Cumulative Time Series Representation for Code Blue prediction in the Intensive Care Unit

Area of Science:

• Biomedical Engineering
• Clinical Informatics
• Data Science in Healthcare

Background:

• Hospital patient monitors generate numerous false alarms, negatively impacting patient care and clinician workload.
• Previous research developed 'SuperAlarms' combining monitor alarms and lab tests for code blue prediction.
• A need exists for improved methods to accurately predict critical events like code blue events.

Purpose of the Study:

• To develop and evaluate a novel time series representation, weighted accumulated occurrence representation (WAOR), for predicting code blue events.
• To compare the performance of WAOR with traditional term frequency (TF) methods in predicting critical events.
• To assess the algorithm's effectiveness in an intensive care unit (ICU) setting.

Main Methods:

• Developed a novel time series representation, WAOR, incorporating cumulative effects and temporality.
• Utilized both TF and WAOR representations as inputs for an L1 regularized logistic regression classifier.
• Evaluated the algorithm's performance online using an independent dataset, measuring sensitivity, work-up to detection ratio, and false alarm rates.

Main Results:

• The WAOR representation demonstrated superior performance compared to the TF approach.
• The cumulative representation achieved sensitivity comparable to previous methods.
• Key metrics such as the work-up to detect ratio and false alarm proportion were improved with the WAOR method.

Conclusions:

• The novel WAOR time series representation is effective for predicting code blue events in the ICU.
• WAOR offers an improvement over traditional methods by better accounting for cumulative effects and temporality.
• This approach has the potential to reduce false alarms and enhance patient care in critical settings.