Loky do
★    

Egypt,
2021-06-16 02:49
(1016 d 14:45 ago)

Posting: # 22414
Views: 4,303
 

 Outlier in fully replicate BE study [Outliers]

Dear colleagues,

During bioanalysis of a fully replicate BE study for drug known to has a potential for back conversion (clopidogrel), one volunteer shows drug concentrations below LLOQ in 3 phases and only one phase (Reference product) shows very low drug concentration, no probability of not swallowing the tablets (mouth check was assured) what could be the reason of this results :confused:? Knowing that the ISR results for the study is accepted?
Thanks in advance
drgunasakaran1
★★  
avatar

2021-06-16 10:59
(1016 d 06:35 ago)

@ Loky do
Posting: # 22416
Views: 3,860
 

 Outlier in fully replicate BE study

Dear Mr Loky do,

❝ During bioanalysis of a fully replicate BE study for drug known to has a potential for back conversion (clopidogrel) ,one volunteer shows drug concentrations below LLOQ in 3 phases and only one phase (Reference product) shows very low drug concentration, no probability of not swallowing the tablets (mouth check was assured) what could be the reason of this results :confused:? Knowing that the ISR results for the study is accepted?


As per regulatory guidance, you can exclude this subject if the following criteria is fulfilled;
A subject with lack of any measurable concentrations or only very low plasma concentrations for reference medicinal product. A subject is considered to have very low plasma concentrations if its AUC is less than 5% of reference medicinal product geometric mean AUC (which should be calculated without inclusion of data from the outlying subject).
The exclusion of data due to this reason will only be accepted in exceptional cases and may question the validity of the trial.


Reference: EMA's Guideline on the Investigation of Bioequivalence

Also, be informed that Genetic Polymorphisms of Cytochrome P450 enzymes (CYP2C19, CYP3A, CYP2B6 and CYP1A2) and Esterases will affect the Pharmacokinetic profile of clopidogrel in some subjects.

Dr S Gunasakaran MBBS MD
Disclaimer: The replies/posts are my personal opinions and it does not represent my company views on the same.
Loky do
★    

Egypt,
2021-06-16 17:38
(1015 d 23:57 ago)

@ drgunasakaran1
Posting: # 22418
Views: 3,782
 

 Widening for clopidogrel BE study

many Thanks, drgunasakaran1 for your reply

Dears
I also have a question regarding Clopidogrel BE study design, as per ema questions and answers, is widening of Cmax accepted in clopidogrel BE studies with fully replicate design or not?

thanks in advance


Edit: link restored. [mittyri]
mittyri
★★  

Russia,
2021-06-16 18:04
(1015 d 23:30 ago)

@ Loky do
Posting: # 22419
Views: 3,825
 

 Widening for clopidogrel BE study

Dear Loky do

❝ is widening of Cmax accepted in clopidogrel BE studies with fully replicate design or not?


You linked the EMA Q&A where it is stated unequivocally:
Under these circumstances, the widening of 90% confidence intervals for Cmax is not recommended.

Kind regards,
Mittyri
Loky do
★    

Egypt,
2021-06-16 19:08
(1015 d 22:27 ago)

@ mittyri
Posting: # 22420
Views: 3,834
 

 Widening for clopidogrel BE study

❝ You linked the EMA Q&A where it is stated unequivocally:

Under these circumstances, the widening of 90% confidence intervals for Cmax is not recommended.


Ok, in my case, the study fails as per EMA guidance, but if we used the FDA method for calculations it passes, regarding a drug as clopidogrel with many therapeutic issues, could it be accepted by different authorities, considering FDA guidelines (sponsor requests to submit an appeal to authorities to use calculations as per FDA guidelines as the protocol uses EMA guidance for statistical calculations)?
thanks
mittyri
★★  

Russia,
2021-06-16 19:27
(1015 d 22:07 ago)

@ Loky do
Posting: # 22421
Views: 3,754
 

 Widening for clopidogrel BE study - cherry-picking?

Dear Loky do,

❝ Ok, in my case, the study fails as per EMA guidance, but if we used the FDA method for calculations it passes, regarding a drug as clopidogrel with many therapeutic issues, could it be accepted by different authorities, considering FDA guidelines (sponsor requests to submit an appeal to authorities to use calculations as per FDA guidelines as the protocol uses EMA guidance for statistical calculations)?


The product-specific guidance is quite old and was issued before RSABE first appeared as an option.
BTW that smells like cherry-picking, sorry
Most probably it will be rejected taking into account audit trail of the Protocol changes/dates of analysis.

