# PharmPK Discussion - Significant digits / rounding

PharmPK Discussion List Archive Index page
• On 31 Jan 2002 at 10:10:49, David_Bourne (david.-a-.boomer.org) sent the message
`[A few replies - db]From: "Edward O'Connor"  Date: Wed, 30 Jan2002 21:31:27 -0600To: david.aaa.boomer.orgSubject: Re: PharmPK Significant digits / roundingThe following message was posted to: PharmPKthe curve then acutally goes from 0.500 ng/mL to 300 ng/mL (three sigfigs)---From: "Edward O'Connor"  Date: Wed, 30 Jan2002 21:58:31 -0600To: david.-at-.boomer.orgSubject: PharmPK Re: Statistical weight for calibration dataThe following message was posted to: PharmPKr is misleading when the range of values is great-over several orders ofconc and when the fit is non linear. Weighting using 1/y or 1/y2 (1/x,1/x2 adjusts for the bias imparted by the points at the higher ends ofthe curve upon the predictive values at the lower end. When using alinear model weighting may be used balancing the positive impact of theweighting on the accuracy and the negative impact it has upon the r. Innon linear relationships, weighting's impact is again balanced againstthe fit, this time as the SSQ or MSQ of the fit. Using a polynomial fitbecomes challenging particularly where both a high and a low value X cangive the same response. In those cases practical limits have to be seton the extenst of the fit range. The simplest fit is always a choice butweighting and other fit are valid if justified, and what do we use tojustify? we use accuracy, correlation (ssq or msq), and logic. 5 or 6order fits will connect all the dots but.........---From: "Weber, Cornelia {PDC5~Basel}"  Date:Thu, 31 Jan 2002 08:40:02 +0100To: david.-a-.boomer.orgSubject: RE: PharmPK Significant digits / roundingThe following message was posted to: PharmPKDear Jaime Ilha,I fully agree with your concerns and the way we handle it in our companyis as follows:We ask any analytical laboratory that is analysing samples for ourPK-studies (in-house and external) to report analytical data withsignificant figures. The number of significant figures of course alwaysdepends on the precision of the analytical assay and may therefore bedifferent for different assays. We then report all PK parameters thatdepend on the concentrations measured (e.g. Cmax, Cmain, AUC etc.) withthe same level of precision, meaning significant figures.I am personally not aware of any official guidance or publication onthis issue.Kind RegardsDr. Cornelia WeberClinical Science PharmacologyHoffmann-La Roche Ltd.(52/906)BaselE-mail cornelia.weber.aaa.roche.com---From: jan-georg.moeller.jm.-at-.bayer-ag.deDate: Thu, 31 Jan 2002 10:46:26 +0100To: david.aaa.boomer.orgSubject: Antwort: PharmPK Significant digits / roundingThe following message was posted to: PharmPKDear Jaime,bioanalytical methods are characterized by a wide working range -sometimes upto 10^4. (= Factor 10000). Within this range an accuracy and precisionof (normally) 15% need to be achieved. Thus the number of SIGNIFICANTdigits is important - not the number of digits. Using your examples onewould report a working range of 0.500 (including the zeros) to 296mcg/L. Using four significant digits you would report, e.g.: 0.5000 to296.3 mcg/L. Other examples are 0.1234, 123.4 or 1234000 (rounded).>From my point of view at least 3 or 4 significant digits should be used(reflecting a precision of about 15%). But of course, this is dependenton theerror and / or uncertainty of the method. However, I have not found aclear guideline on this and would appreciate any comments / hints onguidelines if 3significant numbers are sufficient.Best regards,Jan`
Back to the Top

• On 31 Jan 2002 at 12:12:43, Casey Laizure (claizure.at.utmem.edu) sent the message
`The following message was posted to: PharmPKIn Response to Jan's Comment:If the range of a standard curve is 0.5 to 300 and the precision iswithin 15%, does this mean that a concentration of 300 is 300±45. Giventhis, isn't 3 significant figures too many, 2 would seem moreappropriate. I find that we often, myself included, report assay resultsthat are optimistically precise.Casey`
Back to the Top

• On 31 Jan 2002 at 13:36:43, "Chaurasia, Chandra S" (CHAURASIAC.at.cder.fda.gov) sent the message
`In response to Casey's Comments:With the precision of +/-15%, 300 is indeed 300 +/-45.With regards to the number of significant figures, the sensitivity ofthe analytical method (or lowest amount quantitated) need to beconsidered among other things. For example, if the LOQ is at pico- orfemento gram/mL level, and plasma analyte concentrations are given inng/mL, one may express analyte concentrations in significant figuresrepresenting up to the pico or femento level.Chandra`
Back to the Top

