6 replies to this topic

[Anthony J Frend]

When performing NCA with sparse sampling enabled, the output gives standard errors for AUC and Cmax parameters.

 

Are the standard error values estimates of inter individual variability or of some other variability?

 

If so, what variability has been estimated?

 

Many thanks,

 

Tony

[Simon Davis]

Hi Tony,

     You can also find this information in the help files;

 

Standard error of the mean Cmax will be calculated as the sample standard deviation of the y-values at time Tmax divided by the square root of the number of observations at Tmax.

 

Standard error of the mean AUC will be calculated as described in Nedelman and Jia (1998), using a modification in Holder (2001), and will account for any correlations in the data resulting from repeated sampling of individual animals.

 

  Nedelman and Jia (1998). An extension of Satterthwaite's approximation applied to pharmacokinetics. J Biopharm Stat 8(2):317-28.

  Holder (2001). Comments on Nedelman and Jia's extension of Satterthwaite's approximation applied to pharmacokinetics. J Biopharm Stat 11(1-2):75-9.

 

  Simon.

[Helmut Schütz]

Hi Simon!

There's a little flaw in the evaluation if data points are missing. All references require complete datasets, whereas both Phoenix/WNL and the 'old' versions of WNL give the arithmetic mean of concentrations in any case. The underlying distribution in sparse sampling NCA is pretty complicated - results may be biased. I know that Thomas Jaki of Lancaster University (co-author of R-package PK) is working on this problem, but I would suggest that at least in the manual a warning should be given.

[EDIT - thanks Helmut, I've noted this under QC_PHX_9292. Simon]

[Anthony J Frend]

Hi Simon,

 

Thanks for your response.

 

I was trying to get a feel of whether one could legitimately say that the standard error was a measure of inter-individual variability.

 

From what your response it seems that one can. Please confirm if this is so.

 

On a related topic, we plan to use this functionality in a regulated study so we need to demonstrate that this function is working according to specification.

 

Do you have any reference data-sets (time vs concentration) and expected output results that we could put through the application to demonstrate that the software is calculating the various standard errors correctly.

 

We have already pre-ordered the Validation suite from Pharsight but this will not be available for a while.

 

Many thanks,

 

Tony

[Simon Davis]

Hi Tony,
To answer your first question; "I was trying to get a feel of whether one could legitimately say that the standard error was a measure of inter-individual variability.

From what your response it seems that one can. Please confirm if this is so."
I concur the answer is YES

the algorithms being used haven't changed since they were implemented in 5.x, and the Phoenix references the same papers. Therefore if you validated 5.x then you can use the data files included in that folder to do a quick review your self

the WNL Classic INPUT files, SparseSamplingChaioYeh.pmo and SparseSamplingChaioYeh.pwo can be found here.

C:\Program Files\Pharsight\WinNonlin Validation Suite\Tests\WNL 5.2\NCA_Models\Model_200\Sparse_Sampling\Sparsesamplingchaioyeh\Input

the reference OUTPUT here;
C:\Program Files\Pharsight\WinNonlin Validation Suite\Tests\WNL 5.2\NCA_Models\Model_200\Sparse_Sampling\Sparsesamplingchaioyeh\Reference

I would suggest that to manually check Phoenix 6.1 results against the validated 5.x results you could simply load up these input files, execute in Phoenix and then compare the results.

If you do find some issue I would recommend submitting it to support if you suspect a bug, just to delineate between what these boards are for versus Support

If it works fine I am sure other readers would like to hear about your anecdotal testing ;0)

Best regards,
Simon.

PS are you going to make it to the course in London next month?

-http://www.pharsight.com/training;
Introduction to Phoenix v6.1: 13-15 Apr 2010 London UK

[Anthony J Frend]

Hi Simon,

 

Many thanks for your response. I should have given you more details on our current setup.

 

Our current validated version of WinNonlin is V4.1 which (as you know) does not provide sparse sampling analysis.

 

We are currently planning to update to V6.1 and have purchased the Validation Suite for V6.1 and are awaiting it's release (Ed Dix of Pharsight can confirm this order). Therefore, I do not have access to the Validation Suite files that you refer to in your response

 

In the meantime we need to perform a sparse sampling analysis for a regulated study so need to put through a reference data-set through the sparse sampling analysis to demonstrate fitness for purpose.

 

Would it be possible for you to send me a zip file containing the two files referred to in your previous response?

 

Many thanks,

 

Tony

[Simon Davis]

Hi Tony, I have sent you directly the Input and Reference files for this test case from the 5.2 version of the Validation Suite. I trust this helps you until the Phoenix 6.1 Validation Kit is released.

               Simon.