Table 1 Important transparency considerations expressed at the workshop.
From: Transparency of artificial intelligence/machine learning-enabled medical devices
Stakeholder | Transparency consideration |
|---|---|
Patients | • Patient-centric labeling that includes information like training data demographics, technical requirements for use of device, etc. • Consistency in what information is shared • Variety of information delivery mediums (user interface, video, graphics, training, etc.) • Management of bias and patient-focused communication explaining these efforts • Appropriate notification of changes in device performance • Standard notation in electronic health records to indicate AI/ML device use, connecting events that could impact health (device recalls, device performance, etc.) • Trusted source(s) of device information |
Healthcare providers | • Healthcare provider-specific labeling, specifically including intended use and performance in different populations • Consistency in what information is shared, including how a device can be used within the clinical workflow • Publicly available, detailed decision summary that includes information on physical or social characteristics of the clinical validation data • Standard testing before a device receives marketing authorization • Trusted source(s) of device information and user training resources • Notification to providers of device modification, malfunction, or performance changes • Real-world performance monitoring |
Payors | • Assurance of performance across environments and populations • Real-world performance monitoring |
Industry | • Appropriate level of regulatory oversight • Transparency aligned with proprietary needs • Additional guidance for device marketing authorization submissions • Post-market pathways to expand device claims |
