Summary:
Le gouvernement australien a alloué 39,9 millions de dollars pour renforcer la réglementation des logiciels médicaux et de l’IA. La Therapeutic Goods Administration (TGA) est responsable de la régulation de ces dispositifs, visant à clarifier des définitions et responsabilités tout en maintenant une approche technologique agnostique. Les consultations avec les acteurs de la santé ont révélé un besoin d’amélioration dans la compréhension des obligations réglementaires. La TGA continuera de revoir et d’adapter les lois tout en s’alignant avec les normes internationales.
Original Link:
Original Article:
Report: Clarifying and strengthening the regulation of Medical Device Software including Artificial Intelligence (AI)
Executive Summary
In the 2024-25 federal Budget, the Australian Government provided $39.9 million over 5 years to develop policy and capability across government to support Safe and Responsible AI. This includes clarifying and strengthening existing laws and addressing risks and harms from Artificial Intelligence (AI), including a review specific to the health and aged care sectors.
The Therapeutic Goods Administration (TGA) is tasked with regulating therapeutic goods such as software and AI models when classified as medical devices under the Therapeutic Goods Act 1989. In 2021, classification levels for these devices were clarified, and “carve-outs” for very low-risk software were introduced. Since then, ongoing monitoring has identified areas requiring further review and adjustment due to the rapid development and deployment of AI in healthcare.
The TGA’s 2024 review focused on examining existing legislation and identifying measures to enhance regulations for medical software and AI, recognizing emerging risks and opportunities. This involved consultation with various stakeholders, including the medical device industry, clinicians, and consumers, as well as mapping the existing legislative framework against the proposed mandatory guardrails from the Department of Industry, Science and Resources (DISR).
Key findings indicate that while the legislative framework is largely fit for purpose, AI’s evolving role in healthcare necessitates ongoing refinement. The TGA will ensure their regulatory approach remains risk-based and technology-agnostic, collaborating with international regulators and stakeholders to adapt to new developments. Further consultations will be conducted during 2025 and 2026 to refine regulations and compliance activities.
Findings
The TGA’s review provided findings in the following areas:
1. **Legislation**
– **Definitions**: Stakeholders recommended reviewing legislative definitions, such as “manufacturer” and “sponsor,” to align with evolving AI-related roles like “developer” and “deployer.” Refinements or clarifications through guidance may be necessary.
– **Supply**: The definition of “supply” needs revision to ensure that digital and virtual access to software-based devices is covered within regulations, especially for software hosted internationally.
– **Responsibility**: Language in the Act should ensure appropriate accountability for AI’s outputs, particularly when replacing traditional human roles.
– **Compliance**: Targeted actions are required to improve understanding of regulatory obligations among developers and deployers. Issues like the inappropriate use of AI in healthcare settings and resistance to complying with regulations were identified. A review of digital scribes is needed to assess compliance.
2. **Regulations**
– **Classification Rules**: Current rules for AI-based devices, particularly those making predictions or prognoses, were deemed broadly appropriate. However, future reviews are necessary as reliance on these devices increases in patient care decisions.
– **Essential Principles**: No changes to these principles are recommended, but additional guidance is required regarding adaptive and generative AI, dataset usage, and performance reporting.
– **Regulation of Specific AI Subtypes**: Regulation should remain technology-agnostic, without differentiating between subtypes of AI like Large Language Models (LLMs) or Neural Networks.
3. **Legislative Instruments**
– **Excluded Software**: Some exclusions, such as digital mental health tools, require urgent review, while others remain appropriate. Stakeholders need clearer guidance on exclusions.
– **Advertising and Transparency**: Improved measures are needed to enhance transparency, providing information like AI functionality and dataset details. Provisions in advertising regulations should be reviewed to achieve this.
4. **Guidance**
– **Website and Resources**: Stakeholders highlighted the need for better organization and accessibility of TGA resources. Specific landing pages for consumers and professionals should be added, along with educational content for understanding regulatory obligations.
– **Adaptive AI**: Explicit guidance is required for managing continuous change control in adaptive AI models. Methods like Pre-Approved Change Control Plans (PCCPs) may streamline assessments.
– **Datasets of Unknown Provenance**: Standards and validation procedures for datasets of unknown origin should be established.
– **Performance Monitoring**: Ongoing monitoring systems are critical to ensuring AI models maintain accuracy and reliability, with feedback loops from healthcare providers.
– **Healthcare Use of AI**: Users need help understanding the limitations and risks associated with AI-enabled therapeutic goods. Resources must be developed for consumers, clinicians, and developers to promote safe and effective use.
5. **International Harmonisation**
– Alignment with international standards is crucial to minimizing regulatory burden and ensuring timely access to innovative devices. Continuous engagement with overseas regulators and adoption of globally recognized guidelines should remain a priority.
Background
The TGA regulates therapeutic goods in Australia under the Therapeutic Goods Act 1989. This includes pre-market and post-market measures to ensure ongoing safety and performance. The agency remains committed to technology-agnostic, risk-based, and principles-driven regulation, ensuring flexibility and responsiveness to emerging challenges like AI-enabled medical devices.
Consultation
The TGA conducted extensive consultations during this review, including targeted stakeholder sessions, joint webinars with DISR, and public consultations. Contributions focused on identifying challenges and opportunities for refining Australia’s therapeutic goods regulatory framework.
Future Work
The TGA will ensure its framework for medical devices, including AI-enabled products, remains fit for purpose by pursuing ongoing developments in legislation, regulations, and guidance. Complementary collaboration with other Australian agencies and international regulators will further strengthen the regulatory system.