Information Technology Reference
In-Depth Information
Predicting Nosocomial Infection
by Using Data Mining Technologies
Eva Silva, Luciana Cardoso, Filipe Portela, António Abelha,
Manuel Filipe Santos, and José Machado
Algoritmi Research Centre, University of Minho, Portugal
{a 58508,a55524}@alunos.uminho.pt, {abelha,jmac}@di.uminho.pt,
{cfp,mfs}@dsi.uminho.pt
Abstract. The existence of nosocomial infection prevision systems in
healthcare environments can contribute to improve the quality of the healthcare
institution and also to reduce the costs with the treatment of the patients that
acquire these infections. The analysis of the information available allows to
efficiently prevent these infections and to build knowledge that can help to
identify their eventual occurrence. This paper presents the results of the
application of predictive models to real clinical data. Good models, induced by
the Data Mining (DM) classification techniques Support Vector Machines and
Naïve Bayes, were achieved (sensitivities higher than 91.90%). Therefore, with
these models that be able to predict these infections may allow the prevention
and, consequently, the reduction of nosocomial infection incidence. They
should act as a Clinical Decision Support System (CDSS) capable of reducing
nosocomial infections and the associated costs, improving the healthcare and,
increasing patients' safety and well-being.
Keywords: Clinical Decision Support System; CRISP-DM; Data Mining;
Knowledge Discovery in Databases; Nosocomial Infection.
1 Introduction
A nosocomial infection is an infection that occurs during the 48 hours after the
patient's hospitalization, during three days after his/her discharge or during the 30
days after a surgery. Also, this infection was not present or in incubation at the
moment of the patient's admission [1] [2] [3]. These infections also include healthcare
institutions' occupational infections [1].
A patient with a nosocomial infection stays more time hospitalized resulting in an
additional financial burden for the healthcare institution [2] [3]. Moreover, these
infections have a great impact on patients' morbidity and mortality [2] [3]. Thus, it is
very important to prevent nosocomial infections.
The present work arises from the need of preventing the occurrence of nosocomial
infections. This prevention can be accomplished by performing predictions using data
stored in the database and containing attributes capable to characterize the patient
health status as well as his/her hospitalization period and the procedures performed
Search WWH ::




Custom Search