Fusion of pain avoidance and the contingent negative variation induced by punitive condition predict suicide ideation in a college population

Abstract

This study examined behavioral and ERP features involved in pain processing as predictors of suicide ideation. Twenty-seven depressed undergraduates with high suicide ideation (HSI), 23 depressed undergraduates with low suicide ideation (LSI), and 32 healthy controls (HCs) completed the clinical Scales. The amplitudes of LPP, P2, P3, CNV, FRN, power in the beta, theta, and delta bands in the SAID task were multimodal EEG features. A machine learning algorithm known as support vector machine was used to select optimal feature sets for predicting pain avoidance, depression, and suicide ideation. The accuracy of suicide ideation classification was significantly higher for multimodal features (78.16%) which pain avoidance ranked the first and the CNV ranked the fifth than a single ERP feature model (66.62%). Pain avoidance emerged as the most optimal feature of suicide ideation classification than depression. And the CNV elicited by punitive cues may be a biomarker in suicide ideation. Pain avoidance and its related EEG components may improve the efficacy of suicide ideation classification as compared to depression.

Keywords: Machine learning; Psychological pain; Self-reference affective incentive delay task; Suicide; Time/time-frequency features of ERP.