Resumen de Feature Selection and Machine Learning for Predicting Multi-drug Resistance just After ICU Admission
J.Tarancón Rey, I. Mora Jiménez, J. Álvarez Rodríguez, C. Soguero Ruiz
and greatest threats to the global health system assuming an increase in the mortality rate as well as health care costs. This situation is especially relevant in intensive care units (ICU) where the serious state of health of patients makes them more vulnerable. It is therefore necessary to identify MDR bacteria with the aim of taking appropriate action as soon as possible. Since cultures require 48-72 hours to obtain results, in this paper machine learning (ML) techniques are proposed to identify earlier whether a patient will develop MDR in the first 48 hours of admission. The use of the 48-hour threshold is due to the fact that it is the threshold used by clinical experts to discern between patients who have acquired the MDR in the ICU and those who have not. For this purpose, clinical and demographic features from the patient are considered (12 in total). The results of the feature selection identified a representative set of 24 features which can be used to identify MDR and non-MDR patients. That means that almost 75% of the initial features (95 after data preprocessing) are not informative for this task. We can conclude that the application of ML techniques contributes to the classification of patients according to the development of MDR during their first hours in ICU stay.
