The association between early reports to Child Protective Services and developmental trajectories through middle childhood

Abstract

Background:

Longitudinal studies on resilience among children who have experienced maltreatment indicate that resilience is multi-dimensional. However, most research consolidates diverse developmental domains comprising resilience into a single score, which does not allow for detection of potentially heterogeneous associations between risk factors and outcomes of resilience processes.

Objective:

This study seeks to improve our understanding of the association between early child maltreatment and development through middle childhood (6–12 years) using individual domains considered to be outcomes of resilience processes.

Participants and Setting:

Participants are 499 children from the Longitudinal Studies of Child Abuse and Neglect.

Methods:

We used latent growth curve models to explore patterns of socialization and daily living skills, and internalizing and externalizing behaviors – outcomes of resilience processes -- across three time points in middle childhood, and their association with early maltreatment, defined as referral to Child Protective Services (CPS) before age 6.

Results:

In fully adjusted models, children experiencing early maltreatment had poorer baseline scores in activities of daily living (−4.22, 95% CI [−7.38, −1.46]) and externalizing behavior (2.95, 95% CI [1.05, 4.86]), but maltreatment was not associated with change over time in these domains. However, maltreatment was associated with increases in internalizing behavior over time (0.42, 95% CI [0.06, 0.77]).

Conclusion:

Heterogeneity in patterns of association between maltreatment and outcomes of resilience processes support the utility of examining developmental domains individually, versus as a composite, to identify specific targets for intervention.

1. Introduction

Child maltreatment is a serious threat to public health and child development in the United States (U.S.). Researchers estimate that 12.5% of children under 18 years of age in the U.S. experience maltreatment; the mean age that children are first reported to Child Protective Services (CPS) is 6.2 years (Wildeman et al., 2014). A review of literature documenting the consequences of child maltreatment in high-income countries identified evidence of a range of physical and mental health challenges, including behavior problems, post-traumatic stress disorder, depression, attempted suicide, alcohol misuse, and chronic health conditions (e.g., ischemic heart disease, lung disease, liver disease, and cancer) (Gilbert et al., 2009). By understanding the risk and protective factors underlying variation in these outcomes, research can inform opportunities to promote resilience – generally considered the ability to thrive in spite of adversity (Cicchetti, 2013; Masten, 2018) – among children who experience maltreatment. However, progress in the literature on resilience is hampered by the absence of consensus on the definition or measurement of resilience (Cicchetti, 2013; Happer, Brown, & Sharma-Patel, 2017; Yoon et al., 2019).

1.1. Defining and Measuring Resilience

Researchers commonly define resilience as competence – or performance within a standardized normal or non-clinical range – in one or more domains at a given time (Yoon et al., 2019). However, competence may be a misleading definition, as it suggests having reached a permanent status or goal. Studies suggest that classifying a child as resilient at one time point may not hold true across future developmental time points (Dubowitz et al., 2016; Happer et al., 2017; Jaffee & Gallop, 2007; Yoon et al., 2019). For example, a study of resilience patterns among a nationally representative sample of school-age children (8–16 years) who were referred to CPS defined resilience as competence in mental health (e.g., internalizing and externalizing behaviors), social skills, or academics (e.g., math and reading achievement) at any time during a 36-month post-investigation follow-up period. Findings indicated that 37% to 49% of children demonstrated resilience at any time over the study period, but only 14% to 22% consistently demonstrated resilience across the entire study period (Jaffee & Gallop, 2007), suggesting fluidity over time rather than static competence. Similarly, in an analysis of young children (ages 4–6) from the Longitudinal Study of Child Abuse and Neglect (LONGSCAN) sample, 45% of children demonstrated social competence – a composite inventory of social, communication, adaptive behavior, psychomotor and cognitive skills – at age 4 and 77% demonstrated social competence at age 6, but only 38% of children demonstrated social competence at both ages (Dubowitz et al., 2016). Maltreatment (operationalized as a report to CPS before age 4) was associated with a decrease in the odds of a child having social competence across ages (Dubowitz et al., 2016).

