Abstract: Non-critical patient deterioration tends not to alter vital signs, and this is addressed through the compensatory reflex. However, severe deterioration alters a patient’s physiology noticeably. In this paper, we present an unsupervised clustering approach to identify abnormal vital sign trends. To accomplish this, we used unsupervised random forest dissimilarity matrix with day-level vital sign abstraction and k-means clustering. The clustering approach identified groups of vital signs frequently observed among expired patients.
Learning Objective 1: Formulate an annotation strategy using abnormal vital sign patterns for labeling patient deterioration
Learning Objective 2: Evaluate feasibility of using an unsupervised learning approach to cluster abnormal clinical variables
Dae Hyun Lee (Presenter)
University of Washington
Meliha Yetisgen, University of Washington