You can't nudge your way out of a clinical judgment problem
By Craig Joseph, MD ·
A little over two years ago, I got enthusiastic about a chair. The occasion was a randomized trial at Parkland Memorial Hospital, which found that a chair placed within three feet of the bed and turned toward the patient did something no memo has ever reliably done: it changed how physicians behaved. When the chair was already in position, 63% of doctors sat down with their patients. When they had to go dig one out of the closet, 8% did. Patient satisfaction followed the seating. I wrote about it admiringly at the time, and I still admire it.
A trial published in JAMA this spring, though, has me dwelling on what the chair cannot do.
For a decade now, the dominant strategy for changing clinician behavior has been to change everything around the clinician instead. Pre-checked boxes in the order set. Best-practice alerts. The chair. Behavioral economists file all of it under choice architecture, and the logic they inherited from Richard Thaler and Cass Sunstein is genuinely sound and well evidenced: stop lecturing the human and redesign the environment the human operates in. It is cheap, it scales, and it frequently works.
But look at the assumption tucked underneath it. The clinician is treated as a constant, a person whose surroundings we can engineer but whose judgment we quietly assume we cannot. We have gotten very good at improving defaults. The harder question, the one we have mostly declined to ask, is whether we have put anything like the same effort into helping physicians make better decisions.
What if the problem was never knowledge?
The JAMA trial went at exactly that question. A team at the University of Pittsburgh took 800 board-certified emergency physicians working at non-trauma centers and randomized them to either the usual continuing education or a theory-driven video game called Night Shift. The doctors played on iPads: two hours to start, then 20 minutes every quarter. The game runs on case-based narratives and pattern-recognition puzzles under time pressure, all built to retune the split-second judgment a physician uses when deciding whether an injured older adult belongs at a trauma center.
Undertriage of severely injured patients 65 and older dropped from 57% in the control arm to 49% in the intervention arm. That is a model-adjusted seven percentage points, and it came with no rise in overtriage. It happened across 1,147 hospitals with no EHR build, no new committee, and no army of consultants. Put concretely, about one in 13 severely injured older adults who would have stayed put under usual care instead reached a trauma center, where the literature puts the mortality reduction somewhere between 10% and 25%.
Here is the number that should stop a chief medical officer cold: 94% of these physicians had completed Advanced Trauma Life Support, the criterion standard for trauma care. They had the training. They knew the guidelines. And more than half of them still undertriaged. Whatever this is, it is not a knowledge deficit, and we have known as much for a generation. A 1999 JAMA meta-analysis found that the didactic continuing medical education we still pour billions into, the lecture halls and the conference ballrooms, produces no meaningful change in what physicians actually do. The interactive kind, the kind that makes clinicians practice, does.
Nudges manage clinicians. Training invests in them.
So why would a game beat the classroom? Start with the base rate. The median physician in this trial saw two severely injured older adults in a year. Two. Stretch that across a 30-year career and you get something like 60 repetitions of a high-stakes pattern-recognition task. We would never ask an athlete, a pilot, or a musician to stay sharp on those terms, yet physicians are the one class of elite performer we expect to maintain rare-event skills entirely through live performance on real patients. The game did what real practice almost never does: it compressed a career’s worth of uncommon cases into a single afternoon and handed back immediate feedback on each one. None of this should be a surprise. A meta-analysis of simulation-based education with deliberate practice already found it substantially better than traditional clinical education for building skill.
The authors frame the contrast cleanly. A serious game treats the clinician as a partner in sharpening their own judgment. A nudge works on behavior from the outside and leaves the decision-maker’s underlying competence exactly where it found it.
I spent years on the other side of this problem, as a chief medical information officer. Anyone who has tried to configure drug-drug interaction checking knows the trap. The interaction alert that might genuinely save a brand-new attending is pure noise to the hospitalist twenty years into the work. The warning that belongs in front of a primary care physician is an insult to the cardiologist who co-prescribes those two drugs on purpose, every week, with her eyes wide open. The system cannot tell them apart, because choice architecture is often forced to treat every prescriber as the same interchangeable decision point.
Across 34 studies, override rates for these alerts land between 55% and 98%. And nudges tend to shrink the moment they leave the well-tended garden of the academic study. An analysis of all 126 trials run by the two largest US nudge units, a sample covering 23 million people, pegged the real-world effect at 1.4 percentage points against the 8.7 points that show up in journals. Still worth having, just roughly six times smaller than the literature might lead you to expect.
Underneath the operational question sits a quieter one about what kind of professionals we think we are employing. A nudge says, in effect: we have arranged your surroundings so that your unexamined reflexes land on the right answer. Training says something different: we are investing in your judgment because you are the instrument, and the instrument is worth maintaining. One of those is how you treat a decision point. The other is how you treat a physician. And judgment travels in a way furniture never will. It rides along in the clinician’s head, shows up at every hospital where she picks up a shift, and never once generates a maintenance ticket. The chair only helps in the rooms that happen to have one.
Honesty demands a caveat, and it is an important one. The trial’s composite clinical endpoint, 30-day mortality or readmission, did not improve. The study was not powered to move it, and the survival case for getting trauma patients to trauma centers rests on decades of separate evidence, so anyone waving this trial around as proof of a mortality benefit is out ahead of the data. What it does establish is narrower and still remarkable: physician judgment, the thing we have long filed under fixed, turns out to be trainable at national scale for the cost of an iPad and three hours a year.
What this means for your portfolio
None of this is a case against nudges. It is a case against running a one-instrument orchestra. Choice architecture shines when the right answer is already known and the only obstacle is friction or forgetting, which is precisely the chair, the default, the pre-checked box. Judgment training shines somewhere else entirely: when the decision is rare, time-pressured, and pattern-based, and when feedback is far too sparse for ordinary experience to teach it. Think trauma triage, pediatric sepsis in a community emergency department, stroke recognition.
So what can a chief medical officer or chief health information officer actually do on Monday? Start with an audit of the alert inventory, asking one question of each entry: is this compensating for a friction problem or a calibration problem? An alert that fires constantly and gets overridden constantly is a strong sign you are using furniture to solve a judgment problem. Then pick a single low-base-rate, high-stakes decision in your system and pilot judgment training against it, dosed like the therapeutic it is. In this trial the effect was strongest inside the first 30 days after gameplay and faded from there, which means the quarterly boosters were not filler; they were doing real work. Training is not an event. It has a half-life, and you have to redose it.
So keep placing the chair. It works, it costs almost nothing, and there is no reason to stop. Just stay honest about what it did and what it never could. The chair changed where the doctor sat. It was never going to change what the doctor saw.