Extracting Clinical Relationships from Patient Narratives
Author Information
Author(s): Roberts Angus, Gaizauskas Robert, Hepple Mark, Guo Yikun
Primary Institution: Department of Computer Science, University of Sheffield
Hypothesis
Can supervised machine learning techniques effectively extract clinical relationships from patient narratives?
Conclusion
The study demonstrates that important clinical relationships can be extracted from text using supervised machine learning techniques with accuracy levels approaching those of human annotators.
Supporting Evidence
- The system achieved an average F1 score of 72% for relation extraction.
- Performance was evaluated using a gold standard corpus of oncology narratives.
- Different features were tested to improve extraction performance.
Takeaway
Researchers created a computer program that can read patient notes and find important connections between medical terms, similar to how a human doctor would.
Methodology
The study used support vector machines to train a model on a gold standard corpus of oncology narratives annotated with clinical relationships.
Potential Biases
The reliance on human-annotated data may introduce bias in the training set.
Limitations
The study's results are based on a relatively small sample size and may not generalize to other clinical domains without further validation.
Participant Demographics
The narratives were selected from clinical records of over 20,000 cancer patients.
Digital Object Identifier (DOI)
Want to read the original?
Access the complete publication on the publisher's website