Artificial Intelligence in Behavioral Science ESC

Artificial Intelligence in Behavioral Science ESC - Boston May 4, 2017 Chris Poulin Principal Partner Disclaimers: This material is based upon work supported by the Defense Advanced Research Projects Agency (DARPA), and Space Warfare Systems Center Pacific under Contract N 6600111 -4006. Also supported by, the Intelligence Advanced Research Projects Activity (IARPA) via the Department of Interior National Business Center contract number N 10 PC 20221. The opinions, findings and conclusions or recommendations expressed in this material are those of the authors(s) and do not necessarily reflect the views of the Defense Advanced Research Projects Agency (DARPA) and Space, the Naval Warfare Systems Center Pacific, or IARPA/ DOI/NBC, the U. S. Government, or the Microsoft Corporation. © 2013 -2016 Patterns and Predictions

Why Behavioral and Mental Health? The Highest Health Care Cost In America? Mental Disorders Huffington Post – May 23, 2016 The Neglect of Mental Illness Exacts a Huge Toll, Human and Economic Scientific American – May 1, 2012 U. S. Suicide Rate Surges to a 30 -Year High New York Times - April 22, 2016 Slide 5

Predictive Health Care The math is like that of Moneyball (Lewis), where instead of statistics on your players, your looking at statistics on your consenting users. Big Data + Analytics provides a; • Greater risk assessment over time for individuals • Greater number of individuals that can be reached (e. g. rural communities) • Cost Reductions: As we can; • Reach millions with CURRENT systems • Cut health care costs by isolating those at TRULY risk Slide 6

Medical Context for Suicide • 50 -66% of individuals who die by suicide • NCVS pilot data: 65% of suicidal military never mention or deny suicidal personnel in mental health treatment intent/ideation prior to their deaths “looked suicidal” when assessed for (Bush et al. , 2003; Coombs et al. , 1992; Hall et al. , 1999; Kovacs et al. , 1976) suicidal beliefs, even though they had • Highest risk individuals more likely to be denied suicidal ideation on standard missed because they experience frequent screening tool and rapid fluctuations in intent (Witte et al. , 2005) -------------> • Assessing “suicidal beliefs” instead of suicidal intent can improve detection of risk even when the service member denies intent • Slide 3

The Durkheim Project • The project was named in honor of Emile Durkheim, a founding sociologist whose 1897 publication of Suicide defined early language analysis for suicide explanations. • Our team was comprised of a multidisciplinary team of artificial intelligence, & medical experts (Dartmouth & VA) • Funded by DARPA from mid 2011 - early 2013 • Based on mathematics funded prior by IARPA • Delivered a data-powered mental health screening system Slide 5

Our Technology Approach • Opt-In is critical • Technical Problem: How to build a system that collects, stores, analyzes, and allows clinicians to react at Internet scale? Phase 1: • Machine Learning (Basic) • Scalable Machine Learning Phase 2: • Opt-In Interface Layer • Data Collection Layer • Storage Layer Phase 3: • Automated Intervention Slide 7

Machine Learning (with VA) • We developed linguistics-driven prediction models to estimate the risk of suicide. • These models were generated from unstructured clinical notes • From the clinical notes, we generated datasets of single keywords and multi-word phrases • We were able to initially predict suicide with 65%* accuracy on a small dataset. • * We have since reached 70% Slide 18

Example Features • Red terms are from Suicide Positive Cohort • Yellow terms are from Psychiatric/Non-Suicidal Control • Green terms are from Control Group Slide 9

Methodology (Results) Predicting the risk of suicide by analyzing the text of clinical notes. PLOS ONE. March 20, 2014 Artificial Intelligence in Behavioral and Mental Healthcare. Elsevier/Academic Press. October 2, 2015 Slide 10

Dashboard Solution (for DARPA) Dashboard • Provides real-time monitoring of patients by Clinical Professionals • Also enables a buddy-system • (Live at Dartmouth Hitchcock Medical Center) Comparative Risk Monitoring • By Cohort • Daily Tracking (think Fitbit™ for mental health) Slide 11

Opt-In Interface (App) • Simplified user participation for consent, and privacy control • Consent based access to online resources: Slide 12

Storage Layer • The data is secure & protected • Specifically, the PII scrubbed data persists at a medical center, behind a HIPAA compliant firewall (e. g. Dartmouth Hitchcock Medical Center below) • The highest professional security standards Slide 13

Technology Stack Slide 2

Facebook Partnership On July 2013, in conjunction with Facebook, Inc. , we publically launched our project, with the aim of raising awareness, as well as direct recruiting of subjects. https: //www. facebook. com/notes/us-military -on-facebook/durkheim-projectlaunch/472400246179609 … And since, we have been covered by NPR, NBC. com, Time. com, Fast Company, The Boston Globe, Daily Mail UK, Mashable, CIO, and CNN. com Slide 15

Intervention Automated systems are coming online for potential patients and families seeking treatment, as well as passive intervention strategies (i. e. ‘safety plan’narratives). Interventions that are designed to be timely and appropriate Slide 16

Interventions Sequence Slide 17

Preview 1 (*Upcoming User Applications) *Unpublished Results Slide 17

Preview 2 (*Interventions Timing) *Unpublished Results Slide 17

Takeaways Suicide and Mental Health Risks: An opt-in risk assessment and intervention routing system ALREADY exists. We just need to deploy this at scale, through the responsible interface with medical practitioners and social workers. The challenges are mostly medical inertia (cost of care, liability, etc. . . ) *Insider Threat: Also consider that: "Germanwings co-pilot Andreas Lubitz saw 41 doctors in five years; . . . " - ABC "Germanwings Co-Pilot Andreas Lubitz Was Treated for Suicidal Tendencies" - WSJ "Andreas Lubitz repeatedly set the same plane for an unauthorized descent earlier that day. " - BBC "He told former girlfriend he was planning an act so horrifying his name would be remembered forever. " - Dailymail. co. uk Slide 19

Partners Slide 2

Thank you Chris Poulin Patterns and Predictions +1 617. 755. 9049 And please remember: National Suicide Prevention Lifeline 1 -800 -273 -TALK (8255) Slide 19

Artificial Intelligence in Behavioral Science ESC — Boston