The TGA is quietly looking into how AI in healthcare is controlled, and they need your input.
AI development is moving fast in healthcare, and the Therapeutic Goods Administration has shown it is along for the ride.
The TGA is seeking feedback on a draft guidance document outlining its approach to predetermined change control plans – a pre-approved roadmap for how a medical device, usually software or AI, can be modified after it’s on the market, without needing a full new regulatory submission each time.
The guidance shows how PCCPs can gain premarket regulatory approval to implement a change at a future stage. It signals a shift from approving each software update to approving how devices are allowed to change over time.
This article originally ran on TMR’s sister site, Health Services Daily. TMR readers can sign up for a discounted subscription.
The TGA gives several examples of how a PCCP might apply.
The first is an AI-powered eye diagnostic tool where the manufacturer wants to retrain the model after two years of real-world Australian data to improve its ability to detect eye conditions. Provided there are no changes to the intended purpose or user population, the intended PCCP could apply.
Another example is AI software that analyses chest x-rays to flag areas for radiologists. In its initial application, it might be compatible with two x-ray machine models. However, the manufacturer plans to expand compatibility to two more models in three years. A PCCP can cover the testing, validation, labelling, and training process needed to do that.
The third example involves a mobile app initially built for iOS, with plans to add Android compatibility within three years. The approach depends on the device’s risk classification: Class III software medical devices need a new application for the Android version, while Class IIb and below can use a PCCP.
“A PCCP allows manufacturers to more rapidly update their software, with faster access for users and patients, and to responsibly evolve their software in response to new data and technology. PCCPs can help balance innovation and regulatory oversight without compromising patient safety,” the TGA guidance suggests.
Importantly, it doesn’t remove the need for clinical evidence or verification and validation to be performed and documented.
“PCCPs may affect how, when, and if we review the clinical evidence or the verification and validation results, according to risk,” the guidance highlights.
Only pre-defined, low-risk changes are allowed that include ensuring the device’s purpose doesn’t change, safety-critical functions aren’t modified, and all strict performance, safety, and testing criteria are met.
A PCCP could be suitable for:
- Retraining AI models;
- Adding new data sources;
- Tweaking algorithms;
- Updating interfaces;
- Routine maintenance.
Post-market monitoring will be used to make sure changes implemented under the PCCP remain safe and effective.
Related
“Sponsors must maintain sufficient records to determine whether a reported issue or adverse event is associated with a software change implemented under a PCCP. This includes traceability to the relevant pre-authorised change, associated verification and validation evidence, and risk management documentation,” the guidance says.
In a statement, a spokesperson for the Department of Health, Disability and Ageing told The Medical Republic:
“The TGA recognises that software plays a critical role in the delivery of healthcare and develops at a rapid pace when compared to traditional technologies.
“Our proposed approach to PCCPs has been developed to be broadly consistent with draft guidance drafted by the International Medical Device Regulators Forum.
“Key benefits of PCCPs allow software developers to obtain regulatory approval more rapidly when they update their software, enabling faster deployment and access for users and patients.
“For patients and clinicians, the improved speed to market for pre-approved changes support continuity of care through timely access to improved medical device software.
“Risks will be mitigated as PCCP approval is proposed to be limited to relatively low risk manageable changes.”
The consultation is open until 5 June 2026, as the TGA seeks to finalise its approach to regulating how AI-driven medical software evolves in practice.
