An Alarm System for Death Prediction

An Alarm System for Death Prediction

Rüdiger Brause (Computer Science Department, Goethe University, Frankfurt, Germany) and Ernst Hanisch (Department of Surgery, Asklepios Klinik Langen, Academic Teaching Hospital of Goethe University,Frankfurt, Germany)
DOI: 10.4018/ijmstr.2013040102
OnDemand PDF Download:
$37.50

Abstract

The clinical treatment of sepsis is one of most severe issues in hospitals. Unfortunately, until now it has not been possible to significantly reduce the mortality rate of severe forms of sepsis like septic shock, which is as high as 50-60% worldwide. Often, the diagnosis and awareness for possible implications of sepsis can be facilitated by an automated online diagnosis. This contribution reports the development of a monitoring alarm system for the individual prediction of death based on the data of 382 patients with septic shock. The paper discusses the pros and cons of such a prediction system used in a medical environment, its principal usage issues and implementation.
Article Preview

2. The Alarm System

The goal of the data analysis was the development of a prediction system for the individual mortality prognosis. Such systems can be used in the emergency case or for advances in treatment by automatic state monitoring. In our case, the analysis goal was two-fold: first, we liked to trace back the causes and influences of several clinical variables like coagulation or thrombocyte level to the patient outcome, and second, we aimed to build an alarm system which rings an alarm as soon as a bad state is reached. In this contribution, we focus on the latter case.

The main problem for a medical alarm system is the availability of data. Unfortunately, there is no clinical standard for bedside monitoring or patient data bases in Germany. For this reason, for our analysis of septic shock (which is a rare event) we had to initiate a multi-center study and concentrated on those 140 variables only, which are currently available in clinical routine. All gene tests or other special features were practically out of reach.

The next problem after obtaining the data is the question: What kind of analysis system should we use? It is well known that most metabolic processes are non-linear. Therefore, the usage of all linear methods like correlation analysis is not adequate. Instead, we used the method of formal neural networks.

Complete Article List

Search this Journal:
Reset
Open Access Articles: Forthcoming
Volume 5: 4 Issues (2017): 2 Released, 2 Forthcoming
Volume 4: 4 Issues (2016)
Volume 3: 4 Issues (2015)
Volume 2: 4 Issues (2014)
Volume 1: 4 Issues (2013)
View Complete Journal Contents Listing