Today, the headlines that dominate family medicine news are of two themes: first, the crisis in family medicine and doctor shortages; and second, the news of emerging AI technology and how it can revolutionize medicine. These topics are linked as AI technology is being increasingly promoted as a potential solution to the family medicine crisis. However, like any innovation that impacts the care we deliver, we (family physicians) must be at the forefront of how this technology is developed and implemented in practice. Whether we like it or not, AI is here and will have a profound impact on patient care. We cannot be idle bystanders, lest we repeat what has happened with EMRs, which were also promoted as revolutionary administrative tools yet have become a source of frustration and burnout amongst physicians.

So how do family physicians shape the future of AI in our own practices? The first step is to be informed and the second step is to try AI solutions for ourselves and provide early feedback on their functionality. My goal in this blog post is to provide a primer on AI in medicine, and how to audition these tools for use in your own practice.

Generally, the perception of AI has been positive with some caveats. The Ontario Medical Association (OMA) encourages the use of AI in medicine, with the OMD CEO Robert Fox stating that “AI technology has the potential to be an important part of a sustainable healthcare system in Ontario.” The OCFP emphasizes that AI is becoming an essential part of family medicine, whether to help decrease physicians’ time on administrative tasks or enhance the patient experience. Both organizations discuss AI’s potential to decrease physicians’ time on administrative tasks and paperwork to focus on patient care and experience.

It would be a lengthy task to catalog all the ways AI is being used in medicine, but AI is being used in everything from administrative support, to patient-facing tools, to generative AI, and all the way to precision medicine. In family medicine, most of the tools that exist today focus on administrative support, specifically “AI scribes,” designed to record and document the encounter. There is also ongoing development of chat bots that are meant to interact with patients to facilitate administrative tasks (such as scheduling). Importantly, our patients are using AI tools like ChatGPT to inform their care.

Currently, the landscape puts the majority of the liability on the physician. We are expected to navigate privacy concerns and security measures. Physicians are expected to also vet all the details of how/where/how long data is stored. These are challenging tasks, and it is expected that AI vendors will be vetted/regulated by medical organizations in the future, such as the OMD (similar to how EMRs get certified). However, it may be several months or years before this is implemented, so in the meantime, here is the CMPA’s framework when it comes to adopting AI technology:

 

1) Consent: Given that this is an emerging technology, obtaining consent is strongly recommended. To obtain informed consent, there should be a clear explanation of the AI tool’s purpose and functionality. A form should not be seen as a substitute for having a discussion with a patient to explain the AI tool where necessary.

 

2) Privacy: As doctors in Ontario, we should ensure that the AI tools are compliant with the Personal Health Information Protection Act (PHIPA). Physicians should evaluate whether the tool has implemented appropriate safeguards to ensure confidentiality and protection of patient data. AI companies should have conducted a formal privacy impact assessment (PIA) before collection of personal information and data, which is an analysis of how personally identifiable information is handled to ensure compliance with PHIPA, determine the privacy risks associated with information systems or activities, and evaluate ways to reduce the privacy risks.

 

3) Quality and accuracy:  AI technology is supposed to make our jobs easier. If the AI we use produces inaccurate notes that require us to heavily edit, then this is counterproductive. We need to insist on having high standards when it comes to accuracy.

 

4) Accountability:  As physicians, we are the most responsible party. When it comes to using AI, whether for charting, or for additional tasks, we must ensure that there is a system in place to ensure safety. In charting, this simply means that we must read and approve the final note. For AI systems that go beyond charting, such as providing clinical insights, physicians are expected to exercise their own clinical judgment and make the final decisions. 

Step 1: Schedule time in your calendar to test out 3-5 AI scribes on your own or with a colleague. Take notes of how they perform (do they distinguish between voices well? How accurate is the SOAP note they produce?), and what you like and dislike about them.

 

Step 2: Examine privacy and security measures–here are some key questions (from the CMPA) that should be in their documentation or should be readily answered by the company:

• What specific measures have you taken to ensure privacy and security?
• Is patient data hosted entirely on Canadian servers? (Does it leave Canadian servers for processing?)
• Do you re-purpose the data for third-party use (such as marketing)?
• Have you completed a Privacy Impact Assessment?
• How long is the data stored for? What is the process for deleting the data?
• Companies may advertise that they are compliant with PHIPA or PIPEDA. Show me the measures that you have taken to ensure compliance with these laws?

 

Step 3: Deployment into clinic:

1. Check that you have the right hardware (microphone).
2. Prepare signage and the consent process: This does not need to be an onerous task. As family physicians, we obtain consent from our patients frequently (whether we are getting records or prescribing a treatment or doing a procedure). In the case of AI scribes, we discuss with the patient that we are using an AI scribe that transcribes the audio and produces a medical note. This can be explained as “voice-to-text” software. We review the benefits (e.g. “This helps me document the encounter without having to type everything up myself and I can focus my attention on our conversation.”) We discuss potential risks (e.g. “I review the note to make sure it is accurate” “The software company indicates that they follow necessary precautions to ensure that the data is private and secure” “the audio-recording is auto-deleted”). We offer the patient the opportunity to ask questions and ensure that they are agreeable to the plan. Consider having a poster in the exam room that you can refer to if needed (see example).
3. Try out the AI!See if it makes your workflow better. Feel free to give feedback to these companies (they should love feedback)! 

 

The above steps may take a few hours. But the potential return on that time invested can be hugely impactful (up to 2 hours a day!). AI is here, likely to stay. As family doctors, we have the opportunity to shape AI as it emerges as an everyday tool in our workflows–we should seize that opportunity and make tools that help us work, not give us even more work.

by Dr. Amjed Kadhim-Saleh

July 1st, 2024

Dr. Amjed Kadhim-Saleh is a community family physician and clinic owner in Toronto. He has a deep understanding of the joys and pains of practicing family medicine, building an efficient practice, and creating workflows that work for family physicians. Dr. Kadhim-Saleh developed Pippen (https://pippen.ai/) so that family physicians could claim back their time, feel supported, and have a great day at work. Dr. Kadhim-Saleh also wants to help physicians discover the wonder of AI, which is poised to revolutionize how physicians deliver care.