Re: [IP] FDA *DOES NOT* require 20% (LONG!!!)
Jan quoted and asked:
> James Handsfield, of the CDC reports: "Under the Clinical Laboratories
> Improvement Amendments of 1988 (CLIA '88) all home test meters must
> meet the same minimum accuracy and precision requirements. Their
> tests must be within 5% of a laboratory standard 50% of the time. That
> may be approximated as +/- 20%.
> So, if it is 50% of the time, what about the other 50%?
My reply:
Per my long post: This is an OVER-SIMPLIFICATION, because *NO* meter is
anywhere near this accurate at low bGs. The mfrs and the FDA all
understand this issue very well. And so, the FDA has LOOSER standards
for accuracy on readings taken at low "actual" bG. They also want the
supporting study data to actually INCLUDE a significant number of
readings in the Hypo and Hyper ranges. If the meter under test is
"frequently" reporting bGs which would lead to mis-treatment or
erroneous non-treatment, then the FDA will be VERY concerned. And the
amount of concern varies according to the FDA-perceived risk of
Fatality/Serious Injury due to that mistreatment/nonTreatment protocol
error.
Handsfield's quote says nothing about the bG testing range which the FDA
wants to see, or the wider error allowance which they allow at lower
bGs. If you ask him, I am *SURE* that he will agree that the FDA *DOES*
allow a wider error allowance, and that there *ARE* concerns about the
range of "actual" (i.e., YSI or equivalent analyzer) bG ranges tested. A
meter company which submitted a new meter with average error less than
3%, but with data taken only at YSI glucose between 100mg/dL and
150mg/dL, would be thrown out of the office and told to "give us numbers
we can use" almost immediately.
- - - - -
Now for your question: The other 50% of readings WILL differ from the
YSI reading of the same sample by 5% or more of the YSI value. Since a
"sampling" process is being used, the range of possible readings for any
given sample is almost infinitely wide: Even a very good bG meter, if
you took ENOUGH readings, would eventually show a really bad "350" value
when YSI-measured bG is "90". The odds are very close to zero, but the
"bell-shaped curve" probability never actually reaches zero chance on
ANY finite "reading" value. (bG is a Poisson distribution, not a
"bell-shaped curve", but statisticians use Gaussian math as an
approximation in most circumstances. The meters themselves are
programmed to give a "LOW" or "HIGH", instead of a number, when the
"reading" occurs in a REALLY low-probability part of the distribution--
you'll see "HIGH" instead of "562". And don't forget, the YSI is also
doing it's measurement based on a sample, so even the "actual" bG being
used as reference really consists of a "tighter" curve, but still a
curve, of ERRORS surrounding the "true" value.
On a Gaussian Distribution, a "bell-shaped curve" where there's a 50/50
chance of being within 5% of the "true" value will have about an 80%
chance of being within one Standard Deviation, that's what he was saying
when he said "That may be approximated as +/- 20%." That's just how
these "bell-shaped" normal distribution curves are defined by the
underlying math.
- - - - -
Most meters have errors of a "consistent" type in comparison with YSI,
this where the "accuracy" versus "precision" differential comes up. A
meter's agreement with ITSELF. for multiple readings using the same
sample, is "precision". The meter's agreement with YSI (or other)
assumed-to-be-"true" value is "accuracy". My meter, for example, has
"precision" of 3.2% ("Coefficient of Variance") when you're using strips
from the bottles in the same "Production Run" (bottles with the same
calibration code, manufactured at the same time). But that precision
widens to 4.4% CV when you use strips from several different bottles
created on different days.
And, it's also an example of my MAIN POINT: the 3.2%
Within-Production-Run, 4.4% Across-Runs figures are for bG "samples"
with average bG of 44-45. If you look instead at at samples with average
bG of 364-366, the error percentage falls to just 1.6% CV within-run,
and 2.4% Across-run. Almost TWICE as good-looking! Accuracy AND
precision, when expressed as "% of error versus YSI", is DRAMATICALLY
HIGHER at low bG levels. The FDA *DOES* consider this.
My strips, One-Touch Ultra, present a much more complete and useful
"Performance Characteristics" Patient Information package insert sheet
than most others. Most other strips, if they gave FULL data like this,
would be readily seen to be quite a bit worse. That's why they conceal
the details, preferring to show just one or two figures which were
heavily dominated by lots of "Normal" and "High" measurements, where
it's EASY to get low error percentages.
- - - - -
hmmm, I seem to have gone from LONG!!! to EVEN LONGER!!!!!!
.
----------------------------------------------------------
for HELP or to subscribe/unsubscribe/change list versions,
contact: HELP@insulin-pumpers.org
