Joshua Liu, MD
Co-founder & CEO at SeamlessMD
LinkedIn: Joshua Liu
X: @joshuapliu
Co-host: The Digital Patient Podcast
Musings and Insights
I was talking to CMIOs about whether access to AI co-pilots would help or hurt medical training.
→ If medical students use AI Scribes from Day 1 and never have to write their own note… how does that impact how they learn to process and think about a patient encounter?
→ If radiology residents can lean on AI co-pilots for medical imaging from Day 1… will that impact how much they learn to read imaging independently?
→ What if beyond automating clinical documentation, we implement AI automates a differential and drafts an assessment/plan… will our next generation of physicians fall quickly into automation complacency?
So some might say “trainees should learn the fundamentals, and then later get access to the AI co-pilots”.
Of course, I can also see many benefits to giving early access to AI tools to trainees:
→ Self-motivated learners will actually use AI to learn faster. E.g. once there is a ChatGPT interface next to clinical documentation, students could make queries to ask: “What’s the evidence for X suggested treatment?”. It’ll lower the friction for self-directed learning.
→ It’ll free up time for more learning. If trainees don’t need to spend all day doing documentation or scut work they often are tasked to do, it frees up more time for hands-on learning – e.g. being in the operating room more for surgical trainees.
The general consensus from the CMIOs I spoke to is that the low-hanging fruit for AI are the administrative pieces – with AI scribes being the closest thing to “clinical” that clinicians are comfortable automating… for now.
They believe more safety and efficacy studies need to be done with using AI for those next steps – e.g. drafting an assessment and plan based on the clinical data. And until those results are good, then those items won’t be in the hands of trainees anytime soon.
But personally I think we’ll see draft assessments and plans in the hands of trainees faster than we think (it might already be happening).
My take is that we can’t (and shouldn’t) shield medical trainees from the AI that we will increasingly embed in the clinical workflow. Instead we should be thoughtful about how to leverage AI to accelerate clinical training and self-directed learning.
For example: most clinical references (e.g. UpToDate, ClinicalKey) now have ChatGPT like interfaces to make queries. Assuming well tested and very low-risk of hallucination, these interfaces should be embedded into the EHR and allow trainees and clinicians to ask questions and get answers immediately at the point of care.
What do you think?
How do we implement AI in healthcare in such a way that it helps and does not harm clinical training?
I was talking to CMIOs about whether access to AI co-pilots would help or hurt medical training.
→ If medical students use AI Scribes from Day 1 and never have to write their own note… how does that impact how they learn to process and think about a patient encounter?
→ If… pic.twitter.com/d4rUiCpYtH
— Joshua Liu (@joshuapliu) July 16, 2024
The Digital Patient
The Digital Patient takes an “edu-taining” approach to all things digital patient care. On this show hosts Dr. Joshua Liu, and Alan Sardana talk with healthcare, technology, and innovation leaders about the latest advancements in digital health, trends in digital transformation, and strategies for optimizing the patient experience.