Suicide Risk in 8- to 12-Year-Old Children: Retrospective Review of Emergency Department Electronic Health Records

Abstract

Objective

As of 2021, suicide has increased to the third leading cause of death among children 8 to 12 years of age. Children presenting to the emergency department (ED) for suicide thoughts and behaviors (STB) are at high risk for recurrent behavioral health (BH) concerns. This study leveraged electronic health records (EHR) to identify risk factors for STB and return ED visits.

Method

EHR for 920 patients 8 to 12 years of age (N = 1,310 visits) who indicated STB or BH concerns or had a psychiatric consultation in the pediatric ED of a metropolitan hospital were reviewed from 2020 to 2023. Structured data (eg, demographics, diagnoses) were combined with clinician free-text notes.

Results

STB were frequently indicated among BH patients (65%). Ingestion was the most common injury cause from acute suicide behavior (71%). Most suicide behavior was linked to family/social triggers (56%); 21% included conflict over social media/technology. Psychosocial factors differentiated STB from other BH cases, including bullying, impulsivity, irritability, sleep, and LGBTQ+ identity (odds ratio = 2.09-6.46). Of the patients, 13% had multiple visits implicating STB. Most subsequent ED returns were within 3 months of discharge (median = 89 days). Prior psychiatric hospitalization was the strongest predictor of ED returns for subsequent STB within 1 year (odds ratio = 3.27).

Conclusion

Suicide risk is common among 8- to 12-year-old children seeking psychiatric emergency services. EHR provide critical data for understanding this public health challenge and can guide enhancement of pediatric screenings for suicide risk, including detailed assessment of symptoms (eg, impulsivity, sleep disruption) and psychosocial context (eg, bullying, LGBTQ+ identity). Helping families increase safety around ingestible lethal means remains a key point of intervention.

Diversity & Inclusion Statement

We worked to ensure sex and gender balance in the recruitment of human participants. We worked to ensure race, ethnic, and/or other types of diversity in the recruitment of human participants. One or more of the authors of this paper self-identifies as a member of one or more historically underrepresented sexual and/or gender groups in science. One or more of the authors of this paper received support from a program designed to increase minority representation in science. We actively worked to promote sex and gender balance in our author group. While citing references scientifically relevant for this work, we also actively worked to promote sex and gender balance in our reference list. The author list of this paper includes contributors from the location and/or community where the research was conducted who participated in the data collection, design, analysis, and/or interpretation of the work.

Plain language summary

Suicide risk is the third leading cause of death for children aged 8 to 12 years. This study examined electronic health records from emergency department encounters of children with suicidal thoughts or behaviors or with behavioral health concerns (N = 920 youth). Ingestion or overdose attempts were the most common injury type (71%). Most suicidal behavior was linked to family/social triggers (56%); 21% involved conflict over social media/technology. The authors also identified multiple psychosocial factors associated with suicidal thoughts and behaviors. Children with prior psychiatric hospitalizations were most at risk for repeated emergency room visits. Helping families increase safety around ingestible lethal means remains essential.

Rates of suicide ideation, self-harm, and suicide attempts have been on the rise in recent years among children.^1,2 Based on our analysis of Centers for Disease Control and Prevention (CDC) estimates (Supplement 1, available online),² suicide was the sixth leading cause of death among children 8 to 12 years of age in 2001 to 2005. As of 2021, suicide had increased to the third leading cause of death, such that it accounted for 10% of deaths in this age range.² Furthermore, analysis of CDC estimates indicates that self-harm accounts for an increasing number of emergency department (ED) visits for nonfatal injuries among children 8 to 12 years of age, showing a 10-fold increase over the past 2 decades.²

The ED is often the first point of contact for youth to receive psychiatric care for suicide risk.³ ED visits associated with nonfatal self-harm for 8- to 12-year-olds in 2021 were estimated to cost an accumulated $83 million in medical expenses and $150 million in quality-of-life costs over the following year. Among adolescents (∼12-18 years of age) receiving inpatient or ED psychiatric care, ∼10% will attempt suicide and 30% will exhibit suicidal behaviors within 3 to 6 months of discharge.4, 5, 6, 7 Yet, there is far less research on preadolescent suicide risk, and thus there is a critical need to understand the clinical presentation of suicidal thoughts and behaviors (STB) in this age group and to identify risk factors for recurrent STB after discharge. Increased understanding of the suicide risk in this population may lead directly to the implementation of actionable preventive intervention approaches.

