Neurocognitive and Temperament Predictors of Changes in Child and Adolescent Suicidal Thoughts and Behaviors for Youths With and Without Attention-Deficit/Hyperactivity Disorder

Abstract

Suicidal thoughts and behaviors (STBs) are an urgent concern, with elevated risk in attention-deficit/hyperactivity disorder (ADHD). Drawing on the developmental psychopathology framework and socio-ecological model of suicide risk, this study examined trajectories of STBs in youth with and without ADHD and their prediction via individual and interpersonal factors. Data from 849 participants (n_ADHD=509), aged 7–13, were collected over 13 years using multiple informants (parent, teacher, youth) and methods (cognitive tasks, self-report, five-minute speech sample). Trajectories were best captured by a piecewise latent growth curve model representing STBs in childhood and adolescence. Higher parental criticism predicted persisting childhood STBs. In adolescence, female sex and less temperamental fear predicted increasing adolescent STBs, while greater temperamental openness predicted increasing adolescent STBs for females and youth with lower baseline working memory. ADHD symptoms alone did not drive risk; instead, individual and interpersonal factors emerged as key predictors. Prevention efforts should integrate these multidimensional considerations.

Suicide is a leading cause of death among youth and emerging adults in the United States, accounting for 17–19% of deaths in 2022 (CDC, 2022). Suicide rates increased by more than 50% over the past two decades (Curtin, 2020; Keyes & Platt, 2024), and nearly 10% of U.S. high school students report a suicide attempt in the past year (Verlenden et al., 2024). Youth with Attention-deficit/hyperactivity disorder (ADHD) experience higher rates of both suicidal thoughts and behaviors (STBs) compared to their peers (Balazs & Kereszteny, 2017; Garas & Balazs, 2020; Liu et al., 2022; Nigg, 2013), including being over three times as likely to make a suicide attempt by age 18 (Chronis-Tuscano et al., 2010). Greater rates of co-occurring psychiatric conditions in ADHD offer some explanation, though the direct association between ADHD and STBs persists in adolescence after accounting mental health difficulties in most studies (Rother et al., 2025), highlighting the need to identify other predictors of this vulnerability.

Across clinical and community samples, predictive modeling of individual-level risk has identified between 17–41% of those who experience ideation go on to attempt suicide (Liu et al., 2022; McHugh, Corderoy, et al., 2019). Yet, most scales yield low positive predictive value (Carter et al., 2017), and there is little evidence that current approaches improve clinical decision-making or outcomes (Kessler et al., 2020). Multimodal approaches are needed to advance suicide risk research and directly link assessment to intervention, including the integration of clinical, cognitive, biological, and contextual factors (Kessler et al., 2020).

The socio-ecological model provides one such conceptualization, situating risk and resilience across nested levels of influence, including individual, interpersonal, community, and societal contexts (Prades-Caballero et al., 2025). The developmental psychopathology framework complements this view by examining how individual vulnerabilities and environmental stressors interact over time to shape distinct trajectories of STBs (Glenn et al., 2016; Séguin et al., 2014). Together, these frameworks underscore that suicide prevention requires both attention to mapping risk and also clarifying which predictors are most informative for whom and at what developmental stages, thereby narrowing targets for mechanistic study and informing prevention (Oppenheimer et al., 2022).

Suicidal Thoughts and Behaviors Across Development

A first step in improving the prediction of suicide risk is to clarify the course of STBs across development, particularly in high-risk populations. The largest longitudinal studies have used definitions that examine STBs as a combined dimension, identifying that both suicidal thoughts and suicidal behaviors can emerge early in life and meaningfully persist throughout development. In a sample of youth oversampled for depression, 11% of children ages 3 to 7 years old experienced STBs and 75% of these continued to do so at ages 7 to 12 years old (Whalen et al., 2015). Zhu et al. (2019) followed a sample of 715 students in China from middle childhood (3^rd and 4^th grade) into early adolescence (6^th and 7^th grade), with 7.1% following a trajectory of increasing suicidal ideation, 6.5% exhibiting a high but decreasing trajectory, and 86.4% remaining at near-zero levels. Prospective examination of the Whalen et al. (2015) childhood depression sample through age 19 has also identified varying patterns of STB change, with 6% demonstrating a steady and linear increase, 6% described as late onset with a sharp increase starting at age 10 that peaks around age 15 and then decreases, and 78% exhibiting persistently low levels (Whalen et al., 2022).

Multiple developmental trajectories for both suicidal thoughts and behaviors have also been identified through adolescence and young adulthood. In general population adolescents ages 11–19, high likelihood of suicidal ideation and moderate likelihood of suicide attempt each show a decreasing trajectory to near-zero levels over the subsequent decade, whereas moderate suicidal ideation tends to persist over this timeframe (Xiao & Lindsey, 2021). Adolescents recently discharged from a psychiatric inpatient unit also exhibit varied patterns of suicide ideation and attempt, including 1) increasing risk in adulthood (11%), 2) high in adolescence and moderate in adulthood (12%), 3) high in adolescence and decreasing in adulthood (33%), and 4) persistently low (44%) (Goldston et al., 2016). There is yet to be an examination of the emergence of, and changes in, STBs that considers youth with ADHD as a particularly high-risk group. It is also not yet clear if childhood STBs increase risk for persistent or recurring suicidal thoughts and behaviors in adolescence (Whalen et al., 2015), or if child and adolescent STBs reflect distinct vulnerabilities.

Predictors of Suicidal Thoughts and Behaviors

Individual Factors - Neurocognitive Development

Risk for STBs may be partially explained by neurocognitive features, with executive function differences often found in ADHD (Nigg, 2017) being related to short- and long-term outcomes including worse academic functioning (Fried et al., 2019), peer victimization (Chou et al., 2018), and substance use (Nigg et al., 2006). Two areas of cognitive difficulty often associated with ADHD, working memory (the ability to maintain and manipulate information in the focus of attention; (Kasper et al., 2012; Martinussen et al., 2005) and response inhibition (the ability to suppress one’s thoughts or actions to maintain goal-directed behavior; (Barkley, 1997), are also associated with STBs (Huang et al., 2022; Meza et al., 2016; Moniz et al., 2017; Richard-Devantoy et al., 2015).

In depressed patients, working memory difficulties are broadly associated with suicidal ideation (Huang et al., 2022) and differentiate those who attempt suicide within both clinical and community control groups (Richard-Devantoy et al., 2015). In a subset of the sample used for the current study, working memory impairments mediated ADHD’s relationship with the presence of suicidal ideation in a cross-sectional analysis of youth ages 7–12 years old (Bauer et al., 2018), but prospective prediction models have not yet been examined for this sample.

Facets of response inhibition may also differentiate risk for suicide attempts among depressed patients. Several studies have identified greater commission errors during a go-no-go task for depressed participants with a history of attempted suicide compared to those without, implicating a mechanistic risk via lack of behavioral inhibition (Moniz et al., 2017; Richard-Devantoy et al., 2016). Similarly, greater commission errors on the Continuous Performance Task (CPT) predict young adult self-harm or STBs in ADHD (Meza et al., 2016; Miller et al., 2012). Slower response time within response inhibition tasks has also been reported for youth with a history of self-harm or suicidal behavior via meta-analytic review (McHugh, Lee, et al., 2019), though this relationship is less clear for suicidal ideation in ADHD (Bauer et al., 2018).

