by

**lmh** » Tue Feb 12, 2019 4:24 pm

I don't know what the regulators would say, but statistically, the right thing is to use all measurements. The reason is this:

All that the LOQ is doing, is telling you the level of result that would have an X% standard deviation. Smaller results will be worse, bigger results better. Therefore if it is important to you that the result is accurate to X%, then you need to have a value bigger than the LOQ.

Imagine someone gives you a sample exactly at the LOQ and you measure it repeatedly. This will give a normal distribution of values centred on the LOQ and with the s.d. that is your target for the LOQ. If you reject all values that are less than the LOQ, and report only those values that are greater than the LOQ, the mean will be greater than the LOQ and the s.d. smaller than the target s.d. for the LOQ. Your results will be biased, and over-precise.

In fact the effect is very small and probably irrelevant, but it's never a good idea, knowingly to come up with a strategy that biases results or artificially influences errors.

Being fussy, it also doesn't make much sense to talk about the LOQ in the context of a result that must be < 0.10%. The decimals imply the result will be reported to a maximum precision of 0.01 on a target value of 0.1. The expected precision therefore cannot exceed 10% regardless of how well the underlying method and instrument have measured, because the instructions are to cut off all precision better than 10%. In effect, the standard deviation of the measurements no longer depends on the instrument or the method; it depends only on the reporting conditions! [In fact if your instrument and method are good enough, then a true underlying result of 0.10% will be reported with near perfect accuracy, while a true underlying result of 0.105% will always be reported with (nearly exactly) a 5% error].

So what you should do is keep all measurements at full precision until you've got the final result, and then report it to the number of decimals specified, sigh about the illogicality of the analytical world, and be happy that your method is much better than it needs to be!