Once a year, many health systems send physicians a survey. A few months later, a single figure appears in a board pack: the share of doctors reporting at least one symptom of burnout. If it has gone down, there is quiet relief. If it has gone up, there is a wellness initiative.
That number is worth tracking. It is also the least actionable metric an organisation can collect about its doctors. It tells leaders how much pain there is. It says almost nothing about where it comes from, or what to change on Monday.
The headline is improving, which is a risk in itself
In the United States, the American Medical Association’s national data show burnout falling from its pandemic peak. In 2024, 43.2 per cent of physicians reported at least one symptom of burnout, based on the AMA’s Organizational Biopsy survey of physicians across participating health systems. That is better than the worst years. It still means more than four in ten doctors.
A falling headline creates a temptation to declare victory and redirect attention. But a burnout rate is a lagging indicator. By the time it rises, physicians have already cut their hours, started looking for other jobs or left. Leaders who manage only the lagging number are always responding to damage already done.
There is a subtler problem too. A national or system-wide rate can fall because conditions improved, or because the most exhausted doctors have already left and are no longer in the survey. The second kind of improvement looks identical on a chart and means something very different.
Why it matters financially
The business case is well documented. A 2019 study in the Annals of Internal Medicine by Shasha Han, Tait Shanafelt, Christine Sinsky and colleagues estimated that about US$4.6 billion a year in US costs from physician turnover and reduced clinical hours is attributable to burnout. At the organisational level, that worked out to roughly US$7,600 per employed physician per year.
That estimate covers only turnover and reduced hours. It does not include effects on errors, patient experience or the colleagues left to absorb the extra workload.
Scale it to a single organisation. A health system employing 500 physicians would, on that average, be carrying around US$3.8 million a year in burnout-related turnover and lost hours. The figure will differ from place to place, but it is the right order of magnitude to set against the cost of a scribe programme, a documentation clean-up or an extra nurse in a busy clinic. Most wellness budgets are a small fraction of it, which says something about how seriously the problem is being priced.
The metrics that actually point somewhere
If the burnout rate is the thermometer, leaders need metrics that work more like a diagnosis. Some of the most useful are already sitting in systems that organisations own.
The first is time on the electronic health record, especially outside clinic hours. A widely cited 2016 time-and-motion study by Sinsky and colleagues, also in the Annals of Internal Medicine, found that in ambulatory practice, for every hour physicians spent with patients, they spent nearly two more on EHR and desk work. Participants also reported one to two hours of work each night at home. EHR vendors now make this kind of activity data available. Tracking after-hours EHR time by department is one of the clearest signals of workload that has overflowed the working day.
Do the sum. An hour a night, five nights a week, across roughly 46 working weeks is more than 200 hours a year, the equivalent of five or six extra working weeks done unpaid at the kitchen table. When a department’s after-hours time creeps up by twenty minutes a day, that is not noise. It is a week of someone’s life.
The second is inbox volume per physician. Patient messages, results and administrative requests are a major, often unmeasured part of the job, and their growth should be tracked like any other workload. The third is intent to leave and intent to reduce hours. These are closer to the decisions that cost money than burnout symptoms are, and they arrive earlier than resignation letters.
Fourth, unit-level results rather than the system average. Burnout clusters. A system average of 40 per cent can hide departments at 20 and departments at 60, and the high ones are where action pays off. Fifth, control over schedule and work: whether physicians feel they have a say in how their work is organised is both a driver of burnout and something managers can change. And sixth, leadership behaviour scores. How physicians rate their immediate leader is a measurable, improvable factor, and one of the few that sits directly within a manager’s authority.
How a driver metric changes a decision
Consider how the conversation differs. A board is told that burnout has risen from 41 to 45 per cent. The natural response is a programme: a resilience course, a wellbeing app, perhaps a retreat. Nobody knows which department it should target or what it is meant to change.
Now the same board is told that after-hours EHR time in two departments has risen sharply since a new documentation template went live, that inbox volume per physician in primary care has climbed while support staff numbers have not, and that intent to leave in those same units is well above the system average. The response writes itself: review the template, fix the message routing, add support staff where the volume is. Each action has an owner and a number to move. The same pain, measured differently, produces a completely different decision.
