Apparently Effect Modification is something "built into" the variables you are trying to study. It is NOT a bias and thus cannot be easily controlled for by randomization. (I believe)
That question you refer to... using OCPs will increase relative risk of breast cancer in those with family history because:
thinking it through on the most basic of terms:
If there is an increased chance of breast cancer to begin with.. any cancer tissue that does show up will be amplified by the estrogens... leading to more breast cancer being detected.
At least that is my understanding. Hope that helps and hope someone can give you a better explanation