Individual Factors - Temperament

Although cognitive features have been of primary interest in ADHD, differences in temperament may yield additional information about clinical risk (Goh et al., 2020; Karalunas et al., 2019). Temperament can be defined as differences in emotional reactivity and regulation, related to the broader construct of personality (Rothbart et al., 2014). Here, we focus on temperament traits as defined in Rothbart’s widely used model and measured by the Temperament in Middle Childhood Questionnaire (TMCQ; Simonds & Rothbart, 2004). This structure includes 17 temperament domains subsumed within three high-order factors: Surgency (related to extraversion and positive emotionality), Negative Affect (including fear, anger, sadness, and low soothability), and Effortful Control (related to attention control and behavioral inhibition).

Differences in these higher-order temperament domains may be relevant in understanding heterogeneity within ADHD and risk for STBs. As a group, adolescents with ADHD are characterized by greater surgency and negative affect and less effortful control compared to their non-ADHD peers (Krieger et al., 2019). Yet, youth with ADHD vary in their temperament profiles in ways that might help predict clinical outcomes. Between-child differences in negative affect prospectively predict worsening patterns of comorbidity in the sample used for the current study (Karalunas et al., 2014, 2019). Further, cross-sectional findings in community-sampled youth ages 6–9 suggest that higher negative affect is associated with suicidal ideation, with no differences between groups in surgency or effortful control (Sheftall et al., 2021).

Among the domains that contribute to negative affect, irritability (or anger dysregulation) has been of particular interest (Shaw et al., 2014). Within the general population, elevated childhood irritability is associated with increased risk for STBs in adolescence (Orri et al., 2019) and adults (Pickles et al., 2010). Notably, the cross-sectional association between irritability and suicidal ideation in adulthood may be specific to males (Skala et al., 2012). At least one study found that childhood irritability partially explains the cross-sectional relationship between ADHD symptoms and STBs (Levy et al., 2020) and children with both high irritability and more ADHD symptoms experience higher rates of STBs in adolescence (Galera et al., 2021). Within the ADHD sample reported on here, variation in irritability prospectively predicts high, persistent depression after accounting for other facets of negative emotion (Karalunas et al., 2023), but STBs have not been examined. Overall, current research suggests that high negative affect, and perhaps specifically irritability, may predict risk for STBs in ADHD. Questions remain about whether it predicts independently of cognitive differences and after accounting for characteristics of oppositional defiant disorder (Bauer et al., 2018), and whether it differentially predicts STBs in males and females.

Temperamental impulsivity is also associated with STBs and overlaps with core ADHD symptoms. Some literature suggests that impulsivity may be associated with greater likelihood of youth’s lifetime suicidal ideation (Janiri et al., 2021). Alternatively, impulsivity may increase capability for suicidal action but not suicidal ideation (Hadzic et al., 2020). Impulsivity is also often characteristic of ADHD and may be related to greater likelihood of suicide attempts in this population (Conejero et al., 2019), and particularly for young women with ADHD (Swanson et al., 2014). Thus, it is not clear if impulsivity is related to early development of ideation or specific to suicidal action, and if these relationships vary between those with and without ADHD or by sex.

Related work in adolescent and adult personality may also provide insight into risk for STBs, as an extension of youth temperament. Several studies have reported a link between STBs and openness to experiences, a personality trait described as being imaginative, creative, and liking to try new things. For example, Pawar & Palve (2021) identified a cross-sectional association between greater suicidal ideation and low adolescent openness, which may include a tendency to be dogmatic, resistant to change, and close minded. At the other extreme, risk for a future suicide attempt in young adulthood may be greater for adolescents with prior ideation who are high in openness, including a tendency to be imaginative and in touch with feelings (Mars et al., 2019a). Mixed and null findings regarding the relationship between openness and suicidal ideation (Baertschi et al., 2018) may be partially explained by evidence that extremes of openness, either high or low, are associated with greater STBs when in the context of high neuroticism/emotional instability (Planellas & Calderón, 2022).

Individual Factors - Sociodemographics

Several individual sociodemographic factors are associated with STBs across both childhood and adolescence, including impulsivity, anxiety, depression, and externalizing symptoms; whereas other features appear more stage-specific. In childhood, greater STBs are associated with gender (boys), ODD diagnosis, and higher impulsivity or higher externalizing symptoms (Whalen et al., 2015). Other studies have idenfied that greater social anxiety and depression predict risk for childhood STBs (Whalen et al., 2022; Zhu et al., 2019). In adolescence, additional risk factors include lower academic achievement and female biological sex (Goldston et al., 2016; Miranda-Mendizabal et al., 2019; Whalen et al., 2022; Zhu et al., 2019), as well as increased risk among sexual minority and Black or other racially minoritized youth (Xiao & Lindsey, 2021). Greater risk for suicide attempts in adolescence is also associated with receiving public assistance (Xiao & Lindsey, 2021) and, in high income countries, lower levels of parental education (Chen et al., 2022).

Differential risk models by biological sex suggest that suicidal ideation remains high for some females through adolescence but decreases for males (Adrian et al., 2016). In the context of ADHD, several studies have indentified that boys with ADHD in childhood are at increased risk for STBs in adolescence, whereas girls with ADHD do not differ from non-ADHD girls in STB risk (Forte et al., 2020; Lin et al., 2024). However, both male and female adolescents and young adults with ADHD may indeed be at greater risk for suicide attempts compared to their peers (Huang et al., 2018)

Interpersonal Factors

Interpersonal factors are a well-established source of risk for STBs in youth, including parent mental health problems, harsh parenting, and adverse family experiences (Orri et al., 2022; Pickles et al., 2010; Whalen et al., 2015, 2022; Xiao & Lindsey, 2021). Adolescence is a developmental period when social connections gain importance and sensitivity to rejection intensifies (Andrews et al., 2021; Veenstra & Laninga-Wijnen, 2023). Compared to their peers, youth with ADHD have fewer reciprocal friendships and these friendships tend to be of worse overall quality and include more conflict (for review: Neprily et al., 2025). In combination with the emotion regulation difficulties often found in ADHD (Christiansen et al., 2019), these patterns suggest that youth with ADHD may be particularly vulnerable to relational conflicts. Disruption of social relationships play a critical role in suicide risk with interpersonal conflict strengthening the association between non-suicidal self-injury and STBs in both clinical and community samples (Horváth et al., 2020). In ADHD, bullying victimization is associated with greater depression (Simmons & Antshel, 2021) and STBs (Yen et al., 2014) and partially mediates the relationship between ADHD and STBs (Lin et al., 2024).

Family factors are also implicated in risk for STBs. Youth with ADHD experience greater parental criticism (expressed emotion) than peers, which predicts persistence of difficulties across development (Musser et al., 2016). In cross-sectional studies, parental criticism and rejection are associated with youth internalizing problems (Rea et al., 2020) and adolescent STBs (Wedig & Knock, 2007). Prospectivly, adolescents who go on to experience increasing suicidal ideation report greater parental rejection (characterized as hostility, criticism, and lack of warmth) compared to those whose ideation remains low (Shen et al., 2024). Youth who blame themselves for inter-parental conflict also exhibit greater internalizing symptoms and mental health problems in adolescence (Fear et al., 2009; Olatunji & Idemudia, 2021), with male adolescents experiencing more self-blame compared to females (Hafsa et al., 2021). Perceived lack of family support or low satisfaction with family relationships may further mediate risk in ADHD (Chen et al., 2019; Chou et al., 2016). Yet few studies have examined how peer and family factors interact with child characteristics to predict STBs, or how these influences combine with neurocognitive differences.

