Ensemble-Based Medical Relation Classification


About

Medical Relation Classification, an information extraction task in the clinical domain that was defined in the 2010 i2b2/VA Challenge, involves determining the relation between a pair of medical concepts (problems, treatments, or tests) such as a treatment improves a problem, a test reveals a problem, etc. Eleven types of intra-sentential pairwise relations are annotated in the 2010 i2b2/VA corpus. A brief description of these relation types are provided in the following table.


Id Relation Example
1 TrIP: Treatment improves medical problem Her pain resolved after surgery
2 TrWP: Treatment worsens medical problem treated with Zofran with no relief
3 TrCP: Treatment causes medical problem Transdermal nitroglycerin caused headache
4 TrAP: Treatment is administered for medical problem start on Decadron 4 mg q6 to prevent swelling
5 TrNAP: Treatment is not administered because of medical problem His Avandia was discontinued secondary to the side effect profile
6 NTrP: No relation between treatment and problem with sutures intact and no erythema or purulence noted.
7 TeRP: Test reveals medical problem A postoperative MRI revealed no remarkable findings
8 TeCP: Test conducted to investigate medical problem An ultrasound was done to rule out cholestasis
9 NTeP: No relation between test and problem Throughout the stay his labs remained normal and his pain controlled.
10 PIP: Medical problem indicates medical problem with a moderate-sized, dense, fixed inferior defect indicative of scar
11 NPP: No relation between paired medical problems He is somewhat cantankerous and demanding of the nurses.
Table - The 11 medical relation types for classification.

Our ensemble-based approach to the Medical Relation Classification task is described in the following paper.

Jennifer D'Souza and Vincent Ng. 2014. Ensemble-Based Medical Relation Classification. In Proceedings of COLING. pp. 1682-1693.

This page is the access site of the complete set of cue phrases developed for the task of Medical Relation Classification.


Download

Download file with cue phrases for Medical Relation Classification.

Funding Statement

This work was supported in part by NSF Grants IIS-1147644 and IIS-1219142. Any opinions, findings, or conclusions expressed above are those of the authors and do not necessarily reflect the views or official policies of NSF.

Questions

Questions, feedback, and suggestions for improvement are welcome via email contact.