Quality, why 15%?

[dp_kloos]

I was wondering about the following; bioanalytical guidelines prescribe that a deviation of 15% for accuracy and/or precision is allowed. But on what grounds was this decided? Why 15%? Was it purely a consensus made in Crystal City in the 90s, or is there more to it?

Also for a control sample at LLOQ, allowed deviation is 20%, again, on what basis was this decided?

Is there any statistical reasoning behind this? Seems that this is a broadly accepted guideline, but nobody knows why we do this..

[lmh]

A lot of figures are plucked from thin air, but there can be a sensible thought-pattern. There is no point, for example, in insisting on 2% measurement precision if the underlying thing you're measuring has a biological variability between "identical" samples of 30%. I'd guess the reasoning behind the 15% value is based on the biological variability of the intended samples. It'd be nice to know though. Very often, SOPs and rules come without any indication of Why.

[Steve Reimer]

Many times that 15% or 20% is for reasons of consistency. There is no absolute dividing line so a consistent, easy to remember number is selected. It is not really out of thin air, but it is often arbitrary.

[Peter Apps]

As with everything in life, it's a compromise. A larger imprecision/accuracy would reduce how useful the measurement result was for decisions about quality, compliance with legislation, or safety. A smaller imprecision would push up costs - as a matter of course metrology labs achieve accuracy and precision ten times better than routine labs, but at probably 100 times or more the cost per sample.

Peter

[James_Ball]

If you take a standard GC/MS and inject the same standard 20 times, you will probably find that the result will deviate 5% on a good day, and more if the sample has a complex matrix. If you set your limit to the absolute best the instrument can achieve, you will probably be limited to analytical runs consisting of a calibration curve, calibration verification standard and maybe one or two samples before you must start over if the matrix is more complex than DI water and clean extraction solvents.