Back to the Top
Dear discussants,
I'm a clinician involved in conduct and review of clinical PK studies for submission to regulatory authorities such as the FDA. We found your excellent discussion forum while trying to answer the question, How many significant digits should we be presenting in the text of study reports for PK parameters? (We often receive particularly mindless PK sections and stat/PK tables from CROs that imply precision to 6 or even more digits, even in studies with group sizes of n=6. This does not pass the 'red face' test, in my opinion. Their frequent defense that 'It is in our SOP' seems even more brainless, we think.)
So I have one question and one suggestion for the consideration of the PK and stats experts weighing in here (which I certainly am not):
1) Does anyone disagree with the rules provided in the Wikipedia chapter on 'Significance arithmetic'?
http://en.wikipedia.org/wiki/Significance_arithmetic
If not, or contrariwise if there is a vocal disagreement on some aspect, a sort of 'Delphi' approach could be used here. That is, perhaps you all could come to an agreement by iteration on the proper rules, which could even result in a published consensus on how to handle significant digits and precision for presentation of raw data such as Cmin and Tmax. One suggestion I might make is that it is the BA lab's job to provide us with understandable information on the degree of precision of the concentration data, and the clinical site's job to provide accurate sampling times (most have moved beyond nominal time by now); so these could be stipulated as being available in then proceeding to recommendations. I suspect you would find that there is fairly solid agreement on preserving the raw data from the lab and site in listings and in computations, and that significant digits, not decimal places, is the proper way to think about precision -- but a Delphi process would settle that.
2) Although the discussion in the 'propagation of uncertainty' chapter in Wikipedia makes clear that determining precision after complicated transformations, of which AUCinf might be a good example, is not easy, a follow-on chapter suggests that there is statistical methodology and software to do this:
http://en.wikipedia.org/wiki/List_of_uncertainty_propagation_software
My question is: Has any of you, or a publication we could all access, performed simulations with these types of software/methods to determine the degree of precision one can reasonably claim for parameters like geometric mean Cmax, AUCinf, t-half, etc, based on the inputs of 1) precision of the concentration data, 2) precision of the time of collection, and probably most importantly, 3) number of subjects contributing the data to the parameter estimate? If so, might you then see if there is agreement that 1) the method put forward is acceptably trustworthy to form a basis for recommendations on 2) the number of significant digits for PK parameters that should be presented in regulatory submissions and scientific publications? It would be a nice contribution if this forum could accomplish that. If a software program in, eg R, could be identified that would perform simulations/estimations based on an individual study's size, etc, that would be valuable to know too.
Thanks very much for considering these two questions,
Rich Reeves, MD
RAR Consulting LLC
rareevesmd.-a-.comcast.net
Back to the Top
Dear Dr. Reves:
You quoted
"A trivial example is a concentration of 0.10 ng/mL, which is the exact same as 100 pg/mL; but having the first result in hand does not allow expressing the second as 100.00 pg/mL, "
This is precisely my frustration anytime a scientific argument was presented then people bring an alternate argument and creates lot noise (academic mulch- is my favorite word), in the ensuing noise the actual point of discussion gets drowned (as Bill Clinton quoted " The problem with democracy is that there are too many opinions expressed in multiple voices, in the middle of all these voices original issue was drowned or lost its relevance"). Two decimal points are to a unit (mg,ng,pg whatever) that you are primarily expressing your measure, meaning units on which your calibration curve is established. If your calibration curve is established in ng/mL and you are getting concentrations in mg/mL (what is the purpose of having calinbration units in ng/mL when you expect concentrations in mg/mL???)there is a need to redefine the calibration units and express your measured concentrations. Once you define proper units then express concentrations to two decimal points this way we are keeping precis
Back to the Top
Dear Dr Tata,
thank you for your quick reply, and thoughts.
I tend to agree with you that presenting concentration data from an HPLC in a format that implies an accuracy to about 1 % is probably about right, overall. However, as lots of people pointed out in the thread from which this discussion arose, 1% is not at all the same as 2 decimal places. [A trivial example is a concentration of 0.10 ng/mL, which is the exact same as 100 pg/mL; but having the first result in hand does not allow expressing the second as 100.00 pg/mL, which implies a lot of precision that simply wasn't there in the original number.] Thus, number of decimal places just does not work for conveying precision, as it is not a reliable surrogate for significant digits. The number of decimal places is in many cases solely dependent on the arbitrary choice of units the writer makes, and those words (the units) can by themselves say nothing about teh HPLC's and lab's measurement precision (or pseudoprecision, which is my beef with a lot of the PK reports I see). From what I can see o
|
Want to post a follow-up message on this topic?
If this link does not work with your browser send a follow-up message to PharmPK@boomer.org with "Digits and decimals in concentration reporting" as the subject | Support PharmPK by using the |
Copyright 1995-2011 David W. A. Bourne (david@boomer.org)