Impulsivity is a broad construct, commonly defined as an individual difference, that manifests as a preference for immediate over delayed reward, a failure to plan, and impaired self-control (Beauchaine et al., Reference Beauchaine, Zisner and Sauder2017). Other definitions of this construct emphasize elevated urgency and sensation seeking and deficient premeditation and perseverance (e.g., Whiteside & Lynam, Reference Whiteside and Lynam2001). Extensive evidence has linked impulsigenic traits to heightened psychopathology (for a meta-analysis, see Berg et al., Reference Berg, Latzman, Bliwise and Lilienfeld2015). In fact, impulsivity is a criterion of 18 DSM-5 psychiatric disorders (American Psychiatric Association, 2013). Although critically involved in the development and maintenance of psychopathology, and particularly “externalizing” psychopathology (e.g., opposition, noncompliance, defiance, aggression), few studies have examined whether childhood preventive interventions targeting such problems of conduct mitigate risk for impulsivity, specifically, and associated maladaptive outcomes. The main objective of this study was to disentangle between-person heterogeneity in growth of impulsivity across childhood and adolescence in nonintervention and intervention participants recruited from five harmonized longitudinal preventive intervention trials. In addition, we also aimed to observe whether profile membership was associated with adult psychopathologies.
Impulsivity as an underlying process
Impulsigenic traits have been identified as transdiagnostic phenotypes for multiple externalizing psychopathologies, including attention-deficit/hyperactivity disorder (ADHD), oppositional defiant disorder, and conduct disorder in early and middle childhood, as well as substance use disorder and antisocial personality disorder in adolescence and adulthood (Beauchaine et al., Reference Beauchaine, Hinshaw and Pang2010, Reference Beauchaine, Zisner and Sauder2017). Impulsivity has also been identified as a specific measurement item that bridges (i.e., a symptom from one dimension that connects to symptoms from another dimension; Cramer et al., Reference Cramer, Waldorp, Van Der Maas and Borsboom2010) aggressive and delinquent behavior across childhood and into early adolescence (Goulter et al., Reference Goulter, McMahon, Lansford, Bates, Dodge, Crowley and Pettit2022). Indeed, impulsivity has been hypothesized to contribute to externalizing heterotypic continuity, which refers to distinct behavioral manifestations of the same underlying liability (Beauchaine et al., Reference Beauchaine, Hinshaw and Pang2010, Reference Beauchaine, Zisner and Sauder2017). In addition to underpinning multiple forms of externalizing psychopathology, meta-analytic work has identified impulsivity as a mediator between childhood maltreatment and suicidal behavior (Pérez-Balaguer et al., Reference Pérez-Balaguer, Peñuelas-Calvo, Alacreu-Crespo, Baca-Garcia and Porras-Segovia2022). Facets of impulsivity are also associated with suicidal behavior in young people and may represent critical markers of suicidal risk (Brezo et al., Reference Brezo, Paris and Turecki2006; McHugh et al., Reference McHugh, Lee, Hermens, Corderoy, Large and Hickie2019).
Etiological models of impulsivity point to dysfunction in early-maturational subcortical emotion-generation and later-maturational cortical emotion-regulation systems (Casey, Reference Casey2015; Cubillo et al., Reference Cubillo, Halari, Smith, Taylor and Rubia2012; Macdonald et al., Reference Macdonald, Goines, Novacek and Walker2016). Specifically, imaging research has shown that the mesolimbic dopamine system and its feedforward and feedback projections to and from the mesocortical dopamine system represent neural substrates characterizing dispositional impulsivity (Gatzke-Kopp, Reference Gatzke-Kopp2011). Molecular genetic research has also shown that dopamine functioning plays an important role in vulnerability to approach-related affect and impulsive behavior (Neuhaus & Beauchaine, Reference Neuhaus, Beauchaine, Beauchaine and Hinshaw2017). In addition to these neurobiological influences, environmental factors have been implicated in the susceptibility of impulsive behavior. Particularly, childhood adversity (e.g., parental maltreatment) can alter functioning of cortical structures, such as the prefrontal cortex, further promoting impulsive behavior (Hanson et al., Reference Hanson, Chung, Avants, Shirtcliff, Gee, Davidson and Pollak2010; Mead et al., Reference Mead, Beauchaine and Shannon2010). Thus, interventions with components addressing childhood cognitive, affective, and behavioral difficulties, in addition to targeting change in adverse environmental factors (such as social interactions within the home environment), have the potential to decrease the likelihood of the development of psychopathology through reductions in levels of impulsivity.
Childhood preventive interventions
Although few studies have examined childhood preventive intervention effects on impulsivity specifically, there is accruing evidence of the positive impact of intervention on aggressive and disruptive behavior through childhood, adolescence, and into early adulthood. For example, the Johns Hopkins trials found that children who received the Good Behavior Game [GBG]—a classroom-based behavior management intervention that aims to decrease disruptive behaviors—exhibited lower levels of aggressive behavior across childhood and adolescence, relative to children in a control condition (Dolan et al., Reference Dolan, Kellam, Brown, Werthamer-Larsson, Rebok, Mayer, Laudolff, Turkkan, ford and Wheeler1993; Ialongo et al., Reference Ialongo, Werthamer, Kellam, Brown, Wang and Lin1999, Reference Ialongo, Domitrovich, Embry, Greenberg, Lawson, Becker and Bradshaw2019). Further, participation in the Linking the Interest of Families and Teachers (LIFT) intervention, which included a version of the GBG, was associated not only with reduced aggression on the playground but also reduced teacher-rated problem behavior during the middle school years (Eddy et al., Reference Eddy, Reid, Stoolmiller and Fetrow2003). Similarly, SAFE Children, a family-focused preventive intervention, significantly reduced children’s aggressive and disruptive behaviors (Tolan et al., Reference Tolan, Gorman-Smith and Henry2004), as did Fast Track (FT), a multimodal school-based preventive intervention (Conduct Problems Prevention Research Group, 2020).
Some of these interventions had long-term and multifaceted impacts. For example, in young adulthood, the GBG showed effects across multiple outcomes, including a reduction in the use of school-based mental health services (Poduska et al., Reference Poduska, Kellam, Wang, Brown, Ialongo and Toyinbo2008), and lower incidence of substance use disorders (Kellam et al., Reference Kellam, Brown, Poduska, Ialongo, Wang, Toyinbo, Petras, Ford, Windham and Wilcox2008), antisocial personality disorder (Petras et al., Reference Petras, Kellam, Brown, Muthén, Ialongo and Poduska2008), and suicidal ideation and attempts (Wilcox et al., Reference Wilcox, Kellam, Brown, Poduska, Ialongo, Wang and Anthony2008). Among those with the highest kindergarten-assessed risk of conduct problems, FT participants assigned to intervention, relative to control, showed significant reductions in delinquency and a reduced number of conduct disorder diagnoses in childhood and adolescence (Conduct Problems Prevention Research Group, 2007, 2011). The FT intervention also decreased the probability of suicidal ideation and hazardous drinking in adolescence and adulthood, and opioid use in adulthood (Godwin et al., Reference Godwin, Coie, Crowley, Dodge, Greenberg, Lochman, McMahon and Pinderhughes2020), as well as decreasing externalizing and internalizing psychopathology, substance use problems, and substance and violent convictions in adulthood (Dodge et al., Reference Dodge, Bierman, Coie, Greenberg, Lochman, McMahon and Pinderhughes2015).