Measures of resilience as competence are also inconsistent. A recent systematic review identified seven studies published over the past decade that measured resilience in school-age children (ages 6–12 years) (Yoon et al., 2019). Studies included in this review operationalized resilience as either a lack of psychopathology, including internalizing and externalizing behaviors, competence across multiple social, emotional and behavioral domains, or as outcomes or processes related to adaptive functioning (Yoon et al., 2019). While each study included in this review operationalized the construct differently, many of the studies that assessed resilience as competence in social, emotional and behavioral development used multidomain composite measures (Yoon et al., 2019). Findings from this review are consistent with a previous review, which concluded that resilience is often determined based on measures of social, emotional and behavioral domains but how those domains are measured and the methods used to classify resilience differ considerably (Klika & Herrenkohl, 2013).

To address these conceptual and measurement challenges, leaders in the field of resilience science are advocating for a common, scalable definition of resilience as “the capacity of a dynamic system to adapt successfully through multisystem processes to challenges that threaten the function, survival, or development of the system” (Masten, Lucke, Nelson, & Stallworthy, 2021, p. 524). A capacity to adapt could vary across developmental periods and contexts, and variation in capacity may be reflected in heterogeneous patterns of social and behavioral skills. Therefore approaches that measure and model individual developmental outcomes over time as proxies for the resilience process may be more informative than combining measures and modeling resilience as a single construct (Kaufman, Cook, Arny, Jones, & Pittinsky, 1994; Kim-Cohen, Moffitt, Caspi, & Taylor, 2004; Vanderbilt-Adriance & Shaw, 2008). We embrace this definition of resilience as a capacity to adapt across dynamic systems and operationalize measures of psychopathology and positive development as outcomes of the adaptation process.

1.2. Theoretical Framework

Developmental systems theory posits that development occurs in the context of relationships between systems and individuals, and that resilience processes are functions of those relationships (Masten, 2018). The Family Stress Model demonstrates how two contextual factors, maternal depression and poverty, increase risk for both child maltreatment and poor developmental outcomes and act as confounders of the association between child maltreatment and subsequent development. (Conger, Rueter, & Conger, 2000; Conrad-Hiebner & Byram, 2020).

Maternal depressive symptoms are a well-documented risk factor for child maltreatment. For example, a longitudinal study of mothers involved with CPS found an increased risk of psychological aggression towards children in the 12-month period following the onset of maternal depression (Conron, Beardslee, Koenen, Buka, & Gortmaker, 2009). Maternal depression may also influence of maternal emotional availability to support a child’s development. A meta-analysis of studies examining the association between maternal depression and child outcomes synthesized findings suggesting that maternal depressive symptoms are associated with increased risk for child internalizing problems, externalizing problems, general psychopathology, and negative affect/behavior (Goodman et al., 2011).

Poverty is also consistently recognized as a risk factor for child maltreatment and is associated with increased risk for children’s poor health, achievement and behavioral outcomes (Brooks-Gunn & Duncan, 1997; Drake & Jonson-Reid, 2014). Although all families have inherent strengths, lower income children, compared to wealthier children, are more likely to live in households that experience greater material hardship and have parents who are more stressed and depressed (Berger, Paxson, & Waldfogel, 2009), which elevates the possibility of maltreatment (Conger et al., 2000; Conrad-Hiebner & Byram, 2020). Poverty is also associated with children’s language development, aggression, behavior problems, and internalizing behaviors (i.e., withdrawn, anxious) (Berger et al., 2009). It is important to control for maternal depressive symptoms and poverty as confounders when examining associations between child maltreatment and developmental outcomes.

1.3. Study Aim

This study examines the association between child maltreatment, defined as referral to CPS, and outcomes of resilience processes, operationalized as four individual, repeated measures of social and behavioral outcomes (socialization skills, daily living skills, internalizing behaviors and externalizing behaviors). We use latent growth curve models to examine linear trajectories of these outcomes over three time points during middle childhood (ages 6, 8, and 12 years), a key developmental period for growth in reasoning and self-regulation. Using latent growth curve models allows us to expand on the current understanding of developmental risks at discrete time points by taking the average of individuals’ standardized developmental scores at baseline and the rate of developmental change over time. By modeling each developmental outcome separately, we also advance a more nuanced understanding of the multidimensional resilience process we are approximating.

