segnal to noise

[pcocco]

Dear all, I have a question about signal to noise in LC/MS.
Which is the criteria of appraisal of SN?
For example 1000RMS is a good value for LLOQ?
Thank's

[tom jupille]

For *any* technique, the usual approximations are:

LOD (Limit of detection) is S/N > 3
LLOQ (Lower limit of quantitation) is S/N > 10

The noise in question is peak-to-peak noise (which is about 3X the RMS noise), so multiply those numbers by 3 is you're measuring RMS.

Note that the above are only approximations. They must be verified by running standards at the expected levels.

[adam]

Tom

I'm just curious: what approach do you generally recommend to verify LOD and LOQ. In our lab we typically prepare a standard at the calculated concentrations; then run it several times and determine an RSD. Typical acceptance criteria would be:

- 8% RSD for LOQ and 25% RSD for LOD

Strictly speaking I don't think this really "verifies" it; I think it just gives it somewhat more credibility. There is no way that I am aware of arriving at a "sacred" LOD or LOQ. I think one can only apply the standard formulas (and hope that the insrument software - which calculates the noise for you - is doing it's job right) and the result is what it is. Another lab...another software package...etc...and you get a diffrerent result.

[tom jupille]

You're right, it's a "can of worms".

My take on it from a scientific point of view (not regulatory) is that the best approach involves the following:
1. Specify what level of repeatability constitutes LOQ and LOD. This is a more contentious issue with LOQ than with LOD, but your values of 8% and 25% seem reasonable to me.
2. Run replicate injections at various levels and generate a "CV vs A" plot (usually log-log, plotting the CV as a function of the amount of analyte). Such a plot will usually be flat over most of the range (dominant errors due mainly to things like sample prep and injection) and turn upward near the bottom (dominant errors due mainly to integration and noise). Identify where the plot crosses your target CVs for LOQ and LOD.
3. Run replicates at those levels to verify that the CV's meet the desired criteria.

From a regulatory perspective, you can use either 10 and 3 for target S/N levels (and I agree that these should be viewed as estimates, not as sacred values!) or better, from 10 and 3.3 times the standard error of the y-intercept. In effect, this uses the standard error of the y-intercept as a surrogate for "noise", but a surrogate that is (hopefully) less biased.

Finally, LOQ and LOD are not fixed characteristics of a method (although we tend to treat them that way ). As you pointed out, they are dynamic and can vary tremendously from day to day, based on the noise characteristics of our system and how the data system deals with it. That's why verification by running standards at those levels is important.

[Kwet]

I agree with the above discussion focused on LOQ determination with standards. Although, It's not sufficient to validate the LOQ only with standards analysis especially for complex matrixes (like blood or vegetables for example). You should also realize spiked matrix samples at LOQ level (with a matrix that does not contain the analyte to determine). Evalute the S/N ratio on these samples is more realistic than with standards. You can indeed have more noise in your real sample matrix than in the standards. You can also have matrix effect (especially with MS detection) or looses during sample preparation / purification which change signal of analyte.

As I'm working in pesticide residue analysis in respect with European legislation, I don't really known what are the rules in US pharmaceutical or other legislation for that point.