When a worker reports depression, what makes it reasonable to describe the condition as work-related? The answer matters for prevention, surveillance and compensation. Yet it is difficult to separate a workplace pathway from depression that affects someone’s experience of work regardless of where it began.
Yangwook Kim and colleagues approach that problem with an inventive comparison. Their analysis of Korean wage workers asks whether exhaustion and low work engagement distinguish depression attributed to work from depression attributed to other causes. The result is suggestive: exhaustion does, modestly; low engagement does not. The design is stronger than a conventional cross-sectional association, but it is still a signal rather than a diagnosis or causal verdict.
Why use a negative control?
The study uses the seventh Korean Working Conditions Survey, conducted in 2023. Its complete-case sample contains 25,776 wage workers, including 270 people reporting work-attributed depression and 271 reporting depression not attributed to work. The survey’s complex sampling design and weights are retained in the main analysis.
Non-work-attributed depression serves as a negative-control outcome. It is reported by the same kind of respondent, in the same instrument and at the same time, so it shares many sources of reporting bias. But, under the design’s key assumption, it should not be reached through an occupational pathway. If exhaustion merely reflects depressed mood or a general tendency to report problems, its association should look similar in both depression groups.
The authors analyse exhaustion and engagement separately, adjust for demographic and employment factors, and test the findings through alternative adjustment sets, stratification and propensity-score matching. They are unusually explicit about assumptions that the cross-sectional data cannot verify, including the temporal position of engagement and the ambiguous role of sleep problems.
A double dissociation—with small numbers behind it
Each point on the study’s 0–8 exhaustion scale is associated with higher odds of both forms of depression. The association is stronger for work-attributed depression: the fully adjusted specificity odds ratio is 1.17 per point, with a 95% confidence interval from 1.01 to 1.36. Mean exhaustion rises from 3.8 among workers without depression to 4.1 in the non-work-attributed group and 4.6 in the work-attributed group.
Engagement behaves differently. Higher engagement is associated with lower odds of both depression outcomes, but does not distinguish between them: the specificity odds ratio is 0.97, with a confidence interval from 0.88 to 1.07. In the authors’ terms, exhaustion and engagement form a double dissociation. One is differentially associated with the worker’s attribution to work; the other looks like a more general correlate of depression.
The pattern survives many sensitivity analyses. The exhaustion contrast stays between roughly 1.15 and 1.23 across alternative models, although it becomes borderline rather than conventionally significant in several specifications. Pairwise propensity-score matching reproduces the direction: matched workers with work-attributed depression report more exhaustion than those with non-work-attributed depression, while the negative-control comparison does not distinguish non-work-attributed depression from no depression on exhaustion.
This triangulation is a real strength. The study does not depend on a single adjusted regression, and the authors interpret the graded pattern more carefully than the largest categorical estimate, whose interval is very wide.
What remains unresolved?
All of the central variables are self-reported at one time. A worker who experiences their job as exhausting may be more likely to attribute depression to work; depression may make work feel more exhausting; occupational conditions may produce both; or all three processes may operate. The negative control reduces some explanations based on general low mood and common reporting conditions. It cannot establish direction.
The depression measure is a single survey item rather than a validated clinical assessment, and attribution is also reported rather than independently established. The two depression groups are small relative to the full sample. Dividing them for categorical and subgroup analyses produces imprecise estimates, and the apparently stronger association among women and younger workers is not supported by significant interaction tests. Complete-case analysis, residual confounding and the structural differences between the two depression groups remain concerns.
There is also a conceptual risk in calling exhaustion a discriminating surveillance indicator. A measure can distinguish groups in a population without being accurate enough to identify an individual case or justify an employment decision. Used badly, a workplace survey could individualise a problem—classifying an exhausted employee—while leaving staffing, schedules, control and insecurity untouched.
From an exhausted person to an organised activity
My autopoietic ecology (AE) interpretation treats exhaustion as a possible sign that an activity is continuing by drawing down the capacity needed to continue it. A service can still be delivered and a worker can still appear productive while recovery time, attention or bodily resilience is being depleted. The important unit is not an isolated score but the organised relation among demands, time, authority, support and recovery.
The study cannot demonstrate that relation. Its exhaustion items are work-framed, but it does not directly observe workload allocation, supervisory response, schedule changes or whether reporting exhaustion alters the next decision. The AE distinction between performance and renewed capacity therefore suggests questions for a later design rather than a conclusion from this one.
A conventional Job Demands–Resources account is a strong alternative. Exhaustion may lie on a health-impairment pathway shaped by job demands, whereas engagement may be a broader motivational state with substantial overlap with depressive symptoms. The reported pattern can be interpreted within that framework without AE. The useful comparison is not which vocabulary sounds more comprehensive, but which design specifies modifiable conditions and predicts what changes when those conditions are altered.
A stronger test would follow workers over time with validated depression and exhaustion measures, administrative or observational indicators of hours and demands, and repeated records of control, recovery and organisational response. A workplace intervention that changes scheduling or workload would be especially informative. If exhaustion falls and later depression risk changes while relevant alternatives are addressed, the causal case would strengthen. If the association disappears with prospective measurement, the present interpretation would need revision.
Use the signal at the right level
The practical implication is not to diagnose workers from an exhaustion score. At group level, a graded rise can help identify teams or settings that deserve investigation. That investigation should ask what demands are present, who can change them, whether workers can report problems safely and whether a report produces a timely response.
The paper improves on a familiar claim that “burnout is linked to depression” by asking whether the association has occupational specificity. Its answer is careful and worth pursuing: exhaustion contains a work-related signal, but the route from signal to cause—and from cause to repair—remains to be shown.
Reference: Kim, Y., Yun, B.-Y., Lee, W.-T., Park, H.-J., Oh, J.-Y., Lee, J.-M., Lee, J., & Yoon, J.-H. (2026). Exhaustion, Not Low Engagement, Differentiates Work-Related and Non-Work-Related Depression in Korean Wage Workers: A Negative-Control Analysis. Healthcare, 14(18), 3113. Published 21 September 2026. https://doi.org/10.3390/healthcare14183113.
EdgeLab is an independently organised research network. It is not formally recognised as a research group, centre or unit by the Faculty of Education or the University of Cambridge. Institutional affiliations refer to individual contributors and do not imply institutional endorsement.
Leave a Reply