Electronic health records (EHR) are a large and powerful data source to understand childhood suicide.8, 9, 10 Yet, identifying suicide risk in EHR can be complex, especially for pediatric patients. STB may be indicated across different EHR sources (eg, chief complaint, diagnostic codes, structured assessments, clinician free-text notes). Examining both diagnostic codes and chief complaints can be effective at uncovering youth with self-injurious thoughts and behaviors, whereas either alone might underestimate these rates.⁹ Prior EHR work in adults and adolescents has sought to improve longitudinal prediction of future suicide behaviors and attempts.11, 12, 13 Large-scale EHR data across ages (ie, nonspecific to children) have been examined to predict suicide deaths within 90 days of ED or outpatient visits,^11,14 and often implicate sociodemographics, psychiatric diagnoses, and medication factors in predicting suicidal behavior.¹² Yet, there is limited EHR research to understand suicide risk in children.8, 9, 10

The current study examined suicide risk among children 8 to 12 years of age presenting with behavioral or psychiatric concerns to the pediatric ED of a large tertiary care hospital in a major metropolitan area. EHR were used to identify the rates of STB among children in the ED, and to examine sociodemographic and clinical factors that related to STB presentation relative to other behavioral health (BH) concerns, building on prior work.^1,15, 16, 17, 18 Clinician notes were probed for the presence of psychosocial risk factors of interest, such as family and interpersonal issues or prior and outside psychiatry care. We hypothesized that patients in the ED with indication of STB would be more likely to be female, of low socioeconomic background, and experiencing interpersonal stressors than patients receiving care for other BH concerns. We anticipated that cutting and ingestion would be the most common forms of injury from among current suicide behaviors.⁸ Furthermore, this study examined the rate of return visits to the ED in this population for BH and STB-related concerns, and probed predictors of return post discharge within 1 year of discharge. Building on prior work,⁸ we hypothesized the following: (1) an ED return rate of ∼10% (which is less than that observed in prior studies of adolescents), and (2) that more severe STB would predict greater risk for return.

Method

Overview

This was a retrospective study examining EHR from pediatric patients visiting the ED at one major metropolitan hospital between the local initiation of Epic Systems on February 1, 2020, and December 31, 2023. This study was approved by the local Institutional Review Board (Protocol AAAU8923). All analyses were conducted in R version 4.3.0 (R Project for Statistical Computing).¹⁹

Sample

Records for all pediatric patient (<18 years of age) visits were extracted based on the presence of a psychiatric consultation, chief complaint description, and/or an International Classification of Diseases, Tenth Revision (ICD-10) code for suicide-related concerns or mental and behavioral health diagnoses (F01-F99, R45, T14.91, T36-T50, T65.92, X78, W33, Z72.89, Z86.5). Records generally indicated one ICD-10 code per case. This yielded 3,696 unique patients with 4,818 visits. Data were filtered to include only patients with at least one visit between 8 and 12 years of age (additional visits outside this age range were retained if a patient also had at least one other visit between ages 8 and 12 years). This yielded 920 unique patients with 1,310 visits for analysis. Figure 1 summarizes the flow of patients and results, and Supplemental Material, available online, provides workflow details.

Figure 1.

Study Overview

Note:This figure provides an abbreviated summary of study methods and results. Selection of electronic health records (EHR) and identification of behavioral health (BH) and suicide thoughts and behaviors (STB) groupings are outlined. A schematic EHR is displayed in panel 2. Highlights regarding current suicide behavior (SB) cases are summarized in panel 3. Highlights regarding group differences are shown in panel 4 (seeTables 1and2for details). Factors related to return to the emergency department (ED) are shown in panel 5 (seeTable 3for details).

STB Characteristics

Visits were grouped based on the presence of suicide-related concerns using an ICD-10 code, structured assessments, and/or free-text clinician notes. Clinical notes (free-text input and standardized flowsheet information) from the BH team were the main source of information, as well as initial medical provider notes. Patients presenting for a current intentional self-harm, suicide attempt, or broader suicide behavior (SB) were identified based on ICD-10 diagnosis code (eg, T14.91) and/or indication in clinician notes of aborted, interrupted, or actual attempts, intentional ingestion, and related terms.

Additional cases indicating prior SB were identified by Columbia–Suicide Severity Rating Scale (C-SSRS)²⁰ item 6, the Suicide Assessment Five-Step Evaluation and Triage (SAFE-T),²¹ and/or behavioral health notes (eg, “Hx of suicide attempt via OD”). The C-SSRS Screen Version was added to the EPIC flowsheet in November 2021 and was administered to every patient over the age of 6 years at triage.

Cases indicating recent/current suicidal ideation (SI) were detected based on the ICD-10 code (R45.851), notes (eg, “pt reports thoughts of death”), past-month C-SSRS (yes on items 1-5), SAFE-T (SI in the past months), or the mental status examination (MSE). Report of lifetime (without recent) SI indicated on the SAFE-T was not examined. Patients with current or prior SB or recent/current SI (STB group) were compared to those with other non-STB BH causes. Additional analyses probed characteristics of youth brought to the ED for current SB. The accuracy of identification or labeling of all current SB visits and ∼20% of other visits were checked by 3 coauthors (DP, JSK, JG).

