jag009
★★★

NJ,
2015-09-15 22:29
(3116 d 20:07 ago)

Posting: # 15401
Views: 20,742
 

 Baseline correction vs non-baseline cor­rection [Regulatives / Guidelines]

Hi guys,

Maybe a stupid question.. What is the purpose of determining BE for both baseline corrected and uncorrected data for drug what exist endogenously? Isn't baseline corrected sufficient since the data is more relevant?

Thanks
John
Lucas
★    

Brazil,
2015-09-16 16:16
(3116 d 02:19 ago)

@ jag009
Posting: # 15405
Views: 19,270
 

 Baseline correction vs non-baseline correction

Hello John!

I don't think that this is a stupid question at all. I've always wondered that. When you ask for both results but says that only the corrected should be considered in BE determination, then why present both??

I'm not sure if in the US both are mandatory, but here only corrected data determine BE, and uncorrected should be presented. That's maybe only to confuse, because I've seen different conclusions for corrected and uncorrected data...

I'm also not sure about the correction based on three basal sampling points, taken very close to the drug administration. We've conducted a estradiol study recently and the basal levels for it were a nightmare, really really bizarre. We observed curves that were almost completely below the basal level, implying that the basal level is not even near constant through the Pk sampling schedule.

So if anyone can shed a light on that, I would also be pleased.

:confused:
thx

Lucas
Relaxation
★    

Germany,
2015-09-18 12:25
(3114 d 06:11 ago)

@ Lucas
Posting: # 15428
Views: 18,788
 

 Baseline correction vs non-baseline correction

Hi to everybody.

❝ I don't think that this is a stupid question at all. I've always wondered that. When you ask for both results but says that only the corrected should be considered in BE determination, then why present both??


I think a possible explanation for requiring both data sets (presented) may also be that there seems to be a lot of "errors" during evaluations.
I haven't searched for that, but thinking about the maybe last four publications/reports I saw dealing with a baseline and BE: all of them naturally claimed the use of baseline corrected data.
But in reality none of them did.

Some of them clearly stated the erronous way of calculation and presented data for corrected and uncorrected concentrations, that were not plausible, if you knew where to look at (meaning: if the quotient of means (arithmetic, because who would present geometric ones for a PK parameter ;-)) is 0.6, but the point estimate is a 0.9, well...
So there might be just some misunderstandings.

Some of them "forgot" to present all data (even down to give the point estimates/CI and some nice graphs on the maximum time scale only) and described the evaluation in a lot of words with little content.

No way to separate misconduct from unintentional failure here, but the best way to check that is by having both data sets available, so that may be a reason to ask for presentation of both?

Best regards,

Steven.
Helmut
★★★
avatar
Homepage
Vienna, Austria,
2015-09-16 17:16
(3116 d 01:19 ago)

@ jag009
Posting: # 15407
Views: 19,357
 

 Baseline correction or supra-thera­peu­tic dose

Hi John,

first of all I agree with Lucas. I don’t get the idea of basing BE on uncorrected data. The EMA’s BE-GL is less ambiguous:

4.1.4 Study conduct
Standardisation

In bioequivalence studies of endogenous substances, factors that may influence the endogenous baseline levels should be controlled if possible (e.g. strict control of dietary intake).
Sampling times
For endogenous substances, the sampling schedule should allow characterisation of the endogenous baseline profile for each subject in each period. Often, a baseline is determined from 2–3 samples taken before the drug products are administered. In other cases, sampling at regular intervals throughout 1–2 day(s) prior to administration may be necessary in order to account for fluctuations in the endogenous baseline due to cir­ca­dian rhythms (see section 4.1.5).

4.1.5 Characteristics to be investigated
Endogenous substances

If the substance being studied is endogenous, the calculation of pharmacokinetic parameters should be per­formed using baseline correction so that the calculated pharmacokinetic parameters refer to the additional concentrations provided by the treatment. Administration of supra-therapeutic doses can be considered in bioequivalence studies of endogenous drugs, provided that the dose is well tolerated, so that the additional concentrations over baseline provided by the treatment may be reliably determined.
If a separation in exposure following administration of different doses of a particular endogenous substance has not been previously established this should be demonstrated, either in a pilot study or as part of the pivotal bioequivalence study using different doses of the reference formulation, in order to ensure that the dose used for the bioequivalence comparison is sensitive to detect potential differences between for­mu­la­tions.
The exact method for baseline correction should be pre-specified and justified in the study protocol. In gen­er­al, the standard subtractive baseline correction method, meaning either subtraction of the mean of in­di­vi­du­al endogenous pre-dose concentrations or subtraction of the individual endogenous predose AUC, is pre­ferred. In rare cases where substantial increases over baseline endogenous levels are seen, baseline cor­rec­tion may not be needed.
In bioequivalence studies with endogenous substances, it cannot be directly assessed whether carryover has occurred, so extra care should be taken to ensure that the washout period is of an adequate duration.

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
Ohlbe
★★★

France,
2015-09-16 19:55
(3115 d 22:41 ago)

@ Helmut
Posting: # 15409
Views: 19,106
 

 Baseline correction or supra-thera­peu­tic dose

Dear Helmut,

❝ first of all I agree with Lucas. I don’t get the idea of basing BE on uncorrected data.


I can see one theoretical situation which could be tricky. I don't know whether you have come across anything like this.

Many endogeneous compounds follow a circadian rythm. What if the concentrations at 8 AM are higher than in the evening ? Negative concentrations after baseline correction will be replaced with 0. Couldn't this artificially decrease the difference between formulations, compared with a non-baseline-corrected analysis ? I mean, if formulation A gives concentrations 50 % lower than formulation B, both negative after baseline correction and reset to 0...

Regards
Ohlbe
Helmut
★★★
avatar
Homepage
Vienna, Austria,
2015-09-16 20:15
(3115 d 22:21 ago)

@ Ohlbe
Posting: # 15410
Views: 19,239
 

 Baseline profile prior to each period?

Dear Ohlbe,

❝ ❝ … I don’t get the idea of basing BE on uncorrected data.


❝ I can see one theoretical situation which could be tricky. I don't know whether you have come across anything like this.


Not only theoretically. ;-)

❝ Many endogeneous compounds follow a circadian rythm. What if the concentrations at 8 AM are higher than in the evening ? Negative concentrations after baseline correction will be replaced with 0. Couldn't this artificially decrease the difference between formulations, compared with a non-baseline-corrected analysis ? […]



Good point! In such a case one has to sample an entire blank profile (as the GL suggests) – not only one (or a few pre-dose) blanks in the morning. If I understand the GL correctly, subtraction of the pre-dose AUC from the post-dose AUC is suggested. I don’t get it. What about Cmax? I always prefer to subtract concentrations.

Another point: The GL suggest a blank profile preceding each period (“to account for fluctuations in the endogenous baseline due to cir­ca­dian rhythms”). This might be difficult to achieve given the limited blood volume (especially in higher order cross-overs or replicate designs). What I did in the past (and which was accepted after a scientific advice):
  • Sample an entire blank profile prior to administration in period 1 only.
  • Sample three pre-dose samples in the other periods. Keep the interval short (i.e., maximum 30 minutes pre-dose). Take the median (my preference) or the mean (GL-disciples).
  • Correct the blank profile for the ratio of pre-dose in period 1 and the value from above.
  • Subtract these corrected blank-profiles from other period’s concentrations.

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,
2015-09-17 17:44
(3115 d 00:51 ago)

@ Ohlbe
Posting: # 15423
Views: 19,245
 

 Baseline correction examples

Dear Ohlbe,

two examples of baseline-corrections.
  1. A stable baseline

    The theoretical T/R for AUC is 90%. The baseline is 25. What we measure:

    [image]

    If we work with uncorrected concentrations the AUCt-ratio with 94.87% will be biased (here the T looks better, but can be the other way ’round as well). The Cmax-ratio is also affected (92.00%). Extrapolate? Forget it. The estimated λz is only ~10% (!) of the true value… Instead of the theoretical extrapolated fraction of 0.33% we get ~45%!

    Let’t subtract the baseline:

    [image]

    As expected everything matches the theoretical values. No bias for any PK metric. Mission accomplished.

  2. A circadian rhythm in the baseline

    I followed your example (if I understood it correctly). The maximum of the base­line is at 6 a.m (50) and the minimum at 6 p.m. (20). The baseline at 8 a.m (administration) is 48 and at 8 p.m 22. The theoretical T/R for AUC and Cmax is 50%; tmax for both at 2.56.

