Exhaustion, Not Low Engagement, Differentiates Work-Related and Non-Work-Related Depression in Korean Wage Workers: A Negative-Control Analysis

Background/Objectives: Workplace mental health surveys routinely measure exhaustion and work engagement, but whether either distinguishes work-attributed depression from depression in general remains unclear. We examined this question among Korean wage workers using a negative control outcome design. Methods: We analyzed the seventh Korean Working Conditions Survey (2023), a nationally representative sample of wage workers (complete-case N = 25,776; 270 work-related and 271 non-work-related depression cases). Past-year depression was classified by workers’ attribution, with non-work-related depression as the negative control outcome. This control shares the respondent, instrument, and reporting occasion—and thus the principal reporting biases—but is presumed not to be reachable through an occupational pathway. For each exposure, we estimated the specificity contrast and assessed robustness across adjustment models, propensity-score matching, and stratification. Results: The exposures showed a double dissociation. Exhaustion was associated with both depression types, but was more strongly with work-related depression (fully adjusted specificity odds ratio, 1.17 per point; 95% confidence interval, 1.01–1.36), increasing monotonically with severity. Engagement was inversely associated with both, but did not distinguish them (0.97; 0.88–1.07). The dissociation persisted across all adjustment and sensitivity models and under propensity-score matching, where exhaustion did not distinguish between non-work-related depression and no depression—the expected negative-control behavior. Conclusions: Exhaustion, but not low engagement, distinguished work-attributed from non-work-attributed depression and may serve as a discriminating surveillance indicator. These cross-sectional associations are consistent with occupational specificity, but cannot establish causality, and they rely on single-item self-reported measures in small subgroups. Thus, replication with validated instruments and longitudinal data is required.