Predictors

Sociodemographic information was extracted from structured EHR fields, including age, sex, race (categorized as White, Black, and other identities; <10 participants identified as Asian and <10 as American Indian, Native Hawaiian, or Pacific Islander), Hispanic ethnicity, and use of private insurance. ED stay time was calculated as the number of hours between arrival and discharge. Antidepressant medication use was detected based on an existing medication variable or in notes, including tricyclic antidepressants, selective serotonin reuptake inhibitors (SSRI), serotonin–norepinephrine reuptake inhibitors (SNRI), and monoamine oxidase inhibitors (MAOI) (full list in Supplemental Material, available online). Notes were also screened for mention of comorbid attention-deficit/hyperactivity disorder (ADHD), oppositional defiant disorder (ODD), or conduct disorder (CD).

Home ZIP codes from EHR were used to derive whether the patient lived close (<5 miles) to the local ED. Home ZIP codes from EHR were also linked to Social Deprivation Index (SDI)²² scores based on the 2015 to 2019 American Community Survey from the US Census Bureau (https://www.census.gov/programs-surveys/acs/). The SDI is available online as a validated resource to quantify neighborhood levels of social determinants of health in a given geographic area. SDI scores range from low to high (1-100), with higher scores indicating more deprivation. The score for New York County (Manhattan) is 82 and for ZIP code 10032 is 97.

Exploratory indicators were ascertained from keyword text searches (ie detection of regular expressions in R) of terms evident in clinical notes (details and sample code in Supplemental Material, available online). Binary variables were created to indicate the absence or presence of certain terms, given prior conceptual links with mental health and suicide. Keywords were derived via both inductive and deductive methods; lexicons were initiated with a priori terms of interest23, 24, 25, 26 and then expanded and refined through review of EHR and a consensus of authors. Risk factor identification code was iteratively confirmed and enhanced by 3 coauthors (DP, JSK, JG). Risk factors included text related to the following: abuse/neglect (eg, “sexual assault,” “child protective services”), aggression (eg, “aggressive outburst”), bullying (eg, “cyberbullying,” “teasing”), family concerns (eg, “foster care,” “custody dispute”), impulsivity (eg, “poor impulse control”), irritability (eg, “moody,” “DMDD”), LGBTQ+ identity (eg, “non-binary,” “bisexual”), existing outpatient care (eg, “weekly therapy”), prior hospitalization (eg, “previous inpatient stay”, “in residential treatment”), psychosis (eg, “hearing voices”), school concerns (eg, “failing grade,” “IEP and 504 plan”), sleep problems (eg, “insomnia”), and social media/ technology usage or concerns (eg, “Instagram,” “phone privileges revoked”). Substance use (eg, “vaping”) was explored but was relatively infrequent (3% of visits) and thus was not included in statistical analyses. Identification of risk factors was agnostic to reporter (child, parent, clinician observation) and timeframe (current vs past), as this was not always specified in notes.

Data Analyses

For primary patient-level analyses (n = 920), one visit was selected per individual to examine their first STB visit (or earliest visit if they never presented with STB). Patient characteristics were compared across groups (STB vs BH) using the scipub package²⁷ with a χ² and t test for categorical and continuous characteristics, respectively, and accompanying effect sizes. Logistic regressions were used to examine multiple factors simultaneously in relation to STB group and likelihood of ED return for STB within 1 year of discharge. Adjusted odds ratios (aOR) were presented for logistic regressions models; unstandardized aOR were presented for continuous predictors (eg, age in years). Stepwise selection was used to select predictors for inclusion in models predicting ED return (stats::step R function with both backward and forward selection–based reduction in Akaike information criterion [AIC]). This selection was confirmed using a LASSO regularized regression approach (glmnet R package²⁸; alpha = 1, 100-fold cross-validation). Model fit was assessed using the performance package.²⁹ Cox proportional hazards regression models were used to explore the time-to-return in a survival analysis framework.³⁰

Results

Overview

Leveraging pediatric ED EHR from 2020 to 2023, a total of 920 patients (8-12 years of age) were identified with BH and/or STB at 1,310 visits (Table 1, Figure 1). A total of 629 patients (850 visits) indicated STB (recent/current SI and/or prior SB). Of these, 32 patients (34 visits) indicated a current SB event (eg, were brought to the ED following a suicide attempt). The remaining 291 patients (460 visits) included other non-STB BH causes. Overall, the sample was 34.67% White, 25.33% Black, and 56.09% Hispanic, which was roughly representative of Washington Heights, New York (ZIP code 10032) (data.census.gov; 2022 American Community Survey 5-year estimates: 21.5% White, 16.9% Black, 13.1% multi-racial, 65.7% Hispanic ethnicity).

