Computer Science · Signal Processing
ICU mortality prediction using time series motifs
Sean McMillan, Chih-Chun Chia, Alexander Van Esbroeck, Ilan Rubinfeld, Zeeshan Syed
See on the mapAbstract
In this paper, we explore the application of motif discovery (i.e., the discovery of short characteristic patterns in a time series) to the clinical challenge of predicting intensive care unit (ICU) mortality. As part of the Physionet/CinC 2012 challenge, we present an approach that identifies and integrates information in motifs that are statistically over-or under-represented in ICU time series of patients experiencing in-hospital mortality. This is done through a three step process, where ICU time series are first discretized into sequences of symbols (by segmenting and partitioning them into periods of low, medium and high measurements); the resulting sequences of symbols are then searched for short subsequences that are associated with in-hospital mortality; and the information in many such clinically useful subsequences is integrated into models that can assess new patients. When evaluated on data from the Physionet/CinC 2012 challenge, our approach outperformed existing clinical scoring systems such as SAPSII, APACHEII and SOFA, with an event 1 score of 0.46 and an event 2 score of 56.45 on the final test set.