    [image]

    The thin green line is the baseline. Thin blue/red lines are theoretical profiles of R and T. Thick lines are observed profiles.

    Let’s see what we get for the two baseline-correction methods.

    [image]

    If we subtract the pre-dose (that’s ~48 at 8 a.m.) we get the thin lines. Humbug. T/R for AUCt is 33.30% (worse than the true value). The T/R for Cmax is 44.92%. Good luck with extra­polation. Not only AUC and Cmax are affected but tmax is shifted towards earlier time points (T 1.8, R 2.1).

    If we subtract the entire baseline, we get the theoretical profiles (thick lines) and no bias in PK-metrics.

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
Lucas
★    

Brazil,
2015-09-18 16:19
(3114 d 02:17 ago)

@ Helmut
Posting: # 15430
Views: 18,877
 

 Baseline correction examples

Thanks for the graphic representation Helmut!

That does help a lot, and I agree with the method. Two questions though:
1. Do you know a example of a endogenous substance that has a constant (or nearly constant enough) baseline level in the first example of yours?
2. I see that you only considered one baseline profile. We wouldn't have to measure the baseline level before each period of the study because we ate assuming that the cycle is the same through time. Right? But what about feminine hormones produced in a 28-day cycle?

I've also seen a lot of what you've described in the last graph, our triangulation points (or embedded concentrations) that we've discussed. :-D

Regards

Lucas
Helmut
★★★
avatar
Homepage
Vienna, Austria,
2015-09-18 19:28
(3113 d 23:07 ago)

@ Lucas
Posting: # 15432
Views: 18,983
 

 Baseline correction examples

Hi Lucas,

❝ Two questions though:

❝ 1. Do you know a example of a endogenous substance that has a constant (or nearly constant enough) baseline level in the first example of yours?


I’m too lazy to browse the FDA’s product specific guidances, but you will find examples.
If you have evidence that the BL is stable over the day you may subtract average predose levels – measured in each period. This is what my example is based upon. The EMA’s GL states:

The exact method for baseline correction should be pre-specified and justified in the study protocol. In gen­er­al, the standard subtractive baseline correction method, […] subtraction of the mean of in­di­vi­du­al endogenous pre-dose concentrations […]

Which mean? Though often applied I hope not the arithmetic one. The geometric would be better but what if at least of the values is BQL? I always use the median and never got any requests.
BTW, be aware of softwares’ interpretation of non-numerics like "BQL". Example: 5.0, 7.0, "BQL"

                  Excel  PHX/WNL   R
arithmetic mean    6.0     6.0    NA
geometric mean     5.9     5.9    Error
median             6.0     6.0    "7.0"

Excel ignores the "BQL" and works with the numbers only. Same in PHX/WNL but we get the information that 2 values are used and 1 is missing. R essentially gives up.

❝ 2. I see that you only considered one baseline profile. We wouldn't have to measure the baseline level before each period of the study because we ate assuming that the cycle is the same through time. Right? But what about feminine hormones produced in a 28-day cycle?


Hey funny – coincidentally this afternoon I was working on this post and missed yours. ;-) In a 2-period study likely I would sample a complete baseline prior to each period myself. However, if there are more periods you would end up with an extreme blood volume. Agencies’ standard phrase “improve the analytical method” sometimes is not realistic.

If I switch the amplitudes of my other example (flucations within the day lower than within the 28-days cycle) I got:

method     1     2      1     2 
period       AUCt         Cmax   
  1      +3.24 -0.10  +1.59 -0.08
  2      -0.74 -2.28  -0.22 -0.55
  3      -3.20 -3.22  -1.59 -0.25
  4      -0.74 -2.28  +0.22 +0.21

No method is “perfect”. What about performing such a study in postmenopausal women?

❝ I've also seen a lot of what you've described in the last graph, our triangulation points (or embedded concentrations) that we've discussed. :-D


Don’t remind me, please. :cool:


PS. Too bad that GL’s don’t allow modeling… Should be an easy task to estimate parameters even for composite sinusoidal baselines. We need three for the circadian rhythm and four for the one with a lower frequency. Try this:

a <- c( 3.5,  1.5, 1)     # parameters for circadian rhythm
b <- c(15  , 10  , 0, 28) # parameters for 28-days cycle
t  <- -24:(27*24)
BL <- a[1]+a[2]*sin(pi/180*(t+a[3])/24*360) +
      b[1]+b[2]*sin(pi/180*(t+b[3])/24*360/b[4])
plot(t/24, BL, type="l", ylim=c(0, sum(c(a[1:2], b[1:2]))), xlab="days")
abline(v=0, lty=3)

If you have data of a study with one full baseline profile and predose concentrations in the other periods, give it a try in Phoenix!

