Bioequivalence and Bioavailability Forum

Hello Mithu,

❝ kindly share your thoughts on Repeat analysis based on the ISR data. During ISR sometimes we found that all the selected samples from any specific subject get failed to meet the acceptance criteria and failing with the huge margin with some specific trend i.e. positive or negative.

"All samples" means what?
Usually for ISR you only have a few from a subject period.

Come to BioBridges in Prague next month and someone may mention something having a wee bit of remote resemblance to your situation.

❝ In such cases, scientists use to go for the investigation and run same samples of ISR again to find out the reason for the failure. During investigation they get the same values which they had got at the time of ISR analysis and not matching with the initial analysis.

Be careful. If you are submitting to FDA then be aware that some of their regulators take the guidance very literally. The guidance says no repeats without a root cause, meaning that you do not repeat a sample as part of an investigation. You only repeat a sample if the investigation has given you a root cause (I am not in favour of the literal interpretation of the FDA guidance, but that aspect is entirely is in the hands of powerful people. I am aware of a few EU regulators who are or were enforcing the same policy).

❝ Based on this they conclude error in the first/ initial analysis and they reject the original batch which was otherwise technically passing the batch acceptance criteria.

If there is a root cause, sounds fair. But if they reject original the batch on basis of the outcome of the repeat only, then it may get tricky.
Some CROs or labs have some very elaborate SOPs with all sorts of funky decision trees allowing e.g. N repeats and if the median or mean or something is more than X % from the original then the reported value is so-and-so and blahblah. This gets regularly questioned by inspectors. Much safer is to report the original unless you have a root cause which in turn isn't founded in the set of repeat values.
Someone hammering your API5000 with a baseball bat and your detector going bat shit crazy, that's a root cause justifying a repeat with reporting of the repeat. A deviation of whatever%
from the original value isn't in itself guaranteed to be a root cause that merits reporting of the repeat, even if the deviation is 250% and verified with a tripple repeat.