Objective: About 80% of patients who commit suicide do not report suicidal ideation the last time they speak to their mental health provider, highlighting the need to identify biomarkers of suicidal behavior. Our goal is to identify suicidal behavior neural biomarkers to classify suicidal psychiatric inpatients.
Methods: Eighty percent of our sample [suicidal (n = 63) and non-suicidal psychiatric inpatients (n = 65)] was used to determine significant differences in structural and resting-state functional connectivity measures throughout the brain. These measures were used in a random forest classification model on 80% of the sample for training the model.
Results: The model built on 80% of the patients had sensitivity = 79.4% and specificity = 72.3%. This model was tested on an independent sample (20%; n = 32) with sensitivity = 81.3% and specificity = 75.0% for confirming the generalizability of the model. Altered resting-state functional connectivity features from frontal and middle temporal regions, as well as the amygdala, parahippocampus, putamen, and vermis were found to generalize best.
Conclusion: This work demonstrates neuroimaging (an unbiased biomarker) can be used to classify suicidal behavior in psychiatric inpatients without observing any clinical features.
Keywords: biostatistics; magnetic resonance imaging; neuroimaging; self-harm; suicide.
Published 2019. This article is a U.S. Government work and is in the public domain in the USA.