2024 was the year for ambient artificial intelligence medical scribes. As Fierce Healthcare Staff Writer Emma Beavins forecast in December, this year, the healthcare industry will expect more out of AI.
AI agents and AI copilots have been dominating headlines and taking over showroom floors at recent healthcare conferences.
Electronic medical record company Canvas Medical also is advancing AI technology but is taking a distinctive approach. It built an open-source AI-enabled clinical copilot for clinicians, called Hyperscribe, and developed it with the Canvas software development kit (SDK).
Canvas also aims to set a higher standard for AI governance and performance with publicly accessible source code and performance benchmarks.
Canvas customers can modify the Hyperscribe copilot to fit their own needs, integrate into any large language model and customize as needed, all with deeper functionality, according to executives.
Adam Farren, CEO of Canvas, described Hyperscribe as a "new paradigm" for how providers document in an EMR, but the technology goes beyond an AI medical scribe to take action on behalf of the user.
Canvas, launched in 2015, developed a new EMR architecture for primary care and specialty care providers and then built developer tools and bidirectional FHIR application programming interfaces. Data scientist Andrew Hines founded Canvas Medical 10 years ago to develop a "human-centered EMR" after observing his wife, a family nurse practitioner, grapple with bad medical software.
Canvas initially focused on building the core EMR platform with revenue cycle, practice management, scheduling and claims capabilities.
In 2021, the company launched its developer dashboard to enable developers to build tools on top of that core platform. Farren, who joined the company as president and chief operating officer in 2023, stepped into the CEO role last year, and Hines transitioned to chief technology officer.
Canvas gained traction with healthcare organizations that found an out-of-box EMR didn't meet their needs, Farren said. The company offers an EMR that can be customized for providers' workflows and to meet clinical and financial needs. In December, the company made its Canvas SDK generally available, laying the groundwork for more workflow automation, ecosystem integration and the adoption of AI copilots and agents.
Hyperscribe is the "next phase" of innovation for Canvas' EMR platform, Farren noted.
"We are seeing explosive demand for AI agents that can do work in Canvas, in collaboration with the care team. With Hyperscribe we are accelerating into that future, demonstrating a new level of compounding value from agents working together on the Canvas platform, made possible with our publicly-available developer tools," he said.
Hyperscribe is currently available in beta for early adopters via an application process.
Canvas' copilot uses ambient audio to write clinical documentation and orders while working as a team, so to speak, with other AI agents, executives noted.
A key feature of Hyperscribe is advanced processing time and speed to enable continuous updates to the patient record and notes during an appointment to give clinicians a live, up-to-date view of documentation and orders, executives said. Hyperscribe also taps into the patient’s full medical record and ongoing updates from prebuilt data integrations to give clinicians complete patient context.
"The output of Hyperscribe is aware of that context, which is really powerful as far as accuracy and time savings for the end user," Farren said.
The technology goes beyond drafting orders to executing tasks. It can use clinical informatics guardrails to safely interpret conversations and accurately infer details about vocalized observations and care plans, executives said.
"Because Hyperscribe is built on top of the Canvas platform, it has these built-in guardrails; there's a human user in the loop. There's a bunch of safety logic built into Canvas that it can take advantage of that are both organizational and system-driven. Some of those guardrails are logic that Canvas has created out of the box, others are things that organizations control," Farren noted.
With Hyperscribe, multiple AI agents can collaborate, what Hines refers to as "chaining." This means outputs from one task can immediately trigger and provide information for the next task, which can help streamline processes like patient referrals, eligibility verification, preparing insurance claims and supporting clinical decisions.
As an example, during a patient exam, a clinician might tell the patient she wants to follow up in a week. Hyperscribe will translate that audio and infer the details to create and schedule a secure portal message to that patient that will be sent in seven days, even adjusting for time zones or holiday schedules, Hines noted.
"For that to happen, it has to chain with the scheduling agent that's then going to pick that up and execute that. That's what Hyperscribe does," he said.
Hines provides another example of a clinician examining a child's ear infection and telling the child's parent in the room that he is prescribing an antibiotic. Hyperscribe will update the clinician note and draft a prescription medication order, with the specific drug and dose and preferred patient pharmacy. The clinician just has to review it and click send, he noted.
"It does it in a way that observes contraindications and observes allergies and puts the patient at no harm. These are examples of chaining with medication safety agents, drug-drug interaction and drug allergy interaction. And that's the world that Hyperscribe ushers in for Canvas users, being able to rely on that kind of multi-agent collaboration to get the executable details right," Hines said.
As a unique point of distinction, Canvas used open-source development for Hyperscribe, as the code and evaluation framework are available for anyone to use. This enables organizations to enhance it on their own, integrate their foundation model of choice or customize their own solutions for specific use cases, executives said.
Canvas aims to be open and transparent about what its building, Farren noted, and makes its performance benchmarking and evaluations accessible.
AI governance in healthcare is becoming increasingly crucial as organizations both broaden and deepen their adoption of AI-powered products.
Organizations such as the Coalition for Health AI and the Joint Commission are moving forward to define AI governance frameworks.
Hines contends that practical examples of transparent evaluations and open benchmarks are virtually nonexistent.
"It comes down to those business motivations and realities that governance in AI deployment in clinical settings is—I would characterize it as abysmal. It is a time bomb. It is a liability for these caregiver organizations and it ultimately lands with the clinician," Hines said. "That's just an unacceptable solution. We feel like it's actually a market advantage for Canvas to say we're going to pick up more of that for you."
He added, "We are going to provide total transparency on how AI is being used, which AI is being used and not only total transparency, but also control. You want to use a different LLM? Maybe you've done some fine-tuning, maybe you're working with a researcher that has a new LLM that's disease-specific. You want to put that in for your kidney disease patients? Great. You can do that. You can see everything. You can reproduce our evaluations."
Canvas aims to set a new governance standard by publishing reproducible evaluation code, the underlying evaluation data set and a growing set of benchmarks comparing different foundation models with Hyperscribe and other third-party AI scribes and copilots, executives said.
Advancements in technology can often lead to increased burdens for providers if not implemented properly or if providers don't trust the tech tools, Farren noted.
"Done wrong, AI doesn't help make clinicians more efficient, AI is actually a burden on those clinicians. Or, they just don't adopt it, and that's not how it should work. We want to put our money where our mouth is, so to speak, and one way is by holding ourselves accountable to standards, publishing those standards and making them open," he said.
In 2021, Canvas Medical picked up a $17 million series A funding round, followed by $24 million in series B financing in 2022. That round was led by M13 with participation from existing investors including Inspired Capital, IA Ventures and Upfront Capital. The same year, its EMR became federally certified by the Office of the National Coordinator for Health IT.