Software & AI/ML — FDA Policy (SaMD)
What is SaMD?
Software as a Medical Device (SaMD) is software that performs a medical device function without being part of a hardware medical device. The definition is aligned with the IMDRF SaMD framework.
Examples:
- Software that analyses medical images to detect pathology
- AI-powered ECG interpretation software
- Clinical decision support software that drives clinical decision-making
FDA's regulatory framework for software
FDA uses three categories:
| Category | Description | Regulatory status |
|---|---|---|
| Device software functions | Software that meets the SaMD definition | Subject to FDA oversight |
| Non-device software functions | Administrative, general wellness, non-clinical decision support | Outside FDA oversight |
| Exempt CDS software | Clinical decision support that is not the primary basis for clinical decision | Not a device (under 21st Century Cures Act) |
SaMD classification
SaMD is classified using the same Class I/II/III framework. FDA's Digital Health Software Precertification Program (2019–2022) explored an alternative approach but did not result in a permanent regulatory pathway.
Key guidance documents:
- FDA Software Functions Guidance (2023) — determines if software is a device
- Clinical Decision Support Software Guidance (2022)
- AI/ML Action Plan (2021)
- Predetermined Change Control Plan (PCCP) Guidance (2024)
Predetermined Change Control Plan (PCCP)
For AI/ML-based SaMD that is intended to learn and change, FDA has established the PCCP framework:
- Submitted as part of a 510(k), De Novo, or PMA
- Specifies the types of modifications anticipated
- Describes the algorithm change protocol
- Specifies performance monitoring requirements
A device with an approved PCCP can implement pre-specified changes without submitting a new 510(k) for each change.
Official resources
- FDA — Digital Health Center of Excellence
- FDA — Software as a Medical Device
- FDA — PCCP Guidance (2024)
510(k) Submission for SaMD
Most SaMD products follow the 510(k) pathway and must:
Identify an appropriate predicate device — the predicate may be another SaMD product, a cleared medical device software component, or a substantially equivalent hardware device with the same intended use
Demonstrate substantial equivalence — show that the SaMD is as safe and effective as the predicate device in terms of performance, indications for use, and technological characteristics
Submit software documentation — including software specifications, algorithm descriptions, cybersecurity documentation, and clinical/analytical validation data
Note: Identifying a predicate device is more challenging for novel AI/ML algorithms; sponsors may pursue De Novo classification if no suitable predicate exists.