Shock Prediction Using Vital Sign Time Series

• Activity Number: 470 - Biomarker Evaluation and Winning Student Papers on Medical Devices and Diagnostics
• Type: Contributed
• Date/Time: Wednesday, July 31, 2019 : 8:30 AM to 10:20 AM
• Sponsor: Section on Medical Devices and Diagnostics
• Author(s): Iris Bennett* and Bill Rand
• Companies: North Carolina State University and North Carolina State University
• Keywords: time series; causal state models; vital signs; shock; decision support

Abstract

Early identification of shock can decrease mortality among trauma patients, but abnormal vital signs occur too late in shock to be useful predictors on their own. Instead of waiting for vital signs to become abnormal, we leverage their patterns over time to predict future abnormal values and shock. Five vital signs were included in the analysis: heart rate, peripheral capillary oxygen saturation, systolic blood pressure, diastolic blood pressure, and the ratio of pulse pressure to systolic pressure. Time series data taken at five-minute intervals on the vital signs of 111,260 ICU patients were used to build time-series models for each vital sign, which we used to predict vital signs up to 25 minutes into the future. Predictive accuracy was compared between the full testing set and a subset of observations near a patient entering shock, called critical regions. While trends in accuracy over amount of past data used and prediction time ahead were similar between the two cases, accuracy was higher in the full testing set than the critical regions. Our goal is to create a decision support system that, combined with standard protocols, can help doctors rapidly respond to critical cases.