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
Mauricio Sampaio
★    

Brazil,
2016-02-07 04:05
(2972 d 13:31 ago)

@ Helmut
Posting: # 15939
Views: 16,768
 

 Baseline correction examples

Hi Lucas and Helmut

»No method is “perfect”. What about performing such a study in postmenopausal women?

Yes! This is my doubt! Why worry with high fluctuations in the endogenous baseline due to cir­ca­dian rhythms to some hormones, as Estradiol, for example, if is possible performing the study with healthy, physiologically or surgically postmenopausal females, single-dose and blood samples colected no more than 96 hours (http://www.fda.gov/Drugs/GuidanceComplianceRegulatoryInformation/Guidances/UCM238055)? :confused:

The levels should be low and stable, because the circulating estradiol after menopause is approximately 10-20 pg / ml and the greater part of which is derived from peripheral conversion of estrone.
Lucas
★    

Brazil,
2016-02-10 12:48
(2969 d 04:48 ago)

@ Mauricio Sampaio
Posting: # 15969
Views: 16,536
 

 Baseline correction examples

Hi Mauricio!

The issue with that, at least here in Brazil, is recruiting post-menopause healthy subjects. We see a lot of failed pre study exams results like cholesterol, BMI out of range, etc. So it is not that easy to recruit these women, usually taking a lot of extra time in recruitment and screening. But it would be far more easy to conduct BE studies of female hormones with them, and at least ANVISA accepts it so far. I wouldn't know about FDA and EMA, but I think that they'd accept as well. Baseline levels of estradiol are a nightmare in healthy fertile women...

Regards

Lucas
Helmut
★★★
avatar
Homepage
Vienna, Austria,
2016-02-10 13:26
(2969 d 04:10 ago)

@ Lucas
Posting: # 15970
Views: 16,541
 

 Postmenopausal women

Hi Lucas & Mauricio!

❝ I wouldn't know about FDA and EMA, but I think that they'd accept as well.


We successfully performed a couple of studies in postmenopausal women for European submission in the past. Recruitment was tricky in the beginning but after a while we had a nice pool in our volunteer’s database. Was easy then.

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
Mauricio Sampaio
★    

Brazil,
2016-02-11 02:38
(2968 d 14:58 ago)

@ Lucas
Posting: # 15973
Views: 16,613
 

 Baseline correction examples

Hi Lucas

❝ So it is not that easy to recruit these women, usually taking a lot of extra time in recruitment and screening


I share of same situation explained by Helmut. "Recruitment was tricky in the beginning but after a while we had a nice pool in our volunteers database".

We performed two studies with Estradiol in postmenopausal women for Brazilian submission. One fail and one success! In my opnion your bad result was in part from formulation and/or in part from analytical phase (Main point! Wary!!!!!).

❝ I wouldn't know about FDA and EMA, but I think that they'd accept as well.


Look:http://www.fda.gov/Drugs/GuidanceComplianceRegulatoryInformation/Guidances/UCM238055

Regards
Dr_Dan
★★  

Germany,
2016-02-11 10:22
(2968 d 07:14 ago)

@ Lucas
Posting: # 15975
Views: 16,496
 

 Baseline correction examples

Dear all
If you regard BE as an in-vivo quality test of your formulation it would be consequent to use male subjects for studies of female hormones. By this you circumnavigate the challenge of baseline correction. I tested oral contraceptives in men and it was accepted by FDA and EMA.
I hope this Information helps.
Kind regards
Dr_Dan

Kind regards and have a nice day
Dr_Dan
Mauricio Sampaio
★    

Brazil,
2016-02-12 04:24
(2967 d 13:11 ago)

@ Dr_Dan
Posting: # 15979
Views: 16,398
 

 Postmenopausal women

Dear Dan,

❝ I tested oral contraceptives in men and it was accepted by FDA and EMA.


the assay was regulatory submission or an exploratory test?
Without any doubt of your information, but is there a public assessment document from FDA and EMA?

