How to determine the limit of detection?

[Masspecs]

Hello. I am developing an HPLC method for the quantitation of a substance at a particular wavelength. I had no idea as to the limit of detection, as this is a new product, so I ran 5 calibration standards: 0 ppm (solvent blank), 62.5 ppm, 125 ppm, 250 ppm, and 500 ppm. Using the calibration curve from the responses, and the formulae LOD = 3*[(std deviation)/(slope)] & LOQ = 10*[(std deviation)/(slope)], I found my LOD and LOQ to be 102 and 340 ppm, respectively.

These seem to be exorbitantly high, compared with the smallest non-blank standard of 62.5, which gave me a well-behaved chromatogram. My boss wants to know why the LOD and LOQ are what they are compared to the 62.5 ppm standard, and I have no answers for him. I don't know if I'm doing something wrong or if I just lack the statistical knowledge to explain why my work is correct (if it is). The software I'm using is old and doesn't offer any help in the way of automatic LOD determination. With this information, can someone tell me if I'm OK with those LOD and LOQ numbers, or did I mess up somewhere?

[James_Ball]

Standard deviation is the key to why the numbers are so large. If the SD is too high because the numbers are not reproducible enough then the LOD and LOQ will be high by this type of calculation.

What concentration would give you a signal to noise ratio of the analyte peak that is in the 3-5 range? That would be the instrument detection limit. Make a calibration curve that starts at that level then make a set of seven LOD checks at double the low standard and then take the SD of those and see what values you have for LOD and LOQ. Spiking too high can cause problems just as spiking too low can.

[Masspecs]

Standard deviation is the key to why the numbers are so large. If the SD is too high because the numbers are not reproducible enough then the LOD and LOQ will be high by this type of calculation.

What concentration would give you a signal to noise ratio of the analyte peak that is in the 3-5 range? That would be the instrument detection limit. Make a calibration curve that starts at that level then make a set of seven LOD checks at double the low standard and then take the SD of those and see what values you have for LOD and LOQ. Spiking too high can cause problems just as spiking too low can.

How can I determine the noise for the s/n ratio if the blank doesn't give any response for the analyte?

[James_Ball]

I inject the low standard then look at the height of the noise before the peak versus the actual peak height as shown in this post

viewtopic.php?t=20977

You estimate from the standard you run, say you current low standard, to what would give you the proper signal to noise ratio, then make a standard near that level and inject and see what you get. Then you can make a more accurate estimate and inject again until you see the proper s/n ratio. It has to be based on injections of analyte, not from blanks.

[lmh]

I assume your calibration curve is linear? If, in fact, it's curvy, then the deviations between the points and a straight line put through them will show up as "errors" and give you a large standard deviation, which will mess up your LOD/LOQ calculation when using the method you describe.

You can also assess whether the standard deviation is genuine or an artefact of a curvy curve by running one standard (of the right sort of concentration, somewhere in the LOD-LOQ range) three or four times (or more if you can) to check how reproducible a single value is.

If your LOD calculation is correct in this case, you will probably be struggling to see your 62.5ppm peak; if you can see it easily and integrate it easily, then it is a clue that the LOD could be wrong. If you're having to integrate it manually with the eye of faith, then the LOD is probably telling the truth.