QC +/- 2SD or 15% ????

[tayzyboy]

We have always ran QCs with a limit of 15% accuracy (nM concentration biological) for several years.
Recently we were looking at an SOP from another centre and they use a +/- 2SD from the mean for acceptance.
Now... could someone explain this to me? I dont want to sound stupid infront of collegues but heres what I thought it means.

I have a set of data which is very good - concentration should be 10nM and I have results as follows.
10.0 , 10.0 , 10.0 , 10.0 and 10.1
This means Mean +/- 2SD is 10.02+/-0.089 and so my 10.1 fails.

If I have a horrible set - concentration should be 10nM and I get
100.0 , 2.0 , 5.0 , 10.0 and 15.0
The Mean+/- 2SD is 33.4+/-82.8 meaning all QCs pass.

I know this cant be correct.... please help anyone

Thanks in advance

[tayzyboy]

ammendment - to get the same values for the second set , the last value should be 50.0 not 15.0

[Mark Tracy]

What you need to do is neither approach. First, decide, based on the intended use of the product what the required performance limits should be. The 15% rule is a typical default decision for this question, but it may or may not be acceptable (read your regulations). Second, you need to decide what is an acceptable defect rate, and calculate how many standard deviations that represents. For instance, 90% yield works out to 2SD. Third, measure the SD of your process. This is part of your validation process. Then compare the performance limits to the SD requirements; if the limits fall within the SD range, you are good to go. If not, you need to tighten your process, or rethink your requirements.

[yangz00g]

85% to 115% recovery of spiked value or SRM /QCS is widely accepted for biological/clinical analysis across the world. When the concentration near the quantification limit, 20% is acceptable.

2/3 SD associated with Z-score is an approach most widely used in assessing PT results from different laboratories or assessing SRM vlaues (eg. NIST uses a similar approach to determine the "true" value of their SRMs). Simply using 2SD is not a technically applicable approach as exampled in your post. You can get more information at www. westgard.com.

[mbulentcakar]

I have a set of data which is very good - concentration should be 10nM and I have results as follows.
10.0 , 10.0 , 10.0 , 10.0 and 10.1
This means Mean +/- 2SD is 10.02+/-0.089 and so my 10.1 fails.

If I have a horrible set - concentration should be 10nM and I get
100.0 , 2.0 , 5.0 , 10.0 and 15.0
The Mean+/- 2SD is 33.4+/-82.8 meaning all QCs pass.

SD of your method is a good way to measure variation. We know from statistics +/- 2SD will contain roughly %95 of your results (with the inherent variation). A value far above this indicates some other error.

As for the first set (10.02+/-0.089 ); if you have a method this good (providing very small variations) than it is better to stick with 2SDs. Probably it will be hard to obtain this much small variations (SD) with biological/clinical samples. I'll advise you to check with realistic examples.
A method which has a SD of your second set example would obviously be not a valid approach for the analysis. So QC check will be meaningless.

As a general rule SD will increase with decreasing concentration of the analyte. For biological assays upto %20 RSD is not uncommon. For some residue analysis it can be upto %30.