Evaluating a Tool for Extracting Medical Information from Text
Author Information
Author(s): Zeng Qing T, Goryachev Sergey, Weiss Scott, Sordo Margarita, Murphy Shawn N, Lazarus Ross
Primary Institution: Brigham and Women's Hospital, Boston, MA, USA
Hypothesis
Can the HITEx tool accurately extract principal diagnosis, co-morbidity, and smoking status from discharge summaries?
Conclusion
The HITEx tool demonstrated promising accuracy in extracting medical information from complex discharge summaries.
Supporting Evidence
- HITEx achieved 82% accuracy for principal diagnosis extraction.
- HITEx achieved 87% accuracy for co-morbidity extraction.
- HITEx achieved 90% accuracy for smoking status extraction.
- The combination of HITEx and ICD9 improved accuracy for co-morbidities to 89%.
- HITEx performed better than ICD9 in diagnosis extraction.
Takeaway
Researchers created a tool to read medical notes and found it did a good job at figuring out patients' main health issues and smoking habits.
Methodology
The HITEx tool was evaluated by comparing its extractions from 150 discharge summaries to those made by a human expert.
Potential Biases
The evaluation may be biased due to over-sampling of asthma and COPD cases.
Limitations
The study relied on a single expert for the gold standard, which may introduce bias, and the evaluation was biased towards asthma and COPD cases.
Participant Demographics
Patients with known history of asthma or COPD from a large health care data repository.
Digital Object Identifier (DOI)
Want to read the original?
Access the complete publication on the publisher's website