Predicting suicidal ideation from depression screening data: A network‑augmented machine learning approach

Suicidal ideation often goes undetected in brief screenings. We developed a model that infers risk from routine depressive symptom data by integrating machine learning with symptom-network features. Using nationally representative data from US adults (NHANES; N = 44,922), we predicted ideation (PHQ-9 item 9 ≥1) under three specifications: (1) PHQ-8 total score; (2) eight PHQ-8 items; and (3) items plus individualized network features. Models used 10-fold cross-validation; precision-recall AUC (PR AUC) was the primary metric. Both item-level and network-augmented models outperformed the total-score baseline. The network-augmented XGBoost performed best (PR AUC = 0.37) while meeting screening criteria (sensitivity ≥0.80; specificity ≥0.50). The most influential signals were the centrality and severity of depressed mood and worthlessness/guilt, network density, and selected edges. External validation across five datasets (N = 808,023) yielded normalized PR AUCs of 0.32-0.51, supporting generalizability. Findings suggest that network augmentation can improve interpretability without compromising model performance.