Physicians in the emergency department must rapidly gather and synthesize large amounts of data from disparate sources in order to make treatment decisions, all while constantly switching between patients. Subtle changes in clinical parameters over time may be easily overlooked by even the most vigilant clinician. This paper proposes the design of algorithms for real-time surveillance of patients medical records. The challenge is to make accurate inferences from the diverse set of signals available for each patient, putting these signals in the context of the patients existing medical history. We describe initial results for a surveillance algorithm applied to early detection of sepsis.