Sleep Disturbance Trajectories During Childhood and Early Adolescence Associated With Increased Suicide Risk

Abstract

Objective

Sleep disturbances have been linked to suicide risk, but few studies have explored these effects during the transition from childhood to adolescence. This study examined whether specific trajectories of sleep disturbance across childhood and early adolescence were associated with greater suicidal thoughts and behaviors (STB) for youth in the Adolescent Brain Cognitive Development℠ (ABCD) Study.

Method

Data from 11,864 participants in the ABCD Study® (Data Release 5.1) were used in this study. Youth STB were assessed by the Schedule for Affective Disorders and Schizophrenia for School-Age Children suicidality module. The Sleep Disturbance Scale for Children measured sleep disturbance. Latent class growth analysis was used to identify sleep trajectories, and Bayesian ordinal regression models were used to examine whether sleep trajectories were differentially associated with STB during 2-, 3-, and 4-year follow-up.

Results

Latent class growth analysis identified 3 latent sleep profiles (low-stable, high-decreasing, and moderate-increasing). Sleep profiles with greater disturbance, including both high-decreasing (odds ratio 1.75, 95% CI [1.06, 2.89], p = .030) and moderate-increasing (odds ratio 2.34, 95% CI [1.72, 3.14], p < .001) profiles, were linked to higher likelihood of more severe STB outcomes compared with the low-stable group.

Conclusion

This study identified distinct developmental trajectories of sleep disturbance across childhood and early adolescence linked to STB. Specifically, early high sleep disturbances that improved and moderate disturbances that worsened over time were both associated with greater STB severity. Difficulty falling and staying asleep and excessive sleepiness were common in both patterns. These findings highlight the need to identify and address early and/or worsening sleep problems as a potential target for suicide prevention strategies.

Study registration information

Analysis Code for: “Sleep Disturbance Trajectories During Childhood and Early Adolescence Associated with Increased Suicide Risk”; https://osf.io/pf49z.

Plain language summary

Sleep disturbance has been associated with suicidal risk. This study examined whether sleep problems from childhood to early adolescence were linked to suicidal thoughts and behaviors. Using data from over 11,000 youth in the Adolescent Brain Cognitive Development℠ (ABCD) Study, researchers found those with persistent or worsening sleep disturbances, such as trouble falling or staying asleep and daytime sleepiness, were more likely to report suicidal thoughts and behaviors. Early and ongoing sleep problems may be important warning signs for suicide risk and could help guide prevention efforts.

Adequate sleep during childhood and adolescence is essential for promoting healthy cognitive and psychosocial functioning and related brain development.¹ The American Academy of Sleep Medicine recommends that children ages 6 to 12 sleep 9 to 12 hours and adolescents ages 13 to 18 sleep 8 to 10 hours each night.² According to these guidelines, the majority of youth are sleep deprived, as the Youth Risk Behavior Surveys found that only 1 in 4 middle school and high school students reported getting at least 8 hours of sleep.³

Sleep patterns and circadian rhythms undergo substantial changes throughout childhood and adolescence, influenced by both biological and social/environmental factors.^4,5 Biologically, after puberty, the homeostatic drive to sleep accumulates more slowly, making it harder for youth to fall asleep and easier for them to stay awake at night.⁶ Additionally, melatonin secretion shifts to a later time during adolescence, causing a delay in the circadian rhythm.⁷ Social and environmental factors, such as reduced parental monitoring, increased screen time, caffeine use, and early school start times also play a critical role in disrupting sleep during these years.⁸ Whereas these changes in sleep are typical during adolescence, insufficient sleep is linked to poorer academic performance and an increased risk of emotional and behavior problems.⁹

Lack of sleep and sleep disturbance are particularly concerning because they are associated with poor emotional regulation and increased risk of self-harm.^9,10 In youth without psychiatric disorders, inadequate sleep has been associated with more symptoms of depression¹¹ and anxiety.¹² Sleep problems during childhood have also been shown to predict higher levels of depression and anxiety in adolescence.¹³ Furthermore, sleep disturbances are linked to a higher risk for suicide in adolescents and adults.^14,15 Meta-analyses have provided preliminary support that some sleep disturbances (eg, insomnia, nightmares) are independent risk factors for suicidal ideation (SI), suicide attempts (SAs), and death by suicide.¹⁴ Moreover, sleep disturbances are linked to reduced attention, inhibition, and emotion regulation,^16,17 which may contribute to difficulties in controlling and acting on thoughts of suicide.¹⁸

Although the connection between sleep problems and suicide risk has been established in adults and adolescents, less is known about the role of sleep disturbance in preadolescents and the emergence of suicidal thoughts and behaviors (STB). Most research examining sleep and suicide risk has focused on epidemiological samples of adults and adolescents,¹⁴ with few longitudinal studies following children over time through adolescence. Prospective studies, such as the Adolescent Brain Cognitive Development℠ Study, can offer valuable insights into how sleep disturbances impact behavioral health trajectories, including increasing risk for STB.