Kind regards,
Mittyri
drgunasakaran1
★★  
avatar

2021-06-17 14:01
(1015 d 03:33 ago)

@ Loky do
Posting: # 22422
Views: 3,690
 

 Widening for clopidogrel BE study

Dear Mr Loky Do,

❝ Ok, in my case, the study fails as per EMA guidance, but if we used the FDA method for calculations it passes, regarding a drug as clopidogrel with many therapeutic issues, could it be accepted by different authorities, considering FDA guidelines (sponsor requests to submit an appeal to authorities to use calculations as per FDA guidelines as the protocol uses EMA guidance for statistical calculations)?


Most of the times, the regulatory agency may not accept the change in the statistical plan after the Statistical Evaluation for the reason since it fails as per EMA guidance.

Dr S Gunasakaran MBBS MD
Disclaimer: The replies/posts are my personal opinions and it does not represent my company views on the same.
Helmut
★★★
avatar
Homepage
Vienna, Austria,
2021-06-19 14:32
(1013 d 03:02 ago)

@ Loky do
Posting: # 22423
Views: 3,677
 

 Hypotheses

Hi Loky do,

in addition to what Mittyri and Dr Gunasakaran wrote, a general remark about confirmatory studies:
  1. You state a null hypothesis H0 and an alternative hypothesis H1.
    In bioequivalence H0 is inequivalence – which you desire to reject.
  2. You state an appropriate statistical method, most commonly the confidence inclusion approach.
  3. You perform the study and assess #1 by #2.
    The outcome is dichotomous, i.e., either the study passed (H0 rejected) or it failed (H0 not rejected).
What you must not do: The study failed according to the planned conditions and then you change #1 and/or #2 in order to make it pass. That’s the cherry-picking Mittyri was talking about.
In simple terms: The entire \({\small{\alpha=0.05}}\) was already ‘spent’ in the original analysis. Hence, any ‘alternative’ evaluation will increase the patient’s risk, which is not acceptable.

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
Loky do
★    

Egypt,
2021-07-06 13:51
(996 d 03:43 ago)

@ Helmut
Posting: # 22458
Views: 3,397
 

 Hypotheses

Thanks, Helmut for your reply

but I am quite confused, some products show high intrasubject variability more than anticipated or published in the literature and I think this variability affect the reliability of results, for example, I have a BE study for (lansoprazole), partially replicate design, the published intrasubject variability ~ 40% but the practical intrasubject variability we had was near 85% :confused: study protocol also stated using ema guidelines for scaling but the study failed, when I used FDA method it passes :confused: does this high variability have a role in this confusing results? also in the future protocol can I specify that if the intrasubject variability is more than 50% can I switch to the FDA method for calculation, as many authorities we submit our studies to follow ema guidelines?

Thanks in advance
Helmut
★★★
avatar
Homepage
Vienna, Austria,
2021-07-06 18:17
(995 d 23:17 ago)

@ Loky do
Posting: # 22460
Views: 3,411
 

 ABEL ≠ RSABE

Hi Loky do,

❝ but I am quite confused, some products show high intrasubject variability more than anticipated or published in the literature …


If you design studies for 80% power and all of your assumptions are exactly realized (T/R-ratio, CV, dropout-rate), one out of five will fail be pure chance. That’s life.
If the T/R-ratio is worse, and/or the CV higher, and/or the dropout-rate higher, you loose power though you might still pass (see there). Happens all the time.

❝ … and I think this variability affect the reliability of results, …


Not necessarily.

❝ … for example, I have a BE study for (lansoprazole), partially replicate design, the published intrasubject variability ~ 40% …


That’s on the lower end for lansoprazole. I’ve seen studies with substantially higher CVs.

❝ … but the practical intrasubject variability we had was near 85% :confused:


Bad luck if you designed the study for 40%. Was risky.

❝ … study protocol also stated using ema guidelines for scaling but the study failed, when I used FDA method it passes :confused:


We are going in circles. Since you stated in the protocol that the study will be assessed for the EMA’s approach, you failed. Full stop. The fact that you would pass with RSABE is irrelevant.

❝ … does this high variability have a role in this confusing results?


