
 


#2 Simon Davis

Simon Davis

    Advanced Member

  • Administrators
  • 1,316 posts

Posted 20 June 2016 - 09:20 AM

Douglas, how does the rest of your fit look? is it possible you can provide the actual project, if not here on the forum then directly to me to have a look at  in confidence?

 

  many thanks, Simon.


Edited by Simon Davis, 20 June 2016 - 09:20 AM.


#3 BP1968

BP1968

    Newbie

  • Members
  • Pip
  • 8 posts

Posted 25 April 2018 - 04:35 PM

Hi, I realize this is an old post, but was there any resolution/advice that can be shared? I am running into an identical problem with all negative etas and a theta value that seems too large. I tried bounds, but the -2LL and AIC are much larger (and the estimate is my upper bound).



#4 Simon Davis

Simon Davis

    Advanced Member

  • Administrators
  • 1,316 posts

Posted 25 April 2018 - 05:08 PM

Hi Bruno, generally Phoenix models perform better without bounds, if the estimate is hitting your upper bound i would suggest you increase or remove that bound.

If you are using a bound I suggest only on one 'side' e.g zero to prevent the algorithm going into a physiologically impossible negative space.
 

Beyond these general comments we would need to see your project to help further i think.

 

 Simon.



#5 BP1968

BP1968

    Newbie

  • Members
  • Pip
  • 8 posts

Posted 25 April 2018 - 06:46 PM

Simon,

 

Thank you for your response. I started without bounds, and the tv estimate I get is unlikely large (based on individual models, previous knowledge, and naive pooled approach) and the etas are all negative. I would be happy to share the project off forum if you are willing to take a look.



#6 BP1968

BP1968

    Newbie

  • Members
  • Pip
  • 8 posts

Posted 27 April 2018 - 03:01 PM

Simon,

 

Thank you for your response. I started without bounds, and the tv estimate I get is unlikely large (based on individual models, previous knowledge, and naive pooled approach) and the etas are all negative. I would be happy to share the project off forum if you are willing to take a look.

I should add that I have individual models that fit the data well, and that the tv estimate for the volume of the central compartment (without bounds) is much closer to the largest individual estimate than to the mean or median (this is a small study using a fairly uniform sample, so it is unlikely that covariates would play a role).






0 user(s) are reading this topic

0 members, 0 guests, 0 anonymous users