GRS News

The advancement of machine learning and artificial intelligence in healthcare has opened up revolutionary opportunities for enhancing patient care. However, these technologies' complexity and data-driven nature necessitate careful consideration, particularly in terms of transparency and regulatory practices. Recently, the FDA, Health Canada, and the UK's MHRA have released a comprehensive set of guiding principles focused on transparency for MLMDs. These principles aim to foster international harmonisation and underscore the importance of transparency throughout the life cycle of MLMDs.

 

Background into the developments

In 2021, the U.S. Food and Drug Administration (FDA), Health Canada and the United Kingdom’s Medicines and Healthcare products Regulatory Agency (MHRA) jointly identified 10 guiding principles for good machine learning practice (GMLP).  As the adoption of these tools advanced, the 3 regulatory bodies further identified guiding principles for transparency for Machine Learning-enabled Medical Devices (MLMDs). These principles build upon the GMLP principles, especially:

• principle 7: Focus is placed on the performance of the human-AI team.
• principle 9: Users are provided with clear, essential information.

 

Why Transparency Matters

Transparency in MLMDs is not just a buzzword; it's a necessity. It refers to providing stakeholders with contextual information about the device, its intended use, and how it functions within healthcare settings. This information is crucial for building trust among healthcare professionals and patients, aiding in informed decision-making, and ensuring the safe and effective use of these devices. Troy Tazbaz, the director of the Digital Health Center of Excellence at FDA’s Center for Devices and Radiological Health, emphasises that transparency throughout a device's life cycle promotes trust and proper utilisation, ultimately benefiting patient outcomes.

 

Defining Transparency and Logic

In this document, “transparency” describes the degree to which appropriate information about a MLMD (including its intended use, development, performance and, when available, logic) is clearly communicated to relevant audiences. “Logic” refers to information about how an output or result was reached or the basis for a decision or action. The degree to which this logic can be explained in a way that a person can understand is known as “explainability”. Logic and explainability are aspects of transparency. 

 

The document further expands transparency in effective transparency added with the value of adapting human-centred design into the development of these MLMDs.

 

Key Components of Transparency

1. Context of Use: Information provided should detail who uses the device, why it is used, and the specific healthcare settings in which it operates. This includes detailing the intended users, the purpose of the device, and its role in treating specific conditions or diseases.
2. Communication Strategies: The principles suggest optimising software user interfaces to ensure the conveyed information is personalised, adaptive, and reciprocal. Effective communication should cater to different stages of the device’s lifecycle and address the needs of various stakeholders, from healthcare providers to patients.
3. Human-Centred Design: Employing human-centred design principles helps develop transparent and user-friendly MLMDs. This involves understanding the users, their environments, and workflows and incorporating iterative design, validation, monitoring, and communication strategies.

Detailed Device Information

Good practices for describing an MLMD include providing information about:

• Intended Use and Users: Clearly outline who the device is for and what it aims to achieve.
• Function and Workflow: Explaining how the device works within the healthcare process, its impact on clinical decisions, and any limitations or risks.
• Updates and Lifecycle: Notifying users of updates and changes throughout the device's lifecycle, ensuring continuous transparency.

 

The Ten Guiding Principles

1. Leveraging Multi-Disciplinary Expertise: Ensuring a comprehensive understanding of the clinical workflow and patient risks to maintain the safety and effectiveness of MLMDs.
2. Implementing Good Software Engineering and Security Practices: Adopting robust design, data quality, and cybersecurity practices to manage risks and ensure data integrity.
3. Representative Clinical Study Participants and Data Sets: The data sets must be representative of the intended patient population to manage biases and ensure generalisable performance.
4. Independence of Training and Test Sets: Maintaining independence between training and test datasets to assure unbiased performance.
5. Using the Best Available Methods for Reference Datasets: Employing the best methods to develop reference datasets ensures data relevance and robustness.
6. Tailored Model Design: Designing models that fit the available data and intended use, emphasising mitigating known risks.
7. Focus on Human-AI Team Performance: Addressing human factors to optimise the performance of the human-AI team, not just the AI in isolation.
8. Clinically Relevant Testing Conditions: Developing test plans that reflect real-world conditions and patient diversity.
9. Clear User Information: Provide accessible, relevant information about the device's use, performance, and limitations.
10. Monitoring and Managing Deployed Models: Ensuring continuous monitoring and management of deployed models to maintain safety and performance, addressing issues like dataset drift and bias.

The Path Forward

The FDA, Health Canada, and MHRA are committed to advancing these principles through international collaboration and continuous feedback from stakeholders. This collaborative approach aims to harmonise regulatory practices, promote health equity, and encourage the adoption of innovative, transparent MLMDs. The ongoing dialogue and engagement with the public and international partners will help refine these principles and support responsible innovation in AI-powered healthcare.

 

What These Guidelines Mean for MLMD Manufacturers and Developers?

These guiding principles carry significant implications for manufacturers and developers as they aim to ensure that MLMDs are developed, communicated, and maintained in a safe, effective, and trustworthy way. Here's what these principles mean for the industry:

 

1. Enhanced Responsibility Throughout the Product Lifecycle

Manufacturers must take a holistic approach, ensuring transparency at every stage of the product lifecycle, from design and development to deployment and post-market monitoring. This includes:

• Initial Design and Development: Integrating multi-disciplinary expertise to understand the clinical workflow and patient needs.
• Deployment: Clearly communicating the intended use, user groups, and device functioning in clinical settings.
• Post-Market Monitoring: Continuously monitor device performance and address identified risks, biases, or data drift.