Of course, it does. In ABEL you impose an upper limit to expanding the limits, which is at CVwR = 50% for the EMA (max. expansion 69.84–143.19%) and CVwR ~57.4% for Health Canada (66.7–150.0%). For the FDA’s RSABE there is no such restriction. Hence, for any CV, the sample size for RSABE will be lower than the ones for the variants of ABEL. Or the other way ’round: For a given CV and sample size, RSABE has more power than ABEL. See there.

library(PowerTOST)
CV     <- sort(c(seq(40, 90, 10), 85))
theta0 <- 0.95
design <- "2x2x4"
target <- 0.80
res    <- data.frame(CV = CV, n = NA_integer_,
                     power.ABEL = NA_real_, power.RSABE = NA_real_,
                     power.ABEL.n1 = NA_real_, power.RSABE.n1 = NA_real_)
for (j in 1:nrow(res)) {
  tmp                <- sampleN.scABEL(CV = CV[j]/100, theta0 = theta0,
                                       design = design, targetpower = target,
                                       details = FALSE, print = FALSE)
  res$n[j]           <- tmp[["Sample size"]]
  res$power.ABEL[j]  <- tmp[["Achieved power"]]
  res$power.RSABE[j] <- power.RSABE(CV = CV[j]/100, theta0 = theta0,
                                    design = design, n = res$n[j])
  if (j == 1) {
    res$power.ABEL.n1[j]  <- res$power.ABEL[j]
    res$power.RSABE.n1[j] <- res$power.RSABE[j]
  } else {
    res$power.ABEL.n1[j]  <- power.scABEL(CV = CV[j]/100, theta0 = theta0,
                                          design = design, n = res$n[1])
    res$power.RSABE.n1[j] <- power.RSABE(CV = CV[j]/100, theta0 = theta0,
                                         design = design, n = res$n[1])
  }
}
names(res)[5:6] <- c(paste0("power.ABEL (n=", res$n[1], ")"),
                     paste0("power.RSABE (n=", res$n[1], ")"))
res$CV <- sprintf("%.0f%%", res$CV)
print(res, row.names = FALSE)

Gives:

  CV  n power.ABEL power.RSABE power.ABEL (n=20) power.RSABE (n=20)
 40% 20    0.82131     0.86422           0.82131            0.86422
 50% 22    0.84475     0.90013           0.79996            0.86790
 60% 24    0.80957     0.90602           0.71029            0.85313
 70% 30    0.81094     0.91987           0.57715            0.82659
 80% 38    0.82330     0.92529           0.43569            0.79581
 85% 40    0.80422     0.91998           0.36923            0.78044
 90% 44    0.80858     0.92106           0.30796            0.76509

That’s why your study failed with ABEL and may have passed with RSABE.

❝ … also in the future protocol can I specify that if the intrasubject variability is more than 50% can I switch to the FDA method for calculation, as many authorities we submit our studies to follow ema guidelines?


No, you can’t. That’s data-driven. OK, you can but it will not be accepted.

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
BEQool
★    

2023-10-09 16:47
(171 d 00:48 ago)

@ Helmut
Posting: # 23747
Views: 1,676
 

 ABEL ≠ RSABE

Hello!

I have seen this or similar explanation several times:

❝ If you design studies for 80% power and all of your assumptions are exactly realized (T/R-ratio, CV, dropout-rate), one out of five will fail be pure chance. That’s life.


... but I still can't get my head around this.

Lets say I plan study with theta0=0.95, CV=0.20, design="2x2", targetpower=0.8 (all except CV are default settings in PowerTOST), so with sampleN.TOST I get N=20 and power=0.834680
> sampleN.TOST(theta0=0.95, CV=0.20, design="2x2", targetpower=0.8, print=FALSE)
  Design alpha  CV theta0 theta1 theta2 Sample size Achieved power Target power
    2x2  0.05   0.2  0.95    0.8   1.25          20      0.8346802          0.8


If all of my assumptions in a study are exactly realized, doesnt it mean that I would 100% always get bioequivalent formulations (and the study would never fail)?
If all of my assumptions in a study are exactly realized (pe=0.95, CV=0.20, design="2x2", n=20) then I would get the following confidence interval:
> CI.BE(pe=0.95,CV=0.20,n=20)
    lower     upper
0.8522362 1.0589787


So because a confidence interval is completely within limits 80-125%, my formulations would always be bioequivalent because I would always get this confidence interval with same numbers (CV,n,pe)?

What am I not understanding here?
Thank you and best regards
BEQool
mittyri
★★  

Russia,
2023-10-11 14:42
(169 d 02:52 ago)

@ BEQool
Posting: # 23749
Views: 1,565
 

 power.TOST.sim and uncertainty

Dear BEQool!

❝ Lets say I plan study with theta0=0.95, CV=0.20, design="2x2", targetpower=0.8 (all except CV are default settings in PowerTOST), so with sampleN.TOST I get N=20 and power=0.834680

> sampleN.TOST(theta0=0.95, CV=0.20, design="2x2", targetpower=0.8, print=FALSE)

❝   Design alpha  CV theta0 theta1 theta2 Sample size Achieved power Target power
❝     2x2  0.05   0.2  0.95    0.8   1.25          20      0.8346802          0.8


❝ If all of my assumptions in a study are exactly realized, doesnt it mean that I would 100% always get bioequivalent formulations (and the study would never fail)?

