control of type I error?

Hello forum,

I am seeking your advise. Here is the problem:

In a clinical trial, we will collect MRI imaging data from each patient participating in the study. The imaging data will be read by 3 independent image readers. So we have 3 sets of evaluations from the 3 readers. Then we run a statstical test based on each data set. Now the protocol states that if at least 2 out of the 3 tests are significant, then we can claim the study to be positive.

To calcuate the sample size needed, do you think we need to control the type I error rate in this situation?
It is a non-inferiority test comparing with placebo. So we will have 3 p-values from the 3 readers evalutions. If we had just one reader, then the trial is positive is the p-value is less than 0.05. The question is for the current proposal (i.e. at least 2 out of the 3 p-values is less than 0.05, then we claim the trial as positive), is there an issue with inflating the type I error rate?

What test are you using? Please provide some more detail.

Do you have a gold standard to confirm the positive?


Omega Contributor
Not completely my forte, but I get what you thinking. This would be similar to multiple comparisons. What do the people in your field do? Can you find anyone else using this approach in the literature?