Related Experiment Videos
Building ICU artifact detection models with more data in less time.
C L Tsien1, I S Kohane, N McIntosh
1BWH/MGH Harvard Affiliated Emergency Medicine, Boston, MA, USA.
Proceedings. AMIA Symposium
|February 5, 2002
Summary
Reducing false intensive care unit (ICU) alarms is crucial. Using aggregated 1-minute monitor data for decision tree models effectively detects artifacts, saving processing time without losing accuracy.
Area of Science:
- Biomedical Engineering
- Clinical Informatics
- Data Science
Background:
- High rates of false alarms in intensive care units (ICUs) lead to alarm fatigue.
- Effective artifact detection is essential for reliable patient monitoring.
Purpose of the Study:
- To investigate the impact of data granularity on decision tree models for artifact detection.
- To compare the performance of models built using 1-second versus 1-minute temporal data.
Main Methods:
- Decision tree induction was used for multiple signal integration of temporal monitor data.
- Models were developed using 1-minute data and tested on 1-second data.
- Temporal data compression via arithmetic mean was employed.
Main Results:
- Models developed from 1-minute data maintained effectiveness when tested on 1-second data.
- Processing time for model development was significantly reduced using 1-minute data.
- Data compression did not compromise the learning or artifact detection capabilities.
Conclusions:
- Aggregating temporal data to 1-minute intervals is an effective strategy for improving the efficiency of ICU alarm artifact detection models.
- This approach allows for processing more monitoring hours in less time, enhancing knowledge discovery without sacrificing performance.