Predicting Readmissions, Mortality, and Infections in the ICU Using Machine Learning Techniques

Serie

  • Documentos de Trabajo Quantil

Resumen

  • Health care at the Intensive Care Unit (ICU) is both expensive for hospitals and strenuous for doctors. Early detection of risk factors associated to readmissions, mortality, and infections in the ICU, can improve patient care quality and reduce costs in the long-run. In this article we use machine learning techniques to predict those three outcomes using patient-level data of the ICU of a high complexity hospital in Colombia. Our results show that pathologies of the aorta, cancer, neurologic and respiratory diseases as well as invasive procedures such as dialysis, tracheostomy, and bronchoscopy are positively correlated to the probability of readmission, death, and catheter infections in the ICU. The area under the ROC curve for the first outcome ranges between 71 and 74%, for the second between 76 and 81%, and for the third between 88 and 92%. We estimate a model that competes against the APACHE II scoring system and achieve the same predictive power using less information about the patient.

fecha de publicación

  • 2016-08

Líneas de investigación

  • Catheter Infections
  • Intensive Care Unit
  • Machine Learning
  • Mortality
  • Readmissions

Issue

  • 15074