Societal Factor – COVID-19 Pandemic

The societal-level influence of the coronavirus disease 2019 (COVID-19) pandemic may also impact individual-level risk for STBs. Pandemic-related disruptions to schooling, services, and social relationships have been linked to increased stress and mental health difficulties for many youth, with particularly pronounced effects for those with pre-existing conditions such as ADHD (Becker et al., 2020; Miranda et al., 2020; Sibley et al., 2021). This context is critical to consider alongside individual and interpersonal predictors of risk.

Present Study

Many risk factors for STBs in childhood and adolescence have been identified, though prior work has not adequately distinguished between predictors of overall risk (intercept) and those that predict developmental change over time (slope). Further, the role of ADHD is yet to be thoroughly considered in these risk models. Guided by socio-ecological and developmental psychopathology perspectives, the present study uses latent curve models in a large sample of youth with and without ADHD to 1) identify the developmental trajectory of STBs across middle childhood and adolescence and 2) test prospective prediction of individual differences in these trajectories via a set of theoretically informed and multifaceted predictors. By integrating multiple levels of influence with attention to developmental timing, this approach aims to clarify which factors predict persistent versus changing patterns of risk and infom prevention and intervention targets.

Transparency and Openness

Preregistration

There was no preregistration of the present study.

Data, Materials, Code, and Online Resources

Preprint and analytic code are available at https://doi.org/10.17605/OSF.IO/NHKM3.

Reporting

We report how we determined our sample size, all data exclusions, all manipulations, and all measures in the study. For additional details about the cohort please see Nigg et al. (2023).

Ethical Approval

This paper describes secondary analysis of previously-collected data that were de-identified before being shared with the first author. The Oregon-ADHD-1000 cohort was collected under IRB protocol #00002560 through Oregon Health & Science University; the first author was added to this IRB to conduct the present analyses.

Method

Participants

Data were drawn from the Oregon-ADHD-1000 cohort, here we include the 849 participants (ages 7–13; n_ADHD = 509) who were followed longitudinally across 13 timepoints. The sample was 81% White, non-Hispanic and 62% biological males, consistent with ADHD sex distributions. Families were recruited from the community with oversampling for ADHD (Nigg et al., 2023). Sample demographics are presented in Table 1.

Table 1.

Participant Baseline Demographics

	All (n = 849)		ADHD (n = 509)		non-ADHD (n = 340)
	n	%	n	%	n	%
Age M(SD)	9.44 (1.56)		9.51 (1.53)		9.34 (1.60)
Sex at Birth
Male	525	61.84%	359	70.53%	166	48.82%
Female	324	38.16%	150	29.47%	174	51.18%
Race
Am. Indian/AL Native	2	0.24%	1	0.20%	1	0.29%
Asian/East Indian	15	1.77%	11	2.16%	4	1.18%
Black	20	2.36%	19	3.73%	1	0.29%
White/Middle Eastern	725	85.70%	425	83.50%	300	88.24%
Biracial/Multiracial	84	9.93%	50	9.82%	34	10.00%
Missing	3	0.35%	3	0.59%	0	0.00%
Ethnicity
Hispanic/Latino	52	6.12%	29	5.70%	23	6.76%
Non-Hispanic/Latino	796	93.76%	479	94.11%	317	93.24%
Missing	1	0.12%	1	0.20%	0	0.00%
Family Income
< $50,000	208	24.50%	142	27.90%	66	19.41%
$51,000 - $100,000	349	41.11%	202	39.69%	147	43.24%
$100,000 - $150,000	165	19.43%	94	18.47%	71	20.88%
$150,000+	69	8.13%	41	8.06%	28	8.24%
Missing	58	6.83%	30	5.89%	28	8.24%
Highest Parent Education
High School	22	2.59%	13	2.55%	9	2.65%
Some College	220	25.91%	157	30.84%	63	18.53%
4-year degree	289	34.04%	173	33.99%	116	34.12%
Graduate degree	308	36.28%	160	31.43%	144	42.35%
Missing	10	1.18%	6	1.18%	4	1.18%
ADHD Rating Scale M(SD)
Parent	20.62 (15.10)		30.41 (10.38)		6.02 (7.16)
Teacher	17.23 (14.47)		25.71 (11.97)		4.18 (5.50)

Note. Am. Indian/AL Native = American Indian/Alaska Native. ADHD Rating Scale total raw scores prior to standardization

Procedure

Following screening, youth completed a single, in-person clinical eligibility visit including brief cognitive testing (WISC-IV), semi-structured clinical interview (Kiddie Schedule for Affective Disorders and Schizophrenia (K-SADS-PL; Puig-Antich & Ryan, 1986), multi-informant questionnaires, and clinical observation. Parent and teacher measures included the Conners Rating Scales-Revised (CRS-R; Conners, 2003), Strengths and Difficulties Questionnaire (SDQ; Goodman, 1997), and the ADHD Rating Scale (ADHD-RS; DuPaul et al., 1998). A board-certified child psychiatrist (over 25 years’ experience) and licensed child neuropsychologist (>10 years’ experience) independently reviewed data and assigned diagnoses using a best-estimate procedure (Kostn & Rounsaville, 1992) with inter-rater reliability > .80 for all diagnoses with base rate >5%. Disagreements were conferenced to reach consensus. Research diagnoses were made according to both DSM-IV and DSM-5 criteria for ADHD (including subtype), mood disorders, anxiety disorders, disruptive behavior disorders, and learning disabilities.

To count ADHD symptoms, the clinicians used the following rule: If both parent and teacher ratings exceeded T ≥ 60 on at least one ADHD scale and both rated at least 3 symptoms as “often” or “very often” on the ADHD rating scale (or for parents, were counted present on the K-SADS), the “or” algorithm could be employed (Lahey et al., 1994). When either of the informant’s rating fell below this mark, and clinicians judged that this was not explained by successful medication treatment at school, then the case was rejected as failing to meet the DSM requirement of substantial symptoms present in more than one setting. All other DSM criteria were enforced (a) impairment (identified on K-SADS and SDQ), (b) onset prior to age 7, (c) persistence > 1 year, and (d) not better accounted for by comorbid conditions, trauma history, or other confounds.

These best-estimate research diagnoses were combined with parent-reported medical history to determine study eligibility. Baseline exclusionary criteria were intellectual disability (estimated IQ < 75), psychosis, mania, Tourette’s syndrome, autism, non-stimulant psychotropic medications, history of seizures or head injury, and discrepancies in parent-teacher ratings that precluded clinical diagnostic confirmation. A current major depressive episode was also exclusionary at baseline, due to the profound effects of depression on cognition, which would have precluded answering some of the study’s main questions about associations of ADHD with cognitive abilities. A history of major depression was not exclusionary for study participation and depression was not exclusionary in follow up visits. As expected, depression developed in many participants over 12 subsequent timepoints (see Table S1), enabling study of depressed mood and related outcomes over time. Please see Nigg et al. (2023) for additional details regarding participant eligibility assessments and diagnostic determinations.

