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