Researchers have used latent profile analysis of ABCD Study® data to delineate distinct sleep trajectory groups based on shared sleep patterns (eg, sleep duration, wakefulness periods, sleep efficiency) and their associations with various outcomes. For example, Cooper et al.¹⁹ identified 4 distinct sleep profiles in the ABCD Study: low disturbance, sleep onset/maintenance problems, mixed disturbance, and high disturbance across 2 time points. Youth in the sleep onset/maintenance problems, mixed disturbance, and high disturbance profiles were found to have higher internalizing and externalizing symptoms. Another study using latent profile analysis of data obtained from Fitbit devices in a subset of ABCD participants found that youth with both low sleep duration and poor sleep efficiency had the highest behavioral problems, including rule-breaking behaviors, attention issues, and social problems.²⁰

Despite the established link between sleep disturbance and STB in both adolescence and adulthood, few prospective studies have focused on STB as a primary outcome variable. Most research examining sleep disturbance and suicide risk in youth has relied on cross-sectional analyses, which limits the ability to identify sleep disturbance trajectories that may contribute to the emergence of STB. A recent study by Gowin et al.²¹ used ABCD data to examine whether sleep disturbances in childhood predicted STB at the 2-year follow-up. Findings indicated that disturbed sleep at ages 9 and 10, particularly nightmares and excessive daytime somnolence, was associated with increased risk for STB within the following 2 years. Although this study found that childhood sleep disturbances were linked to longitudinal risk for STB, more prospective studies that examine sleep patterns across both childhood and adolescence are needed. Using latent profile modeling to identify distinct subgroups and individual sleep disturbance trajectories over time may thus provide a more nuanced understanding of the temporal nature of sleep–suicide associations and inform personalized treatment approaches.

To address these gaps, the current study aimed to examine whether sleep disturbances in childhood and early adolescence are associated with STB during adolescence. Specifically, this study examined how patterns of sleep disturbance evolve from childhood to early adolescence and whether certain sleep trajectories are associated with greater STB in adolescence. We hypothesized that persistent sleep disturbances across childhood would be associated with higher STB risk during adolescence.

Method

Participants

Participant data were from the ABCD Study, a 10-year study conducted at 21 US data collection sites. The ABCD Study enrolled 11,876 9- and 10-year-olds at baseline from 21 initial sites primarily through school-based recruitment.²² Data used in the current study were from ABCD Data Release 5.1, which included complete data from baseline; 1-, 2-, and 3-year follow-up visits; and approximately half of the 4-year follow-up data. Centralized institutional review board approval was obtained from the University of California San Diego, and study sites obtained approval from their local institutional review boards. Caregivers provided written informed consent and permission, and children provided written assent. Data collection and analysis complied with all ethical regulations.

ABCD Data

Sleep Disturbance Trajectories

Caregivers completed the 26-item Sleep Disturbance Scale for Children (SDSC) at each annual visit, assessing the youth’s sleep-related behaviors in the past 6 months.²³ Each item is rated on a 5-point scale, which is summed to a total score and 6 subscale scores, where higher scores indicate greater sleep disturbance. The 6 subscales include disorders of arousal/nightmares, disorders of initiating and maintaining sleep, disorders of excessive somnolence, sleep breathing disorders, sleep hyperhidrosis, and sleep-wake transition disorders. Data from the SDSC at all time points were used to determine sleep trajectories.

Suicidal Thoughts and Behaviors

STB were measured using the computerized version of Schedule for Affective Disorders and Schizophrenia for School-Age Children (K-SADS).²⁴ Youth participants completed the suicide module at annual visits, which consists of questions to assess passive thoughts of suicide, active thoughts of suicide (SI), and SAs. Based on participant responding and endorsement of symptoms, participants were coded as having recent STB during the past 2 weeks using the following levels: 0 = no suicidality; 1 = passive ideation; 2 = active ideation including nonspecific active SI, active SI, or active SI with plan; 3 = SA including interrupted SA, aborted SA, or SA. Data from the 2-, 3-, and 4-year follow-up assessments were used in the primary analyses for modeling STB outcomes at each respective time point.

