A data-driven approach to categorizing early life adversity exposure in the ABCD Study

The present study used the National Data Archive, ABCD version 2.01 baseline data set collected between 2016 and 2018 from the ABCD study, the largest longitudinal neuroimaging study of youth development. Over 10,000 youth from 21 different research sites in the United States are enrolled in this 10-year longitudinal study [17]. Procedures, sampling and recruitment [17‐19] for the ABCD study have been described previously. Caregivers provided written informed consent and children provided assent for participation in the study. All procedures were approved by a central institutional review board, and each site has a detailed protocol in place to address reported adversity exposure. The University of California, Los Angeles, institutional review board has indicated that analyses using the publicly released ABCD Study data are not human subjects research and therefore do not require their own approval.

The prevalence of at least one form of ELA was 92.1% among our sample of 11,566 9- and 10-year-olds across the following domains: 1) physical and sexual violence; 2) parental psychopathology; 3) neighborhood threat; 4) prenatal substance exposure; 5) scarcity; and 6) household dysfunction. 81.4% of youth were exposed to parental psychopathology, 42.1% reported household dysfunction, 19.9% experienced neighborhood threat, 11.7% faced scarcity, 10.6% were exposed to prenatal substance use, and 7.3% reported physical and sexual violence exposure. Youth with ELA and controls were statistically different from one another across the following sociodemographic characteristics: sex (χ² = 7.44; p = 0.006), race and ethnicity (χ² = 136.25; p < 0.0001), family income (χ² = 16.64; p = 0.002) and primary caregiver’s education (χ² = 12.29; p = 0.015) (see Table 1). Additionally, youth with ELA endorsed more internalizing, externalizing and total problematic behaviors (χ²(5, N = 11,566) = 8.84, p = 0.012).

No evidence for singularity and multicollinearity was found for the correlation matrix. Evaluation of the correlation matrix showed a correlation of the items between -0.089 and 0.691.

The determinant of the matrix was not equal to the identity matrix (< 0.00001). Bartlett’s test of sphericity was significant (χ²(1176,N = 11,566) = 80,020.79, p < 0.01), suggesting that there was enough variability in the items to perform the factor analysis. The Kaiser Meyer Olkin measure of sampling adequacy was acceptable at 0.80 and indicated common variance among the items.

Out of the 47 variables included in the factor analysis, 30 variables loaded on six unique domains. A six-factor solution was identified for the final factor analysis utilizing a principal factor solution and oblique rotation. The eigenvalue for the first six factors ranged from 1.02 to 3.45 and explained 22.4% of the variation in this construct. All other eigenvalues were less than 1 and accounted for less than 10% of the variation. Our selected model’s fit corresponds to: root mean square error of approximation (RMSEA) value = 0.02; and Tucker-Lewis index (TFI) and comparative fit index (CFI) values > 0.85. No variables loaded on more than one factor. As shown in Table 2, this solution gives clearly interpretable factors entitled: 1) physical and sexual violence; 2) parental psychopathology; 3) neighborhood threat; 4) prenatal substance exposure; 5) scarcity; and 6) household dysfunction.

Given the prevalence of parental psychopathology, a closer examination was conducted demonstrating that the greatest weight comes from the following three sub-types of parental psychopathology exposure: parental hospitalization due to mental health concerns (factor loading: 0.657; prevalence: 36.7%), parent utilization of mental health counseling due to mental health concerns (factor loading: 0.630; prevalence: 70.3%), and parental depression (factor loading: 0.605; prevalence: 61.3%). Additionally, there is a moderate correlation between parental hospitalization and parental utilization of mental health counseling (r(10,649) = 0.34, p < 0.0001). These questions were completed by the youth’s primary caregiver in regard to the youth’s biological parent; the two of which were not always the same.

The presence of clinically significant internalizing behaviors was reported in 17.7% of youth with at least one form of adversity exposure as opposed to 6.6% of controls (χ²(1,N = 11,566) = 74.28, p < 0.0001); clinically significant externalizing behaviors were reported in 11.1% of youth with at least one form of adversity exposure as opposed to 2.3% of controls (χ²(1,N = 11,566) = 70.84, p < 0.0001); total clinically significant problematic behaviors were evident in 13.0% of ELA youth versus 2.7% among controls (χ²(1,N = 11,566) = 83.45, p < 0.0001).

