A Little More Meat on The Scribe Bone
Scribe vs. Scribe vs. No-Scribe
It’s all scribes all the time, now – the magical tool that will “save” clinicians from the most burdensome chores of their profession.
It has been mostly assumed that ambient scribes are roughly in the ballpark as human scribes, and most folks like their human scribes – so, in the same vein as the 510(k) program, the ambient scribes must also be good?
In the pre-Thanksgiving NEJM AI, there are two prospective scribe trials. The first, by Lukac et al, randomizes clinicians in the outpatient arena to a control arm, a DAX arm, and a Nabla arm. The basic outcomes are fairly consistent with the previously reported observational data – generally less time in notes, a Better Life – but there’s also some fun tidbits in the supplement.
For example, how often are clinicians perceiving problematic notes? Fairly often:
Or, a small accounting of feedback from clinicians who never used the ambient scribe despite being assigned:
The second study is Ashfar et al, a 24-week stepped-wedge rollout of Abridge. The same general findings applied – less time in notes, Better Life:
Grossly, these trials are very similar. However, one notable difference was that Abridge integrated directly into Epic, whereas DAX and Nabla required copy-and-paste for notes. This likely contributed to the incrementally better results observed in Ashfar et al, where clinician uptake was stronger and the measures of workload reduction manifested more reliably.
There’s also an editorial by Kim et al, “AI Scribes Are Not Productivity Tools (Yet)”. The “(Yet)” part is likely the key, here. What will scribe use look like in a year? When clinicians fully adapt their practice to the nature of ambient AI? When integration improves? When customization becomes readily accessible to those without specialized prompting skill? Also notable: the outpatient arena is a challenging setting into which “productivity” can be injected. Emergency departments, with their ready supply of work, are likely a better environment in which to evaluate “productivity”.