You are referring to the MDD or minimum detectable difference of a method or
assay. This value depends on the precision of the assay and a common way to
compute it is MDD = 2*sqrt(2)*SD. The value of 2.8 comes from the product of
2 and the Sqrt 2. I have seen this approach used to compute detection
limits, although it seems not to be commonly used. You're right, the value
of the SD (or CV if you use it)often depends on the concentration. This
makes sense theoretically and often in practice as an assay's ability to
resolve two closely spaced levels is not the same over the dynamic range of
the assay.
See Chapter 9 [Principles of Measurement] in J.K. Taylor's book 'Quality
Assurance of Chemical Methods'. Lewis Publishers, 1987.
Also check out some of the environmental and analytical chemistry literature
as this concept is often used to evaluate Youden analyses of shared samples.
Hope this helps out.
Timothy Foley
New England Newborn Screening Program
Boston, MA USA
[log in to unmask]
-----Original Message-----
From: [log in to unmask] [mailto:[log in to unmask]]
Sent: Thursday, June 17, 1999 4:20 AM
To: [log in to unmask]
Subject: Lowest detectable difference
Dear list members!
The lowest detectable difference for a quantitative method (LDD (%))has been
explained to me to be LDD (%) = 2.8 * total CV (%). This is also used to
define
the limit of detection for a method. My wonder is what the conditions are
for
applying this. Can this be applied also when CV reaches 30 %? CVs in this
range
would than mean that two quantitative values can be discriminated with 95%
confidence when the higher value is twice the lower value. My feeling is
that
the equation does not take in acount the increase in standard deviation of
the
error with higher values, which is inherent in our methods and that the
distribution of the error is likely to have a positive skewness when CV is
high.
I would also appreciate references over this matter.
With kind regards
Goran Brattsand
M.D.
Clinical Chemistry
Umeå University Hospital
S-901 85 Umea
Sweden
e-mail : <[log in to unmask]>
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
|