Are we ready for AI to graduate up the data ladder from scribing notes to access patient analytics in our country’s biggest GP patient management system to help GPs optimise their patient consults and billings?
On Thursday the country’s biggest GP patient analytics business integration, Cubiko, announced it had combined with one of the country’s two fastest growing AI scribe startups, Lyrebird, and our largest GP patient management system, Best Practice, to automatically optimise patient lists, pre-load data on patients for a consult, identify patients for chronic disease care plans and optimise the MBS numbers to bill such plans.
What could possibly go wrong with that?
According to Lyrebird’s website, nothing – because in the end, it’s ultimately the responsibility of the clinician to decide on who they put on a plan, and to decide whether to use prompted billing codes for patients who they might be putting on that plan.
The AI is merely acting as an intelligent assistant.
This probably puts the initiative beyond the reach of the TGA in terms of classifying any of what is happening as software as a device with all the heartache in testing, governance and potential liability that might cause an AI vendor.
As far as AI scribing is concerned so far, it’s just summarising notes, and those notes must be read by the clinician before they are entered as any sort of formal record. The liability sits with the clinician for any mistakes.
The question some AI healthcare experts are asking is, if the process becomes too repetitive, and the clinician begins to trust the process, will they really keep reviewing the notes or simply start trusting the AI?
The same problem feels like it could occur with the sort of automation this new AI driven productivity tool on offer from Lyrebird, Cubiko, Best Practice is offering.
Lyrebird assures potential users of the service that all the patient data is secure, although it’s clear that this extension of AI into “helpful administration tasks” into the core Best Practice Software means that Lyrebird is accessing the data that Cubiko is surfacing in the form of care prompts in Best Practice.
The AI isn’t accessing all the detailed patient data that Cubiko can, but it is using selected patient data that Cubiko is generating, and then interacting with the core Best Practice patient management software to complete its automation tasks for the GP.
Cubiko is used and trusted by thousands of GP practices and already has a chronic care prompt alert system. All that the Lyrebird integration is doing is automating that part of the Cubiko intelligence, adding a little specificity on MBS billing around that process, presumably using data from Best Practice, and inserting it into the workflow of Best Practice to make everything smoother for a doctor.
That means that the patient data that Lyrebird is accessing for its processes is as good as the data Cubiko is outputting, so the chances that the AI would do any sort of hallucination that you see reported regularly these days for generative AI applications is near non-existent.
Maybe nothing will go wrong and we will have taken a much bigger step on the part of AI and GP productivity, which is pretty much the marketing spiel from all the companies on involved in the initiative.
A press release from Cubiko reads:
“This new integration securely shares Cubiko’s Care Prompts with Lyrebird Health via Best Practice, allowing Lyrebird to intelligently recommend the most relevant consult templates based on a patient’s potential service eligibility. The result is a smarter, faster workflow for GPs, reducing administrative burden while ensuring patients receive the care they need, when they need it most”.
Maybe this isn’t such a big deal and there is just nervousness because we are talking about AI now manipulating some sets of patient data.
Related
Most of the major GP PMS vendors have touted that their initial AI scribing functions would likely move upstream, not all the way to clinical decision support, but into the just the sort of administrative function announced by Cubiko, Lyrebird and Best Practice in this initiative.
Coincidentally, this week we published an opinion piece by the immediate past Chief Product Officer at Best Practice, Danielle Bancroft, that may have been touching on the possible issues you could run into when you build outwards in the manner, speed and singular focus that these three groups seem to be travelling with this first foray into AI administrative help for GPs.
The Bancroft piece is about the need for “whole of system thinking” when building clinical systems software that, and she warns that:
“It’s common to see health tech companies zero in on a single task like clinical notes, bookings, or tasks, and try to optimise it in isolation.
“But without mapping the full workflow end to end, including the parallel automations and conditional logic behind the scenes, they risk introducing friction, duplicating effort, or worse, disrupting clinical safety”.
She goes on to say that design has to “reflect clinical thinking, not develop logic”, noting that clinicians using the software are usually juggling cognitive load, time pressure, compliance frameworks, safety guidelines and legal risks (i.e., they’ve not got much time or headspace to consider what the software might be doing under them).
Notwithstanding, Bancroft was in favour of AI moving upstream into administrative tasks when she worked at Best Practice.
It was just a few weeks ago that Cubiko announced that it had developed a smart billing module which was a form of competition with long-term billing automation vendors such as KPeyes and Surgical Partners, some of whom have successfully completed integrations with the big banks.
That launched the group into the heady world of tax compliance, not just payroll tax compliance, but potentially soon, compliance with rapidly evolving federal tax law and regulation on “deemed employees”.
The competitors stayed pretty silent on Cubiko’s foray into billing automation, but did note to Medical Republic that the billing, accounting and tax side of automation was a notoriously complex world in which to play because of tax and governance requirements so the group would need to take care.
Other raised eyebrows on this AI automation initiative might come from the Medicare regulator AHPRA, of the Department of Disability, Health and Ageing Medicare, both of which spent much time over the past decade discouraging any sort of automation of care plans by individual practitioners.
On the one hand, such automation, if it works out should be welcomed by the healthcare system as GPs catching far more patients for longer term chronic care management. On the other, it’s going where no one has really been before.
If it were truly efficient and worked at scale, which it likely could being inside Best Practice, it feasibly could cause some upward movement in the fee for service funding envelope.
While there are only two MBS items for chronic care plans for GPs, there are very complex rules around co billing, and types of consults.
If the data munchers deep inside Medicare start seeing payment patterns they don’t like, that’s likely to reverberate back through one of those nasty AHPRA nudge letters.
It’s a bold and potentially revolutionary step taken by these three leading GP software vendors.
Hopefully we will look back and see it as a step in the right direction.