Artificial intelligence (AI) encompasses algorithms designed for tasks such as learning, decision-making, and prediction. Machine learning (ML), a subset of AI, allows algorithms to build models from data without explicit programming. This technology holds great potential in healthcare, particularly through machine learning-enabled medical devices (“MLMDs”). On February 5, 2025, Health Canada released a pre-market guidance on MLMDs (“Guidance”), outlining the information manufacturers should provide to demonstrate safety and effectiveness, ensuring that these devices meet regulatory standards and contribute to improved patient care.
It is important to note that Health Canada can impose terms and conditions on medical device licences. These limitations are based, among others things, on the safety profile of your MLMD and how it was presented to Health Canada. This bulletin is meant to help you navigate important aspects of the new Guidance so you can be in the best position to obtain a licence and hopefully prevent limitations.
Overview
Subject to certain exceptions, manufacturers must submit an application and obtain a medical device licence issued by Health Canada in order to import, sell or advertise a MLMD in Canada. The application includes detailed information such as a list of standards followed in its design and manufacture, as well as a description of the medical conditions, purposes, and intended uses. The Guidance provides additional clarification and specific information for manufacturers to consider when preparing a license application for an MLMD.
Health Canada emphasizes the importance of providing product lifecycle information to demonstrate the safety and effectiveness of MLMDs, establishing specific criteria for this purpose.
Good Machine Learning Practice
Design
Manufacturers must clearly outline the device's intended medical purpose within the application. A comprehensive device description must also be provided, detailing the ML systems used to achieve this intended purpose. This should include an explanation of the methods, training algorithms, and data used to develop and train the ML system. Additionally, the PCCP must document any modifications to the MLMD, ensuring it remains within its intended use. The PCCP should include:
- A change description detailing the initial design and proposed changes;
- A change protocol outlining policies and procedures for managing these changes;
- An impact assessment evaluating the potential benefits and risks of the changes.
Risk Management
Robust risk management practices must be implemented throughout the MLMD lifecycle. This includes addressing potential issues such as erroneous outputs, biases and degradation in the performance of the ML system. Manufacturers are encouraged to consider ISO 14971, Medical devices - Application of risk management to medical devices.
Data Selection and Management
The quality of the datasets used to develop an MLMD directly impacts the quality of the device. Manufacturers should outline inclusion and exclusion criteria, address data imbalances, and ensure data quality and accuracy as more fully described in the Guidance.
Development and Training
Manufacturers must provide clear descriptions of the ML development and training, ensuring transparency and clarity in the methods used.
Testing and Evaluation
Comprehensive testing and evaluation strategies should be implemented to assess the safety, efficacy, and performance of the MLMD throughout its lifecycle.
Clinical Validation
In the case of Class III and IV MLMD, manufacturers must submit appropriate clinical evidence, including clinical validation studies, to demonstrate the device's safe and effective use. This evidence should encompass the type of studies conducted, clinical data, usability and human factors testing, real-world evidence (RWE), and post-market clinical experience.
Transparency
Transparency refers to the extent to which clear and appropriate information about a device is communicated to stakeholders, such as patients, users, and healthcare providers.
It must be maintained throughout the device lifecycle, with clear information provided in device labelling (including instructions for use), software user interfaces, and medical device license applications to help stakeholders make informed decisions.
Post-Market Monitoring
Manufacturers should include detailed descriptions of the processes and plans for post-market surveillance and performance monitoring. This should include strategies for ongoing performance assessment, risk mitigation, and ensuring inter-compatibility of the ML system.
Conclusion
Manufacturers of MLMDs must include very detailed information in their medical device license application to demonstrate compliance with regulatory standards. The Guidance provides detailed direction on the essential information required, ensuring that the safety, effectiveness, and intended use of these devices are clearly outlined and thoroughly evaluated.
It is also important to note that Health Canada can impose terms and conditions on some medical device licenses. These may include, for instance, additional tests to be performed and the submission of the results to Health Canada. Where possible, it is important to submit complete, accurate information in compliance with the Guidance to avoid such limitations.
The Fasken team is closely monitoring all regulatory updates in the Canadian medical device landscape.
