University of California

Rebeca Salas-Boni

Xiao Hu

Yong Bai

Methods for predicting patient deterioration or clinical events by detecting patterns in heterogeneous temporal clinical data streams that are predictive of certain clinical end points and matching the patient state with those patterns are described. The detected patterns, referred to as SuperAlarm triggers, are a predictive combination of frequently co-occurring monitor alarms, conditions and laboratory test results that can predict patient deterioration for imminent life-threatening events. SuperAlarm triggers may also exhibit patterns in the sequence of SuperAlarms that are triggered over the monitoring time of a patient. Sequential patterns of SuperAlarm triggers may also indicate a temporal process of change in patient status. SuperAlarm based alerts will also greatly reduce the number of false alarms and alarm fatigue compared to conventional alarms.

Recognizing predictive patterns in the sequence of superalarm triggers for predicting patient deterioration

Methods for automatically determining whether a patient monitor alarm will sound from a true or false signal, in particular from ventricular tachycardia (VT) and suppressing false alarms without eliminating any true alarms are presented. A multiresolution wavelet is extracted from a raw ECG waveform. Features are then extracted from the wavelets that account for summary statistics, noise, areas under the curve and summary statistics of the KL-divergence of the power spectra density between every two ECG leads. A classifier can be then be trained and its performance measured.

Growing community of inventors

San Francisco, CA, United States of America

Rebeca Salas-Boni