joyjac
★

Philippines,
2006-01-27 00:39
(5627 d 21:08 ago)

Posting: # 65
Views: 15,036

## Deletion of outliers [Outliers]

Several methods of outlier detection have been discussed in the literature. Would you know if US FDA recommends deletion of outliers? May we seek your opinion on this issue.

Thanks.
Helmut
★★★

Vienna, Austria,
2006-01-27 13:51
(5627 d 07:56 ago)

@ joyjac
Posting: # 69
Views: 14,092

## Deletion of outlying subjects

Dear Joy!

» Would you know if US FDA recommends deletion of outliers?

OK. Let's talk about outlying subjects (not 'inadequate profiles' / 'pharmacokinetic outliers').

FDA (like all other regulatory authorities) strongly discourages deletion of outliers based solely on statistics.

You should include a procedure in your protocol (subjects may be excluded prior to statistical analysis if e.g., vomiting or diarrhea was observed). http://www.fda.gov/cder/guidance/5356fnl.pdf FDA states:

'We recommend that data from subjects who experience emesis during the course of a BE study for IR products be deleted from statistical analysis if vomiting occurs at or before 2 times median Tmax. In the case of MR products, the data from subjects who experience emesis any time during the labeled dosing interval can be deleted.'

If you find an outlying subject based on product failure of the reference in a replicate design on only one occasion, the outlier http://www.fda.gov/cder/guidance/3616fnl.pdf may be deleted, but '[...] the retest character of these designs should indicate whether to delete an outlier value or not. Sponsors or applicants with these types of data sets may wish to review how to handle outliers with appropriate review staff.'

Aside from rumours spreading at numerous BA/BE conferences about retesting of outlying subjects (a small study including the outlier and 20% of subjects who showed 'normal' values in the main study), I don't know of an official guideline issued by the FDA covering this topic. One warning: if you decide for retesting, statistics will be very tricky - nothing for M\$-Excel

Maybe these proceedings from the Canadian HFPB may give you further insights:
http://www.hc-sc.gc.ca/dhp-mps/prodpharma/activit/sci-com/bio/sacbb_rop_ccsbb_crd_2001-11-15_e.html
http://www.hc-sc.gc.ca/dhp-mps/prodpharma/activit/sci-com/bio/sacbb_rop_ccsbb_crd_2004-06-03_e.html

If you don't have rules specified in the protocol a-priori, your chances of acceptance are low - it may look like 'data dredging', although the ICH Guideline Statistical Principles for Clinical Trials suggests presenting the full data set (including the outlier), and the reduced data set (outlier excluded), and discussing the implications.

Edit: Links corrected for FDA’s new site. [Helmut]

Dif-tor heh smusma 🖖
Helmut Schütz

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

2006-07-10 09:18
(5463 d 12:29 ago)

@ Helmut
Posting: # 159
Views: 13,385

## Deletion of outlying subjects

» OK. Let's talk about outlying subjects (not 'inadequate profiles' / 'pharmacokinetic outliers').

Dear HS,

Request you to please guide how to handle 'inadequate profiles' / 'pharmacokinetic outliers'. We are facing this problem of outliers in the BE study for an Pantoprazole Tablets for submission to Europe.

Essar
Helmut
★★★

Vienna, Austria,
2006-07-10 10:41
(5463 d 11:07 ago)

@ Essar
Posting: # 160
Views: 13,532

## Deletion of outlying subjects

Dear Essar,

I’m a little bit confused…

» Request you to please guide how to handle 'inadequate profiles’ / 'pharmacokinetic outliers'.

By this term generally ‘strange’ profiles are meant, e.g., rapidly fluctuating values in the area of Cmax/tmax (connected to nonphysiological half lives), or increasing values in the elimination phase (which lead to λz > 0 and AUC = ∞ [sic!]).
Such values often are symptoms of analytical problems: remember, the method was validated by spiking blank plasma, and now we have to deal with ‘real world’ samples. Yes, interferences may occur (most commonly caused by the matrix effect in LC/MS-MS, where ion supression / enhancement by co-eluting compounds may alter the signal). Also the methods’s variance and bias is highest close to the LLOQ…

You should include a plausibility review of analytical data before unblinding the study, and initiate repeated analysis of suspect values:

‘The protocol for repeat analysis should be established a priori. Some aberrant values can be identified which can be attributed to processing errors, equipment failure, poor chromatography or QC samples outside predefined tolerance. Cautious use of a ‘pharmacokinetic fit’ such as a double peak may call for repeat analysis of some samples in the study, but the reasoning should be clearly documented.’[1]
‘A standard operating procedure or guideline for repeat analysis and their acceptance criteria must be established a priori. This SOP or guideline should define acceptable reasons for repeating sample analysis, such as sample processing errors, equipment failure, poor chromatography, etc. Cautious use of “pharmacokinetic fit” such as double peak may call for repeat analysis of some samples in the study. The rationale for the repeat analysis and the reporting of the repeat analysis should be clearly documented.’[2]

But remember: if you are a victim of the matrix effect, repeated analysis most likely will only justify the original value. You should have a procedure stated in the protocol, which allows the exclusion of such values (after repeated analysis!), modify the method, or even substitution of these values by estimates.

» We are facing this problem of outliers in the BE study for an Pantoprazole Tablets for submission to Europe.

OK, this sounds more like the topic already discussed. Outlying subjects are a common phenomenon with all the *prazoles…
If you do not have an a-priori procedure against outlying subjects, I’m afraid, the cards are stacked against you…
European regulators have seen many BE studies with *prazoles where sometimes the reference product showed very low values. But even if this was foreseen in the protocol, the acceptance of such studies differed from country to country:
Sometimes the evaluation excluding the subject(s) was accepted (e.g., if it was possible to demonstrate problems with gastric resistance of the reference product in vitro), rarely a nonparametric method – which is robust against outliers – was accepted, and – mainly – the study was rejected…

My suggestion would be retesting of the outlying subjects (together with 20% of subjects showing ‘normal’ responses in the original study), and keep your fingers crossed!
1. Shah VP, Midha KK, Dighe S, McGilveray IJ, Skelly JP, Yacobi A, Layloff T, Viswanathan CT, Cook CE, McDowall RD, Pittman KA and S Spector
Analytical methods validation:
Bioavailability, bioequivalence and pharmacokinetic studies

Int J Pharm 82, 1-7 (1992)
2. Shah VP, Midha KK, Findlay JWA, Hill HM, Hulse JD, McGilveray IJ, McKay G, Miller KJ, Patnaik RN, Powell ML, Tionelli A, Viswanathan CT and A Yacobi
Bioanalytical Method Validation—A Revisit with a Decade of Progress
Pharmaceutical Research 17(12), 1551-1557 (2000)

Dif-tor heh smusma 🖖
Helmut Schütz

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