Drawing on developmental models grounded in the research literature on impulsivity and ADHD (Beauchaine et al., Reference Beauchaine, Hinshaw and Pang2010, Reference Beauchaine, Zisner and Sauder2017), multimodal interventions designed to target conduct problems, including child-focused skills training, behavior therapy, social-emotional learning, and parent management training (such as the preventive interventions noted above), also might be well suited to reduce impulsivity among children and adolescents. There have been some promising findings in this regard. For example, the SAFE Children study found that the intervention predicted lower latent trajectory class membership of impulsivity for children across a 42-month period (Fowler et al., Reference Fowler, Henry, Schoeny, Gorman-Smith and Tolan2014). Similarly, Musci et al. (Reference Musci, Bradshaw, Maher, Uhl, Kellam and Ialongo2014) found that the Family School Partnership and classroom-centered interventions significantly reduced impulsive (and aggressive) behavior for adolescents across the 6th to the 12th grades. Overall, preliminary findings from longitudinal studies of childhood preventive interventions demonstrate that psychosocial interventions that target and significantly reduce conduct problems may also have beneficial effects on impulsive behaviors specifically.
Integrative data analysis and mixture modeling
Studies of large-scale interventions often use specific methodological approaches to reduce attrition, such as including a concise battery of measures and planned missingness whereby incomplete data are collected by design. These approaches not only help to reduce participant burden—which can increase the overall quality of the collected data—but also reduce the financial cost of conducting the study (Little & Rhemtulla, Reference Little and Rhemtulla2013). Of course, such practices necessitate that some measurement occasions are omitted, and some key constructs are not fully assessed. By synthesizing raw data across multiple trials through pooled data analytic approaches such as integrative data analysis, sample sizes at measurement occasions can be increased and sufficient information about these constructs may be captured (Brincks et al., Reference Brincks, Montag, Howe, Huang, Siddique, Ahn, Sandler, Pantin and Brown2018; Hussong et al., Reference Hussong, Curran and Bauer2013). Because integrated data increase the sample size and statistical power, heterogeneity of population characteristics may also be studied. Importantly, this can both optimize the examination of subgroup effects and enable the examination of impacts on low base-rate behaviors, such as suicidal behaviors. Mixture modeling is a methodological approach often used to parse variance and distinguish subgroups or profiles (Masyn, Reference Masyn and Little2013). In particular, latent class growth analysis (LCGA) may be beneficial in identifying patterns across time.
Longitudinal studies charting the developmental course of impulsivity (and related constructs) have typically revealed decreasing or stable patterns across childhood, adolescence, and into adulthood. For example, research with several population cohorts found that hyperactive-inattentive scores decreased from early childhood through early adulthood (Wootton et al., Reference Wootton, Riglin, Blakey, Agnew-Blais, Caye, Cadman, Havdahl, Gonçalves, Menezes, Wehrmeister, Rimfeld, Davey Smith, Eley, Rohde, Arseneault, Moffitt, Stergiakouli, Thapar and Tilling2022). In another study spanning multiple countries, self-regulation scores increased from preadolescence through early adulthood (Steinberg et al., Reference Steinberg, Icenogle, Shulman, Breiner, Chein, Bacchini, Chang, Chaudhary, Giunta, Dodge, Fanti, Lansford, Malone, Oburu, Pastorelli, Skinner, Sorbring, Tapanya, Tirado, Alampay, Al‐Hassan and Takash2018). However, other models have conceptualized impulsivity (and, more broadly, ADHD symptoms) as stable over time (Sonuga-Barke et al., Reference Sonuga-Barke, Becker, Bölte, Castellanos, Franke, Newcorn, Nigg, Augusto Rohde and Simonoff2022). By applying LCGA to longitudinal data of impulsivity, findings can help to inform understanding regarding heterogeneity in the developmental course of impulsive behavior across critical developmental periods. Using LCGA with large integrated datasets also serves to identify between-person differences in growth of impulsivity with more power and precision than when applied to single studies. In addition, parsing between-person differences in intervention samples can inform for whom interventions are effective (Fowler et al., Reference Fowler, Henry, Schoeny, Gorman-Smith and Tolan2014; Pasalich et al., Reference Pasalich, Craig, Goulter, O’Donnell, Sierra Hernandez and Moretti2022). For example, a robust finding in the literature shows that classroom-based behavior management interventions, such as the Good Behavior Game, are more effective for those with higher initial severity of conduct problems (Kellam et al., Reference Kellam, Brown, Poduska, Ialongo, Wang, Toyinbo, Petras, Ford, Windham and Wilcox2008; McMahon et al., Reference McMahon, Goulter and Frick2021).
The present study
This study had three core aims. First, we aimed to harmonize items related to impulsive behavior from five randomized controlled trials focused on preventing aggressive and disruptive behavior in middle childhood by using individual participant data and applying integrative data processing. Second, we aimed to parse between-person heterogeneity in growth of impulsivity across six timepoints in childhood and adolescence (grades 2, 4, 5, 7, 8, and 10). To understand intervention effects on the development of impulsivity, we examined whether intervention status predicted profile membership in the combined nonintervention and intervention sample (N = 4,975; 53.6% male; 55.1% Black, 37.2% white, 7.5% other race/ethnicity). We also examined between-person differences in impulsivity in the nonintervention (n = 2,492; 53.0% male; 54.3% Black, 38.4% white, 7.2% other race/ethnicity) and intervention (n = 2,483; 54.2% male; 55.9% Black, 35.9% white, 7.8% other race/ethnicity) samples. Finally, we aimed to determine whether distinct profiles of impulsivity across childhood and adolescence were associated with theoretically relevant psychopathologies in adulthood (ranging from 18 to 25 years) and we examined sex, race/ethnicity, and baseline levels of aggression as predictors. Adult outcomes included aggressive behavior, substance use, suicidal ideation/attempts, and anxiety/depression. Findings have the potential to inform understanding of the developmental course and prevention of impulsivity during childhood and adolescence, as well as prognosis regarding psychopathological outcomes in adulthood.
Method
Participants and procedure
The harmonized sample included 4,975 participants from five randomized controlled prevention trials: Johns Hopkins Prevention Intervention Research Center Generation (PIRC) Generation 1 (n = 2,311), PIRC Generation 2 (n = 678), LIFT (n = 671), SAFE Children (n = 424), and FT (n = 891). In each trial, legal guardians provided consent, participants assented to procedures, and all participants were provided monetary compensation for their time. All procedures were approved by the relevant Institutional Review Board (IRB) for each site and the harmonized dataset and analyses were approved by the Johns Hopkins IRB.