In ANVISA guidelines is described: If the drug is indicated for patients with specific characteristics of age and sex, the study should be fully performing in volunteers with these characteristics. I think that nobody male will talk with thin voice after a single dose of estradiol, .... but is black on white!

FDA do not the same position? See link:Product-Specific Recommendations
Dr_Dan
★★  

Germany,
2016-02-15 18:45
(2963 d 22:51 ago)

@ Mauricio Sampaio
Posting: # 15991
Views: 15,831
 

 Postmenopausal women

Dear Mauricio,

❝ the assay was regulatory submission or an exploratory test?


Pivotal studies to get MA for a generic product

❝ In ANVISA guidelines is described: If the drug is indicated for patients with specific characteristics of age and sex, the study should be fully performing in volunteers with these characteristics.


As I said if you regard BE as an in-vivo quality test of your formulation you do not need to reflect clinical use especially with regard to specific characteristics of age and sex. In former times FDA agreed with this approach but I have to admit that in the meantime they revised their product specific recommendations.
Kind regards
Dr_Dan

Kind regards and have a nice day
Dr_Dan
Helmut
★★★
avatar
Homepage
Vienna, Austria,
2015-09-18 19:16
(3113 d 23:20 ago)

@ Ohlbe
Posting: # 15431
Views: 19,067
 

 Baseline; another example

Dear Ohlbe,

another example explaining the method outlined in this post.
There might not only be circadian fluctuations in the baseline but others with a lower frequency super­imposed. The BL [12–25] consists of [10–20] within the day and [2–5] within a 28-days cycle. Let’s explore a four-period study:

[image]

Fat lines are the pre-dose profiles to be sampled prior to administrations in each period according to the EMA’s GL.

What happens if we subtract these values from the the ones on the following days? I plotted the dif­fe­rences (ignoring the if-negative-set-to-zero rule).

[image]

There will be small biases because the GL’s method assumes that baselines on dosing days are es­sen­ti­ally identical to the ones before. The largest absolute bias is observed in periods 1 and 3 because we are close to the inflection points of the 28-days cycle. Less bias in periods 2 and 4 (close to the max­i­mum and minimum).

The alternative method. Sampling only one baseline before period 1 and adjusting for the predose value in each period.

[image]

Not so nice. However, think about the doubled sampling volume in the GL’s method…

Let’s dose! Identical BA in all periods (the true Cmax 2.75 times the maximum baseline). What we would observe (periods stacked):

[image]

First the GL’s method:

[image]

Next the alternative:

[image]

Practically superimposible profiles, methods almost indistinguishible. :-D
Bias (expressed as %RE). GL = 1, alternative = 2:

method     1     2      1     2  
period       AUCt         Cmax    
  1      +0.97 +0.04  +0.48 -0.07
  2      -0.22 -0.40  -0.07 -0.35
  3      -0.96 -1.03  -0.48 -0.03
  4      +0.22 -0.60  +0.07 +0.25

I can live with it; especially given that with the alternative method one needs 37.5% less blood samples.

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
nobody
nothing

2016-02-12 10:10
(2967 d 07:26 ago)

@ Helmut
Posting: # 15980
Views: 16,270
 

 Baseline; another example

Dear Mr. Helmut!

Thanks, very nice indeed! Can we start a diploma thesis with adding some more random effects on the baseline (physiology jumps around like hell in practice, due to health status, sports, sleep, food etc.). Would love to see a large set of simulations taking into account all that!

Some 20 years ago I set up some S-plus simulations "retro-NONMEM" with true values and all kinds of variability to study the potential influence of analytical variability on BE decisions. Unfortunately never finished (end of academic time :-D) and lost the files while moving over from one computer to the next in the pre-NAS era...

Kindest regards, nobody
Helmut
★★★
avatar
Homepage
Vienna, Austria,
2016-02-12 14:37
(2967 d 02:58 ago)

@ nobody
Posting: # 15981
Views: 16,183
 

 Baseline; another example

Hi nobody,

❝ Can we start a diploma thesis…


THX, not interested.

❝ … with adding some more random effects on the baseline (physiology jumps around like hell in practice, due to health status, sports, sleep, food etc.). Would love to see a large set of simulations taking into account all that!


Me too.

❝ Some 20 years ago I set up some S-plus simulations "retro-NONMEM" with true values and all kinds of variability to study the potential influence of analytical variability on BE decisions.


