I’m a radiology trainee so my statistical background is not so solid. I want to design a study when I’ll be comparing three measurements of a variable in imaging obtained in magnetic resonance and then compute an ICC as a measure of agreement since I’ll be obtaining a continuous measurement. This is a novel type of measurements and related to the artificial intelligence field so I have no prior data to compare with. I’m expecting to obtain a clinical useful ICC if this is at least between 0.75 and 0.9 and I’m basing my opinion on the following two papers: ref1, ref2
I used R with the package ICC.sample.size and this is my result
ICC.Sample.Size::calculateIccSampleSize(p=0.75, p0 = 0, k = 3, by = "p", step = 0.1)
[[1]]
N p p0 k alpha tails power
1 6 0.75 0 3 0.05 2 0.8
[[2]]
p N
1 0.0 Inf
2 0.1 286
3 0.2 77
4 0.3 36
5 0.4 21
6 0.5 14
7 0.6 10
8 0.7 7
9 0.8 5
10 0.9 4
11 1.0 2
Given this result can I assume that with 30 patient (my actual sample size) I have enough power to detect an ICC of at least 0.75?