Validity of composite end points in clinical trials