Risk of mortality and cardiopulmonary arrest in critical patients presenting to the emergency department using machine learning and natural language processing