Covariates

We included sex assigned at birth and age (in months at time of visit) variables as covariates given their associations with adolescent suicidality.²⁵ Income-to-needs ratio was also included due to its association with sleep and suicidality^26,27 and was calculated by dividing parent-reported family income by the poverty threshold for a family of that size for the year of the visit.²⁸

Pubertal development was included as a covariate given the previous literature showing an impact on sleep and suicide risk.²⁹ Pubertal status was measured using the youth report on the Pubertal Development Scale (PDS)³⁰; its use in the ABCD Study including scoring information has been previously reported in detail.³¹ Categorical scores are derived based on responses to items on a Likert scale regarding height, body hair, skin, and voice for boys and girls and breast development and menarche for girls, with lower scores representing earlier pubertal stages. The category of early puberty was used as the reference class for the categorical pubertal status variable in the analyses.

Studies have shown the heritability of suicide risk.³² Family history of suicide was measured using a modified version of the Family History Assessment Module Screener (FHAM-S) used in the National Consortium on Alcohol and Neurodevelopment in Adolescence (NCANDA) study.³³ Parents reported on the presence or absence of a parent of the youth ever experiencing an SA or death by suicide (0 = “no”; 1 = “yes”), which was included as a covariate.

Peer victimization has also been shown to be a risk factor for STB that can impact sleep.³⁴ The Peer Experiences Questionnaire (PEQ) was collected annually following the 2-year visit and used to assess whether the youth experienced overt, relational, or reputational victimization from peers.³⁵ Each domain was scored on a 5-point Likert scale ranging from 1 (“never”) to 5 (“a few times a week”). Covariates from the 2-, 3-, and 4-year follow-up visit were used in the primary analyses.

Data Analysis

All analyses were conducted in R version 4.4.2.³⁶ For detailed information regarding missingness and attrition in the current study, see the supplementary methods in Supplement 1, available online.

Identifying Sleep Disturbance Trajectories

Univariate and multivariate testing was used to determine the best fit for sleep trajectories. We analyzed the trajectory of each SDSC subscale individually and determined that the growth patterns followed a linear trajectory across baseline (ages 9-10), year 1 (ages 10-11), year 2 (ages 11-12), year 3 (ages 12-13), and year 4 (ages 13-14) of the study. To achieve this, we used univariate latent class growth analysis (LCGA) to progressively fit linear and flexible spline models with an increasing number of classes, and we observed negligible evidence of nonlinear trajectories within any subscale. Multivariate testing examined the combined trajectories of all SDSC subscales to determine the optimal number of latent classes, testing 1-, 2-, 3-, and 4-class solutions using LCGA. Model fit was evaluated using several fit statistics, including Akaike information criterion, Bayesian information criterion, log likelihood, entropy, and class distribution percentages, and indicated that the 3-class solution was optimal.

Primary Analyses

Bayesian ordinal regressions were conducted using the brms package37, 38, 39 to examine whether STB at the 2-, 3-, and 4-year follow-up visits was differentially associated with the latent trajectory classes of sleep disturbance. The baseline and 1-year follow-up visits were not included in the primarily analyses due to covariate availability, as the PEQ was not collected during these time points. We accounted for the effects of age, sex assigned at birth, income-to-needs ratio, pubertal status, family history of suicide, and peer victimization at each time point given prior evidence of their association with STB. Model type was a cumulative link model, which models the probability of an outcome being in a higher level of STB severity compared with the probability of being in a lower level of STB severity. Thus, odds ratios (ORs) less than 1 indicate a predictor associated with a decreased likelihood of higher level of STB severity, whereas ORs greater than 1 indicate an increased likelihood of a higher level of STB severity. LCGA classes were used as sleep-related predictors, and the largest sleep trajectory class was used as the reference. To assess the robustness of the association between sleep disturbance trajectories and STB, we conducted a sensitivity analysis incorporating depression.

Sensitivity Analyses Incorporating Depression

Depression has been shown to be associated with both sleep disturbance and STB.⁴⁰ To assess the robustness of the association between sleep disturbance trajectories and STB, anxious/depressed symptoms (excluding sleep items) were included in a sensitivity analysis as a covariate to examine the effects of anxiety/depression independent of sleep. Scores from the anxious/depressed subscale of the Child Behavior Checklist (CBCL) were used. The CBCL is a 113-item measure of behavior and dimensional psychopathology over the past 6 months completed by the caregiver.⁵ Each question is rated on a 3-point scale (1 = “not true”; 2 = “somewhat or sometimes true”; and 3 = “very often or always true”). The CBCL standardizes summed scores as a function of age and sex, with standardized scores >65 regarded as at risk for clinical symptoms. CBCL Anxious/Depressed subscale data from the 2-, 3-, and 4-year follow-up visits were used in the sensitivity analysis as a covariate.

