Bioequivalence and Bioavailability Forum • Truncated 72 hours

mmw
☆

India,
2015-02-09 12:12
(3592 d 09:28 ago)

Posting: # 14389
Views: 11,620

Truncated 72 hours [NCA / SHAM]

Dear All,

We have conducted a truncated bio-equivalence study (truncation up to 72 hours as per European GL requirements). But it was found that around one-third of the study population did not turn up for 72.00 hours ambulatory sample.

Is their any impact on calculation of AUC due to number of missed sample? Could these subjects be excluded from statistical analysis? If yes, on what basis we exclude these subjects? However, this was not mentioned anywhere in guideline regarding exclusion of such subjects in truncated studies?

Thanks

MMW

Edit: Category changed. [Helmut]

Dr_Dan
★★

Germany,
2015-02-09 13:59
(3592 d 07:41 ago)

@ mmw
Posting: # 14390
Views: 10,019

Truncated 72 hours

Post reply

Dear MMW
Since the subjects did not show up for the 72 h sample you can not use these subjects for statistical evaluation. As a consequence you have less subjects and wider 90% CI. If the study passes everything is fine, if not you can argue by calculating AUCt for these subjects and all other subjects (AUCt=AUC72) and repeat ANOVA.
I hope this helps.
Kind regards
Dr_Dan

—
Kind regards and have a nice day
Dr_Dan

Helmut
★★★

Vienna, Austria,
2015-02-09 17:10
(3592 d 04:30 ago)

@ mmw
Posting: # 14391
Views: 10,032

Truncated 72 hours

Post reply

Hi MMW,

❝ But it was found that around one-third of the study population did not turn up for 72.00 hours ambulatory sample.

Shit happens. Since AUC₇₂ is a primary PK metric consider hospitalizing subjects in the future. By opting for ambulatory sampling you saved some rupees but risked to substantially loose power.

❝ Is their any impact on calculation of AUC due to number of missed sample? Could these subjects be excluded from statistical analysis? If yes, on what basis we exclude these subjects?

You should have asked yourself these questions before starting the study. It is good practice to expect the unexpected and state a bail-out procedure in the protocol. Whatever you do post hoc might smell fishy. :cherry picking:

❝ However, this was not mentioned anywhere in guideline regarding exclusion of such subjects in truncated studies?

Laking statements in guidelines are never an excuse for own judgments. Since C_max likely is more variable than AUC, it might that you even pass BE after excluding subjects with missing AUC₇₂ data. But one third is an awful lot…

[image]

Not sure whether presenting BE of AUC_t helps. If half lives were really long, maybe. Otherwise you end up with apples-and-oranges statistics.

In my protocols I state to use an estimated value anyway. This not only helps to correct for deviations from scheduled sampling times, but with ‘missings’ as well. In Phoenix/WinNonlin you could opt for a partial AUC_0–72. In order to do so, you need a reliable estimate of λ_z (at least three data points after t_max).

—
Dif-tor heh smusma 🖖🏼 Довге життя Україна!
Helmut Schütz

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

nobody nothing 2015-02-09 23:48 (3591 d 21:53 ago) @ Helmut Posting: # 14393 Views: 9,956	Truncated 72 hours Post reply
	Hi everybody! Just to add the obvious: ❝ ... In order to do so, you need a reliable estimate of λ_z (at least three data points after t_max). ...and really in the terminal phase, please... — Kindest regards, nobody

mmw ☆ India, 2015-02-11 13:57 (3590 d 07:43 ago) @ nobody Posting: # 14401 Views: 9,734	Truncated 72 hours Post reply
	Dear All, Thanks for your reply. MMW

stic-stats
☆

India,
2020-07-13 16:22
(1611 d 06:19 ago)

@ Helmut
Posting: # 21687
Views: 5,713

Truncated 72 hours

Post reply

Dear Helmut and All,

❝ ❝ But it was found that around one-third of the study population did not turn up for 72.00 hours ambulatory sample.

❝

❝ Shit happens. Since AUC₇₂ is a primary PK metric consider hospitalizing subjects in the future. By opting for ambulatory sampling you saved some rupees but risked to substantially loose power.

May I know what type of power loose.
please clarify my dought.

Is their any impact on calculation of AUC due to number of missed sample? Could these subjects be excluded from statistical analysis? If yes, on what basis we exclude these subjects?

Regards,
Stic-Stats.

