Clinical Profiles Associated With Deliberate Self-Harm in Preadolescent Children

Abstract

Objective

Children who self-harm are at high risk for suicide attempts later in life, yet little is known about the clinical profiles of young children who self-harm. The primary aim of this study is to characterize clinical profiles associated with self-harm in preadolescent children to inform risk recognition and prevention efforts.

Method

A retrospective, population-based cohort analysis of children aged 5 to 11 years who presented with self-harm in a medical setting from 2010 to 2020 (N = 878) was conducted using Ohio Medicaid data. Children were followed for 1 year after the initial self-harm event to determine subsequent self-harm events.

Results

Of the 878 children with initial self-harm, 116 children (13%) self-harmed again within the following year. Latent class analyses revealed 3 distinct clinical profiles. Children in class 1 tended to be male, typically presented in mental health settings with both externalizing and internalizing disorders, and commonly reported suicidal ideation. Children in class 2 were largely female, presented with internalizing disorders, and often expressed suicidal ideation. Children in class 2 were less likely than those in class 1 to have a history of mental health treatment but were equally at risk for repetitive self-harming behaviors. Children in class 3 tended to be younger and from non-metropolitan areas, with fewer known mental health conditions.

Conclusion

Study findings show that children who self-harm have distinct clinical profiles that are associated with elevated risk of future self-harm. Findings underscore the importance of comprehensive risk assessment to guide clinical decision making for young children who self-harm.

Plain language summary

This study analyzed the medical profiles of children ages 5 to 11 who presented with self-harm from a medical setting from 2010 to 2020. The authors found children who self-harm have distinct clinical profiles associated with risk for future self-harm. Findings underscore the importance of comprehensive risk assessment to guide clinical decision making for young children who self-harm.

Child suicide is a major public health problem. Although it is rare before 10 years of age, suicide is the sixth leading cause of death among children aged 5 to 11 years in the United States.¹ In 2020, for the first time, suicide became the 10th leading cause of death for children aged 5 to 9 years.¹ Suicidal ideation (SI) and suicide attempt (SA) are surprisingly common in childhood, with an estimated 15% of children having experienced SI and 2.6% having attempted suicide by 11 years.^2,3 Estimates are even higher in clinical samples aged 3 to 7 years, with SI reported by 19% and self-harm by 3.5%.^4,5

Deliberate self-harm, defined as both non-suicidal self-harm behaviors and suicide attempt (SA), is a strong risk factor for suicide,⁶ and children who self-harm often experience persistent SI and self-harm across adolescence and into adulthood.^7,8 Moreover, risk for SA is higher for individuals who begin self-harming at younger ages and for those with repeated incidents of self-harm.^8,9 Only a few studies have examined why some children have repeated incidents of self-harm and others remit,^4,8 with most research focused on repeated self-harm in adolescents.^10,11

Identifying children at risk for self-harm is challenging, as there are no robust singular predictors of risk, and as self-harm is a transdiagnostic behavior with myriad risk factors. Within the overall population of children who self-harm, there are likely distinct subgroups of individuals who share similar clinical profiles in terms of psychopathology and social and contextual factors. A better understanding of these clinical profiles may help to identify which children are at risk for self-harm, and could inform comprehensive risk assessments about potential mechanisms that could contribute to repeated self-harm incidents.

Risk factors for child suicide include psychopathology, family conflict, unstable caregiving, maltreatment,^12,13 and residence in poorer communities.¹⁴ Risk factors associated with SI and self-harm in young children include attention-deficit/hyperactivity disorder (ADHD), impulsivity, internalizing psychopathologies such as anxiety and depression,^4,15,16 and intellectual and developmental disorders.¹⁷ Some work suggests that externalizing disorders such as ADHD may be more prominent risk factors for preadolescent suicide relative to depression, the most relevant psychopathological risk factor for adolescent suicide.¹³

Much of the existing literature has focused on independent predictors of suicide and self-harm in children; this is problematic, given that some risk factors, such as ADHD, occur in about 10% of the child population,¹⁸ and lifetime prevalence of anxiety by age 11 years is about 25%.¹⁹ Self-harm has been associated with combinations of risk factors, including internalizing with externalizing disorders,20, 21, 22 depressive symptoms with irritability,^4,23 and maltreatment in girls with ADHD.²⁴ These combinations of risk factors may represent clinical profiles that better explain why some children tend toward self-harm, with these risk factors contributing independent and interactive effects. Singularly, a given factor may not confer significant risk for self-harming behavior including SAs, but a combination of risk factors may represent a more complex diathesis for self-harming tendencies.

