Health Canada has published a new pre-market guidance document for Machine Learning-enabled Medical Devices (MLMDs), introducing important regulatory expectations for manufacturers developing AI/ML-based medical technologies. The guidance was officially published on 1 April 2026 and reflects the increasing global regulatory focus on artificial intelligence within healthcare and Software as a Medical Device (SaMD).
The guidance applies to Class II, III, and IV machine learning-enabled medical devices and outlines expectations related to design, risk management, data quality, clinical validation, transparency, post-market monitoring, and lifecycle management.
As AI-enabled medical technologies continue to evolve rapidly, regulators globally are moving toward a lifecycle-based oversight approach rather than focusing only on initial product approval. Health Canada’s latest guidance aligns closely with evolving international frameworks from IMDRF and other global regulators.
Key Areas Covered in the Guidance
The guidance introduces several important areas manufacturers should carefully evaluate during product development and regulatory submission preparation.
1. Explicit Identification of AI/ML Usage
Health Canada now expects manufacturers to clearly state within the submission cover letter that the device uses machine learning technology. For devices incorporating a Predetermined Change Control Plan (PCCP), this must also be explicitly declared.
This highlights the regulator’s emphasis on transparency and proactive regulatory communication for AI-enabled products.
2. Introduction of Predetermined Change Control Plans (PCCP)
One of the most important additions in the guidance is the introduction of PCCPs. A PCCP allows manufacturers to define certain planned post-market modifications to ML systems without requiring repeated regulatory submissions for every change.
The guidance explains that PCCPs should include:
- Change descriptions
- Change protocols
- Impact assessments
- Risk mitigation approaches
- Ongoing monitoring strategies
This reflects the growing recognition that AI/ML systems may continuously evolve over time and require a more flexible lifecycle-based regulatory framework.
3. Increased Focus on Dataset Quality and Representativeness
The guidance places strong emphasis on dataset quality, representativeness, and bias management. Manufacturers are expected to justify that datasets appropriately represent the intended patient population, including considerations related to sex, gender, race, age, and other demographic factors.
Health Canada also expects manufacturers to address:
- Dataset imbalance
- Data integrity
- Bias mitigation
- Inclusion and exclusion criteria
- Canadian population relevance
This is particularly important for AI systems where model performance may vary significantly across different patient populations.
4. Lifecycle-Based Risk Management Expectations
The guidance reinforces that risk management for ML-enabled devices should extend across the entire product lifecycle. Specific AI-related risks identified include:
- False positive and false negative outputs
- Automation bias
- Alarm fatigue
- Model overfitting and underfitting
- Performance degradation over time
- Bias-related risks
Manufacturers are expected to implement appropriate controls and maintain ongoing monitoring processes throughout commercialization.
5. Transparency and Explainability Expectations
Health Canada also highlights the importance of transparency for users, healthcare professionals, patients, and regulators. The guidance recommends providing clear information regarding:
- How the ML system works
- Intended use and limitations
- Performance characteristics
- Known failure modes
- Update frequency
- Clinical workflow integration
The regulator additionally encourages inclusion of structured summaries such as model cards or data cards to improve transparency.
6. Clinical Validation and Real-World Monitoring
For Class III and IV devices, manufacturers are expected to provide robust clinical evidence supporting the safe and effective use of the MLMD in the intended population.
The guidance also emphasizes post-market surveillance and continuous performance monitoring, particularly to identify performance drift, changes in input data distribution, and interoperability concerns after deployment.
Global Regulatory Trend Toward AI Governance
Health Canada’s new guidance reflects a broader international movement toward stronger AI governance in healthcare. Regulatory agencies worldwide are increasingly focusing on:
- Continuous lifecycle oversight
- AI transparency
- Dataset governance
- Human oversight
- Real-world performance monitoring
- Cybersecurity and software assurance
- Bias mitigation
Manufacturers developing AI-enabled medical devices should therefore prepare for increasing global expectations beyond traditional software validation approaches.
Final Thoughts
The publication of Health Canada’s MLMD guidance marks another important step in the global evolution of AI/ML medical device regulation. Manufacturers developing AI-enabled medical technologies should carefully evaluate their current development, validation, risk management, transparency, and post-market processes to align with these evolving expectations.
Organizations that proactively establish structured AI governance, lifecycle monitoring, and robust data management practices will likely be better positioned for future global regulatory compliance.
Need support for AI/ML medical device regulatory compliance?
MedOrdyn Solutions supports medical device and SaMD manufacturers across:
- AI/ML regulatory strategy
- SaMD compliance
- IEC 62304
- ISO 14971 risk management
- EU MDR
- FDA regulatory support
- Technical documentation
- QMS implementation
- Global regulatory compliance support
Connect with us to discuss your regulatory requirements.
Source reference:
