This is a new column about a layer of healthcare that doesn’t get written about much, even though it shapes the care patients receive every day.
I’m calling it the practice layer.
The place where one clinician watches another work, signs off on the work, and learns or fails to learn something from the watching. In healthcare, the formal name for this is supervision. Most of the public conversation about it is bureaucratic: how to staff it, how to bill for it, how to stay compliant with whichever state’s rules apply.
But that framing misses what the supervisory layer is for. Done well, it’s where clinicians grow over time and where patient care gets better. Done poorly, it’s where small drift becomes larger patterns that show up downstream in patient outcomes and clinician burnout. These kinds of quality failures look random until you trace them back to a feedback loop that never happened.
This is a corner of healthcare most people will never read about. The language inside it is jargon-heavy, and supervision sounds like an HR topic until you understand what it shapes. But it matters more every year.
The patient population in the US keeps growing. The supply of physicians isn’t keeping pace. And a lot of the care being delivered, especially in primary care, mental health, urgent care, and women’s health, is happening through nurse practitioners and physician assistants working under physician supervision. The way that supervisory relationship is designed shapes whether the people doing more of the country’s care are getting better at the work or just getting through the day.
If you work in clinical quality, operations, or leadership inside a health system or medical group, this column is written for you. It won’t be the five-step plan version. The work is too hard for that. The goal is to name what’s broken in this layer of healthcare, walk through what the research shows about how to improve it, and translate both into something you can use inside your own organization.
For about a decade I’ve worked in applied science and research, mostly in healthcare and AI: studying the systems that shape what people do, and building products inside them. The supervisory layer is newer territory for me.
At Zivian, I’m working on a product that runs chart reviews for organizations that employ nurse practitioners and physician assistants across many states at once. Coming into a new space from outside means I notice things. Some of which are things that the people who’ve worked inside this corner for years have stopped seeing.
The first challenge jumped out almost immediately: the data exists, the systems collect it, but almost no one is producing learning from it. This isn’t a failure of any individual. It’s a system-level gap that needs to be addressed accordingly.
Where the gap actually is
Most quality teams we work with are sitting on more data about their own clinicians than they could ever review. Chart review comments, exception reports, peer feedback, audit findings, attestations, signature logs.
Some of it lives in EHRs. Some lives in spreadsheets. Some lives in the heads of supervisors who haven’t written it down. Almost none of it gets aggregated into something a clinician could use to understand their own practice patterns, or that an organization could use to spread what its highest performers are doing.
This isn’t because feedback doesn’t work. Healthcare has a large research literature on whether clinical feedback improves practice. The Cochrane Review of Audit and Feedback has tracked it since the 1990s and been updated several times, most recently in 2025. That update pulled together 292 controlled trials.
On average, feedback produced about a six-percentage-point improvement in clinician compliance with recommended practice. What the research is clearest about, though, is the conditions that make the effect bigger: feedback works better when baseline performance is low, when it comes from a supervisor or peer rather than an outside body, when it happens more than once, and when it includes a specific target and an actual plan to act on.
None of that is a secret. However, meeting those conditions takes infrastructure most organizations were never set up to build. The resulting cost impacts both healthcare operations and patient care.
The supervisory layer is where care gets shaped. A nurse practitioner who isn’t getting useful feedback isn’t growing in the directions their patients need them to grow. A physician supervising a dozen advanced practice clinicians across three states, with no infrastructure for noticing patterns across that group, can’t really mentor at scale. They can only sign things. The patient who walks in with the kind of presentation those three clinicians have been mishandling is unlikely to be the one who benefits from anyone catching it.
That’s the gap I care about. From the engineering side, it’s a data problem with a human problem hiding inside it. From the practice side, it’s a human problem with a data problem hiding inside it. Each discipline keeps trying to fix it from its own angle and missing the other half.
I came to this corner of healthcare because the questions inside it pull from more disciplines than any single field owns. Behavioral science is one lens among these.
I’ll cite research as I go. Most of what I’ll write about has been studied. The gap is between what the evidence says and what organizations have built.
This column will come back to certain questions: Where data already exists and learning doesn’t. Why some teams improve and others stay flat. What high-performing supervising physicians do differently, and how organizations might spread that without making it punitive. What patients downstream of the supervisory layer get and don’t get because of how that layer is designed. And what’s owed to the people carrying the cognitive load of all of it.
Next month, a piece on positive deviance: what learning from your own high performers looks like inside a single quality team, and why moving from spotting one to spreading what they do is harder than it sounds.
— Eden
