I am trying to understand more about non-inferiority trials and one thing that I am struggling with (for binary outcomes) is when it might be more appropriate to choose either an absolute (i.e. risk difference) vs relative (RR/OR) for establishing the non-inferiority margin. Are there any good reasons for choosing one over the other? Intuitively (for me) an absolute difference feels easier to work with…
An absolute scale is harder to work with because every patient will have a different absolute risk reduction (ARR is a function of baseline risk).
And yet the majority of NI studies uses ARR, right?
Doesn’t this also make use of RR/OR difficult since whether or not an RR/OR is “large” depends on baseline risk?
An OR changing over the range of baseline risk when RR is held constant or vice versa does not tell us much. The best way to answer this question is to take a couple of hundred trials, all with different baseline risk, and see if baseline risk is associated with the reported magnitude of effect.