Measures

Outcome - Suicidal Thoughts and Behaviors

The highest level of STBs endorsed by either youth or parent at each visit was coded using a 5-point Likert scale based on items from the parent-report K-SADS-PL (all ages), youth-report K-SADS-PL (ages 15–17) and the youth self-report Structured Clinical Interview for DSM Disorders (SCID, ages 18+; First & Gibbon, 2004), one item from the Children’s Depression Inventory (CDI; Kovacs, 2011), and two items from the Youth/Adult Self Report questionnaires (YSR/ASR, ages 11+; Achenbach & Rescorla, 2001; Achenbach & Rescorla, 2003).

Suicidal behavior (4) was coded based on endorsement of this behavior on the K-SADS-PL or SCID, or if YSR/ASR item 18 (“I deliberately try to hurt or kill myself”) was rated as a ‘2’ (“Very True or Often True”). Suicidal intent (3) was coded based on endorsement of this behavior on the K-SADS-PL or SCID, or if CDI item 9 was endorsed as a ‘2’ (“I want to kill myself”), or if YSR/ASR item 18 was rated as a ‘1’ (“Somewhat or Sometimes True”). Suicidal ideation (2) was coded based on endorsement of this behavior on the K-SADS-PL or SCID, or if CDI item 9 was rated as a ‘1’ (“I think about killing myself but would not do it”), or if YSR/ASR item 91 (“I think about killing myself”) was endorsed (either 1 or 2). Thoughts of death/own death (1) was coded based on endorsement on the K-SADS-PL or SCID. Respondents who did not endorse any of the above items via parent or youth report were coded as none (0) at each timepoint.

This operationalization of STBs was created to include as much data as possible and noting that false-positive reports for suicidal thoughts or behaviors are unlikely in a research study. Agreement between youth and parent reports was calculated using Gwet’s AC₂ for each age with at least paired 30 observations (McHugh, 2012), revealing substantial to excellent agreement (see Table S2). Temporal stability of STB reports was also excellent according to youth (AC₂ = .87), parents (AC₂ = .83), and for the combined multi-informant variable (AC₂ = .86).

Individual Factors

Neurocognitive Development

Youth completed a series of laboratory measures testing neurocognitive performance. Within the present analyses, we examined Working Memory (i.e., WISC-IV digit span forward/backward; CANTAB spatial span forward/backward; N1-back/N2-back) and Response Inhibition (i.e., D-KEFS Color and Word Naming Interference Trial residual after score regressed on color and word naming scores, D-KEFS Trail Making Test Letter-Number Sequencing Trial residual after score regressed on number and letter sequencing scores, Stop Signal Reaction Time). See Nigg et al. (2018) for more details about latent variable construction and validation.

Temperament

Caregivers completed the Temperament in Middle Childhood Questionnaire (Simonds & Rothbart, 2004), originally comprising 157 items rated on a 5-point Likert scale. For the present analyses, we used a published revision of the TMCQ lower-order solution (Kozlowski et al., 2022) that addresses limitations in replicability and psychometrics of the original form (Kotelnikova et al., 2017; Lipska et al., 2022; Nystrom & Bengtsson, 2017). This revision retains 90 of the original items across 12 factors, labeled Activity, Affiliativeness, Anger, Assertiveness, Attentional Focus, Fear, Pain Sensitivity, Impulsivity, Reading, Openness, Perceptual Sensitivity, Sadness, and Shyness. Within this solution, internal consistencies are acceptable for the retained factors (ω’s ≥ .76), with all but two factors greater than .80. Notably, we chose to also include the fear scale (ω = .68, α = .73) considering its theoretical importance to the research question and omitted the reading scale due to limited conceptual meaning (see original paper for more discussion of these scales).

ADHD Symptoms

Child ADHD characteristics were measured via parent- and teacher-report on the ADHD Rating Scale (ADHD-RS; DuPaul et al., 1998). Respondents rate items on a four-point Likert scale: Never/Rarely, Sometime, Often, and Very Often. Internal consistency was excellent for parents (ω = .96) and teachers (ω = .96). Parent and teacher total raw scores were standardized within the sample and averaged to form a composite measure of ADHD characteristics, with informant scores strongly correlated (r = .627, p < .001).

Caregivers’ report of their own ADHD characteristics was measured by the Conners’ Adult ADHD Rating Scales ADHD Index T-score (CAARS; Conners et al., 1998). Internal consistency was acceptable (ω = .83).

Demographics

Demographic information was provided via a caregiver questionnaire, including youth age, biological sex, race/ethnicity, combined family annual income, and caregivers’ highest level of education. A socioeconomic status (SES) composite variable was created as the average of the family income and parent education variables, which were each standardized within the sample.

Interpersonal Factors

Family Environment

Family environment was measured using three variables designed to capture different aspects of parenting style, support, and conflict. The 9-item Family Environment Scale, conflict subscale (FES-CS) was used as a caregiver-report measure of the degree to which family members express anger and interactions that stimulate anger (Moos, 1990).

Caregivers also completed the Five Minute Speech Sample (FMSS), providing an audio-recorded description of their child and relationship with their child for five minutes (Gottschalk & Gleser, 1969). Two independent expert raters, blind to data about the child’s clinical features, coded the FMSS transcripts and tapes following the standard protocol (Magaña et al., 1986). Inter-rater reliability for 60 dual-coded tapes and transcripts was k = 0.78, indicating substantial agreement. Professional coding of caregiver’s expressed emotion (EE) was completed for 574 of the 810 available recordings. The present study focuses on the number of critical statements (EE-Criticism) made by the parent about the child, which ranged from 0–7 within the present sample.

Finally, youth completed the Children’s Perception of Inter-parental Conflict scale for Younger children (CPIC-Y), including 5 true/false questions that assess child self-blame for parental conflict, which was the primary outcome here (Grych et al., 1992). Internal consistency was evaluated using tetrachoric correlations given the binary item format, which indicated acceptable reliability (ordinal ω = .80).

Peer Relationships

Peer relationship difficulties were measured via parent- and teacher report on the Strengths and Difficulties Questionnaire, Peer Relationship Problems Subscale (SDQ-PR; Goodman, 1997). Respondents rate items on a three-point scale: Not True, Somewhat True, or Certainly True. Given this three-point scale, internal consistency was estimated using ordinal reliability indices based on polychoric correlations, which were within the acceptable range for parents (ordinal ω = .79) and teachers (ordinal ω = .83). Parent and teacher total raw scores were standardized within the sample and averaged to form a composite measure of social connection and related difficulties, with informant scores moderately correlated (r = .483, p < .001).

Societal Factor

COVID-19 Pandemic

A binary variable indicated if data collection occurred before (0) or during (1) the pandemic based on a cutoff of March 11, 2020, which is the date the World Health Organization declared COVID-19 to be a pandemic.

