Page 1 of 1
GLP "control of bias"
Posted: Mon Mar 17, 2008 9:59 pm
by jdlh199
The GLP 21 CFR 58 regulations state that a non-clinical study should include:
Each study shall have an approved written protocol that clearly indicates the objectives and all methods for the conduct of the study. The protocol shall contain, as applicable, the following information:
...
(6) A description of the experimental design, including the methods for the control of bias.
For a standard GCMS analytical assay, how would you generally address the "control of bias"?
Posted: Mon Mar 17, 2008 11:01 pm
by tom jupille
Think of it this way: there are three components of error:
- bias (slope is wrong)
- offset (intercept is wrong)
- blunders (we all know what those are)
For bias, probably best to document that your recovery is constant across the working range of your assay.
Next best (and it's a loooong way down!) would be to document that your recovery is predictable across the working range (and to include an appropriate correction factor).
Posted: Mon Mar 17, 2008 11:16 pm
by jdlh199
So here's a question then. I have a study with a GCMS method that includes:
1. Calibration curves with back-calculated concentrations and recoveries
2. QC samples (across range) with back calculated concentrations and recoveries
3. Study samples
Is a statement that essentially says "control of bias is covered through calibration/QC data meeting ±15% acceptance criteria" enough to satisfy FDA scrutiny?
Posted: Tue Mar 18, 2008 12:06 pm
by JGK
I think "control of Bias" in GLP 21 Part 58 is primarily aimed at pre-clinical studies on animal based test sytems. It's a wide ranging document and is applicable to a host of study types.
However, as part of you validation, you should run sample blanks and test that the presence of the sample matrix (non analyte components), if different from the STD solution matrix, does not affect analyte peak response.
Thus avoiding, or being able to correct for, any "bias" to sample measurement.
Posted: Tue Mar 18, 2008 3:51 pm
by jdlh199
Whatever it's aimed at, our QA dept is wanting me to address it in a sample analysis study.
Validation has already been completed. "Control of bias" was not included in that either (and that seemed to get through QA no problem).
What you mention about sample blanks/matrix sounds more like Selectivity than "Control of Bias". The validation already had calibration curves, QC's, blank samples, spike samples, etc etc....
So I'm a little stumped as to what QA wants.... and of course when I ask what they want it's "that's up to you, you're the lead scientist!"
Posted: Tue Mar 18, 2008 4:15 pm
by Dan
I am not all that familiar with that particular reference in the 21 CFR.
However, my two cents interpretation is that this may be a terminology issue. It seems to me that the "experimental design", "methods" and "control of bias" that are mentioned in the excerpt that jdlh has given are references to statistical evaluation of the study data and not to 'analytical methods'.
Of course, I could be entirely wrong here. I don't currently have access to the reference material to make a full detailed review.
Regards,
Dan
Posted: Tue Mar 18, 2008 5:12 pm
by jdlh199
Here's the only online link I could find:
http://www.fda.gov/ora/compliance_ref/b ... alrule.htm
and here's the full text:
§ 58.120 Protocol.
(a) Each study shall have an approved written protocol that clearly indicates the objectives and all methods for the conduct of the study. The protocol shall contain, as applicable, the following information:
(1) A descriptive title and statement of the purpose of the study.
(2) Identification of the test and control articles by name, chemical abstract number, or code number.
(3) The name of the sponsor and the name and address of the testing facility at which the study is being conducted.
(4) The number, body weight range, sex, source of supply, species, strain, substrain, and age of the test system.
(5) The procedure for identification of the test system.
(6) A description of the experimental design, including the methods for the control of bias.
(7) A description and/or identification of the diet used in the study as well as solvents, emulsifiers, and/or other materials used to solubilize or suspend the test or control articles before mixing with the carrier. The description shall include specifications for acceptable levels of contaminants that are reasonably expected to be present in the dietary materials and are known to be capable of interfering with the purpose or conduct of the study if present at levels greater than established by the specifications.
(8) Each dosage level, expressed in milligrams per kilogram of body weight or other appropriate units, of the test or control article to be administered and the method and frequency of administration.
(9) The type and frequency of tests, analyses, and measurements to be made.
(10) The records to be maintained.
(11) The date of approval of the protocol by the sponsor and the dated signature of the study director.
(12) A statement of the proposed statistical methods to be used.
I agree it's a statistical evaluation of the study data, but I'm not too sure on how to address it. I'm presuming that I can simply show the mean/%RSD of each QC level, any significant difference between nominal and observed mean would be some kind of bias. If this bias is within the ±15 acceptance criteria of the method, then surely I can say "bias" is "controlled"??
