jag009
★★★

NJ,
2014-09-18 23:32
(3479 d 01:47 ago)

Posting: # 13535
Views: 11,535
 

 Reporting of plasma concentration data etc [Bioanalytics]

Hi everyone,

I have 2 questions...

1) How should concentration data be reported in terms of the # of signficant digits or decimal places? Which is the proper way?

I received data from the a lab that reports the values in 3 significant figures only and do not take into account the decimal places. Example, if the value is 12.333, they present it as 12.3. If the value is 0.33333, they will report it as 0.333. If the value is 100.203, they will report it as 100.

Is their approach correct or incorrect? I mean what if the 90% ratio is 86.55 - 125.05 and the upper end would have passed if the lab uses 3 decimal places instead of 3 significant figures for data presentation? I think they should report the values in 3 decimal places.

2) Lets say you analyze the data with 1) SAS, 2) Winnolin. The SAS result shows a 90% CI of 85.22 - 125.1 and the Winnonlin result shows 85.15 - 124.98???? :confused::confused: I hate this decimal games...

Thanks
John
Ohlbe
★★★

France,
2014-09-18 23:50
(3479 d 01:30 ago)

@ jag009
Posting: # 13536
Views: 10,625
 

 Reporting of plasma concentration data etc

Hi John,

I'll let the stats gurus answer the second point and will focus on the first one.

❝ 1) How should concentration data be reported in terms of the # of signficant digits or decimal places? Which is the proper way?


Scientifically, having a fixed number of significant digit makes sense. I was taught at university that you can't report more significant digits than the figures that you used for your calculations. It makes no sense to report 100.203 if you took a sample of 10.1 mg of reference substance to prepare the stock solutions. Not even to mention your 5, 10 or 15 % CV and accuracy.

If you have a fixed number of decimals: considering the wide range of the calibration curve, you may get 3 significant digits at the LLOQ and 5 or 6 at the ULOQ. At the highest concentration levels your decimals will be nothing more than random numbers.

❝ I mean what if the 90% ratio is 86.55 - 125.05 and the upper end would have passed if the lab uses 3 decimal places instead of 3 significant figures for data presentation? I think they should report the values in 3 decimal places.


I would report the concentrations with 3 significant digits (or more probably 4, actually) but I would calculate AUC with non-rounded concentrations. You can put a note in the report explaining that the concentrations in the tables are rounded but that PK parameters were calculated with full precision to avoid questions from reviewers.

Regards
Ohlbe
jag009
★★★

NJ,
2014-09-19 00:22
(3479 d 00:57 ago)

@ Ohlbe
Posting: # 13537
Views: 10,629
 

 Reporting of plasma concentration data etc

Hi Ohlbe,

Okay.. So what if the dose given is 20mg? I mean we don't know the decimals right? The reason I ask is what if the concentration data shows high variability and you see values such as 0.0543, 1.1253, 20.3521? So you would go with 0.0543, 1.13, 20.4? But yet with the 90% CI rule the goal post is 80.00 - 125.00 and 2 decimal places needed to be reported (?)

❝ I would report the concentrations with 3 significant digits (or more probably 4, actually) but I would calculate AUC with non-rounded concentrations. You can put a note in the report explaining that the concentrations in the tables are rounded but that PK parameters were calculated with full precision to avoid questions from reviewers.


What would you do with Cmax if the individual values range from 10.xxx to 100.xxx?

Thanks
John
Ohlbe
★★★

France,
2014-09-19 14:26
(3478 d 10:54 ago)

@ jag009
Posting: # 13540
Views: 10,651
 

 Reporting of plasma concentration data etc

Hi John,

❝ Okay.. So what if the dose given is 20mg? I mean we don't know the decimals right?


True, but the 20 mg is not used for calculations of PK parameters or bioequivalence.

❝ The reason I ask is what if the concentration data shows high variability and you see values such as 0.0543, 1.1253, 20.3521? So you would go with 0.0543, 1.13, 20.4?


