Dynamic Bayesian Networks to predict sequences of organ failures in patients admitted to ICU.

Abstract

Multi Organ Dysfunction Syndrome (MODS) represents a continuum of physiologic derangements and is the major cause of death in the Intensive Care Unit (ICU). Scoring systems for organ failure have become an integral part of critical care practice and play an important role in ICU-based research by tracking disease progression and facilitating patient stratification based on evaluation of illness severity during ICU stay. In this study a Dynamic Bayesian Network (DBN) was applied to model SOFA severity score changes in 79 adult critically ill patients consecutively admitted to the general ICU of the Sant'Andrea University hospital (Rome, Italy) from September 2010 to March 2011, with the aim to identify the most probable sequences of organs failures in the first week after the ICU admission. Approximately 56% of patients were admitted into the ICU with lung failure and about 27% of patients with heart failure. Results suggest that, given the first organ failure at the ICU admission, a sequence of organ failures can be predicted with a certain degree of probability. Sequences involving heart, lung, hematologic system and liver turned out to be the more likely to occur, with slightly different probabilities depending on the day of the week they occur. DBNs could be successfully applied for modeling temporal systems in critical care domain. Capability to predict sequences of likely organ failures makes DBNs a promising prognostic tool, intended to help physicians in undertaking therapeutic decisions in a patient-tailored approach.