Posted: Tue Mar 18, 2008 5:12 pm
by JGK
I think the following may be of help
I foound it in
http://www.isoqol.org/plenarysession.pdf which is not strictly 21 CFR part 58 bu can be considered applicable.
"Bias = systematic error that results in deviation of results from “trueâ€
Posted: Tue Mar 18, 2008 6:20 pm
by Dan
Thanks for the references. I found some also, but nothing yet that seems to address the issue.
Link to the text of 21 CFR 58 from the GPO (goverment printing office):
http://frwebgate1.access.gpo.gov/cgi-bi ... n=retrieve
Link to an FDA guidance on the GLP topic:
http://www.fda.gov/cder/guidance/old004fn.pdf
This guidance gives a lot of questions and answers, but, unfortunately, not the one you need (at least I haven't found it there in my reading of it).
I am still of the opinion that the 'control of bias' applies to the study results and not to the validation of the analytical method.
However, it seems that your QA group is being insistent. I agree with your conclusion and that which JGK stated. It makes sense as it is in keeping with ICH and FDA regulations. Meeting your method validation acceptance criteria gives you control of bias for your method.
Again, just my 2 cents worth (I am more aware of GMP than GLP).
Regards,
Dan
Posted: Tue Mar 18, 2008 6:49 pm
by JGK
I have worked in GLP environments with analysis methods for over 20 years (god I'm getting old!).
With a valid analytical method there are checks and balances at every turn which do nothing else but guarantee the validity and lack of bias in your data.
Chances are with ever analytical sequence you produce you will have:
Proof of system functionality from the system suitability test;
Linearity, precision and accuract within criteria from the calibration data.
Intra run accuracy (and possibly precision) from QC analyses.
all of which must meet specified criteria for the sample results to be considered valid. This is your demonstrable control of bias
Posted: Tue Mar 18, 2008 6:56 pm
by jdlh199
I am still of the opinion that the 'control of bias' applies to the study results and not to the validation of the analytical method.
Yes that's right, this is for a GLP sample analysis study using a method that has already been validated.
With a valid analytical method there are checks and balances at every turn which do nothing else but guarantee the validity and lack of bias in your data.
Chances are with ever analytical sequence you produce you will have:
Proof of system functionality from the system suitability test;
Linearity, precision and accuract within criteria from the calibration data.
Intra run accuracy (and possibly precision) from QC analyses.
all of which must meet specified criteria for the sample results to be considered valid. This is your demonstrable control of bias
I think that's kind of what I thought as well, perhaps our QA is being insistent in simply requiring a statement that says "this will reflect control of bias" with those magic words in the protocol. At the moment it just says something along the lines of "calibration data must be within XX, QC data must be within XX etc etc" without mentioning "control of bias"
I think all I need to add is a simple statement along the lines of "Control of bias is assessed through the accuracy (%recovery) and precision (%RSD) of QC data at each level". For the study report following analysis I can then just show the mean and %RSD of each QC level as an examination of control of bias.
That'll do.....
Thanks everyone for the comments!
Posted: Wed Mar 19, 2008 7:18 am
by Peter Apps
"Bias" has two meanings in this context. The first applies to analytical results and is also known as systematic error - results being higher or lower than the "true" value, and the means of sets of results not getting any closer to the true value no matter how many results are in a set (in contrast to random errors that trend to the "true" value as means of larger sets are taken). The best that can be done to eliminate this bias is to validate the method for recovery and correct for it if necessary, and use internal standards that are a very good match to the analyte (isotope labelled analogues are first choice).
The second meaning of "bias" applies to interpretation of results - human nature being what it is the developer of a drug say might be biased towards getting a favourable result for efficacy or safety for example. This is handled by using statistical tests rather than subjective opinions, and by experimental designs that include e.g placebos and having double blind and crossovers built into the study plan. None of this has anything to do with analytical chemistry.
Peter
Posted: Wed Mar 19, 2008 12:16 pm
by JGK
On a personal note, having worked with a few QA units, you get ones that are aware that definitions and terminology vary and will be flexible. (in this case they can see the "controls" (acceptance criteria) within the validation and subsequent assay and accept them.
Or you get an extremely rigid "no exceptions" approach the most obvoius one I can recall had to do with test article characterisation. Instead of a "certificate of analysis", I was provided with a "certificate of quality control" which, despite containing all of the information you would expect in a certificate of analysis, my QA (at the time) refused to accept as an alternative document insisting I place test article characterisation as a GLP exclusion in my report. I eventually won out in the argument, but only after it was elevated to the Senior VP management level.