 

2. Adoption of Robust Software Engineering Practices

Developers need to implement strong software engineering practices to ensure the integrity and security of MLMDs. This involves:

• Data Quality Assurance: Ensuring that training and testing datasets are representative and independent.
• Cybersecurity Measures: Implementing robust cybersecurity practices to protect patient data and device integrity.
• Methodical Risk Management: Employing systematic risk management processes to capture and mitigate potential issues throughout the development and operational phases.

 

3. Commitment to Representativeness and Bias Management

Data collection protocols must ensure that datasets represent the intended patient population. Manufacturers must:

• Ensure Diversity in Data: To avoid bias, collect data that accurately represent various demographics (age, gender, race, etc.).
• Validate Model Performance: Regularly test the model’s performance across different subgroups to ensure equitable outcomes.

 

4. Transparency in Communication and User Information

Providing clear, comprehensive, and accessible information to all stakeholders is essential. Manufacturers should:

• Detail Device Usage and Limitations: Clearly outline the device’s intended use, performance metrics, limitations, and potential risks.
• User Interface Design: Optimise the software user interface to convey information effectively and intuitively to users.
• Regular Updates and Notifications: Ensure that users are kept informed about updates, modifications, and any real-world performance data.

 

5. Human-Centred Design Principles

The integration of human-centred design principles is vital for developing user-friendly MLMDs. This includes:

• Understanding User Needs: Conducting thorough research to understand the environments and workflows where the device will be used.
• Iterative Design and Validation: Continuously improving the device based on user feedback and performance data.
• Training and Support: Providing adequate training and resources to ensure users can effectively and safely operate the MLMD.

 

6. Focus on Human-AI Collaboration

Manufacturers need to design MLMDs with a focus on enhancing the performance of the human-AI team. This means:

• Human Factors Consideration: Ensuring that the device’s outputs are interpretable and usable by healthcare professionals.
• Performance Testing: Conducting tests that reflect real-world clinical conditions and the interaction between humans and the AI.

 

7. Clear Guidelines for Reference Datasets and Model Design

When developing and testing MLMDs, manufacturers should use the best available methods for creating reference datasets and tailoring model designs to available data. This involves:

• Developing Robust Reference Standards: Using clinically relevant and well-characterised data to develop and test models.
• Avoiding Overfitting and Performance Degradation: Implement strategies to mitigate risks like overfitting and ensure the model remains effective over time.

 

8. Ongoing Collaboration and Feedback

Manufacturers are encouraged to engage in continuous dialogue with regulatory bodies and stakeholders to refine best practices and standards. This collaborative approach ensures:

• International Harmonisation: Aligning practices with international standards to promote global consistency in MLMD development.
• Continuous Improvement: Using feedback to drive improvements in device transparency and performance.

 

What actions should the manufacturers/developers take?

Manufacturers of Machine Learning-Enabled Medical Devices (MLMDs) need to take several specific actions to align with the newly released guiding principles. These actions are:

1. Incorporate Multi-Disciplinary Expertise to ensure the device is designed with a comprehensive understanding of clinical workflows and patient needs.
2. Implement Robust Software Engineering Practices to ensure the integrity, security, and reliability of the MLMD.
3. Ensure Data Quality and Representativeness To avoid biases and ensure the model performs well across diverse patient groups.
4. Tailor Model Design to Data and Intended Use to ensure the model’s clinical validity and reliability
5. Focus on human-AI team Performance to optimise the human-AI team's performance and ensure practical utility in clinical settings.
6. Engage in Continuous Feedback and Collaboration to align with evolving best practices and regulatory expectations.
7. Adopt Human-Centred Design Principles To ensure the device meets its users' needs and is intuitive and effective in real-world settings.
8.  Prepare for Regulatory Compliance To facilitate smoother regulatory approval and market entry.

In summary, the guiding principles for transparency in MLMDs provide a robust framework to ensure these advanced technologies are developed, communicated, and utilised in a way that prioritises patient safety, trust, and effective healthcare delivery. By adhering to these principles, developers and regulators can work together to harness the full potential of machine learning in medicine.

 

These guiding principles for MLMD manufacturers and developers represent a roadmap to developing safe, effective, and transparent medical devices. By adhering to these guidelines, manufacturers can build trust with healthcare professionals and patients, ensure regulatory compliance, and ultimately contribute to the advancement of AI in healthcare. Embracing these principles will enhance the quality and safety of MLMDs and foster innovation and acceptance in the rapidly evolving field of digital health.

 

By taking these actions, MLMD manufacturers can not only comply with regulatory expectations but also enhance their products' trust, safety, and effectiveness, ultimately contributing to better patient outcomes and the advancement of healthcare technology.

 

At Global Regulatory Services, we can help you with:

• Regulatory Submissions: Prepare regulatory submissions in compliance with the FDA, Health Canada, and MHRA guidelines.
• Risk Management: Develop and implement risk management frameworks that address the unique risks of MLMDs, including data security and integrity.
• Clinical Evaluation Report: Effectively report the design and conduct clinical evaluations to demonstrate the safety and effectiveness of MLMDs.
• Post-Market Surveillance: Establish robust post-market surveillance systems to monitor the performance of MLMDs and manage any arising issues effectively.

By leveraging expertise at Global Regulatory Services (GRS), MLMD manufacturers can ensure they meet the high standards set by international regulatory bodies, thereby enhancing their products' safety, effectiveness, and marketability.

 

Reach out to us by sending an email to grs@globalregulatoryservices.com or by submitting a web form - https://globalregulatoryservices.com/global-regulatory-services-contact-details