Rare disease knowledge enrichment through a data-driven approach.

Area of Science:

• Medical Informatics
• Rare Disease Research
• Computational Biology

Background:

• Current rare disease diagnostic resources are often literature-based and limited for clinical application.
• Early suspicion of rare diseases is challenging, delaying the use of available resources.
• A data-driven approach is needed to augment existing rare disease information.

Purpose of the Study:

• To enrich existing rare disease resources by extracting phenotype-disease associations from electronic medical records (EMR).
• To improve the differential diagnosis of rare diseases through data mining.

Main Methods:

• Association rule mining algorithms were applied to EMR data to identify significant phenotype-disease associations.
• Existing rare disease resources (Human Phenotype Ontology and Orphanet) were enriched.
• Phenotype-disease bipartite graphs were constructed for HPO-Orphanet, EMR, and the enriched knowledge base.
• A case study on Hodgkin lymphoma was performed to compare diagnostic performance.

Main Results:

• The enriched knowledge base (HPO-Orphanet+) demonstrated improved performance in differential diagnosis compared to HPO-Orphanet and EMR alone.
• Sensitivity and specificity values were evaluated against a gold standard (eRAM rare disease encyclopedia).
• The HPO-Orphanet+ graph showed promising results when compared with established diagnostic tools like eRAM and Phenomizer.

Conclusions:

• The data-driven approach successfully enriched rare disease knowledge resources using EMR data.
• The enhanced resources provide valuable support for rare disease differential diagnosis.
• Mining EMR data is a viable strategy to improve clinical decision-making for rare conditions.