Back to the Top
Dear All
I found that there is a round off issue for CI in the recent BA/BE
guideline in August 1999. Could you please give me some suggestion? If the
upper limit of my generic products 1.2499 or lower limit is 0.7999. Then;
the conclusion should be BE or not BE. Thank you for your comments.Asst.
Prof. Dr. Korbtham Sathirakul
Mahidol University
Bangkok, Thailand
Back to the Top
The limits are 80 to 125. 79.9999999999999999999 is not 80 and 125.00000000001
is not 125. OK?
Art Straughn
Back to the Top
Although different people have different opinion. First of all, I think it
is a good question. Just how much attention ought we devote to the issue
of valid decimal places?. Imagine u r doing HPLC msrmnt of a chemical,
the computer spits out a zillion numbers including the decimal place,
and you
calculate the AUC, guess what u take 8 numbers regardlessly. NOw you
export it to the excel spreadsheet, guess what, it adds decimal places
just because it is from a different package. Now you give it to the
statistician, who probably uses SAS, he/she ask u, well, how much decimal
place should I use? U said, PICKACHU, how about 2. Given all the
confusion, u come up with 79.999999 and 120.000000000000001, how in the
world can you be so sure that it is not Bioequivalent? OK?!!!
Plus, r u saying if u r one cent richer than Bill Gates, then u r richer?
I guess my question is, was there a rule in taking numbers? I know
in other descipline, there is recognized convention. Please enlighten
me.
Back to the Top
The guidance did not say a value which rounds off to 80 is acceptable. It says
80. If the FDA were to accept 79.51, who is to say they should not accept
79.49. In my courses, a student whose final average is 79.95000 (Excel round
off to 80) will receive a B-, while one whose average is 79.9499999 (round off
to 79) receives a C+. Therefore, my cutoff is really 79.5 not 80. This cutoff
and round off is clearly stated in my course syllabus. The C+ student will
invariably end up in my office with a reason for changing the grade to B-. The
grade debate become much more lively when the "round off" places them closer
and closer to probation (i.e. outside the acceptable limits). I will probably
make a concession to the student. I doubt the FDA would change the grade.
Art Straughn
Back to the Top
[Two replies - db]
From: harry.mager.hm.at.bayer-ag.de
To: " - *PharmPK.at.boomer.org"
Subject: Re: BA/BE draft guidance 1999
Date: Thu, 11 May 2000 09:14:01 +0200
In my understanding, it's just a "rule of thumb", in some sense comparable to
the famous 5% significance level. Of course, 4.45% would work equally well. It
goes without saying that the specified limits nether will be the correct ones,
even when neglecting covariates and interindividual variability. The interval
may be to wide for drug A and to narrow for drug B, and even the whole concept
might be disputable in some cases. Thus, as you mentioned, it seems to be a
convention maybe based on averaged experience.
Harry Mager
---
Date: Thu, 11 May 2000 07:00:53 -0700 (PDT)
From: Richard Molitor
To: PharmPK.-at-.boomer.org
Subject: Re: PharmPK Re: BA/BE draft guidance 1999
From the "real world" of clinical practice here's a thought:
Why split hairs (or fractions) about trying to meet the minimum acceptable
standard? Spend a few more dollars and make the product more like 95-105%
of the reference agent. This would be especially prudent if it's a narrow
therapeutic index agent.
Richard Molitor, R.Ph.
http://www.angelfire.com/wa/pharmacist/
---
From: Peter Gingras
To: "'PharmPK.-at-.boomer.org'"
Subject: RE: PharmPK Re: BA/BE draft guidance 1999
Date: Thu, 11 May 2000 11:22:10 -0400
The limit is 80 to 125%. Please refer to USP 23, p 3, Significant Figures and
Tolerances. While 79.999999 may not be 80, when rounded to the required
tolerance of 80 (ie not 80.000000) it complies with the requirements. USP is
very clear about the acceptability of rounding to the required tolerance.
Peter Gingras
pgingras.aaa.apotex.ca
Back to the Top
The precision of numbers in an assay and in any subsequent application such
as, statistical analysis and PK are generally governed and restricted by the
precision associated with the standard preparation. Reported numbers should
have the same precision (or less). Interim numbers used in calculations may
have more, but should be consistent so that reviewers, QA and the FDA get the
same results when the reported numbers are checked. This consideration is
not trivial. When calcuating CI, the same considerations must be made.
Back to the Top
This is a very interesting and challenging question. From a regulatory
perspective the answer is clear: report the answer with the same number of
decimal places as in the specification. The specification is legal, not
mathematical.
The rules of sig figs and rounding are, amazingly enough, not universally
agreed upon and not well founded mathematically. Sig fis and rounding are
only one expression of uncertainty; statistics gives us others; the latter
are more useful.
If you are interested in this subject, contact the chemistry faculty at the U
of Oregon. They have studied this and publisheed.
Stan Alekman
Back to the Top
I believe the original question was something to the effect, "Is it acceptable
to round your calculated CI such that 79.95% is reported as 80%, and can you
claim your result meets the stated requirement of 80%?"
