At 2001-12-13 12:07 +0000, David Halsall wrote:
>I'm in the process of writing up a PCR based assay for detecting exon
>deletions. The outcomes are either 0,1 or 2 copies per genome. Obviously
>the assay needs to be able to detect 1 copy per genome from 0 (Limit of
>detection?) but also needs to have a sharp enough dose-response curve to
>tell 1 and 2 copies appart. What would be the safest use of the 's' word
>in this situation? What stats do I need to collect to define assay
>performance in this respect? Presumably this question was answered in the
>recent thread on this topic, can anyone refresh me?
Roger is right, this is not a matter of sensitivity. The performance should
be evaluated in the same way as other methods giving whole-number answers
e.g. dipsticks.
If 0 is true the anwers 1 and 2 are false.
If 1 is true, the answers 0 and 2 are false.
If 2 is true, the answers 0 and 1 are false.
As far as I know there is no commonly accepted name for this performance
parameter, let us therefore call it "reliability" (r)
r = (number of true answers)/(total number of answers)
As said before I prefer to use the complement, i.e. the "irreliability" (ir)
ir = (number of false answers)/(total number of answers)
In a method like this any result but the correct is false. For a dipstick
one unit from the correct may be acceptable. Then all results deviating
more than one unit from the correct are false.
The (ir)reliability may be different at 0, 1 and 2 copies and in such a
case you must declare it for each of these steps.
Mr Sten Öhman, PhD
|