Ignoring the matching variables in cohort studies – when is it valid and why?