 # Where do I find relevant RSE(%) values?

3 replies to this topic

### #1 MojDa01

MojDa01

Member

• • Members
•  • 13 posts

Posted 18 May 2018 - 02:21 PM

Dear all,

I want to report one of my models, but I am not sure where to find the RSE(%) values that are almost always reported in POP PK papers. May be you could help me find these values?

I have some suggestions...

1. Theta estimate & RSE (%):

--> In the Theta table under the colum name "Estimate"

--> The RSE(%) is equal to the colum "CV%", right?

2. Eta estimates % RSE(%)

--> In the Omega table. Eta estimates are given as variances and can be transformed to CV%. ok!

--> But, where do I find the RSE(%) values for these eta estimates??

3. What is the difference between the correlation given in the "Omega" table and the correlations given in the "ThetaCorrelation" table?? Which values should be reported?

Best

Daniel

### #2 smouksassi1

smouksassi1

• • Members
•   • 169 posts
• LocationMontreal

Posted 22 May 2018 - 02:44 PM

Hi Daniel,

Posted 18 May 2018 - 10:21 AM

Dear all,

1. Theta estimate & RSE (%):

--> In the Theta table under the colum name "Estimate"

--> The RSE(%) is equal to the colum "CV%", right?

yes these are RSE (%) not sure why the software call it CV % but these are computed as the square root of the diagonal of the varcov matrix to get SE then you divide by parameter value.

2. Eta estimates % RSE(%)

--> In the Omega table. Eta estimates are given as variances and can be transformed to CV%. ok!

--> But, where do I find the RSE(%) values for these eta estimates??

you get two sheets Omega and OmegaStderr you should compute the RSE(%) yourself if you want to report it.

depending on your Eta model transforming a variance into a CV% might be misleading or not straightforward care should be taken.

let me ask a question what does a  50 % CV (random BSV variability ) on CL really mean clinically ?

it would be much more useful to report how much percent of your distribution would fall within say 0.8-1.2 fold of your reference clearance but that will be another post...

3. What is the difference between the correlation given in the "Omega" table and the correlations given in the "ThetaCorrelation" table?? Which values should be reported?

ThetaCorrelation is the one from variance covariance and it is just another way to output the same information as in ThetaCovariance this will give you how much these parameters are correlated in their uncertainties. Think this can be useful to you if you want to simulate drom the variance covaraince distribution of CL / V  ( population values)

now  individual clearnace might also depend on eta

CLi = CL *exp(etaCL)

Vi = V *exp(etaV)

so once you selected a set of correlated CL/V now you want to simulate from the omega matrix of omega CL/V nad possibility there is a correlation so what the Omega sheet give you is the correlations between the variances of omega within the omega matrix. These do not give you how much the population CL/V are correlated in their estimation uncertainty.

These are concepts that would require some course work and practice and a post will not be sufficient to clarify everything.

bests,

Samer

### #3 MojDa01

MojDa01

Member

• • Members
•  • 13 posts

Posted 23 May 2018 - 05:26 AM

Dear Samer,

thanks a lot for your response!

one more thing where I am not sure how to calculate...

you said

"yes these are RSE (%) not sure why the software call it CV % but these are computed as the square root of the diagonal of the varcov matrix to get SE then you divide by parameter value."

In one of my models I have a Vc estimate of 4442 mL, a Stderr of 204 and a CV% of 4.6%. The eta of Vc is 0.0599. the eta square root is 0.244. The CV (0.244/4442) then would be 0.0055% and not 4.6%. Does Phoenix calculate with Vc in L instead of mL??

Big thanks,

Daniel

### #4 Simon Davis

Simon Davis

• • • 1,119 posts

Posted 15 July 2019 - 03:10 PM

This came up again recently so to summarise;

CV in PHX output is the same as RSE in NONMEM

By definition

CV = relative standard deviation = SD/mean

RSE = relative standard error = SE/mean

Since "standard error" is literally "the standard deviation of the distribution of some estimate", for theta context they are interchangeable,

(SD and SE on raw data are calculated differently but for parameters they are used interchangeably)

SImon

PS and sorry Daniel I missed you last question but hopefully you worked it out!  The units of outputs in a Phoenix model are driven by the input units, there is no unit conversion within the Phoenix model engine.  The best advice I can give is to make sure you standardise units of dose to match units of measurement, then it's clear what units volume etc are.

#### 0 user(s) are reading this topic

0 members, 0 guests, 0 anonymous users 