Symposium 61 - Leveraging Deep Learning to Improve Suicide Risk Detection

Chair(s)

• BA

  Brooke Ammerman, Ph.D. (she/her/hers)

  Assistant Professor
  University of Notre Dame
  Notre Dame, Indiana, United States
• TB

  Taylor Burke, Ph.D. (she/her/hers)

  Assistant Professor
  Harvard Medical School / Massachusetts General Hospital
  Providence, Rhode Island, United States

Discussant(s)

• RJ

  Ross Jacobucci, Ph.D.

  University of Wisconsin-Madison
  Madison, Wisconsin, United States

Presenter(s)

• 

  Kelly Zuromski, Ph.D.

  Research Associate
  Harvard University
  Cambridge, Massachusetts, United States
• BA

  Brooke Ammerman, Ph.D. (she/her/hers)

  Assistant Professor
  University of Notre Dame
  Notre Dame, Indiana, United States
• TB

  Taylor Burke, Ph.D. (she/her/hers)

  Assistant Professor
  Harvard Medical School / Massachusetts General Hospital
  Providence, Rhode Island, United States
• 

  Kate Bentley, Ph.D. (she/her/hers)

  Assistant Professor
  Massachusetts General Hospital/Harvard Medical School
  Boston, Massachusetts, United States

Suicide rates have continued to rise over the past decade, highlighting the complexity in suicide risk prediction. The integration of deep learning techniques in assessing suicide risk represents the potential for significant advancement. By leveraging large datasets and complex neural network architectures, these approaches can uncover subtle patterns and predictors of suicide risk that may be imperceptible to human clinicians or traditional statistical methods. Deep learning models excel at processing and analyzing heterogeneous data types, including textual data from social media and electronic health records, as well as medical images. This capability allows for identifying high-risk individuals with greater accuracy and the potential for personalized intervention strategies. Furthermore, deep learning applications can continuously learn and adapt to new data, improving their predictive capabilities over time. Thus, the use of deep learning in elucidating suicide risk indicators offers unparalleled advantages in precision, scalability, and adaptability, heralding a new era of targeted and effective interventions.

This symposium will begin with an overview of deep learning provided by our discussant, Dr. Jacobucci, who has expertise in deep learning and the specific application to suicide research. Dr. Jacobucci will provide a descriptive overview of the specific deep learning approaches used in this symposium’s talks. In our first talk, Dr. Zuromski will present data from a recently developed algorithm aimed at identify posts with suicide-related content on a military-specific social media platform. Findings demonstrate the feasibility of using social media posts to identify at-risk Servicemembers and Veterans. Dr. Ammerman will then present research from near-continuous (i.e., every five second) smartphone screenshot captures. These data aim to highlight the promise of intensive passive smartphone monitoring as an indirect proxy for self-reported suicide risk. Following, Dr. Burke will present preliminary evidence suggesting computer vision approaches can be validly used to automate the identification and assessment of medical images of self-injury scarring. Dr. Burke will subsequently discuss the potential promise of applying this novel approach to inform suicide risk prediction in high-risk settings.  Finally, Dr. Bentley will conclude the symposium by outlining findings that highlight the application of deep learning to clinician assessments. More specifically, she will present Suicide Assessment Five-Step Evaluation and Triage data from a large healthcare system, which has been transformed into a predictive model to examine relationships with suicide-related outcomes.

Presentations:

• 8:30 AM – 10:00 AM EST

  (SYM 61) Detecting Suicide Risk Among U.S. Servicemembers and Veterans: A Deep Learning Approach Using Social Media Data
  

  Speaker: Kelly L. Zuromski, Ph.D. – Harvard University

  Co-author: Daniel Low, M.A. – Harvard University

  Co-author: Noah Jones, M.S. – Massachusetts Institute of Technology

  Co-author: Daniel Kessler, M.A. – Massachusetts Institute of Technology

  Co-author: Carlos Madden, B.A. – RallyPoint Networks, Inc

  Co-author: Satrajit Ghosh, Ph.D. – Massachusetts Institute of Technology

  Co-author: Dave Gowel, B.S. – RallyPoint Networks, Inc

  Co-author: Matthew K. Nock, Ph.D. (he/him/his) – Harvard University
• 8:30 AM – 10:00 AM EST

  (SYM 61) An Application of Screenomics to Improve Real-time Suicide Risk Detection
  

  Speaker: Brooke A. Ammerman, Ph.D. (she/her/hers) – University of Notre Dame

  Co-author: Ross Jacobucci, Ph.D. – University of Wisconsin-Madison

  Co-author: Nilam Ram, Ph.D. – Stanford University
• 8:30 AM – 10:00 AM EST

  (SYM 61) Employing Computer Vision to Objectively Assess the Severity of Self-injury to Augment Suicide Risk Prediction
  

  Speaker: Taylor A. Burke, Ph.D. (she/her/hers) – Harvard Medical School / Massachusetts General Hospital

  Co-author: Akash Nagaraj, Ph.D. – Brown University

  Co-author: Richard T. Liu, Ph.D. – Harvard Medical School

  Co-author: Kathryn R. Fox, Ph.D. – University of Denver

  Co-author: Thomas Serre, Ph.D. – Brown University
• 8:30 AM – 10:00 AM EST

  (SYM 61) Leveraging Clinician Suicide Risk Assessment to Predict Suicidal Behavior in a Large Healthcare System
  

  Speaker: Kate H. Bentley, Ph.D. (she/her/hers) – Massachusetts General Hospital/Harvard Medical School

  Co-author: Chris Kennedy, PhD (he/him/his) – Massachusetts General Hospital/Harvard Medical School

  Co-author: Pratik Khadse, MSBA (he/him/his) – Massachusetts General Hospital

  Co-author: Emily Madsen, BS (she/her/hers) – Massachusetts General Hospital

  Co-author: Jordan Smoller, MD, ScD (he/him/his) – Massachusetts General Hospital/Harvard Medical School

  Co-author: Taylor A. Burke, Ph.D. (she/her/hers) – Harvard Medical School / Massachusetts General Hospital