The uncomfortable incentive
Why do organisations focus on the headline rate and wellness programmes rather than these drivers? Partly because the drivers implicate the organisation. After-hours EHR time points at staffing, documentation requirements and IT design. Inbox volume points at how patient access is organised. Low schedule control points at management.
A resilience workshop, by contrast, locates the problem in the individual physician. It is cheaper, easier to announce and asks nothing of the institution’s workflows. It is also, for most doctors, mildly insulting.
None of this means individual support is worthless. Access to confidential counselling and peer support matters, particularly after adverse events. The problem is sequence and proportion. When individual support is the whole response, the message physicians hear is that the job is fine and they are the ones who need fixing. Most doctors can tell the difference between an organisation that is helping them cope and one that is asking them to cope with more.
The strongest objection
There is a serious argument against all this measurement. Physicians already feel watched. Turning EHR logs and inbox counts into dashboards can feel like surveillance, and a metric used to judge individuals will be gamed or resented. Some doctors will reasonably ask whether after-hours data will be used to support them or to question their efficiency.
That concern partly holds, and it should shape how the data are used. Driver metrics work when they are reported at team and department level, when they are used to diagnose systems rather than rank people, and when physicians can see the numbers and what was done about them. The same log data can support a case for more help or a case for performance management. Which one it becomes is a governance choice, and it should be made explicitly, before the dashboard is built.
A second objection is that the data are imperfect. EHR activity logs count time a record is open, not time spent thinking, and they vary between vendors. True, but imperfect operational data tracked consistently over time still show direction, and direction is what boards need.
The practical answer to both objections is to start small and in the open. Pick one or two departments, agree with their physicians which measures will be tracked and how they will be used, publish the results to the people measured before anyone else sees them, and act visibly on what they show. A pilot that produces one fixed template or one extra member of support staff will do more to build trust in measurement than any policy document about data governance.
Second-order effects
Measuring drivers changes more than the burnout rate. Once after-hours EHR time is visible, it becomes an input to IT decisions: a new template or alert can be judged partly by what it does to documentation time. Once inbox volume is visible, it becomes part of how patient access is designed and how support staff are allocated. The burnout conversation stops being a human resources topic and becomes an operational one, which is where it belonged all along.
There is a cultural effect as well. Physicians who see an organisation measure and act on the things that make their days harder tend to trust its surveys more, which improves the quality of the next round of data.
What this means for the people who can act
For hospital chief executives and chief medical officers, the practical step is to change what goes to the board: two or three driver metrics, by department, alongside the headline rate, each with a named executive responsible. For chief information officers, after-hours EHR time and inbox volume should be treated as service quality measures for their own work, not just clinicians’ problems.
For payers and regulators, documentation requirements are a major source of the clerical load behind these numbers. Every new reporting or prior authorisation demand has a cost measured in physician evenings, and it is worth asking for that cost to be estimated before the requirement is imposed. For founders building ambient documentation and inbox tools, the metrics above are the ones customers should hold them to. A product that claims to reduce burnout should be able to show falling after-hours EHR time in the departments that use it, not just satisfaction scores.
What to watch in the next two years
Watch whether the AMA’s next national figures continue the decline, and whether the improvement holds across specialties or is concentrated in a few. Watch the evidence on ambient AI scribes, which health systems are now deploying widely: the question is whether independent evaluations show reductions in after-hours EHR time, not just in how physicians feel about documentation. Watch whether health systems begin reporting driver metrics publicly, in the way some now report staffing or safety data. And watch the inbox. As patient messaging grows, organisations that do not measure and staff for it will find it quietly undoing whatever gains they make elsewhere.
Making metrics useful
A few principles separate measurement that changes things from measurement that decorates reports. Measure where you can act, at department and team level, not only system level. Pair survey data with operational data, so that self-reported burnout sits next to EHR logs and inbox counts. Report drivers to the board, not just the outcome; a board that sees after-hours EHR time rising will ask better questions than one that sees a single percentage. Close the loop by telling physicians what was learned and what will change, because surveys without visible action produce survey fatigue and cynicism. And assign ownership, so that each metric has an executive responsible for moving it.
The takeaway
The burnout rate tells a health system how its doctors feel. It does not tell it why, or what to fix. Organisations that want to keep their physicians should spend less time admiring or lamenting that one number and more time tracking the hours, messages and decisions that produce it.
Which burnout-related metric does your organisation track that has actually led to change? Share it in the comments.
