If Victorian results could be replicated nationwide almost 2500 clinically unnecessary non-urgent care visits and almost 20,000 ED visits could be avoided each year.
AI-powered triage was able to recommend the right level of care for people who called seeking advice on whether to go to ED or elsewhere, according to research based on a Healthdirect trial completed last year.
“Virtual triage is proving to be the front door Australia needs to connect people to the right care, at the right time, through the right channel,” Healthdirect Australia CEO Bettina McMahon said on social media.
“Emergency department intent dropped 33% as patients shifted to appropriate (and lower cost) virtual and urgent care options.
“Patient uncertainty about where to seek care fell 99.6%.
“Non-urgent care engagement more than doubled, and virtual emergency care adoption surged 1613% as these services matured,” said Ms McMahon.
“This isn’t about replacing clinical judgment. Our nurses agreed with AI recommendations in 83% of cases.
“It’s about giving clinicians better tools and patients clearer guidance—especially during the 69% of encounters that happen outside business hours when access is most limited.”
Virtual triage and care referral uses AI to assess a patient’s needs, triage them, and if appropriate, refer them to the correct service for the level of care they need, that isn’t an in-person emergency department.
Healthdirect introduced one of these systems – Infermedica’s AI-driven VTCR engine – “to serve as a digital or virtual front door to care and to improve patient access to and engagement of health care information and services available 24/7/365”.
For the purposes of evaluation, every triage decision was also undertaken by a nurse.
The evaluation study, carried out by Healthdirect and Infermedica, was published this week in the journal Mayo Clinical Proceedings: Digital Health.
The AI-powered VTCR offered people who contacted Healthdirect an assessment of their symptoms and, based on that, personal referral recommendations to emergency services, urgent care, non-urgent general practice, pharmacy, mental health, midwifery care or self-care.
It did not provide a diagnosis.
Healthdirect’s aim was to give people certainty about where they should go for treatment and to prevent people unnecessarily seeking a higher level of care than they needed (and the costs associated with that).
The AI-powered virtual front door improved patient experience and accurately triaged people, diverting them to the correct treatment service, according to the evaluation findings.
These were based on 1,552,592 VTCR patient encounters across all Australian states and territories except Queensland (which has its own phone line), from 1 April 2023 to 31 March 2025. They were all asked prior to their interaction with the VCTR where they intended to go for care and then asked again after.
The results showed that the VCTR almost entirely eliminated people’s uncertainty about where they should be going to get the right care for their problem (from 670,502 to 2557 patients, down 99.6%).
The number of people who said they were going to an in-person ED went down from 119,414 (36.7%) to 105,349 patients (24.6%) (–12.1 PP; P<.01).
Virtual triage and care referral more than doubled the number of patients who selected appropriate, lower acuity non-urgent care from 330,279 (21.3%) to 820,800 (52.9%), an increase of 31.6 percentage points.
The number deciding to use virtual (rather than in-person) ED went from 612 to 11,840 patients (+10.1 PP) and to use nonurgent virtual care went from 20,467 to 26,289 patients (+2.9 PP; P<.01).
Victoria had the highest rate of adoption of the new care pathways, the study found. The number of people intending to go to in-person non-urgent care went down 44%, and the number intending to go to in-person ED went down 47.7%.
More Victorians decided to access non-urgent telemedicine (up 8.7%), take up new urgent care services (0 to 2353) and virtual emergency care (up from 228 to 8424 encounters).
“If the magnitude of change in acuity level of care intent achieved in Victoria is applied to encounters occurring across all Australian states annually and the volume of use remains constant, 2409 clinically unnecessary in-person non-urgent care visits would be avoided per annum, and 19,286 unnecessary in-person emergency care visits would be avoided yearly once VTCR implementation reaches maturity in all states and territories,” the study said.
The AI’s recommendations were largely correct, the evaluation found.
Nurses concurred with 83.3% of AI triage recommendations. Where they did not agree, nurses generally set a higher triage level.
That “may be due to additional clinical information available to them, such as patient distress, not captured by VTCR,” the researchers said.
“It is also possible that this difference may reflect VTCR error. This warrants further investigation, given that VTCR is configured to overtriage to minimize patient risk.”
They said that prior research on the Infermedica VTCR engine showed that it provided “safe recommendations in 97.8% of cases” – comparable to triaging by human nurses using rules-based triage protocols.
In the 2.2% of cases where the VTCR’s triage was not correct, there “was a maximum of 1 level less conservative than expected, which is therefore not significantly different from expectations”. This was lower than the level of under-triage for nurses using a clinical decision support system (3.7%) and GPs using a clinical decision support system (7.3%), the authors wrote.
During the period studied, most people called the helpline (and used the AI tool) out of normal office hours (69%). Sixty percent were female and 70% were under the age of 44. Just under 6% were Indigenous, with 41% of those under the age of 18 (compared to 34% of non-Indigenous) and 23% aged 18-29 (compared to 17% of non-Indigenous callers).
“What strikes me most: Aboriginal and Torres Strait Islander patients engaged with virtual triage at higher rates than the general population. This technology can reduce – not widen – healthcare inequities when designed and implemented with care,” Ms McMahon said.
Of course, what people say they will do and what they will actually do may not be the same thing, and that requires further study, the authors noted.
They also pointed out that it is yet to be proven that using a VTCR improves clinical outcomes and reduces health costs.
Read the full article here.
