Most responses to clinician burnout start with the person. Resilience training, wellness curricula, an employee assistance line, a mindfulness session at the start of a meeting. Each of these can help, and none of them should be withdrawn.
But they share an assumption: that the problem lives mainly in the people doing the work. In many practices, a large part of it lives in the work itself. In a message queue that routes every question to the most expensive person on the team. In a schedule template that has not been revisited since the patient panel doubled. In a handoff that nobody owns, so everyone checks it twice. If you run a practice, a residency clinic or a department, that is the part you control.
When the design of the work is poor, people compensate. They stay late, they absorb tasks that belong elsewhere, they work around a broken process because the patient in front of them cannot wait. For a while, the outcomes hold. That is exactly what makes the problem hard to see. A stable outcome does not always mean a stable system. Sometimes it means people are quietly paying the difference.
So the useful question for a healthcare leader is not only "how resilient is my team?" It is this:
Where is your system asking people to compensate for poor design?
What the evidence does and does not say
The research on burnout is large, and most of it is observational. That matters, because it means the honest language is associated with, not caused by. Stated carefully, the evidence is still hard to ignore.
After-hours electronic health record work is one example. In a national survey of 9,653 upper-year family medicine residents, 32.3% reported averaging three hours or more per night on the ambulatory record after usual clinic hours. After adjusting for resident characteristics, that level of after-hours work was associated with higher odds of burnout (odds ratio 1.61) and lower odds of professional satisfaction (odds ratio 0.61).¹ The study was cross-sectional, so it cannot say the record work produced the burnout. It is a strong reason to look.
Control over the work is another. In a multi-institution survey of U.S. physicians, poor control over patient load, team composition, clinical schedule, workload, and the areas a physician is held accountable for were each independently associated with burnout.² The sample was drawn from practices with more than 100 physicians, so it may not represent smaller ones. Again cross-sectional, and again a reason to look rather than a proof.
And the associations reach beyond the individual. A meta-analysis of 170 observational studies of 239,246 physicians found that, in the 25 studies measuring it (32,271 physicians), burnout was associated with roughly three times the odds of turnover intention, with wide variation between studies.³ For an organization, that is a workforce and budget question, not only a wellbeing one.
None of this means all burnout is organizational. It is multifactorial, and some of it belongs to life outside work. The point is narrower: some of the drivers are modifiable by the organization, and those are the ones a leader can act on.
Where should a leader look first?
The literature describes the organizational drivers in several overlapping ways. They reduce to five practical domains, each with one question worth asking in your own setting:
- Workload and resources, or Capacity. Can available resources reasonably meet expected demand?
- Autonomy, or Control. Do people have appropriate influence over the work for which they are accountable?
- Task flow and the record, or Workflow. Does work reach the right person at the right time with minimal avoidable friction?
- Community, roles and teamwork, or Team. Are roles, responsibilities, delegation and handoffs intentionally designed?
- Organizational support and supervision, or Leadership. Do leaders identify and remove barriers to effective work?
Two cautions travel with this list. First, it is a way of deciding where to look, not a validated instrument. The framework tells you where to look. It does not diagnose the organization for you. Second, the domains should be judged separately and never added into a single resilience score. A total hides exactly the variation that tells you where to act, and it invites comparisons the method cannot support. When different roles rate the same domain differently, that disagreement is itself useful information.
Why do burnout interventions miss?
The most common failure in this work is jumping from symptom to solution. Portal messages are taking longer to close, so the answer must be more staff. Turnover is rising, so the answer must be a retention bonus. Sometimes that is right. Often it treats the wrong domain.
A better sequence starts with observation and holds off on cause. See what is visible: messages aging, after-hours record time rising, continuity falling, tasks escalated inconsistently. Diagnose which domain appears most involved, what evidence would confirm it, and what competing explanation you would need to rule out. Only then redesign one modifiable condition: standardize message routing, clarify escalation criteria, protect review time, remove a duplicate review step, define who owns a handoff.
This sequence does not replace an organization's improvement method. It sits inside PDSA, the Model for Improvement or Lean, and its only job is to make workforce design visible within quality improvement work that is already happening. It is also the approach Energized Vision takes in operational improvement work with healthcare organizations.
What this looks like in a clinic
Desert Valley Family Medicine Residency is an illustrative teaching case built for the STFM session. The clinic and every number below are fictional. They are case parameters, not benchmarks.