Yes. Or actually go for 4 digits if possible.

❝ But yet with the 90% CI rule the goal post is 80.00 - 125.00 and 2 decimal places needed to be reported (?)


Yes... Which is indeed in contradiction with what I wrote previously (5 digits for 125.00, more than the 4 I use for concentrations). In any case it is a bit artificial here: the rounding of 90 % CI to two decimals is just for reporting, but Helmut would say that the exact value should be used to conclude on bioequivalence...

I usually use full precision for any calculation, and only round at the end. If you round each value before the calculations, inaccuracies will add on.

❝ What would you do with Cmax if the individual values range from 10.xxx to 100.xxx?


I would report 10.xx to 100.x...

Regards
Ohlbe
d_labes
★★★

Berlin, Germany,
2014-09-19 10:46
(3478 d 14:33 ago)

@ Ohlbe
Posting: # 13538
Views: 10,538
 

 Reporting of plasma concentration data etc

Dear Ohlbe,

❝ I would report the concentrations with 3 significant digits (or more probably 4, actually) but I would calculate AUC with non-rounded concentrations.


I wouldn't do that that way. Imagine you have only the reported values with 3 or 4 significant digits and are going to recalculate the AUC's. You will surely end with some differences which nitpicking QC personal will surely throws you into your face.

See also this thread back into 2008.

Regards,

Detlew
nobody
nothing

2014-09-19 12:31
(3478 d 12:49 ago)

@ d_labes
Posting: # 13539
Views: 10,513
 

 Reporting of plasma concentration data etc

As pointed out: You can report/calculate with 17 sig digits, however, most of it is random noise.

Face it: if a postive BE decision depends on stuff like that, your product is, eeehm,... imperfect ***cough***

(...it might even be possible to pick a single analytical run, "adjust" the calibration curve and make the products "bioequivalent"...)

Kindest regards, nobody
jag009
★★★

NJ,
2014-09-19 17:26
(3478 d 07:53 ago)

@ nobody
Posting: # 13541
Views: 10,591
 

 Reporting of plasma concentration data etc

❝ As pointed out: You can report/calculate with 17 sig digits, however, most of it is random noise.


❝ Face it: if a postive BE decision depends on stuff like that, your product is, eeehm,... imperfect ***cough***


True but we live in a world where these bloody #s mean pass or fail and you will not be surprised how many "successful" BE studies are borderline passing with low end of 90%CI like 80.1, 80.05 or high end like 124.9, 124.8...

I just don't like the inconsistency of the labs. Some go with 3 decimal places, some go with 3 significant figures.

John
nobody
nothing

2014-09-19 18:01
(3478 d 07:19 ago)

@ jag009
Posting: # 13542
Views: 10,547
 

 Reporting of plasma concentration data etc

❝ I just don't like the inconsistency of the labs. Some go with 3 decimal places, some go with 3 significant figures.


Isn't there anything in the guidances? ;-)

Start a controlled correspondence with the FDA?

The "4 digits" approach is reasonable, ask your lab in the analytical study plan to provide the data in the format you like, or are you the poor one who get's the data after all is finished? :-(

PS: Found it...

"A presentation of numeric data that is consistent with instrumental capabilities and acceptance criteria. The method should indicate what format to use to report results (e.g., percentage label claim, weight/weight, and weight/volume etc.) with the specific number of significant figures needed.

The American Society for Testing and Materials (ASTM)E29 describes a standard
practice for using significant digits in test data to determine conformance with specifications."

http://www.fda.gov/downloads/Drugs/GuidanceComplianceRegulatoryInformation/Guidances/UCM386366.pdf

In the bioanalytical method validation guidance there is nothing on significant digits...

Kindest regards, nobody
jag009
★★★

NJ,
2014-09-26 18:29
(3471 d 06:51 ago)

@ nobody
Posting: # 13597
Views: 10,283
 

 Reporting of plasma concentration data etc

Hmm...

