Controlling type I error rates in multi-arm clinical trials: A case for the false discovery rate