Analytic Plan

Analyses were run in Mplus version 8 (Muthén & Muthén, 2017). To examine trajectories, the STB variable was transformed from a wave-based structure (e.g., Year 1, Year 2, […] Year 13) to age-based variables (e.g., age 7, age 8, […] age 23). See online supplement for details. Information on age of onset and offset for STBs was used to fill the age-based data matrix when possible (Horwitz et al., 2015). Several iterations of modeling were attempted with the tails of the distributions condensed as needed to optimize model fit (e.g., combining the highest STBs endorsed at ages 7 and 8).

A longitudinal growth model was used to examine STB change across age, accounting for repeated measurement and missingness (FIML). Model fit was evaluated using the following cutoffs: Tucker-Lewis index (TLI) ≥ 0.90, comparative fit index (CFI) ≥ 0.90, and root-mean square error of approximation (RMSEA) ≤ 0.08 and preferably ≤ 0.05 (Hu & Bentler, 1999). A cutoff of < 0.08 for the standardized root-mean-square residual (SRMR) was also considered but not solely used for model rejection as it is not well-suited for growth models (Wu et al., 2009). To improve model fit, correlated residuals were added within the same variable over time based on modification indices. Model comparisons used chi-square difference tests based on log-likelihood values and scaling correction factors from the MLR estimator. Following initial model selection, COVID variables were entered into the longitudinal model as time-varying covariates; their correlations were set to 0 with each growth factor but permitted to covary with each other given interrelation of pandemic occurrence across timepoints.

From this model, individually varying intercepts and slopes were exported and used as outcomes to identify predictors of change in STB trajectories. Linear regression models were conducted using hypothesized predictors. Change in STBs (individually varying slopes) was modeled while accounting for missingness (FIML). All variables were continuous, except for the binary variable of biological sex (1 = Male, 2 = Female) and COVID-19 covariates. Continuous predictors were examined for normality assumptions relevant to FIML (i.e., skew < 2, kurtosis < 7; West et al., 1995); kurtosis within parent criticism (FMSS) and youth self-blame (Y-CPIC) variables was addressed using square root transformations. Temperament variables were entered as z-scores standardized to the sample mean. All models included clustering within family (using the CLUSTER command in Mplus) to statistically account for similarities among siblings (n = 152 families with multiple youth, representing 697 families total).

To identify significant predictors of change in STBs, hypothesized variables measured at baseline were examined stepwise. An initial prediction model (M1) was conducted with covariates only—STBs intercept, youth biological sex, and youth total ADHD symptoms (ADHD-RS composite). Change in childhood STBs (S1) was also included as a predictor of change in adolescent STBs (S2) consistent with prior literature and study aims. Subsequent models included these covariates and separately tested each predictor set: Working Memory (M2), Response Inhibition (M3), Temperament (M4), Interpersonal Factors (M5), and Demographic variables (M6). In cases of multiple hypothesized predictors (i.e., Temperament, Interpersonal, Demographics), non-significant variables were removed using backward elimination until confidence reached the trend level (p’s < .1). A combined model (M7) was then run with predictors of change from each prior model to examine their collective impact. Finally, interactions between domains (e.g., temperament*sex) were examined within the combined model given prior work suggesting variability in findings by sex and ADHD status. Within this model, interactions were individually tested with confidence adjusted by setting the p threshold at .005. Significant interactions were added to the multivariate environment to form the final model (M8).

Results

Trajectories of Suicidal Thoughts and Behaviors

The age-binned STB variable included N = 3826 observations from ages 7–23 years old (see Table 2). Across the study timeframe, 43% of participants experienced suicidal thoughts and behaviors (see Table S3 for frequencies by type and diagnostic group). On average, youth with ADHD experienced greater STBs (M = 1.12, SE = 0.06) compared to those without ADHD (M = 0.90, SE = 0.07), Wald χ2(1) = 5.89, p = .015.

Table 2.

Suicidal Thoughts and Behaviors Endorsed by Participant Age

		Behavior		Intent		Ideation		Thoughts of [Own] Death		None
Age	Total Obs.	n	%	n	%	n	%	n	%	n	%
7	78	1	1%	1	1%	22	28%	7	6%	49	63%
8	235	0	0%	5	2%	51	22%	8	3%	171	73%
9	353	0	0%	6	2%	61	17%	8	2%	278	79%
10	377	2	1%	4	1%	52	14%	7	2%	312	83%
11	324	6	2%	3	1%	39	12%	6	2%	270	83%
12	315	5	2%	9	3%	36	11%	12	4%	253	80%
13	356	10	3%	17	5%	45	13%	8	2%	276	78%
14	363	12	3%	24	7%	64	18%	4	1%	259	71%
15	352	12	3%	28	8%	56	16%	5	1%	251	71%
16	288	8	3%	21	7%	50	17%	4	1%	205	71%
17	243	9	4%	16	7%	43	18%	8	3%	167	69%
18	212	9	4%	11	5%	42	20%	4	2%	146	69%
19	160	5	3%	6	4%	32	20%	3	2%	114	71%
20	107	2	2%	3	3%	10	9%	2	2%	90	84%
21	53	0	0%	2	4%	5	9%	0	0%	46	87%
22	9	0	0%	1	11%	0	0%	0	0%	8	89%
23	1	0	0%	0	0%	0	0%	0	0%	1	100%

Visual examination of reported STBs across ages revealed two distinct parts of the growth trajectory, so linear, quadratic, and piecewise models were tested. The piecewise growth model included a knot at age 10 years with a linear slope (S1) from ages 7/8 until age 10 and a second piece with both linear (S2) and quadratic (Q) terms from ages 10 through ages 19+ (see Figure 1). The piecewise model was selected to best represent the data due to not differing significantly from the non-piecewise linear model with a quadratic term (χ²(5) = 9.931, p = .077) and for enhanced interpretability of individually varying slopes in planned analyses. Piecewise model fit was further improved by correlating residuals between STBs experienced at ages 13–17 (χ²(4) = 28.009, p < .001) and adding COVID-19 time-varying covariates (χ²(42) = 226.153, p < .001). The final model was an acceptable fit to the data (CFI = .956, TLI = .952, RMSEA = .017, SRMR = .108) and indicated decreasing STBs from ages 7/8 until age 10 (S1: M = −.141, SD = .031, p < .001) but no overall change across the sample from age 10 through ages 19+ (S2: M = .019, SD = .035, p = .580; Q: M = .007, SD = .005, p = .179). There was minimal variance predicted within the quadratic term (M = .001, SD = .000, p < .001) so it was not used for additional analyses. See Table 3 for model fit details.

Figure 1.

Piecewise Growth Model of Suicidal Thoughts and Behaviors

Table 3.

Model Fit Comparisons

Shape	Growth Factors	RMSEA	CFI	TLI	SRMR	AIC	BIC	Additions
Piecewise*	I S1 S2 Q	0.017	0.956	0.952	0.108	7707.990	8049.052	CR + TVC
Piecewise	I S1 S2 Q	0.029	0.944	0.939	0.072	9482.711	9624.749	CR
Piecewise	I S1 S2 Q	0.039	0.893	0.889	0.082	9534.906	9658.005
Linear	I S Q	0.050	0.811	0.819	0.113	9615.827	9715.253
Piecewise	I S1 S2	0.055	0.768	0.778	0.144	9656.242	9755.669
Linear	I S	0.059	0.718	0.745	0.159	9707.580	9788.068

Note.

