AnnoDash, a clinical terminology annotation dashboard

Area of Science:

• Health Informatics
• Bioinformatics
• Clinical Data Management

Background:

• Standard ontologies are essential for health data interoperability and multisite analysis.
• Current methods for mapping concepts to ontologies are often manual, time-consuming, and lack contextualization.
• Ad hoc approaches for contextualizing concepts within source data hinder efficient analysis.

Purpose of the Study:

• To introduce AnnoDash, a flexible dashboard designed to streamline the annotation of clinical concepts with ontology terms.
• To enhance the accuracy and efficiency of mapping clinical data to standard ontologies.
• To provide a user-friendly tool for non-technical users to improve health data harmonization.

Main Methods:

• Developed AnnoDash, a web-based dashboard utilizing text-based similarity and large language models for ontology term matching.
• Integrated a visualization interface to display concept-associated observations and time-series plots for disambiguation.
• Evaluated AnnoDash qualitatively using MIMIC-IV measurements against standard ontologies like SNOMED CT and LOINC.

Main Results:

• AnnoDash effectively identifies likely ontology matches for clinical concepts, improving ranking through large language models.
• The dashboard facilitates concept disambiguation by visualizing associated data and clinical measurements.
• Qualitative evaluation demonstrated AnnoDash's utility in annotating concepts across multiple standard ontologies.

Conclusions:

• AnnoDash is an advanced clinical terminology annotation tool that enhances health data harmonization.
• The tool simplifies the mapping of clinical data to standard ontologies, promoting data consistency.
• AnnoDash is freely available with modular code for user customization and extension.