The FDA shared Advice project On the design and development of medical and marketing devices compatible with AI for these devices.
The importance of engaging early
The sponsors of AI compatible devices should commit to the FDA early to ensure that tests to support the marketing submission of an AI compatible apparatus reflect the total life cycle of agency products, an approach based on Risks.
THE Advice project Has recommendations concerning these areas of content for marketing submissions (organized in step in the life cycle of products):
- Development: Risk assessment, data management and description of the model
- Validation: Data management and validation
- Description of the final device: Description of the device, description and development of the model, user interface and labeling and summary of the submission of the public
- Postal market management: Device monitoring of the device and cybersecurity.
Emphasize transparency and wrestling
The project of directives highlights the importance of transparency. He also seeks to reduce biases. The FDA claims that companies should keep in mind these imperatives of the first step in the design of devices throughout the product marketing authorization.
Transparency: The FDA says that transparency can guarantee “that important information is accessible and functionally understandable”. In content labeling, the FDA recommends companies:
- Include information on how AI helps to carry out the expected use of a device.
- Details and explain the inputs and outputs of the model and development.
- Describe performance validation data, device performance measures, performance monitoring tools and known to the AI compatible device, AI-DEVICE software or model (s).
Bias: The FDA describes the bias as “the potential tendency to produce incorrect results in a systematic way, but sometimes unpredictable”. This “can have an impact on the safety and efficiency of the device in any or a subset of the planned use population”. To protect itself against prejudices, the FDA recommends to companies:
- Ensure representativeness in the collection of data in the development, tests and monitoring of the device throughout the life cycle of the product, as well as in performance evaluation, in all relevant demographic groups of expected use (For example, race, ethnic, sex and age).
- Collect evidence to assess whether a device benefits all relevant demographic groups in a similar way to ensure that these devices are safe and effective for their planned use.
Covering the entire product life cycle
Directives make recommendations concerning the performance of the product which extend over the life cycle of products, including in the postal market parameter.
The objective is to solve problems such as data drift, which occurs when the inputs for compatible AI devices change over time from development to real deployment so as to have an impact on performance.
For example, the FDA encourages sponsors to consider using a predetermined changes control plan (CCCP).
As discussed in FULL Guide of the FDA in December 2024Companies can use a PCCP to find a pre-commercial authorization for the changes provided for in a compatible AI device software function. They can do so without first submitting additional marketing bids or obtaining additional FDA authorization.
A piece in a range of AI advice
The FDA does not intend that this orientation project works alone. The agency is considering rather that companies also consult previous guidelines related to devices compatible with AI and specific technologies, including advice on:
What is the next step?
The FDA requests comments on the guide project by April 7, it can therefore consider the comments because it develops a final version.
Our team can help companies prepare and submit comments. We can also help sponsors looking for FDA comments on marketing submission documentation via the agency’s submission program Q.