Sample size estimation for a retrospective study

[Iman Ghoneim]

Hi

I'm trying to draft a proposal for my study in a retrospective setting. A population of a fixed number, has been classified into four classes (mainly one with no markers and three with combination of the other 3 markers) using three bio-markers and accordingly assigned to treatment groups . I need to test for a new marker in this population and correlate it's level of expression to favorable clinical and pathological parameters. My problems are:

 What method of sample size calculation would be suitable for this type of study with reasonable power?

What type of statistical tests would be suitable for correlation?

 I'm selecting from a patient data base within one year ( around 350 patients) if not sufficient then will expand the selection from the following year.

 I've  reviewed similar studies, usually somewhere around 100 but the method of sample calculation is not mentioned only the comparison tests

 . ( My budget limits is there and not much lower so as not affect the power and significance of the study).

I've tried the online calculator but I'm not sure is it considered one or two tailed. Also I don't have a fixed power of 0.8 

 I've consulted a statistician at my educational institute . She suggested to use Pockoc's formula ;  at power of 0.8. I'm not sure how this works though.

I've also sought help from more statistic oriented colleagues . She showed Cohen tables and tried the above formula and another that has two equal arms:

The sample size calculated using ( Snedecor & Cochran 1989) formula

                                           C s 2  

             n1 = n2 =2 +_______________

                                             D2

  s = 8 from similar study (SD)

 D =  change in the mean from similar study

 C= constant from Cohen tables at power of 0.8 

Any help would be greatly appreciated.

Iman