NONMEM Users Network Archive

Hosted by Cognigen

[NMusers] eigenvalues

From: Pavel Belo <nonmem_at_optonline.net>
Date: Wed, 04 Nov 2015 11:55:02 -0500 (EST)

Hello NONMEM Users,
 
I try to make sense of the results and one of the ways to do it is to
compare the same or similar models across software packages.  5x5 full=
 
omega matrix is used because it was prohibitive to remove some
insignificant correlations from the matrix without removing significant
correlations (All recommended ways to do it were tested. Diagonal omega
was also tested, of course).  Adding correlations has little effect on=
 
PK parameters, but it has some effect on simulations. 
 
NONMEM provides all eigenvalues in one pocket.  Here is an example.=

 *********************************************************************=
***************************************************
 ********************        =
            =
            =
            =
            =
            =
        
********************
 ********************        =
        STOCHASTIC APPROXIMATION
EXPECTATION-MAXIMIZATION        =
      ********************
 ********************        =
            EIGENVAL=
UES OF COR MATRIX OF
ESTIMATE (S)          =
        ********************
 ********************        =
            =
            =
            =
            =
            =
        
********************
 *********************************************************************=
***************************************************
 
             1=
         2    =
     3         =
4         5    =
     6        =
 
7         8    =
     9        10=
       11     =
  12
             13=
        14     =
   15        16  =
     17        18=
       
19        20    =
   21        22  =
      23
 
         3.36E-05  5.69E-03=
3.40E-02  6.32E-02  9.19E-02  1.24E-01 
1.53E-01  2.79E-01  3.20E-01  4.32E-01  5.74E-01  =
6.45E-01
          7.25E-01  7.67E=
-01  9.73E-01  1.08E+00  1.42E+00  1.63E+00 
1.86E+00  2.14E+00  2.31E+00  3.12E+00  4.26E+00
 
Monolix provides them in 3 pockets:
 
PK parameters: Eigenvalues (min, max, max/min): 0.22  2  9.2
OMEGA (diagonal) and SIGMA: Eigenvalues (min, max, max/min): 0.66  1.5=
 
2.2
OMEGA (correlations):  Eigenvalues (min, max, max/min): 0.097  2.=
5  25
 
Even though the results look similar, eigenvalues look different. 
Taking into account that max/min ratio is frequently reported, it is
important to understand the difference.  It almost look like different=
 
sets of parameters are estimated separately in the Monolix example,
which most likely is not the case.  Even if we combine all eigenvalues=
 
in one pocket, max/min looks good.   It is impressive that=
 max/min ratio
for OMEGA correlations may look OK even though there are small
correlations such as -0.0921, SE=0.064, RSE=70%.
 
What is the best way to report estimate and report max/min ratios?
 
Take care,
Pavel

Received on Wed Nov 04 2015 - 11:55:02 EST

The NONMEM Users Network is maintained by ICON plc. Requests to subscribe to the network should be sent to: nmusers-request_at_iconplc.com. Once subscribed, you may contribute to the discussion by emailing: nmusers_at_globomaxnm.com.