Very little is known about the clinical profiles of preadolescent children who go on to self-harm, and whether certain profiles are more likely to be associated with subsequent self-harm incidents. The present study examined profiles based on psychiatric and contextual risk factors among children aged 5 to 11 years with a self-harm event. We focus on this age group specifically because children who begin self-harming before adolescence are at much higher risk for subsequent self-harm events and SA compared to those with later onset of self-harm.^8,9 Understanding how individual risk factors may combine and interact in children who self-harm and in those who persistently self-harm may inform early intervention efforts that could prevent self-harm and suicide.

Method

A retrospective cohort design was used to identify which clinical profiles were most at risk for repeated self-harm events in the year following onset of self-harm. The study population included all children aged 5 to 11 years with a self-harm event between calendar years 2010 and 2020 (n = 1,323) who were continually enrolled in Medicaid for >10 months before and after the index claim for self-harm (n = 938). The index self-harm event was the first claim related to self-harm and was defined using International Classification of Diseases (ICD) codes (Supplement 1, available online; Table S1). The 1-year period preceding the index self-harm was chosen to allow for sufficient diagnoses with varied symptom duration diagnostic criteria to be captured. The 1-year period following the index self-harm event was chosen because this is a high-risk period for repeated self-harm events.¹⁰ Excluded were children who (1) died by suicide with no other medically documented self-harm (n = 6); (2) had no documented service use claims in the year before self-harm (n = 11); and (3) who were missing data on race or ethnicity (n = 43). The final sample included 878 children (see Supplement 1, available online, and Figure S1 for a sampling methodology flow chart). This study was approved by the institutional review board of Nationwide Children’s Hospital.

Data for this study were abstracted from Ohio Medicaid claims and eligibility data and death certificate files, which were not publicly available because of ethical and privacy restrictions. In 2021, Medicaid provided coverage to 38% of the children in Ohio.²⁵ Three broad enrollment categories qualify children for Medicaid: children <19 years whose family income is at or below 200% of the federal poverty level (FPL); (2) children with a disabling condition whose family is at or below 64% of the FPL; and (3) children in foster care receiving adoption assistance or in institutional placements. Medicaid claims data included information on paid inpatient hospitalizations, office- or hospital-based physician visits, other outpatient service claims, and up to 16 diagnostic codes for each date of service use. Medicaid eligibility files included information on monthly enrollment status and demographic characteristics of enrollees. Death certificate data were used to identify suicide deaths.

Measures: Psychiatric and Medical Diagnoses

Medicaid claims data spanning the year preceding children’s index self-harm event on psychiatric diagnoses were extracted from the Medicaid claims data files to create 8 categorical indicators used in the latent class analysis (LCA): suicidal ideation; fear/anxiety internalizing disorders (eg, anxiety, phobias, compulsions); distress internalizing disorders (eg, depression, adjustment, and stress disorders); disinhibited externalizing disorders (eg, oppositional defiant disorder); impulsivity (eg, impulsivity disorders, ADHD with impulsivity or combination type); early-childhood adversity (ECA; defined as diagnoses related to abuse, maltreatment, neglect, and/or documented history of residence in foster or adoptive care); intellectual and developmental disorders (eg, autism spectrum disorder); and thought disorders (eg, schizoaffective disorder, bipolar disorders). Psychiatric diagnoses were collapsed into transdiagnostic categories described in the Hierarchical Taxonomy of Psychopathology (HiTOP), a hierarchical classification of psychopathology that groups emotional and behavioral symptoms into dimensional categories.^26,27 Diagnoses falling within these categories tend to be highly comorbid and share symptoms that respond similarly to treatment.^26,27 Greater detail regarding specific disorders included in each indicator can be found in Supplement 1, available online, Table S2.

