The special term pairwise
indicate that the model should be fitted individually for
every tested level in comparison to the reference level.
Details
Let's ARM
being a factor with level A, B, C; let's be B the reference level,
a model calling the formula including pairwise(ARM)
will result in two models:
A model including only levels A and B, and effect of A estimated in reference to B.
A model including only levels C and B, the effect of C estimated in reference to B.