Haha, welcome to the club! I had even to crazy idea to optimize the number and location of calibrators based on PK (objective function: Minimize the RE of back-calculated concentrations). Trashed.

❝ Unfortunately never finished (end of academic time :-D) and lost the files while moving over from one computer to the next in the pre-NAS era...


Do you mean the NSA? The impact of bioanalytics on the outcome BE-studies is overrated anyhow. Unless the method is (really!) lousy, analytical variability contributes little to total variability (mainly biological noise).*


  • Gaffney M. Variance Components in Comparative Bioavailability Studies. J Pharm Sci. 1992; 81(4): 315–7. doi:10.1002/jps.2600810402.

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
nobody
nothing

2016-02-12 15:08
(2967 d 02:28 ago)

@ Helmut
Posting: # 15982
Views: 16,255
 

 Baseline; another example

Hi anybody!

❝ "The impact of bioanalytics on the outcome BE-studies is overrated anyhow."


That was my gut feeling when starting, but better to have some solid numbers, or? ;-)

NAS = network attached storage... since this time I never loose any files anymore :-D

Optimizing calibration concs is more based on the weighting factors/variance distribution and principles of linear regression, not that difficult, disagreeeeee?

I wanted to use this simulation framework as a starting point to study other influences on BE and other pk parameters. Especially the nonsense terminal half-life...

Kindest regards, nobody
Helmut
★★★
avatar
Homepage
Vienna, Austria,
2016-02-12 16:47
(2967 d 00:48 ago)

@ nobody
Posting: # 15983
Views: 16,423
 

 Slowly going OT: BE study simulations

Hi nobody!

❝ ❝ "The impact of bioanalytics on the outcome BE-studies is overrated anyhow."


❝ That was my gut feeling when starting, but better to have some solid numbers, or? ;-)


Agree. See this stupid example (slides 46–49) from the early days of LC/MS-MS (no stable isotope IS and hit by the matrix effect). Slide 49 compares results of the first 12 (of 24) subjects obtained by the lousy LC/MS-MS method with stable isotope IS GC/MS. Does it matter? No.
In his paper Mr. Gaffney replicated everything (administrations, analytics) in order to estimate the contribution of variances. Bioanalytics = least important. IMHO, you could screw more easily up in the clinical phase. But that’s difficult to assess by assessors/inspectors. Hence, they concentrate on the weakest link of the chain, bioanalytics (numbers, numbers…).

❝ Optimizing calibration concs is more based on the weighting factors/variance distribution and principles of linear regression, not that difficult, disagreeeeee?


Yes and no. If the model is not established yet (linear, quadratic,…) the optimal locations would be equally spaced. Note that equally spaced calibrators are mandatory in most accredited methods in environmental analysis. Once a model is confirmed to be linear, the optimal* placement of calibrators would be ½ at the LLOQ and the other half at the ULOQ. Yes, nothing in between. Try it.
Of course the BMV-GL does not allow that. :-D

The optimal weighting factor (Fisher information…) is 1/s2. Impossible if one runs just duplicates (you may always loose one). I know two labs routinely validating methods both with 1/s2 and 1/y2. The first weighting scheme is the standard and the SOP allows to switch to the second one if they end up with a single determination within a batch. One of them faced a deficiency letter stating that “samples were not treated equally”. Under protest they recalculated the entire study with 1/y2. Difference in the BE-outcome? Third decimal. Rounded away…
The exponentially increasing spacing of calibrators likely goes back to the dark ages where weighted regression was not available (pocket calculator, Excel?) in an attempt to shift the inaccuracy/imprecision (which due to the hyperbolic CI is minimal at x|y) towards lower values. Duno. No spare time to check.

❝ I wanted to use this simulation framework as a starting point to study other influences on BE and other pk parameters.


Can imagine. Was my point as well.

❝ Especially the nonsense terminal half-life...


Ahem. What to you mean by nonsense – apart from the word “terminal”?


  • Optimal = lowest SEs of estimated slope and intercept.

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
nobody
nothing

2016-02-12 17:15
(2967 d 00:21 ago)

@ Helmut
Posting: # 15984
Views: 16,274
 

 Slowly going OT: BE study simulations

eehhmmm, more than slightly OT, yesssss, definitely

❝ Note that equally spaced calibrators are mandatory in most accredited methods in environmental analysis.