Prevention Intervention Research Center Generation 1 (PIRC 1)
In 1985, two successive first-grade cohorts (n1 = 1,196; n2 = 1,115; total N = 2,311) were recruited from 43 classrooms among 19 elementary schools located in five distinctly different sociodemographic urban areas of eastern Baltimore. Within each area, three or four schools were matched by socioeconomic status, size of the school, and ethnicity. In first grade, the students’ mean age of participants was 6.5 years (SD = 0.48), and 52% received free or reduced-priced lunch—a proxy for low family income. Schools were randomly assigned to receive either the GBG intervention (Barrish et al., Reference Barrish, Saunders and Wolf1969), the Mastery Learning curriculum intervention (Dolan et al., Reference Dolan, Kellam, Brown, Werthamer-Larsson, Rebok, Mayer, Laudolff, Turkkan, ford and Wheeler1993), or no intervention. In addition, within schools, teachers and students were randomly assigned to intervention or no intervention conditions. The GBG is a whole-class behavior management strategy that aims to decrease disruptive behaviors by assigning children to teams and only allowing the teams that do not exceed a specified criterion of precisely defined off-task, disruptive, and aggressive behaviors to “win.” Participants assigned to the GBG condition received the intervention over the course of first and second grades, with the game being played several times per week. The Mastery Learning intervention involved extensive enrichment of the reading curriculum, with a special focus on clear instructional statements, communication of expectations, and a group-based approach to mastery. For the present analyses, participants were clustered into intervention (GBG + Mastery Learning: n = 1,339; 49.0% male; 68.2% Black, 30.1% white, 1.6% other race/ethnicity) and nonintervention (n = 972; 50.4% male; 63.6% Black, 35.0% white, 1.5% other race/ethnicity) groups.
Prevention Intervention Research Center Generation 2 (PIRC 2)
In 1993, 678 urban first graders were recruited from 27 classrooms in 9 elementary schools primarily located in western Baltimore. At baseline, participants ranged in age from 4.9 to 8.1 years with a mean age of 6.2 years (SD = 0.37), and 68.9% of the children received free or reduced-priced lunch. Classrooms were randomly assigned to the classroom-centered intervention (adapted from the GBG and Mastery Learning interventions), the Family-School Partnership intervention, or no intervention. The classroom-centered intervention was designed to reduce the early risk behaviors of poor achievement and aggressive behavior through enhancements to the curriculum, improvements in teacher instructional and classroom behavior management practices, and specific strategies for children not performing adequately (Ialongo et al., Reference Ialongo, Werthamer, Kellam, Brown, Wang and Lin1999). The Family-School Partnership intervention was developed to improve collaboration between parents, teachers, and school mental health professionals and to enhance parents’ behavior management skills (Ialongo et al., Reference Ialongo, Werthamer, Kellam, Brown, Wang and Lin1999). For these analyses, participants were clustered into intervention (classroom-centered + Family-School Partnership: n = 459; 54.5% male; 86.5% Black, 13.3% white, 0.2% other race/ethnicity) and nonintervention (n = 219; 51.1% male; 85.8% Black, 14.2% white, 0% other race/ethnicity) groups.
Linking the interest of families and teachers (LIFT)
Across three cohorts recruited between 1991 and 1993, a total of 671 first or fifth-grade children were enrolled from 12 public elementary schools in three adjacent school districts in the Pacific Northwest. Either the entire set of first-grade classes (which included some second-grade students) or the entire set of fifth-grade classes (which included some fourth-grade students) within each school were invited to participate. School eligibility was based on living in “at-risk” neighborhoods, characterized by elevated rates of juvenile contact with police relative to local norms. Schools were randomly assigned to a multimodal intervention condition or a nonintervention condition. At baseline, approximately 25% of families were receiving some type of public financial assistance. In the intervention condition, the classroom component was intended to improve child social and problem-solving skills through 20 hour-long social-emotional learning skills training sessions over the course of the intervention; the playground component was a modified version of the GBG which reinforced child positive social behaviors during free (unstructured) play time; the family component was a group-based version of cognitive-behavioral parent management training emphasizing effective communication, positive reinforcement, supervision, discipline, and problem-solving skills. Parents met with a parent management training coach once a week, for 6 weeks, as well as received support via phone between sessions; finally, the linking component encouraged various avenues for teacher–parent communication while the other components were in play (Eddy et al., Reference Eddy, Reid and Fetrow2000; Reid et al., Reference Reid, Eddy, Fetrow and Stoolmiller1999). The present study included intervention participants (n = 382; 51.3% male; 1.3% Black, 85.9% white, 11.0% other race/ethnicity) and nonintervention participants (n = 289; 46.4% male; 2.4% Black, 81.3% white, 16.3% other race/ethnicity).
SAFE children (SAFE)
In 1997, 424 families consented to participate in the first-grade intervention of the SAFE Children study set within seven inner-city schools in Chicago (Tolan et al., Reference Tolan, Gorman-Smith and Henry2004). A majority of the sample (59%) had a family income of below $20,000 per year. Randomization occurred at the individual level, weighing toward the intervention condition because of expected lower retention rates for the prevention program. The first component was a multiple-family group approach that focused on parenting skills, family relationships, understanding and managing expectable developmental and situational challenges to the families, increasing support among parents in the group, issues in engaging as a parent with the school, and managing issues such as neighborhood problems. This component was combined with a phonics-based reading tutoring program and relates to academic improvement among high-risk children during first grade (Conduct Problems Prevention Research Group, 2020). Participants included intervention (n = 225; 45.8% male; 39.1% Black, 5.3% white, 54.7% other race/ethnicity) and nonintervention (n = 199; 52.3% male; 46.2% Black, 3.0% white, 50.8% other race/ethnicity) groups.
Fast track (FT)
Between 1991 and 1993, 55 high-risk schools were selected based on crime and poverty rates of their communities across four sites (Durham, NC; Nashville, TN; rural Pennsylvania; and Seattle, WA; Conduct Problems Prevention Research Group, 2020). Within each site, clusters of schools were randomly assigned to intervention and nonintervention conditions. Children were selected for inclusion into the sample based on teacher and parent reports of behavior problems moving from the highest score downward until desired sample sizes were reached within sites, cohorts, and groups. This multi-stage screening procedure resulted in a total of 891 high-risk children. During the elementary school phase of the intervention (grades 1–5), all families were offered parent management training with home visiting, academic tutoring, and child social skills training. Parent and child group interventions were conducted during a 2-hour enrichment program. These sessions included social skill training “friendship groups” (Bierman et al., Reference Bierman, Greenberg, Coie, Dodge, Lochman and McMahon2017), parent-training groups for parents, and guided parent–child interaction sessions (McMahon & Slough, Reference McMahon, Slough, Peters and McMahon1996). Paraprofessionals also provided tutoring, as well as peer-pairing session to improve friendships with classmates. In addition, a universal intervention (a modified, grade-level version of the PATHS curriculum; Greenberg & Kusché, Reference Greenberg and Kusché2011) was provided to the classrooms in intervention schools through the elementary school years to promote social and emotional competence. The universal intervention included weekly teacher consultation for lessons and classroom behavior management (Bierman et al., Reference Bierman, Greenberg, Coie, Dodge, Lochman and McMahon2017). During the middle and early high school phase (grades 6-10), there were three standard prevention activities offered to all FT intervention children: the middle school transition program, parent and youth groups on adolescent topics, and youth forums. Adolescent developmental issues were addressed with four meetings for parents and youth during grade 6. Parent groups focused on issues such as positive involvement and monitoring, and youth groups focused on issues such as coping with peer pressure. Parents and youth met together in groups to address romantic relationships and sex education, substances use and vocational goal setting. In grades 7 and 8, eight Youth Forums based on Oyserman’s (Reference Oyserman2000) program were held with youth in small groups to address vocational opportunities, budgeting and life skills, job interview skills, and summer employment opportunities. In grades 7–10, individualized intervention plans were developed and implemented with each youth, based on regular assessments of risk and protective factors, conducted three times during each year. The present study used data from the intervention (n = 445; 72.4% male; 53.0% Black, 44.5% white, 2.5% other race/ethnicity) and high-risk nonintervention (n = 446; 66.4% male; 48.4% Black, 48.9% white, 2.7% other race/ethnicity) samples.
