pash413
★

India,
2016-07-11 09:04
(1975 d 11:09 ago)

Posting: # 16481
Views: 28,789

## Outliers in Replicated Crossover Study [Outliers]

Hello All,
This is regarding outlier(s) observation in 2 treatment 2 sequence 4 period Replicated Crossover Study.

As per in the post, I have worked out outlier calculation on my data of 12 subjects.

Data:

Subject  Seq             PK Values                       A1      A2      B1      B2  1      ABAB   1286.8  2469.1   259.9    69.8  3      ABAB   2718.0   839.2  2521.4  1427.9  6      ABAB   2094.3  1724.2  2448.7  2120.5  8      ABAB   1611.2  1940.5  1777.9    58.1  9      ABAB   1918.2  3773.1   411.5  2338.0 12      ABAB  29682.3  1405.9  3319.2  4370.1  2      BABA   1543.0  1118.1  2327.8   149.2  4      BABA   3690.7   952.7   902.7   996.5  5      BABA   2364.1   782.6  1914.7  1789.8  7      BABA   1279.1  1926.6   794.1  1135.5 10      BABA   3769.9  3190.9  3288.9  5244.2 11      BABA   1315.0  2088.0  1079.9   217.4

Calculated Residuals:

Subject  Seq              Residuals                     sShi     sSFhi    sDThi    sDRhi  1      ABAB  -1.7575   1.9362  -1.0832   0.5320  3      ABAB   0.2143  -1.3030   0.6589  -0.0029  6      ABAB   0.5426  -1.2493  -0.2763  -0.3074  8      ABAB  -1.0723   0.9247  -0.6391   2.0422  9      ABAB   0.1487   0.1226  -1.1069  -1.6561 12      ABAB   1.9241  -0.4312   2.4466  -0.6078  2      BABA  -0.8265  -0.1165  -0.1546   1.5592  4      BABA  -0.1650  -0.2546   0.8297  -0.4815  5      BABA   0.1115  -1.3952   0.5925  -0.3622  7      BABA  -0.3054  -0.4608  -0.8524  -0.6670 10      BABA   1.4984  -1.2477  -0.3028  -0.7451 11      BABA  -0.7975   0.3769  -0.9027   0.7386

Above are Residual values for Subject Outlier, Subject by formulation Outlier, Test formulation outlier and Reference formulation Outlier respectively.

Grubb's Value 2.411559518

So, as Residual value of Subject 12, A1 (2.4466) is greater than Grubb's value 2.411559518, it can be considered as Outlier.

So, according to the suggested method in the paper above, is my calculation and result correct?

Edit: Tabulators changed to spaces and BBcoded; see also this post #6. [Helmut]
Helmut
★★★

Vienna, Austria,
2016-07-11 12:56
(1975 d 07:18 ago)

@ pash413
Posting: # 16482
Views: 18,235

## Which agency?

Hi Pash,

please give complete information in the future; see also this post #4. Which jurisdiction / method are you targeting at (the FDA’s RSABE, the EMA’s ABEL)?

» As per in the post, I have worked out outlier calculation on my data of 12 subjects.

Did you read the post in its entirety (especially the linked RTR-guidance)? Keep in mind that FDA’s guidance you maybe had in mind is 15 years old. Even then, you should have contacted the FDA before running the analysis.

If you want to submit to the FDA:

The minimum sample size for RSABE is 24 (dosed). I doubt that you had 13 drop outs in the study and that the FDA would accept removal of an outlier of the test product. Even if the FDA generously would accept removal of subject #12 the study would still fail (PE >125.00%, critbound >0). So why all that fuzz?

If you want to submit to the EMA:

The minimum sample size is 12 eligible subjects. Bad luck with 11. Statistical tests to assess outliers are not acceptable. Furthermore, only the reference product should be assessed for potential outliers by box plots (in your case it was after test).

If this was a pilot study of course you are free to do anything you prefer. However, you should be very wary and not close your eyes / cross fingers by removing an outlier after the test product!

Dif-tor heh smusma 🖖
Helmut Schütz

The quality of responses received is directly proportional to the quality of the question asked. 🚮
Science Quotes
pash413
★

India,
2016-07-11 17:01
(1975 d 03:12 ago)

@ Helmut
Posting: # 16483
Views: 17,997

## Which agency?

Hi Helmut,
Thanks for quick response.

» please give complete information in the future; see also this post #4. Which jurisdiction / method are you targeting at (the FDA’s RSABE, the EMA’s ABEL)?

Pardon me for missing out to mention the details . It is Pilot study for FDA's RSABE, however we want to know whether the outlier detection method which we had worked out according to mentioned in the paper is correct or not ?

» If this was a pilot study of course you are free to do anything you prefer. However, you should be very wary and not close your eyes / cross fingers by removing an outlier after the test product!

This data is of pilot study and we want to confirm the calculation only, whether it is for test product or reference product so that it can help for future pivotal studies. Specifically, if you could guide and check whether the residual values we calculated are correct or not!!