❝ If all of my assumptions in a study are exactly realized (pe=0.95, CV=0.20, design="2x2", n=20) then I would get the following confidence interval:

> CI.BE(pe=0.95,CV=0.20,n=20)

❝     lower     upper

❝ 0.8522362 1.0589787


I think exactly realized means a bit different thing.
Take a look at power.TOST.sim() function in PowerTOST package
The description says:
Power is calculated by simulations of studies (PE via its normal distribution, MSE via its associated χ2 distribution) and application of the two one-sided t-tests. Power is obtained via ratio of studies found BE to the number of simulated studies.
So exactly realized means that the variance and centers of distributions used in power estimation were close to the true values, nothing more.

> power.TOST.sim(n = 20, CV = 0.2, theta0 = 0.95, nsims = 1000)
[1] 0.832
# easy to find non-BE replicate
> set.seed(5)
> power.TOST.sim(n = 20, CV = 0.2, theta0 = 0.95, nsims = 1, setseed = FALSE)
[1] 0

Kind regards,
Mittyri
BEQool
★    

2023-10-15 13:29
(165 d 04:05 ago)

@ mittyri
Posting: # 23758
Views: 1,522
 

 power.TOST.sim and uncertainty

Hello Mittyri,

thanks for the answer and explanation!

So power takes into account probability distributions of PEs (normal distribution) and CVs (Chi-squared distribution) and then shows probability that we will be bioequivalent based on the variabilities of PEs and CVs?

Regards
BEQool
mittyri
★★  

Russia,
2023-10-20 23:47
(159 d 17:47 ago)

@ BEQool
Posting: # 23765
Views: 1,483
 

 power.TOST.sim and uncertainty

Hi BEQool,

yes, seems correct to me

Kind regards,
Mittyri
Helmut
★★★
avatar
Homepage
Vienna, Austria,
2023-10-23 12:48
(157 d 04:46 ago)

@ mittyri
Posting: # 23766
Views: 1,457
 

 Distribution of PEs

Hi Mittyri & BEQool,

❝ yes, seems correct to me


I think that the documentation of power.TOST.sim() is confusing and needs some improvement in the next release of the package.
With its (default) argument logscale = TRUE the simulations are performed with log(theta0) – following a normal distribution, which gives lognormal distributed point estimates.

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
BEQool
★    

2023-10-23 21:55
(156 d 19:39 ago)

@ Helmut
Posting: # 23767
Views: 1,458
 

 Distribution of PEs

Mittyri and Helmut, thank you both for your answers.

Maybe a basic questions but it came to mind when I was thinking about this.
Distributions (both normal and log-normal) are determined by their mean and variance. As discussed above, in order to get power we need distribution of PE; the mean is of course PE but what about the variance? Is the variance of PE distribution MSE? Or something else?

BEQool
mittyri
★★  

Russia,
2023-10-31 16:40
(148 d 23:55 ago)

@ BEQool
Posting: # 23770
Views: 1,367
 

 power.TOST.sim code

Dear BEQool!

❝ Distributions (both normal and log-normal) are determined by their mean and variance. As discussed above, in order to get power we need distribution of PE; the mean is of course PE but what about the variance? Is the variance of PE distribution MSE? Or something else?


I think it is better to look into the code

  # simulate point est. via normal distribution
  pes   <- rnorm(n=nsims, mean=diffm, sd=se.fac*sqrt(mse))
  # simulate mse via chi-squared distribution
  mses  <- mse*rchisq(n=nsims, df=df)/df


So the variance is calculated based on MSE, design constant and the number of subjects:
 
power.TOST.sim <- function(<...>) {
  <...>
  d.no  <- .design.no(design)
  <...>
  ades  <- .design.props(d.no)
  <...>
  nc <- sum(1/n)
  <...>
  se.fac <- sqrt(ades$bkni * nc)
  <...>
}

Kind regards,
Mittyri
BEQool
★    

2023-11-02 12:22
(147 d 04:13 ago)

@ mittyri
Posting: # 23772
Views: 1,274
 

 power.TOST.sim code

Dear Mittyri,

thank you very much, everything is clear now!

BEQool
UA Flag
Activity
 Admin contact
22,957 posts in 4,819 threads, 1,638 registered users;
76 visitors (0 registered, 76 guests [including 7 identified bots]).
Forum time: 16:35 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