4
$\begingroup$

I'm doing a clinical trial and I'm comparing a certain number before treatment and after treatment. I recorded the numbers for each patient before treatment and after treatment and take the difference(after-before). But somebody commented that in the previous studies, it is possible that the study was conducted on extremely selected limited subjects, so there is no guarantee that the same parameter is observed as a result of the study. Therefore, we have to consider a more conservative way, like using ratio or percentage instead of the difference.

My question is, do I have to change the sample size here? I calculated the sample size using the sample size formula using type I error and type II error. I personally don't think the sample size needs to be changed, but I would love confirmation.

Improve this question
$\endgroup$

1 Answer 1

Reset to default
4
$\begingroup$

Comparing before and after treatment is usually meaningless for assessing the causal effect of treatment, unless you know that without treatment the before and after change would have been exactly 0. This is nicely discussed e.g. in the ICH E10 guidance. For that reason, in many settings a before-and-after-study is not the design to choose, or will need at the very least historical or external controls for analysis and interpretation (but more commonly a concurrently randomized control group is preferable).

Whether a ratio or percentage change is a better way of looking at something and whether that is a "more conservative way", at all, depends very much on the specific situation. In general, this should somewhat change the distribution of the outcome somewhat so that samples size calculations would change. They might change even more, if the study populations between two studies are very different (e.g. more variable outcomes in a wider/less controlled population is a possibility). When reviewing clinical trial protocols I would consider it a problem, if the sample size calculation does not match the intended main analysis (unless good arguments are presented why this is not an issue).

Improve this answer
$\endgroup$
5
  • 1
    $\begingroup$ Thank you for the answer. I will take a look at the guideline! I guess there might be a problem if I didn't change the sample size when switching to ratio or percentage. I will think about it. $\endgroup$
    – user10386405
    Commented Jan 13, 2022 at 12:59
  • $\begingroup$ I have one more question tho. Do I need to change the sample size if I want to to a randomized block allocation? $\endgroup$
    – user10386405
    Commented Jan 13, 2022 at 13:00
  • 3
    $\begingroup$ As alluded to above, this kind of pre-post design is very unlikely to answer your questions, and if attempting to publish, good journals won't accept it. It is the weakest of all designs save the use of historical controls and having no concurrent controls. I'm not sure I would even call pre-post a clinical trial. $\endgroup$
    – Frank Harrell
    Commented Jan 13, 2022 at 13:12
  • $\begingroup$ Regarding your second comment: Randomized block allocation (e.g. having blocks of two patients that either are assigned Treatment + Control or Control + Treatment) makes only minimal differences in most practical settings and can usually be ignored in sample size calculations. $\endgroup$
    – Björn
    Commented Jan 13, 2022 at 13:42
  • $\begingroup$ Thank you very much! $\endgroup$
    – user10386405
    Commented Jan 13, 2022 at 22:11

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.