Guiding principles identified by US FDA, Health Canada, and UK MHRA for ML-enabled medical devices
➔ 1. Leverage Multi-Disciplinary Expertise
Understand clinical workflow, benefits, and patient risks. Ensure safety and effectiveness across the product lifecycle.
➔ 2. Implement Good Software Engineering and Security Practices
Focus on solid software engineering, data quality, and cybersecurity. Use risk management to capture and communicate design and implementation decisions.
➔ 3. Use Representative Clinical Study Participants and Data Sets
Ensure data reflects the intended patient population (age, gender, race, etc.). Manage bias and promote generalizable performance.
➔ 4. Maintain Independence of Training and Test Sets
Keep training and test datasets independent. Address all potential sources of dependence to ensure reliability.
➔ 5. Base Reference Datasets on Best Available Methods
Use best methods for reference datasets. Ensure clinically relevant and well-characterized data. Understand limitations.
➔ 6. Tailor Model Design to Available Data and Intended Use
Design models suited to available data and mitigate risks. Set performance goals that align with clinical benefits and risks.
➔ 7. Emphasize Human-AI Team Performance
Consider human factors and model interpretability. Focus on the Human-AI team’s performance, not just the model.
➔ 8. Test Under Clinically Relevant Conditions
Develop sound test plans. Include patient population, subgroups, clinical environment, and potential confounders.
➔ 9. Provide Clear, Essential Information to Users
Offer clear, relevant info to healthcare providers and patients. Include intended use, performance, data characteristics, limitations, and updates.
➔ 10. Monitor Deployed Models and Manage Re-training Risks
Ensure models can be monitored in real-world use. Control risks of overfitting, bias, and performance degradation.
Follow for daily insights into tech product development.
#MedicalDevices #MachineLearning #AI #Compliance #Safety #MedTech #SimplifyingInnovation #Unplex
Choctaw Nation of Oklahoma |Traverse Project MMIW | Certified Project Director | Operational Advisor | Human & Athletic Performance Specialist | Retired SOF
7moWhen folks like Dr. Preston Cline Ed.D and Dan Dworkis MD PhD speak I make damn sure I listen. Mission Critical Team Institute and The Emergency Mind Project put out a wealth of knowledge and I would recommend they are required reading/listening for any team.