As we approach the end of 2024, First Opinion is publishing a series of essays on the state of AI in medicine and biopharma.
Get a group of primary care physicians together, and there’s a pretty good chance they will start talking about the potential of AI scribes to reduce documentation burden and improve the clinician-patient office interaction. These programs use ambient listening to record clinician-patient interactions and generative AI to filter extraneous conversation and compose cogent progress notes. There is an ever-growing list of companies and start-ups actively marketing these tools to medical practices.
While AI scribes score high on the “wow” scale, I worry that they are 1) not solving the exam room problems that are most tedious and time-consuming, 2) like any type of scribe, reinforcing our electronic health record-exacerbated obsession with elaborate chart notes, and 3) creating a new task of reviewing and editing imperfect AI-generated text.
Writing chart notes has been a bane of clinicians’ existence for some time, especially in the electronic documentation era. While some has been foisted upon us, elaborate chart notes are in large part a problem we have created for ourselves. “Note bloat,” as it is sometimes sardonically termed, was exacerbated by Medicare’s strict billing documentation guidelines established in 1997 as part of the Balanced Budget Act. In 2021, these rules were simplified such that evaluation and management billing codes are now determined only by the complexity of medical decision-making or time spent during a visit. Curiously, in spite of this rule reprieve, most clinicians have not backed away from excessive documentation, perhaps due to fear of litigation or a spurious sense imbued in medical culture that note comprehensiveness equals clinical excellence. Further, there seems to be little concern over the under-recognized burden of reading these notes. At this point, despite the illogic, this appears unlikely to change, so it makes sense for AI scribes to provide a life raft for those clinicians who need one.
EHR note-writing may also be somewhat unfairly blamed for lack of eye contact and attentive engagement with patients in the exam room. While there is undeniable truth to this, I personally recall handwriting in paper charts competing with active listening as well. Of course, quiet note-taking in a manila folder is far less distracting than clackety typing on an obtrusive computer terminal.
I have worked to become a more pragmatic and efficient progress note documenter, yet there are many other unavoidable aspects of the EHR-assisted patient encounter that I hope new technology can eventually help facilitate a solution for. Patients often require numerous laboratory and imaging tests, consultation requests, in-office procedures like vaccination, all of which require manual order entry by the provider. I also find myself spending a lot of visit time looking for things in the electronic chart — test results, consultants’ notes, even my last visit note. When AI scribes or other technologies have mastered these tasks, which many companies are actively pursuing, I will be first in line to sign on.
So AI scribes will certainly reduce note-writing burden, which, whatever the true underlying cause, is a significant source of after-hours task work and clinician burnout. But they also create a new task: reviewing the AI-generated text. A large-scale AI scribe implementation at the Permanente Medical Group in 2023 improved care experience of both clinicians and patients overall, yet there were inconsistencies requiring clinician editing and some instances of AI hallucination — that is, making things up. Most clinicians in this study did not find this to be a major impediment. Further, elaborate chart notes, whether typed by clinicians or scribes and regardless of accuracy, will continue to be a time drain to read and review.
AI scribes are an extraordinary technology that has great potential to reduce clinicians’ documentation burden and lessen distraction in the exam room. It behooves us, though, to honestly examine the root causes of the behaviors that AI scribes aspire to improve. Dialogue around this issue can help health systems advocate for standards that balance detail and efficiency. In the case of chart documentation, clinicians have in some ways been their own worst enemy.
Jeffrey Millstein is a primary care physician and regional medical director for Penn Primary Care.