Association between objective sleep structure and suicidal ideation in patients with depression: a study based on polysomnographic regression and cluster analysis

BackgroundDepression is a common mental disorder, and current suicidal ideation (SI) is a clinically important symptom. Sleep disturbance is common in depression, but the relationship between objective sleep architecture and current SI remains unclear. This study examined associations between objective sleep structure and current SI and explored whether sleep-clinical clustering could support exploratory SI stratification.Methods287 patients meeting DSM-5 criteria for depressive disorder underwent clinical assessment and overnight polysomnography (PSG). Current SI was assessed using item 3 of the HAMD-17. Binary logistic regression evaluated factors associated with current SI. K-Means clustering (K = 3) was performed using standardized HAMD score, PSQI score, AHI, total awakenings, N3%, R%, and sleep efficiency. Quantitative cluster validation, sensitivity analyses excluding HAMD from clustering, and ROC-based performance metrics were additionally conducted.ResultsCurrent SI was present in 50.52% (145/287) of patients. In the individual-variable regression model, only HAMD total score was independently associated with current SI (OR = 1.683, p < 0.001). Cluster analysis identified three subgroups with distinct current SI rates: Cluster 1 (deep-sleep dominant, 37.04%), Cluster 2 (high-arousal-apnea, 70.37%), and Cluster 3 (low-arousal-subjective insomnia, 60.80%). The K = 3 solution showed WSS = 1450.71, mean silhouette = 0.204, and Davies-Bouldin index = 1.679. In the demographic-adjusted model, membership in Clusters 2 + 3 was associated with current SI (OR = 3.152, 95% CI: 1.883-5.274, p < 0.001; AUC = 0.700, 95% CI: 0.639-0.760). When HAMD was excluded from clustering, the association remained (OR = 2.669, 95% CI: 1.598-4.458, p < 0.001), whereas additional adjustment for HAMD total score attenuated the primary cluster association (OR = 0.842, 95% CI: 0.411-1.725, p = 0.639).ConclusionsDepression severity was the primary factor associated with current SI. PSG-derived sleep-clinical clusters may help characterize heterogeneous presentations of current SI, but their incremental value beyond depressive severity should be interpreted as exploratory and validated in larger longitudinal samples.