FDA AI Device TPLC Guidance: Introduction

Draft guidance notice: This learning path summarises FDA's January 2025 draft guidance distributed for comment only. Draft guidances describe the Agency's current thinking and are not legally binding; they do not establish enforceable responsibilities unless specific statutory or regulatory requirements are cited.

For Hong Kong healthcare professionals evaluating AI medical devices — whether sourced from the United States, used in cross-border telemedicine, or referenced in local procurement — understanding how the U.S. Food and Drug Administration (FDA) expects AI-enabled devices to be managed across their lifecycle provides a practical benchmark for vendor questions and governance discussions.

FDA's total product lifecycle (TPLC) approach

FDA has long promoted a total product life cycle (TPLC) approach to medical device oversight. AI-enabled devices pose distinct challenges because models are data-driven, may change with real-world use, and can be opaque to end users. This draft guidance continues FDA's TPLC work by providing lifecycle management and marketing submission recommendations for devices that include AI-enabled device software functions (AI-DSFs).

The guidance reflects a comprehensive approach spanning:

• Development — risk assessment, data management, model description and development
• Validation — data management and performance validation
• Description of the final device — device description, model details, user interface and labeling, public submission summary
• Postmarket management — device performance monitoring and cybersecurity

Transparency and bias control should be incorporated from the earliest design stage through decommissioning.

Key definitions: device software functions and AI-DSFs

For purposes of this guidance:

• A device software function is software that meets the U.S. device definition under the Federal Food, Drug, and Cosmetic Act — including Software as a Medical Device (SaMD) and Software in a Medical Device (SiMD).
• An AI-enabled device includes one or more AI-enabled device software functions (AI-DSFs) — device software functions that implement one or more AI models to achieve their intended purpose.
• A model is the mathematical construct generating an inference or prediction from new input data.

When the guidance refers to an "AI-enabled device," it means the whole product; "AI-DSF" refers only to the AI-using function; "model" refers to the mathematical construct alone.

Scope of marketing submissions

The guidance addresses documentation for marketing submissions — including 510(k), De Novo, PMA, HDE, and BLA pathways — and some recommendations may apply to Investigational Device Exemption (IDE) submissions. It also applies when a combination product's device constituent part includes an AI-DSF.

FDA takes a risk-based approach: specific testing and documentation vary by device type, intended use, and automation level. The guidance does not replace device-specific guidances or the broader "Content of Premarket Submissions for Device Software Functions" guidance.

Relationship to GMLP and PCCP

This draft guidance builds on related FDA and international efforts:

Good Machine Learning Practice (GMLP)

FDA co-developed ten GMLP guiding principles with Health Canada and the UK's MHRA. GMLP promotes safe, effective, high-quality ML across the device lifecycle. This TPLC guidance operationalises many of those concepts into concrete marketing submission content — for example, representative data, human–AI team performance, and post-deployment monitoring.

When reviewing vendor materials, Hong Kong clinicians may find GMLP principles in plain-language summaries and TPLC-specific evidence in FDA submission structures.

Predetermined Change Control Plan (PCCP)

AI models may need updates after deployment to maintain or improve performance. FDA's separate PCCP guidance describes how manufacturers can prospectively specify and seek premarket authorisation for intended AI-DSF modifications — for example, retraining to improve performance — without a new marketing submission for each change, when implemented consistent with an approved PCCP.

The TPLC guidance cross-references PCCP for devices that automatically or continuously update, and for postmarket change decisions.

Terminology: FDA vs the AI community

The guidance explicitly clarifies terminology differences. In medical device submissions:

• "Validation" follows 21 CFR 820.3(z) — confirmation with objective evidence that requirements for a specific intended use are consistently fulfilled. Avoid using "validation" in submissions to mean training or tuning.
• "Development" refers to training, tuning, and tuning evaluation (often called "internal testing" in AI research).
• "Test data" supports verification and validation activities and is not part of development.

FDA's Digital Health and Artificial Intelligence Glossary provides additional definitions.

What this learning path covers

The following articles summarise the draft guidance's major submission sections in plain language, with practical implications for Hong Kong healthcare settings evaluating AI devices. Each article includes a short quiz.

Source: U.S. FDA — Artificial Intelligence-Enabled Device Software Functions: Lifecycle Management and Marketing Submission Recommendations (January 2025, draft guidance)