Leveraging Collaborative Filtering to Accelerate Rare Disease Diagnosis.

Area of Science:

• Medical Informatics
• Computational Biology
• Rare Disease Research

Background:

• Rare diseases affect fewer than 200,000 patients in the USA.
• Misdiagnosis and delayed diagnosis are significant challenges for rare disease patients.
• Lack of provider knowledge and experience contributes to diagnostic delays.

Purpose of the Study:

• To develop and evaluate a system for accelerating rare disease diagnosis.
• To leverage patient phenotypic information from electronic medical records (EMR).
• To assist diagnosis by characterizing patient phenotypes.

Main Methods:

• Proposed a collaborative filtering system enhanced with natural language processing (NLP) and semantic techniques.
• Utilized four similarity measurements and two neighborhood algorithms.
• Analyzed an unstructured patient cohort from Mayo Clinic (2010-2015).

Main Results:

• Demonstrated the effectiveness of collaborative filtering using phenotypic information.
• Showed the ability to stratify patients with similar rare diseases.
• Preliminary results indicate potential for improved diagnostic accuracy.

Conclusions:

• Phenotypic characterization from EMR data can aid rare disease diagnosis.
• Collaborative filtering offers a promising approach for identifying similar rare disease cases.
• Further development can enhance diagnostic support for rare conditions.