RE: Low Positive Doses in Personnel Monitoring

["Sandy Perle" <sandyfl@ix.netcom.com>]

Mike Bielby made some excellent points. The question as to whether to
use an LLD is an interesting one, moreso based on the methodology used
to make these determinations for the various ANSI N13.11-1993
Categories. While the LLD is important, as to what the lower dose that
should be reported, it is sometimes a misleading indicator. The
ability to report a low dose is highly dependent on the variance
within the current system. Some processors may increase the positive
bias of the system to mitigate poor variance. This is a serious
mistake, for evn with a positive bias, there will still be the chance
false positives and false negatives. If you want to conduct good
dosimetry, then significant care is needed to assure minimal variance
within the system. A system with a decent bias but a poor variance is
a very BAD system overall. Without statistical control, which is more
dependent on the variance than the bias, hit or miss dosimetry is the
result.

The issue of faslse positives is one issue. Another important factor 
is the degree of the falseness of the result, i.e., a false positive 
at or just above the LLD is one thing. A false positive at 30 mrem or 
higher is another. In that case, I say that there is a problem with 
the processing system, and all of the quality control checks that are 
conducted, or, there is one serious problem with the specific 
dosimeter in question, looking at the characteristics I posted 
earlier, be it film or TLD. All of the various factors need to be 
considered.

One could have excellent bias and have a very poor system. An example
of this is a system with a 1.00 and a single Std. Dev. of 5%. The same
system could have a bias of 1.00 and a single Std. Dev. of 15%. The
shape of the normal distribution is broader, the system has the same
bias, but the Std. Dev. is way out of whack! Therefore, it is
impossible to determine the % of false positives solely from the p
value of .95. I submit that without statistical control, more serious
problems will occur. And, if a system is within statistical control, a
5% false positive rate should not and would not be acceptable. If you
are getting a 5% false positive, or false negative rate, then the
Quality System has failed, and there needs to be some incremental
improvement, or even major improvement in the system. 

If the mean and Std. Dev. of the unirradiated dosimeters is
below the LLD value, say around 5 +/- 4 mrem (for an LLD of 10 mrem),
the USL of 10 would not be exceeded. I also suspect that for the
unirradiated dosimeters, the shape of the distribution is more
Binomial in characteristics than normally distributed. This would be
expected with a preponderance of the data points at the minimal
threshold of the unirradiated dosimeter. If this is the case, then
there should be very few false positives.

If the system is in statistical control, than 99.97% of all data
points should fall within the statistical specification limits. If
this is assured, than there is only that 0.03% chance that there will
be a data point outside of these limits, and those woild be what I
consider to be the false positives or negatives. The only way to get
to this point is to minimize the variance within the system. 

If there are other comments regarding my understanding of the 
statistics involved, I would be plesed to discuss these fine points.
We all should be willing to learn and improve.

------------------
Sandy Perle
Technical Director
ICN Dosimetry Division
Costa Mesa, CA 92626
Office: (800) 548-5100 x2306 
Fax:    (714) 668-3149
  
mailto:sandyfl@ix.netcom.com
mailto:sperle@icnpharm.com

Personal Homepages:
http://www.geocities.com/CapeCanaveral/1205 (primary)
http://www.netcom.com/~sandyfl/home.html (secondary)

"The object of opening the mind as of opening 
the mouth is to close it again on something solid"
              - G. K. Chesterton -