Abstract: Descriptive data mining techniques, such as temporal pattern mining and clustering analysis, may provide clinicians with deeper understanding of the associations and/or correlations among clinical variable that may predict patient outcomes. Temporal patterns and clustering of patient data evolve due to changes in the patient’s clinical status and proper analytical techniques for their analysis should be informative and give stable results over time. Here we report the application of our Frequency pattern mining, Clustering, and Graph database (FCG) method to extract temporally stable and accurate patterns and clusters of patients, as described by clinical and demographic variables measured in an ICU setting. The accuracy of patterns and clusters is measured by their ability to correctly discriminate between positive (survival) and negative (not-survival) cases at any given time.

Learning Objective 1: The integration between pattern mining and clustering analysis is an effective descriptive tool.

Learning Objective 2: The clusters of patients are evolved over the time due to patient changes of status.

Learning Objective 3: Using patient outcome is a great measurement to identify high quality clusters.


Samir Abdelrahman (Presenter)
University of Utah

Julio Facelli, University of Utah
Bruce Bray, University of Utah
Rashmee Shah, University of Utah
Guilherme Del Fiol, University of Utah

Presentation Materials: