Homepage > Research abstracts > The effect of an automated knowledge-based textual summarization system, on the routine work at the ICU
The effect of an automated knowledge-based textual summarization system, on the routine work at the ICU
Researchers: Yuval Shahar1
- Ben-Gurion University of the Negev
Background: Creating clinical summaries when transferring patients is important, requires much time and effort, and needs support to include all key details.
We developed an innovative knowledge-based summarization system, which automatically generates a textual summary from an electronic medical record.
We developed an innovative knowledge-based summarization system, which automatically generates a textual summary from an electronic medical record.
Objectives: 1. Compare the quality and completeness of automatically generated textual summaries to that of summaries composed by care-providers.
2. Assess the functionality of automatically generated textual summaries, regarding potential use by the target clinicians.
2. Assess the functionality of automatically generated textual summaries, regarding potential use by the target clinicians.
Method: We collected textual discharge summaries and corresponding patient data from the Hadassa ICU’s Metavision system.
We acquired ICU-specific medical knowledge.
We then performed the following process:
1. A temporal Abstraction module created knowledge-based temporal-abstractions from the time-stamped raw data;
2. Missing significant data were abducted by an Abductive Reasoning module;
3. A Pruning component, selected, based on the context, relevant data;
4. Several modules grouped and organized the data and transformed the structured data into a textual output format appropriate for the ICU domain.
We acquired ICU-specific medical knowledge.
We then performed the following process:
1. A temporal Abstraction module created knowledge-based temporal-abstractions from the time-stamped raw data;
2. Missing significant data were abducted by an Abductive Reasoning module;
3. A Pruning component, selected, based on the context, relevant data;
4. Several modules grouped and organized the data and transformed the structured data into a textual output format appropriate for the ICU domain.
Findings: The summaries generated by the system, comparing to summaries composed by the care-providers without the system’s support:
1. Are more complete, although not quite of the same quality;
2. Are functionally equivalent to a composed letter;
3. When used as drafts to be edited by a clinician, have the potential to save time and increase the satisfaction of the staff with the process of generating a discharge summary.
1. Are more complete, although not quite of the same quality;
2. Are functionally equivalent to a composed letter;
3. When used as drafts to be edited by a clinician, have the potential to save time and increase the satisfaction of the staff with the process of generating a discharge summary.
Conclusions: An automated system can generate summaries of high completeness and functionality. Initial results suggest that using them as drafts to be edited by the clinician, can exploit the best of human and machine.
Recommendations: Automated summary generation systems have a potential to be effectively used for performing routine tasks such as generating discharge summaries, admission letters, and transfer reports.
Research number: A/7/2015
Research end date: 12/2017