Multidomain correlates of burnout: A population-based study using supervised machine learning

Burnout is a growing challenge for individuals and society and understanding risk factors is crucial to develop targeted prevention strategies. Thus, we aimed to determine whether and how burnout relates to different organisational factors, psychosocial employment conditions, sociodemographic variables and mental health.

We analysed cross-sectional survey data collected in the 2023 wave with focus on occupational health within the German population-based DigiHero cohort. Using linear associations and xg-boost machine learning, we explored the relations of multidomain variables with burnout severity, which was measured with the Maslach Burnout Inventory. Machine learning results were interpreted through SHAP values.

We included 27,020 participants, aged 18-67 years (median 49), 65.2% female, in the analysis. The strongest predictors of burnout were effort-reward-imbalance, work-life-interference, overcommitment and poor general mental health, particularly symptoms of depression and anxiety. Compared to psychosocial employment conditions and individual mental health, organisational and sociodemographic factors were less important. However, occupation mattered and other aspects, such as extended remote work, had small, but noticeable effects. Age and income were the only sociodemographic factors that contributed. Results from xg-boost were compared to results from univariate statistics and remained robust when adjusting for sample bias through case weights.

The study emphasises the importance of psychosocial employment conditions and individual mental health in the context of burnout, which outweigh the effects of organisational and sociodemographic factors. For practice this suggests that positive social work environments may prevent burnout while mental health problems increase vulnerability. For causal insights, prospective longitudinal studies are required.