Table 1.

Sociodemographic and Clinical Characteristics by Suicide Risk

Characteristic	Behavioral health (n = 291)	Suicide risk (n = 629)	Group difference	p	Effect size
Age, y	10.25 (1.45)	10.78 (1.41)	t = 5.17	<.001	d = 0.37
Sex, female	141 (48.45%)	357 (56.76%)	χ2 = 5.19	.02	OR = 1.40
SDI	87.47 (22.7)	91.76 (16.72)	t = 2.82	.005	d = 0.21
Distance to ED <5 miles	203 (69.76%)	531 (84.42%)	χ2 = 25.61	<.001	OR = 2.35
Race, White	107 (36.77%)	212 (33.7%)	χ2 = 0.70	.40	OR = 0.87
Race, Black	73 (25.09%)	160 (25.44%)	χ2 = 0.00	.97	OR = 1.02
Ethnicity, Hispanic or Latinx	165 (56.7%)	351 (55.8%)	χ2 = 0.03	.85	OR = 0.96
Insurance, private	61 (20.96%)	172 (27.34%)	χ2 = 3.96	.05	OR = 1.42
Antidepressant medication	17 (5.84%)	122 (19.4%)	χ2 = 27.45	<.001	OR = 3.87
ED stay time, h	9.01 (14.56)	16.25 (23.49)	t = 5.71	<.001	d = 0.37
Admitted	60 (20.62%)	196 (31.16%)	χ2 = 10.49	.001	OR = 1.74
Return to ED	27 (9.28%)	170 (27.03%)	χ2 = 36.20	<.001	OR = 3.62

Note: Sample characteristics are summarized and compared for behavioral health cases without indication of suicide risk vs cases with indication of suicide risk. One emergency department (ED) visit is included per patient (first visit with suicide risk or first visit if never indicating suicide risk). Table S1, available online, shows these differences across all visits (allowing for multiple visits per person). Continuous variables (age, SDI, stay time) are summarized with their group mean and SD. Group differences are tested via t test with their accompanying Cohen d effect size. Categorical variables are summarized by count and subgroup percentage. Group differences are tested via χ² test and accompanying odds ratio. Significant group differences (p < .05) are shown in boldface type. N = 32 missing SDI scores. ED = emergency department; OR = odds ratio; SDI = Social Deprivation Index (rated from 0 to 100).

Across all cases, the most frequent ICD-10 codes used were for SI (R45.851; 274 visits), aggression or behavioral concerns (R46.89; 246 visits), evaluation for psychiatric services (Z00.8; 92 visits), and depression (F32; 81 visits). Only 8 visits were coded for a suicide attempt (T14.91), and no visits included codes for firearms accidents or injury (W32-W34, X72-X74, Y22-Y23). There were 42 visits with codes for self-harm or nonsuicide self-injury [NSSI] (R45.88, Z72.89).

Current Suicide Behavior

A total of 32 patients (34 visits) were brought to the ED for current SB (n = 24 [71%] female). Of those who completed the C-SSRS, all indicated moderate or high suicide risk. These cases mostly included ingestion incidents (71%; predominantly over-the-counter or prescribed pills), as well as fewer instances involving knives, hanging, self-strangulation, or walking into traffic. From clinician descriptions, these cases involved a range of severities. For example, some children indicated a clear attempt to die but ingested a nonlethal number of pills, whereas others used a more dangerous method but indicated unclear or impulsive reasoning. Most patients (76%) were admitted for BH or inpatient services at this visit for SB (24% discharged home to outpatient care).

Most patients presenting with SB (56%) reported a psychosocial stress trigger (eg, family conflict, bullying, school), whereas a clear trigger was not captured or logged for the remaining patients (44%). Of the cases, 21% also indicated technology-related triggers (eg, parent revoking access to phone, computer, or video games). Twenty SB cases (59%) indicated a history of prior suicide behavior. For 41% of these patients, this was their only ED visit in the current dataset.

Current Suicide Ideation and Prior Suicide Behavior

An additional 597 patients (816 visits) implicated SI and/or prior SB (but were not brought to the ED for acute SB). Of these children, 36% reported SI and prior SB, whereas 58% reported SI but not prior SB. Of 850 STB visits, 287 (34%) were denoted by their clinician as moderate, high, or imminent risk for suicide or self-injury. Of those completing the C-SSRS, 59% of prior SB cases reported this on CSSRS item 6; the remaining cases were identified via clinical notes. Of patients with recent/current SI, 35% of these patients had a primary ICD-10 code for SI (R45.851), 30% were identified by the C-SSRS, 11% by the MSE, and 24% by other clinical notes.