❝ The "4 digits" approach is reasonable, ask your lab in the analytical study plan to provide the data in the format you like, or are you the poor one who get's the data after all is finished? :-(


Yes, I source the study out to the CRO as a sponsor (fortunately).
Then they bring up the stupid SOP thingy "Oh our SOP states 3 significant figures" Oh well.


John
SDavis
★★  
Homepage
UK,
2014-09-25 17:10
(3472 d 08:09 ago)

@ jag009
Posting: # 13585
Views: 10,473
 

 Reporting of plasma concentration data etc

Hi sorry to be late to the conversation but I was intrigued by this part of your question,

❝ 2) Lets say you analyze the data with 1) SAS, 2) Winnolin. The SAS result shows a 90% CI of 85.22 - 125.1 and the Winnonlin result shows 85.15 - 124.98????


Are you sure the differences are due to decimal places - I would think it more likely there is some difference in your model settings to explain this - (although of course it's a great advert if WNL can somehow make borderline products BE ;0)

In all seriousness, if possible, I'd love to see your Phoenix project and corresponding SAS code to see where this difference may have arisen since the results should be identical,

Simon

PS On the more general subject topic I've always used the maximum precision supplied (sure if full precision concs are supplied, some of it maybe 'random' numbers in the decimals but that too should 'round out') and rounded only at the end to minimise compounding of rounding errors.

Senior Scientific Consultant
Pharsight- A Certara™ Company
Get tips and discuss Pharsight products with other users; www.pharsight.com/extranet
___________________________________
https://www.certara.com/training
Intro (Tue-Wed) & Intermediate (Thur-Fri) NLME with Phoenix 18-21 Nov 2014, Paris, France
Intro to Phoenix WinNonlin (Mon-Tue) & Intro to NLME (Wed-Thur) NLME with Phoenix 1-4Dec 2014, Barcelona, Spain
Helmut
★★★
avatar
Homepage
Vienna, Austria,
2014-09-25 18:11
(3472 d 07:08 ago)

@ SDavis
Posting: # 13587
Views: 10,377
 

 Reporting of plasma concentration data etc

Hi Simon,

I never have seen any difference between SAS and PHX/WNL – if the ‘right’ coding was used. One exception are parallel designs, where PHX/WNL assumes equal variances (Linda posted a workaround on the Extranet; see also this thread). Let’s see what v6.4 will do (any news about the release date?)… ;-)

Agencies want to have the CI in % rounded to two decimals. PHX/WNL’s core-output (which many people copy/paste to their reports) might state “Failed to show bioequivalence…” – even if the rounded CI would pass – because the software compares the CLs to the acceptance range in full precision. Might be confusing. I prefer to set up a series of transformations which also the assess­ment based on rounded results and construct a table.

❝ PS On the more general subject topic I've always used the maximum precision supplied (sure if full precision concs are supplied, some of it maybe 'random' numbers in the decimals but that too should 'round out') and rounded only at the end to minimise compounding of rounding errors.


True, that “noise” means out. But: Sometimes regulators don’t have access to the raw data and recalculate a study based on what is given in the report (limited precision). In the worst case results differ – questions arising, :blahblah:

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
jag009
★★★

NJ,
2014-09-26 18:22
(3471 d 06:58 ago)

@ SDavis
Posting: # 13596
Views: 10,308
 

 Reporting of plasma concentration data etc

Hi Simon,

❝ Hi sorry to be late to the conversation but I was intrigued by this part of your question...


The dataset was from a 3-way study (2Ts vs R). Analysis was carried out to compare A vs C, B vs C.

I used SAS and Phoenix to do the above comparisons. SAS was carried out in Proc GLM.

The only caveat I could think of is the dataset (n=56) because the concentrations were reported in 3 significant figures and have values ranging from 0.xxx to x0.x. Note that for this investigation I only focused in Cmax and not AUC (so to eliminate any differences in computation).

