Chromatography Forum

Averaging an OOS results with other in specification can be a problem for FDA. Several defficient letter can be found regarding this topic (see links below). For replicates of injection according OOS FDA guidance should not be a problem, however, this can be interpreted in several ways by an auditor. In all cases, I prefer, in fact I do, compare the individual results with the specification and only I do a average if all samples meet the specifications.


http://www.fda.gov/downloads/drugs/guid ... 170584.pdf
http://www.fda.gov/ICECI/EnforcementAct ... 182206.htm

StephenO, you are right that if you make two measurements and one is OOS, the other OK, you don't know which to trust.

But think of all those who only make a single measurement. They have only half the information of those who make two, but are never in doubt about the result. How can half as much information offer an infinite increase in certainty?

Understanding this is actually key to understanding the whole of this thread...

Ok so how do you know which injection is representative of the sample, the OOS or the injection which is within the specification?

An instrument malfunction could of occured on either of the injections and you dont know which injection that is, how can you trust your result?

If your duplicate injections don't agree, then I would be very suspect of just assuming that one injection would suffice.

My wife says she prefers two injections.

the point I'm getting at is that two injections will never agree. If you make single measurements, you are relying on knowing the standard deviation of "typical" measurements from when you did the validation. If you make triplicate measurements, you know the standard deviation for the actual measurements made. In either case you can check the probability of the sample having a particular value, so you can estimate whether it is OOS with a known probability of misclassification.

If your sample is so close to the cut-off value that some measurements are above, some below, then your measurements are not good enough to be sure which side of the cut-off it is. This honestly doesn't matter, if the cut-off was selected taking the measurement error into consideration. For example, if the sample must not exceed X, but measurements have an error s.d., then depending on how certain you want to be that no sample exceeds X, you test that it does not exceed Y = X-n*(s.d.) (choose "n" depending on how sure you want to be). In this instance, if your sample comes out so close to Y that you're not sure whether it's a pass or fail, it doesn't matter which way you classify it. The question you're worrying about is "am I going to be wrong 50% of the time when comparing my sample to Y?", but the underlying situation is whether you're going to be wrong, say, 0.99% of the time, or 1.01% of the time, when comparing your sample to X. So from a theoretical perspective, the right thing to do is to classify based on single measurements or means of multiple measurements according to the basis on which you chose your cut-off value Y, and not to worry in the faintest about borderline cases. Generally, if it mattered which side of the border they fell, then there should have been a bigger safety margin (or a more precise assay).

But of course regulatory bodies trump everything, and are not obliged to be logical. You must ultimately do what is required.

Here at my company - we make OTC dental pharma and also disinfectants (Health canada, FDA and ISO regulated depending on product/shipping) - this is our common practise:

X blanks
6 x system suit - <=2% RSD of area
1 x standard (usually system suit prep)
2 x sample prep 1
2 x sample prep 2
1 x standard
2 x sample and so on...

The average of the bracketing standard area is used to calculate the 4 injections of sample (per preparation, not individually). All area %RSD must be less than 2 and %difference for RESULT of sample preps must be less than 3%. I admit having the 3% requirement helped me hone my sample prep skills for certain products when I first started here.

ANY failing result is a total fail, regardless of average. Also any % difference fail is a rerun regardless of result being in spec or not.

For any impurity methods we do single sample prep with duplicate injection....allllways duplicate injection to show injector is still precise...I guess?

I think its overkill and I'd like to find a better way to bracket so that i can set up the reports in Empower for a one click full report.

But of course regulatory bodies trump everything, and are not obliged to be logical. You must ultimately do what is required.

Ain't that the truth !!!

Cody - we use the same injections of calibration standard for system suitability and as analytical standard. We run 9 samples (duplicate injections) then a single calibibration standard to bracket, then 9 more samples if there are that many (etc.).

Next project: let's define "zero".

CPG that's cool. In R&D (where I am) I bracket triple preps for validations or more for investigations etc. And for all new formula investigation work I just make one prep.

I think the idea in QC is to bracket each batch/lot so that if something happens there's less rerun - which is complete bull because QA will force reruns all the time for things that are ok but they don't understand. They'll also get like 10 analysts to rerun OOSs jusssst to make sure...auditor laugh at all the extra stuff we do. I think we should get cited for it.

We have OOS so infrequently so re-running is not common. Our procedures are very simple preparations and autoamted chromatography, so largest source of error is the weighing of the sample.

Most of our OOS results are from old samples that have dried out, so results are OOS high. I say since they fail the appearance test and are OOS because of that, that we shouldn't even assay them, because they are known OOS. But no one really listens....

Haha such is life..

If that was happening here their justification would be that its still ok just looks ugly...IE we can change the appearance spec or release under deviation and still make money off it! Actually, now that I think of it we have done that for a number of products...sometimes I wonder what the purpose of a quality system is.

Lol