Results

Participant Characteristics

For detailed characterization of patterns of missingness and attrition in the current study, see the supplementary results in Supplement 1, available online. Participant characteristics, including descriptive statistics for demographic variables and SDSC scores across all time points in the current study, are summarized in Table 1.

Table 1.

Characteristics of the Sample by Assessment Time Point

Variable	Baseline (N = 11,864)		Year 1 (n = 11,220)		Year 2 (n = 10,973)		Year 3 (n = 10,338)		Year 4 (n = 4,754)
	Mean	(SD)	Mean	(SD)	Mean	(SD)	Mean	(SD)	Mean	(SD)
Age, y	9.91	(0.62)	10.92	(0.64)	12.03	(0.67)	12.91	(0.65)	14.08	(0.68)
	n	(%)	n	(%)	n	(%)	n	(%)	n	(%)
Sex at birth, female	5,674	(47.8)	5,349	(47.7)	5,217	(47.5)	4,911	(47.5)	2,265	(47.6)
	Mean	(SD)	Mean	(SD)	Mean	(SD)	Mean	(SD)	Mean	(SD)
Income-to-needs ratio	3.68	(2.34)	3.81	(2.30)	3.88	(2.26)	3.95	(2.23)	3.93	(2.18)
	n	(%)	n	(%)	n	(%)	n	(%)	n	(%)
Race
Asian	252	(2.1)	241	(2.1)	231	(2.1)	221	(2.1)	110	(2.3)
Black	1,782	(15.0)	1,597	(14.2)	1,561	(14.2)	1,361	(13.2)	509	(10.7)
Hispanic	2,409	(20.3)	2,221	(19.8)	2,164	(19.7)	2,077	(20.1)	979	(20.6)
Othera	1,249	(10.5)	1,176	(10.5)	1,156	(10.5)	1,087	(10.5)	474	(10.0)
White	6,172	(52.0)	5,985	(53.3)	5,861	(53.4)	5,592	(54.1)	2,682	(56.4)
STB level
No SI	11,516	(97.1)	10,896	(97.1)	10,679	(97.3)	9,925	(96.0)	4,492	(94.5)
Passive SI	66	(0.6)	47	(0.4)	29	(0.3)	21	(0.2)	11	(0.2)
Active SI	163	(1.4)	121	(1.1)	121	(1.1)	226	(2.2)	157	(3.3)
SA	47	(0.4)	31	(0.3)	30	(0.3)	28	(0.3)	11	(0.2)
	Mean	(SD)	Mean	(SD)	Mean	(SD)	Mean	(SD)	Mean	(SD)
SDSC scores
DA	3.44	(0.92)	3.35	(0.78)	3.31	(0.73)	3.25	(0.65)	3.19	(0.53)
DIMS	11.76	(3.75)	11.95	(3.79)	12.05	(3.78)	12.23	(3.94)	12.10	(3.84)
DOES	6.95	(2.44)	7.18	(2.59)	7.13	(2.59)	7.14	(2.63)	7.29	(2.75)
SBD	3.77	(1.26)	3.74	(1.20)	3.69	(1.16)	3.60	(1.07)	3.54	(0.99)
SHY	2.44	(1.18)	2.39	(1.08)	2.34	(0.98)	2.29	(0.89)	2.25	(0.82)
SWTD	8.18	(2.63)	8.04	(2.50)	7.78	(2.41)	7.49	(2.18)	7.25	(1.95)

Note: Participants included those who had complete SDSC data and were assigned to a latent class growth analysis class. DA = disorders of arousal/nightmares; DIMS = disorders of initiating and maintaining sleep; DOES = disorders of excessive somnolence; SA = suicide attempt; SBD = sleep breathing disorders; SDSC = Sleep Disturbance Scale for Children; SHY = sleep hyperhidrosis; SI = suicidal ideation; STB = suicidal thoughts and behaviors; SWTD = sleep-wake transition disorders.

^aIncludes participants who identified as Native Hawaiian, Pacific Islander, Alaskan Native, American Indian, or multiracial.

Sleep Profiles

Identification of sleep trajectories using LCGA resulted in 3 latent classes of sleep disturbance across childhood and early adolescence (Figure 1). Comprehensive participant characteristics by class membership across assessment time points are provided in Table S1, available online. In brief, class 1 (high-decreasing) comprised approximately 4.8% (n = 573) of the sample and was characterized by high initial sleep disturbance across all subscales at baseline, which significantly decreased over the 4-year follow-up period. Class 2 (moderate-increasing), comprising 11.8% (n = 1,410) of the sample, demonstrated moderate baseline sleep disturbance on sleep-wake transition disorders, sleep breathing disorders, and disorders of arousal/nightmares subscales, which continued to remain moderate over the 4-year follow-up, and high sleep disturbance on the disorders of initiating and maintaining sleep and disorders of excessive somnolence subscales, which increased over time. Class 3 (low-stable), representing the majority of participants (n = 9,881; 83.4% of the sample), consistently exhibited low levels of sleep disturbance across all assessment points in the current study. Estimated slopes and intercepts for each trajectory are detailed in Table 2.

