Medical Device Alarms - The Statistician

Abstract

The high rate of false alarms triggered by medical devices raises concerns over the reliability of current clinical alarm systems. In close cooperation between physicians, statisticians and engineers within the Collaborative Research Centre SFB 475, we have worked on improved alarm algorithms for online monitoring in intensive care. In current clinical practice there are procedures based on simple fixed thresholds, which give many false alarms because of e.g. measurement artifacts and short-term fluctuations around the pre-specified thresholds. Most of these unnecessary alarms can be avoided by applying robust filtering techniques. A further challenge is the high dimension of the data, which can be handled by suitable techniques for dimension reduction. The new alarm algorithms have been developed and validated against large annotated clinical data sets.