Can Personal Electronic Communications Identify Suicide Risk in Real Time?
Glenn, Jeffrey, Psychology - Graduate School of Arts and Sciences, University of Virginia
Teachman, Bethany, Department of Psychology, University of Virginia
Suicide is a serious public health problem and a leading cause of death around the world. Despite its tragic toll on society, our methods for identifying individuals when they are at highest risk of suicide remain ineffective. As such, objective tools to dynamically assess level of suicide risk are sorely needed to determine not just who but when someone is at imminent risk of killing themselves. The proliferation and ubiquity of mobile phone text messaging offers a promising avenue for gaining insight into possible novel, real-time markers of suicidal behaviors. In this pilot investigation, we utilized a within-subject, laboratory-based research design to identify and better understand real-time patterns in communication unique to periods preceding suicide attempts when suicide risk is especially high.
Individuals reporting a history of suicide attempt (N=33) were recruited from the Psychology Department’s participant pool and the UVA/Charlottesville community. After collecting their phone text messaging data (SMS), participants were asked to retrospectively identify and describe past suicide attempts, as well as periods of suicide ideation, depressed mood, and positive mood. An automated language analysis software package (LIWC) was then used to produce scores for each text message capturing five psychological constructs of interest: self-focus, sentiment, social engagement, time-orientation, and cognitive performance. Within-subjects analyses were performed to test whether these characteristics differed in general (mean differences) and over time (slope differences) just before a suicide attempt (high risk), relative to other periods when participants had suicidal thoughts but did not attempt (moderate risk), or were depressed but not suicidal or during periods of positive mood (low/minimal risk).
In terms of overall mean differences, results indicated that high suicide risk was associated with messages indicating greater anxiety, sadness, and orientation towards or focus on the future, as well as more complex and higher-status communication, though these language features did not uniquely and consistently differentiate suicide attempts from other episodes. In terms of differences in patterns over time, high suicide risk was associated with language indicating greater increases over time in self-focus, elaborated fluency, and orientation towards the future, though, as with mean differences, pairwise comparisons between episode types were mixed. Most notably, however, results suggested that anger increased and positive emotion decreased to a greater extent as one approached a suicide attempt, relative to the other episode types, potentially providing unique markers of high suicide risk.
Overall, these results indicate that personal electronic communication has the potential to provide clues into the suicidal mind and offer temporally sensitive markers of suicide risk. Specifically, emotional content of language (and anger in particular), when examined over time, may represent unique psychological features indicative of high suicide risk. In the short term, findings from this study may be utilized to construct machine learning models that attempt to predict acute suicide risk. In the long term, such predictive models may serve as the basis for developing objective tools to determine level of suicide risk in real time and provide at-risk individuals the help they need before they attempt suicide.
PHD (Doctor of Philosophy)
Suicide, Suicide Risk, Suicide Prediction, Mental Health, Data Science, Machine Learning, Natural Language Processing, Linguistic Analysis, Sentiment Analysis, Personal Electronic Communication, Text Messaging (SMS), Social Media
Presidential Fellowship for Data Science, University of Virginia
All rights reserved (no additional license for public reuse)