Multimodal brain connectome-based prediction of suicide risk in people with late-life depression

Suicidal ideation, plans and behavior are particularly serious health issues among the older population, resulting in a higher likelihood of deaths than in any other age group. The increasing prevalence of depression in late life reflects the urgent need for efficient screening of suicide risk in people with late-life depression. Employing a cross-sectional design, we performed connectome-based predictive modelling using whole-brain resting-state functional connectivity and white matter structural connectivity data to predict suicide risk in late-life depression patients (N = 37 non-suicidal patients, N = 24 patients with suicidal ideation/plan, N = 30 patients who attempted suicide). Suicide risk was measured using three standardized questionnaires. Brain connectivity profiles were used to classify three groups in our dataset and two independent datasets using machine learning. We found that brain patterns could predict suicide risk in the late-life depression population, with the explained variance up to 30.34%. The functional and structural connectivity profiles improved the classification-prediction accuracy compared with using questionnaire scores alone and could be applied to identify depressed patients who had higher suicide risk in two independent datasets. Our findings suggest that multimodal brain connectivity could capture individual differences in suicide risk among late-life depression patients. Our predictive models might be further tested to help clinicians identify patients who need detailed assessments and interventions. The trial registration number for this study is ChiCTR2200066356.