Intelligent alarms for patient supervision

Abstract

This work presents a proposal for creating intelligent alarms that offer more efficient support to medical staff than the threshold alarms currently available in commercial monitors. Our alarms make it possible to handle the uncertainty and imprecision that are characteristic of the medical domain, reason over the temporal evolution of physiological variables, and incorporate information from a number of variables into one single alarm. The proposal is based on a structural pattern recognition model (the MFTP model) which allows certain monitoring criteria in a computational representation to be captured and identified over the evolution of patients' physiological variables. The description of the criteria is carried out using a graphical tool (TRACE), which is sufficiently intuitive to be used by physicians without the need for assistance.