5 or 6 injections for system suitability

[ym3142]

I run 5 injections for system suitabilty though the protocol says 6 injections. And I have finish the whole validation. I am trying to say 5 should be equal to 6 here. Does any one know some reference for this. Thanks in advance.

[DR]

Ongoing debate - - -
5 or 6, for us, it depends on the type of assay. Also to be argued - do 5 (or 6) have to be consecutive and up front to satisfy suitability requirements, or can you just average standards from all over the run?

We use 5-6 up front, then additional standards (1-3) after each group of 6-12 samples.

[Dan]

ym,

I don't know of a reference for saying that 5 equals 6, but the USP requirements for system suitability may be interpreted as saying they are not equal. The USP states that there are to be 5 injections when the RSD criteria is 2.0% and 6 injections if an RSD critieria is more than 2.0%.

So, I don't think that you can consider that 5 injections is equivalent to using 6 injections. However, statisticis is not my strong point. Maybe someone else can give a good statistical equivalence answer.

I think that the main point here is that the protocol called for 6 injections. It would seem that you deviated from that and someone will probably be asking why. To help answer that, you need to see if there was a specific reason for the prcotocol to ask for 6 injections (is it truely a requirement for your particular method?). Find the answer and then you can determine if 5 can be used instead.

DR,

The FDA addressed the issue of consectutive/up front injections in an issue of Human Drug CGMP Notes in September 1997. Here is the link:
http://www.fda.gov/cder/hdn/cnotes97.pdf
Basically, the FDA stated that a minimum of 3 replicate standard injections should be mabe prior to injections of the samples. The remaining 2 (or 3) standard injections can be made during or at the end of the run. Note that the FDA doesn't use the word "consecutive". They just stated "replicate", I am not sure if they interpret replicate=consecutive.

Regards,
Dan

[ym3142]

Thanks, Dan and DR. That's good.

[jdlh199]

Or do 6 and delete the one that doesn't pass

[Sunjay]

Dear ym3142,

This is a deviation from the protocol and if your RSD limit is 2.0% or less then i think you can justify it as per USP. The USP requires 2.0% RSD for 5 injections. If your RSD limit is more than 2.0% then 6 injections are required as per USP.

But in no way we can say 5 is equal to 6 (as per USP).

Regards

[DR]

Or do 6 and delete the one that doesn't pass

To those of us in regulated industries, running compendial methods etc. I say Noooooooooo! (Unless you delete the first in the series and can blame lack of column equilibrium or something like that).

Employing Dixon's Q and other rho type tests to eliminate outlier results is what got Barr Labs soundly slapped around in their suit vs. the FDA several years ago. We don't do that.

[Sunjay]

I agree with DR. We can not delete any injection midway just because it is not meeing the criteria or it is outlier.

We are required to inject consecutive rplicate injctions for SST.

So if for example out of 5 injections, 3rd injection is outlier, then we have to give 3 more injections to have 5 consecutive rplicate injctions.

We are following this practice in our lab.

Regads

[leadazide]

Or do 6 and delete the one that doesn't pass

To those of us in regulated industries, running compendial methods etc. I say Noooooooooo! (Unless you delete the first in the series and can blame lack of column equilibrium or something like that).

Employing Dixon's Q and other rho type tests to eliminate outlier results is what got Barr Labs soundly slapped around in their suit vs. the FDA several years ago. We don't do that.

Is it OK to delete a run if you loudly and in public say "Practice Run" as soon as you see the chromotogram is a bit wierd?? If not it definetly should be!! It would have saved my rear end more then once.

[HW Mueller]

Gee, ..... seeing all this quality control I am getting constantly more paranoid about swallowing all those pills....

[Peter Apps]

jdlh199's run 6 and delete one is a bit of heavy massaging (aka falsification), but it was joke, right ??

