Predicting sepsis: A comparison of analytical approaches

Abstract

Sepsis is a significant cause of mortality and morbidity and is often associated with increased hospital resource utilization, prolonged intensive care unit and hospital stay. With advances in medicine, there is now aggressive goal oriented treatments that can be used to help patients that may be at risk for sepsis. To predict this risk, we hypothesized that commonly used univariate and multivariate models could be enhanced by using multiple analytic methods to providing greater precision. As a first step, we analyze data about patients with and without sepsis using multiple regression, decision trees and cluster analysis. We compare the predictive accuracy of the three different approaches in predicting which patients are likely (or not likely) to develop sepsis. The precision analysis suggests that decision trees may provide a better predictive model than either regression methods or cluster analysis. © 2011 ICST Institute for Computer Science, Social Informatics and Telecommunications Engineering.