One of the markers of quality, such as it’s commonly understood in medicine, is reducing variability. This is why we have best practices that are often not really the “best” in truth but are, essentially, a reasonable consensus approach. The primary intention is of course to raise the floor (and not to reduce the ceiling), because the reality is that variability and noise in general are a real problem in all human endeavors, and medicine is no exception.
Over the years since I began working in radiology, one of the ways that’s manifested has been in the promotion of structured reporting: making radiology reports look the same, organized in the same way, and, in many cases, using the same language. The drive there is obvious and intuitive. Even as a radiologist, I find it easier to find what I’m looking for in a prior report if all prior reports look the same. When I grade resident work from overnight, I generally find it simpler to take it in a structured report at a glance than to read a long narrative.
Now, I’ll be the first to admit that the benefit requires a reasonable degree of structure (and not dozens of entry fields), and even then, part of the benefit is because long unstructured narratives tend to be, well, long, and if somebody did a great job dictating concisely, the delta would be not quite so high. The right person can make terrible reports in any format.
I generally agree with the goal of tamping down outliers: some reports are so terse they look lazy and incomplete; others are so long as to be functionally unreadable. Balance in all things, as they say.
The great irony is that our push for standardization, in essence, also functions to remove personality from the radiology report. This isn’t a bad thing per se. I don’t necessarily need to see that a report was written by Todd or Kevin instead of Sharon based on how long it is, how many words they use, or the weird quirks of language that we all accumulate, refine, and cling to as we dictate our thousands upon thousands of cases.
But a standard report with concise language also sands down the edges that make human work distinctly human. It makes it easier for us to become interchangeable cogs in a reporting machine, where a CT head report is a CT head report is a CT head report. And in the nascent age of AI drafting, it also means that, in many cases, it will likely not be possible to know at a glance if the work was really done by a human or done by a machine on the human’s behalf. And to be clear, many will almost certainly prefer the generic machined product (because, if we’re being honest, some idiosyncrasies are just annoying).
That will be a very weird thing indeed. Already, I find myself irritated by so much of the writing I see online, so much of the inbound I receive, because it’s clearly generated by AI. We’ve divorced the written word from the effort it takes to generate it, and that’s a brave new world, even in these early days. I wonder how that will feel as the wave continues to climb and ripple through all of medicine.