Edit: Full quote removed. Please delete everything from the text of the original poster which is not necessary in understanding your answer; see also this post #5. Standard quotes restored; see also this post #8. [Helmut]

Helmut
★★★

Vienna, Austria,
2020-07-13 19:07
(1611 d 03:33 ago)

@ stic-stats
Posting: # 21689
Views: 5,693

Truncated 72 hours

Post reply

Hi stic-stats,

❝ ❝ […] Since AUC₇₂ is a primary PK metric consider hospitalizing subjects in the future. By opting for ambulatory sampling you saved some rupees but risked to substantially loose power.

❝

❝ May I know what type of power loose.

Fewer eligible subjects will always decrease power. To which extent is unclear (mmw posted more than five years ago) and without the CV and sample size we would need a crystal ball.
Example (2×2 crossover, T/R 0.95, target power 0.80):

library(PowerTOST) CV <- seq(20, 40, 5) # intra-subject CV in percent dor <- seq(0, 15, 5) # dropout-rate in percent res <- data.frame(CV = rep(CV, each = length(dr)), n = NA, dor = dor, elig = NA, power = NA) for (j in 1:nrow(res)) { tmp <- sampleN.TOST(CV = res$CV[j]/100, print = FALSE, details = FALSE) res$n[j] <- tmp[["Sample size"]] if (res$dor[j] == 0) { res$elig[j] <- res$n[j] res$power[j] <- tmp[["Achieved power"]] } else { res$elig[j] <- floor(tmp[["Sample size"]]*(100-res$dor[j])/100) res$power[j] <- suppressMessages(power.TOST(CV = res$CV[j]/100, n = res$elig[j])) } } res$power <- signif(res$power, 5) names(res)[1:4] <- c("CV (%)", "dosed", "dropouts (%)", "eligible") print(res, row.names = FALSE)

Gives:

CV (%) dosed dropouts (%) eligible power 20 20 0 20 0.83468 20 20 5 19 0.81324 20 20 10 18 0.79124 20 20 15 17 0.76365 25 28 0 28 0.80744 25 28 5 26 0.77606 25 28 10 25 0.75766 25 28 15 23 0.71729 30 40 0 40 0.81585 30 40 5 38 0.79533 30 40 10 36 0.77239 30 40 15 34 0.74666 35 52 0 52 0.80747 35 52 5 49 0.78311 35 52 10 46 0.75583 35 52 15 44 0.73545 40 66 0 66 0.80525 40 66 5 62 0.77978 40 66 10 59 0.75824 40 66 15 56 0.73466

❝ Is their any impact on calculation of AUC due to number of missed sample?

Yes. Since generally studies are planned for the PK-metric with the highest variability (likely C_max) the overall impact on power should be limited.

❝ Could these subjects be excluded from statistical analysis? If yes, on what basis we exclude these subjects?

Everything is possible – if stated in the statistical protocol. You could exclude them (only AUC whilst keeping C_max), use an estimate (if λ_z is reliable), use the AUC to the last time point where you have concentrations for both T and R,…
Please search the forum; lots of posts.

—
Dif-tor heh smusma 🖖🏼 Довге життя Україна!
Helmut Schütz

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

stic-stats ☆ India, 2020-07-15 08:17 (1609 d 14:23 ago) @ Helmut Posting: # 21697 Views: 5,595	Truncated 72 hours Post reply
	Dear Helmut, Thank you for your response. I am fresher in the field. few days back I joind BEBAC Forum. Edit: Full quote removed. Please delete everything from the text of the original poster which is not necessary in understanding your answer; see also this post #5. [Helmut]

M.tareq
☆

2020-07-16 03:36
(1608 d 19:04 ago)

@ Helmut
Posting: # 21706
Views: 5,612

Truncated 72 hours

Post reply

❝ In my protocols I state to use an estimated value anyway. This not only helps to correct for deviations from scheduled sampling times, but with ‘missings’ as well.

Is it possible to use a model-based approach to estimate the overall/typical value of Ke or CL and use that for prediction of such points ?

Or is it simpler to just perform a log-linear regression on averaged/all terminal points (3xTmax) and use that estimate to account for such deviations as defined in protocol.

Thanks

Mahmoud

Helmut
★★★

Vienna, Austria,
2020-07-16 13:56
(1608 d 08:44 ago)

@ M.tareq
Posting: # 21711
Views: 5,589

Truncated 72 hours

Post reply

Hi Mahmoud,

❝ Is it possible to use a model-based approach to estimate the overall/typical value of Ke or CL and use that for prediction of such points ?

