The CARE framework for AI dataset documentation in clinical laboratories: a comprehensive checklist and data lineage...

Area of Science:

• Clinical laboratory science
• Artificial intelligence in healthcare
• Data management

Background:

• Traditional regulatory frameworks for clinical laboratories do not adequately address data documentation for artificial intelligence and machine learning (AI/ML).
• AI/ML solutions inherit biases and limitations from training data, posing risks in clinical settings.
• A structured approach to data documentation is crucial for reliable AI/ML applications in laboratories.

Purpose of the Study:

• To develop and implement a comprehensive data documentation checklist and data lineage methodology for AI/ML in clinical laboratories.
• To integrate this framework within the existing Clinical AI Readiness Evaluator (CARE) framework.
• To enhance transparency, reproducibility, and regulatory compliance of AI/ML solutions in clinical settings.

Main Methods:

• Developed a data checklist and lineage methodology building upon the CARE framework.
• Conducted a comprehensive review of existing AI/ML data documentation practices.
• Engaged with AI operations teams and refined the checklist through iterative, multidisciplinary efforts.
• Assessed the checklist's effectiveness on three diverse AI/ML projects within the CARE framework.

Main Results:

• The CARE Data Checklist and Data Lineage provide a structured method for documenting critical dataset aspects.
• Key documented elements include data composition, demographics, collection, labeling, usage, and maintenance.
• The framework addresses unique data challenges in AI/ML, promoting transparency, reproducibility, and compliance.

Conclusions:

• The CARE Data Checklist and Data Lineage function as technical guides and communication tools for stakeholders.
• Early implementation within the AI lifecycle helps laboratories proactively manage data challenges.
• This approach saves resources and improves the reliability of AI-augmented laboratory solutions.