Outlier in Bioequivalence

Phoenix Certara Forums Phoenix WNL basics
1

Hello,

I was wondering in which way we should detect outliers in bioequivalence studies. I mean, considering a 2x2x2 study, should I check which aberrant observation for each point in the time scale?

Is WNL able to detect outliers?

What is the common practice in bioequivalence studies, just to eliminate the outlier?

Thanks.

2

Dear Ailton,

out of curiosity: did you state in the Protocol/SAP outliers detection before submission to some agency?

What is the goal of outliers detection?

No known agency requires outlier detection in 2x2x2 (some roumors about Russians only…), so PLEASE don’t do it!

If you have some strong forces behind which insist on it you are free to discover and implement any of existed methods (there are tons of aproaches: Wald test, Studentized residuals, Grubbs test and more) for resulted metrics.

I would suggest to look into Shein-Chung Chow, Jen-pei Liu-Design and Analysis of Bioavailability and Bioequivalence Studies, 3rd edition (Chapman & Hall Crc Biostatistics Series) (2008) for more details if needed.

Since the values in ‘concentration’ column are not just values but the result of some ‘magik’ related to chemistry with unavoidable constraints of the detection method, it is impossible to define ‘aberrant observation’. Please also note that the human is not a PK model, so you never know which value is just a ‘lab fantasy’.

Bests,

Mittyright

3

Dear Mittyright

Thanks for your reply.

I was interested in a way to find subjects with aberrant reading, so that we could go to his/her clinical record and see if something happened.

I will take a look at your suggestions.

Regards

4

Dear Mittyright,

I was wondering, if there might be outliers in the study. Please, see the attached files. We conclude non-bioequivalence, but as you can see the geral plots (see attacment), show the curves pretty much close.

What do you think?

Thanks.

itraconazol - graficos gerais.docx (39.6 KB)Results.docx (16.1 KB)

5

Dear Ailton,

sorry to say but…

a Swedish inspector once say at a DIA meeting: don’t worry, it’s too late…

As far as I can see there’s a real bioinequivalence and you are right that’s a good idea to try to find the reason (by the way this study seems to be rejected anyway)

I suspect you’ll find some subjects with very low profile after T treatment. Please look at Ratios Test in your Bioequivalence object result, that’s an easiest way. The subjects with lowest ratios are influencing more than others in your case. After that you can compare their T/R profiles to undestand the reason.

Another way is to try to plot the results as it is described here:

http://forum.bebac.at/mix_entry.php?id=1296#p1298

BR,

Mittyright

6

Hi Mittyright,

I am convinced that even having outliers we cannot do much about it.

This another study (see attachment) have a very close BE, regardless the AUC_inf parameter which is not used to conclude BE, the AUC_last shows an IC of 79,72% - 86,05%, which has almost 80% as the lower bound. In this case, I would conclude BE, but the legislation fixes 80% as lower bound, statistically, can we check something else to argue in favour of BE?

Ailton.

Results - study 2.docx (16 KB)

7

Hi Ailton,

Unfortunately to seek outliers after having assessed BE makes it very hard to justify that you aren’t ‘cherry-picking’ the data. I don’t think you can do anything futher with this study; since the cut-off boundary is set and that is the end of the conversation in this analysis/protocol. I can’t see a good argument for a subsequent change from your analysis plan to squeeze in to the legislative cut-off.

I would actually suggest your formulation is consistently giving lower exposure; since 79.72% - 86.05% is pretty tight and you have to look at reformulation.

Simon

8

Hi Ailton,

I agree with what Simon wrote above about cherry-picking. Something creeps into my mind: Did you expect such a low variability? If yes, did you also expect a low T/R-ratio (say 0.85?). If the answer to both is yes, it would explain the sample size of ~52 subjects. IMHO, the IEC should have rejected the protocol. The design smells of “forced bioequivalence”. Don’t play games.

9

Thanks guys for the suggestions. I am new in BE, so I needed to clarify some practical issues. I’ve rejected BE, BTW. I wanted just to be sure that I was not being unfair. Now I have the picture how things work in BE studies.

Regards.