Improving search over Electronic Health Records using UMLS-based query expansion through random walks

Area of Science:

• Medical Informatics
• Natural Language Processing
• Information Retrieval

Background:

• Electronic Health Records (EHRs) contain vast amounts of unstructured text.
• Searching EHRs requires careful query formulation due to vocabulary variations like synonyms.
• Existing search methods may struggle to bridge the gap between user queries and clinical documentation.

Purpose of the Study:

• To demonstrate that automatic query expansion using the Unified Medicine Language System (UMLS) Metathesaurus enhances EHR search performance.
• To evaluate the effectiveness of a query expansion method based on graph representations and random walks over the UMLS Metathesaurus.
• To improve the accuracy and efficiency of retrieving patient information from EHRs.

Main Methods:

• Utilized a graph representation of lexical units, concepts, and relations from the UMLS Metathesaurus.
• Implemented a query expansion technique based on random walks initiated from query terms within the UMLS graph.
• Tested the method on the TREC Medical Record track datasets from 2011 and 2012.

Main Results:

• Achieved significant improvements in search results over a robust baseline.
• Demonstrated enhanced performance on both the 2011 and 2012 TREC Medical Record datasets.
• Validated the effectiveness of the query expansion approach in a realistic clinical data setting.

Conclusions:

• Automatic query expansion using the UMLS Metathesaurus effectively overcomes vocabulary gaps in EHRs.
• The method successfully expands queries with topically related terms, not just synonyms.
• This approach offers a valuable strategy for improving patient cohort identification through enhanced EHR search.