Learning to Identify Rare Disease Patients from Electronic Health Records.

Area of Science:

• Medical Informatics
• Computational Biology
• Rare Disease Research

Background:

• Identifying rare disease patients in large electronic health records (EHRs) is challenging due to limited confirmed diagnoses.
• Existing prediction models struggle with the scarcity of 'gold standard' labeled data.

Purpose of the Study:

• To present a novel cascade learning methodology for accurate prediction model induction from noisy 'silver standard' labeled data.
• To enable robust learning from unconfirmed diagnostic evidence in population-scale databases.

Main Methods:

• Combines unsupervised feature selection, supervised ensemble learning, and unsupervised clustering.
• Utilizes a cascade learning approach to handle noisy labels effectively.
• Applies the methodology to a case study for lipodystrophy patient detection in EHRs.

Main Results:

• The proposed algorithm demonstrates superior performance compared to state-of-the-art prediction techniques.
• Successfully identifies lipodystrophy patients in a large-scale EHR database.
• Facilitates the discovery of previously undiagnosed patients.

Conclusions:

• The cascade learning methodology offers a robust solution for rare disease patient identification in EHRs.
• This approach enhances the accuracy of prediction models derived from noisy data.
• Enables scalable and effective rare disease case finding in real-world healthcare data.