We hypothesize that children referred to CPS before age 6 years, compared to children who are never referred to CPS by age 12 years, will have poorer socialization and daily living skills and higher internalizing and externalizing behavioral problem scores at age 6 (i.e., intercept). Based on previous literature demonstrating greater increases in behavioral problems from age 3 to age 9 years among children who experienced maltreatment compared to children who did not experience maltreatment (Font & Berger, 2015), we hypothesize that the rate of change (i.e., slope) of each outcome domain will be greater in magnitude among children who were referred to CPS before age 6 compared to those children who were never referred to CPS and gaps between developmental outcomes will widen over time. As the mean age of first report to CPS is 6.2 years (Wildeman et al., 2014) and the majority of CPS reports occur prior to age 6 years (U.S Department of Health & Human Services, Administration on Children and Families, Administration on Children Youth and Families, & Children’s Bureau, 2020), by measuring CPS referrals before age 6 we capture the majority of cases ever reported to CPS.

2. Methods

We describe four social and behavioral developmental domain trajectories in a high-risk sample of children from age 6 to 12 years using latent growth curve models. We examine whether referral to CPS before age 6 is associated with differences in the mean baseline score (intercept) of each domain at age 6 years and the rate of change (slope) of domain trajectories from age 6 to age 12 years.

2.1. Sample

We analyzed data from a sub-sample of LONGSCAN participants. The LONGSCAN study uses the ecological developmental framework to situate relationships between maltreatment risk/history, child and parent/family characteristics, family functioning, extrafamilial relationships, community ecology, systems of care, and child outcomes (Bronfenbrenner, 1994; Runyan et al., 2014). LONGSCAN is a consortium of five different study samples that were each recruited based on varying degrees of risk. Two study sites recruited all participants based on referral to CPS. The remaining three sites recruited samples based on other risk factors including socioeconomic status, family structure, young maternal age, and low birth weight (Hunter & Knight, 1998). We limited the current sample to children who participated in data collection at the age 6-year visit (n=1176). We used data on the outcome measures from the age 6-year, 8-year and 12-year visits. We excluded children with CPS contacts between age 6 and age 8 visits and age 8 and age 12 visits (n=67), as we are interested in the unique effects of early childhood maltreatment rather than repeated or chronic maltreatment throughout middle childhood. We also excluded children from two of the study sites (Southwest and Northwest) (n=511) because these study sites recruited families based on previous CPS contact, therefore excluding the possibility of their belonging to our reference group of children without CPS contact before age 6 years. Finally, we excluded one child from the analytic sample because they were missing all demographic information. Due to missing data on the age 6 confounders (i.e., maternal depressive symptoms, income-to-needs ratio), the final model included 499 children (Table 1). Children who were excluded from the sample due to missing data on age 6 confounders did not significantly differ from children who were included in the final sample on any of the reported demographic factors or CPS referral status. From this sample of 499 children, we identified an exposure group of children who were referred to CPS before age 6 (n=198) and a control group of children who were never referred to CPS before age 12 (n=301) (Figure 1).

Table 1.

Sample Description

	Total N (%)	CPS Referral N (%)	No CPS Referral N (%)	Test of significance
Total	499	198 (39.7%)	301 (60.3%)
Center				χ2=0.25 (p=0.88)
East	183 (36.7%)	70 (35.4%)	113 (37.5%)
Midwest	153 (30.7%)	62 (31.3%)	91 (30.2%)
South	163 (32.7%)	66 (33.3%)	97 (32.2%)
Child Sex				χ2=1.08 (p=0.30)
Female	268 (53.7%)	112 (56.6%)	156 (51.8%)
Male	231 (46.3%)	86 (43.4%)	145 (48.2%
Child Race				χ2=4.97 (p=0.29)
Non-Hispanic White	83 (16.6%)	26 (13.1%)	57 (18.9%)
Non-Hispanic Black	360 (72.1%)	151 (76.3%)	209 (69.4%)
Hispanic	21 (4.2%)	7 (3.5%)	14 (4.7%)
Other	35 (7.0%)	14 (7.1%)	21 (7.0%)
Annual Household Income (Age 6)				χ2=8.12 (p=0.004)
<$5,000	105 (21.0%)	45 (22.7%)	60 (19.9%)
$5,000 – $9,999	129 (25.9%)	64 (32.3%)	65 (21.6%)
$10,000 – $14,999	89 (17.8%)	33 (16.7%)	56 (18.6%)
$15,000 – $19,999	55 (11.0%)	20 (10.1%)	35 (11.6%)
$20,000 – $24,999	39 (7.8%)	12 (6.1%)	27 (9.0%)
$25,000 – $29,999	28 (5.6%)	10 (5.0%)	18 (6.0%)
$30,000 – $34,999	17 (3.4%)	8 (4.0%)	9 (3.0%)
>$35,000	37 (7.4%)	6 (3.0%)	31 (10.3%)

Figure 1.

