Simplest and fastest way to estimate LOD from chromatogram.

[along]

Dear all,
I worked with GC since few months ago, I would like your opinion on the way I determine my LOD, please consider the procedure below.

2ul of my standard (0.01ppm, this is the lowest concentration I use to build my calibration curve) is injected into the GC, assuming that the peak is at 7.00 min. With the help of Chem station software, I determine the average noise level from 3-5 min and 8-11 min. The injection is repeated for 3 times to obtain the averge S/N ratio.

The average S/N ratio is 325 and through a simple mathematical calculation, I estimate that my LOD is 0.00009 ppm (Concentration at the S/N ratio =3) and the LOQ is 0.0003 ppm (Concentration at the S/N ratio =3).

Is this fine for me to estimate my LOD and LOQ?
Thanks.

[tom jupille]

Estimate, yes. Evaluate as part of method validation, no.

The problem is that you're basing the estimate on an injection 30X higher than the LLOQ. That leaves a lot of room for weirdness in the calibration plot.

[along]

Thanks Tom, what do you mean by my injection is 30X higher than LLOQ, do you mean if I calculate my LOD and LOQ base on the S/N ration that obtain by injecting much more lower concentration, it will help?

Do you have any ideal how am I going to determine the LOD and LOQ, I try to look for the detail prosedures on how other lab has carried out, but I can't find it, would appreciate if anybody can help me.

[vimal]

Precision study can be done for the estimate of LOD and LOQ u arrived at to prove that the value estimated is the LOD/LOQ.

[tom jupille]

along: LLOD is most commonly defined as S/N=3; LLOQ is most commonly defined as S/N=10. You are estimating these from a peak with S/N=325 (i.e., more than 30X LLOQ and more than 100X LLOD. That is a tremendous extrapolation from your lowest standard. Hence my comment that you can use it as an estimate, but it falls short of what you need to demonstrate for validation. Vimal is right; if LLOD or LLOQ are important, you really need to determine the precision at (or at least near) that level.

[JI2002]

Most people consider S/N=3 as LOD or IDL (instrument detection limit) or even MDL, shouldn't it be S/(standard deviation of blank signal)=3?

Usually it's difficult or even impossible to calculate standard deviation of blank signal, maybe that's why somebody came up with the idea of runing 7 low level standards and calculating the standard deviation out of 7 replicates, then using the result as standard deviation of blank signal. This method is based on the assumption that the standard deviation of the response doesn't change with the concentration of the standard, but that's not the case. Standard deviation gets bigger at high concentration, it also gets bigger at very low concentration.

Another shortcoming of the S/N approach is that it only considers Type 1 error alpha (i.e. predicting false positive) and doesn't consider Type II error Beta (i.e. predicating false negative). If you have a standard corresponding to the signal of 3X the stadard deviation(LOD), you'll have 50% of the chance the response will be below 3X the stadard deviation (predicting false negative).

[vimal]

Hello JI2002,
Do u consider the SD value arising out of 7 replicates of low concentration of standard as signal corresponding to blank.

[JI2002]

Vimal,

Personally, the answer is no, but in practice that's what people do in environmental field.

[JI2002]

One more thing: the standard deviation out of the 7 low level standard is used as standard deviation of blank signal by people in the environmental field.