Variability analysis with analytics applied to physiological data streams from the neonatal intensive care unit

Late onset neonatal sepsis (LONS) is one clinical condition that shows promise for earlier onset detection through the analysis of physiological signals. However, current work on Heart Rate Variability (HRV) analysis does not discuss the impact of narcotics and other drugs on early identification of sepsis. We present results of a pilot retrospective data mining study of neonatal intensive care unit patients using a dataset of 30 second spot readings. We derive analytics by creating temporal abstractions of hourly summaries for HRV and respiratory rate variability (RRV). Using representative patient examples, we illustrate an analytics user interface design that shows 1) the potential in using our HRV analytics for early identification of LONS with 30 second spot readings; and 2) that based on initial pilot results, reporting analytics for HRV and RRV concurrently adds value to HRV analysis by distinguishing between patients with low HRV due to imminent sepsis and those patients with low HRV due to the presence of confounding factors such as surgery and narcotics.