FDA has kicked off the new year by publishing two highly-anticipated draft guidance documents concerning artificial intelligence (AI): Artificial Intelligence-Enabled Device Software Functions: Lifecycle Management and Marketing Submission Recommendations (“Device AI Draft Guidance”); and Considerations for the Use of Artificial Intelligence to Support Regulatory Decision-Making for Drug and Biological Products (“Drug AI Draft Guidance”). The publication of these two draft guidance documents checks off topics from FDA’s device and drug center guidance agendas, respectively.
The Device AI Draft Guidance follows longstanding efforts to develop guidance and resources for a total product life cycle (TPLC) approach to the oversight of AI-enabled devices. It provides FDA’s recommendations on the data and information that should be included in marketing submissions for devices that include AI-enabled software functions. The Device AI Draft Guidance recommends that sponsors include, among other things:
- A device description including information on model inputs (e.g., whether the model uses data from imaging devices, in vitro diagnostics, etc.) and outputs, how AI is used to achieve the device’s intended use, and an explanation of the degree of automation used by the model in comparison to the current standard of care.
- A description of the data on which any AI models were trained.
- Information on the user interface, including any on-screen information or alerts to communicate risks about the AI-enabled device, information provided to users about the AI model and its performance characteristics, and examples of the output format.
- A risk management file addressing risks related to AI-enabled devices, such as lack of information, unclear information, or misunderstood or unavailable data.
- An explanation of data management practices, including for training and testing data.
- Information about the model design, including biases and limitations and any quality control criteria or algorithms.
FDA suggests that sponsors communicate this information in the form of a “model card” that concisely describes key aspects of AI-enabled devices, such as characteristics, performance, and limitations. In addition to the information to be included in a marketing submission, the Device AI Draft Guidance describes the information that should be available in public submission summaries, such as 510(k) Summaries.
Moreover, the Device AI Draft Guidance contains a number of labeling recommendations for AI-enabled devices, including that the labeling should state that AI is used in the device, describe the model inputs and outputs, explain the intended degree of automation, describe the methods and architecture used to develop any AI models, and describe the development and performance validation data. Although it may not be required for certain devices, FDA also recommends that sponsors elect to employ proactive performance monitoring plans. Finally, the Device AI Draft Guidance contains appendices with recommendations on transparency design, performance validation, and usability to aid in the design and development of AI-enabled devices.
The Drug AI Draft Guidance is the first FDA guidance document dedicated to the use of AI in the development of drugs (including both human and animal drugs) and biological products, and it is the latest development in a multi-year effort at the Agency to develop policy in this area. FDA released discussion papers on this topic in 2023 and has sponsored workshops and collected public comments.
The Drug AI Draft Guidance addresses a discrete but important issue: for a sponsor that proposes to use an AI model to produce information to support regulatory decision-making regarding safety, effectiveness, or quality for drugs or biological products, how should the sponsor determine whether the AI model is adequate for a specific use? As FDA highlights, this determination is crucial because of the unique challenges posed by AI, such as the complexity of AI models and of data used for model training. The approach proposed by the draft guidance is to follow a seven-step, risk-based framework for establishing and assessing the credibility of an AI model output for a specific context of use–a framework informed by similar approaches developed in the medical device context. Notably, the Drug AI Draft Guidance does not apply to AI used in drug discovery, another increasingly common use of AI in the development of drugs and biological products.
These draft guidance documents are issued in the final days of FDA under the Biden Administration. When finalized, they will constitute non-binding recommendations to the industry. Stakeholders will be watching closely to see how FDA under the Trump Administration considers comments it receives and whether the new Administration will take a different approach to any of the policy issues addressed in these or other AI-related guidance documents.
Comments on the guidance documents are due by April 7, 2025.
If you have any questions concerning the material discussed in this client alert, please contact the members of our Digital Health and Food, Drugs, and Devices practice.
Back
