PROJECT SUMMARY / ABSTRACT The candidate requests support for a five-year program of training and research to better understand how electronic health record phenotyping and other computational methods applied to existing medical record data can bolster detection and prediction of suicide attempts by children ages 10 to 17. In the proposed training plan, the candidate will build upon her previous experiences in social psychology, clinical informatics, and clinical child and adolescent psychiatry to perform a multidisciplinary project at the University of California, Los Angeles Health System. Her training plan includes developing skills and knowledge in 1) analysis of natural language (text) data, 2) development of risk algorithms in healthcare settings to improve suicide prevention, 3) basic qualitative research skills including modified Delphi Panel approach, and 4) the responsible conduct of research. Suicide is the second leading cause of death of young people over 10 years old in the United States and suicide attempts among children are common, costly and preventable. There is an urgent need to close the gap between risk prediction algorithms and clinically-useable tools that can enhance medical decision- making for providers and families. This study proposes that electronic health record phenotyping, a method of standardizing case detection using clinical note text and structured medical record data, may offer improved detection and personalized risk prediction for children, thus complementing existing suicide prevention efforts. In the proposed research, using a cross-sectional design, Aim 1 will focus on adaptation of electronic health record phenotyping to detect emergency department visits for suicide attempts by children using electronic health records. Then, using a case-control design, Aim 2 will focus on development of risk prediction models of emergency department visits for suicide attempts by children using longitudinal electronic health records over two years. Aim 3 will focus on assessment of the validity, acceptability, usability, feasibility, and overall utility of a personalized risk prediction prototype with case simulations using a modified Delphi panel approach. This plan will parallel a training plan building skills and knowledge to bridge informatics, computational methods, and clinical child psychiatry. In the long term, this research is an initial step to enhance signal detection and support prediction of suicide attempts, in turn, setting the stage for deployment of personalized approaches to prevention in clinical settings where providers, youth, and families may directly benefit.

PROJECT NARRATIVE Suicide is the second leading cause of death among U.S. children: suicide attempts are common, costly, and preventable, and reduction in youth suicide is a national and international priority area. Computational approaches to predict risk of future suicide attempts using medical records have emerged as a promising means to address the crisis of youth suicide by identifying children who could benefit from early, evidence- based intervention before suicidal behavior occurs or escalates. This study seeks to improve detection and prediction of emergency department visits for suicide attempts among children aged 10 to 17 by newly adapting a computational method applied to existing medical records (electronic health record phenotyping), and then assessing the perceived added clinical benefit of an electronic health record-integrated personalized risk prediction software prototype among potential end users.