Ensuring the comparability of comparison groups: is randomization enough?