At their earliest visit, compared to patients presenting for BH (n = 291), patients with STB (n = 629; 32 current SB + 597 SI or prior SB) tended to be older (Cohen d = 0.37), were more likely to be female (odds ratio [OR] = 1.40), were more likely to be admitted to the hospital or inpatient pediatric or psychiatric unit; OR = 1.74), stayed in the ED longer (Cohen d=0.37), tended to live close to the ED (OR = 2.35) (Figure S1, available online), were more likely to be on antidepressant medications (OR = 3.87), and were more likely to have multiple ED visits (OR = 3.62) (Table 1; Table S1, available online).

Examining keyword matches in clinical notes (Table 2; Table S2, available online), compared to non-STB BH visits, visits with STB were more likely to indicate prior hospitalization(s) (OR = 2.72) and existing outpatient care (OR = 4.10), family or school concerns or abuse/neglect (OR = 2.09-2.93), irritability, impulsivity, or sleep concerns (OR = 2.09-3.30), bullying or social media/technology–related text (OR = 3.19-3.93), and LGBTQ+ identity (OR = 6.46). Logistic regression models combined these structured (eg demographics) and unstructured (clinical notes) data factors as simultaneous predictors of STB group, with most factors remaining significant (Table S3, available online).

Table 2.

Risk Factors in Clinical Notes by Suicide Group

Characteristics indicated in clinical notes	Behavioral health (n = 291)	Suicide risk (n = 629)	Group difference, χ2	p	Effect size, OR
Abuse/neglect	38 (13.06%)	191 (30.37%)	30.96	<.001	2.90
ADHD	42 (14.43%)	127 (20.19%)	4.02	.04	1.50
Aggression	70 (24.05%)	175 (27.82%)	1.26	.26	1.22
Bullying	40 (13.75%)	212 (33.7%)	38.85	<.001	3.19
Family concerns	54 (18.56%)	252 (40.06%)	40.49	<.001	2.93
Impulsivity	33 (11.34%)	187 (29.73%)	35.97	<.001	3.30
Irritability	70 (24.05%)	290 (46.1%)	39.69	<.001	2.70
LGBTQ+ identity	4 (1.37%)	52 (8.27%)	15.35	<.001	6.46
ODD/CD	19 (6.53%)	80 (12.72%)	7.31	.007	2.08
Outside mental health care	58 (19.93%)	318 (50.56%)	75.95	<.001	4.10
Prior hospitalization	25 (8.59%)	128 (20.35%)	19.00	<.001	2.72
Psychosis symptoms	17 (5.84%)	62 (9.86%)	3.59	.06	1.76
School concerns	65 (22.34%)	236 (37.52%)	20.15	<.001	2.09
Sleep disruption	57 (19.59%)	212 (33.7%)	18.49	<.001	2.09
Social media/ technology use or concerns	22 (7.56%)	153 (24.32%)	35.22	<.001	3.93

Note: Potential risk factors from clinical notes are summarized and compared for behavioral health cases without indication of suicide risk vs cases with indication of suicide risk. One visit is included per patient (first visit with suicide risk or first visit if never indicating suicide risk). Table S2, available online, shows these differences across all visits (allowing for multiple visits per person). All variables indicate the absence vs presence of related text and are summarized by count and subgroup percentage. Group differences are tested via χ² test and accompanying odds ratio. Significant group differences (p < .05) are shown in boldface type. ADHD = attention-deficit/hyperactivity disorder; CD = conduct disorder; ODD = oppositional defiant disorder; OR = odds ratio.

Visit Timings Exhibited Temporal and Seasonality Effects

Compared to BH visits with no STB indications, STB visits were more frequent during day or evening hours (9 am to 9 pm vs late night to morning), especially on weekdays (vs weekends), and during fall/winter/spring (September-May) vs summer (Figure S2, available online).

Repeat Visits for STB

Among the 920 patients, 723 were seen for one ED visit, whereas 197 (21%) had multiple ED visits for BH or STB (range, 2-16 visits) (Figure S3, available online). Of these, 119 patients (13%) had ED multiple visits implicating STB. The median time to return to the ED for a subsequent STB visit was 89 days (between all successive events; range, 1-1,274 days). Among returns for STB, 25% were within 30 days, 50% within 90 days, and 89% within 1 year of discharge (Figure S4, available online).

