Number of subjects in cohort?

I have to write a fictive research protocol for a prospective cohort in which the goal is to study "...if an intervention program with individual exercise coaching substantially can reduce the number of subjects with low physical activity in the group of prefrail subjects aged 50-64?". Earlier studies show that there are around 37% prefrail in the age group 50-64. For the expected outcome of such an intervention, I can create a fictive prestudy.

The part I'm struggeling with, is how to determine how many subjects I need for this study, and how to defend that decision. I dont believe that the standard 1/√n for 95% confidence intertval will suffice... Also, what analytical test would be best for such a set of data?

In other words:
How to select representative sample size for this kind of cohort?
What analytical test to use?

Best regards