Youth with higher factor scores across the following domains had more internalizing problems: physical and sexual violence, parental psychopathology, and scarcity. Conversely, individuals with higher factor scores across the following domains had higher externalizing problems: neighborhood threat, prenatal substance exposure, and household dysfunction (see Table 3).

While controlling for age, sex, race/ethnicity of the youth, and family income, all forms of adversity exposure except for scarcity were significantly associated with greater internalizing, externalizing and total problematic behaviors (p < 0.0001) (see Tables 4, 5 and 6). In particular, parental psychopathology, household dysfunction and neighborhood threat demonstrated the greatest association with problematic behaviors, while controlling for age, sex, race, ethnicity and family income. A cumulative adversity exposure score was calculated for each youth, created by summing the adversity scores across the six domains. The relationship between cumulative adversity exposure and problematic behaviors is included in Table 6.

This study is, to our knowledge, the largest retrospective source of ELA derived from a population-based study of youth development. An EFA yielded a 6-factor solution corresponding to distinct domains of ELA, including: 1) physical and sexual violence; 2) parental psychopathology; 3) neighborhood threat; 4) prenatal substance exposure; 5) scarcity; and 6) household dysfunction. Our findings reveal that ELA prevalence among 9-and 10-year-old youth is largely driven by the incidence of parental psychopathology. The lifetime prevalence of any adult psychiatric disorder per DSM-IV diagnostic criteria has been estimated at 46.4% [24]. Our sample’s greater proportion (81.4%) may in part be attributable to a different measure being used to capture psychopathology and that the measure was not just completed by the youth’s biological parent but by the primary caregiver, which was not always the same. Therefore, parental psychopathology as an exposure may reflect both genetic and behavioral influences on our clinical outcomes and does not always equate with behavioral exposure in the instances where youth do not have contact with their biological parent(s) at baseline (n (%) = 443, 3.8%). Lastly, biological parental psychopathology reported by the caregiver is not equivalent to a clinical diagnosis.

Sociodemographic characteristics significantly differed between youth with adversity exposure and controls, specifically, sex, race/ethnicity of youth, primary caregiver’s education and family income. These findings are supported by previous research showing a higher incidence of ELA among racial and ethnic minorities, and among individuals identifying with low SES [4], the latter also associated with an increased risk for mental and physical health problems [25]. Adversity exposure was significantly associated with greater problematic behaviors, specifically, parental psychopathology, household dysfunction and neighborhood threat.

A 6-factor solution corresponding to 6 domains of ELA were derived from an EFA performed on 47 variables both youth and caregiver-reported across 14 measures. Seventeen variables were not included in the final EFA due to sparse endorsement of the variables which can in part be explained by the sensitive self-identifying nature of the questions, which were primarily caregiver-reported, as well as narrow time constraints referenced in the question, i.e., within the past 6 months. While the 6 domains of ELA are similar to the original ACEs, our domains differ in two prominent areas: incarceration of household member and neighborhood threat. We hypothesized that adversity domains derived from the EFA would overall align with and complement the domains established by the CDC-Kaiser ACE’s Study given that the original categorizations of exposure were broad yet discrete in nature. At baseline, the ABCD Study did not capture information on youth, caregiver, or household member incarceration. Given that one in three Americans will have an encounter with the criminal justice system, with racial and ethnic minorities carrying a significantly greater risk [26], capturing incidences of arrest, detainment, juvenile confinement, and adult incarceration are necessary to comprehensively catalog exposures that impact youth development. Not only does incarceration of a caregiver or family member constitute the removal of a source of support, a youth’s direct involvement with the justice system is associated with significant disadvantages (e.g., educational, economic, social, emotional, general health and wellbeing) throughout the lifespan [27]. Our study was, however, able to capture both youth and caregiver reported neighborhood threat. National survey data indicate that adolescent exposure to community violence is on par with adversity exposure within the home [28]. Irrespective of direct harm, community violence exposure constitutes a pervasive threat that accelerates biological aging and contributes to detrimental quality of life outcomes [29]. Despite not being captured in the original ACEs Study, more recent studies examining ELA are including measures of neighborhood or community threat and or violence [30]. Our findings support the literature detailing the increased incidence of problematic behaviors following neighborhood threat and community violence exposure [31]. Lastly, our EFA resulted in the combination of physical and sexual violence exposure into one domain versus two discrete categories of exposure. This may in part be explained by the minimal endorsement of these exposure types as well as that the same questionnaire (i.e., Kiddie Schedule for Affective Disorders and Schizophrenia (KSADS-5)) was used to measure physical and sexual violence exposure.