Covariates included age at first self-harm event, residence in a non-metropolitan area (1 = non-metropolitan, 0 = metropolitan area), use of mental health services within 30 days before self-harm (1 = mental health services were used, 0 = no record of services), race (1 = non-Hispanic White, 0 = non-White including American Indian or Alaskan native, Asian Americans, Native Hawaiians or other Pacific Islanders, and non-Hispanic Black, Hispanic), and sex (1 = female, 0 = male). Method and lethality of self-harm are provided for the overall sample and latent classes (LCs) but were not included in the analysis. Methods of self-harm were used to define lethality (see Supplement 1, available online, Table S1 for ICD coding), similar to previous research. High lethality was defined as multiple methods used, firearm, or hanging/ strangulation; low lethality was defined as poisoning, cutting, or other/unspecified means (eg, transportation-related).²⁸ Race and ethnicity were defined based on self-reporting and Medicaid classification. Descriptive information for all racial and ethnic groups is provided in Table 1; these groups were collapsed for the analysis. Although defining race using only 2 categories is not preferred, less than 1% of the present sample was of a race other than non-Hispanic White or non-Hispanic Black, and a more diverse sample is needed to test more culturally relevant models.²⁹ Given how little is known about young children who self-harm and the racial and ethnic disparities that exist in suicide rates at younger ages, we see this study as an important first step.

Table 1.

Sample Clinical and Demographic Characteristics

Characteristic	n	%
Age, y
5-8	196	22.3
9-11	682	77.7
Sex
Female	469	53.4
Male	409	46.6
Race/ethnicity
Black (non-Hispanic)	258	29.4
White (non-Hispanic)	604	68.8
Othera	16	0.7
Self-harm method
Cutting	277	31.5
Drowning/submersion	2	0.2
Firearm	1	0.1
Hanging/suffocation	41	5.7
Jumping	8	2.2
Multiple methodsb	47	5.4
Other/undefined meansc	224	25.5
Poisoning	278	31.7
Clinical factors
Disinhibited externalizing	436	50.0
Distress internalizing	694	79.0
Fear/anxiety internalizing	316	36.0
Impulsivity	454	51.7
Intellectual or developmental	110	12.5
suicidal ideation	427	48.6
Thought disorders	260	29.6
Contextual factors
Early adversity	223	25.4
Any mental health services used	592	67.4
Inpatient	54	6.2
Outpatient	330	37.6
Emergency room	83	9.5
Non-metropolitan residence	166	18.9

Note:

^aOther race and ethnicities included in the “Other” category are Hispanic, American Indian or Alaska native, Asian American, Native Hawaiians or other Pacific Islanders.

^b“Multiple methods” was defined as having used more than 1 method, such hanging and poisoning.

^c“Other” includes transportation-related or poorly defined methods.

Data Analysis

Descriptive statistics were calculated to examine the demographic and clinical characteristics of the study sample using SPSS version 28.³⁰ LCA using Mplus 8³¹ was used to identify homogenous subgroups based on response patterns to the 8 categorical indicators. LCA is a form of mixture modeling using an algorithmic approach to probabilistically assign individuals into homogenous subgroups, or latent classes (LCs).³² Probability estimates represent the probability that an individual assigned to that LC was diagnosed with 1 or more of the diagnoses encompassed within that categorical indicator in the year before self-harm. Model fit statistics were compared using commonly accepted fit indices for models testing 1 to 4 classes using the following: Bayesian Information Criterion (BIC), Akaike Information Criterion (AIC), sample-size adjusted BIC (CAIC), the Lo–Mendell–Rubin likelihood ratio test (LRT), and entropy comparing the improvement in model fit between the number of classes.^32,33 We further considered the mean probability of belonging to the most likely assigned class, the sample sizes of each class, and how derived classes aligned with prior research as is recommended to avoid over-extracting LCs.^32,33

After selecting the best-fitting model, we used the manual 3-step Bolck, Croon, and Hagenaars method³⁴ (BCH) to test the association between LC membership and subsequent self-harm events within the following year. This method is advantageous over other approaches, as it accounts for classification uncertainty in the estimation of distal outcomes. Covariates and the distal outcome were included in the estimation of the final model; thus, differences in distal means of subsequent self-harm events across classes are adjusted for covariates and classification error.^33,34 Posterior probability estimates were used to aid in understanding the characteristics of the derived LCs; it should be noted that these estimates do not account for classification error. Finally, we compared the measurement model and the final model including the covariates and distal outcome for potential shifting of LCs.

