Limit of detection and Limit of quantification

[DeviSirisha]

Can we consider the standard deviation of precision for calculating LOD and LOQ.

[Hollow]

Probably yes.

See this 2nd article of a 5 part series from John Dolan's LCGC-Troubleshooting articles

http://www.lcresources.com/tsbible/hs2900.htm

[James_Ball]

Probably yes.

See this 2nd article of a 5 part series from John Dolan's LCGC-Troubleshooting articles

http://www.lcresources.com/tsbible/hs2900.htm

Depends on what regulating agency if any the data is reported to. Once the lawyers get involved the science can get thrown out in favor of protocol.

[lmh]

I notice the ICH are currently rewriting Q2(R1) and will be producing a Q2(R2) which will include this sort of topic.

[DeviSirisha]

Probably yes.

See this 2nd article of a 5 part series from John Dolan's LCGC-Troubleshooting articles

http://www.lcresources.com/tsbible/hs2900.htm

In this article it was mentioned to take the standard deviation of calibration curve.
In method validation PRECISION is one parameter, so, can we consider the standard deviation of that precision as sigma value and substitute in the LOD formula.

[Fernando]

Dear DeviSirisa

You can use this formula:

CV= 50 / S/N

Where S/N is the signal to noise ratio

Taken from Prctical HPLC Method Development Snyder et al.

Best regards

Fernando

[tom jupille]

Be careful with that one, Fernando! It is only an approximation, and uses a different measure for S/N from that used by the USP.

In any case, the reason for using the standard error of the y-intercept to evaluate LOD and LLOQ is that it avoids the issues involved with S/N measurements (e.g., the exact definition of S/N, where you measure the noise, over how long a interval you measure noise, etc.)

[Fernando]

Dear Tom

Yes, as always you are right, it is only an aproximation but in certain cases a usefull one. Thank you for your explanation.

Best regards.

Fernando

[Mattias]

From a scientific point of view, I think using the precision and accuracy of the method to define LOD/LOQ is far better than the other methods. Depending what the method will be used for, the acceptable precision/accuracy will be different.

The S/N method is completely unrelated to the actual results of the method.

But that being said, the guidelines seem to favor the S/N measurement and I always include that in my validations. Otherwise I will have to do it anyway when the authorities come back with questions..

[lmh]

I have had difficulties with S/N since discovering that of the three integrators provided with Thermo's Xcalibur some years ago, the two that reported S/N gave values that were orders of magnitude different, for the same chromatogram! Which raises the question of which to use...
Also S/N ratio is very vulnerable if the data have been collected with a signal threshold (if all points near a weak peak have been rounded to zero, then there is no noise, in which case the S/N ratio is infinite. It's also vulnerable in weak data, because weak data can have a very low S/N ratio, but still be very variable, in which case the limit of quantification will be drastically underestimated.
But the method based on the s.d. of the calibration curve is also far from perfect. Errors are harder to measure than means (i.e. the uncertainty of the s.d. (its own RSD) is generally a greater fraction of the s.d. than the s.d. is of the mean).

[Fernando]

From a scientific point of view, I think using the precision and accuracy of the method to define LOD/LOQ is far better than the other methods. Depending what the method will be used for, the acceptable precision/accuracy will be different.

The S/N method is completely unrelated to the actual results of the method.

But that being said, the guidelines seem to favor the S/N measurement and I always include that in my validations. Otherwise I will have to do it anyway when the authorities come back with questions..

Hi Mattias

It is very interesting, how do you define the LOD/LOQ with the precision and accuracy (I think you use them from the assay)?

Greetings

Fernando

[Fernando]

I have had difficulties with S/N since discovering that of the three integrators provided with Thermo's Xcalibur some years ago, the two that reported S/N gave values that were orders of magnitude different, for the same chromatogram! Which raises the question of which to use...
Also S/N ratio is very vulnerable if the data have been collected with a signal threshold (if all points near a weak peak have been rounded to zero, then there is no noise, in which case the S/N ratio is infinite. It's also vulnerable in weak data, because weak data can have a very low S/N ratio, but still be very variable, in which case the limit of quantification will be drastically underestimated.
But the method based on the s.d. of the calibration curve is also far from perfect. Errors are harder to measure than means (i.e. the uncertainty of the s.d. (its own RSD) is generally a greater fraction of the s.d. than the s.d. is of the mean).

Hi lmh

My experience is only with the Chemstation of the HP1100, I run a placebo for say 20 minutes, many times, and use the ASTM method of noise calculation.

Best rregards,

Fernando