Abstract 219

Bedside data integration in ICU environments is an unrealized Pediatric Informatics dream. Databases to integrate clinical information must accommodate time-dependent, highly variant, repeated measures data with many missing (null) values. Some database engineers suggested that null values be disallowed; impossible for clinical databases. Critical care data exists in a clinically and legally deterministic date/time matrix. Titrations of ventilator support, fluid/vasopressor infusions and electrolyte balance create constant change. Because any change can potentially crash the patient, current practice is to try to maintain a timely sequential database by creating a paper flowsheet to help bedside caregivers detect and diagnose deterioration. The flowsheet charting can consume hours/day of finding, downloading and physically copying needed values from a series of stand-alone computerized devices and data flows. Commercial database structures must be reprogrammed to accept multiple, timed streams of monitoring, pump and site-specific ancillary data. Comprehensive computerization of critical care data lags far behind business process automation.

Our database design represents a logical model that is structured for expansion to capture the 100 to 200 NICU clinical variables that neonatologists use for real-time clinical decision making. The exact requirements for organizing data imported directly from computerized bedside devices can't be precisely known yet because open communication standards for ventilators & infusion pumps are a work in progress. Institutional constraints, data vocabulary and dictionaries will differ by unit. For maximum flexibility in logical and practical design, we created new lookup tables at each point where options occurred for adding expansibility. Attributes common to sub-entities in "is-a" hierarchies were stored in the highest entity. A similar design strategy was used for fluids, blood products, medications and laboratory tests/results. The relational database is implemented in Oracle8 with a minute-by-minute date/time dimension that was hard-wired to simultaneous happenings by defining "startDT" (date/time) and "endDT" as part of the primary key. When any parameter changes the "endDT" becomes the "startDT" of the new state. Oracle8 supports unitary date/time notation and arithmetic well. The database "testbed" was neonatal ECMO research data generated by the care of 29 babies in a hypertension study that had 64 data variables observed hourly. When run on a 400 MHz Oracle PC server, the database sorts defined hypertensive values from 6500 rows in far less than 1 second. The speed and low cost of PC computers, new functionality in Oracle8 and the promise of industry acceptance of IEEE device standards suggest that real-time bedside integration of automatically computerized clinical data is nearly an achievable goal.

Supported by NLM Grant 5T15LM07124.