To examine class-specific differences in subsequent self-harm events, our distal outcome, an omnibus Wald χ² test was conducted in Mplus.^33,35 Tests for post hoc pairwise intercept and slope differences were specified using model constraints. Significant differences between class-specific intercepts indicate significant differences in risk for subsequent self-harm incidents.³³

Results

Of the 878 children, the mean age was 9.67 years (SD = 1.57) at the time of the index self-harm event (Table 1). Most children were non-Hispanic White (69%); 82% qualified for Medicaid based on income; and 81% resided in a metropolitan area. Roughly an equal proportion of children were male and female (Table 1). The most common method of self-harm was poisoning (31.7%), followed by cutting (31.5%), and 5.4% of children used multiple methods (eg, hanging and cutting). Distress internalizing disorders (78%) and impulsivity (52%) were the most frequently endorsed indicators, and 13% of the sample had a documented subsequent self-harm incident within the year following the index self-harm event.

Nearly two-thirds of class 1 children utilized mental health services within 30 days before the first self-harm event (.63), and many of these children received outpatient services (.57). See Figure 1 for graphical display of the proportion of children with positive endorsements for each indicator within the three LCs.

Figure 1.

Proportion of Positive Psychiatric and Contextural Factors Within Latent-Class

Latent Classes of Child Self-Harm

Model fit indices BIC, AIC, and CAIC were lower (Table 2), entropy was higher, and the LRT was significant in the 3-class solution compared to the 2-class solution, suggesting that the model was significantly improved with the addition of a third class compared to the 2-class solution. Although entropy increased with the addition of a fourth class, BIC, AIC, and CAIC increased, suggesting there was no significant improvement with the addition of a fourth class. The additional fourth class was also not substantively different from the existing classes. Thus, we selected the 3-class solution as the best-fitting model for the data.

Table 2.

Summary of Latent Class Model Identification and Fit Statistics

No. of classes	Entropy	LMT	AIC	BIC	CAIC
1	—		8451.17	8489.39	8463.98
2	0.66	<.00	7906.98	7988.20	7934.21
3	0.69	<.00	7723.86	7848.07	7765.51
4	0.76	.04	7882.71	8169.36	7987.81

Note: AIC = Akaike Information Criterion; BIC = Bayesian Information Criterion; CAIC = sample-size adjusted BIC; LRT = Lo–Mendell–Rubin likelihood ratio test.

Table 3 presents the LCs and the probability children assigned to that LC responded positively to that indicator. These probability estimates can also be interpreted as the proportion of children within that LC having at least 1 diagnosis captured within that categorical indicator. Posterior probability estimates were used to further describe the LCs in terms of demographics (provided in the lower portion of Table 3). In an exploratory manner, we used posterior probability estimates to describe the type of mental health services used and the mean number of days until subsequent self-harm within LCs. The 3 classes were roughly equal in size, and classes generally aligned with prior research in terms of comorbidity and demographic characteristics. We examined the mean probability of belonging to the most likely assigned class; the most likely assigned class probability for all classes ranged from 0.85 to 0.89.

Table 3.

Probability Estimates Within Latent Classes of Children Who Self-Harmed

Probability	Class 1			Class 2				Class 3
	n = 329, 37%				n = 255, 29%				n = 294, 33%
	Est	SE	p		Est	SE	p	Est		SE	p
Distress	0.92	0.03	.00		1.00	0.00	1.00	0.46		0.07	.00
Suicidal ideation	0.61	0.05	.00		0.80	0.09	.00	0.07		0.02	.00
Adversity	0.34	0.04	.00		0.38	0.04	.00	0.05		0.03	.17
Intellectual/dev	0.24	0.03	.00		0.03	0.02	.09	0.09		0.02	.00
Fear/anxiety	0.46	0.03	.00		0.51	0.05	.00	0.12		0.03	.00
Thought	0.60	0.04	.00		0.21	0.06	.00	0.06		0.02	.00
Disinhibited	0.90	0.03	.00		0.35	0.08	.00	0.21		0.04	.00
Impulsivity	0.97	0.05	.00		0.26	0.07	.00	0.29		0.04	.00
Characteristics using posterior probability estimates
Age, y, at DSH	9.54				10.24			9.31
Subsequent DSH	15.5%				18.6%			6.0%
Lethality	13.4%				13.7%			6.8%
Mental health services	63.2%				39.2%			22.8%
Emergency room	17%				8%			2%
Inpatient	10%				8%			1%
Outpatient	57%				31%			21%
Non-metropolitan	15.5%				16.1%			25.2%
Sex, Female	38.6%				72.9%			53.1%
Race, White	67.5%				66.3%			72.4%

Note: dev = Developmental; DSH = deliberate self-harm.

