Hi Jesse

You are comparing two independent groups.  Therefore, you have to determine an effect size, or the "difference that makes a difference".  Let's assume your mean is 100 and your standard deviation is 10.  This means that someone with a score of 90 is one standard deviation below the mean.  In fact, Cohen's seminal text on power analysis (1988) uses that exact explanation.  Here is a web page to help you: https://www.stat.ubc.ca/~rollin/stats/ssize/n2.html 

I plugged in the following numbers: mean for the control, 100. Mean for sepsis 90, standard deviation 10, alpha 0.05, power of 0.80, and received the result you would need 16 people in each group.  Bear in mind that if the effect is very small, that number will go up...a lot!  I plugged in a mean of 100 for the control, and a mean of 95 for sepsis with the same standard deviation (10) and got an n of 63 per group!

Good luck, the web site is quite nice

Rich

Hi Richard, thanks for your fast answer, appreciate it!

The fact is, that we want to determine the levels op sepsis mean, so it's an unknown factor. And after this level determined, we would like to determine the difference between the sepsis levels and the non-sepsis levels.

So actually, our question, is it possible te determine/estimate the sample size needed without knowing the sepsis mean and SD, just the control mean and SD?

Or do I envision something that in statistically impossible?

Hi Jesse,

Knowing the control mean (sd) is very helpful; now you, as the researcher, must determine when you will be alarmed / interested / encouraged/ discouraged by a finding.  I will create an example.

 Protein "x" has a normal level of 10 and a standard deviation of 1.  If you tested a septic patient and received a score of 9, would you be worried?  Probably not, it is within 1 standard deviation of your normal score of 10.  You test a second septic patient and get a score of "2". Now you are quite worried, as this is 5 standard deviations below the mean.  in other words, the chance of a score of 2 due to random chance is less than 1 in 10,000.  

So, as the researcher, you need to decide what the level of the protein is to make you take notice.  Usually, unless this is brand new research, someone has written an article that says "In septic patients, protein "x" is decreased to a level of 2" (or some number).  This is called the effect size, and you, the researcher, based on the literature and your judgement determine how great a change is clinically significant.  Once you know that, you can calculate your sample size.

If you want to email outside of researchgate you can reach me at [email protected]

Cheers!

Rich

I have send you an email!