Measures
Impulsivity (grades 2, 4, 5, 7, 8, and 10)
Impulsivity was assessed with four measures including the Parent Observation of Classroom Adaptation-Revised (POCA; Ialongo et al., Reference Ialongo, Werthamer, Kellam, Brown, Wang and Lin1999), the Teacher Observation of Classroom Adaptation-Revised (TOCA; Werthamer-Larsson et al., Reference Werthamer-Larsson, Kellam and Wheeler1991), the Child Behavior Checklist (CBCL; Achenbach, Reference Achenbach1991a), and the Teacher’s Report Form (TRF: Achenbach, Reference Achenbach1991b). The POCA is a 42-item structured interview assessing child behavior and adaptation to the family environment. Items are scored on a 4-point scale (1 “never/almost never,” 2 “sometimes,” 3 “often,” 4 “almost always”). A corresponding measure to the POCA, the TOCA is also a 42-item structured interview in which teachers report on their experiences with the child regarding different behaviors and characteristics. Items are scored on a 6-point scale (1 “never/almost never,” 2 “rarely,” 3 “sometimes,” 4 “often” 5 “very often,” 6 “always”). The CBCL is an empirically derived 113-item self- and parent-rated assessment of a child’s behavior problems and social competence. Items are scored on a 3-point scale (0 “not true,” 1 “somewhat or sometimes true,” 2 “very or often true”). The TRF is the teacher’s adaptation of the CBCL, which also includes 113 items assessing child behavior problems and social competence. Items are also scored on a 3-point scale (0 “not true,” 1 “somewhat or sometimes true,” 2 “very or often true”). Impulsivity items included: waits for turn, interrupts others, blurts out answer, can't sit still, concentrates, pays attention, and stays on task (see Supplementary Table S1 for which measures were administered at which grades, and Supplementary Table S2 for a list of item wordings for each measure). These items have been used in other studies examining impulsivity with high predictive validity (Musci et al., Reference Musci, Bradshaw, Maher, Uhl, Kellam and Ialongo2014).
Adult outcomes
Aggressive behavior (ages 18–21)
In addition to items drawn from the above measures, self-report measures (and an adaptation of the Diagnostic Interview Schedule’s [DIS] Antisocial Personality Module [Robins et al., Reference Robins, Helzer, Croughan and Ratcliff1981] was used in PIRC 1 and 2) were used to assess aggression including items such as: gets into fights, takes others’ property, lies, yells at others, hurts others physically, and destroys property.
Substance use (ages 18–25)
Substance use was defined as use of illicit substances in the past 12 months including opioids, cocaine, inhalants, stimulants, hallucinogens, and sedatives-tranquilizers. The origin of the substance use measurement varied across studies but was primarily derived from Composite International Diagnostic Interview (CIDI; Kessler et al., Reference Kessler, Wittchen, Abelson, Mcgonagle, Schwarz, Kendler, Knäuper and Zhao1998), the National Survey on Drug Use and Health (U.S. Department of Health and Human Services), or the National Longitudinal Study of Adolescent Health (Bureau of Labor Statistics, US Department of Labor, 2002).
Suicidal ideation/attempts (ages 18–23)
Suicide ideation and attempts were measured by asking two questions: whether the participants had (1) any serious/repeated thoughts about suicide and (2) deliberately tried to hurt or kill themselves in the past 12 months. Questions were drawn from the CIDI (Kessler et al., Reference Kessler, Wittchen, Abelson, Mcgonagle, Schwarz, Kendler, Knäuper and Zhao1998), DIS (Shaffer et al., Reference Shaffer, Fisher, Lucas, Dulcan and Schwab-Stone2000), and the Adult Self-Report (Achenbach, Reference Achenbach1997).
Anxiety/depression (ages 18-21)
In addition to items drawn from the above measures, self-report measures were also used to create an anxiety/depression score including items such as: I feel sad, I feel like crying, nothing makes me happy, I am tired.
Predictors
Intervention status, baseline aggression, binary sex (1 = Male, 0 = Female), and race (1 = Black, 0 = white) were included as predictors. Baseline aggression was assessed with items from the same measures as impulsivity including items such as: gets into fights, takes others’ property, lies, yells at others, hurts others physically, and destroys property.
Analytic approach
Analyses proceeded in two main steps: integrative data processing and mixture modeling.
Integrative data processing
To integrate impulsivity data across datasets, similar items from measures were identified and recoded as binary to adjust for differing Likert scales (such that, for the POCA and TOCA, 1 was recoded as 0 and all other scores were recoded as 1; and for the CBCL and TRF, 2 was recoded as 1). Items were then coded as an overall binary variable, such that if a participant endorsed any item, they received a score of 1 (otherwise a score of 0).Footnote 1 Overall binary variables were then mean scored for each timepoint. For baseline aggression and both adult aggression and anxiety/depression, items were z-scored and then mean scored within year (and across years for adult outcomes). Substance use and suicidal ideation/attempts were coded as a binary variable, such that if a participant endorsed these items at any timepoint, they received a score of 1 (otherwise a score of 0).
Mixture modeling
This step was completed separately for combined, nonintervention, and intervention samples. Using Mplus version 8.6 (Muthén & Muthén, Reference Muthén and Muthén2021), the manual Bolck, Croon, and Hagenaars (BCH) method was conducted to identify the number of impulsivity latent profiles and to compare profiles on predictors and adult outcomes (Asparouhov & Muthén, Reference Asparouhov and Muthén2014). The BCH method is recommended for examining associations between identified profiles and distal outcomes because it accounts for measurement error (i.e., classification uncertainty) of the latent profiles and the unequal variances among the variables by using posterior probabilities as weighting to account for individual differences (Nylund-Gibson et al., Reference Nylund-Gibson, Grimm and Masyn2019). In the first step, LCGA were conducted to identify groups of individuals based on impulsivity scores across six time points during childhood and adolescence. The optimal number of latent profiles was determined by comparing model fit across five models. Fit statistics included the Bayesian information criterion (BIC), sample-adjusted Bayesian information criterion (aBIC), Akaike information criterion (AIC), the Lo-Mendel-Rubin likelihood ratio test (LMR-LRT), and the Bootstrapped likelihood ratio test (BLRT) (Nylund et al., Reference Nylund, Asparouhov and Muthén2007). The model with the lowest BIC, aBIC, and AIC values is preferred, a nonsignificant χ 2 value (p > .05) for the LMR-LRT and BLRT statistics suggests that a model with one fewer class is preferred (i.e., k-1), and higher entropy values closer to 1.00 indicate clearer classification and greater power to predict class membership. Multiple random starting values (500 for each model) were used to avoid a local maximum. In addition, classification, accuracy, parsimony, and interpretability were also considered to ensure models fit with theory and previous research. In the second step, classification errors were computed, and the inverse logit of the error rates were transformed into weights. In the final step, using the established weights, we regressed class membership on intervention status (in the combined sample) and predictors (i.e., baseline aggression, sex, race/ethnicity), reported odds ratios (OR), and compared classes on adult outcomes (i.e., aggression, substance use, suicidal ideation/attempts, anxiety/depression) using difference testing. Full-information maximum likelihood (FIML) techniques were used to attain model estimates, which provides full-sample estimates that accommodate all observations regardless of whether missing data occurs for certain variables (Rubin & Little, Reference Rubin and Little2002). Data were considered missing at random qualifying our use of FIML. For integrative data analytic techniques, data is argued to be missing at random when trials have “different measures, different follow-up times, and missingness on conditions to which they were not randomly assigned (e.g., active intervention if assigned to control)” (Brincks et al., Reference Brincks, Montag, Howe, Huang, Siddique, Ahn, Sandler, Pantin and Brown2018 p. 64). FIML is also highly robust when there are large amounts of missing data (Brown, Reference Brown1990; Lavori et al., Reference Lavori, Brown, Duan, Gibbons and Greenhouse2008; Siddique et al., Reference Siddique, Brown, Hedeker, Duan, Gibbons, Miranda and Lavori2008).
