FDA Prepares New Guidance for SaMD

- The FDA's evolving strategy for SaMD includes the January 2026 withdrawal of its 2017 guidance on "SaMD: Clinical Evaluation," which was based on principles from the International Medical Device Regulators Forum (IMDRF). This signals a shift from a single, prescriptive framework toward more tailored, risk-based clinical evidence strategies for demonstrating safety and effectiveness. - For AI/ML-based SaMD, the FDA's 2021 Action Plan established a "Total Product Lifecycle" approach, which includes the use of a Predetermined Change Control Plan (PCCP). This allows a manufacturer to pre-specify and get approval for planned modifications to a learning algorithm without needing a new regulatory submission for each change. - The distinction between Software *in* a Medical Device (SiMD) and SaMD hinges on whether the software can perform its medical function independently of a hardware device. SiMD is integral to a physical device and validated as part of that system, while SaMD is regulated as a standalone product. - Under its risk-based framework, the FDA has authorized over 1,000 AI/ML-enabled devices as of early 2025. The vast majority of these have been classified as Class II (moderate-risk) and cleared via the 510(k) or De Novo pathways. - While formal guidance on digital twins is still evolving, the technology is gaining traction with the FDA as a tool to support regulatory submissions. Digital twins can be used for virtual modeling, testing, and post-market monitoring of a SaMD's performance in a controlled digital environment. - The FDA's approach has been shaped by legislation like the 21st Century Cures Act, which clarified that certain software functions—such as those for maintaining a healthy lifestyle or some forms of clinical decision support—are excluded from the medical device definition.