0
$\begingroup$

I know how to conduct a sample size calculation comparing two independent means in a randomized controlled trial. However, I am confused by the following situation.

There are 2 groups: intervention (IG) and control (CG). The outcome of interest is a severity score (0-100, the higher the worse).

The baseline score in both groups is 20. Within a year the score increases by 5 in the general population. The intervention is supposed to cut the score by 50%. So, given a baseline of 20, the IG is expected to have a score of 22.5 after a year, while the CG is expected to have a score of 25. Let‘s assume the standard deviation of the score in both groups is 10. Aiming for an alpha of 5% (two-sided test) and a power of 80% what sample size is needed to identify the difference?

Using the standard formula 2*(Ζα∕2 + Ζ1-β)² devided by (mean1-mean2)/SD gives about 250 per group.

However, I was told the approach is not correct and I should instead calculate the sample size needed to identify that the score is 50% lower in the IG than in the CG. How can that be done? Isn’t that the same?

Best, Joey

$\endgroup$
5
  • $\begingroup$ Since the score scale [0,100] is limited on both sides and have only 101 vales, the starting mean 20 and standard deviation of 10 don't adequately describe the population. I don't believe there can be any closed form solution that might give us the sample size, especially if you are going to use a non-parametric test, which is strongly preferred here. If you have a real-world distribution of the score in different age groups it might serve a basis for a Monte Carlo simulation to attempt to estimate the sample size for the target power of 80%. $\endgroup$
    – Alex
    Oct 1, 2022 at 10:45
  • $\begingroup$ Thanks, Alex. The numbers were just an example. Disregarding that a mean of 20 and SD of 10 may not be plausible, what would be the correct formula to calculate the sample size for a percentage change (as in the example, i.e. „the IG experiences a 50% lower increase in the score than the CG“)? $\endgroup$
    – Joey
    Oct 1, 2022 at 12:37
  • $\begingroup$ By there is no closed form solution I meant no formula :-) $\endgroup$
    – Alex
    Oct 1, 2022 at 14:23
  • $\begingroup$ It seems you need to clarify with whomever is running the study what "50% reduction" means. Do they mean the score in control will be 50% of the treatment? Do they mean the change in scores will be 50% of what would have been otherwise? You need clarity on what the estimand should be. $\endgroup$ Sep 18, 2023 at 19:44
  • $\begingroup$ Percent change is an asymmetric measure so is not appropriate for use as a dependent variable. See this. $\endgroup$ Sep 18, 2023 at 19:55

1 Answer 1

1
$\begingroup$

It seems like the problem is how you understand "The intervention is supposed to cut the score by 50%." It looks like your calculation reflects the situation when the intervention cuts the increase by 50% instead of the score. However, that sentence is unclear, leading to the intervention value to possibly be either 10 or 12.5 after one year. Did I get it right?

$\endgroup$

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.