Results
Descriptive statistics
Missing data, descriptive statistics, and correlations of main study variables are presented in Supplementary Tables S4 (combined sample), S5 (nonintervention sample), and S6 (intervention sample). (Missing data, descriptive statistics, and correlations of main study variables are also provided for each trial in the Supplementary Materials [see Supplementary Tables S7-S11].)
Impulsivity profiles through childhood and adolescence
Model fit statistics are presented in Table 1. LCGA indicated an optimal four-class impulsivity solution for the combined sample, the nonintervention sample, and the intervention sample.
Model fit statistics from the BCH method
Note. Bolded text indicates the best-fitting model. BIC = Bayesian information criterion, aBIC = sample size adjusted Bayesian information criterion, AIC = Akaike information criterion, LMR-LRT = Lo-Mendel-Rubin likelihood ratio test, BLRT = Bootstrapped likelihood ratio test.
For the combined sample, profiles are shown in Figure 1 and described as: high (61.9%; [intercept = .86, p < .001; linear slope = .10, p < .001; quadratic slope = −.02, p < .001]), moderate (12.2%; [intercept = .82, p < .001; linear slope = −.02, p = .001; quadratic slope = −.00, p = .017]), low (10.8%; [intercept = .67, p < .001; linear slope = −.11, p < .001; quadratic slope = .01, p < .001]), and low-increasing (15.3%; [intercept = .32, p < .001; linear slope = −.24, p < .001; quadratic slope = .05, p < .001]).
Impulsivity profiles from grades 2, 4, 5, 7, 8, and 10 – combined sample.
For the nonintervention sample, profiles are shown in Figure 2a and described as: high (63.6%; [intercept = .86, p < .001; linear slope = .10, p < .001; quadratic slope = −.02, p < .001]), moderate (12.1%; [intercept = .80, p < .001; linear slope = −.02, p = .116; quadratic slope = −.00, p=.010]), low (11.4%; [intercept = .71, p < .001; linear slope = −.14, p < .001; quadratic slope = .02, p < .001]), and low-increasing (14.3%; [intercept = .34, p < .001; linear slope = −.26, p < .001; quadratic slope = .05, p < .001]).
a . Impulsivity profiles from grades 2, 4, 5, 7, 8, and 10 – nonintervention sample. b . Impulsivity profiles from grades 2, 4, 5, 7, 8, and 10 – intervention sample.
For the intervention sample, profiles are shown in Figure 2b and described as: high (59.8%; [intercept = .86, p < .001; linear slope = .09, p < .001; quadratic slope = −.02, p < .001]), moderate (12.9%; [intercept = .83, p < .001; linear slope = −.03, p = .003; quadratic slope = −.00, p=.395]), low (11.4%; [intercept = .64, p < .001; linear slope = −.09, p < .001; quadratic slope = .01, p < .001]), and low-increasing (15.9%; [intercept = .30, p < .001; linear slope = −.22, p < .001; quadratic slope = .04, p < .001]).
Impulsivity profile comparisons on predictors and adult outcomes
Table 2 depicts the means and standard errors of adult outcomes for each profile for each sample. Table 3 depicts profile comparisons on predictors and adult outcomes. In the combined sample (rows 1-8), there were no intervention effects (row 1). Participants with higher levels of baseline aggression were more likely to belong to the high, moderate, and low profiles relative to the low-increasing profile (row 4, columns 3, 5, 6). Regarding adult outcomes, the high, moderate, and low profiles endorsed greater levels of aggression compared to the low-increasing profile (row 5, columns 3, 5, 6). The moderate and low profiles endorsed greater suicidal ideation/attempts than the high- and low-increasing profile (row 7, columns 1, 2, 5, 6). Finally, all profiles significantly differed on anxiety/depression (except for the moderate and low profiles) with the moderate profile showing the highest levels of anxiety/depression followed by the low, high, and low-increasing profiles (row 8). There were no significant differences in substance use (row 6).
Means and standard errors of adult outcomes for each profile from the BCH method
Profile comparisons from the BCH method
Note. Class membership is regressed on intervention status (in the combined sample) and predictors (i.e., baseline aggression, sex, race/ethnicity) and odds ratios (OR) are reported. Classes are compared on adult outcomes (i.e., aggression, substance use, suicidal ideation/attempts, anxiety/depression) using difference testing. H = high; M = moderate; L = low; LI = low−increasing; Intervention: 1 = Intervention, 0 = Nonintervention; Sex: 1 = Male, 0 = Female; Race: 1 = Black, 0 = white.
In the nonintervention sample (rows 9-15), similarly, participants with higher levels of baseline aggression were more likely to belong to the high, moderate, and low profiles relative to the low-increasing profile (row 11, columns 3, 5, 6). Regarding adult outcomes, the high, moderate, and low profiles endorsed greater levels of aggression compared to the low-increasing profile (row 12, columns 3, 5, 6). The moderate profile endorsed greater suicidal ideation/attempts relative to the high- and low-increasing profiles (row 14, columns 1, 5), and the low profile endorsed greater suicidal ideation/attempts relative to the low-increasing profile (row 14, column 6). The moderate and low profiles reported higher levels of anxiety/depression compared to the high- and low-increasing profiles (row 15, columns 1, 2, 5, 6). There were no significant differences between profiles on substance use (row 13).
In the intervention sample (rows 16–22), as well, participants with higher levels of baseline aggression were more likely to belong to the high, moderate, and low profiles relative to the low-increasing profile (row 18, columns 3, 5, 6). Regarding adult outcomes, the high, moderate, and low profiles endorsed greater levels of aggression compared to the low-increasing profile (row 19, columns 3, 5, 6). The moderate profile endorsed greater suicidal ideation/attempts relative to the low-increasing profiles (row 21, column 5), and the low profile endorsed greater suicidal ideation/attempts relative to the high- and low-increasing profile (row 21, columns 2, 6). The moderate profile reported higher levels of anxiety/depression compared to the high- and low-increasing profile (row 22, columns 1, 5), and the low profile reported higher levels of anxiety/depression relative to the low-increasing profile (row 22, column 6). There were no significant differences between profiles on substance use (row 20).