^*Selected for present analyses. I = intercept. S = linear slope. Q = quadratic. S1 = linear slope 1. S2 = linear slope 2. CR = correlated residuals. TVC = time varying covariates.

Predictors of Change in STBs

Baseline Model

The initial model identified greater ADHD symptoms as related to both decreasing childhood STBs (S1: β = .077, SE = .034, 95% Confidence Interval (CI); .010, .144) and increasing adolescent STBs (S2: β = .074, SE = .035, 95% CI: .007, .142) and also a significant effect for females experiencing increasing STBs in adolescence (S2: β = .182, SE = .034, 95% CI: .116, .249). Decreasing childhood STBs (S1) also predicted increasing adolescent STBs (S2: β = −.195, SE = .056, 95% CI: −.306, −.085). These variables were retained as covariates, and included along with the intercept, in all subsequent models.

Model Building

Stepwise modeling retained several variables related to decreasing childhood STBs (S1), including greater temperamental openness (S1: β = .071, SE = .033, 95% CI: .006, .135) and more peer difficulties (S1: β = .078, SE = .041, 95% CI: −.001, .161). Greater parental criticism was associated with childhood STBs that remained high and did not resolve (S1: β = −.068, SE = .034, 95% CI: −.135, −.002). Retained predictors of increasing adolescent STBs (S2) were greater temperamental sadness (S2: β = .057, SE = .032, 95% CI: −.007, .121), more temperamental openness (S2: β = .077, SE = .028, 95% CI: .021, .133), less temperamental fear (S2: β = −.084, SE = .028, 95% CI: −.138, −.029), more youth self-blame (S2: β = .076, SE = .039, 95% CI: .−.001, .153), and lower working memory abilities (S2: β = −.060, SE = .030, 95% CI: −.119, −.001). Additional retained variables based on the p < .10 cutoff were temperamental sadness (S1: β = .072, SE = .032, 95% CI: .009, .134) and family SES (S2: β = −.071, SE = .038, 95% CI: −.149, .005).

Three significant interactions were identified (see Analytic Plan), which were added to the final model. Sex interacted with both temperamental openness and youth self-blame to predict adolescent STBs (S2). Working memory also interacted with temperamental openness to predict adolescent STBs (S2). See Table S4 for full model building results.

Final Model

Within the final model, less decrease in (more persistent) childhood STBs was related to greater parental criticism (S1: β = −.070, SE = .034, 95% CI: −.136, −.003). Youth with greater temperamental openness experienced decreasing childhood STBs (S1: β = .066, SE = .033, 95% CI: .002, .131); the effect of temperamental sadness was no longer significant (S1: β = .060, SE = .031, 95% CI: −.002, .122). Youth’s ADHD symptoms, sex, and peer relationships were not significant predictors of childhood change in STBs within the multivariate model (p’s ≥ .151).

Predictors of increasing adolescent STBs (S2) included decreasing childhood STBs (S2: β = −.211, SE = .054, 95% CI: −.317, −.104), female sex (S2: β = .080, SE = .039, 95% CI: .004, .157), and lower temperamental fear (S2: β = −.092, SE = .028, 95% CI: −.146, −.037). Interaction effects revealed sex-specific relationships such that increasing adolescent STBs was related to greater temperamental openness (S2: β = .221, SE = .091, 95% CI: .043, .400) and higher youth self-blame (S2: β = .373, SE = .130, 95% CI: .118, .628) for females but not males. There was also a stronger relationship between greater openness and increasing adolescent STBs for youth with lower baseline working memory abilities (S2: β = −.090, SE = .028, 95% CI: −.144, −.036). Youth ADHD symptoms, temperamental sadness, and family SES were not reliably related to changes in adolescent STBs when all other variables were accounted for within the final model (see Table 4).

Table 4.

Predictors of Change in Suicidal Thoughts and Behaviors

	(1) Covariates				(7) Combined				(8) Final
Youth (S1)	β	SE	95% CI		β	SE	95% CI		β	SE	95% CI
i	−0.413	0.051	−0.512	−0.313	−0.412	0.051	−0.513	−0.312	−0.412	0.010	−0.513	−0.312
ADHD	0.077	0.034	0.010	0.144	0.047	0.039	−0.029	0.123	0.047	0.039	−0.028	0.123
Sex (F)	0.055	0.033	−0.010	0.120	0.043	0.034	−0.013	0.109	0.043	0.034	−0.023	0.109
Sadness					0.060	0.032	−0.002	0.122	0.060	0.031	−0.002	0.122
Openness					0.066	0.033	0.002	0.131	0.066	0.033	0.002	0.131
Peer Context					0.060	0.042	−0.022	0.142	0.060	0.042	−0.022	0.142
Criticism					−0.069	0.034	−0.136	−0.003	−0.070	0.034	−0.136	−0.003
Adolescence (S2)	β	SE	95% CI		β	SE	95% CI		β	SE	95% CI
i	−0.020	0.056	−0.131	0.090	−0.042	0.054	−0.147	0.064	−0.054	0.052	−0.156	0.048
S1	−0.195	0.056	−0.306	−0.085	−0.206	0.055	−0.314	−0.099	−0.211	0.054	−0.317	−0.104
ADHD	0.074	0.035	0.007	0.142	0.014	0.038	−0.061	0.089	0.015	0.038	−0.058	0.089
Sex (F)	0.182	0.034	0.116	0.249	0.163	0.034	0.096	0.230	0.080	0.039	0.004	0.157
WM					−0.048	0.030	−0.107	0.011	−0.048	0.030	−0.108	0.011
Sadness					0.055	0.033	−0.010	0.120	0.055	0.033	−0.010	0.121
Openness					0.070	0.029	0.014	0.126	−0.138	0.086	−0.307	0.031
Fear					−0.090	0.028	−0.145	−0.035	−0.092	0.028	−0.146	−0.037
Self-Blame					0.061	0.038	−0.014	0.136	−0.294	0.117	−0.523	−0.065
SES					−0.059	0.039	−0.136	0.019	−0.062	0.039	−0.137	0.014
WM*Openness									−0.090	0.028	−0.144	−0.036
Sex*Openness									0.221	0.091	0.043	0.400
Sex*Self-Blame									0.373	0.130	0.118	0.628

Note. i = intercept. S1 = slope 1. S2 = slope 2. WM = working memory latent variable. Additional model building results available in supplemental material.

Discussion

Suicide is a significant public health concern among adolescents and emerging adults, and those with ADHD are at heightened risk, in part due to the interplay of cognitive, emotional, and environmental factors across development. This study draws on prior applications of the socio-ecological and developmental psychopathology frameworks for suicide risk STBs (Glenn et al., 2016; Prades-Caballero et al., 2025; Séguin et al., 2014) to characterize the developmental course of suicidal thoughts and behaviors (STBs) from middle childhood into emerging adulthood in a large cohort oversampled for ADHD and identify key childhood predictors associated with individual trajectories of worsening or improving STBs. Results identified early onset but decreasing childhood STBs, followed by a nonsignificant increase in adolescence. Persisting childhood STBs were linked to greater parental criticism, whereas more temperamental openness predicted decline. In adolescence, increasing STBs was predicted by female sex and less temperamental fear in childhood. Sex interacted with hypothesized predictors, such that greater temperamental openness and greater youth self-blame for parental conflict in childhood predicted increasing adolescent STBs in females but not males. Temperament and cognitive features also interacted; greater temperamental openness predicted increasing adolescent STBs most strongly when childhood working memory abilities were low. ADHD symptoms were not reliably associated with changes in childhood or adolescent STBs after accounting for these other factors, underscoring the complexity of predicting STB risk.

