Cognitive DDx Assistant in Rare Diseases.

Area of Science:

• Medical Informatics
• Computational Biology
• Rare Disease Research

Background:

• Over 6,000-7,000 rare diseases exist, posing diagnostic challenges due to their infrequency and complexity.
• Diagnosing rare diseases is often time-consuming, resource-intensive, and requires specialized centers, limiting access for many patients.
• General practitioners may encounter rare disease patients infrequently, highlighting the need for advanced diagnostic support.

Purpose of the Study:

• To develop and evaluate a cognitive assistant for differential diagnosis in rare diseases.
• To leverage knowledge graph technology for comprehensive rare disease information and evidence-based diagnosis.
• To improve the accuracy and efficiency of rare disease identification in clinical practice.

Main Methods:

• Constructed a knowledge graph integrating data from sources like ICD-10, Orphanet, PubMed, and patient data.
• The system utilizes knowledge graph technology to provide a weighted list of potential diagnoses.
• Tested the cognitive assistant with 101 published rare disease cases to assess diagnostic accuracy.

Main Results:

• The knowledge graph comprised over 500,000 nodes, representing extensive rare disease data.
• Achieved 79.5% accuracy in identifying the correct rare disease within the top 1% of diagnostic possibilities.
• Further refinement improved results, with 74% of cases correctly diagnosed within the top 6 suggestions on average.

Conclusions:

• Knowledge graph technology offers a powerful solution for cognitive assistance in rare disease differential diagnosis.
• Data curation is a critical factor influencing the accuracy of rare disease diagnostic tools.
• The developed system demonstrates potential for practical clinical application to aid in diagnosing rare conditions.