• On 31 Jan 2002 at 14:35:46, "Tata, Prasad N" (Prasad.Tata.-a-.TycoHealthcare.com) sent the message
`The following message was posted to: PharmPKI would like to throw my two cents in this thread. Philosophically thereis no hard and fast rule for significant digits and rounding. Typicallyyou have to set your own rule keeping in view of the study objectives inmind. rounding off the assay values should not introduce error in yourfinal parameters and clinical interpretations. Imagine a situation whereyou round off plasma concentrations and then round of pk parmaeters andthen round of statistics introducing additive error at every stage.Bottom line is that we have to % CV at the lowest possible end (accepted% CV is < 20%) and rounding off should not artificially present higherCV.In my honest opinion it would be advisable to basic analytical measuresupto two decimal points (eg. 2.12 ng/mL or ug/ml). There is one and onlyofficial guidance available in literature is under Significant Figuresand Tolerances, USP 2002 page 4.It clearly states "Limits which are fixed numbers not rounded off" andit continues stating " When rounding off is required consider only onedigit in the decimal place to the right......" and lists few examplesClearly folks argue about the pros and cons of reporting values upto twodecimal points and there is no end for it. When designing studies orprograms responsible individuals should discuss all of the above pointsand make a determination on the reporting strategy, to avoid confusionit is better to have a uniform policy of reporting across all thedevelopmental projects.There can be exceptions but exceptions can not make rules.Hope my two cents is sensible and helps this discussion.Regards,Prasad TataTyco Healthcare-Mallinckrodt, Inc.`
Back to the Top

• On 31 Jan 2002 at 15:05:56, Laszlo Endrenyi (l.endrenyi.-at-.utoronto.ca) sent the message
`The following message was posted to: PharmPKFrankly, I do not think so. Usually 3 significant digits (at times only2, occasionally 4) are meaningful regardless of the distance from theLOQ. Goingdownwards to different units would introduce 3 or even 6 additionaldigits. Thenonsignificant digits could be represented by zeros or, better, theexponentialnotation could be used.Laszlo EndrenyiUniversity of Toronto`
Back to the Top

• On 31 Jan 2002 at 15:38:13, Andrew Somogyi (andrew.somogyi.-a-.adelaide.edu.au) sent the message
`The following message was posted to: PharmPKI suggest people have a read of Gerhard Levy's Letter to the Editor:Significant figures or significant nonsense? Clinical Pharmacology andTherapeutics 59: 363, 1996.Andrew SomogyiAssoc Prof Andrew SomogyiDepartment of Clinical and Experimental Pharmacology University ofAdelaideAdelaide 5005Australiaemail: andrew.somogyi.-a-.adelaide.edu.au`
Back to the Top

• On 1 Feb 2002 at 11:38:36, "Hans Proost" (J.H.Proost.aaa.farm.rug.nl) sent the message
`The following message was posted to: PharmPKDear colleagues,With respect to rounding and significant digits, three comments:1) Better using too many digits than too few, for at least two reasons:using too many never introduces inaccuracies (the false feeling ofaccuracy does not hurt, except in final results), and in many cases thenumbers are used further in the process (e.g. analytical data in PKanalysis). We all know that rounding should be done ONLY in the finalresult, e.g. the parameters of a population model. As a rule of thumb,all interim-results should have at least one more significant digit.2) In presenting results, avoid using '1234000' if you mean 1.234E+6,and using '0.5' if you mean '0.500'. On the other hand, it is quiteappropriate to describe the range of a calibration curve as, e.g., 0.5to 20000. In this case (not a result) everybody will understand that theconcentration at the lower end was not 0.46 and that the upperconcentration was perhaps 20007.3) Prasad Tata wrote:>In my honest opinion it would be advisable to basic analytical measuresupto two decimal points (eg. 2.12 ng/mL or ug/ml).I don't understand this recommenation. '2.12' means three significantdigits. The number of decimal points is irrelevant (exept for opticalreasons).best regards,Hans ProostJohannes H. ProostDept. of Pharmacokinetics and Drug Delivery University Centre forPharmacyAntonius Deusinglaan 19713 AV Groningen, The NetherlandsEmail: j.h.proost.-at-.farm.rug.nl`
Back to the Top

• On 6 Feb 2002 at 12:31:12, "Daniel Sitar" (sitar.-a-.Ms.UManitoba.CA) sent the message
`	It is invalid to extend precision beyond the accuracy of the toolsused for theproduction of the data. The increased precision is meaningless and leadsto a false senseof robustness of the calculations. Variance around a particulardetermination is a usefulway to decide on an acceptable number of significant digits.	D. Sitar, University of ManitobaDaniel S. Sitar, PhDProfessor and HeadDepartment of Pharmacology and Therapeutics sitar.-at-.ms.UManitoba.ca`
Back to the Top

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 "Significant digits / rounding" as the subject

Copyright 1995-2010 David W. A. Bourne (david@boomer.org)