Session SA1.2

Acute Hypotension Episode Prediction Using Information Divergence for Feature Selection and Non-Parametric Methods for Classification

PA Fournier*, JF Roy

Carré Technologies Inc.
Montréal, Canada

Acute hypotension is a critical event that can lead to irreversible organ damage and death. When detected in time, an appropriate intervention can significantly lower the risks for the patient. The objective of this work is to find an automated method to predict acute hypotension episodes, using vital signs such as blood pressure and heart rate acquired in the hours before the event.
We first detail the problem of having more features than samples in the PhysioNet/CinC Challenge 2009 training set. We constrain our analysis to the largest common subset of features available for all patients. We then use information divergence (or Kullback-Liebler divergence) between two distributions to identify the most discriminative features. We use the two most discriminative features in each training set to classify the samples in the test sets using a nearest neighbors algorithm. With this method, we obtained a score of 8/10 for event 1, and 31/40 for event 2. Our preliminary results show that our method leads to significantly better than random results, therefore it increases our information about the samples in the test sets.

(Abstract Control Number: 277)