Children in class 1 (n = 329, 38%) were more often male, younger, and used more mental health services compared to children in class 2. Children in class 1 were older, were less likely to be from non-metropolitan communities, and more often used mental health services than those in class 3. Additional results from the covariate multinomial logistic regression results are displayed in Table 4. A large proportion of children belonging to class 1 had impulsive (0.97) or disinhibited (0.90) behavioral disorders and distress internalizing disorders (0.92), with more than half having a history of SI (0.61). This class had the highest proportion of children with intellectual and/or developmental disorders (0.24). Nearly two-thirds of class 1 children used mental health services within 30 days before the first self-harm event (0.63), and many of these children received outpatient services (0.57).

Table 4.

Covariate Multinomial Logistic Regression

Class 1 vs class 3				Class 1 vs class 2			Class 3 vs class 2
	OR	SE	95% CI	OR	SE	95% CI	OR	SE	95% CI
Age	0.88	0.06	[0.76, 1.00]	1.53	0.17	[1.22, 1.91]	0.58	0.06	[0.47, 0.71]
White	1.02	0.26	[0.62, 1.69]	0.82	0.23	[0.47, 1.43]	1.26	0.35	[0.73, 2.16]
Female	2.21	0.54	[1.37, 3.57]	5.25	1.46	[3.05, 9.05]	0.42	0.12	[0.25, 0.72]
Non-metro	2.16	0.68	[1.16, 3.88]	1.24	0.46	[0.62, 2.57]	1.70	0.56	[0.89, 3.25]
MHS	0.12	0.03	[0.08, 0.19]	0.30	0.08	[0.18, 0.51]	0.40	0.11	[0.23, 0.71]

Note: Race is coded as 1 = non-Hispanic, White. Sex is coded as Female = 1. MHS = mental health services; Nonmetro = nonmetropolitan; OR = odds ratio.

Children belonging to class 2 (n = 255, 29%) were more often female and older compared to children in the other classes. Children in class 2 used mental health services more often compared to those in class 3 but less often compared to those in class 1. Children in class 2 had a high proportion of distress internalizing disorder diagnoses (1.0), SI (0.80), anxiety internalizing disorders (0.51), and documented history of ECA (0.38). Compared to the other 2 classes, class 2 had fewer children with impulsive disorders (0.26). Although class 2 had the highest proportion of SI before the self-harm event, less than half of these children used mental health services (0.39), and even fewer used outpatient mental health services (0.31).

Children in class 3 (n = 294, 33%) were more likely to be male than those in class 2 and were more likely to be female than those in class 1. Overall levels of psychopathology were lower in class 3 children, and only a small minority had previously reported suicidal thinking compared to the other 2 classes; children in class 3 were more likely to be younger and less likely to use mental health services. The most common diagnostic categories within this class were distress internalizing disorders (0.46) and impulsivity (0.29). Posterior probability estimates found this class to have the smallest proportion of children who used mental health services (0.23), with most services being outpatient (0.21).

Predicting Subsequent Self-Harm Events From Latent Classes

There were minimal shifts across LCs with the addition of the distal outcome and covariates, and interpretation of LCs was no different. The omnibus Wald test supported a statistically significant association between the LC membership and the distal outcome (χ²[2] = 19.73, p < .001). The pairwise comparisons supported a statistically significant difference between the distal means (or the proportion of children with subsequent self-harm events) across LCs, while also controlling for covariates and misclassification of LC membership. The mean difference (MD) in risk for subsequent self-harm incidents between class 1 (mean = 0.13) and class 3 (mean = 0.01) was significantly higher (MD = 0.12, p < .00, SE = 0.03), and MD in risk for subsequent self-harm incidents between class 2 (mean = 0.16) and class 3 was significantly higher (MD = 0.15, p < .00, SE = 0.04). The MD between risk for subsequent self-harm events between class 1 and class 2 was not significantly different (MD = −0.04, p = 0.43, SE = 0.05). In an exploratory analysis, we examined the mean number of days until subsequent self-harm incidents occurred in class 1 (mean = 48, SD = 86), class 2 (mean = 63, SD = 99), and class 3 (mean = 55, SD = 71). Days between first and subsequent self-harm were not included in the LCA and should not be interpreted.