Discussion
In this study, we integrated items representing impulsive behavior across several measures included in five preventive interventions targeting childhood conduct problems, as well as items representing aggression, substance use, suicidal ideation/attempts, and anxiety/depression across ages 18–25 years. We performed LCGA to parse between-person heterogeneity in growth of impulsivity through childhood and adolescence, in addition to testing profile membership with regards to adult outcomes.
Impulsivity profiles through childhood and adolescence
LCGA evidenced four profiles of impulsivity through childhood and adolescence. Specifically, across samples, profiles were represented by high (59.8–63.6%), moderate (12.1–12.9%), low (10.8–11.4%), and low-increasing (14.3–15.9%) impulsive levels. A general decreasing pattern of impulsivity across childhood and adolescence was observed, which is consistent with research examining similar constructs such as self-regulation (Steinberg et al., Reference Steinberg, Icenogle, Shulman, Breiner, Chein, Bacchini, Chang, Chaudhary, Giunta, Dodge, Fanti, Lansford, Malone, Oburu, Pastorelli, Skinner, Sorbring, Tapanya, Tirado, Alampay, Al‐Hassan and Takash2018). Although criticized as an oversimplified perspective of neurobiological development (Casey et al., Reference Casey, Galván and Somerville2016), the dual systems model suggests that increasing self-regulation (or decreasing impulsivity) across childhood and into late adolescence/early adulthood may be explained by the later-maturating cognitive control system inhibiting approach and risky behaviors (and balancing the early-maturating socioemotional system, which amplifies affinity to risky activities; Steinberg et al., Reference Steinberg, Icenogle, Shulman, Breiner, Chein, Bacchini, Chang, Chaudhary, Giunta, Dodge, Fanti, Lansford, Malone, Oburu, Pastorelli, Skinner, Sorbring, Tapanya, Tirado, Alampay, Al‐Hassan and Takash2018). Identification of the low-increasing profiles might highlight a reporter bias of impulsive behavior as children age, such that impulsivity in early childhood is considered more normative than similar behaviors expressed in later childhood and adolescence.
Relative to other related symptoms, such as inattention, impulsivity is considered to manifest with greater stability and persistence (Willcutt et al., Reference Willcutt, Nigg, Pennington, Solanto, Rohde, Tannock, Loo, Carlson, McBurnett and Lahey2012). However, our findings add to a growing literature that undulations (see grade 5 in the Figures) or later onsets in neurodevelopmental traits may be more common than originally thought (Sibley et al., Reference Sibley, Arnold, Swanson, Hechtman, Kennedy, Owens, Molina, Jensen, Hinshaw, Roy, Chronis-Tuscano, Newcorn and Rohde2022). Shifts in the expression of such traits may be the consequence of time-varying environmental factors (e.g., the dip at grade 5 may reflect changes in the environment during the transition from primary to middle school) or underlying neurobiological susceptibilities (Sibley et al., Reference Sibley, Arnold, Swanson, Hechtman, Kennedy, Owens, Molina, Jensen, Hinshaw, Roy, Chronis-Tuscano, Newcorn and Rohde2022; Sonuga-Barke et al., Reference Sonuga-Barke, Becker, Bölte, Castellanos, Franke, Newcorn, Nigg, Augusto Rohde and Simonoff2022).
Findings may also relate to our measurement of impulsivity. Trials included in the present study were developed in the late 1980s and early 1990s. Although transdiagnostic and dimensional perspectives of mental health and illness were increasingly discussed during this time, it is only more recently that there has been a greater appreciation for processes that underlie psychopathology at multiple levels (Insel et al., Reference Insel, Cuthbert, Garvey, Heinssen, Pine, Quinn and Wang2010; Kotov et al., Reference Kotov, Krueger, Watson, Achenbach, Althoff, Bagby, Brown, Carpenter, Caspi, Clark, Eaton, Forbes, Forbush, Goldberg, Hasin, Hyman, Ivanova, Lynam, Markon, Miller, Moffitt, Morey, Mullins-Sweatt, Ormel, Patrick, Regier, Rescorla, Ruggero, Samuel, Sellbom, Simms, Skodol, Slade, South, Tackett, Waldman, Waszczuk, Widiger, Wright and Zimmerman2017). Despite our ability to capture several items relevant to impulsive behavior, most operationalizations of impulsivity emphasize a multidimensional structure. For example, the UPPS structures impulsivity through four or five distinct but correlated dimensions: urgency (negative and positive), lack of premeditation, lack of perseverance, and sensation seeking (Whiteside & Lynam, Reference Whiteside and Lynam2001). Other researchers have disentangled impulsivity from disinhibition (Joyner et al., Reference Joyner, Daurio, Perkins, Patrick and Latzman2021). Specifically, the authors found that impulsive traits were associated with both externalizing problems and negative affect, whereas disinhibition was only associated with externalizing psychopathology. Thus, while our study provides information that represents an advance in understanding of impulsigenic traits throughout childhood and adolescence, findings are constrained by our limited measurement. Clearly, similarly large-scale studies are needed that employ a more comprehensive assessment, including indicators of multiple dimensions of impulsivity.
It is also important to consider the design specifics of each trial and the integrative data processing when interpreting findings that synthesize multiple trials. As shown in the Supplementary Materials, there were a greater number of measures within trials in earlier timepoints and greater missing data levels across trials in later timepoints. Both factors may have contributed to the decreasing patterns observed in the high and moderate profiles. In addition, studies that draw on integrative data processing of multiple trials introduce many “researcher degrees of freedom” or require many decisions at every integrative stage (Simmons et al., Reference Simmons, Nelson and Simonsohn2011), each of which can have substantial impacts on final results. For example, we also coded our impulsivity items as z-scores and on a 0-2 Likert scale before creating mean scores across items. Both forms of coding resulted in poorer model fit (see Supplementary Materials), as well as differences in the shape of profiles. However, the psychopathology field has moved away from dichotomous coding except where necessary. Here, we want to loudly voice the importance of retaining continuous scores where possible in these emerging integrative data methodological practices. Integrative data analytic studies are a burgeoning approach and much more research is needed testing the implications of various decision-making steps (we refer the reader to the following papers for in-depth discussions of integrative data processing; Brincks et al., Reference Brincks, Montag, Howe, Huang, Siddique, Ahn, Sandler, Pantin and Brown2018; Hussong et al., Reference Hussong, Curran and Bauer2013).
Impulsivity profile comparisons on predictors and adult outcomes
We observed no significant intervention effects on profile membership. Collectively, our interventions included similar components targeting conduct problems underpinned by similar theoretical models. Nonetheless, trials differed in the number of years intervention components were applied, and we took an inclusive approach by clustering interventions within trials with multiple intervention components (e.g., PIRC1: GBG + Mastery Learning; PIRC2: classroom-centered + Family-School Partnership), which may have contributed to null findings.Footnote 2 However, there is extensive evidence of intervention effects on conduct problems from individual analyses of each trial (e.g., Conduct Problems Prevention Research Group, 2020; Eddy et al., Reference Eddy, Reid, Stoolmiller and Fetrow2003; Ialongo et al., Reference Ialongo, Domitrovich, Embry, Greenberg, Lawson, Becker and Bradshaw2019; Tolan et al., Reference Tolan, Gorman-Smith and Henry2004).
