Dichotomising outcome variables in clinical trials: results of a simulation study