AAMI TIR34971-2023 PDF

Full title and description

AAMI TIR34971:2023 — Application of ISO 14971 to machine learning in artificial intelligence — Guide. This Technical Information Report provides guidance and practical interpretation of ANSI/AAMI/ISO 14971:2019 specifically for medical devices and systems that incorporate machine learning (ML) or artificial intelligence (AI), addressing ML‑specific hazards, risk controls, verification/validation and lifecycle risk management practices tailored to ML/AI-enabled medical technology.

Abstract

This guide explains how to apply the ISO 14971 risk management framework to medical devices/systems that include ML or AI components. It highlights ML/AI‑specific risk sources (for example, data quality and drift, training/validation errors, model updates and distributional shifts), describes appropriate risk control and verification strategies, and gives practical recommendations for documentation, change management and post‑market monitoring for ML/AI-enabled devices. The document is intended as a companion to ISO 14971 rather than as a replacement.

General information

• Status: Published (active guidance document).
• Publication date: 14 March 2023 (AAMI publication).
• Publisher: Association for the Advancement of Medical Instrumentation (AAMI); a joint BS/AAMI edition is also published by BSI.
• ICS / categories: Medical equipment in general (11.040.01); Information technology in general (35.020).
• Edition / version: 2023 edition (TIR34971:2023).
• Number of pages: 33 pages (PDF).

Scope

The scope of AAMI TIR34971:2023 is to provide targeted guidance for applying the principles and process of ISO 14971:2019 to devices and systems that incorporate ML/AI. It covers lifecycle risk management activities where ML/AI characteristics are material to safety and performance: hazard identification and analysis for ML failure modes, data collection and management, model training and validation, change control for model updates, verification/validation approaches appropriate for ML, clinical evaluation considerations, and post‑market performance monitoring and mitigation. The guidance is intended for use within a medical device quality and regulatory framework and complements existing ISO/AAMI requirements.

Key topics and requirements

• Applying ISO 14971 risk management processes to ML/AI components (hazard identification, risk analysis, risk evaluation and risk control).
• Identification of ML/AI‑specific hazards (data bias, data drift, out‑of‑distribution inputs, model degradation, concept drift, adversarial inputs).
• Data quality, dataset curation, representativeness and documentation requirements for training/validation datasets.
• Verification and validation strategies tailored to ML (performance metrics, hold‑out and external validation, continuous monitoring methods).
• Change management for model updates (planning for anticipated changes, predetermined change control-like approaches for ML lifecycle changes).
• Post‑market surveillance and real‑world performance monitoring to detect drift, safety signals and usability issues after deployment.
• Documentation, traceability and rationale for risk acceptability decisions specific to ML design choices.

Typical use and users

Primary users are medical device manufacturers and their multidisciplinary teams (risk managers, software engineers, data scientists, regulatory affairs specialists, clinical experts) who design, develop, validate or maintain ML/AI-enabled medical devices. Regulators, notified bodies, clinical engineers, hospital technology assessment teams and third‑party evaluators also use the guidance to assess ML/AI risk management practices. The TIR is commonly used alongside ISO 14971 and other software/clinical standards when implementing or auditing ML/AI risk processes.

Related standards

Key related documents include ISO 14971:2019 (Medical devices — application of risk management to medical devices), AAMI CR34971:2022 (consensus report that informed this TIR), AAMI TIR45:2023 (guidance on AGILE practices for medical device software), and the joint BS/AAMI 34971:2023 publication. The AAMI consensus work that preceded this TIR has also been recognized in regulatory discussions and informed international BSI alignment.

Keywords

AI, artificial intelligence, machine learning, ML, risk management, ISO 14971, medical device software, verification and validation, data quality, post‑market surveillance, change management, model updates, AAMI.

FAQ

Q: What is this standard?

A: AAMI TIR34971:2023 is a Technical Information Report (guide) that explains how to apply the ISO 14971 risk management framework to medical devices and systems that include ML or AI components. It is a companion guidance document rather than a replacement for ISO 14971.

Q: What does it cover?

A: It covers ML/AI‑specific risk topics across the device lifecycle: dataset and data‑quality considerations, ML‑specific hazards (drift, bias, unexpected inputs), verification and validation approaches, planned change management for model updates, and post‑market monitoring and mitigation strategies.

Q: Who typically uses it?

A: Device manufacturers (risk managers, data scientists, software engineers), regulatory and quality professionals, notified bodies, clinical engineering groups and hospital technology evaluators use the TIR to design, assess and audit ML/AI risk practices.

Q: Is it current or superseded?

A: The 2023 TIR is the current guidance edition published by AAMI (TIR34971:2023). It builds on earlier AAMI consensus work from 2022 and the joint BSI/AAMI publication; users should check with AAMI/BSI for any amendments or newer editions.

Q: Is it part of a series?

A: Yes. The TIR is related to a broader set of AAMI guidance on software and AI in medical devices (for example CR34971:2022 and TIR45:2023) and a coordinated BS/AAMI publication (BS/AAMI 34971:2023) that aligns the guidance internationally.

Q: What are the key keywords?

A: AI, machine learning, ISO 14971, risk management, data quality, model validation, change control, post‑market monitoring, medical device software.