Modeling Dynamic Patients Variables to Renal Failure in the Intensive Care Unit Using Bayesian Networks

Abstract

Renal failure in the intensive care unit (ICU) is associated with high morbidity and mortality. The Sequential Organ Failure Assessment (SOFA) score is applied in the ICU to track the progression of organ dysfunction. The renal component of the SOFA score employed serum creatinine and urine output to define the stage of its dysfunction. This study aims to explore the relationship between commonly available variables in the ICU together patients' gender and comorbidities to renal failure employing Bayesian Network. The process of building Bayesian Networks involved variable selection, data discretization, and aggregation before structural learning method. The dataset was discretized using equal distance technique into 3 intervals before it was fed into unsupervised structural classification learning techniques. The highest overall precision of 85.1 % was achieved using the unsupervised learning Taboo Order Bayesian Network. Other than creatinine, heart rate, systolic blood pressure, temperature, diabetes mellitus, and hypertension are directly connected with renal failure in this Bayesian Network.