TY  - JOUR
PY  - 2023//
TI  - Detection of suicidal behavior and self-harm among children presenting to emergency departments: a tree-based classification approach
JO  - AMIA Joint Summits on Translational Science proceedings
A1  - Edgcomb, Juliet B.
A1  - Tseng, Chi-Hong
A1  - Pan, Mengtong
A1  - Klomhaus, Alexandra
A1  - Zima, Bonnie
SP  - 108
EP  - 117
VL  - 2023
IS  - 
N2  - Suicide is the second leading cause of death of U.S. children over 10 years old. Application of statistical learning to structured EHR data may improve detection of children with suicidal behavior and self-harm. Classification trees (CART) were developed and cross-validated using mental health-related emergency department (MH-ED) visits (2015-2019) of children 10-17 years (N=600) across two sites. Performance was compared with the CDC Surveillance Case Definition ICD-10-CM code list. Gold-standard was child psychiatrist chart review. Visits were suicide-related among 284/600 (47.3%) children. ICD-10-CM detected cases with sensitivity 70.7 (95%CI 67.0-74.3), specificity 99.0 (98.8-100), and 85/284 (29.9%) false negatives. CART detected cases with sensitivity 85.1 (64.7-100) and specificity 94.9 (89.2-100). Strongest predictors were suicide-related code, MH- and suicide-related chief complaints, site, area deprivation index, and depression. Diagnostic codes miss nearly one-third of children with suicidal behavior and self-harm. Advances in EHR-based phenotyping have the potential to improve detection of childhood-onset suicidality.<p />  <p>Language: en</p>
LA  - en
SN  - 2153-4063
UR  - http://dx.doi.org/
ID  - ref1
ER  -