The overarching goal of the proposed research is to develop methods to automate electronic health record (EHR) cohort discovery of children and adolescents newly presenting for suicide-related emergency care.
Improving capacity to detect suicide-related emergency encounters is critical: Emergency departments provide an important safety net for youth experiencing suicidal thoughts and behaviors. In children and adolescents, the behavioral presentation (phenotype) of suicidality is affected by development, family and peer relationships, and social determinants of health to which youth are uniquely vulnerable, such as bullying, foster care, and emergence of sexual and gender identity. Although EHRs have emerged as a major source of data for suicide research, there is no clearly established method to identify all suicide-related emergency encounters by children and adolescents from within a large EHR dataset. Phenotype algorithms using EHR data to classify patients with specific disease and outcomes are well-established in informatics research, successfully applied to depression, diabetes mellitus, inflammatory bowel disease, multiple sclerosis, rheumatoid arthritis and other disease conditions. However, no such algorithms currently exist to classify children and adolescents with suicide-related behavior.
In the proposed research, we seek to identify cases of new-onset suicidality by developing an algorithm that learns from data, i.e., a machine learning model, jointly informed by 14 years of EHRs from UCLA's Integrated Clinical and Research Data Repository and child psychiatrist expertise. The algorithm will use both coded data (demographics, medications, diagnoses, and universal triage screening for suicidality) and text data from physician notes to classify emergency encounters for suicidal thoughts, behaviors, and self-injury. The algorithm will then be tested on a subset of encounters to ascertain performance for varied presentations (subphenotypes) of suicidality.
If funded, this research would form the basis of a proposal to the National Institute of Mental Health (NIMH) for a Mentored Patient-Oriented Research Career Development Award (K23) dedicated to a scalable, systematic, and developmentally-informed method to identify each instance of emergency care for suicidal thoughts and behavior in youth. In turn, this research would enable rapid identification of case positives of suicide-related behavior variants in massive EHR data, greatly increasing the predictive value of current suicide risk prediction models and identifying potentially modifiable predictors of suicide-behavior escalation.