Thank you.
pash413
★

India,
2016-07-14 08:00
(1972 d 12:13 ago)

@ pash413
Posting: # 16490
Views: 17,917

## Residuals and Critical value Calculation?

Dear Friends,
El Maestro, DavidManteigas, Ohlbe and Helmut. Thank u for your discussion, but still my question is not addressed and answered.

» This data is of pilot study and we want to confirm the calculation only, whether it is for test product or reference product so that it can help for future pivotal studies. Specifically, if you could guide and check whether the residual values we calculated are correct or not!!
»
» Thank you.

I agree with all points that you all put here, but the question is: Is my calculation correct???
My calculated residuals and Grubb's critical value is correct?

It would be great help and learning for me.
DavidManteigas
★

Portugal,
2016-07-12 16:55
(1974 d 03:18 ago)

@ pash413
Posting: # 16484
Views: 17,989

## Outliers in Replicated Crossover Study

What is the point of detecting outliers in a bioequivalence study anyway? Unless they're really extreme values (possibly indicating an error) you should never remove them from your analysis. The only point of detecting outliers in replicate designs is to assess if the CV obtained is not due to an outlier instead of real variation in data. Or in a pilot study, to assess the robustness of your estimated CV, for instance.

Outliers do exist and they're expected 5% of the times :)
Ohlbe
★★★

France,
2016-07-12 17:50
(1974 d 02:23 ago)

@ DavidManteigas
Posting: # 16485
Views: 17,918

## Outliers in Replicated Crossover Study

Dear David,

» The only point of detecting outliers in replicate designs is to assess if the CV obtained is not due to an outlier instead of real variation in data.

My understanding is that this is precisely what Pash is willing to do.

Regards
Ohlbe
ElMaestro
★★★

Denmark,
2016-07-13 18:49
(1973 d 01:24 ago)

@ pash413
Posting: # 16487
Views: 17,993

## Outliers in Replicated Crossover Study

Hello Pash413,

why bother?
I mean, the regardless of whether this value represents real life or not, I am not sure it changes anything for you. The matrix likelihood does change, but I don't think it has implications for the within subject variance for Ref, does it?
It is the latter that you can use for planning of new studies.

If the aberrant value is 'real' then it has implications for the point estimate, of course. But in a pilot study your sample is typically so small that you do not know the point estimate well (your CI's are big), therefore it might not at all be healthy to use the point estimate for anything. You can't make good decisions from the observed point estimate until the sample size is like in a pivotal trial. You can say that in a pivotal trial the sample size is so big that you can justify making a decision about the point estimate. It isn't when the sample size is pilot scale. It is almost a definition, but not one any company likes to discuss broadly.

Pass or fail!
ElMaestro
Helmut
★★★

Vienna, Austria,
2016-07-13 19:56
(1973 d 00:17 ago)

@ ElMaestro
Posting: # 16488
Views: 18,125

## I would stop

Namaste ElMaestro,

» why bother?

I agree in general. But here we have after exclusion of subject #12 almost proven inequivalence (PE: 221%, lower 90% CL: 123.34%). The CVwR depending on the calculation method is >155%. One has to be an outstanding optimist to continue.

Dif-tor heh smusma 🖖
Helmut Schütz

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

India,
2017-10-05 15:46
(1524 d 04:27 ago)

@ pash413
Posting: # 17863
Views: 15,016

## Outliers in Replicated Crossover Study

Dear All,

Apologies for opening this old thread but hope this could help me in one of my present study.

We have conducted one Pivotal BE study for NTI drug.

Study is passing in all the criteria except that upper limit of the 90% equal-tails confidence interval for σWT/σWR is coming more than 2.5.

The reason behind the same is that there is difference in intrasubejct variability between test and reference product for one of the PK parameter.

Incase we exclude one subject from the study (suspected outlier) then variability is coming almost similar for test and reference product.

For this subject there is five fold difference in PK parameter for the same treatment which is also not inline with other subject data.

Can mentioned method be used for outlier detection. Has anyone used the mentioned method for FDA dossier submission.

Apart from the above, incase pilot study is conducted in full replicate design with adequate number of subject and we want to plan Pivotal study considering the variability of test and reference product, is there any program to calculate the sample size for which upper limit of the 90% equal-tails confidence interval for σWT/σWR can be lesser than 2.5.

As such the σWT/σWR ratio won't change but f value and degree of freedom would change based on the sample size and accordingly upper limit of the 90% equal-tails confidence interval will change.

Regards
Pjs
d_labes
★★★

Berlin, Germany,
2017-10-05 17:25
(1524 d 02:48 ago)

@ pjs
Posting: # 17864
Views: 15,100

## Sample size for the FDA NTID BE decision

Dear pjs,

» ... is there any program to calculate the sample size for which upper limit of the 90% equal-tails confidence interval for σWT/σWR can be lesser than 2.5.

Not exactly what you have requested, but the whole BE decision. Namely: For deciding BE the study must pass the 95% upper CI of the linearized RSABE criterion <0, the conventional ABE test and additionally the test that the ratio of sWT/sWR is <= 2.5.
R package PowerTOST has the functions power.NTIDFDA() / sampleN.NTIDFDA()

Have a look if this fits your needs.

Regards,

Detlew