Knowledge Models for Cancer Clinical Practice Guidelines: Construction, Management and Usage in Question Answering

Area of Science:

• Oncology
• Medical Informatics
• Natural Language Processing

Background:

• Existing automated algorithms for Cancer Clinical Practice Guidelines (CPGs) have limited scope and struggle with the complexity of varied cancer types.
• There is a need for more robust automated methods to extract and structure knowledge from CPGs for improved usability and updates.

Purpose of the Study:

• To propose an improved automated knowledge modeling algorithm for creating structured knowledge models from National Comprehensive Cancer Network (NCCN) CPGs across different cancer types.
• To develop an algorithm for comparing guideline versions to identify changes in treatment protocols.
• To create and evaluate a question-answering (Q&A) framework utilizing these knowledge models.

Main Methods:

• Developed an enhanced automated knowledge modeling algorithm to process NCCN CPGs for multiple cancer types.
• Implemented an algorithm to compare different versions of CPGs, detecting modifications in treatment protocols.
• Constructed a Q&A framework using the generated knowledge models as an augmented knowledge base.
• Evaluated the Q&A framework using 32 question-answer pairs for Non-Small Cell Lung Cancer (NSCLC) treatment.

Main Results:

• The proposed algorithm successfully created knowledge models from NCCN CPGs for four distinct cancer types.
• The guideline comparison algorithm effectively identified changes between CPG versions.
• The Q&A framework achieved 54.5% accuracy for answers derived from the treatment algorithm and 81.8% accuracy from the discussion section of the NSCLC NCCN guideline knowledge model.

Conclusions:

• The improved automated knowledge modeling algorithm enhances the extraction and structuring of complex information from CPGs.
• The developed framework provides a viable method for querying CPG knowledge and tracking guideline evolution.
• This approach has the potential to improve access to and understanding of cancer treatment guidelines.