Since PK modeling is not acceptable in BE, no. Furthermore, an average value might be misleading. Imagine the drug is subjected to polymorphic metabolism. The 90% extensive metabolizers have an average t_½ of 4 hours and the 10% poor metabolizers 16 hours. You end up with an overall (geometric mean) t_½ 4.59 hours. Now you could think about basing the decision to which group the subject belongs (i.e., instead of the average, use 4 or 16 hours) on the AUC. Spice the data with high between subject variability and you are at a loss.

❝ Or is it simpler to just perform a log-linear regression on averaged/all terminal points (3xTmax) and use that estimate to account for such deviations as defined in protocol.

That’s exactly what I do (and performed in Phoenix/WinNonlin when you specify a partial AUC_0–72 with missings and/or deviations from the scheduled 72 hours sampling).

—
Dif-tor heh smusma 🖖🏼 Довге життя Україна!
Helmut Schütz

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

M.tareq
☆

2020-07-16 19:05
(1608 d 03:35 ago)

@ Helmut
Posting: # 21717
Views: 5,479

Truncated 72 hours

Post reply

❝ Since PK modeling is not acceptable in BE, no. Furthermore, an average value might be misleading. Imagine the drug is subjected to polymorphic metabolism. The 90% extensive metabolizers have an average t_½ of 4 hours and the 10% poor metabolizers 16 hours. You end up with an overall (geometric mean) t_½ 4.59 hours. Now you could think about basing the decision to which group the subject belongs (i.e., instead of the average, use 4 or 16 hours) on the AUC. Spice the data with high between subject variability and you are at a loss.

There have been some publications: Model‐based analyses of bioequivalence and Link : 2 about using model-based approach to drive BE, though as you said it's not acceptable in BE "yet", do you think in the near future we will see model based BE studies approved by regulators?

Helmut
★★★

Vienna, Austria,
2020-07-16 20:30
(1608 d 02:10 ago)

@ M.tareq
Posting: # 21719
Views: 5,458

PK modeling in BE

Post reply

Hi Mahmoud,

❝ There have been some publications: […] about using model-based approach to drive BE, though as you said it's not acceptable in BE "yet", do you think in the near future we will see model based BE studies approved by regulators?

I’m not very optimistic, esp. when dealing with missing data or – even worse – sparse sampling. Mats Karlson (quoted in your second reference) is extremely clever and France Mentré has my deepest respect when it comes to PopPK. However, last year at the 10^th American Conference on Pharmacometrics (20–23 October, Orlando) she gave a presentation “Evaluation of Model Based BioEquivalence (MBBE) statistical approaches for sparse designs PK studies” with a whoopy mean (!) Type I Error of 7.6%. :-(

I hope that we can clarify things later this year.

—
Dif-tor heh smusma 🖖🏼 Довге життя Україна!
Helmut Schütz

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

mittyri
★★

Russia,
2020-07-19 02:24
(1605 d 20:16 ago)

@ Helmut
Posting: # 21744
Views: 5,272

PK modeling in BE

Post reply

Hi Helmut,

❝ However, last year at the 10^th American Conference on Pharmacometrics (20–23 October, Orlando) she gave a presentation “Evaluation of Model Based BioEquivalence (MBBE) statistical approaches for sparse designs PK studies” with a whoopy mean (!) Type I Error of 7.6%. :-(

Are you referring to this one?
Stan approach is a bit better, but still questionable

—
Kind regards,
Mittyri

Helmut
★★★

Vienna, Austria,
2020-07-19 14:35
(1605 d 08:05 ago)

@ mittyri
Posting: # 21749
Views: 5,265

PK modeling in BE

Post reply

Hi mittyri,

❝ Are you referring to this one?

Nope, an even worse one. I’ll send it to you by PM.
If you have nothing better to do, watch this recording (navigate to 01:27:02 for the presentation “Improved bioequivalence assessment through model-informed and model-based strategies”) of the 2020 Generic Drug Regulatory Science Initiatives Public Workshop.

Slide 14 (left) shows the Type I Errors for the standard NCA and the model based approach. Why should be even their average in NCA >5%? Sumfink stinks here.

—
Dif-tor heh smusma 🖖🏼 Довге життя Україна!
Helmut Schütz

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