RE: Applicability of SPC to Intra and Inter Assay controls

["Jonathan Siegel" <jmsiegel@yahoo.com>]

Shawn,

I think you're right. Running a control chart on the controls ("known"
samples) makes perfect sense to me (with a separate chart for each assay
method and including each operator, equipment, etc.). The 3-5 controls
you use each run would be a perfectly good number to make a logical
subgroup for the control chart. If you wanted to have one point per day
and have many runs per day, you might want to pick the control set to
use as that day's control chart sample randomly from among all the
control samples for each assay type that day. This would address
fluctuations in the method over months and years.

You might well want to have a more aggressive charting regime than one
point per day, perhaps a chart per assay type with one point for each
individual control sample set, in order run. This would be particularly
helpful if you want more real-time feedback on whether conditions are
changing and want to keep a closer look at what is going on than a
once-per-day peek could provide. 

While not applicable to a single assay done in isolation, charting makes
sense once the same assay is done repeatedly over a period of time. I
don't believe there's any way we could know the reliability of an assay
method based only on a single example with 3-5 controls. No statistical
method, traditional or not, can do that. 

For this reason, I believe you are not really using "traditional
statistical methods" to assure the reliability of your method, but are
instead simply relying on the representation of reliability from the
method's originator. Hopefully there is one. Hopefully it is backed by
the method's having been validated on a large number of samples and over
a wide range of conditions that include the ones you're working under.
Hopefully your conditions do not result in variation not accounted for
in the original validation. 

It might be better to know than to hope.

Sincerely,

Jonathan Siegel
jmsiegel@yahoo.com
Senior Statistician (Consulting), Bristol-Myers Squibb
Venturi Partners, Inc.