Technical considerations for evaluating clinical prediction indices: a case study for predicting code blue events with MEWS