Desert Valley is a community-based academic practice: six core faculty, eighteen residents, three advanced practice providers, four nurses, nine medical assistants, five front-office staff and a practice manager. Over eighteen months, demand grew. Portal messages, refill requests, same-day requests and visit volume all rose. Clinical support staffing did not, and panel sizes and workflows were never systematically rebalanced.
At the quarterly operations review, nothing looks alarming. Visit completion moved from 94% to 93%. The quality composite moved from 82% to 81%. Patient experience slipped from the 86th to the 82nd percentile. Underneath, the system is moving faster: third-next-available appointment went from 7 to 11 days, staff turnover from 11% to 19%, and resident continuity from 58% to 48%. The outcome measures are holding while the measures that describe the system are not.
The obvious diagnosis is a staffing problem, and it may be part of one. But follow a single portal message: "My home blood pressure readings are still high. Should I change my medication?" It travels from the portal to the front office, to a medical assistant, to a resident, to faculty, back to the same resident, and finally to the patient. Seven steps, and the resident handles it twice. Some of those steps are required, because supervision in a residency clinic is clinically and educationally necessary. Not all of them are.
Two numbers separate a capacity problem from a workflow problem. If each message is handled about once and most are resolved at first touch, the work is arriving in greater volume, and more hands or fewer messages is the answer. If messages are handled repeatedly and few are resolved at first touch, the same work is being done more than once, and adding staff adds people to the loop. In the case, messages averaged 3.8 touches and 34% were resolved at first touch. That points first to Workflow and Team, with Capacity still under genuine pressure and Control as the lever that makes a redesign possible.
So the first test is not a hiring request. In one pod, for thirty days, messages are routed by type: administrative work to the front office, protocol-based clinical work to a medical assistant and nurse pathway within approved scope, clinical judgment to the resident or physician, and urgent issues along a defined escalation pathway. Explicit escalation criteria, a ten-minute huddle and a weekly review of misrouted messages hold it together.
If the hypothesis is right, touches per message fall and first-touch resolution rises. If physician inbox time falls while nursing workload rises by the same amount, the test has moved the burden rather than removed it, and it does not scale. A routing result that holds does not replace the staffing case. It makes that case stronger, because it shows the existing capacity is being used well.
Leave with one test, not five ideas
The instinct after a diagnosis is to fix everything at once. Resist it. Do not leave with five ideas. Leave with one test.
Keep the first test small: one pod, one clinician, one workflow, one message category, for a few weeks. Small is not timid. Small is what makes the result interpretable and a failure survivable.
Measure it in four ways: a workforce measure, an operational or process measure, a care or quality measure, and a balancing measure. The fourth is not optional. Reducing physician burden by moving it to nursing is burden transfer, not improvement. The balancing measure is what catches it, and in settings where support staff have the least slack, it catches it first.
Then scale only when the evidence justifies it. A mixed result is information, not permission. And once something works, decide who owns it, which measure stays on a dashboard, and what signal would show it slipping. An improvement with no owner and no measure is a pilot that will quietly revert.
Support people. Redesign the work.
Individual support and system redesign are not competing strategies. People under strain deserve support now, whatever the underlying cause. But support alone asks people to keep compensating for conditions nobody has examined.
Three things to take into Monday: look upstream of burnout, diagnose the work before selecting the intervention, and support people while redesigning the work.
This piece is adapted from a session Dr. Bains presented at the 2026 STFM Conference on Practice and Quality Improvement. The full Workforce Design Toolkit from that session, including the diagnostic, a 30-day test worksheet, an implementation guide and an evidence library, is free and ungated at energizedvision.org/stfm.
References
- Barr WB, Peterson LE, Fleischer S, Bazemore A. Pajama time and burnout: the burden of after-hours electronic health record use on family medicine residents. Acad Med. 2026;101(3):312-318. doi:10.1093/acamed/wvaf092
- Sinsky CA, Brown RL, Rotenstein L, Carlasare LE, Shah P, Shanafelt TD. Association of Work Control With Burnout and Career Intentions Among U.S. Physicians: A Multi-institution Study. Ann Intern Med. 2025;178(1):20-28. doi:10.7326/ANNALS-24-00884
- Hodkinson A, Zhou A, Johnson J, et al. Associations of physician burnout with career engagement and quality of patient care: systematic review and meta-analysis. BMJ. 2022;378:e070442. doi:10.1136/bmj-2022-070442