Temporal Signal Within Vital Signs Precedes Delayed Cerebral Ischemia Diagnosis
Video
Team Information
Team Members
Kalijah Terilli, Research Assistant, Program for Hospital and Intensive Care Informatics, Columbia University Medical Center
Murad Megjhani, Associate Research Scientist, Program for Hospital and Intensive Care Informatics, Columbia University Medical Center
Faculty Advisor: Soojin Park, Associate Professor of Neurology, Program for Hospital and Intensive Care Informatics, Columbia University Medical Center
Abstract
Delayed Cerebral Ischemia (DCI) has been implicated as the most morbid secondary injury after SAH, but difficult to predict and detect. Current prediction is static and relies on scales based on imaging at admission. We seek to develop tools that update and get more accurate over time to aid in earlier identification of patients who will develop DCI, as well as those who are likely to not be affected. We used multilevel linear regression of vital signs and relationships between pairs of vital signs (cross-correlations) to investigate temporal signals as to patients' states embedded within the timeseries data. We then used these vital signs, along with the static characteristics to train machine learning models to classify DCI. We used the best performing model to generate risk scores. These scores correctly identified patient outcomes 70% of the time 12 hours prior to the DCI event. We conclude that there is a time-dependent signal within these vital signs that may add useful information to a dynamic DCI detection tool.
Contact this Team
Contact: Kalijah Terilli (use form to send email)