Figure 1.

Predicted Sleep Disturbance Trajectories for Latent Class Growth Analysis (LCGA) Classes

Note:DA = disorders of arousal/nightmares; DIMS = disorders of initiating and maintaining sleep; DOES = disorders of excessive somnolence; SBD = sleep breathing disorders; SDSC = Sleep Disturbance Scale for Children; SHY = sleep hyperhidrosis; SWTD = sleep-wake transition disorders. Please note color figures are available online.

Table 2.

Sleep Trajectory Class Estimates and Intercepts

SDSC score	Classa	Intercept			b
		Estimate [95% CI]	SE	p	Estimate [95% CI]	SE	p
DA	1	4.61 [4.52, 4.71]	0.047	<.001	−0.60 [−0.66, −0.55]	0.027	<.001
	2	3.34 [3.13, 3.56]	0.109	<.001	−0.06 [−0.09, −0.02]	0.016	<.001
	3	2.34 [2.18, 2.51]	0.084	<.001	−0.08 [−0.09, −0.07]	0.004	<.001
DIMS	Class 1	15.54 [15.22, 15.86]	0.163	<.001	−0.21 [−0.25, −0.17]	0.020	<.001
	Class 2	15.40 [15.22, 15.57]	0.091	0.115	0.33 [0.29, 0.37]	0.021	<.001
	Class 3	13.69 [13.55, 13.82]	0.069	<.001	0.03 [0.02, 0.04]	0.004	<.001
DOES	Class 1	9.50 [9.30, 9.70]	0.103	<.001	−0.23 [−0.27, −0.19]	0.020	<.001
	Class 2	9.27 [9.11, 9.42]	0.080	0.004	0.38 [0.34, 0.42]	0.020	<.001
	Class 3	7.82 [7.70, 7.94]	0.061	<.001	0.01 [−0.00, 0.01]	0.004	0.167
SBD	Class 1	5.12 [5.01, 5.23]	0.054	<.001	−0.40 [−0.44, −0.36]	0.022	<.001
	Class 2	3.92 [3.74, 4.09]	0.089	<.001	0.04 [0.01, 0.07]	0.015	0.009
	Class 3	3.23 [3.09, 3.37]	0.072	<.001	−0.07 [−0.08, −0.06]	0.004	<.001
SHY	Class 1	6.07 [5.91, 6.23]	0.082	<.001	−1.19 [−1.25, −1.14]	0.029	<.001
	Class 2	0.73 [0.48, 0.99]	0.128	<.001	0.14 [0.12, 0.17]	0.013	<.001
	Class 3	0.37 [0.13, 0.62	0.126	<.001	−0.03 [−0.04, −0.03]	0.004	<.001
SWTD	Class 1	12.19 [11.98, 12.40]	0.108	<.001	−0.65 [−0.70, −0.61]	0.023	<.001
	Class 2	10.93 [10.75, 11.12]	0.095	<.001	−0.06 [−0.10, −0.03]	0.018	0.001
	Class 3	9.34 [9.20, 9.48]	0.071	<.001	−0.13 [−0.14, −0.13]	0.004	<.001

Note: Intercepts index baseline mean scores on each SDSC subscale; b coefficients index yearly changes in SDSC scores; p values are derived from Wald tests for each parameter. DA = disorders of arousal/nightmares; DIMS = disorders of initiating and maintaining sleep; DOES = disorders of excessive somnolence; SBD = sleep breathing disorders; SDSC = Sleep Disturbance Scale for Children; SHY = sleep hyperhidrosis SWTD = sleep-wake transition disorders.

^aClass 1 = high decreasing; class 2 = moderate increasing; class 3 = low stable.

