Forecasting Demand for Pediatric Critical Care

Johns Hopkins University, Baltimore MD

Investigators

Abstract

This award is funded under the American Recovery and Reinvestment Act of 2009 (Public Law 111-5) ************************************************************************************ This grant provides funding for the development of a pediatric intensive care unit (PICU) resource demand forecasting model ('PICU-Forecast'). The model will forecast demand for nurse staffing and bed resources in real-time in 3 hour increments up to 72 hours into the future. PICU-Forecast is comprised of patient inflow and outflow models and will be designed using retrospective data collected from multiple hospital information systems. The inflow model will predict unscheduled arrivals to the PICU, their probability of admission and nursing care requirements. The outflow model will predict PICU patient length-of-stay based on physician orders (i.e., medications, ventilation, etc.). Inflow and outflow models will be integrated to create the global 'PICU-Forecast' model. PICU-Forecast will be cross-validated at the original site, then re-calibrated and externally validated at an alternate site. Successful completion of the research will result in the development of a generalized tool that optimizes scarce critical care resources to improve access and outcomes for children who are severely ill or injured. An accurate PICU-Forecast model has potential to; (1) improve daily management decision making, (2) improve coordination across the children?s hospital system, and (3) facilitate proactive interventions to eliminate bottlenecks, avoid underutilization and improve access. The forecasting methods, which use readily available patient and hospital information, are potentially applicable to any hospital unit with capable information technology. In addition, an expected methodological contribution of the study will be the application of survival analysis methods in real-time to produce updated forecasts based on continuously changing patient and hospital information.