Bioequivalence and Bioavailability Forum • Outlier test for reference replicate crossover study design

abhimanyu
☆

Banglore,
2019-08-13 17:11
(2158 d 17:21 ago)

Posting: # 20494
Views: 3,535

Outlier test for reference replicate crossover study design [Regulatives / Guidelines]

Hello Helmut/All,

I have question regarding outlier test (Using Lund's method) for reference replicate crossover study design.

How can we find outlier? which one will be appropriate scenario for any regulatory point of view?

Scenario 1 : can we use test and individual reference (R R) to calculate standardized residual?

Scenario 2 : Can we use test and average of reference (average of R R) to calculate standardized residual?

Thanks in advance!

—
Abhimanyu

ElMaestro
★★★

Denmark,
2019-08-13 17:28
(2158 d 17:05 ago)

@ abhimanyu
Posting: # 20495
Views: 2,925

Outlier test for reference replicate crossover study design

Post reply

Hi Abhimanyu,

❝ I have question regarding outlier test (Using Lund's method) for reference replicate crossover study design.

Sounds like a bad idea.

❝ How can we find outlier? which one will be appropriate scenario for any regulatory point of view?

The trick is to avoid generating outliers. In my small universe, there is no problem that merits a solution involving identification and and removal of outliers.

—
Pass or fail!
ElMaestro

Helmut
★★★

Vienna, Austria,
2019-08-13 18:40
(2158 d 15:52 ago)

@ abhimanyu
Posting: # 20496
Views: 2,939

Murky waters

Post reply

Hi Abhimanyu,

first of all I agree with ElMaestro.

❝ How can we find outlier? which one will be appropriate scenario for any regulatory point of view?

EMA and others following its ABEL method:
An outlier test is not acceptable. Have to justify that the estimated CV_wR (driving the expanded limits) is not caused by outliers. Aha. Can be explored by box plots of studentized residuals of the EMA’s model. Some people set the limits to 2×IQR. If outliers are detected, remove them and recalculate CV_wR. However, the analysis has still to be done with the complete data set (the outlier is not removed). Since the expanded limits will be narrower you shoot yourself in the foot.
FDA:
No way.
Health Canada:
See the guidance (Sections 2.3.5 and 2.7.4.1). Outliers can be completely removed from the data set if
- identification is performed before the BE assessment (i.e., no CIs calculated already),
- the studentized model residual is outside {–3, +3} (irrespective whether T or R),
- identified for all PK metrics (that’s a big obstacle, especially if one has to deal with partial AUCs of multiphasic products),
- not more than 5% of subjects (or only one subject if the sample size is 20 or lower),
- the procedure is stated unambiguously in the protocol.

—
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