Content of Premarket Submissions and Lifecycle Management for Artificial Intelligence and Machine Learning-Enabled Medical Devices (DRAFT)
This guidance provides recommendations for lifecycle management and marketing submission content for AI-enabled device software functions. It covers documentation requirements for premarket submissions (510(k), De Novo, PMA, HDE, BLA) and lifecycle considerations for AI-enabled devices, including transparency and bias control throughout the total product lifecycle (TPLC). The guidance applies to devices that include one or more AI-enabled device software functions to achieve their intended purpose.
This is a draft guidance. Not for implementation.
Recommended Actions
- Develop comprehensive documentation of AI model development, validation and risk management
- Implement rigorous data management practices including data quality controls and bias assessment
- Establish performance monitoring plan for post-market surveillance
- Design transparent user interfaces and labeling that clearly communicate device capabilities and limitations
- Conduct appropriate clinical and non-clinical validation studies with independent test datasets
- Implement cybersecurity controls specific to AI-enabled devices
- Create detailed labeling including model characteristics, performance metrics and limitations
- Establish change management procedures for software updates
- Consider transparency and bias throughout device lifecycle
- Engage early with FDA through Q-Submission program for novel approaches or technologies
Key Considerations
Clinical testing
- Clinical validation studies should demonstrate device performance across intended use population and subgroups
- Study protocols and statistical analysis plans should be pre-specified
- Independent test datasets should be used for validation
- Human-device team performance should be evaluated when appropriate
Non-clinical testing
- Software verification and validation testing required
- Model development and testing data should be independent
- Precision studies (repeatability/reproducibility) may be needed
- Bench testing and simulations may be appropriate based on device risk
Human Factors
- Human factors validation required for devices with critical tasks
- Usability evaluation recommended to assess information interpretation and user interface
- User tasks and knowledge tasks should be evaluated throughout device use continuum
- Training program effectiveness should be validated if used as risk control
Software
- Software documentation per “Content of Premarket Submissions for Device Software Functions” guidance
- Model architecture and development process should be described
- Software version history and changes should be documented
- Quality system considerations for software development
Cybersecurity
- Cybersecurity controls for AI-specific risks like data poisoning, model inversion, evasion attacks
- Security testing including malformed input and penetration testing
- Data vulnerability and leakage prevention controls
- Secure update mechanisms
Labelling
- Clear description of AI use and model characteristics
- Performance metrics and limitations
- User instructions and training requirements
- Risk information and warnings
- Model development and validation data descriptions
Safety
- Comprehensive risk assessment required
- Risk controls for AI-specific hazards
- Performance monitoring plan recommended
- Change management procedures
Other considerations
- Transparency in device design and outputs
- Bias assessment and mitigation strategies
- Data management practices
- Model maintenance and updates
Relevant Guidances
- Content of Premarket Submissions for Device Software Functions
- Cybersecurity in Medical Devices: Design, Implementation, and Premarket Submissions
- Applying Human Factors Engineering and Usability Engineering to Medical Devices
- Requests for Feedback and Meetings for Medical Device Submissions: The Q-Submission Program
Related references and norms
- ANSI/AAMI/ISO 14971: Medical devices - Applications of risk management to medical devices
- ANSI/AAMI HE75: Human factors engineering - Design of medical devices
- AAMI CR34971: Guidance on the Application of ISO 14971 to Artificial Intelligence and Machine Learning
Original guidance
- Content of Premarket Submissions and Lifecycle Management for Artificial Intelligence and Machine Learning-Enabled Medical Devices
- HTML / PDF
- Issue date: 2025-01-07
- Last changed date: 2025-01-06
- Status: DRAFT
- Official FDA topics: Medical Devices, Digital Health, Postmarket, Drugs, Premarket, Biologics
- ReguVirta summary file ID: 5ff8871fa2a74f5278ff55b31ac20ecc
This post is licensed under CC BY 4.0 by the author.