Sample selection. Boxes with dashed lines indicate respondents who were removed from the sample.

2.2. Variables

2.2.1. Outcome Variables.

We modeled four domains of social and behavioral development as outcomes in this study: (1) socialization skills, (2) daily living skills, (3) internalizing behaviors, and (4) externalizing behaviors. As discussed above, these domains are commonly used to measure resilience outcomes to indicate normative child development and the presence or absence of psychopathology. Based on our understanding of resilience as a process, we do not define a cutoff at which a child would be considered resilient on these social and behavioral outcomes, but rather use these measures to further our understanding of developmental trends during the resilience process following maltreatment experiences in early childhood. We examine trends in each of these outcome domains individually, rather than as composite measures, to better assess potential heterogeneity in developmental trajectories.

Socialization and Daily Living: Vineland Screener.

Socialization and Daily Living scores are continuous variables derived from the Vineland Screener Socialization and Daily Living domains (Sparrow, Carter, & Cicchetti, 1993). Interviewers administered the Vineland Screener in a semi-structured interview format with respondent caregivers. Each Vineland Screener domain includes 15 items measuring developmentally appropriate skills. Each item is scored (0=skill is not performed by child, 1=emergent skill performance, 2=satisfactory and habitual skill performance) and a total raw score is calculated by summing all domain items. Raw scores for the Socialization and Daily Living domains are age-standardized based on a normative sample (Mean=100, SD=15). Higher domain scores indicate more advanced skills as reported by the respondent caregiver. The Vineland Screener has demonstrated high interrater reliability (α=0.98) and domain validity (Socialization: r=0.92, Daily Living: r=0.93) (Sparrow et al., 1993).

Internalizing and Externalizing Behaviors: Child Behavior Checklist.

Internalizing and externalizing problems were measured using the Child Behavior Checklist/4–18 (CBCL) (Achenbach, 1991). Respondent caregivers completed the CBCL at the age 6-, 8-, and 12-year visits. We used age-standardized T scores for total internalizing and total externalizing problems for the present analysis. T scores below 60 for both scales are in the normal (non-clinical) range, while scores ranging from 60–63 are borderline clinical and scores greater than 63 are in the clinical range (Achenbach, 1991). To maximize variability and nuance in internalizing and externalizing trajectories, we maintained the continuous scales rather than categorizing scores as non-clinical, borderline, or clinical. The CBCL has demonstrated reliability, content, construct and criterion-related validity, and was normed on a national sample of children representative of the contiguous 48 states based on socioeconomic status, ethnicity, geographic region, and urban/suburban/rural residence (Achenbach, 1991).

2.2.2. Exposure Variable.

Early Child Maltreatment: CPS Referral.

We coded CPS referral prior to the age 6 visit as a dichotomous variable with 1 indicating one or more referral(s) to CPS dated before the age 6-year visit.

2.2.3. Confounders.

Maternal Depressive Symptoms.

Interviewers measured maternal depressive symptoms at the age 6-year visit using the Center for Epidemiological Studies Depression Scale (CES-D). The CES-D contains 20 items, each measured on a 4-point scale (0=rarely or none of the time, 3=most or all of the time), that are summed for a total possible score of 60 with higher scores indicating more depressive symptoms. The CES-D is a reliable (α=.085 to 0.90) and valid measure of depressive symptoms across racial, gender, and age demographics (Radloff, 1977). The CES-D maternal depressive symptom score is a continuous variable in the present analyses.

Income-to-Needs Ratio.

We used an income-to-needs ratio as a proxy measure of poverty as the majority of the LONGSCAN sample is low-income. We calculated the income-to-needs ratio by dividing the total household annual income by the total number of household members dependent on that income as reported at the age 6-year visit. LONGSCAN documented the total household annual income as a categorical variable ranging from <$5,000 to >$50,000 per year in $5000 increments. We used the median income for each annual household income category in our ratio calculations.

Race.

Respondent caregivers reported the racial or ethnic group that best described the child. For the present analysis, we dummy coded the child’s race/ethnicity (non-Hispanic Black, non-Hispanic White, Hispanic, or other) with non-Hispanic Black race coded as the reference group. We grouped children who identified as Mixed Race, Native American or Asian with children whose race was identified as “Other” due to small cell sizes.