Associations Between Sleep Disturbance Trajectories and STB

Occurrence rates of STB at the 2-, 3-, and 4-year follow-up assessment time points by sleep trajectory class are provided in Table S2, available online. Examining associations between sleep trajectory class and STB revealed that youth in the high-decreasing sleep disturbance class (class 1) were more likely to report higher levels of STB severity (OR 1.75, 95% CI [1.03, 2.96], p = .037) during the 2-, 3-, and 4-year follow-ups compared with youth in the low-stable class (class 3). Furthermore, youth in the moderate-increasing sleep disturbance class (class 2) showed an even greater likelihood of reporting more severe STB level, with substantially higher odds (OR 2.34, 95% CI [1.71, 3.18], p < .001) relative to the low-stable class. Direct comparisons between the high-decreasing class and moderate-increasing class showed that youth with increasing sleep problems over time (class 2) tended to report higher levels of STB severity, though this difference was not statistically significant compared with youth in initially severe but improving sleep profiles (class 1; OR 1.33, 95% CI [0.77, 2.38], p = .316).

Several covariates in the model were also found to be associated with higher likelihood of reporting more severe levels of STB (see Table S3, available online, for full model results including covariates). Specifically, age was positively related to STB risk, with each 1-SD unit increase in age associated with greater likelihood of reporting more severe STB levels (OR 1.38, 95% CI [1.14, 1.68], p < .001). Pubertal status showed a similar pattern, with youth reaching late puberty being more likely to report more severe levels of STB (OR 3.38, 95% CI [2.13, 5.45], p < .001), and youth in postpuberty showing the greatest likelihood (OR 4.62, 95% CI [2.43, 9.01], p < .001), relative to youth in early puberty. Male sex at birth was associated with lower likelihood of reporting more severe levels of STB (OR 0.54, 95% CI [0.39, 0.73], p < .001) compared with female sex at birth. Additionally, experiences of relational (OR 1.16, 95% CI [1.09, 1.23], p < .001), reputational (OR 1.15, 95% CI [1.08, 1.22], p < .001), and overt (OR 1.29, 95% CI [1.19, 1.40], p < .001) peer victimization all were significantly associated with greater likelihood of reporting higher levels of STB severity. Notably, family history of suicide and socioeconomic status (ie, income-to-needs ratio) were not significantly associated with suicidality outcomes in this model.

Sensitivity Analyses Incorporating CBCL Anxious/Depressed Scale Scores

To assess the specificity and robustness of the association between STB and sleep trajectory classes, the mean-centered CBCL anxious/depressed subscale score and its interaction with sleep trajectory class were incorporated into a Bayesian ordinal regression model paralleling the primary analyses. In this model, higher anxious/depressed symptoms were significantly associated with greater STB severity (OR 1.11, 95% CI [1.09, 1.13], p < .001), and sleep trajectory main effects were no longer significant (p > .05). However, a significant interaction (class 1 × CBCL anxious/depressed score; OR 0.95, 95% CI [0.90, 0.99], p = .036) indicated that the anxious/depressed–STB association was weaker in the high-decreasing trajectory (class 1) relative to the low-stable reference class (class 3). See supplementary results for full model estimates and details in Supplement 1, available online.

Discussion

This study leveraged data from the ABCD Study to examine trajectories of sleep disturbance during the transition from childhood to adolescence and whether specific sleep profiles were associated with STB risk. The 3 latent sleep profiles identified (low-stable, high-decreasing, and moderate-increasing) demonstrated that childhood sleep disturbances follow different trajectories across development. The majority of youth were classified in the low-stable group, characterized by minimal sleep problems that remained consistently low over time, which likely reflects typical sleep development. The high-decreasing class exhibited elevated sleep disturbance across all SDSC subscales at baseline, which improved steadily over time but remained in the low to moderate range during adolescence. This pattern is in line with prior research showing that early sleep problems during childhood often decline throughout development into adolescence.⁴¹ In contrast, youth in the moderate-increasing class had moderate sleep disturbance that worsened over time, particularly for disorders of initiating and maintaining sleep and disorder of excessive somnolence.

We found that specific trajectories of sleep disturbance during childhood and early adolescence were associated with greater likelihood of reporting higher levels of STB severity. Sleep profiles with greater disturbance (high-decreasing, and moderate-increasing) were linked to elevated likelihood of reporting higher levels of more severe STB compared with the low-stable group. Although youth in the high-decreasing class demonstrated improvements in sleep disturbance over time, they still showed higher likelihood of more severe STB, suggesting that early severe sleep disturbance may have lasting implications for suicide risk. Notably, youth in the moderate-increasing class had the highest risk with more than double the likelihood of having higher levels of STB severity relative to the low-stable class, highlighting the potential impact of worsening sleep patterns throughout development. Whereas the likelihood of more severe STB was higher in the moderate-increasing class than the high-decreasing class, the difference was not statistically significant. Importantly, both sleep disturbance trajectories were characterized by disturbances in disorders of initiating and maintaining sleep and disorders of excessive somnolence, indicating that early and worsening sleep problems in these specific domains may contribute to elevated suicide risk.

