Neural Networks: A New Tool for Predictive Models

Abstract

Intensivists face continual pressures, as both practicing clinicians and managers of a limited service, to provide quantitative assessment of patient outcome. In the daily practice of intensive care medicine, patients and their families often expect and may even demand explicit estimates concerning their “chances” of survival from critical illness. In their role as ICU managers, intensivists perform triage to determine which patients will best benefit from ICU admission. Furthermore, increased financial constraints have forced many to increasingly examine the vast amount of expenditures which are allocated to critical care services and ICUs, especially since hard scientific evidence to conclusively validate the clinical benefit of intensive care medicine is often lacking. From a macroeconomic perspective, ICU care is disproportionately high in all developed nations relative to many other inpatient expenditures. In the United States for example, intensive care beds account for 5 to 10% of all hospital beds, yet consume between 20 and 34% of hospital budgets. When extrapolated, this amounts to over 1% of the American gross domestic product (GDP), or $ 67 billion [1].