Some preliminary findings also support intervention impacts within these trials on impulsive behavior specifically (e.g., Fowler et al., Reference Fowler, Henry, Schoeny, Gorman-Smith and Tolan2014; Musci et al., Reference Musci, Bradshaw, Maher, Uhl, Kellam and Ialongo2014). Numerous questions remain, however, regarding the role of impulsivity in intervention effects. For example, is impulsivity a target or indirect target within these interventions? Emerging decomposition research on mechanisms of intervention-related change found that approximately a third of FT’s impact on antisocial outcomes was accounted for by improvements in both social and self-regulation skills (including prosocial behavior, emotion regulation, and problem-solving; Sorensen et al., Reference Sorensen and Dodge2016). Compared to research on moderators (Goulter & Fleming, Reference Goulter and Fleming2023; McMahon et al., Reference McMahon, Goulter and Frick2021), much less work has been conducted investigating mechanisms of change. Further research is needed testing theoretical mechanisms of action, and the cascading effects of these mechanisms on change in impulsivity and change in other constructs, such as conduct problems, seems of prime interest. Although we observed no significant intervention effects on profile membership, which may be in part due to decisions made in the integrative data processing stages as alluded to earlier, this form of research tapping questions such as “what outcomes?” and “under what conditions?” is important for advancing this field.
Across all samples, participants with higher levels of baseline aggression were more likely to belong to the high, moderate, and low profiles relative to the low-increasing profile, and participants in these profiles also reported higher levels of adult aggression than the low-increasing profile. These findings point to impulsivity as critical in externalizing psychopathology (Beauchaine et al., Reference Beauchaine, Hinshaw and Pang2010, Reference Beauchaine, Zisner and Sauder2017). However, across all samples, there were no significant differences between profiles on adult substance use. Relative to the role of impulsivity in substance use during childhood and adolescence, these findings may suggest that impulsivity does not underlie illicit substance use in adulthood.
By applying an integrative process to data from five randomized controlled trials, we produced a dataset with a large sample size and statistical power to optimally test subgroups, as well as increasing the number of participants with low base-rate behaviors such as suicidal ideation/attempts. In preventive interventions designed to address conduct and disruptive behavior disorders, this enables better evaluation of “cross-over” effects on outcomes not specifically targeted by the intervention (Ayer et al., Reference Ayer, Stevens, Reider, Sims, Colpe and Pearson2023; Goulter et al., Reference Goulter, McMahon and Dodge2019; Musci et al., Reference Musci, Kush, Masyn, Esmaeili, Susukida, Goulter, McMahon, Eddy, Ialongo, Tolan, Godwin, Coie, Crowley, Dodge, Greenberg, Lochman, McMahon, Pinderhughes and Wilcox2023; Reider et al., Reference Reider, Robertson and Sims2014). The lack of research across these preventive trials on suicidal ideation/attempts may be because individually these studies were not adequately powered to test for such cross-over effects—a limitation that can be circumvented by integrating multiple studies. Across samples, in general, the moderate and low profiles endorsed greater levels of adult suicidal ideation/attempts relative to the high- and low-increasing profiles. It may be the case that the interventions impacted adolescent suicidal behaviors, but not adult suicidal behaviors which tend to be less impulsive in nature—an important avenue for future research (Godwin et al., Reference Godwin, Coie, Crowley, Dodge, Greenberg, Lochman, McMahon and Pinderhughes2020). In addition, although meta-analytic work has identified impulsivity as an important marker of suicidal risk (McHugh et al., Reference McHugh, Lee, Hermens, Corderoy, Large and Hickie2019; Pérez-Balaguer et al., Reference Pérez-Balaguer, Peñuelas-Calvo, Alacreu-Crespo, Baca-Garcia and Porras-Segovia2022), another meta-analysis has stressed that many of the impulsivity measures used in this research have not adequately captured the impulsivity construct (Moore et al., Reference Moore, Doughty, Neumann, McClelland, Allott and O’Connor2022). Across samples, almost all profiles differed in adult anxiety/depression with the moderate profile showing the highest levels of anxiety/depression followed by the low, high, and low-increasing profiles.
Research and clinical implications
In this study, we leveraged data from multiple large, longitudinal databases, and in doing so, we have added to a growing pool of studies capitalizing on integrative data processes to answer important developmental research questions (e.g., Brown et al., Reference Brown, Brincks, Huang, Perrino, Cruden, Pantin, Howe, Young, Beardslee, Montage and Sandler2018; Musci et al., Reference Musci, Kush, Masyn, Esmaeili, Susukida, Goulter, McMahon, Eddy, Ialongo, Tolan, Godwin, Coie, Crowley, Dodge, Greenberg, Lochman, McMahon, Pinderhughes and Wilcox2023). As stated by Hussong et al., (Reference Hussong, Curran and Bauer2013, p. 3), “this amassing of rich data archives and advanced statistical analysis intersects with a scientific zeitgeist emphasizing collaborative (particularly transdisciplinary) efforts, fueling a “big science” initiative.” Integrative data processing, in conjunction with group-based analyses such as mixture modeling, opens the door for investigating effects that may not emerge in smaller samples, such as heterogenous subgroups, predictors and moderators of subgroup membership, lower base-rate behaviors, and cross-over effects.
From a clinical standpoint, one implication is that our findings provide important information regarding prevalence and developmental course of impulsivity, as well as prevalence of psychopathologies related to impulsivity. In addition, recent research emphasizes the importance of noncognitive skills across intrapersonal (e.g., emotion regulation) and interpersonal (e.g., prosocial behavior) domains (Godwin et al., Reference Godwin, Coie, Crowley, Dodge, Greenberg, Lochman, McMahon and Pinderhughes2020; Sorensen et al., Reference Sorensen and Dodge2016), and our findings add to this growing literature of possible noncognitive skills in relation to later outcomes. Our findings also point to potential differences in sex and racialized constructs/ethnicity (see Table 3, rows 2, 9, 16 and rows 3, 10, 17), respectively. It is important to note that race and ethnicity may serve as proxies for systemic inequalities. In our study, we were limited by the use of a binary sex and race variables, and further research is needed with samples of greater diversity and from the Majority World (Yu et al., Reference Yu, Goulter, Godwin and McMahon2023).
An important future direction includes the examination of trait versus state impulsivity. Recent advances in technology (e.g., smart phones and wearables) and intensive longitudinal statistical approaches have produced a flurry of studies testing the dynamics or fluctuations of emotional and behavioral problems at micro timescales (Zheng & Goulter, Reference Zheng and Goulter2023; for a review, see Urben et al., Reference Urben, Constanty, Lepage, Rosselet Amoussou, Durussel, Turri, Wouters, Mürner-Lavanchy and Plessen2022). Some research has also identified state-like fluctuations in other neurodevelopmental traits that were considered relatively stable (i.e., callous-unemotional traits; Goulter et al., Reference Goulter, Cooke and Zheng2023). Future intervention work might consider incorporating a measurement burst design combining micro and macro longitudinal methods to test the effects of environmental experiences on short-term impulsive dynamics and relations with long-term change.
