AI in health care: a practical guide for Australian GPs

Before implementing an AI documentation tool, there are some key medicolegal issues to note.

Ambient scribes that simply capture and translate talk into text do not generally require an ARTG listing. However, if a scribe examines or interprets clinical discussions (eg by offering differential diagnoses or treatment recommendations you did not suggest), it falls into the AI medical device category and must be registered with the Therapeutic Goods Administration (TGA).

It’s a criminal offence to record a consultation without consent in most Australian states and territories (Victoria, Queensland and the Northern Territory have partial exceptions).

To avoid a breach, you must obtain explicit, informed consent from all parties in the consultation — including any medical students, trainees, or family members as well as the patient — before activating an ambient scribe.

You also need to evaluate whether the solution complies with Privacy Act healthcare obligations and ensure health information is de-identified. Removing the patient’s name and date of birth from recordings or transcripts may not be enough if they can be otherwise recognised.

Artificial intelligence in medicine is constantly evolving, and Ms Nikolic notes mistakes can still happen.

“All AI tools can be prone to hallucinations, incorrect interpretations of data and errors. The important thing is that there is always a ‘human in the loop’ to pick up on these.”

The RACGP recommends getting advice from your Medical Defence Organisation or a lawyer to ensure you’re complying with relevant laws for using AI scribes.

Some medical software solutions listen to consults and suggest Medicare Benefits Schedule (MBS) item numbers to bill, including co-claimable items, chronic disease management codes, or telemedicine descriptors. By catching missed billing opportunities, tools like MediBetter and MBSPro may help optimise your earnings.

You remain responsible for billings under your provider number, so it’s essential to evaluate the AI’s suggestions, only claim clinically relevant items, and ensure your records accurately reflect MBS descriptors.

Using AI to search the academic literature

The New England Journal of Medicine and the Journal of American Medical Association (JAMA) have created with other partners an LLM trained on medical literature. ‘Open Evidence’ calls itself an AI copilot for doctors. Like all AI models it can misinterpret research or hallucinate, but it has been described as a “valuable but evolving tool” in this article by doctors and medical students in the US.

Clinical decision support and diagnostics

Machine learning in healthcare is also incorporated in specialised tools that help clinicians make clinical choices. These solutions typically use pattern recognition and predictive analytics to detect anomalies and clarify diagnoses.

For example, AI diagnostic imaging software can evaluate skin lesions during skin checks. Other AI models can analyse chest X-rays and brain CT scans. Australian research has shown AI can detect breast cancer in screening mammograms, helping stratify risk.

Currently, these models are mainly used by specialists, but “In the future, we foresee that diagnostic and decision-support tools will help GPs with clinical practice,” Ms Nikolic said.

Clinicians need to be aware of AI bias, she adds, including potential algorithmic bias introduced in model training. “If the data is not representative of the GP’s patient cohort, the tool that relies on it will not be optimal.”

AI in healthcare: Pre-adoption checklist

The Australian Digital Health Agency (ADHA) stresses that robust clinical governance helps ensure digital health solutions function as intended, reducing risk of harm and supporting good outcomes.

Ms Nikolic suggests GPs thinking about deploying AI ask themselves three things: “Am I educated and informed on AI use in general, and this tool more specifically? Do I have the right governance and risk management frameworks in place? Am I involving the patient in informed decision-making about the AI tool I am using for their care?”

Who is legally responsible if an AI tool makes an error?

You are. Ahpra makes it clear the practitioner is personally and professionally accountable for all clinical decisions, findings, and medical records.

Do I need explicit patient consent to use an AI scribe during a consultation?

Yes. You must obtain informed patient consent before using any tool that records, transcribes, or processes a patient’s personal or health information.

Is it legal to use an AI scribe if I don't record the audio permanently?

Yes, but temporary processing still constitutes recording under Australian surveillance laws. You must obtain explicit consent from everyone in the room before activating the tool, regardless of whether the audio file is deleted immediately after text is generated.

Can I trust an AI scribe to suggest the correct Medicare Benefits Schedule (MBS) item number?

While AI tools can suggest MBS item numbers, it is the treating practitioner’s responsibility to ensure any services billed to Medicare satisfy their requirements.

Disclaimer: The AI tools discussed in Insight+ may be inaccurate. They are not a substitute for professional or clinical expertise. Responsibility for patient care rests solely with the treating clinician. Content relating to AI scribes and Medicare Benefits Schedule (MBS) items is for general information only and does not constitute billing or legal advice. Practices remain responsible for ensuring compliance with Medicare requirements and verifying the accuracy of all documentation and claims.