Identifying and Understanding Suicide Risk among Young Children

About This Grant

Project Summary Despite widespread prevention efforts, suicide remains a major public health concern. Most extant research has focused on suicide in adolescents and adults; yet suicide risk among children is understudied. Suicide is now the 3rd leading cause of death among 8-12-year-olds, reflecting a clear public health crisis. The pediatric emergency department (ED) is often the primary point of clinical contact to manage suicidal thoughts and behaviors (STB) for children and the months post-ED discharge are a high-risk period for STB to reoccur. Improving identification and prediction of suicide risk for children in the ED is a critical gap. This R21 will leverage rich electronic health records (EHR) from 5 pediatric EDs across the United States that have been harmonized within the Pediatric Emergency Care Applied Research Network (PECARN). We anticipate ~7,000 EHR from 8-12-years-old patients yielding sufficient power for nuanced analytics. Preliminary data highlight that STB are unfortunately common among children receiving psychiatric evaluation in the ED; ~20% of children in our ED reported a history of suicide behaviors or attempts and another ~40% expressed lifetime suicide ideation. To improve understandings of STB in children, the following aims will be addressed. First, to identify sociodemographic and clinical factors related to STB risk, we will directly compare children in the ED for suicide-related concerns with patients admitted for non-suicide-related psychiatric concerns (e.g., depression). Second, we will examine predictors of return ED visits for STBs within 1-year of discharge. Our preliminary data suggest that ~15% of children will return to the ED within 1-year. Risk factors for return ED visits will be ascertained from both structured EHR data (e.g., insurance status) as well as clinician free-text notes. In preliminary data, we developed lexicon-based approaches to parse clinician notes and extracted novel risk factors for analysis. Preliminary results show that clinician notes can be used to identify key factors of interest, including a history of prior psychiatric hospitalization, and impulsivity. Third, we will apply natural language processing using large language models trained on independent EHR to identity severity and frequency of domains of risk, including negative affect and sleep disturbance, which will further improve prediction of return visits to the ED for STBs. Finally, these aims will be integrated into a practical risk calculator to aid in screening and triaging potential suicide risk for children in the ED. Together, this work will lead to a future multi-center, prospective project designed to improve screening and intervention for high-risk, suicidal children.