A completely different piece of cake, normally. They have not such a high calibration range (using dilution) and they (often) have a certain "point of interest", i.e. the limits given in certain guidelines.

❝ The exponentially increasing spacing of calibrators likely goes back to the dark ages where weighted regression was not available (pocket calculator, Excel?)


You can "teach" weighted linear regression to Excel. OK, not nowadays, but... ;-)

❝ Ahem. What to you mean by nonsense – apart from the word “terminal”?


I don't like this parameter, it's often used/compared in senseless ways. Clearance, volume, f is closer to the clinical reality. But doctors (and many others) can handle "times" (hours), but not "liters per hour" (but AUC divided by dose they find highly intuitive, which is 1/clearance ... :-D )

Kindest regards, nobody
Helmut
★★★
avatar
Homepage
Vienna, Austria,
2016-02-12 18:04
(2966 d 23:32 ago)

@ nobody
Posting: # 15985
Views: 16,252
 

 Religious debate

Hi nobody,

❝ ❝ Ahem. What to you mean by nonsense – apart from the word “terminal”?


❝ I don't like this parameter, it's often used/compared in senseless ways. Clearance, volume, f is closer to the clinical reality. But doctors (and many others) can handle "times" (hours), but not "liters per hour" (but AUC divided by dose they find highly intuitive, which is 1/clearance ... :-D )


On David Bourne’s PKPD-list we have the annual religious debate (aka chicken vs. egg). The thread “A question of clearance” (part 1: 2009, part 2: 2010) set the record since 1995 (!) with an amazing 152 posts. I don’t belong to any of the churches (parameterization of models in terms of rate constants or clearances).
Both are mechanistic and purely empirical. Volume(s) of distribution and clearance(s) are theoretical constructs as are rate constants. Is the body a well-stirred beaker? C’mon! This is the fundamental (but delusionary) world of both churches. I can’t agree that any of the parameters is “closer to the clinical reality”. There is one exception: Hemodialysis, where the volume of distribution chances. This is the only case where I tend to lean towards the V & CL-church. In all other cases and when it comes to NCA [sic] I’m happy with both.
What I definitely don’t like is that PK software in evaluating data of an extravascular dose spit out CL/ƒ, V/ƒ, etc. Since ƒ is unknown, this kind of “information” is meaningless. When I review manuscripts I have a polite boilerplate for that. :-D

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
nobody
nothing

2016-02-12 18:26
(2966 d 23:09 ago)

@ Helmut
Posting: # 15986
Views: 16,120
 

 Religious debate

❝ parameterization of models in terms of rate constants or clearances


For models I don't care. I care for the discussions deduced from these parameters. And often the "half-life" discussions are... ehmmm... undercomplex... :-D

❝ What I definitely don’t like is that PK software in evaluating data of an extravascular dose spit out CL/ƒ, V/ƒ, etc. Since ƒ is unknown, this kind of “information” is meaningless


Nope. Sometimes this is as close to "reality" as you can get (if you lack the info on f for example). There are other cases you simply can not get closer to the information you want. Take metabolite pk in the central compartment. In your logic it's all trash, as you will normally not know Vd. And amounts are usually what's interesting (where does the drug go? Sometimes you have urine data, rarely fecal data, and even if, for non-iv application without iv data: how did it end up in feces?), concentrations without Vd normally are of limited value (you can assume Vd metabolite to be smaller than Vd parent, but that doesn't bring you that far, huh?).

There are other cases where you can not obtain all parameters you might want. Deal with it and take as much info out of your data as you can get (on a scientific basis). Better than a lot of (NONMEM) models overparameterized ad nauseam, months of runtime, highly correlated parameters, in the end completely meaningless nonsense. But 6 or 8 compartments, most of them without any meaningful datasets to determine the related parameters...

Kindest regards, nobody
ElMaestro
★★★

Denmark,
2016-02-13 13:53
(2966 d 03:43 ago)

@ nobody
Posting: # 15987
Views: 16,038
 

 Religious debate

Hi nobody,

❝ (...) (NONMEM) models overparameterized ad nauseam, months of runtime, highly correlated parameters, in the end completely meaningless nonsense.


PK in drug development is all about impressing the assessor with complex models having a million effect estimates of dubious value. It is not about making sense. Just sayin'...

