The best way would be to do a first study to evaluate this prevalence, even roughly, and then decide which prevalence decrease/increase you expect (I suppose this is the effect size you mention).

If not possible, you have to make some « reasonable » guesses.

"and thus prove your sample size is sufficient" - what for?

You have some data, so you can use this to estimate the prevalence(s) and also give a precision for the estimate(s).

The precision can be improved using larger sample sizes. To judge the required sample size you need to know how precise your estimate(s) need to be. The precision should be high enough to see "relevant" effects. But what is relevant? - You need to decide this. If you can't, then take whatever you have and try to interpret it (the data will be unconclusive for effects below the precision; but this is a general fact)

And forget about "proving" something. You will never prove anything with empirical data. Data may be arbitrarily unlikely under a specific model/parameters, but the model itsef may be arbitrarily wrong, too...

Thanks so much for your fantastic input. I have a clearer picture of how to approach it right now. Take care.

If you are doing it for a power analysis (and as Jochen says, not to prove something, but to show you sample size is the conventional size to detect some effect), then the effect size that you want to detect may have nothing to do with prevalence estimates. You need to decide what effect size you want to be able to detect, not what effect is "out there". If you need to detect a tiny effect, that is different than just needing to an effect if it is very large.

Jochen is bang on the money! You can measure the precision of your estimates post hoc. The width of the confidence interval tells you that.

One way of thinking about this is as follows. Consider an effect size D which is important. How look at the following outcomes with respect to D and the confidence interval. In the following graphs ( is the lower confidence limit and ) is the upper confidence limit:

A)     --------------(-----0-------)----D-----                 

B)    ---------------(-----0------------D---)----

C)   ----------------------0--(-----)---D--------

D) ------------------------0--(---------D--)-----

E) -------------------------0------------D--(--)--

Outcome A has an effect which is not statistically significantly different from 0, but which is statistically significantly less than the effect size which is important. You have sufficient information.

Outcome B has an effect which is not statistically signfiicantly different from zero, and which may be above or below the important effect size. You do not have sufficient information.

Outcome C has an effect which is not statistically significantly different from zero, but which is significantly below the important effect size. You have sufficient information.

Outcome D has an effect which is significantly different from zero, but which may be above or below the important effect size. You do not have sufficient information.

Outcome E has an effect which is statistically significantly greater than zero and statistically significantly greater than the important effect size.

Notice two things:-

The yes / no result of a statistically significant result is resolved into five (not 2!) different outcomes with different consequences;

The confidence interval tells you whether or not you have sufficient precision.

Of course, all of this is predicated on there being some sort of minimally important effect size. Usually there is, and if there isn't it means you do not know how you will interpret your results. 

Power is used in the design phase, not the interpretation phase. And to calculate power you need to know what matters (what effect size is important), what effect size you are likely to see, and how variable your measurements are. If yoiu don't know all of these, you have no basis for calculating power, and no basis for setting sample sizes. If you don't have any idea about these, then you need to do some sort of pilot study - or take a leap in the dark.  

I hope that this help-s :)

I think that I am in agreement with others who have offered comments when I suggest a a pilot study to produce an estimate for prevalence. Additionally, I would suggest an examination of the literature for effect sizes associated with conditions that have similar symptoms and biological mechanisms. That last bit might sound a bit silly to those outside psychology...

Thanks Carl. It makes sense to me.