Thanks
John
Ohlbe
★★★

France,
2014-09-28 15:53
(3469 d 09:27 ago)

@ jag009
Posting: # 13604
Views: 10,231
 

 Reporting of plasma concentration data etc

Hi John,

❝ I used SAS and Phoenix to do the above comparisons. SAS was carried out in Proc GLM.


❝ The only caveat I could think of is the dataset (n=56) because the concentrations were reported in 3 significant figures and have values ranging from 0.xxx to x0.x. Note that for this investigation I only focused in Cmax and not AUC (so to eliminate any differences in computation).


That's still a hell of a difference, and I certainly would not expect it to be that large with n=56. But in any case if the same dataset with the exact same figures was used in SAS and Phoenix, I fail to see how the number of significant figures or decimals could change anything in the 90 % CI ?

Regards
Ohlbe
ElMaestro
★★★

Denmark,
2014-09-28 17:07
(3469 d 08:13 ago)

@ SDavis
Posting: # 13605
Views: 10,312
 

 Reporting of plasma concentration data etc

Hi both,

❝ ❝ 2) Lets say you analyze the data with 1) SAS, 2) Winnolin. The SAS result shows a 90% CI of 85.22 - 125.1 and the Winnonlin result shows 85.15 - 124.98????


❝ Are you sure the differences are due to decimal places - I would think it more likely there is some difference in your model settings to explain this - (although of course it's a great advert if WNL can somehow make borderline products BE ;0)


This could be the case if there is one or more subjects having some period data missing; didn't I read somewhere that WNL will do a max likelihood optimisation by default (=essentially a mixed model) whereas SAS will try Proc GLM (normal linear model).
The mixed model versus normal linear model would explain the subtle difference which then isn't a matter of decimals carried.

Pass or fail!
ElMaestro
Helmut
★★★
avatar
Homepage
Vienna, Austria,
2014-09-28 19:21
(3469 d 05:59 ago)

@ ElMaestro
Posting: # 13606
Views: 10,397
 

 SAS vs. PHX/WNL

Hi to all,

❝ This could be the case if there is one or more subjects having some period data missing; didn't I read somewhere that WNL will do a max likelihood optimisation by default (=essentially a mixed model) whereas SAS will try Proc GLM (normal linear model).

❝ The mixed model versus normal linear model would explain the subtle difference which then isn't a matter of decimals carried.


Correct for WinNonlin. SAS doesn’t try “Proc GLM” – it will use it if stated in the code (actually what John did). If we have incomplete data (period missing) we should get identical results if all data are kept:
  • SAS Proc mixed = WinNonlin’s default random effects model evaluated by REML.
  • SAS Proc GLM = WinNonlin all fixed effects model: random effect subject(sequence) deleted and added to the fixed effects.
If incompleted cases are excluded (as sugested by the EMA for nonreplicated studies), any software / model should give the same results anyway. See also there.

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
Helmut
★★★
avatar
Homepage
Vienna, Austria,
2014-09-28 21:34
(3469 d 03:46 ago)

@ jag009
Posting: # 13607
Views: 10,239
 

 Numbers should reflect the precision of the instrumentation

Hi everyone,

just discovered a goodie in the “Instructions for Authors” of The AAPS Journal and AAPS PharmSciTech:

Numbers should be reported to reflect the precision of the instrumentation utilized. Calculated numbers, such as means and standard deviations, should be expressed to no more than 1 sig­ni­fi­cant digit beyond the precision of the instrument. Normally, data reported to more than 3 significant figures should be justified. The precision of the variability (e.g., standard deviation) should not exceed that of the reported mean value.


I like that very much!

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
UA Flag
Activity
 Admin contact
22,957 posts in 4,819 threads, 1,636 registered users;
72 visitors (0 registered, 72 guests [including 7 identified bots]).
Forum time: 00:20 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