Sunjay is effectively doing the same thing - except that he throws out the first two runs along with the suspect one in order to get five in a row that pass the criterion. This does not measure the performance of the system, which is the whole point of system suitability, it merely demonstrates that when the expected answer is known it is possible to edit data to match a preconceived model. The problem is that when real samples are run there is no expected answer, so how do you know which results to throw out ?

Strictly, if you want to throw out the first run (or any others) you have to decide to do it before you see the data. The most common reason for rejection; a column conditioning run at the beginning of a batch, has already been mentioned, but it could also include a fumbled injection or finger operation etc.

The more replicates that you run, the more robust is your estimate of rsd. You can get a rough idea of rsd from duplicates, but the rsd will differ markedly from duplicate to duplicate. rsds from triplicates will vary less, and so on up.

Peter

[Molly]

I have a related question: when the std was used for quantitation, do you use the average of the std areas across the run (5 or 6), or just use the average of the 3 before the sample was injected. Also can anyone tell me FDA's rationale of having three replicate before the sample? If you see a drift in the signals, is evenly distribute your stds aross the run the worst case, meaning give you the worst RSDs?

Thanks, Molly

[jdlh199]

jdlh199's run 6 and delete one is a bit of heavy massaging (aka falsification), but it was joke, right ??

Luckily it was a joke, maybe I need to add more 's next time?

For the majority of our analyses, we will run a system suitability BEFORE we analyse any samples/QC's/calibration curves. The system suitability will consisit of a diw, a matrix blank, then 5 replicate injections of our low QC (usually 3x lowest std). If the %RSD for peak area (and usually retention time) is lower than laid out in the method/protocol (usually 2% or 5%) then we can proceed with the analysis.

Our main runs usually consist of mixed unknown and QC samples surrounded by two calibration curves. We then use the mean response of the two curves (corrected to any internal standard) and a 1/x weighting regression formula to generate final data for the unknown and QC samples.

One question that has come up though is what is the appropriate response to a failed system suitability? i.e. should an investigation into why it failed be carried out, or is it acceptable to just keep running the system suitability until it passes?

[KarenJ]

Hi,

I've just been reading the USP definition of system suitability because I have some questions about it myself. The discussion about 5 replicates for RSD < 2 and 6 replicates for RSD > 2 is of course correct, but it was always my understanding that using 5 replicates was preferred because HPLC assays should be able to meet that criteria. I was always told that there needed to be justification if you went the 6 rep. route because it is less stringent. It also says in the USP that those 5 reps can be made all together or interspersed throughout the run.

My question about system suitability is in regards to the theoretical plate criteria. If you only have a single peak in your run so that you don't have a resolution measurement, do you have to include a theoretical plate criteria? If so, how is that criteria usually set? What about tailing? Does everyone always have tailing criteria? How is that typically set?

One more question about theoretical plates and system suit. How critical is it that the theoretical plates be the same between two instruments or two labs? For those who do this, what does "the same" mean?

Thanks,

KarenJ

[DR]

Joke? In a regulated lab, that joke is about like yelling "Hi, Jack" as you board an airplane.

To Karen - the 5/6 issue comes up as a function of the type of assay you're doing. For trace level impurities, initial samples for dissolution profiles - anything where you expect the peak areas to be pretty small, it's OK to use 6, >2% RSD. For uniformity assays, API assays etc., 5, <2% RSD is the rule of thumb.

Tailing, plate and resolution requirements are a function of the particular monograph you're looking at. They are generally whatever the original method sponsor put in. Some prefer taililng, others like plate counts. As long as you 1) meet the method criteria and 2) have no reason to suspect that things are going South (based on your previous experience with a given method, you should be able to tell if something is 'off') - you're good to go.

Tailing criteria are generally set to some window approximating what was seen during development and validation of a method. I have seen some really bad ones.

Plate counts will vary as a function of mobile phase composition, instrument used (differences in dwell vol.) and, to a lesser extent, lot-to-lot variation of healthy columns. I'd venture that "same" is probably ±10%.