Two stepwise logistic regressions (Table 3; Table S4, available online) tested multiple simultaneous factors in relation to whether patients returned for a subsequent STB visit within 1 year of discharge. Sociodemographic and other risk factors were included or removed based on AIC change (forward and backward selection) to refine core indicators of interest. Among all 1,310 visits, 258 visits (20%) were linked to a subsequent STB return within 1 year. Greater likelihood of a subsequent ED visit for STB was related to race, female sex, living closer to the ED, antidepressant medications, family concerns, sleep concerns, and prior hospitalization (including a prior visit to the local ED). Results were largely conserved when examining the subset of 850 STB visits. Both models exhibited a value of R² = 0.14 (Tjur). Hosmer–Lemeshow goodness-of-fit tests (10 bins) showed that both models fit well (p > .12). All collinearity was low (variance inflation factor [VIF] < 1.30), and simulated residuals were uniformly distributed (p > .75). The same significant predictors were identified via least absolute shrinkage and selection operator (LASSO) regression selection. Confirmatory box–Cox survival analyses examined prior hospitalization as the strongest predictor identified above. Prior hospitalization was associated with faster return to ED for STB (hazard ratio = 5.28, z = 13.83, p < .001) (Figure S5, available online), particularly in the first 9 months after discharge.

Table 3.

Stepwise Logistic Regression Results Related to Return Emergency Department Visits Within 1 Year

Characteristic	All visits (n = 1,310), 20% return for STB within 1 y			STB visits (n = 850), 13% return for STB within 1 y
	aOR	z	p	aOR	z	p
Race, Black	0.90	–0.47	.64	1.07	0.26	.79
Race, other identities	1.49	2.26	.02	1.57	2.19	.03
Distance to ED <5 miles	2.85	3.96	<.001	2.62	2.99	.003
Antidepressant medication	1.76	3.17	.002	1.78	2.80	.01
Sex, female	1.49	2.50	.01	—	—	—
Private insurance	1.29	1.46	.15	—	—	—
ODD/CD	1.53	1.99	.05	—	—	—
Family concerns	1.58	2.96	.003	1.50	2.32	.02
Sleep disruptions	1.40	2.14	.03	1.37	1.74	.08
Outside care	1.28	1.63	.10	1.38	1.79	.07
Prior hospitalization	2.61	5.86	<.001	3.27	6.09	<.001

Note: Stepwise regression results are presented for two models examining factors related to whether patients returned to the ED for STB within 1 year of discharge. The first model examined all visits. The second model examined the subset of visits with STB. Variables in Tables 1 and 2 were included as potential variables, and only those that improve model fit (Akaike information criterion) were automatically retained or removed (forward and backward). Variables that were tested but removed through stepwise selection included age, ethnicity, admission, aggression, bullying, irritability, impulsivity, LGBTQ+ identity, school concerns, and social media. Race was a 3-level variable, with White as the statistical reference. Prior hospitalization included previous visits to the local ED. Significant associations (p < .05) are shown in boldface type. Table S4, available online, shows raw odds ratios between visits with vs without a subsequent ED return for STB within 1 year. aOR = adjusted odds ratio; ED = emergency department; CD = conduct disorder; ODD = oppositional defiant disorder; STB = suicide thoughts and behaviors.

Discussion

Suicide is an understudied concern for children despite increasing rates in the United States in recent years.¹ The ED is a critical, and often first, point of contact for initiating psychiatric evaluation and care. The period after ED discharge can be a high-risk period for youth to experience re-emergence of STB. Understanding suicide risk presentation and recurrence among children in the ED is an important step toward understanding this public health problem to improve screening and subsequent interventions.

This study conducted a retrospective review of 1,310 EHR of 920 pediatric patients who presented to a large metropolitan pediatric ED for behavioral health concerns and/or who required a psychiatric consultation. Children largely presented from Washington Heights, Harlem, and other areas of upper Manhattan, New York. A majority (65%) of 8- to 12-year-old patients presenting to the ED for behavioral health concerns disclosed thoughts of suicide and prior suicidal behaviors. Approximately 20% of children had multiple visits to the ED for behavioral health concerns within the timeframe of review, with most returns occurring within 3 months of ED presentation. Our data considered only one medical center and thus potentially underestimates the care needed by these youth (ie, at other medical facilities). That said, 23% of cases had mention of a prior history of hospitalizations, inpatient care, or residential treatment in their clinical notes.

Although many visits were coded in structured data for SI (ICD-10: R45.851), additional suicide risk was identified even among those presenting for other concerns, such as aggression or panic attacks. Having multiple assessments in our system helped to ensure ascertainment and documentation of STB (ie, the C-SSRS, SAFE-T, MSE). We also note that current suicidal behavior was often not included as the primary ICD-10 code for these cases. Different hospitals and EHR systems may configure ICD coding differently (eg, the number of allowable codes or separating primary vs secondary concerns). This is important to consider in other work with large-scale EHR datasets, as reliance on ICD codes only can lead to misidentification of STB risk, without examination of clinical notes.

