3

I'm fairly new to statistics and could use some advice.

I have 70 blood serum samples that have had metabolite concentrations tested twice. (Same blood draw, different test dates). I calculated the average, the standard deviation and the coefficient of variation as a percentage for each sample's two measurements. Then I calculated the average %CV of all 70 samples. The average %CV is less than 10%, which according to what I've read is acceptable, but some of the individual %CV's are much higher.

In this case you would expect to see some variation between tests, but a large difference would be indicative of some error in handling the specimen. I'm trying to determine a cut off value over which to reject the sample. The %CV's are not normally distributed (according to Shapiro-Wilks) or else I would say + 2 std.dev. Is there a mathematical way to determine an appropriate cut off?

Thanks!

Bart
  • 320
  • 2
  • 8
  • There actually are environmental regulations and guidance prescribing laboratory data quality objectives in these terms, so I would expect comparable regulations to apply to medical laboratories. What kinds of QA/QC procedures does your lab follow and what is its guidance about this? – whuber Jan 07 '15 at 23:27
  • The tests were preformed at an outside lab. I've emailed them for their input, but I wanted to understand the theory behind the cutoffs rather than just be told what they are. – Bart Jan 08 '15 at 15:04
  • 3
    As far as regulations go, there's no statistical theory at all behind it: the cutoffs are based on the accuracy the agencies think they need. – whuber Jan 08 '15 at 15:15

0 Answers0