Beyond independent effects: stress and body image in adolescent depressive risk

Perceived stress and body image are both associated with adolescent depressive symptoms, but whether body-image-related vulnerability is amplified under greater stress remains unclear. This study examined their joint association using a large, nationally representative sample of Korean adolescents. Data from 201,895 adolescents participating in the Korea Youth Risk Behavior Survey (2022–2025) were analyzed, with 2022–2023 used for model development ( N = 99,235) and 2024–2025 for temporal validation ( N = 102,660). The prespecified analysis used multivariable logistic regression with an ordinal-linear stress × body-image interaction. To examine the functional-form assumption, categorical and quadratic body-image specifications were examined post hoc. Survey-weighted models were used as sensitivity analyses, and a balanced random forest was used as an exploratory nonlinear complement. Classification thresholds were selected in the development data and applied unchanged to temporal validation. Perceived stress was strongly associated with self-reported depressive symptoms (OR = 2.715, 95% CI 2.555–2.885), whereas the prespecified stress × body-image interaction was not significant (OR = 0.996, 95% CI 0.977–1.014, p = .640). Survey-weighted analysis yielded a similar interaction estimate (OR = 0.994, p = .611). Post hoc categorical modeling improved fit and yielded a significant overall interaction, but category-specific effects were not uniform and no individual interaction remained significant after multiplicity correction in the survey-weighted analysis. The quadratic specification was not supported ( p = .168). In temporal validation, logistic regression achieved an AUC of 0.736 (95% CI 0.732–0.739), compared with 0.712 (95% CI 0.709–0.716) for the balanced random forest. The prespecified diathesis–stress amplification hypothesis was not supported. Perceived stress was the most consistent correlate of self-reported depressive symptoms, while post hoc analyses suggested possible category-specific heterogeneity in body-image associations that requires independent confirmation. Exploratory random-forest analyses did not improve discrimination; predicted probabilities from the balanced model overestimated absolute risk and are not calibrated clinical risk estimates.