1 - (SYM 19) Nonverbal Behaviors Exhibited During Suicide Assessments Predict Suicidal Thoughts and Behaviors: A Machine Learning-based Approach

Background: In social interaction, nonverbal behaviors contain salient affective and interpersonal cues. Because they are not easily quantifiable or perceivable to human beings in real-time, nonverbal behaviors have largely been relegated to the margins of efforts to understand clinical encounters empirically. We draw upon recent advances in computational science to objectively and efficiently characterize nonverbal behaviors exhibited during suicide assessments. We test whether facial expressions and head motion velocity of young adults varying in risk for suicide and their clinical interviewers convey concurrent and predictive information about suicide risk.

Method: Participants were 67 young adults ages 18-24 (M = 21.32, SD = 2.09) recruited from the community. Half reported past-year suicidal behavior (i.e., “suicidal participants"); the other half reported no lifetime suicidal thoughts or behaviors (STBs) (i.e., “nonsuicidal participants”). Participants completed an hour-long recorded interview with master’s-level staff, including the Columbia-Suicide Severity Rating Scale (C-SSRS; Posner et al,. 2008). Subsequent occurrence of STBs were assessed 3 months later. Facial actions and head motion associated with affective engagement were automatically measured from the first 3 minutes of the C-SSRS using machine learning-based tool PyAFAR (Hinduja et al., 2023).

Results: Suicidal young adults exhibited significantly greater velocity of eye and mouth openings than did nonsuicidal young adults. Several nonverbal behaviors of interviewers showed strong concurrent and predictive validity. In interviews with suicidal participants, interviewers less frequently used Duchenne smiles (action units 6+12); their non-Duchenne smiles showed greater complexity (i.e., more peaks and valleys in intensity); and their head motion was less animated (i.e., reduced velocity of pitch and yaw). When STBs were evaluated 3-months post interview, nonverbal behaviors of the interviewers but not clinical ratings predicted occurrence of suicidal behavior and severity of suicidal ideation.

Conclusion: Nonverbal behaviors of clinical interviewers during the first 3 minutes of the C-SSRS varied between groups with and without history of STBs and predicted future occurrences of STBs over the following 3 months. Clinical ratings failed to show similar predictive validity.