To assess both the specificity and the robustness of the observed association between sleep disturbance trajectories and STB, we conducted a sensitivity analysis incorporating depression, given its well-established links to both sleep problems and suicide risk. Including depression attenuated the main effect of sleep trajectory class on STB severity, indicating that depressive symptoms may account for part of the association. Importantly, the association was not fully explained by depression. Notably, the association between depression and STB was weaker in the high-decreasing sleep disturbance class, suggesting that elevated STB risk in this group may involve additional or distinct mechanisms beyond depressive symptoms. These results underscore the complex interplay between sleep disturbance and depression in the context of suicide risk. It is likely that these 2 factors share common biological or psychosocial pathways that contribute to STB.⁴² This possibility is further supported by a recent systematic review, which reported that among individuals with depression, sleep disorders, particularly nightmares and insomnia, were associated with increased risk of STB.⁴³ These findings highlight the importance of simultaneously assessing sleep disturbances and depressive symptoms when screening for suicide risk in childhood and adolescence.

These findings build on prior studies examining sleep trajectories using data from the ABCD Study. Previous studies have identified profiles characterized by sleep onset and maintenance difficulties, as well as higher overall sleep disturbance, that were associated with elevated internalizing and externalizing psychopathology symptoms.²⁰ Additionally, low sleep duration and poor sleep efficiency have been linked to behavioral problems, including rule-breaking behaviors, attention issues, and social problems.²⁰ In the present study, we used LCGA and identified a 3-class model as the best fit for sleep disturbance trajectories spanning 5 time points from childhood to adolescence. This differs from prior work using latent profile analysis over 2 time points to identify 4 sleep disturbance profiles.²⁰ The difference in model structure likely reflects methodological distinctions, including the extended developmental window and the trajectory-focused approach by LCGA. By examining changes across 5 time points, our findings provide a more dynamic understanding of how sleep disturbance patterns change throughout development.

These results are also consistent with prior findings from adolescent and adult studies reporting associations between sleep disturbance and STB,^14,15 particularly difficulties with initiating and maintaining sleep. Furthermore, problems with nightmares and excessive daytime somnolence at youth ages 9 and 10 have previously been linked to increased risk of STB within 2 years in the ABCD Study.²¹ Our study extends these findings over a 5-year period, identifying distinct developmental trajectories of sleep disturbance and offering greater specificity in pinpointing which patterns and subtypes of sleep problems, such as disorders of initiating and maintaining sleep and disorders of excessive somnolence, that may contribute to STB risk in adolescence. These results also indicated that several covariates were meaningfully associated with suicidality, aligning with and expanding on prior research. Consistent with previous studies, we found that age²⁵ and pubertal development²⁹ were associated with suicidality with older youth showing greater risk. Although youth assigned male at birth had a lower likelihood of reporting more severe levels of STB, this finding should be interpreted with caution given that the outcome variable combined STB into an ordinal scale. This approach does not allow for distinctions between STB, which have shown different associations with sex.²⁵ In addition, peer victimization was a strong predictor, reinforcing established evidence that negative peer experiences contribute to increased suicide risk during adolescence.³⁴ In contrast to several previous studies, we did not observe significant associations between suicidality and socioeconomic status^26,27 or family history.^32,44

These findings have important implications, as insufficient sleep and poor sleep patterns have become a common issue for youth worldwide. Sleep has not typically been targeted for suicide risk; however, sleep is modifiable, and sleep interventions may inform suicide prevention efforts. Treatments such as cognitive behavioral sleep interventions⁴⁵ have been shown to be effective in improving sleep in adolescents. Further, there is preliminary support that interventions such as triple chronotherapy (1 day of sleep deprivation followed by 3 days of sleep phase advancement and daily bright light therapy) reduce both depression and suicide risk in adolescents with depression.⁴⁶ Few studies have examined the effectiveness of sleep interventions for reducing suicide risk, especially in youth. Targeted sleep treatment may improve and prevent symptoms of psychopathology. Therefore, clinicians need to also consider sleep throughout the assessment and treatment of youth exhibiting these symptoms, as sleep disturbances may increase STB.

In addition to the clinical implications of our findings, it is important to consider these results from a health equity perspective. Sleep disturbances and their associations with STB may disproportionately affect children and adolescents from minoritized and underserved populations due to social, environmental, and structural factors such as housing instability, neighborhood noise, and inequities in access to health care.⁴⁷ Early identification and intervention strategies should be tailored to meet the needs of diverse populations, and programs aimed at prevention of STB may benefit from addressing systemic determinants of sleep health, such as school start times and community resources, alongside individual-level interventions.

