by
lmh » Tue Dec 12, 2017 4:51 pm
Viewing this from a logical, rather than a regulatory perspective:
The LOQ is chosen at a point where your measurements have the largest acceptable random statistical error. Let's assume the LOQ is 1 pmole and the error is 10%. If your sample contains exactly 1pmole, and you measure it twice, it's quite natural that you'll get values in the region 0.9-1.1, and it's just as likely that the value will be less than the LOQ as it that it will be bigger.
If you ignore the smaller value (because you decided you can't quantify it reliably enough) then you bias the result in an upward direction. This may be a safe, cautious thing to do (or not; it depends entirely on the context of the measurement).
If you take the average of both measurements, regardless of their value, then the new estimate should be slightly more reliable than the old estimate, because it's an average, which always reduces the error (you're now looking at a standard error of two measurements; if their standard deviation was 1, the new standard error is 1/root-2).
So the question is what to do with this average? If it's bigger than your LOQ, you're probably happy. But what if it's smaller?
Well, originally in your LOQ you decided you would trust anything that you could measure to +/- 10% or better, and since the new, average value is an average, it is more accurate than the individual measurements - so it might have an error better than 10%.
So the answer is: your LOQ applies to the whole method you use to make measurements, not to individual injections. If you routinely report the amount of X in a single individual based on multiple chromatography runs, you'd be justified (morally, ethically, and scientifically, but not necessarily legally - that's not my domain) to calculate your LOQ based on the same replication, because what matters to you is entirely the accuracy of the result you report, not the accuracy of any intermediate steps in the calculation.
And, critical point: when people start worrying about the behaviour of their samples on a threshold value, I start worrying about whether the method is adequate. If it matters to you whether your sample contains 1 pmole or 0.9pmoles, then you need a method with a LOQ substantially lower than 1pmole. Life should be arranged so that no important contaminant that needs to be quantified is ever anywhere near as low as the LOQ, and no contaminant that is near the LOQ could possibly be considered concentrated enough to be relevant. Measurement Threshold values should be well clear of Decision-point Threshold values.