expsampleN.TOST vs. sampleN.TOST [🇷 for BE/BA]

posted by Ben – 2018-02-01 19:48 (2246 d 17:24 ago) – Posting: # 18325
Views: 2,894

Dear Elena777,

Let me try to answer your questions.

❝ 1. The fundamental difference between expsampleN.TOST and sampleN.TOST?


expsampleN.TOST uses the "expected power" as target power, whereas sampleN.TOST uses the classical power as target power to determine the sample size. That is, expsampleN.TOST allows you to take uncertainty of either the CV or the gMean ratio or even both into account when determining the sample size of your trial. Usually this will give you a higher required sample size as compared to sampleN.TOST. Note that my last sentence assumes that the value for "targetpower" is the same for both sampleN.TOST and expsampleN.TOST. This may not be fair because for expsampleN.TOST it may be reasonable to choose a lower level.

❝ 2. CV in function expsampleN.TOST is CVintra, isn't it?


Depends on the design argument. If design="parallel" then it is the total CV, otherwise it is the intra CV as you said.

❝ 3. In both cases we can choose the value of desired power (for example 0.8 or 0.9). But in description of function expsampleN.TOST the power is characterized as "expected" (sample size based on expected power) and in description of function sampleN.TOST the power is characterized as "given" (calculates the necessary sample size to have at least a given power). What is the difference between expected and given power?


I think here is a misunderstanding. The important thing in "given power" is not the word "given", but the word "power": the description for expsampleN.TOST could also be "... to have at least a given expected power". The sample size functions simply iterate until you have a sample size with which you have either a classical power (sampleN.TOST) or expected power (expsampleN.TOST) greater or equal to the specified value in the argument "targetpower". For the definitions of "power" and "expected power" refer for example to the documentation within the PowerTOST source (inst/doc/ BE_power_sample_size_excerpt.pdf and Expected_Power_for_TOST.pdf, respectively).

❝ 4. Can I use expsampleN.TOST when I did not conduct a pilot study but I have data from other BE studies, information about which was published? If yes, function expsampleN.TOST seems to have advantages over function sampleN.TOST because in a former case I can take into account uncertainty of CV (as we know CVintra is characterized as a parameter with some degree of uncertainty in general sence).


exppower.TOST and expsampleN.TOST have been developed based on the idea that you have performed a pilot trial. However, you can also use it in such a way that you do not directly assume you have one prior pilot trial. So the answer is 'yes' (with some limitation, see my example below). What matters is the specification of the prior distribution for the calculation of the expected power. The functions for expected power and sample size also allow you to directly specify "df" and/or "SEM" (via the prior.parm argument) - those do not rely on a specific size m (or design) of a previous trial. Let us imagine the following situation: let's say you have quite some information from the literature about your variability and you have calculated a pooled CV (e.g. using PowerTOST::CVpooled) which you assume to be fixed. You only want to take uncertainty wrt the GMR into account and you want to define the prior distribution of your treatment difference. In exppower.TOST the prior is assumed to be Normal, and you can specify the standard deviation of this distribution via the SEM argument (prior.parm = list(SEM = xx)). The mean of this prior will be set to log(theta0) where theta0 is the main argument from exppower.TOST (this is a limitation, in a more general setting you could use a different mean for the prior than for planning the new trial, but this is not possible here). Similarly, you can imagine situations for prior.type = "CV" or = "both".


I hope this clarifies a bit.

Best regards,
Ben

Complete thread:

UA Flag
Activity
 Admin contact
22,957 posts in 4,819 threads, 1,639 registered users;
73 visitors (0 registered, 73 guests [including 3 identified bots]).
Forum time: 13:13 CET (Europe/Vienna)

Nothing shows a lack of mathematical education more
than an overly precise calculation.    Carl Friedrich Gauß

The Bioequivalence and Bioavailability Forum is hosted by
BEBAC Ing. Helmut Schütz
HTML5