Software as a Medical Device (SaMD)

Overview

Software is explicitly included in the MedDO definition. Standalone software intended for a medical purpose is regulated as a medical device in its own right. Qualification guidance MDCG 2019-11 applies in Switzerland.

Qualification — Does This Software Qualify as a Medical Device?

Key qualification tests (MDCG 2019-11):

• Does the software perform an action on data beyond storage/archival/communication/search? → Potentially a medical device
• Does its output drive clinical management or treatment decisions? → Likely a medical device
• Is its output solely for administrative purposes? → Generally not a medical device

Classification of SaMD

Apply the 22 MedDO classification rules, especially Rule 11 and Rule 22. Software IVDs use IVDO Annex VIII Rule 7. Typical classes: Class I (low-risk information to professionals), IIa (medium-risk decision support), IIb (high-risk diagnostic decisions), III (life-critical closed-loop control).

IEC 62304 — Software Lifecycle

MedDO Annex I § 17 requires software to be developed in accordance with the state of the art. IEC 62304 defines software safety classes A/B/C and development, maintenance, and change management requirements for each class.

AI and Machine Learning

Swissmedic applies MDCG guidance on AI/ML with Switzerland-specific oversight. Key requirements: AI/ML software with medical intended purpose is regulated as a medical device; training/validation datasets must be fully documented with provenance, representativeness, and limitations disclosed; algorithm updates, retraining with new data, or significant performance changes may constitute modifications requiring reassessment or post-market surveillance notifications.

Official Sources

• MDCG 2019-11 — Software qualification and classification
• MDCG 2020-1 — AI and ML in medical devices

Disclaimer

AI-assisted content for navigation only. Always verify against official Swissmedic and Fedlex sources. Not legal or regulatory advice.

Dataset documentation should include: dataset origin and composition; representative population characteristics; performance across different subgroups; validation methodology; any known limitations or potential bias. Swissmedic expects manufacturers to demonstrate that datasets are appropriate for the intended use case.