The developmental trajectory of STBs was best represented by a piecewise growth model with two distinct phases. During middle childhood (ages 7–10), relatively low baseline levels of STBs further decreased over time, consistent with trajectories of STBs in childhood depression (Whalen et al., 2015; 2022) and suicidal ideation in the general population (Zhu et al., 2019). Although Whalen et al. (2022) attribute the elementary-age peak in STBs to their sampling population, results from this study and Zhu et al. (2019) suggest that childhood STBs may reflect a normative developmental process rather than disorder-specific risk. In adolescence, there was no overall pattern of change in STBs when averaged across the entire sample. Previous research has identified inscreases in both suicidal thoughts and behaviors through adolescence in specific high-risk groups (Goldston et al., 2016; Whalen et al., 2022); here we suggest that not all youth experience increasing STBs in adolescence. This finding highlights the importance of identifying predictors of individual change, rather than relying on group-level trajectories, as well as considering the interplay with interpersonal factors to understand vulnerability for STBs.

Examining predictors of individual change in STBs across development revealed that greater declines in childhood STBs predicted greater increases in adolescent STBs. Children who experience high levels of distress at young ages can often benefit from interventions that decrease their experience of STBs (Wasserman et al., 2021). Our finding is consistent with prior work suggesting that youth who experience childhood STBs are at high risk for reoccurrence and worsening suicidal thoughts and behaviors as they reach adolescence (Whalen et al., 2015). The present results suggest that this risk is present even when there is short-term improvement during the childhood period. These results have critical clinical considerations such that continued monitoring and follow up is important for adolescents who experienced STBs in childhood, even if those concerns have since diminished.

Separately, we visually observed an apparent tapering of STBs severity around age 18. The present sample does not afford sufficient cases beyond this age to statistically examine the trajectory of STBs within young adulthood. Prior work examining emerging and early adulthood has identified trajectories of declining suicidal ideation (Madhavan et al., 2021) and declines across both suicidal thoughts and behaviors (Goldston et al., 2016; Whalen et al., 2022). This is an important area for further inquiry.

Several temperament domains predicted individual variation in trajectories of STBs. Greater openness was adaptive in childhood, highlighting the benefits of curiosity and cognitive flexibility during this developmental period. In adolescence, increasing STBs was associated with less temperamental fear in childhood, described at the item level as liking to take risks and not being afraid of things. Caution is warranted in interpreting this effect given the low internal consistency for this factor, though we note that fear is one component of the higher order negative affect temperament domain that also includes anger. Negative affect is often elevated for youth with ADHD compared to their peers (Krieger et al., 2019), with prior findings suggesting that greater irritability may play a particularly important role in conveying risk (Leibenluft et al., 2023). Prior work by Karalunas et al. (2023) in this sample also identified temperamental irritability as an important predictor of high, persistent depression in adolescence. In other samples, irritability has been identified as predicting greater risk for suicidal thoughts and behaviors (Orri et al., 2019), and as a mediator of the association between STBs and ADHD symptoms (Levy et al., 2020), but these studies have focused on clinical symptoms of irritability (e.g., via ODD items or similar clinical ratings), depression, and anxiety rather than more dimensional measures of temperament that may capture subclinical variation in traits, highlighting the importance of capturing this trait-level variation.

Examination of neurocognitive predictors revealed that childhood response inhibition did not predict change in STBs. Some prior work indicates that lower response inhibition is associated with risk for suicide attempts (Moniz et al., 2017). However, the present results are consistent with other findings that response inhibition is not directly associated with risk for STBs, though it may predict major depression more broadly (Pan et al., 2011; Richard-Devantoy et al., 2016). The relationship between response inhibition and STBs may also depend on emotional valence of stimuli or the emotional context (Allen et al., 2021; Porteous et al., 2021; You et al., 2020). The stimuli used in the present study were not emotionally valenced, but studies in this sample at older ages have found that cognitive performance on inhibitory tasks and neural responses to stimuli requiring inhibition vary across emotional conditions (Figuracion et al., 2023; Karalunas et al., 2020). Additional studies focusing on cognition-emotion interactions may be important.

Lower childhood working memory predicted increasing adolescent STBs univariate models, consistent with prior literature (Richard-Devantoy et al., 2015). In the final model, working memory and temperamental openness interacted to predict increasing adolescent STBs such that working memory deficits amplified the risk conferred by greater temperamental openness. Although openness can foster emotional awareness and curiosity, lower working memory may limit regulation of these emotions, heightening vulnerability to STBs. Facets such as emotional openness and imagination may promote deeper engagement with intense emotional experiences or rumination, consistent with findings linking high openness to increasing STBs in emotionally volatile adolescents (Planellas & Calderón, 2022). Given that working memory is central to executive control and emotion regulation (Groves et al., 2022), deficits may hinder the ability to manage the emotional complexity associated with openness, increasing susceptibility to STBs. Temperamental openness also interacted with biological sex, with higher openness predicting increasing adolescent STBs for females but not males. Thus, adolescent females who experience a greater need for new experiences or emotional states may be particularly vulnerable. Taken together, these results help to clarify mixed findings on whether low or high openness is associated with STBs (Mars et al., 2019a; Pawar & Palve, 2021; Planellas & Calderón, 2022), suggesting that effects depend on both cohort sex distribution and cognitive abilities.

Parental criticism was associated with higher, persisting childhood STBs, consistent with prior work showing that parental criticism can exacerbate childhood suicidal ideation through emotion dysregulation (James et al., 2017). However, we did not replicate a prior cross-sectional association between parental criticism and adolescent STBs in the general population (Wedig & Nock, 2007), suggesting its influence may not extend to longitudinal risk in ADHD. During model building, greater peer conflict in childhood appeared to predict steeper declines in childhood STBs, but this effect did not persist within the multivariate model. It is possible that our parent- and teacher-report measure of peer difficulties may be moreso capturing social withdrawal and isolation, which may reduce exposure to negative peer interactions for children with ADHD such as bullying, which are linked to STBs (Yen et al., 2014). Additionally, given this effect was only present in model building where domains are considered in isolation, it may reflect a general effect of severity. That is, children with more severe and pervasive challenges overall may be more likely to receive intervention, and therefore show greater decreases in STBs.

In terms of youth sex, being female predicted increasing adolescent STBs but not childhood STBs, consistent with evidence that females are more vulnerable to internalizing problems and depression during adolescence (Keyes & Platt, 2024; Miranda-Mendizabal et al., 2019). Additionally, self-blame for parental conflict in childhood predicted increasing adolescent STBs specifically in females, extending prior links between parental conflict and youth internalizing disorders (Fear et al., 2009; Olatunji & Idemudia, 2021) to highlight a sex-specific vulnerability for STBs. Although males experience higher levels of self-blame (a finding that was also replicated in our sample; Hafsa et al., 2021), this factor appeared more consequential for females, suggesting that interpretation of family dynamics in childhood may uniquely increase adolescent females’ risk for STBs.

