Data-driven knowledge acquisition, validation, and transformation into HL7 Arden Syntax.

Area of Science:

• Medical Informatics
• Clinical Knowledge Engineering
• Oncology Decision Support

Background:

• Acquiring clinical knowledge from diverse data sources is challenging.
• Integrating expert knowledge with data-driven approaches is crucial for effective clinical decision support.
• Standardizing treatment protocols for head and neck cancer requires robust knowledge models.

Purpose of the Study:

• To develop a method for acquiring, validating, and refining clinical knowledge for head and neck cancer treatment.
• To create executable clinical knowledge models (R-CKM) that align with published guidelines and real-world data.
• To incorporate these refined models into clinical workflows for enhanced decision support.

Main Methods:

• A data-driven approach using patient datasets to generate a predictive model (PM).
• Validation of the PM against clinical knowledge models (CKM) derived from National Comprehensive Cancer Network (NCCN) guidelines.
• Conversion of the validated model into executable medical logic modules (MLMs) using HL7 Arden Syntax.

Main Results:

• A refined-clinical knowledge model (R-CKM) was developed for oral cavity cancer treatment.
• The predictive model achieved 59.0% accuracy, and the refined model yielded 53.0% accuracy on test datasets.
• The R-CKM successfully integrated real-world practice data with NCCN guidelines.

Conclusions:

• The proposed method effectively creates executable clinical knowledge by combining data-driven insights with expert validation.
• The R-CKM offers a balance between reflecting actual clinical practices and adhering to established treatment guidelines.
• This approach enhances collaboration between physicians and knowledge engineers, leading to improved clinical decision support systems.