2 - (OP5) How Heterogeneous Depression Symptoms Relate to Biased and Inflexible Updating of Ambiguous Information: A Machine Learning Study

Background: Symptom presentations in depression vary widely across individuals, with substantial symptom-level differences in risk factors, heritability, treatment responsiveness, and functional impairment. Nevertheless, most cognitive models and interventions continue to treat depression as a unitary construct. Among the cognitive mechanisms implicated in depression, interpretation bias and interpretation inflexibility have emerged as distinct processes; however, their symptom-level relevance remains unexplored.

Methods: We examined how interpretation bias and interpretation inflexibility relate to individual depressive symptoms using data from 1,995 participants across multiple studies. Cognitive indices were derived from the Emotional BADE Task, and symptoms were assessed using the BDI-II. Three machine learning models—Random Forest, XGBoost, and Elastic Net—were applied to identify symptom-specific associations and assess their robustness across analytical frameworks.

Results: Interpretation inflexibility—both positive and negative—demonstrated stronger and more consistent associations with depressive symptoms than interpretation bias (R2 = 0.37 vs R2 = 0.16–0.17). Suicidal ideation and appetite change were most strongly linked to inflexibility indices across all models. In contrast, interpretation bias showed weaker and less consistent associations, primarily with cognitive-affective symptoms such as self-dislike, self-criticalness, and crying. Importantly, many associations were non-linear, and overall model performance declined when restricted to top predictors, suggesting broad, distributed contributions across symptoms.

Conclusions: Our findings underscore the value of symptom-level approaches in understanding cognitive vulnerabilities in depression. Moreover, these findings reinforce the conceptual distinction between bias and inflexibility and challenge the assumption of depression as a unitary latent construct. Interpretation inflexibility emerged as a particularly promising target, offering potential to enhance existing interventions and support personalized treatment strategies. Future studies should test whether specific symptoms respond differentially to targeted interventions. This study advances a more nuanced, mechanism-informed framework for individualized care in depression.