Bioequivalence and Bioavailability Forum

Hi Angus,

❝ For the purpose of a regulatory report when reporting plasma concentrations in a Table of a PK report one uses the same precision as the numbers provided by the analytical laboratory. These are the numbers that are used for the pk calculations.

Hopefully. With the limited accuracy/precision of bioanalytical methods reporting too many figures is meaning­less. Let’s say you have a really good method (5% AP overall) – does it make sense to report a result like 3.142? Even reporting only three significant figures would imply that the method is able to distinguish between 3.1415 and 3.1424 – a precision of ~0.2%. If data are electronically transfered from the analyst to the biostatistician, the former should not only format the results as in the report, but actually round them. Otherwise there are discrepancies between paper and electronic data which may raise unnecessary questions. Having 3.14 in the report and 3.14159265358979 in the file is stupid. Many analysts have great difficulties in digesting this message. Tell them that with 5% CV the 95% confidence interval of π is 2.83–3.45.

❝ […] An actual time e.g. 16 hours :10 minutes, can be an awkward number as a decimal (16.166666.......hours). It is the full precision available of the actual time number that one uses in the computer calculation, but for a tabulation in a regulatory report I will round the time to 2 decimals for convenience.

I wouldn’t do that. The smallest temporal resolution of an EDCS I know is Parexel’s ClinBase™ with four seconds (if barcode-readers are connected). Even with such data giving time in decimal-hours to more than three decimals is meaningless. If paper CRFs are used (resolution one minute) two decimals are enough. If you opt for full precision, you should either give more decimals (I use five to six) or report time in ISO 8601-format (i.e., hh:mm[:ss]).

hh:mm  h(2)   h(5)    ∆  ∆(2)  ∆(5)
16:09 16.15 16.15000 –1 –0.02 –0.01667
16:10 16.17 16.16667                 
16:11 16.18 16.18333 +1 +0.01 +0.01666

Let’s say the planned sampling is at 16:10 and you have two samples (one minute early/late). From hh:mm it’s clear at the first look. Two decimals might give a false impression; IMHO, more digits are less confusing.

❝ There will be another corresponding Table describing the actual times in hours and minutes in the report.

Why not keeping everything in one place?