Patient Cohort Retrieval using Transformer Language Models.

Area of Science:

• Medical Informatics
• Artificial Intelligence
• Natural Language Processing

Background:

• Patient cohort retrieval (CR) is crucial for clinical research.
• Current methods often require extensive feature engineering and domain expertise.
• Electronic health records (EHRs) contain vast amounts of patient data.

Purpose of the Study:

• To assess the efficacy of deep learning-based language models for patient cohort retrieval.
• To propose a framework for CR using neural language models.
• To eliminate the need for explicit feature engineering and domain expertise in CR.

Main Methods:

• Mapping the CR task to a document retrieval problem.
• Applying various deep neural network models.
• Utilizing deep learning-based language models for EHR data analysis.

Main Results:

• Deep learning models demonstrated superior performance compared to the BM25 baseline.
• Several models achieved significant improvements across various evaluation metrics.
• The proposed framework successfully retrieved relevant patient cohorts.

Conclusions:

• Deep learning language models are effective for patient cohort retrieval from EHRs.
• The proposed framework offers a robust and efficient alternative to traditional methods.
• This approach reduces reliance on manual feature engineering and domain expertise.