Implementing AI at a nationwide scale, notably in such vital industries as healthcare, is not any straightforward feat.
Within the HIMSS24 APAC session, “Implementing AI-Powered Triage for a Nationwide Digital and Digital Entrance Door,” Dr Nirvana Luckraj, chief medical officer of Healthdirect, described how they selected, examined, and deployed an AI-powered medical choice help system to reinforce its nationwide helpline companies.
Healthdirect gives a spread of free 24/7 helpline companies, together with nurse triage and GP telehealth. It has devoted strains for pregnant individuals, the aged, and most not too long ago, folks coping with psychological well being issues. Moreover, Healthdirect supplies data on medicines and curated well being content material on their web site and options on-line reserving and referrals, which immediately combine with nationwide care pathways together with My Well being Report.
THE PROBLEM
Buying a CDSS is a part of the actions Healthdirect took when it not too long ago migrated its expertise stack to its personal working platform.
It required the device to contemplate a number of signs on the outset. “We wanted a CDSS that might tolerate and analyse a number of signs. We additionally wished it to mirror actual session that thought of danger components equivalent to predisposing components, persistent historical past, and journey historical past.”
“We additionally wished one thing that was subsequent era that might present a extra complete and correct triage,” Dr Luckraj added.
Beforehand, the service used a CDSS primarily based on a call tree-type algorithm which required “in depth upkeep” and “may solely tolerate a single symptom enter.”
PROPOSAL
The organisation then went to the market to hunt a supplier of an AI-based CDSS. It initially shortlisted 5 international distributors whose choices had been subjected to medical testing.
To check the AI system’s applicability within the Australian public well being system, Healthdirect labored with AI specialists at Macquarie College’s Australian Institute of Well being Innovation. They adopted a take a look at methodology from a latest examine that used case vignettes to examine a CDSS’s accuracy.
“We ran these vignettes by means of the CDSS and located the efficiency was according to the not too long ago printed examine. We additionally ran the identical vignette by means of our previous CDSS to match and located that the previous and new CDSS had related outcomes however that the brand new one was extra correct,” Dr Luckraj famous.
Ultimately, Healthdirect went with the Polish agency Infermedica’s AI CDSS, which runs on a probabilistic Bayesian inference engine that considers signs, danger components, and ailments, an up to date medical information base, and an API that connects to its CRM.
MEETING THE CHALLENGE
Instantly after launching the brand new CDSS on its helpline in March 2023, the implementing staff encountered resistance from workers. “We had underestimated the change administration required when getting healthcare professionals to make use of new expertise to carry out previous duties,” Dr Luckraj admitted.
From inputting one symptom to the CDSS, their nurses now must enter a number of signs of sufferers. “The nurses initially had been pissed off with the device as a result of it appeared to ask seemingly unrelated inquiries to the presentation of the affected person. Additionally they needed to decide their previous historical past to find out particular person danger components and the questions additionally appeared extra superior than the earlier.”
In coping with nurses’ preliminary frustration, the staff took on their suggestions and assisted them in adapting to the brand new device. One of many key areas for enchancment right here, based on Dr Luckraj, was working with their vendor on a suggestions loop mechanism the place nurses don’t have to agree with each CDSS output.
“It is a help system, not a diagnostic system. So the nurses can both improve or downgrade the triage consequence giving a cause for his or her change of disposition. We additionally checked out these flows the place the nurses disagree with the CDSS output to constantly enhance the device.”
RESULTS
A yr after bettering the CDSS, the implementing staff noticed that nurses are altering triage output much less and agreeing extra with the CDSS outcomes and that they’re extra happy with its new consumer interface. The device’s accuracy is consistently bettering, Dr Luckraj famous, as nurses additionally proceed to offer suggestions. On the buyer facet, no critical medical incidents attributable to the CDSS have been reported; they’re now extra happy with the service, with satisfaction scores reaching the very best in three years.
In December final yr, the identical AI-powered CDSS was launched in Healthdirect’s consumer-facing on-line symptom checker, making it a “digital entrance door to quite a few different care choices, together with direct entry to pressing care companies and digital emergency departments.”
“Sooner or later, we’re additionally trying to present shoppers with tailor-made data particular to their symptom enter to offer a good higher client expertise,” Dr Luckraj shared.
Since incorporating AI into their helpline, Healthdirect claims to divert half of emergency calls to much less acute companies. From July 2023 to June 2024, it linked 44,000 calls to a digital emergency division. It additionally booked almost 5,500 folks on-line for GP-led pressing care companies. In the meantime, nearly 350,000 folks had been suggested on safely managing their signs at dwelling. Total, Healthdirect nurses helped ebook over 370,000 GP service appointments.