We identified 32 patients presenting to the ED for suicide behaviors. This subset focused on children who were brought to the ED directly for the suicide-related incident; it should be noted that other children reported recent suicide behaviors (eg, within the past week) but came to the ED later for other presenting complaints (eg, an aggressive outburst the next week). The majority of current SB were identified based on an ICD-10 code for attempt (T14.91XA) or poisoning for intentional self-harm (eg, T50.902A). Some cases were denoted by ingestion with unknown intent, and a suicidal intent was later determined through psychiatric evaluation and described in clinical notes. In this sample, ingestion was the most frequent method, primarily medications prescribed to the parent or child or an OTC medication in the home (eg, sleep aids), as seen in prior work.^1,8,31 Overall, these cases varied in their potential lethality and the child’s insight regaring the event. Home sanitation education is key to mitigate risk of future self-harm or suicide attempts. Understanding the prevalence of methods can help target interventions specific to different geographic and sociodemographic areas. Research has shown that parent education programs about safe storage of medications can increase efforts to reduce children’s access to lethal means.³² Psychoeducation on home sanitation is key to mitigate risk of future harm to self or suicide attempts.³³ This may include lock boxes for medications, sharps, and cleaning supplies Increasing knowledge about lethal means safety, particularly for medications, can be a key intervention point. It should be noted that none of the cases in our sample implicated firearm injury, and very few of all cases indicated access to firearms in the home (<10). That said, rates of self-harm and suicide by firearm are substantially higher among adults than among children³⁴ and can be a critical intervention point (eg, with “Lock & Protect” ED programs),³⁵ particularly in other states that exhibit less strict firearm legislation and higher rates of firearm injury compared to New York.³⁶

Rates of suicide risk have been increasing among minoritized populations, but the degree of sociodemographic differences has varied in prior work.^37,38 In this diverse patient population, we did not observe significant differences in the presence of STB by race, ethnicity, or insurance status. Youth reporting STB were also more likely to live near the ED, and their home ZIP codes related to higher Social Deprivation Index scores. This is consistent with prior work suggesting that, for youth less than 18 years of age, greater SDI related to higher risk of suicide ideation.³⁹ Furthermore, girls were more likely to indicate STB and were more likely to exhibit multiple or return visits. This is significant to consider, as gender differences in depression and suicide attempts are most often cited as important later in adolescent development,⁴⁰ although there is also growing evidence for gender differences in STB emerging in earlier childhood.⁸ LGBTQ+ identity was not systematically logged in EHR (eg, in a structured input field), but mentions of sexual or gender minority identity in clinical notes were almost exclusively among cases with STB relative to non-STB BH cases. This is also in line with work showing greater and earlier LGBTQ+ identification and consistent mental health disparities for LGBTQ+ individuals.41, 42, 43

Notably, children expressing STB tended to require a higher level of care. Compared to children with BH concerns but low suicide risk, children indicating STB tended to stay in the ED approximately 1.5 times as long. This can be both costly⁴⁴ and distressing. These children were also about 3 times as likely to be taking antidepressants and 4 times as likely to be engaged in outside care (eg, already seeing a community psychiatrist). With the current data, we are unable to ascertain the relative timing of initiation of antidepressant use and first onset of suicide risk to disentangle any contribution of medication effects. It is possible that engagement in outside care may also contribute to ED admission (eg, a therapist may encourage a parent to bring their child to the ED after discussion of suicide risk), and/or these youth may exhibit a higher level of symptoms that escalate to STB.

Building on known risk factors and examination of clinical notes, we developed a text mining (lexicon-based) approach to identify exploratory indicators that are not typically logged in structured EHR data fields but that may inform suicide risk. Thus, we were able to flag the presence of clinician reports for a variety of factors (eg, impulsivity, sleep problems, bullying). Critically, this approach relies on both the patient or family to report the risk factor and then the clinician to log this in the notes. Thus, these variables likely include many false-negative results when a potential risk factor was not discussed or logged. More advanced natural language processing and machine learning approaches continue to be developed and can be leveraged to understand complex EHR.^9,10,13 Such approaches are increasingly useful as sample sizes increase and may yield more abstract characterization (eg, sentiment analysis, or continuous severity ratings). As suicide research in children is still relatively nascent, we first aimed to identify the absence or presence of clearly interpretable risk factors to lead into more complex, complementary approaches in the future (eg, machine learning).^9,23

