Chromatography Forum

An FDA inspector told a contract manufacturer of OTC products that fve injections of standard were required each day prior to injecting samples. Is that what you all understand, and do? Seems to be contrary to the reproduciility of today's automated systems, and definitely is non-environmental...

Regarding system suitability, the ICH guidelines refer you to the Pharmacopoeia for additional info.The USP suggests 5 replicate injections when the limit is 2.0% or less or 6 replicates when the limit is greater than 2.0%. It also states that the injections may be performed before injecting the samples or injected throughout the run but that system suitability must be confirmed throughout the run by injections. The CDER only recomends an RSD of less than 1% for n=5 or greater.

Any system suitability test has to be appropriate for the analysis. We used to perform 6 replicates of the middle standard before assays. Note that limit and related substance tests don't have to meet that requirement.

For quantitative assays, 5 replicates is common, as it's often the minimum number required to achieve the specified precision data.

I'm not sure what the USP says, but the EP lists the following requirements.

Maximal permitted relative standard deviation for replicate injections of the prescribed reference solution do not exceed the values given in Table 2.2.46.-1. This requirement is applicable to assays for content only and does not apply to the test for related substances.

Table 2.2.46.-1. — Repeatability requirements
-------------Number of individual injections
% B----- 3, 4, 5, 6,
-------------Maximal permitted relative standard deviation
2.0----- 0.41, 0.59, 0.73, 0.85.
2.5----- 0.52, 0.74, 0.92, 1.06.
3.0----- 0.62, 0.89, 1.10, 1.27.

Yes, five injections seem a waste, but how else can you be assured that the system is suitable before you start an analytical sequence?.

Please keep having fun,

Bruce Hamilton

Of course the FDA is allowed to change their policies but usually they give notice of those changes.

The FDA addressed this question back when they were publishing the Human Drug CGMP Notes. It was in the September 1997 issue:

http://www.fda.gov/cder/hdn/cnotes97.pdf

Look at pages 1 and 5 of this document.

Basically, the FDA decided that only 3 of the standard injections needed to be at the begininning of the run before any sample injections and that the other 2 (or 3) injections can be made elsewhere in the run, even at the end after all samples have been injected.

However, I have not seen anyone who did it that way. Everyone seems to put the 5 (or 6) injections consecutively at the begininning of the run.

Yes, 1997 was a while ago, but I am not aware that the FDA has reversed/changed this position (possible, but I just haven't seen it). Perhaps they now have and that is why they are telling you. It would be nice if they published a revision.

Regards,
Dan

Bruce I don't understand the table maybe it's the way it's formatted.

While I always run 6 replicates on the first run of a series of runs with an S1 to S2 comparison, it seems reasonable on subsequent runs to use the RSD of the Stds throughout the run (criteria RSD<2%) which is a more stringent test.

^ but if you use nested bracketing, you may actually find (on bad days) that you have some sample groups bracketed by 5 or more standards that meet suitability criteria while other brackets don't. If you limit yourself to a grand average's RSD requirement, you'll have to rerun everything every time a run has a bad portion.

The post by carnie reflects my practical understanding of the way it is supposed to work. (That said, we usually do 5 up front, at least one at the end and insert a standard after every 5-12 samples, depending on the type of assay - I think 3 up front and 2 at the end would be the bare minimum you could hope to get away with.)

CPG - what's your source? Was the statement about 5 replicates in a 483, or was it just something the inspector said?

Our procedure has been to make the 5 (or 6) injections of the calibration standard up front followed by an injection of the check standard. There may also be injections for a resolution solution and a detection sensitivity solution up front. The system suitability (SS) is based on these calibration standards with criteria for RSD, tailing, and resolution (which could be determined in the resolution solution if used). A standard recovery is calculated for the comparison of the two standards. There may also be a detection sensitivity criteria depending upon the method.

In addition to that, we have injections of up to 6 samples bracketed by injections of the calibration standard for the rest of the run with an additional criteria for the RSD of those standards.

As DR pointed out, you have two options for calibration: use all standards throughout the run or use bracketing standards.

With the all standards approach, you would have an overall RSD. And I agree with DR's comment that you may have to throw out an entire run when there is a bad portion within the run. Alternatively, there is a precaution here in that a bad portion of the run can be masked by the use of the overall RSD as a long injection sequence could result in a lower calculated RSD value.

With the bracketing standard approach, you could use a cummulative RSD, calculating a new RSD value for each successive standard injection. This approach does allow for you keep the good portions of a run and invalidate the bad portions.

Our SOP allows for either the bracketing or overall standard approach.

I prefer the use of bracketing standards.

Regards,
Dan

Sorry, the formatting is bad.

The number of injections permitted ranges from 3 - 6, and the maximal permitted RSD should align with the associated number.

I should have defined B%. It's the upper limit of the Assay minus 100, eg a compound with 98-102% specification would have a %B of 2.0, and a compound with 97 - 103 % assay specification would have %B of 3.0.

You chose the appropriate line according to the assay tolerance, and can select also the number of injections to use.

If the monograph does not specify the number of injections and the acceptable RDS, then the above table is used. Some of the monographs specify 6 injections along with acceptable RSD.

Please keep having fun,

Bruce Hamilton

With the bracketing standard approach, you could use a cummulative RSD, calculating a new RSD value for each successive standard injection. This approach does allow for you keep the good portions of a run and invalidate the bad portions.

Our SOP allows for either the bracketing or overall standard approach.

I prefer the use of bracketing standards.

Regards,
Dan

^Unless you have a really lengthy run... If you're 6th standard is 5-10% off the mark, it would be flagged, but if your 26th standard is that far out, it becomes a systematic problem - that a single bad standard is less likely to trigger a suitability fault if it is injected later in the run (not good).

Nested brackets prevent this scenario from coming to pass while allowing for gradual changes in run conditions and results over time (slow changes in RT or areas due to inherent standard instability or temperature changes in the room, for example) to be appropriately ignored. Imagine a lengthy run (100+ samples x20 min ea., many standards, replicate injections due to client not being held enough as a baby...). It could be the case where you would fail an overall RSD requirement of 1% or 2% while having no problem passing RSD for individual brackets.

If you really like reinjecting your runs, write your methods and SOPs so that both grand average and brackets have to meet the same criteria, no matter the length of the injection sequence. [/how not to do it]

I doubt some of my previous client's requests could be attributed to a lack of childhood TLC, more likely one parent held a flaming fork, was red in the face, and had horns...

Issues with long run drift can also generate concern about samples, especially if they originate from process investigations/devlopment, so I tend to choose at least one sample, and reanalyse it every 20 - 30 samples.

The method validation should have sorted out all regulatory quality issues, so I don't use such samples as part of any system suitability process. They are what I call "no surprises" data solely for the analyst.

The biggest issue is that Quality people want to start including some control parameters onto that data, so they have to be actively discouraged.

My argument is that Quality signed off on the protocol, and they get the data. If any "no surprises" data shows problems, then an investigation is performed, but it's not initially "out of specification".

Bruce Hamilton