Additionally, parent- and teacher-rated ADHD symptoms in childhood were related to increasing adolescent STBs in several model-building steps, but this relationship was better explained by other variables in the final model. ADHD diagnosis is consistently associated with greater STBs at the group level (Chronis-Tuscano et al., 2010; Garas & Balazs, 2020; Liu et al., 2022), a finding that we replicated in the present sample. Yet, a recent systematic review identified that the direct relationship between ADHD and STBs only remained significant in 67% of studies after accounting for sociodemographics and co-occurring conditions (Rother et al., 2025). Further, most of this work has focused on the impact of psychiatric conditions and sex/gender on the association between ADHD and STBs with few studies including psychosocial factors and none of the reviewed studies considered neurocognitive or temperament domains (Rother et al., 2025). Similarly, most studies of suicide risk in general community adolescents have focused on individual and then interpersonal factors, failing to contextualize findings within the broader impacts of community and society (Prades-Caballero et al., 2025). Our findings suggest that ADHD characteristics do not reliably predict developmental changes in STBs in the context of other individual and interpersonal factors and while controlling for the societal impact of the COVID-19 pandemic.

Although impulsivity is often considered a central feature of ADHD and has been linked to suicidal ideation and behaviors in prior research (Conejero et al., 2019; Janiri et al., 2021; Swanson et al., 2014), temperamental impulsivity was not retained as a significant predictor of individual change in STBs in our study. This null finding may reflect measurement specificity such that temperament-based impulsivity (e.g., acting without thinking) captures trait-level tendencies that are less proximally related to the emergence or escalation of STBs. Alternatively, the absence of association may stem from our combined outcome of suicidal thoughts and behavior. Impulsivity may be more relevant to transitioning from suicidal thoughts to suicidal behaviors, rather than to the emergence of ideation itself (Hadzic et al., 2020). Thus, the null result may underscore the importance of disaggregating STB outcomes and using multi-method assessments of impulsivity, including both temperament and behavioral inhibition, in future work.

Finally, we did not replicate prior work identifying associations between suicidal ideation and racial minoritiy identity, and between suicidal behavior and family SES (Chen et al., 2022; Xiao & Lindsey, 2021). One possible explanation is that there are distinct vulnerabilities for ideation and behavior according to these sociodemographic factors that cannot be detected within our combined operationalization of STBs. The characteristics of our sample may have also impacted this result, given that we were not well powered to detect race effects.

Limitations

Our study provides valuable insights into developmental trajectories of suicidal thoughts and behaviors (STBs) and their predictors in youth with and without ADHD; yet several limitations should be acknowledged. First, the dataset was not designed to assess STBs as a primary outcome. Although multi-informant reports strengthen validity, the available items do not fully capture the complexity of STBs as described in other studies (Goldston et al., 2016).

Second, although we followed precedent in combining suicidal ideation and behaviors into a single outcome variable (e.g., Galera et al., 2021; Orri et al., 2019; Pickles et al., 2020), this approach may obscure meaningful distinctions between stages of suicide risk. Indeed, the relationship between suicidal thoughts and behaviors is not always linear (Campos et al., 2021), and distinct predictors may underlie the emergence of suicidal thoughts versus the transition to suicidal actions (Commisso et al., 2023; Hennefield et al., 2025; Kyron et al., 2020; Mars et al., 2019b), though most studies rely on a single class of predictors (Bayliss et al., 2022). Future work should prioritize disaggregated outcomes and leverage multimodal predictors (Kessler et al. 2020) to improve precision in modeling suicide risk trajectories.

Third, our analytic approach focused on identifying the most robust predictors across a large candidate set while examining ADHD symptoms using a total score and incorporating impulsivity as a temperament-based subscale. Some literature suggests that inattention may be more closely linked to internalizing symptoms, such as depression and suicidal ideation, whereas the externalizing behaviors and risk-taking associated with hyperactivity/impulsivity may confer risk for suicidal action (Conejero et al., 2019; Hadzic et al., 2020). The total ADHD symptom score was used in the present analyses given the combined outcome across suicidal thoughts and behaviors, though future should clarify whether ADHD subdomains differentiatally predict suicidal thoughts versus behaviors.

Fourth, we examined differences based on biological sex and did not assess gender identity, limiting conclusions for non-binary and gender-diverse youth. Similarly, exclusion of current major depression at baseline (i.e., ages 7–10) may restrict generalizability to higher-risk clinical populations, although some participants had a prior history of depression. Fifth, we utilized time-invariant predictors—baseline measures of hypothesized variables—to predict changes in STBs over time. This approach fits with our goal of identifying early predictors; the results from this model can now inform future studies using time-varying predictors or repeated measures frameworks to more precidely model developmental processes and potential causal mechanisms.

Finally, we accounted for the broad effects of the COVID-19 pandemic using a binary indiactor of pre- versus during-pandemic participation; however, this measure cannot capture the nuanced ways the pandemic influenced risk. Pandemic-related disruptions may have both heighted vulnerabiulity for youth with ADHD via loss of services and remote learning challenges (Becker et al., 2020; Rogers & MacLean, 2023; Sibley et al., 2021) and buffered risk from school-based bullying victimization and other social stressors (Hall et al., 2024; Vaillancourt et al., 2021). More nuanced measurement of environmental disruptions is needed to understand how macro-level stressors intersect with individual and interpersonal risk factors to shape developmental trajectories of suicide risk.

Conclusions

This work advances the understanding of suicidal thoughts and behaviors in the context of ADHD by demonstrating a piecewise trajectory, in which STBs decrease during childhood but stabilize or rise during adolescence. Risk was not explained by ADHD symptoms alone, but by a dynamic interplay of temperament, neurocognitive factors, and interpersonal experiences with sex-specific pathways differentiating risk. These results underscore the complexity of predicting suicide risk, emphasizing that developmental course and context are critical considerations. Integrating these multidimensional influences into prevention efforts and risk assessments will enhance the ability to identify and support youth who experience STBs.

Supplementary Material

Supplementary_Material

Table S1

Mood Disorder Prevalence

Table S2

Suicidal Thoughts and Behaviors Endorsed by Participant Age and Respondent

Table S3

Highest Suicidal Thoughts and Behaviors Across Study Timepoints

Table S4

Model Building Results

Figure 2. Final Model Predicting Individual Changes in Suicidal Thoughts and Behaviors.

Note. Int = intercept, ADHD = ADHD-RS composite, Sex (F) = Female sex assigned at birth, SES = socioeconomic status composite, Self-Blame = YCPIC Self Blame subscale, S1 = Slope 1 (childhood), S2 = Slope 2 (adolescence), WM = Working Memory latent variable, DS FW = Digit Span Forward, DS BK = Digit Span Backward, SS FW = Spatial Span Forward, SS BK = Spatial Span Backward, N1 Back = Nback task 1 back, N2 Back = Nback task 2 back, Peers = SDQ peer relationships composite, EE-Crit = FMSS Parental Expressed Emotion – Criticism, Sad = Temperamental Sadness, Open = Temperamental Openness, Fear = Temperamental Fear. Solid lines represent significant predictors within the final model; weight of the line is scaled to represent the strength of the effect. Empty circles represent interaction effects. * < .05, ** < .01, ⁺ < .001.