A tool opened once
At Johns Hopkins, a clinician who opens an AI scribe for just one visit during six months still counts as an adopter in the study. Not using it for subsequent visits would not change that first counter. This is why I read Tej D. Azad and colleagues’ study of 2901 clinicians through two separate denominators. One counts clinicians who touched the tool; the other counts eligible visits in which they used it. The useful result is the separation itself, rather than a single headline adoption percentage. It exposes the distance between giving a hospital access and having a tool become part of ordinary work.[1]
In community internal medicine, 130 of 195 clinicians tried the tool at least once; in academic general surgery, 12 of 77 did so. Those counts show that different practice environments did not engage with the same software equally. They do not measure whether surgeons benefited less. Patient consent, the shape of a consultation, training and established habits can all stand between availability and even an initial trial. The cohort covers eligible ambulatory encounters, excluding emergency care, inpatient care and procedure-only visits. Extending these proportions to all hospital activity would expand the population beyond the one studied. The workflow behind the denominator matters alongside the result.[1]
Use that persists and use that stops
The second counter captures a different movement. By the end of the period, median use reached approximately 80–90 percent of eligible monthly visits in some community physician and advanced practice provider groups. The researchers also distinguish slow sustained uptake, early use followed by discontinuation, transient trials and late adoption. These trajectories complicate a success story ending at first contact: someone may try and leave, while a late starter may become a regular user. A six-month count of everyone who ever tried the tool combines those movements. An implementation manager tracking only initial trials could miss both sustained integration and subsequent disengagement.[1]
I am also wary of reading these trajectories as fixed clinician personalities. The researchers standardize monthly use and group temporal patterns; the resulting categories are not validated behavioral types. Higher use could reflect consultations that are easier to document with ambient capture. A competing explanation is variation in informal training or peer support. The study does not use a causal design that separates those mechanisms. Access was voluntary, there was no usage-linked incentive, and individualized on-site support was not provided. Knowing those conditions is more useful than attributing low uptake directly to unwillingness. A declining usage curve cannot by itself explain why somebody stopped.[1]
The path to the patient’s note
The accuracy of the note produced in front of a patient sits outside both counters. Usage logs establish that the tool was activated; they do not establish that its text was safe, easy to edit or time-saving. The study explicitly says it did not measure patient benefit, documentation quality or effectiveness. High use might be consistent with perceived benefit, but incorporation into a workflow and improved clinical outcomes are different measurements. I therefore draw no patient-safety ranking here. The narrower contribution is a denominator for considering which clinicians and encounters could actually receive benefits established in other studies, rather than assuming that universal access makes those benefits universal.[1]
I would find the next comparison more informative if it tracked note-editing burden and patient experience alongside the same clinician’s use over time. That is my proposed design, not a test completed in this six-month study. Interviews with people who stopped could help distinguish a missed opportunity from a reasonable choice for their work. Before transporting one system’s results to another hospital, the definition of an eligible encounter needs to be rebuilt there. Rather than declare the application with the most users the winner, I want to know which task it makes easier in which consultation. Understanding the burden on clinicians and patients remains a separate research question after access is switched on.[1]