DeepPhe-CR: Natural Language Processing Software Services for Cancer Registrar Case Abstraction

Area of Science:

• Computational oncology
• Clinical informatics
• Natural Language Processing

Background:

• Manual extraction of cancer details from patient records for surveillance is time-consuming and costly.
• Natural Language Processing (NLP) offers a potential solution for automating the identification of critical information within clinical notes.
• Integrating NLP tools into existing cancer registry workflows is essential for improving data abstraction efficiency.

Approach:

• Developed DeepPhe-CR, a web-based NLP service API, guided by manual cancer registry abstraction processes.
• Validated NLP methods for coding key variables using established workflows and implemented within a container-based architecture.
• Modified existing registry data abstraction software to incorporate DeepPhe-CR results, facilitating computer-assisted abstraction.

Key Points:

• DeepPhe-CR supports single document submission and multi-document case summarization via API calls.
• NLP modules achieve high extraction accuracy (0.79-1.00 F1 score) for topography, histology, behavior, laterality, and grade across various cancer types.
• Initial usability studies confirmed the feasibility and effectiveness of DeepPhe-CR tools for data registrars.

Conclusions:

• The DeepPhe-CR system offers a flexible architecture for integrating cancer-specific NLP tools into registrar workflows.
• Computer-assisted abstraction enhanced by DeepPhe-CR shows promise for improving cancer surveillance data collection.
• Further refinement of user interactions within client tools may enhance the full realization of these NLP-driven approaches.