Back to the Top
Dear All:
I haven’t heard much about this issue in recent days, and I have been wondering what you
all are thinking beside why doesn’t Jelliffe shut up. Ed, you have mentioned accuracy as being a
separate issue from precision, and I had asked if we could start this time with a definition of what
accuracy is. Can you help us?
All the best,
Roger
Roger W. Jelliffe, M.D., F.C.P., F.A.A.C.S.
Professor of Medicine Emeritus,
USC School of Medicine
Founder and Director Emeritus
Laboratory of Applied Pharmacokinetics and Bioinformatics
Consultant in Infectious Diseases,
Children’s Hospital of Los Angeles
4640 Hollywood Blvd, MS #30
Los Angeles CA 90027
Back to the Top
Accuracy is low bias. Precision is low variance (standard error). The 2x2 matrix is illustrated well
here [http://celebrating200years.noaa.gov/magazine/tct/tct_side1.html].
Back to the Top
From: Roger Jelliffe
OK Accuracy is low bias. What is that, precisely? Is it mean error?
Very best regards,
Roger
--
From: Edward O'Connor
Sure Roger, and no one wants anyone to shut up! The recommended measurement of accuracy is
closeness of an observed value to the expected value or nominal. This closeness is measured as
((observed value-expected value)/expected value) X 100. The tolerance is suggested to have limits
of 20% for LC-MS methods and 25% for ligand binding methods. There can be some relaxing of
tolerance based on agreement with FDA.
--
From: stefan.soback.-at-.gmail.com
Hi everybody,
The example of John Tillinghast is useful. In our case we can only work on the sight of the gun.
Changing position would mean for us changing the analytical instrumentation ( eg. from HPLC to
LC/MS/MS).
The analogy is that what we actually increase the distance to the target. The accuracy stays the
same, but there is a bigger spread on the target. Also the precision is the same if we correct it
with distance. I think the situation we are looking at here is when we can't hit the lower part of
the target at all because of the terrain (no instrument signal).
If we continue to calculate accuracy and precision only using those points that we are able to see
in the upper part of the target (because we don't know how many shots were fired), the precision
might actually look better but the accuracy certainly don't.
So it's not like we can increase the target size when the distance grows to count for every shot,
because we can't lift it up. Simply half of it is underground.
Best,
Stefan
--
From: Philip Kiser
Dear Roger,
Would it be fair to say that accurate data has low systematic error but not necessarily random error
and vice versa for precise data (basically rephrasing John's definition)?
Philip
Back to the Top
For curves, In LC-MS there may be two separate curve run, each point is assessed to nominal, there
may or may not be a "mean" rather the individual points are assessed against nominal. A certain
number may fail. For QCs there are at least two QCs at each of 3 levels. Again, the individual
values are assessed against the nominal. The standard rule is that at least four of the 6 QC must
meet acceptance for BIAS and the run is rejected if this is not met; in addition there cannot be a
failure of both QC at the same level- this run would also be rejected.
Ligand binding assays are similar in acceptance and numbers the difference is that there may be two
or three replicates for each Standard or QC. Again there are at least 3 levels of QC and two
samples at each level. In addition the samples are measured in duplicate. Generally, the CV of the
duplicates ( or triplicates must meet acceptance criteria, if they do then a value is reported and a
test for Bias is run. If the CV fails the QC is rejected and no other tests are run; if the bias
fails the point is rejected. Again at least 4 of the 6 QC must pass and there cannot be the loss of
a level. With curves, there should be at least 6 surviving points including the LLOQ and the ULOQ
and there should not be a loss of two adjacent points.
| PharmPK Discussion List Archive Index page |
Copyright 1995-2014 David W. A. Bourne (david@boomer.org)