Our findings that youth with ELA endorsed more internalizing, externalizing, and total problematic behaviors, which is associated with psychopathology risk, is supported in the literature [32]. Unsurprisingly, half of all childhood-onset and about one-third of adolescent-onset psychiatric disorders are associated with early life adversity exposure [33]. Our findings that parental psychopathology, household dysfunction, and neighborhood threat carried the greatest influence on problematic behaviors among 9-and 10-year-olds in our sample suggest identifying sources of resiliency that may combat these specific forms of exposure. For example, resources within the school and community, such as school-based programs, athletic associations, and peer mentorships, may act as sources of support for youth who are experiencing adversity within the home and immediate environment.

Youth with higher factor scores across the following domains had more internalizing problems: physical and sexual violence; parental psychopathology; and scarcity. Conversely, individuals with higher factor scores across the following domains had higher externalizing problems: neighborhood threat; prenatal substance exposure; and household dysfunction. While ELA exposure does not typically occur in insolation [8], these associations suggest possible mechanistic differences in type-specific ELA’s impact on associated behaviors. The mechanistic differences may be attributable to an individual’s neurodevelopmental stage during exposure and or to the neurodevelopmental subtilties in how different forms of ELA are processed in a region-specific manner. Understanding the nuanced relationship between subtypes of ELA and different problematic behaviors may aid in the earlier identification of ELA exposure and targeted interventional efforts, particularly for those that may be less physically-apparent (e.g., parental psychopathology).

Our findings spotlight the need to develop data-driven approaches to the categorization of ELA, highlighting the need to examine nuances of exposure, e.g., type, age of onset, frequency, duration, and relationship with the perpetrator. The youth in our sample endorsed discrete forms of ELA, the incidence of which significantly differed by sex, race, ethnicity and other sociodemographic characteristics. Additionally, different forms of ELA were associated with specific problematic behaviors. The use of broad domains, such as abuse and neglect [13]; active and passive adversity [14]; and threat and deprivation [15] in place of type-specific ELA in an attempt to categorize exposures and outcomes fails to account for the duality of some forms of adversity and the routine co-occurrence of exposures. For example, household dysfunction or family conflict could include both active and passive adversity exposure if the youth witnesses exposure but is also the direct recipient of. Neighborhood threat often co-occurs with deprivation, specifically, a greater prevalence of violence exposure within low SES communities [34], as well as food insecurity and social deprivation [35]. Despite the endorsement of neighborhood threat in our sample, exposure is not routinely examined and is at times even combined with low SES. Metrics of cumulative adversity and type-specific ELA should both be reported given the heterogeneity in sociodemographic associations and behavioral implications of ELA exposure; the utilization of broad categories and domains may inadvertently obscure pertinent associations and homogenize findings. A systematic data-driven approach to measure and categorize ELA in youth could aid not only in establishing a consistent manner with which to define and measure ELA, but could improve study reproducibility, and elucidate nuances in associated outcomes (e.g., behavioral and physiological) to improve evidence-based treatments and interventions. We advocate for the nuanced categorization of type-specific ELA as well as the inclusion of neighborhood threat as a form of exposure.

To foster a systematic data-driven approach to measure and categorize ELA, we suggest the utilization of large publicly-available epidemiological datasets, including, the ABCD Study, Behavioral Risk Factor Surveillance System (BRFSS) surveys, and the National Longitudinal Study of Adolescent to Adult Health (Add Health). We recommend the utilization of rigorous yet generalizable statistical approaches, such as linear and logistic regression analyses with train and test models, in lieu of more dataset-specific methodologies, such as structural equation modeling [36‐38]. While the modeling of complex patterns of variables and relationships is afforded by structural equation modeling, the analytical steps necessary to generate such a model result in findings that well-reflect the specific dataset, and thus, limit the generalizability.