Origin of Suitability Criterion

[Jumpshooter]

A 2% RSD on 5 consecutive injections of a 100% STD is the typically cited way of demonstrating "suitability". Was this number of 2% arbitrary or does it have statistical foundation?
As far as I can surmise, the F-distribution (in Statistics) that controls for Type 1 error is set at 0.05 or 5% of the tail in a distribution.
What is the origin of the 2% RSD criterion?

[tom jupille]

What is the origin of the 2% RSD criterion?

"Because we can." HPLC is fairly comfortable at that level, so no one wants to rock the boat.

If we wanted to do a better job, we would recognize that the precision of a measurement needs to be better than the precision required for control of the process or product. For example, if you need to control the potency of a tablet within +/- 5%, you had better be able to measure to about +/- 2% (or better). There is a whole branch of statistical quality control called "gauge theory" that deals with this (most of which is beyond my skill level!), but one useful concept for estimating how precise a measurement must be is the Wheeler "discrimination ratio". Here's a slide from our "Validating HPLC Methods" course:



Of course, this begets the question "why did they decide that 4 was the minimum for the discrimination ratio?" (I don't know the answer to that one! ).

[Consumer Products Guy]

USP itself is not that consistent. I've noticed that USP allows %RSD of 4 for alcohol test USP <611> and we're most often under 1, maybe that number was from manual injections, but that is an internal standard GC procedure. For parachloroaniline assay in 20% chlorhexidine solution, USP allows a 5% RSD.

[Jumpshooter]

Well this certainly helped to clarify the origin (and dare I say "rationale") for the 2.0% RSD criterion. Thanks for the slide T.Jupille.
Also, I noted that several exceptions to this criterion are well-cited in the literature (e.g. USP)--thanks ConsumerProdGuy.

So, then I must out ouf necessity pose this question: Is the notion of a 2% RSD an assay-dependent criterion? Because evidently the alcohol assay "611" does relax this criterion to 5%. So is it safe to say that it is assay dependent AND that whatever value it is established to be then it better be Less than the variance in the product/process we are measuring?

[tom jupille]

So is it safe to say that it is assay dependent AND that whatever value it is established to be then it better be Less than the variance in the product/process we are measuring?

Yes, I would agree with that.