Pass or fail!
ElMaestro
nobody
nothing

2016-02-14 16:23
(2965 d 01:13 ago)

@ ElMaestro
Posting: # 15990
Views: 15,990
 

 Religious debate

Hi Danmark :-D

❝ PK in drug development is all about impressing the assessor with complex models having a million effect estimates of dubious value. It is not about making sense. Just sayin'...


I heard in the past that each and every piece of modeling (wherever submitted in the EU when any kind of non-national marketing authorization is applied for) is going for review to Sweden.

I would not build any relevant argumentation on models clearly identifiable as nonsense in this situation, huh? ;-)

Kindest regards, nobody
Ohlbe
★★★

France,
2016-02-13 14:10
(2966 d 03:26 ago)

(edited by Ohlbe on 2016-02-14 00:56)
@ Helmut
Posting: # 15988
Views: 16,069
 

 Slowly going OT: BE study simulations

Dear Helmut,

❝ Agree. See this stupid example (slides 46–49) from the early days of LC/MS-MS (no stable isotope IS and hit by the matrix effect). Slide 49 compares results of the first 12 (of 24) subjects obtained by the lousy LC/MS-MS method with stable isotope IS GC/MS. Does it matter? No.


Well, cough... In your example there is not much difference for AUC, agreed. But Cmax ? A difference of 4 points in the point estimate, closer to 1 with the lousy method, with 90 % CI of 78.6 - 99.8 % with the lousy method against 71.1 - 96.4 % with the good method... In this specific example I would say that yes, it could have mattered, and made a difference between a failing study and a passing study !

Regards
Ohlbe
Helmut
★★★
avatar
Homepage
Vienna, Austria,
2016-02-13 16:02
(2966 d 01:34 ago)

@ Ohlbe
Posting: # 15989
Views: 16,028
 

 Slowly going OT: BE study simulations

Dear Ohlbe,

❝ Well, cough... I your example there is not much difference for AUC, agreed. But Cmax ? A difference of 4 points in the point estimate, …


Actually 5.9%!

❝ … closer to 1 with the lousy method, with 90 % CI of 78.6 - 99.8 % with the lousy method against 71.1 - 96.4 % with the good method... In this specific example I would say that yes, it could have mattered, and made a difference between a failing study and a passing study !


Agreed. I guess the difference would have been smaller if we didn’t stop the LC/MS-MS method after 12 subjects (in the good ol’ days of plausibility reviews). The lousy method was validated and all batches passed. Nowadays such a method likely would not survive the ISR. ;-)

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
lizhao
☆    

US,
2016-02-25 18:31
(2953 d 23:04 ago)

@ Helmut
Posting: # 16027
Views: 15,034
 

 Slowly going OT: BE study simulations

Dear All,

I just have a naive question about how to conduct statistical test on the baseline-corrected parameters.

Should I apply the normal method: ln transformation, then ANOVA, then 90% confidence interval?

or should I un-transform the parameters and calculate fieller's confidence interval?


or should I apply non-parametric method to get the confidence interval?


Thanks a lot! I look forward to your reply.
Helmut
★★★
avatar
Homepage
Vienna, Austria,
2016-02-25 18:44
(2953 d 22:51 ago)

@ lizhao
Posting: # 16028
Views: 15,202
 

 Business as usual

Hi lizhao,

❝ I just have a naive question about how to conduct statistical test on the baseline-corrected parameters.


In NCA, PK metrics, pleeeze. ;-) Parameters are reserved for PK models.

❝ Should I apply the normal method: ln transformation, then ANOVA, then 90% confidence interval?


Yes. Don’t get confused by the other thread.

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
lizhao
☆    

US,
2016-02-25 18:59
(2953 d 22:37 ago)

@ Helmut
Posting: # 16029
Views: 15,190
 

 Business as usual

Thank you so much Helmut!!!:-D You are so helpful!I wish I could meet you in person one day.

I just have a last question....

If you do baseline-correction, sometimes the baseline-corrected metrics might be negative (I am talking about the ones that follow circadian rhythem and therefore a pre-dose AUC need to be subtracted from the post-dose AUC). If I make these negative numbers to be zero, then ln-transformation will give a infinity!

How should I deal with this kind of scenario? Thanks.
UA Flag
Activity
 Admin contact
22,957 posts in 4,819 threads, 1,638 registered users;
88 visitors (0 registered, 88 guests [including 7 identified bots]).
Forum time: 17:36 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