Dear Experts:

In fact, I have a few questions regarding MI. I already read/search online for a short while.

Background information:

We are conductiong a clinical trial and the disease is epilepsy. Two arms, drug, vs placebo. The DB period is around 100 days.

From patient's diary, we get seizure count/day. Our primary endpoint is %change from baseline in seizure frequency (scaled to per 28 days). That is, { seizure frequence (scaled to per 28 days) in the end of DB period - baseline seizure frequency (also scaled to per 28 days) }/ baseline seizure frequency *100%. The test will be Wilcoxon rank sum test.

Patients' diary will be missing in some days. Now we plan use MI for the primary endpoint as our sentivitiy analysis.

Q1: someone told me that, "The imputations will be done separately for each treatment group including baseline, age group as additional covariate". I am confused with this.

I think, imputation is by subject, correct? I understand the imputation model concept. That accounts for all the variables which should be included the imputation model.

I means that, if any subject has some missing values, MI will imputate values for this subject. Any subject is on ONE arm. It is oringally 'seperately'. right?

Q2: Our Test is Wilcoxon. After imputation, how can we get a 'whole' reasonable p-value?

I am reading Robib's rule for mean, sterr, etc., and has not figured it out yet.

Thanks!

Xiaoshu Xu