NimbleMiner

NimbleMiner generates lexicons of synonymous and related clinical terms by leveraging word embedding models applied to large-scale clinical text.

Key Features:

• Word embedding models: Uses word embedding models that represent words in continuous vector spaces to capture semantic relationships.
• Clinician interaction: Enables clinicians to interact with embedding models to identify terms similar to specified keywords.
• Lexicon output: Produces approximately 50 top potentially similar terms per validated term for lexicon creation.
• Parameter optimization: Supports parameter experiments; larger word window width sizes (n = 10) were found optimal in reported experiments.
• Corpus-scale processing: Processes large clinical corpora, demonstrated on 1,149,586 homecare visit records extracted from a US homecare agency.
• Rapid vocabulary construction: Enabled building a comprehensive fall-history vocabulary in about two hours.

Scientific Applications:

• Patient fall history extraction: Identifying and expanding terminology related to patient fall history in homecare visit notes.
• Clinical vocabulary enrichment: Expanding clinical vocabulary for nursing documentation and communication by discovering synonymous and related terms.
• Lexicon generation for clinical text analysis: Generating lexicons of synonymous and related terms to support downstream clinical text analysis.

Methodology:

Employs word embedding models that represent words in continuous vector spaces; supports interactive identification of terms similar to specified keywords; includes experiments varying model parameters, where larger word window widths (n = 10) yielded optimal results and provided ~50 top potentially similar terms per validated term.