Automatic trial eligibility surveillance based on unstructured clinical data.

Area of Science:

• Biomedical Informatics
• Clinical Trial Management
• Natural Language Processing

Background:

• Patient enrollment in clinical trials is a significant challenge, impacting research timelines and costs.
• Accurate identification of eligible patients is crucial for successful clinical trial recruitment.

Purpose of the Study:

• To assess the feasibility of automatically detecting patient eligibility for breast cancer clinical trials.
• To map coded eligibility criteria to clinical information extracted from electronic health records (EHRs).

Main Methods:

• Extracted eligibility criteria from clinical notes using pattern matching and natural language processing (NLP) applications.
• Compared extracted criteria against manually abstracted trial descriptions for three breast cancer trials.
• Utilized rule-based, cosine similarity-based, and machine learning-based NLP approaches.

Main Results:

• The machine learning-based NLP application achieved high accuracy in extracting eligibility criteria (90.9% recall, 89.7% precision).
• The machine learning approach demonstrated the highest accuracy in determining trial eligibility, with AUCs ranging from 75.5% to 89.8%.

Conclusions:

• NLP effectively extracts eligibility criteria from EHR clinical notes.
• Automated patient discovery for clinical trials can be achieved with good accuracy, potentially reducing human screening workload.