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