Analysis of binary outcomes from randomised trials including multiple births: when should clustering be taken into account?