Imperatives & Insights: Navigating the FDA's New Regulatory Landscape for SaMD and AI/ML in MedTech
The recent MedTech Roundtable Dinner hosted by Orthogonal included strategics, ecosystem builders and emerging companies. The conversation covered both premarket and commercialization considerations. Kwame Ulmer (Managing Partner at MIP) moderated a discussion with Jessica Richter from Richter Advisory Collective. He highlighted the areas for the FDA when reviewing submissions for Software as a Medical Device (SaMD) and AI/ML-enabled devices.
Here are the top three questions and issues driving the FDA’s review process, and how your team can be prepared:
1. Clarify Your Software Device Function and Risk Profile
The FDA is looking for crystal-clear clarity on what your software does and its risk level. Submissions must move beyond high-level descriptions. You must:
Detail the technical flow: Provide a reviewer-ready description of the inputs, processing, and outputs.
Define boundaries: Clearly state the intended use, including its clinical role (e.g., adjunctive, decision-support) and explicit boundaries of what the software does and, crucially, what it does not do.
Map the architecture: Supply a block diagram of the architecture and data flow, including all modules, data pathways, and integrity controls.
Show the user experience: Include end-to-end user/clinical workflow diagrams to demonstrate integration into real-world practice.
2. Establish Robust and Unbiased AI/ML Evidence
The adequacy of AI/ML evidence remains a major point of contention, particularly regarding clinical validity, generalizability, and bias avoidance. To meet the FDA's scrutiny, companies must:
Ensure data governance: Document dataset governance, focusing on explicit subject-level separation and a clear, auditable split between training, tuning, and validation datasets as required by FDA guidance.
Define 'Ground Truth': Provide a clinically defensible ground truth/reference standard, supported by a clear clinician adjudication approach.
Validate for real-world use: Conduct generalizability and bias evaluations across relevant subgroups and use-conditions to ensure performance is consistent with the intended population and diverse settings.
Set Pre-Defined Criteria: Establish acceptance criteria—performance thresholds—in advance for primary endpoints and all key subgroups/use conditions.
3. Demonstrate Safe Use in Real-World Conditions and Manage Risk
Safety in the hands of the intended user, coupled with strong cybersecurity, is a paramount concern. The focus is on anticipating and controlling risks across the device's lifecycle:
Test across conditions: Demonstrate safe use across representative real-world conditions (e.g., home-use, varying connectivity, lighting, time pressure) and across all claimed user interfaces and configurations.
Prioritize Cybersecurity: Provide robust testing and documentation aligned with the latest FDA guidance on Cybersecurity in Medical Devices.
Plan for the Lifecycle: Detail your lifecycle assurance processes, including versioning, updates, and postmarket monitoring plans designed to prevent regressions in safety, performance, and cybersecurity.
Ready to confidently navigate the FDA's regulatory landscape for your SaMD or AI/ML device? Our team at MedTech Impact Partners (MIP) specializes in translating complex FDA requirements into clear, actionable development and submission strategies.