The challenges in creating critical-care databases.

Abstract

As critical care is critical, the demands for patient monitoring are high. The methods need to function properly over a large variety of possible physiological conditions in an environment full of sources for technical artifacts, and they need to support clinical reasoning. Several iterations are usually required in the method development to meet these demands. First, the proposed method is developed and tested with simulated or optimal test data on a proof-of-principle basis. When this level is passed, the method needs to be tested offline with more realistic and nonoptimal data. Only after passing these levels may clinical trials be conducted. Usually, these developments require a significant period of time. Availability of extensive, well-characterized, and well-documented real physiological data during critical care would potentially reduce the amount of time required for the first phases of the method development. Such databases would also enable bench validation of any new methods and objective comparison between different methods. Recently, there have been some attempts to collect large signal databases during critical care. In this article, we aim to discuss the special requirements for collecting signal databases in critical care and to summarize the main features of the existing databases.