Discussion

Self-harm is not uncommon in children, and is a strong risk factor for SA, particularly at younger ages, warranting efforts to better identify and intervene with children at risk for self-harm and suicide.⁹ The American Academy of Pediatrics recommends routine suicide risk screening beginning at age 12 years and for children between 8 and 11 years when clinically indicated. In youth under 8 years of age, screening is not indicated, noting to assess for suicidal thoughts and behaviors if warning signs are present.³⁶ Yet all children in the present study self-harmed before the recommended age of routine suicide risk screening, with 12% of children being between ages 5 and 7 years.^37,38 Children who self-harm are heterogeneous in terms of clinical and contextual risk factors, and complex clinical presentations are not unusual, making the assessment of risk for self-harm and suicide challenging at younger ages. In addition, little is known about specific warning signs in this age group, making existing guidance for screening children under age 8 years difficult to follow. The present study identified subgroups of children sharing similar clinical profiles to inform recognition and prevention efforts and to contribute to our understanding of the mechanisms of risk for subsequent incidents of self-harm.

Children belonging to class 1 (37%) had high rates of both emotional and behavioral disorders and were largely male, with a majority having a prior history of expressing suicidal ideation and having prior involvement with mental health services. Nearly all class 1 children had internalizing psychopathologies, which often include symptoms such as strong and persistent negative affect, as well as a history of disinhibited and impulsive externalizing behavioral disorders. About one-third of class 1 children also experienced serious life adversities such as maltreatment or foster care placement.

Highly impulsive children may have difficulty accessing and implementing adaptive coping strategies when experiencing strong negative affect, potentially increasing vulnerability to self-harm when experiencing strong negative affect.^39,40 Impulsivity, maltreatment, and residence in unstable homes are common among children who die by suicide,¹² and ADHD in maltreated adolescents (particularly girls) is associated with self-harm.²⁴ Our results add to prior research suggesting that children with a history of impulsivity and internalizing disorders or impulsivity with a history of maltreatment are at elevated risk for self-harm and should be screened for suicide risk at even younger ages.^12,24 In sum, children with profiles consistent with class 1 should be recognizable as being at risk for self-harm and suicide by clinicians, given the combination of impulsive and disinhibited externalizing disorders with internalizing disorders. Such children are likely to present in mental health treatment settings, are most often male, and commonly have a history of serious life adversity or neurodevelopmental disability.

Children belonging to class 2 (29%) represent a group of emotionally distressed children who are predominantly female and relatively likely to express suicidal ideation. All children in class 2 had a recorded history of distress internalizing disorders such as depression or post-traumatic stress disorder, and, although approximately one-third had a history of disinhibited behavior and about one-fourth had a history of impulsivity, externalizing behavioral disorders were lower in children in class 2 compared to those in class 1. Children in both class 1 and class 2 had records consistent with high levels of psychopathology and similar rates of life adversity, and were at similar risk for a subsequent self-harming event in the following year. These differences in clinical presentations between class 1 (39% female) and class 2 (73% female) could be reflective of sex-specific psychopathology and comorbidities. Internalizing disorders tend to emerge earlier and are more common in female individuals, and internalizing disorders typically attract less attention from parents and caregivers.^23,41 The lower prevalence of engagement with mental health services for children in class 2 relative to class 1 is of some concern, given the relatively high prevalence of SI among children in class 2.

Children belonging to class 3 (33%) might not be as readily recognized in clinical settings as being at risk for self-harm relative to those in classes 1 and 2. Very few children in class 3 had a known history of SI or serious life adversity, and although nearly one-half had a history of distress-related internalizing disorders and approximately one-fourth had recognized problems with behavioral disinhibition or impulsivity, this group was relatively unlikely to be engaged with mental health services compared to children in classes 1 and 2. Children in class 3 were more likely to live in more rural areas and were also a bit younger than those in classes 1 and 2. The lower levels of recorded psychopathology and low use of mental health services could be reflective of younger ages or of having fewer and more distal mental health services available to provide treatment in non-metropolitan communities.

Class 1 and 2 children are likely to be the most readily recognizable in clinical settings. Those in class 1 tend to present with readily observable externalizing disorders and behaviors in combination with internalizing disorders, and often report suicidal thinking and present in mental health settings; preadolescent male individuals at risk for self-harm appear most likely to present in this fashion. Class 2 children also have potential to be identified in clinical settings, most commonly presenting with internalizing disorders and frequently reporting suicidal ideation. Predominantly female, class 2 children are a bit less likely than those in class 1 to have a history of specialty mental health treatment, but appear equally at risk for repetitive self-harming behaviors. As such, it is reasonable to target prevention efforts for class 1 and class 2 children within clinical settings and to champion screening for suicide risk for preadolescent children with mental disorders, particularly in mental health settings. This recommendation highlights the urgent need to develop suicide risk screeners for youth under age 8 years, as currently there are no validated measures for this age group.³⁸ A bit more challenging is how to prevent harm in at-risk children in class 3, who are less likely to be recognized in clinical settings. As such, universal suicide prevention efforts relevant to the larger pediatric population, such as those delivered in school settings, may prove to be an increasingly important strategy for preventing youth suicide compared to selective (targeting high-risk groups) and indicated (showing signs of risk for self-harm) prevention strategies alone.

