Multimodal clinical, sleep, and autonomic correlates across the depressive spectrum: a cross-sectional study

Depression is a heterogeneous condition characterized by affective symptoms, sleep disturbance, and altered physiological regulation, but the relative contribution of subjective and objective measures to depressive status remains unclear. In this cross-sectional study, 101 participants were categorized as healthy controls (HC, n = 47), subthreshold depression (SD, n = 30), and major depressive disorder (MDD, n = 24). Demographic characteristics, symptom questionnaires, polysomnographic variables, heart rate variability (HRV) indices, and eye-tracking measures were collected. Group differences were assessed across the three groups, and classification analyses were then performed by defining HC as the non-depressive group and combining SD and MDD as the depressive group. Univariate and multivariable logistic regression analyses were used to identify independent correlates of depressive status, and model discrimination was evaluated using the area under the receiver operating characteristic curve, accuracy, sensitivity, specificity, F1 score, and Brier score. Significant between-group differences were found for HAMD-17, PHQ-9, GAD-7, PSQI, ISI, NDQ, sleep latency, rapid eye movement latency, and RMSSD. After adjustment for age and sex, higher GAD-7 scores, greater NDQ scores, and longer rapid eye movement latency remained independently associated with depressive status. In model comparison analyses, the combined model showed the highest discriminative performance (AUC = 0.942), followed closely by the clinical model (AUC = 0.930), whereas the polysomnographic and HRV models showed lower performance. These findings indicate that depressive status in this cohort was primarily characterized by greater affective and sleep-related symptom burden, while rapid eye movement latency provided additional objective information beyond questionnaire-based assessment. However, these models should be interpreted as cohort-internal, screening-oriented discrimination models within a cross-sectional design rather than prospective, diagnostic, or etiological models.