Strengths and limitations
Methodological strengths of the present study include its integration of five randomized controlled trials. Preventive intervention trials included in the present analysis were large samples drawn from several sites across the U.S. The present study employed an integrative and mixture modeling data analytic process to extract multi-informant information (i.e., parent, child, and teacher reports) across childhood and adolescence. We also examined how profiles compared on a range of self-reported adult psychopathologies. Despite these strengths, some limitations should be considered. As noted earlier, impulsivity is best conceptualized as a multidimensional construct, and future randomized controlled trials are encouraged to attempt to balance the difficult methodological task of including an optimal number of measurement occasions and comprehensive assessment of key constructs while limiting participant burden. Abbreviated versions of multidimensional measures of impulsivity have now been validated in youth samples (Watts et al., Reference Watts, Smith, Barch and Sher2020), which may help to facilitate the assessment of this construct in resource-intensive designs. Specific measures of impulsivity, however, have typically been precluded from large-scale interventions, which require a suite of tools that provide information accessible and actionable to key stakeholders (e.g., prevalence of psychiatric disorders or criminal behavior). Future work might, similarly, extract and harmonize items related to impulsivity across a variety of related measures using integrative data analytic techniques to increase power and optimally test for effects. Another limitation is that although synthesizing data across several trials enables a much larger sample size and we applied appropriate missing data methods, some measurement occasions still had missing data that may not be best estimated with these methods (see Supplementary Materials for missing data levels and Brincks et al., (Reference Brincks, Montag, Howe, Huang, Siddique, Ahn, Sandler, Pantin and Brown2018) for a discussion of missing data methods in integrative designs).
Conclusion
In sum, this study evaluated between-person effects in impulsive characteristics across childhood and adolescence among participants derived from five longitudinal preventive interventions. We observed meaningful differences in stratified profile slopes, as well as how these profiles differed with respect to adult psychopathological outcomes. Overall, this study adds to emerging scientific collaborative efforts linking data across multiple studies to better inform understanding of developmental processes (Brown et al., Reference Brown, Brincks, Huang, Perrino, Cruden, Pantin, Howe, Young, Beardslee, Montage and Sandler2018; Musci et al., Reference Musci, Kush, Masyn, Esmaeili, Susukida, Goulter, McMahon, Eddy, Ialongo, Tolan, Godwin, Coie, Crowley, Dodge, Greenberg, Lochman, McMahon, Pinderhughes and Wilcox2023).
Supplementary material
The supplementary material for this article can be found at https://doi.org/10.1017/S0954579424000828.
Funding statement
Prevention Intervention Research Center Generation 1 (PIRC 1)
During the last 27 or more years this research has been supported by NIMH Grants R01 MH 42,968, P50 MH 38,725, R01 MH 40,859, and T32 MH018834 (with supplements from NIDA for each of the cited research grants), and NIDA grants R01 DA 019984, DA009897, and P30 DA027828.
Prevention Intervention Research Center Generation 2 (PIRC 2)
This research was supported by grants to NI from the National Institute of Mental Health (MH57005), and the National Institute on Drug Abuse (DA11796) as well as a grant from the National Institute on Drug Abuse to Brion Maher (DA036525). This work used data from the Center for Prevention and Early Intervention at the Johns Hopkins Bloomberg School of Public Health. The authors are grateful for the collaboration of the Baltimore City Public Schools, teachers, parents, and students who participated in the study. They also appreciate the work of COGA and SAGE groups. The authors have declared that they have no competing or potential conflicts of interest.
Linking the interest of families and teachers (LIFT)
Support for this project was provided by Grant R01 MH054248 from the Prevention and Behavioral Medicine Research Branch, Division of Epidemiology and Services Research,
NIMH, U.S. PHS; by Grant P30 MH 46,690 from Prevention and Behavioral Medicine Research Branch, Division of Epidemiology and Services Research, NIMH & ORMH, U.S. PHS; P30 DA 023920, Division of Epidemiology, Services and Prevention Branch, NIDA and by a center infrastructure development grant from the McConnell Clark Foundation. Thanks to the children and families who shared their lives with us during the tenure of the LIFT project; to the school administrators, teachers, and staff in the Bethel, Eugene 4 J, and Springfield public school districts, without whose support and efforts this study could not have been done; and to the many OSLC research scientists and professional research and administrative staff members who contributed their time, energy, and talents to the work across multiple decades.
SAFE children (SAFE)
Primary funding for both phases of Schools and Families Educating Children (SAFE) came from the Substance Abuse and Mental Health Services Administration with additional funding from the National Institute of Mental Health (MH048248) and the National Institute on Drug Abuse (R01DA020829-02).
Fast track (FT)
The FT project has been supported by National Institute of Mental Health (NIMH) Grants R18 MH48043, R18 MH50951, R18 MH50952, R18 MH50953, R01 MH062988, R01 MH117559, K05 MH00797, and K05 MH01027; National Institute on Drug Abuse (NIDA) Grants R01 DA016903, R01 DA036523, R01 DA11301, K05 DA15226, RC1 DA028248, and P30 DA023026; National Institute of Child Health and Human Development Grant R01 HD093651; and Department of Education Grant S184U30002. The Center for Substance Abuse Prevention also provided support through a memorandum of agreement with the NIMH. Additional support for this study was provided by a B. C. Children’s Hospital Research Institute Investigator Grant Award and a Canada Foundation for Innovation award (to Robert J. McMahon).
This work used data from the FT project (for additional information concerning FT, see http://www.fasttrackproject.org). We are grateful to the members of the Conduct Problems Prevention Research Group (in alphabetical order, Karen L. Bierman, Pennsylvania State University; John D. Coie, Duke University; D. Max Crowley, Pennsylvania State University; Kenneth A. Dodge, Duke University; Mark T. Greenberg, Pennsylvania State University; John E. Lochman, University of Alabama; Robert J. McMahon, Simon Fraser University and B.C. Children’s Hospital Research Institute, and Ellen E. Pinderhughes, Tufts University) for providing the data and for additional involvement. We are grateful for the collaboration of the Durham Public Schools, the Metropolitan Nashville Public Schools, the Bellefonte Area Schools, the Tyrone Area Schools, the Mifflin County Schools, the Highline Public Schools, and the Seattle Public Schools. We appreciate the hard work and dedication of the many staff members who implemented the project, collected the evaluation data, and assisted with data management and analyses.
Competing interests
Drs. Bierman, Coie, Dodge, Greenberg, Lochman, McMahon, and Pinderhughes are the Principal Investigators on the FT Project and have a publishing agreement with Guilford Publications, Inc. Royalties from that agreement are donated to a professional organization. They are also authors of the PATHS curriculum and donate all royalties from PATHS Program LLC. to a professional organization. Dr Greenberg is a developer of the PATHS curriculum and has a separate royalty agreement with PATHS Program LLC. Bierman, Coie, Dodge, Greenberg, Lochman, and McMahon are the developers of the FT curriculum and have a publishing and royalty agreement with Guilford Publications, Inc. McMahon is a coauthor of Helping the Noncompliant Child and has a royalty agreement with Guilford Publications, Inc. Dr Crowley has no conflicts.
Open access
