Chromatography Forum

Could anyone give me their thoughts on setting 'meaningful' system suitability criteria.

As a standard we always cover duplicate standard prep (with agreement) and injection precision. Any other items we add are set if we observe peak shape or resolution issues during robustness analysis. In other words if we pick up an issue duirng development/validation then we put something in to try and pick it up. If everything is fine then we typically would not put anything in.

What we don't do is stick numbers in for the sake of it, i.e. if the method development/validation generates an efficiency of 4000 to 5000 we don't stick a criteria of say 3000 as we haven't proved that the method is still going to work down at this value, and on the other side it may work fine with an efficiency of 2000!

Hi

I do not think there is an answer that fits all but page 34-35 of this pdf might be a start: http://www.lcresources.com/resources/ex ... extras.pdf

Depending on industry etc different approaches may apply. For instance in the pharma buissness we are more or less forced to use a couple of SSTs typically those recommended by FDA and mandatory in general chapters in USP/EP.

Then we have the purpose of the method (assay, impurities, traces etc), somewhat different SSTs may be used, for instance a LOD or S/N check usually is not relevant to an assay analysis.

As you touched on alreay, the robustness part of the validation but also throughout the whole validation SSTs should be kept in mind so one can extract what is really critical.

Length of analytical sequence (number of injections) is another thing, FDA likes when you bracket standards in a longer assay run, while in a longer impurity run resolution/signal to noise/LOD may be more intresting/critical to recheck in the end.

Also some technical aspects may aspects may be worthwile to keep in mind. For instance the general pharma requirement for LC with regard to RSD is not more than 2,0% but most/many applications/instruments today usually can stay under 1,0% without problem.

On the otherhand there is no point to set too harch SST demand unless the validation points in that direction, because you want some flexibility (column choices, shift of intruments brands, transfer to other laboratories, normal variation etc).

Personally I belive efficiency (N) is a good column/system state indicator together with tailing factor, but N limit does not always has to be set as tight as you stated.

Generally I would start with the following:

Blank evaluation (not to forget)
Precision
N and Tailing factor

Then add others as necessary depending on application/industry:
Resolution
LOD/signal to noise ratio
k´ in some cases
Column dead time (gradients)
correlation factor if standard curve is used
might be others aswell.

I hope it helped a bit

"What we don't do is stick numbers in for the sake of it, i.e. if the method development/validation generates an efficiency of 4000 to 5000 we don't stick a criteria of say 3000 as we haven't proved that the method is still going to work down at this value, and on the other side it may work fine with an efficiency of 2000!"

so you validated method, and it is valid when there are 4000 to 5000 plates, this is SST criterion, you don't know if it will work at 2000 cos it was not validated in that range

"Then add others as necessary depending on application/industry:
Resolution
LOD/signal to noise ratio
k´ in some cases
Column dead time (gradients)
correlation factor if standard curve is used
might be others aswell. "

resolution - when needed, can be peak to valley etc.
LOD/s/n - for impuities
gradient delay - for gradient methods
retention time

Think of it this way: what could go wrong, and what would be the symptoms? Then set your SST accordingly. For example:

-- mobile phase incorrect, or column chemistry bad --> retention windows for key peaks; resolution between critical pair(s).

-- improperly assembled fittings or column head space --> resolution between critical pairs; tailing factor; plate number

-- poor detectability for trace components --> baseline noise

and so on. What you want to have is assurance that if you meet SS, your method will work OK.