loq based on calibration curve

[laanadrie]

Hello all,
i have a question concerning the loq determination based on a calibration curve with samples in the range of the loq.

what is the lower limit in sample concentration allowed to use for this determination? according to my colleague samples below loq are by definition unreliable, therefore a sample with a concentration at, or just below loq should be the lowest point in the calibration curve used. in my opinion, lower points are allowed (if not needed), since the uncertainty in these points is part of the determination. however, there should be a limit, i would say lod.

it is hard to find clear information on the requirements for this determination. for example: the number of calibration points for the curve, the range of the curve.

additionally: is there preference which σ to use? residual standard deviation or the standard deviation of y-intercepts?

[JGK]

http://www.ich.org/LOB/media/MEDIA417.pdf gives the minimums for validation


When I was validating methods to support pre-clinical studies I would use the ICH method (based on the slope) to estimate the LOQ (it suggests σ of response, I used the low CAL STD). I would then test it with a STD solution at that level. However as we were working to GLP the limit of quantification was set by the range of our STD curve.

For example If our routine calibration STDs were 1, 2, 5, 10, 25 and 50 µg/mL and we had a permitted tolerance of ± 5% on the 1 µg/mL STD then any sample with a measured concentration below 0.95 µg/mL would be automatically classified as BLQ (below the limit of quantification).

[lmh]

I'm not a statistician, but I have to agree, and feel that the ICH document lacks clarity on this, and is open to being misread.

It states that sigma can be measured as the standard deviation of the blank, or estimated from a calibration curve "within the range of QL".

I think your colleague may be reading this as meaning you should only estimate the LOQ from points between the upper and lower limits of quantitation, which then makes the process circular, in that you have to first estimate the lower limit, and then look at the points above the lower limit in order to re-estimate the lower limit.

But logically, if you are allowed to estimate sigma from a blank, or from points around and above the lower limit of quantitation, what is wrong with points between the blank and LLOQ? If the blank is a good estimate of sigma, why should points above the blank but below LLOQ be that much worse than either LLOQ or blank?

So it makes more sense to read ICH's instructions as meaning: use points around the LLOQ to estimate LLOQ, and not points taken from a vastly different area. If my LLOQ comes out at 1ng, but my standards were 100ng and upwards, I ought to do it again!