Electronic Health Record-based Detection of Self-harm and Suicidal Behavior in School-age Children
BACKGROUND: Suicide is the second leading cause of death among young people aged 10-24 in the United States and is the ninth leading cause of death among children 6-12 years old. GAP: Emergency department visits for suicide attempts and self-harm behaviors among school-age children are not readily identifiable in electronic health record (EHR) databases, correspondingly, nor are intervention and prevention efforts. HYPOTHESIS: An EHR-based machine learning (ML) classifier jointly informed by a combination of natural language (clinician note text) and structured data (demographics, diagnoses, medications, triage screening, utilization, procedures, orders) will more accurately identify suicide and self-harm related ED visits among school-age children compared with existing means for discovery (International Classification of Disease (ICD) codes and Columbia Suicide Severity Rating Scale scores). METHODS: A retrospective cross-sectional study designed to develop and validate a method of EHR-based case and control ascertainment of self-harm and suicide-related thoughts and behaviors specific to children aged 6-12 years. The primary data source is the UCLA Clinical and Translational Science Institute (CTSI) informatics database. EHRs from 600 visits (400 training, 200 validation) will be reviewed by two clinician abstractors to establish a gold-standard. Algorithms will be trained by adaptive elastic net penalized logistic regression. The outcome variable will be the clinician-derived classification and the predictors will be the EHR data elements and natural language terms. RESULTS: Pending IMPACT: By refining methods to identify suicidal children at a large scale, our research will yield an innovative computational method to further understand the dimensions of suicidality and self-harm to complement basic, clinical, and translational child mental health research at the individual-level.