Enhancing Clinical Information Retrieval through Context-Aware Queries and Indices

Area of Science:

• Health Informatics
• Information Retrieval
• Clinical Data Science

Background:

• The increasing volume of electronic health records (EHRs) necessitates efficient search capabilities for clinical practice, research, and administration.
• General information retrieval (IR) methods struggle with the complexity of clinical data, failing to account for contextual factors like negation and temporality.
• Accurate identification of relevant clinical narratives at the point of care is critical for effective decision-making.

Purpose of the Study:

• To improve the accuracy and efficiency of searching large-scale EHRs.
• To address the limitations of general IR methodologies in the clinical domain by incorporating semantic context.
• To evaluate the impact of integrating contextual information on retrieval quality, execution time, and storage requirements.

Main Methods:

• Implemented a solution using Elasticsearch to incorporate semantic contexts (negation, temporality, subject) into IR indexing and scoring.
• Conducted experiments comparing the new approach against baseline methods.
• Evaluated the impact on retrieval quality (relevance), query execution time, and storage needs.

Main Results:

• Achieved a 5.1-23.1% improvement in retrieval quality (relevance).
• Demonstrated a 35% faster query execution time compared to baseline approaches.
• Observed a 1.5-2 times increase in space requirements and a 3-fold increase in indexing time.

Conclusions:

• Incorporating contextual information significantly enhances the relevance of clinical information retrieval (IR) from EHRs.
• The proposed solution shows promising scalability with manageable increases in time and space costs.
• This approach offers a valuable advancement for clinical decision support systems and medical research.