Survey data suggests nearly four out of five clinicians in New Zealand have started to use AI tools without a clear government framework in place.
The data – collected and compiled by a healthcare AI company – indicates AI tools are mostly being used to save time on documentation work, and have become widespread across New Zealand hospitals despite the country’s relatively AI-sceptic population.
One New Zealand healthcare representative says the technology is saving time, and any faults it may have are no worse than the underlying problems with the country’s healthcare IT system: problems she says we’ll need to fix if we want to make the most of AI.
The Government bought 1000 licences of healthcare AI scribe Heidi last year after a trial in Hawkes Bay and Whanganui found it saved up to 10 minutes per patient.
Now, the same company is releasing data from a survey of 1823 clinicians across 25 countries, which suggests nearly 90 percent of clinicians worldwide are using the technology daily or several times a week. Heidi did not provide Newsroom with a comprehensive breakdown of statistics per country.
Most of the technology’s impact is on the administrative side of healthcare. Most clinicians report spending at least an hour of their workday on admin work; one in three clinicians spend at least twice that amount.
Documentation work across the globe is the biggest time suck. Most clinicians say it’s reducing their ability to spend time with patients, maintain a manageable schedule and stay up to date with current research.
Now, Heidi’s data says 79 percent of New Zealand clinicians are going ahead with the technology without a clear governance framework in place.
Heidi’s head of legal and regulatory affairs Yass Omar says institutions must keep pace with the workforce. “Clinicians are already using these tools and forming their own governance instincts while regulation catches up,” he says.
Simon Kos, the company’s chief medical officer, says the sector is calling for clear governance, confidence in the technology and integration: “They’re not waiting, nor are they asking permission.”
Ben Condon, Heidi’s clinical director, says New Zealand moved through the rollout phase faster than most countries. “AI has been rolled out at scale in Health NZ with clear engagement and institutional governance resulting in clinicians now reporting lower levels of regulatory uncertainty.”
New Zealand still does not have an overarching policy on AI beyond the AI strategy released last year.
Condon says the findings suggest investment in governance, implementation and cultural engagement is helping the country build clinicians’ confidence in the technology.
New Zealand clinicians’ biggest worry is not a lack of guidance, it’s that the AI will hallucinate or provide false information – a fear that rates higher than patient privacy, reliance-forming habits and erosion of clinical judgment.
For Andrew Lensen, an AI expert at Victoria University Wellington, more oversight is needed to ensure the technology is used to the best of its ability. “The use of AI in healthcare is very promising – but, equally, very concerning if it is being used without sufficient oversight and assurance,” he says.
He also questions how much free and informed consent can be given by patients. “On balance, I think the use cases I’ve been exposed to are a net positive for my care,” Lensen says, but asks if all patients can make the same judgement calls.
In either case, some clearer rules would answer these questions.
In New Zealand, a lack of firm AI policy and a generally sceptical population has not stopped the technology’s rollout. Heidi itself was selected by the Government to be introduced across New Zealand; Health Minister Simeon Brown announced the purchase of 1000 licences in November last year.
The trial suggested doctors could save up to 10 minutes per patient, reducing paperwork time from up to 17 minutes to just four.
Sylvia Boys, national vice president of the Association of Salaried Medical Specialists, is a Heidi user and vouches for its efficacy. She says it has its issues, but overall saves time – and wider issues in the health sector’s tech system are a bigger problem.
Being an ambient scribe (a tool that takes notes from patient-clinician conversations) Boys says Heidi sometimes finds it difficult to tell when a new person enters the room and starts talking about a different patient, confusing its notes. Because it doesn’t train on its recorded data, the Australian-built AI can’t recognise or transcribe many words in te reo Māori, which is especially problematic for recording place names. It also can’t recognise non-verbal communication like head shakes.
Boys said it’s even hallucinated in the past, once accidentally noting a patient’s history of rheumatoid fever based on Boys asking if the patient ever remembered being diagnosed with it.
But these are relatively minor issues, and Boys says the wider tech system in the health sector is no better. Even in the documentation space, the previous software Boys used relied on manual entry and didn’t have the ability to spell check. “So if you’re quite a dyslexic typist, then that’s a major problem,” she says.
Most of the clinicians surveyed in Heidi’s report had gone ahead and used AI without official guidance, and Boys says this was true for her colleagues here in New Zealand. When the 1000 Heidi licences were rolled out, the official direction was just “give it a try and see how it goes”, says Boys.
All things considered, “for a product that we’re just trying how it goes, it’s generally worked quite well”, she says.
Boys does not want to see regulation or guidance on how and when to use the technology coming from the government; she thinks it’s something clinicians can sort out for themselves.
“I would hate to see central government get involved with guidance – they are just too removed from the reality.”
The reality of the situation, she says, is an IT infrastructure nightmare. Systems aren’t joined up to seamlessly communicate, patient records are sometimes fragmented across different platforms, and staff are stuck with outdated hardware.
Even something new like Heidi relies on these systems. Boys says if a hospital doesn’t have enough computers, or if their computers are slow, staff aren’t able to sync Heidi to the laptop they work on.
Boys says they have to manually email Heidi’s notes to themselves, then log in to a work station, and cut and paste that email into the patient’s notes. “In that clunky process, you start to lose a lot of the time savings.”