Bioequivalence and Bioavailability Forum

Dear Elena,

❝ 1. A BE study […] planned to enroll 24 subjects. But in fact that study was conducted with 23 subjects. […] If I input such data:

❝

❝ CVfromCI(pe=0.99, lower=0.87, upper=1.13, n=23, design="2x2", alpha=0.05, robust=FALSE)

❝

❝ RStudio gives me a result but with the following note:

❝

❝ CVfromCI(pe=0.99, lower=0.87, upper=1.13, n=23, design="2x2", alpha=0.05, robust=FALSE)


❝ Unbalanced 2x2 design. n(i)= 12/11 assumed.


❝ [1] 0.2617363

❝

❝ Can I use this result for futher calculations of CVpooled by means of using PowerTOST or should I correct my data?

The function does not “know” how many subjects in each of the sequences were dosed. The function tries to keep a 2×2 study with an odd number of subjects as balanced as possible (here 12 subjects in one sequence and 11 in the other) and throws this message.
CVfromCI(pe=0.99, lower=0.87, upper=1.13, n=23, design="2x2")
Unbalanced 2x2 design. n(i)= 12/11 assumed.
[1] 0.2617363
If you know the subjects per sequence and specify them in the argument n, you get the same result but without a message.
CVfromCI(pe=0.99, lower=0.87, upper=1.13, n=c(12, 11), design="2x2")
[1] 0.2617363
But the study might have been even more unbalanced. Let’s try 14 subjects in one sequence and 9 in the other:
CVfromCI(pe=0.99, lower=0.87, upper=1.13, n=c(14, 9), design="2x2")
[1] 0.255524
Given that if you don’t know the subjects / sequence the code’s attempt to keep the sequences as balanced as possible gives you the highest (i.e., most conservative) estimate. For the background see this presentation (slides 25–29).

❝ 2. While calculating CV from CI with data from another study I`ve got the following result:

❝

❝ CVfromCI(pe=0.96, lower=0.9075, upper=1.0089, n=24, design="2x2", alpha=0.05, robust=FALSE)


❝ [1] 0.1071475


❝ Warning message:


❝ sigma based on pe & lower CL more than 10% different than sigma based on pe & upper CL. Check input.

❝

❝ My input is correct because numbers are taken from that application not from my head.

Please check it again. The PE is given by \(\sqrt{lower \times upper}\). In your case that’s \(\sqrt{0.9075 \times 1.0089} = 0.9568577\). The function checks the input for plausibility. Hence, the message is correct since your 0.96 is different to 0.9568577.

❝ Should I pay attention to this warning message

Yes. Check the PE.

❝ and can I use this result for futher calculations of CVpooled by PowerTOST?

I would (after checking the data for correctness) suggest the ones with the highest numeric precision.

❝ 3. While calculating CV from CI is it prefered to use CI for log-transformed parameters (for example LCmax) or for non-transformed parameters(Cmax)?

The former. According to all guidelines the analysis for C_max (and AUC as well) is done on log-transformed data.