Helmut
★★★
avatar
Homepage
Vienna, Austria,
2016-01-28 21:03
(2981 d 15:56 ago)

Posting: # 15886
Views: 7,862
 

 Software validation [NCA / SHAM]

Dear all,

did you validate the software you are using to get PK metrics (by NCA)?

Recently I faced a question from a European agency (don’t want to be more specific). In 2000 (!) I cross-validated the software against the one I had written myself in the early eighties. The agency asked for cross-validation against “commercial software like WinNonlin”. First I thought that this would be an easy task, but it turned out to need detective-work. I’m only aware of the data sets by Sauter et al.1, which I used already in cross-validating software for 2×2×2 crossover studies. The single dose data set contains the intervals used for estimation of λz, the observed and estimated Cz (AUC0–∞ was estimated based on the latter), :blahblah:
Clicked some buttons, got exactly the same eliminations and – not even the AUC0–tz matched in most cases… What?! I forgot that Sauter et al. used a method which was proposed by Schulz and Steinijans in 1991.2 This method went down the drain and therefore, it is not implemented in any software I know. I figured out to set it up in PHX/WNL, but it was not that easy. Let’s have a look at an example (Subject 8, test):

 t    C  
 0  <0.06
 1   0.19
 2   1.27
 3   2.75
 4   3.23
 5   3.36
 6   3.29
 7   3.89
 8   3.25
10   2.73
12   2.12
14   1.53
16   1.21
18   0.95
20   0.68
22   0.54
24   0.42
28   0.19
36   0.15
40   0.10
44  <0.06
48  <0.06
60  <0.06
72  <0.06

λz was estimated in the interval 10–28 hours. Don’t ask me why. Reference data set = take what you get. Looks like this:

[image]


Let’s zoom into the last/lower range. The two methods which are implemented in all versions of WinNonlin:
  1. Use all values (irrespective whether they are used in the estimation of λz or not), extrapolation based on the observed last concentration (0.1 mg/mL).

    [image]

    AUC0–tz 45.875 h×mg/L, AUC0–∞ 46.574 h×mg/L.

  2. Like 1. but extrapolation based on the estimated last concentration (0.04 mg/L).

    [image]

    AUC0–tz 45.875 h×mg/L, AUC0–∞ 46.147 h×mg/L.
Sauter et al. reported AUC0–tz 44.07 h×mg/L and AUC0–∞ 45.58 h×mg/L. Why does not even the AUC0–tz match? Simsalabim:

[image]


Starting with 28 hours all values are dropped from NCA. At 28 hours the measured concentration (2.78 mg/L) is replaced by its estimate (2.83 mg/L), which is also the last value for the trapezoidal rule.

Quoting the paper: “Subtle problems such as […] the AUC extrapolation to infinity in single-dose studies have been dealt with. As pointed out in the previous paper [Schulz and Steinijans 1991] there is this occasional ambiguity in estimating the terminal half-life and its effect on the extrapolated AUC. As, presumably, there is no definite answer to this problem, it is important to document all steps of analysis as it has been done in Tables 4a and 4b of this paper. This will facilitate independent verifications of results by third party such as quality assurance, licensees, reviewers of journals or health authorities.”

Lessons learned: Even if reference data are published, surprises may happen. ;-)


  1. Sauter R, Steinijans VW, Diletti E, Böhm E, Schulz H-U. Presentation of results from bioequivalence studies. Int J Clin Pharm Ther Toxicol 1992;30(Suppl.1):S7–30.
  2. Schulz H-U, Steinijans, VW. Striving for standards in bioequivalence assessment: a review. Int J Clin Pharm Ther Toxicol 1991;29(8):293–8.

Dif-tor heh smusma 🖖🏼 Довге життя Україна! [image]
Helmut Schütz
[image]

The quality of responses received is directly proportional to the quality of the question asked. 🚮
Science Quotes
ElMaestro
★★★

Denmark,
2016-01-28 23:33
(2981 d 13:26 ago)

@ Helmut
Posting: # 15887
Views: 6,520
 

 Software validation

Hi Hötzi,

someone should push the mental Ctrl+Alt+Del buttons and pull the plug on Sauter et al. and instead publish some independent reference datasets... I have recollection of three weirdos doing something like that recently in relation to some other bioequivalence topic that was difficult to handle...

Pass or fail!
ElMaestro
nobody
nothing

2016-01-29 09:16
(2981 d 03:43 ago)

@ ElMaestro
Posting: # 15889
Views: 6,493
 

 Software validation

...you mean the first part of the Schutz-Labes-Birdsong trilogy?

http://link.springer.com/article/10.1208/s12248-014-9661-0

Kindest regards, nobody
Shuanghe
★★  

Spain,
2016-01-29 12:08
(2981 d 00:51 ago)

@ nobody
Posting: # 15890
Views: 6,492
 

 Software validation

❝ ...you mean the first part of the Schutz-Labes-Birdsong trilogy?


http://link.springer.com/article/10.1208/s12248-014-9661-0


Didn't know it's a trilogy! Then this one must be "The Two Towers".

