method validation / LOD

[Ruth]

Dear,

I am working on a LC-MS/MS method for the analysis of urine and whole blood (forensic samples, so quite dirty). I will start validation of my method, but I am still looking how I best can calculate the LOQ and LOD.

Most people do it by using S/N 3 (10 for LOQ), but given the variability of the composition of forensic samples, the noise around the compound peak also has variability. So this approach doesn't look suited for my method.

Any idea how I can evaluate the LOD for my method in a way that it makes sense for forensic samples?

Thanks for your help!

[JGK]

http://www.fda.gov/downloads/Drugs/Guid ... 070107.pdf

The FDA guidance on bioanalytical method validation does not require a determination of LOD, but does have a requirement for the Lower Limit of Quantification (LLOQ) on page 6.

[lmh]

Try LC-GC Europe Feb 2009 No. 22 pp 82-85, and also Pharmaceutical Chemistry Journal 38:212-225 (Epshtein et al.).

There are some very good methods based on the standard deviation of a low-level calibration curve. They have the advantage of not being reliant on measuring noise, and also not being reliant on comparison to a blank sample (how can you integrate a peak that isn't there??). They are highly respectable, based on good theory, and accepted by ICH etc. (see also ICH Q2(R1)).

[Ruth]

Dear,

Thanks for your suggestion. I looked up the papers you mentioned. The most interesting method to me seems the determination of LOD from the standard deviation of the response and the slope of the calibration curve. I read that a calibration curve in the range of the LOD should be used then. Can anybody specifiy this range more? How many samples should be in this calibration curve?
I was thinking of using the zero samples (only containing IS), the LOQ (= lowest calibrator) and the 2 nd lowest calibrator (since already 6 extracts at each level were performed for the quantitative calibration curve).

I tried this and values look as expected... Is this ok?

Thanks!

[Ruth]

This is another method, which I reasoned myself... I would like to hear your opinion about it...

I am working with postmortem forensic samples, which have a very variable composition which can result in variable recovery, matrix effect (together called process efficiency). For quantification, precision, ... this is solved by the use of an internal standard which behaves the same way as the analytes.
However, internal standardisation doesn't compensate for the sensitivity loss caused by variable process efficiency. The estimation of one LOD-value for all postmortem samples is thus very hard...

Therefore I would estimate the LOD for each sample separately.
An IS is added before extraction. So by comparing the signal for the extracted IS and a pure IS, the process efficiency for a sample can be calculated. Since IS and analyte behave the same way, this process efficiency will be the same for the analyte.
We have established the LOD for the analyte in a pure standard solution. If we apply the calculated process efficiency now on the LOD of the standard solution, is this then the LOD of the extracted sample?

It's just an idea...

Thanks for your help!

[JGK]

On Non-MS HPLC methods I was developing, I would regularly use the estimation of the LOD/LOQ based on the calibration line method.However, I would always follow up with a test using a STD solution at the estimated level.

[lmh]

I think your approach is certainly reasonable. It's not perfect, but very few LOD measurements are going to be perfect!
In an ideal world you would do your LOD on a spiked blank matrix; you would use a matrix that has all the problems and none of the analyte expected in a real sample, and spike it with known amounts to make the low calibration curve for the LOD calculations. The difficulty is presumably that (1) you have no such blank matrix available, and (2) you're concerned that every sample matrix is different.

Running the LOD on a clean standard and scaling it according to the internal standard is certainly seems a reasonable approximation to what's going on in a real sample. My main reservation would be if you have peaks in the real samples that nearly coelute, or background ions/signal that appears in real samples but not in sequences consisting only of clean standards. These could increase the difficulties your integrator will have in quantifying the real peak, and could increase the variability of the real peak's area, which would invalidate your LOD. On the other hand, (a) you will probably be able to spot problems like this very easily by eye, and (b) if you get a sample that is apparently nothing, you can always follow JGK's very good advice and spike it with your expected LOD to see if you can indeed detect your LOD (good practise anyway).

Good luck.

[weber]

Hi Ruth,
we usually do LOD/LLOQ determination with the calibration curve method. We estimate the achivable LLOD out of S/N data of a low spiked matrix calibrator. Than we set up a calibration curve using 6 levels including a blank. It is important to use matrix based calibrators because the performance differs a lot from clean standard samples. We didn't use curves from clean standards for years beacause they do not say to much about the behavior of real samples. Good is also to use a matrix pool for praparation of calibrators.
According to our experience this method for determination of the LLOQ best reflects the reality.
Greetings

[lmh]

weber, have you found a solution to the situation where all available matrix also contains the analyte in unknown quantities? I struggle with this...

[weber]

lmh, it depends.
If your samples concetration is in the same range like your background, in principle you can use the same procedure. The LOD you determine will be worse than the real one. If it is good enough, don't worry. If not you can try to use selected samples with an amount as small as possible.
If this is not usefull, you can use samples with a small amount of analyte, and do replicate injections (6 to 10). From the SD you can then determine the LOD/LLOQ.
For determination of the linearity and working range it is no prblem anyway.

[lmh]

Thanks for the clarification: much appreciated. I'm working in a service lab on (mostly) plant-based samples, and have only very limited influence over the material that's available to me, so any hints like yours are very, very useful. I'm also not an analyst by training, so I appreciate input from those with greater background knowledge.