A follow up to "unknown samples calculation"

[lmh]

"unknown samples calculation" has raised the issue of QC samples and how you choose the limits.

I'm trying to sort out an approach to setting QC limits for our lab. We aren't expected to work to ISO9001, but I personally feel we've been more lax than we ought, and I'm tightening up. Nearly all our work is using self-developed methods, or methods taken from academic literature, where you're lucky if the method is recorded thoroughly enough to repeat, let alone containing any QA information.

We need a balance between being professionally-careful and overdoing things. The approach I'm going for is that during method development, we take blank extracts spiked with standard at levels towards the low end of our quantification range, and find the s.d. of the measurements. We then multiply this by about 3 or 4, and use this range as our "pass" range, on the grounds that 97% of measurements should pass if we used a factor of 2, and 4 gives us a little more safety, especially in the case that we do several QC's and a fail of any one is a fail for the entire sequence (endless false failures aren't going to help). For more critical methods I would prefer to use medium and high concentration checks too, but realistically there's a time issue, and we may have to stick to one QC in a region most likely to spot a problem.

Does this sound stupid/over-optimistic/over-lax? What do "real" labs do?

[Don_Hilton]

I will suggest taking a look at http://www.epa.gov/epawaste/hazard/test ... /8000b.pdf. Every method needs QC to be sure that results are suitable for the inteded task. The EPA methods have spcific objecitves, which may differ from yours.

You will note a number of kinds of checks on the quality of data being generated by the methods. The 2 sigma and 3 sigma limits noted in section 8 are typical for control charting - and are widely used.

See if that link will give you some ideas.

[grzesiek]

Hi lmh

Our lab was very unwilling to do any kind of control / quality assurance and that's one of the reasons i no longer work there

In my opinion what you do is quite good but as always there is a question of statistical significance vs practical significance
Imprecise system can make you accept the result even when the difference between measured and true value is too big, and very precise reject even if there is no practically important difference

Look at what you get not only from statistical point of view but also from practical point of view

[lmh]

Thanks, both: I am pursuing your link, Don (I need to open it on a PC with newer version of acrobat; currently the equations come out wrong). I've also read further in the US DoD manual to which someone provided a link in another thread; concrete examples are so much more helpful than the ICH's general overview. It sounds like I should be aiming to get things within 3 s.d. at most, and routinely checking throughout my concentration range.

It is a huge amount of work to move a lab from generic SOPs that cover health and safety but are skimpy on everything else, towards a proper approach to method validation and data handling. There is inertia... I appreciate this forum and its contributors.