Whereas this study has several strengths, it is important to acknowledge its limitations. First, given past research from the ABCD Study showing low correspondence between youth- and caregiver-reported STB,⁴⁸ we used youth-reported STB in our analyses, as parents may not have awareness of youth STB. Moreover, the K-SADS suicidality questions were not completed by parents during the 1- and 3-year follow-up visits. In addition, the evaluation of suicidality was limited to recent symptoms (past 2 weeks) and included K-SADS questions that did not capture other key factors of suicidality (eg, means, deterrents). Although modeling STB as an ordinal outcome allowed us to examine levels of STB severity, this approach may obscure important distinctions between thoughts and behaviors regarding suicide. Given that the recent past 2-week time frame for STB assessment does not fully align with the 6-month retrospective measure of sleep disturbance, the limited temporal overlap may constrain the ability to capture the dynamic nature of STB and interpret temporal associations with precision. Moreover, the STB variable did not include self-harm, which may be associated with the risk of transitioning from thoughts of suicide to action. However, self-harm has been posited as a separate phenomenon⁴⁹ and may have a different association with sleep disturbances. Whereas sleep problems are known to have a bidirectional association with psychopathology during adolescence,⁵⁰ the precise nature of the association between sleep disturbance and depression remains unclear.¹³ It is challenging to disentangle whether sleep problems precede, contribute to, or exacerbate depression. Examining the combined effects of sleep disturbances and depression on STB may offer a more comprehensive understanding and improvement in prevention efforts, as these symptoms rarely occur in isolation.

A key limitation is the exclusive reliance on parent-reported sleep disturbances, as this was the only measure of youth sleep disturbance available across all time points. Although this allowed for the consistent measurement necessary for trajectory modeling, it may not fully reflect the youth’s sleep experience or capture sleep problems that are less observable to caregivers. Symptoms of depression were also reported by parents, as the subscale was not available from the abbreviated youth report version of the CBCL. Although the effect sizes observed in this study are modest, they are not unexpected given that the ABCD cohort represents a community sample and participants were not clinically referred.⁵¹ Whereas the findings are statistically significant, it is important to consider the clinical relevance. Depression remains a critical risk factor for suicide, and the inclusion of sleep disturbances may enhance the precision of suicide risk assessment and treatment.

Future research should address several limitations of the current study to further refine our understanding of how sleep disturbance contributes to suicide risk during adolescence. Incorporation of more nuanced measures of suicidality and self-injurious behavior could provide greater specificity in risk assessment. Given the bidirectional nature of sleep problems and psychopathology, especially depression, further longitudinal analyses using mediation and temporal modeling approaches are needed to clarify whether sleep disturbances precede, co-occur with, or exacerbate depressive symptoms that contribute to STB. Future studies would benefit from including more detailed and objective measures of sleep as well as examining biological, social, and environmental factors that contribute to sleep disturbances trajectories.

In conclusion, this study identified distinct developmental trajectories of sleep disturbance that are meaningfully associated with STB in adolescence. These findings underscore the importance of identifying and addressing early and/or worsening sleep problems as a potential target in suicide prevention strategies. Given the complex interplay between sleep, depression, and suicidality, integrated screening and intervention strategies are needed. Future research should continue to explore these pathways using more nuanced measures and longitudinal modeling to inform targeted prevention efforts.

CRediT authorship contribution statement

Rebekah S. Huber: Writing – review & editing, Writing – original draft, Visualization, Supervision, Resources, Project administration, Methodology, Investigation, Funding acquisition, Conceptualization. Mizan Gaillard: Writing – review & editing, Visualization, Validation, Software, Methodology, Investigation, Formal analysis, Data curation, Conceptualization. Sam A. Sievertsen: Writing – review & editing, Writing – original draft, Validation, Software, Methodology, Formal analysis, Data curation. Jiyoung Ma: Writing – review & editing, Methodology, Formal analysis, Conceptualization. Sara Shao: Writing – review & editing, Validation, Software, Methodology, Formal analysis, Data curation. Dani Y. Del Rubin: Writing – review & editing, Software, Methodology, Formal analysis, Data curation. Scott A. Jones: Writing – review & editing, Visualization, Validation, Supervision, Software, Project administration, Methodology, Formal analysis, Data curation. Anthony R. Hill: Writing – review & editing, Writing – original draft. Rachel Bartholomeusz: Writing – review & editing, Writing – original draft. Erin C. McGlade: Writing – review & editing, Project administration. Perry F. Renshaw: Writing – review & editing, Funding acquisition. Deborah Yurgelun-Todd: Writing – review & editing, Funding acquisition. Bonnie J. Nagel: Writing – review & editing, Resources, Investigation, Funding acquisition.