By Colin Leslie - June 2024, The Medical Post
Some doctors euphoric about new tech that produces SOAP notes automatically at end of patient encounter. But there is a lot to unpack about this fresh technology and how to find the best tools for your practice.
"I will never go back," gushed Dr. Rosemarie Lall, a family physician in Toronto, to Global News. "It decreases the cognitive load you feel at the end of the day!" She said the burden of paperwork had been so bad she was "dreading" her workday until she got an AI scribe.
For Dr. Ali Damji of Mississauga, Ont., the time-saving is astonishing. He started using an AI scribe at the start of the year and found for a clinic day with 14 patients (mix of in-person and virtual), he now spends a total of 30 minutes after the clinic on all documentation including referrals. He estimates using an AI scribe is saving him at least four hours during the week and at least two or three hours over the weekend. "This is almost like getting a whole day of your life back!" he observed.
To be sure, there are doctors who scoff and say that clinicians who finish on time, use templates and don't chart on evenings and weekends, simply don't need them. But the level of enthusiasm from some doctors about this tech is sky-high.
Essentially, AI scribe companies have taken a foundation model from one of the big AI providers like the creator of ChatGPT and then they have tuned it for medical use along with adding other components like speech-to-text and deidentification, explained Simon Ling, executive director for partnerships and stakeholders at OMD (also known as OntarioMD).
The effect for a clinician like Dr. Damji is that now he brings his iPad to each encounter and just talks with the patient while the scribe listens and writes a SOAP note at the end. Dr. Damji glances over that clinical summary and perhaps changes a few things and then, "It is a quick copy and paste into the EMR," he said. (Many doctors use AI scribes that are integrated with their EMR so they'll just have a microphone beside their computer to record encounters.)
"I find the conversation is more fluid, you're looking the patient in the eye," Dr. Damji said. He likes that the interface on his system uses verbal or written prompts like ChatGPT. "You just tell it what you want. For example, I have a note template and I just said, 'AI: write in point form. Be more to the point.' (Or) 'Write out the differential like this. Put a disclaimer at the end that I did a consent process.' And it just does it!" Or he tells it to write a referral based on the encounter and voilà.
Anecdotally, family doctors, psychiatrists and emergency physicians are the specialties most enthused about AI scribes.
Considerations
All of the AI scribes looked at by the Medical Post are Personal Information Protection and Electronic Documents Act (PIPEDA) compliant. For patients who have concerns about the AI scribe, Dr. Damji finds they are usually relieved when he "explains to them what PHIPA (Ontario's healthcare info act) means and that it has the same level of security as our EMRs that I'm charting into." All of the AI scribe companies provide patient consent forms.
"The general sense is it is a low-risk tool but it delivers great results," said Abhinav Kalra, executive vice-president for connected care at Canada Health Infoway. "In the next two or three years it is going to become a commodity."
Experts suggest doctors consider:
-
Average encounter time: For mental health and psychiatry, you want to be sure your system can handle 60-minute encounters.
-
Workflow adaptability: Can the AI scribe pause or consolidate recordings (for example, put a pre-visit dictation and patient visit together)?
-
Machine learning: Do you want your scribe to learn from you and use machine learning at the local level (integrating previous learnings and practices, for example)? Do you want it at the global level (from across your AI scribe network)?
-
EMR push-pull set up: Consider how good the vendor's service is and how easy it is to set up the EMR integration.
The good news is you can't go wrong even if you choose a vendor that doesn't last. Kalra said lock-in or migration isn't an issue like clinicians have reported with EMR systems. AI scribes are rapidly rolling out advances. For instance, Scribeberry provides a clinical decision-making tool supported by primary source data, featuring an autonomous robot that conducts patient interviews prior to their physician visit, and prepares a detailed note for doctors before their clinic day begins.
Mutuo Health's AutoScribe is launching a new product called AutoForm. It ingests scanned PDFs of common forms such as insurance forms, reads the questions and then uses AI scribe dialogue data and previous EMR note data to generate the answers. It inserts those as draft answers into the form, which can then be edited.
Quip is about to beta launch a feature to aid with billing. The feature will recommend billing codes relevant to the consultation content, helping to ensure doctors capture all applicable payments.
The past year has seen prices for AI scribe technology fall and experts say that will continue while quality improves.
"In terms of a year from now it is going to be so much more accurate. It's going to be so much faster—one of the issues is actually the speed of the final AI output, sometimes 45 seconds (for a 15-minute encounter now)," said Dr. Noah Crampton, CEO of Mutuo Health which makes AutoScribe.
Most Canadian EMR vendors are—or are moving to—cloud-based rather than on-premise solutions, which is needed to fully access patient data. "Once you get access to the patient data and you're able to integrate with the EMR in terms of workflow, then you're really getting AI to be your assistant, to take (more) stuff off your plate or off your MOA's plate," said Ling. "Then when you marry it with external data, like the latest guidelines or the latest research data, then you really (have) meaningful clinical decision support."
At that point, you are looking at a system that will say things to you like: "I understand you are recommending this patient for this blood pressure medication, but when I look at the demographic of similar patients within my database it looks like these are the top three medications being prescribed"—along with their efficacy rate.
"(Ultimately) doctors don't care what's AI and what's not," Ling said. "It's just another tool for them, and they just want it to work. They want it to be cheap . . . and they need support in using that tool."