Strengths of the present study include our HiTOP-informed approach, which provides information about broader dimensions of psychopathology and may advance efforts to identify those at risk for self-harm based on symptoms, traits, and labilities that are transdiagnostic.²⁷ We also include a large population-based sample of children under 12 years of age, including a large number of children with self-harm between ages 5 and 7 years, children from underrepresented groups, and children from non-metropolitan regions. However, the study has several limitations. First, because data are from a single state, study findings may not be generalizable to other states or other, non-Medicaid populations. Second, diagnoses were based on clinical judgment and were not subject to expert validation through standardized assessments. Third, because some children may not have been enrolled continuously for the entire year before and after the index self-harm event, some diagnoses may not have been captured. Fourth, claims do not distinguish between self-harm injuries with suicidal intent from those with nonsuicidal intent and do not capture self-harming events that do not result in medical treatment; thus, children in this sample represent children who were treated for self-harm or whose self-harm was documented by health care providers. Finally, information in the claims data was not available on several other important factors that may increase the risk for deliberate self-harm in children, including precipitating stressors such as family conflict or difficulty with peers and school.

This study’s results call attention to the complementary importance of suicide prevention strategies at the level of at-risk individuals (indicated), high-risk groups (selective), and the pediatric population at large (universal). By identifying the clinical presentations of subgroups of children who self-harmed at young ages, the results of this study could support efforts to better identify groups of children at high risk for self-harm and suicide in clinical settings. At a minimum, screening for suicide risk in preadolescent children with distress-related internalizing disorders and those with comorbid disinhibited and impulsive externalizing behavioral disorders appears to be justified in general medical and specialty mental health settings. At present, however, measures such as the Ask Suicide Questions (ASQ) and other validated suicide screening instruments are validated only in children as young as 8 years of age,^37,38 and very little is yet known about what might distinguish self-harming children who continue to self-harm from those who do not. Findings from the present study suggest that children who self-harm at young ages with a history of SI should prompt intervention and continued screenings to prevent subsequent acts of self-harm and escalating suicidal tendencies. Furthermore, more than one-half of the sample in the present study had either or both disinhibited and impulsive psychopathologies, which tend to precede the development of substance misuse disorders,^5,42 a significant risk factor for suicide death later in life.⁴³ Future longitudinal studies are needed to study the developmental progression of psychopathology in young children with self-harming tendencies to identify potential points for intervention and prevention of worsening psychological impairment, substance misuse, SI, and risk for recurring acts of self-harm and suicide.

In conclusion, we identified 3 subgroups of preadolescent children who shared co-occurring dimensional characteristics in the year before first exhibiting self-harming behavior. Our results add to the growing concern that preadolescent children do engage in self-harm at young ages and have heterogeneous clinical profiles. Children presenting with internalizing psychopathologies comorbid with externalizing disorders or a history of maltreatment should regularly be screened for self-harm and suicide. Given that impulsivity is a highly heritable trait, young children could benefit broadly from interventions that target adaptive emotion regulation, arguably a modifiable characteristic at younger ages, to prevent self-harm and suicide.

CRediT authorship contribution statement

Amanda J. Thompson: Writing – review & editing, Writing – original draft, Visualization, Validation, Software, Resources, Project administration, Methodology, Investigation, Formal analysis, Data curation, Conceptualization. John V. Campo: Writing – review & editing, Writing – original draft, Supervision, Conceptualization. Jennifer L. Hughes: Writing – review & editing, Supervision, Conceptualization. Jeffrey A. Bridge: Writing – review & editing, Funding acquisition, Conceptualization. Donna A. Ruch: Writing – review & editing, Conceptualization. Cynthia A. Fontanella: Writing – review & editing, Writing – original draft, Supervision, Resources, Project administration, Methodology, Investigation, Funding acquisition, Data curation, Conceptualization.