The answer to this question is yes, you can round if you follow the
rules of USP
General Notices, Significant Figures and Tolerances, p. 3, USP 23. It is not a
vague "rule of thumb" convention, as Harry Mager suggests, and the method for
rounding is very clearly stated.
Ed Connor's message was on the right track. Ed says that rounding is permitted
for legal reasons, which is true. But there is also a scientific reason. An
analytical result is meant to signify a true value, with a stated amount of
uncertainty, based upon known variability of the measurement process, and
conventions for required precision and accuracy. Some numbers that are
calculated are purely artefactual. They do not represent a true measurement
result, but are derived from a silicon chip. One should not make scientific
decisions based upon artefactual numbers.
I somewhat disagree with Richard Molitor's suggestion that we simply test to
tighter limits, to ensure that we comply with the real limits. To some degree
this is done during the sample size calculation. No one wants to spend all that
time and money, and then marginally fail to meet the CI requirement. However, a
perfect BE study result is one where the mean ratio result (test to reference)
is 100% with confidence limits of 80% and 125%. And the real world reality is
that it is the statisticians job, in a pharmaceutical company, to estimate the
correct sample size, not one too large or too small. By the way, spending more
money on the testing won't improve the product.
In any event, don't pay too much attention to .001, when you can barely see .1,
and keep your stick on the ice.
Peter Gingras
pgingras.-a-.apotex.ca
Back to the Top
[Two replies - db]
Sender: PharmPK.-a-.boomer.org
Reply-To: Stanley110.-at-.aol.com
MIME-Version: 1.0
From: Stanley110.at.aol.com
Date: Fri, 2 Jun 2000 01:06:52 EDT
To: david.-a-.boomer.org
Subject: Re: PharmPK Re: BA/BE draft guidance 1999
USP under General Notices allows rounding up into compliance with a
specification for compendial tests. While releasing a product at a
specification limit is unwise for several reasons, rounding up into
compliance is acceptable practice. This has no bearing on the subject
guidance but it is interesting.
Stan Alekman
---
Alternate-recipient: prohibited
Date: Fri, 02 Jun 2000 07:56:27 -0500 (CDT)
From: Art Straughn
Subject: PharmPK Re: BA/BE draft guidance 1999
To: postmaster
MIME-version: 1.0
Posting-date: Fri, 02 Jun 2000 07:56:28 -0500 (CDT)
Importance: normal
Priority: normal
Sensitivity: Company-Confidential
A1-type: MAIL
Dear All,
A possible answer to the appropriate number of decimal places to use in
reporting the confidence interval values may be found at the end of the draft
guidance from the FDA titled "BA and BE Studies for Orally Administered Drug
Products -- General Considerations". It was issued 8/1999 and posted at their
web site 8/27/1999. This draft guidance states: "Confidence interval (CI)
values should not be rounded off; therefore, to pass a CI limit of 80-125, the
value should be at least 80.00 and not more than 125.00"
Unfortunately, this statement is ambiguous. If rounding off is not allowed,
then 79.999999999 will not pass; but if one uses the given example value of
80.00, then the USP 24/NF 19 rules for significant figures (pages 3-4)
indicates 79.99999 will pass. I think a clarification from the FDA is needed -
do you or don't you round off?
Art Straughn
Back to the Top
80.00 is written to 4 significant figures while 125.00 is written to five. The
nonsense is that these are percents and reflect a set value of 0.80
or 1.25 times
the values returned by the method itself.. The method will return
values to double
precision or however the computer is constrained to report. These
values in turn
are multiplied by the percent limits of 80 or 125. The precision
reported should
not exceed the precision of these multipliers. So how many angels do
fit on the
head of a pin? How small are the angels, how large the pin?
Back to the Top
The following message was posted to: PharmPK
Stan (Alekman),
You recent comments that "the precision reported should not exceed the term
in the
calculation with the least precsion, according to the rules of sig figs and
rounding" is very much to the point. In fact, those who are not clear about
this basic concept should review their Analytical Chemistry 101 to refresh
their memory.
See: page 41 of Quantitative Chemical Analysis by Daniel C. Harris, 4th Ed.,
W. H. Freeman and Company, New York, 1995.
Regards!
Wei-jian
Wei-Jian Pan, Ph.D.
Dept. of Clinical Pharmacokinetics
Abbott Laboratories
Abbott Park, IL 60064-6104
847-935-5335
weijian.pan.aaa.abbott.com
Back to the Top
This discussion about the precision of the reported results is important.
Most of the analytical instruments give three significant figures as the
value. So, any reported final result should be three sinificant figures. In
between calculations, however, should be carried out to the number of
significant figures supplied by the computer/calculator. Otherwise rounding
errors will cumulate excessively.
Instruments like scintillation counters might give six significant figures,
but these are probably not reproducible beyond the third significant figure.
Prof.Dr.Ilbeyi Agabeyoglu
Gazi Univ.Fac.Pharmacy
PharmPK Discussion List Archive Index page
Copyright 1995-2010 David W. A. Bourne (david@boomer.org)