Gender.

LONGSCAN used a binary measure of children’s gender: female (0, reference) or male (1).

Study Site.

LONGSCAN recruited samples from five study sites across the U.S. in the East, Midwest, Northwest, South, and Southwest. As previously mentioned, we excluded children recruited through the Northwest and Southwest sites from the present analysis because recruitment criteria included previous contact with CPS. Because each site used different sampling criteria, it is essential to control for subjects’ study site. We coded the remaining study sites – Midwest, South, and East –as dummy variables with the East study site as the reference group.

2.3. Analysis

In this study, we first examine bivariate demographic characteristics and mean values of the four outcomes of interest at ages 6, 8 and 12 years. We then use latent growth curve models to explore patterns of change in socialization, daily living skills, internalizing behaviors, and externalizing behaviors across three time points. Growth curve models are useful for assessing individual differences observed in developmental indicators over time and produce more flexible models of change over time than traditional ANOVA or hierarchical linear models (Curran & Hussong, 2003; Felt, Depaoli, & Tiemensma, 2017). This is because latent growth curve models estimate the latent intercept and latent slope based on individual growth trajectories, which allows for individual variance in the baseline measure, rate of change and covariance between the baseline measure and rate of change (Curran & Hussong, 2003).

We are interested in the differences in social and behavioral outcomes between children who did and did not experience maltreatment in early childhood, defined as a referral to CPS before age 6 years, as well as within person changes over time in social and behavioral outcomes. In addition to the latent characteristics of this model, predictions are subject to random error. We use a system of equations to define this model. The level 1 equation is

$$Y_{ij}={\beta }_{0j}+{\beta }_{1}ag{e}_{ij}+{\beta }_{2}CPSReport+{\beta }_{4}X+{\epsilon }_{ij}$$

Where Y_ij is the domain score for individual child j at age i, X is the vector of time-invariant independent variables and covariates, and ε_ij is the age-specific random effect.

The level 2 equation is

$${\beta }_{0j}={\beta }_0+{\mu }_j$$

Where β_0j is the individual child intercept and μ_j is the child-specific random effect.

The combined model is

$$Y_{ij}={\beta }_0+{\beta }_{1}ag{e}_{ij}+{\beta }_{2}CPSReport+{\beta }_{4}X+{\mu }_j+{\epsilon }_{ij}$$

We conducted latent growth curve analyses using the PROC MIXED command in SAS 9.4 (SAS Institute Inc., 2017). A latent growth curve model requires at least three time points and can accommodate measurements at time points that are not equally spaced by specifying the correlation parameters (Duncan & Duncan, 2009). In this analysis, time points are unequally spaced. To account for the difference in the two- and four-year gaps, we use an unstructured variance-covariance model to allow unique values for each variance and covariance (Bell, Smiley, Ene, & Blue, 2013). We conducted a complete case analysis, utilizing all cases with non-missing variables at age 6 years. We used a model building approach to first estimate unconditional models looking only at change over time in the outcome domains. Time is measured using an indicator for the age 6, age 8, and age 12 visit that was transformed to allow the age 6 visit to equal zero. We then used a series of four nested conditional models to introduce covariates: 1) referral to CPS + referral to CPS * time, 2) +demographic variables, 3) +maternal depressive symptoms +maternal depressive symptoms * time, and 4) +income-to-needs ratio, +income-to-needs ratio* time. We include interaction terms to model and control for the influence of referral to CPS, maternal depressive symptoms, and the income-to-needs ratio on the rate of change in developmental outcomes over time. Only the results of the final model are presented in this manuscript. This secondary analysis was designated as Non-Human Subjects Research by the Institutional Review Board at University of North Carolina at Chapel Hill.

3. Results

3.1. Sample Demographics

More than one-third of the sample (39.7%, n=198) was referred to CPS before age 6 years (Table 1). Each study site contributed to approximately one-third of the sample, with a slightly larger number of children included from the East study site (East: n=183, 36.7%; Midwest: n=153, 30.7%; South: n=163, 32.7%). Just over half of the sample is female (n=268, 53.7%). The majority of the sample is non-Hispanic Black (n=360, 72.1%); 16.6% (n=83) is non-Hispanic white, 4.2% (n=21) is Hispanic, and 7.0% (n=35) Other (i.e.., American Indian or Alaska Native, Asian, mixed race or another race). These demographic distributions were not significantly different between children who were referred to CPS before age 6 years and those who were not referred to CPS. This high-risk sample is skewed such that a large proportion of the sample (64.7%) reported annual household incomes in the lowest three income categories. A Kruskal-Wallace test of significance revealed differences in annual household income such that the average household income was lower among families referred to CPS (χ²=8.12, p=0.004).