Compared to children with BH concerns but low suicide risk, children with STB were more likely to have mention of nearly all examined risk factors in free-text clinical notes, including impulsivity, irritability,⁴⁵ and sleep problems.⁴⁶ Children with STB were twice as likely as likely to be described as impulsive (or as having poor impulse control). Impulsivity was expressly documented in the MSE and was thus more clearly indicated than other risk factors. That said, research has suggested that a more nuanced characterization of impulsivity may aid in suicide risk prediction: for example, that negative urgency (a tendency to act rashly when experiencing negative emotions) is a particularly problematic domain of impulsivity.^47,48

Increasing research has examined associations between social media, technology, and mental health in youth.^49,50 Among current SB cases, 15% noted a technology-related trigger (including but not specific to social media), primarily as a source of family conflict or as disruption of routines (eg, family conflict after restricting smartphone access). Indication of social media and/or technology-related words were more common across STB case notes; however, more refined assessment is needed to separate effects of technology itself from family-related factors.

Around 20% of children returned to the local ED for multiple BH or STB-related visits, half of whom presented for more than 2 visits in our 3-year dataset. Predicting recurrence of suicide events (eg, preparatory behavior, attempts, hospitalization for suicide) is a difficult and complex goal. We found several significant predictors of return ED visits for STB. Prior history of psychiatric hospitalization was the most robust predictor, relating to 5 to 6 times higher odds of return to the local ED for STB. Models can likely be improved through more in-depth, prospective assessment and more comprehensive understanding of suicide behaviors post discharge that do not lead to a return ED visit.

EHR represent a powerful and underused data source to understand childhood suicide risk, yet several key limitations should be noted for this work. First, all EHR research is inherently limited by the data available in existing records, which may be limited by biases in assigning ICD codes, patient underreporting, and other factors that may lead to false-negative results in identifying STB risk. Free-text notes are also limited, as specificity about the timeframe of symptoms, medication use, and other factors may not be readily available. We aimed to leverage a combination of structured ICD codes, EHR checklists, and free-text notes to provide a more robust assessment of risk. That said, the C-SSRS was implemented in the EHR part way through our analysis timeframe, which limited the sample size of cases with this assessment. Our results build on a growing understanding of childhood STB risk that can be further explored by means of intensive prospective studies with greater temporal precision. Prospective studies can also help to address the limitation of available data about recurrent STB that we likely underestimate, as we have data only on return ED visits. Importantly, this study provided a nuanced examination of EHR from a single medical center; future work can expand on this through multi-center analysis to ensure generalizability. Particularly, our keyword coding may include site-specific language that is not applicable to other medical centers. Our goal herein was to compare cases with STB to other, non–STB-related BH visits, but this can be expanded on in the future to examine other nonpsychiatric control groups. Our dataset also included the peak of the COVID-19 pandemic, which may affect some effects in the data that should be replicated with newer ED data in the future.

Suicide risk is an increasing concern among children, and the ED is a critical point of contact to identify and to intervene with high-risk youth. Substantial efforts have been made to address the rising crisis of suicide among adolescents; however, this is understudied in children. Importantly, a nuanced approach to identify and understand risk is needed, because ICD codes are helpful but not definitive in identifying at-risk cases.⁹ Furthermore, standardized measures (eg, the Patient Health Questionnaire–9 [PHQ-9], Ask Suicide-Screening Questions [ASQ], Columbia–Suicide Severity Rating Scale [C-SSRS]) have been implemented at many clinical sites and can be helpful in gathering concrete risk factors to enhance risk assessment. At the same time, additional research is needed to tailor assessments and prediction of future STB risk for children.¹⁶ Currently, a history of STB and prior hospitalization are key indicators of elevated risk. Yet, other potential risk factors are often not routinely assessed or logged in structured data fields (eg, LGBTQ+ identity, bullying experiences). Prospective assessment of high-risk children in the ED can help to identify risk and protective factors and points of intervention. These efforts may be combined to develop risk calculators for use in clinical practice to estimate suicide risk, and accordingly, to help support clinicians in determining the appropriate ED management and discharge plans.

CRediT authorship contribution statement

David Pagliaccio: Writing – review & editing, Writing – original draft, Visualization, Validation, Supervision, Project administration, Methodology, Investigation, Funding acquisition, Formal analysis, Data curation, Conceptualization. Jaclyn S. Kirshenbaum: Writing – review & editing, Writing – original draft, Methodology, Data curation, Conceptualization. Julia Greenblatt: Writing – review & editing, Validation, Data curation. Megan M. Mroczkowski: Writing – review & editing, Conceptualization. Lauren S. Chernick: Writing – review & editing, Methodology, Investigation, Conceptualization. Peter S. Dayan: Writing – review & editing, Project administration, Methodology, Conceptualization. Randy P. Auerbach: Writing – review & editing, Methodology, Investigation, Funding acquisition, Conceptualization.