non-inferiority trials

[Paolo2000]

Hi everyone,

I was looking for some suggestions on ho to compute sample size for a non-inferiority trial. The problem is that I do not have studies with previous effect sizes of the active comparator to use for establishing a non-inferiority margin. What would you suggest to do in this case?
Thanks in advance

 

[hlsmith]

Is your dependent variable continuous, binary, other? Are there known risks or costs associated with treatments to weight outcomes? Would 110-120% the comparator work for the threshold?

You are gonna have to give us a little more!

 

[Paolo2000]

I am not sure how to answer to some of your questions, but I try to give more details. My DV is continuous (self-care behaviors). There are higher costs associated with the new treatment compared to the control treatment; it is in fact a educational intervention administered through an online platform vs standard education (and I know that this is an important shortcoming). The sample is composed of ostomy patients and so far, there are no studies that have compared active treatment with placebo in the literature, and at the same time that used the same outcome as our trial (they used similar constructs but not the same as ours); therefore the assumption of sensitivity assay and constancy assumptions cannot met under these conditions. In the family of the self-care instruments (range: 0-100) we adopted 8 as a score to determine clinical improvement (validated). My thought was to use half of this score (4) as a non-inferiority margin in case I cannot find studies helping in this determination. I hope I have been clearer now, Thanks in advance!

 

[hlsmith]

I would simulate the do groups values for the instrument. So stare at the instrument and guess how you think they would answer via inputting a range of possible values for each group. and compare to see if the groups would be different given your margin. So you are expecting in-person to have an 8-point shift and online not to have worse than a 4-point shift?

 

[Paolo2000]

Thanks for this sharp suggestion. I will do definitely that. What I dread though, is the fact that this could be a no evidence based approach. What do you think?