Related Experiment Videos
Inter-patient distance metrics using SNOMED CT defining relationships
Genevieve B Melton1, Simon Parsons, Frances P Morrison
1Department of Surgery, The Johns Hopkins Medical Institutions, Baltimore, MD, USA. gmelton@jhmi.edu <gmelton@jhmi.edu>
Journal of Biomedical Informatics
|March 24, 2006
Summary
Patient similarity metrics using ontology and information content show promise but do not yet match expert performance. These tools aid research and patient care by analyzing cases with standardized concepts.
Area of Science:
- Medical Informatics
- Computational Linguistics
- Clinical Decision Support
Background:
- Patient-based similarity metrics are crucial for case-based reasoning in research and clinical care.
- Ontology and information content principles offer potential for developing advanced similarity metrics.
Purpose of the Study:
- To evaluate the effectiveness of ontology and information content principles in developing patient similarity metrics.
- To compare the performance of novel similarity metrics against expert physician assessments.
Main Methods:
- Converted patient cases (1989-2003) to SNOMED CT concepts.
- Implemented five similarity metrics: percent disagreement, average links, weighted links (information content/descendants, information content/term prevalence), and path distance.
- Utilized three physicians as a gold standard for 30 cases.
Main Results:
- Expert inter-rater reliability was high (0.91).
- Metric correlations with expert performance ranged from 0.27 to 0.30.
- Using SNOMED CT Clinical Findings alone improved correlation to 0.37.
Conclusions:
- Ontology and information content principles enhance similarity metrics but do not fully replicate expert judgment.
- Further development is needed to bridge the performance gap between automated metrics and expert clinicians.