Presenting machine learning model information to clinical end users with model facts labels.

Area of Science:

• Medical Informatics
• Clinical Decision Support Systems
• Machine Learning in Healthcare

Background:

• Machine learning (ML) shows promise for enhancing medical diagnosis and prognosis.
• Clinical end-users often lack awareness of potential patient harm from ML models.
• Translating ML models into clinical practice presents significant risks.

Purpose of the Study:

• To introduce the "Model Facts" label for clear communication of ML model information.
• To ensure clinicians understand how and when to use ML model outputs.
• To mitigate risks associated with ML model integration in healthcare.

Main Methods:

• Proposal of a systematic "Model Facts" labeling system.
• Designed as a 1-page collation of relevant, actionable information for clinicians.
• Emphasizes collaboration between practitioners and regulators for standardization.

Main Results:

• The "Model Facts" label provides essential information for clinical decision-making.
• Standardized presentation of ML model information can prevent patient harm.
• Clear communication is crucial for the safe adoption of ML tools.

Conclusions:

• The "Model Facts" label is a vital tool for safe ML integration in clinical settings.
• Standardization of ML information presentation is necessary for patient safety.
• Collaborative efforts are required to implement and standardize ML model communication.