So when can we expect "The Return of the King" (SCNR :-D)? Don't keep us waiting.

All the best,
Shuanghe
Helmut
★★★
avatar
Homepage
Vienna, Austria,
2016-01-29 14:31
(2980 d 22:28 ago)

@ Shuanghe
Posting: # 15892
Views: 6,515
 

 Trilogy?

Hi Shuanghe,

❝ Didn't know it's a trilogy!


So do I. I don’t have the slightest intention to dive into validating NCA-algos. I think everybody should describe unambiguously in the protocol how it will be done (example). In this respect I concur with Sauter’s quote. In the majority (!) of reports I have seen the interval used for estimating λz is not given. If one wants to recalculate results it means a lot of trial and error. Especially how BQLs are treated deserves special attention. How are partial AUCs (or AUC0–72) calculated if a sample is not taken exactly at the cut-off time? Does the software interpolate (and if yes, how)? t75% (“plateau time” or “peak occupancy time 75%”; mandatory for MR in Russia) can be tricky. In the EU the peak-trough-fluctuation in BE should be calculated with Css,min at τ, whereas in BA (new drugs) Css,min within τ should be used. Only the latter is the standard in software.

BTW, if (if!) the algo is described in the protocol, recalculation is possible with a pocket calculator or even paper-pencil-brain. I don’t understand the agitation.

❝ So when can we expect "The Return of the King" (SCNR :-D)? Don't keep us waiting.


Maybe it is like Patrick Rothfuss’ The Kingkiller Chronicle. Fans of fantasy fiction are waiting since 2011 for the third book.

Dif-tor heh smusma 🖖🏼 Довге життя Україна! [image]
Helmut Schütz
[image]

The quality of responses received is directly proportional to the quality of the question asked. 🚮
Science Quotes
Relaxation
★    

Germany,
2016-01-29 19:04
(2980 d 17:55 ago)

@ Helmut
Posting: # 15893
Views: 6,330
 

 Trilogy?

❝ In the majority (!) of reports I have seen the interval used for estimating λz is not given.


Well, some big players in the industry have these always; I would assume the reason actually is the head of pharmacology there.
On the other hand, I haven't seen these in the majority of CRO reports, too. I would guess, the main reasons might be "These are of no use / Nobody does it" (I like that one!) or, in case of those people that have some knowledge in NCA, "Why should we make life easier for other people, if we do not have to" :-D.
Arguments work also, when you try to present geometric means or ask for overlays in a publication :no:.
Helmut
★★★
avatar
Homepage
Vienna, Austria,
2016-01-29 19:51
(2980 d 17:08 ago)

@ Relaxation
Posting: # 15895
Views: 6,374
 

 Trilogy?

Hi Relaxation,

❝ Well, some big players in the industry have these always; I would assume the reason actually is the head of pharmacology there.


Not always. :angry:

❝ On the other hand, I haven't seen these in the majority of CRO reports, too. I would guess, the main reasons might be "These are of no use / Nobody does it" (I like that one!) or, in case of those people that have some knowledge in NCA, "Why should we make life easier for other people, if we do not have to" :-D.


Did you see these examples: #7226, #10005? Can you guess the CRO when I tell you the number of employees (~19,000)? The first example matched exactly your second “reason”. It was a large study and it took me ages to figure out how these blockheads estimated the elimination.

❝ Arguments work also, when you try to present geometric means or ask for overlays in a publication :no:.


All too true. Did you see the bizarre plot of arithmetic means linked in this post? Was an innovator, BTW. Personally I classify publications without spaghetti plots as poo.

Dif-tor heh smusma 🖖🏼 Довге життя Україна! [image]
Helmut Schütz
[image]

The quality of responses received is directly proportional to the quality of the question asked. 🚮
Science Quotes
ElMaestro
★★★

Denmark,
2016-01-29 12:20
(2981 d 00:40 ago)

@ nobody
Posting: # 15891
Views: 6,447
 

 Software validation

Hi Nobody,

❝ ...you mean the first part of the Schutz-Labes-Birdsong trilogy?


Trilogy? Me likey! :-D:-D:-D

Pass or fail!
ElMaestro
jag009
★★★

NJ,
2016-01-29 23:46
(2980 d 13:13 ago)

@ Helmut
Posting: # 15897
Views: 6,333
 

 Software validation

Hi Helmut,

Remember I told you about one lab which would consider Kel as "could not be evaluated" if the last timepoint(s) jumps back up relative to the rest of the points before them in the elimination phase?

John
UA Flag
Activity
 Admin contact
22,957 posts in 4,819 threads, 1,639 registered users;
73 visitors (0 registered, 73 guests [including 7 identified bots]).
Forum time: 13:00 CET (Europe/Vienna)

Nothing shows a lack of mathematical education more
than an overly precise calculation.    Carl Friedrich Gauß

The Bioequivalence and Bioavailability Forum is hosted by
BEBAC Ing. Helmut Schütz
HTML5