A new study by UCLA Health reveals that hospital emergency departments may be missing signs of suicidal thoughts and behaviors in children, boys and Black and Hispanic youth.
The , published in the journal JAMA Open Network, analyzed electronic health records of nearly 3,000 children and teenagers presenting to two emergency departments in southern California for mental health reasons. Using machine learning algorithms, the researchers determined standard medical record surveillance methods miss youth with suicide-related emergencies. These methods disproportionately missed suicide-related visits among Black, Hispanic, male, and preteen youths, compared with other races and ethnicities, female youths, and adolescents.
“Existing methods are missing kids, and not missing them at random,” said Dr. , study corresponding author, associate director of the UCLA Health and assistant professor-in-residence in the UCLA Health Department of Psychiatry. “Without accurate and equitable detection of suicide-related emergencies, it is difficult for suicide prevention strategies to help the populations they aim to serve.”
The findings come amid an ongoing mental health crisis in the U.S. where, between 2008 and 2022, suicide deaths among preteens increased 8% each year. The new study underscores the need to improve detection of suicidality and address these disparities, Edgcomb said.
The standard approach used by emergency departments to detect suicidality among youth uses two factors. One is the chief concern, which is the reason the patient or caregiver gives to the triage nurse for their visit. The second are diagnostic codes that clinicians use to categorize and record the symptoms and conditions they observe.
While diagnostic codes include suicidality, clinicians may code for underlying disorders or symptoms rather than for suicidality itself.
“Suicide is transdiagnostic,” Edgcomb said. “Clinicians often code for depression, trauma, they may use a symptom code like sadness instead of assigning a diagnostic code of suicidality.”
To test the effectiveness of this standard approach, Edgcomb used electronic health record phenotyping, a method that uses characteristics of a person’s medical record to figure out what disease or condition they have.
Researchers obtained electronic health data from 2,700 children and teenagers of ages 6-17 who visited emergency departments at two academic health system campuses from 2017-2019.
In addition to the information gathered in the standard approach, Edgcomb’s team also read tens of thousands of notes that clinicians made during these visits. From this, the researchers determined whether each visit was related to suicide and, if so, categorized each visit by whether the youth presented with suicidal thoughts, preparatory acts toward suicide, a suicide attempt or non-suicidal self-injury, among other categories.
Edgcomb said this resulted in a “criterion” for each of the 2,700 visits to which different detection methods could be compared. The researchers then used a series of machine learning algorithms to try to detect suicide-related visits. They compared standard methods with the more extensive phenotyping approach. Each algorithm’s prediction was compared to what clinicians determined. Additionally, Edgcomb also stratified detection by child demographics to determine how well the algorithms picked up suicide-related visits among different populations.
The comparison found that including additional data from the visit, such as laboratory tests, medications, and mental health diagnoses, improved detection and attenuated disparities among different populations and minority groups.
Across demographics, children were found to be less well detected than teenagers.
“That’s also layered with stigma, with how much clinicians accept younger children can experience suicidality,” Edgcomb said.
Boys were also less detected than females, which Edgcomb said is a problem given that men die by suicide four times more often than women.
Black and Hispanic youth were also less detected than youth from other races.
“If you’re a Black preteen presenting with suicidality, you’re much less likely to be detected than if you are a White adolescent” Edgcomb said.
The study used electronic health data from a single health care system, with evaluations of multiple emergency departments needed to further validate the findings, Edgcomb said.
Edgcomb said the findings show the need to develop improved algorithms for suicide detection, which can be aided by advancements in artificial intelligence for natural language processing that could incorporate doctor notes.
“Without a diagnostic code or chief concern related to suicide, the next clinician is having to read through medical record notes to figure out if the child has had a suicidal thoughts or an attempt,” Edgcomb said. “If we can automate that process and make suicidality more present, more relevant to the clinician, we can do a lot to improve care.”
