Prediction of suicidal ideation and depression in the general population with subthreshold insomnia using machine learning models.

BACKGROUND: Insomnia is a significant independent risk factor for depression and suicidality. However, these conditions often go undetected, particularly in individuals presenting with sleep complaints. This study aimed to develop and validate machine learning (ML) models for the indirect screening of suicidal ideation (SI) and depression and to specifically evaluate their performance in a population reporting at least subthreshold insomnia. METHODS: Data were obtained from a Slovenian nationwide community sample (N = 2,989) via an online questionnaire. Logistic regression models were developed to predict SI (measured by SIDAS) and moderate-to-severe depression (measured by DASS-21) via indirect predictors, including socio-demographics, life satisfaction, behavioral changes, and 14 coping strategies from the Brief COPE inventory. The model performance was tested on a validation sample, which was stratified into groups with (Insomnia Severity Index [ISI] score ≥ 8; n = 917) and without (ISI < 8; n = 819) insomnia symptoms. RESULTS: The models demonstrated strong and consistent predictive performance across both groups. The area under the receiver operating characteristic curve (AUROC) for the SI model was 0.78 in the insomnia group and 0.80 in the non-insomnia group. For the depression model, the AUROCs were 0.79 and 0.82, respectively. The minimal difference in performance indicates that the models are robust and equally effective regardless of the presence of insomnia. CONCLUSION: Our findings demonstrate that ML models using indirect questions can effectively screen for both suicidality and depression simultaneously. The models' robust performance in individuals with insomnia highlights their potential as feasible, ethical, and efficient tools for early detection. Given that sleep complaints are a common reason for seeking healthcare, this approach offers a critical opportunity for timely intervention in a high-risk population, potentially reducing preventable morbidity and mortality associated with suicide and depression.