3.2. Bivariate Analysis of Outcomes

Independent sample t-tests compared children who were referred to CPS before age 6 years and those who were not referred to CPS on each of the four outcomes of interest at ages 6 years, 8 years, and 12 years (Table 2). Children referred to CPS had significantly lower mean socialization scores at age 8 (2.07, p<.05) and age 12 (t=3.13, p<.01), lower mean daily living scores at age 6 (t=−3.71, p<.01), indicating poorer socialization and daily living skills at those time points. Children who were referred to CPS had higher internalizing scores at age 12 (t=−2.70, p<.01) and higher mean externalizing behavior scores at all three time points, (age 6: t=−3.31, p<.01; age 8: t=−2.66, p<.01; age 12: t=−2.41, p<.05) indicating more problem behaviors. There were no differences between children who were referred to CPS and those who were not referred to CPS in socialization scores at age 6 years, in daily living skills at age 8 years or 12 years, or in internalizing scores at age 6 years or 8 years.

Table 2.

Mean (SD) Outcome Scores at Ages 6, 8, and 12 years (N=499)

	6 years			8 years			12 years
	CPS Ref.	No CPS Ref.	t (df)	CPS Ref.	No CPS Ref.	t (df)	CPS Ref.	No CPS Ref.	t (df)
Socialization	89.76 17.3)	92.51 (15.04)	1.81 (369.89)	87.39 (18.38)	90.96 (17.11)	2.07 (440)*	82.20 (22.75)	89.60 (21/26)	3.13 (356)**
Daily Living	88.71 (17.88)	94.46 (14.83)	3.71 (356.56)**	93.43 (19.63)	95.98 (15.66)	1.43 (295.22)	88.62 (19.46)	92.33 (18.70)	1.81 (358)
Internalizing	50.69 (10.33)	50.13 (9.16)	−0.63 (493)	50.84 (49.21)	49.73 (9.90)	−1.11 (441)	51.38 (11.31)	48.30 (9.99)	−2.70 (356)**
Externalizing	55.66 (12.04)	52.29 (9.57)	−3.31 (355.79)**	53.78 (12.27)	50.78 (10.19)	−2.66 (307.34)**	54.36 (12.71)	51.19 (10.63)	−2.41 (266.52)*

^*p <.05

^**p<.01

^***p<.0001

3.3. Latent Growth Curve Analysis

3.3.1. Socialization.

The average socialization score at age 6 years was 99.37 (p<.0001) (age normalized mean=100, SD=15) (Table 3). Significant inter-individual variation (${\tau }_0^2=97.11$, SE=14.55, p<.0001) indicates between person differences in socialization skills at age 6 years. Referral to CPS before age 6 was not associated with socialization scores at age 6 years (−1.90, 95% CI [−4.48, 0.68], p=0.15) when controlling for maternal depressive symptoms and income-to-needs ratio. Socialization scores did not change significantly over time (−0.35, 95% CI [−1.13, 0.43], p=0.38) (Figure 2) and there was not a significant association between referral to CPS and average changes in socialization scores over time (−0.61, 95% CI [−1.35, 0.11], p=0.10).

Table 3.

Fully Adjusted Growth Curve Models (N=499)^a

	Socialization Coef. (95% CI)	Daily Living Coef. (95% CI)	Internalizing Coef. (95% CI)	Externalizing Coef. (95% CI)
Intercept	99.37 (95.97, 102.77)***	96.34 (92.68, 99.99)***	45.15 (43.18, 47.13)***	49.01 (46.68, 51.33)***
Referral before age 6 visit	−1.90 (−4.48, 0.68)	−4.83 (−7.70, −1.97)**	0.01 (−1.50, 1.53)	2.96 (1.18, 4.74)**
Slope	−0.35 (−1.13, 0.43)	0.12 (−0.64, 0.88)	0.27 (−0.11, 0.66)	0.54 (0.14, 0.94)**
Time*Referral	−0.62 (−1.35, 0.11)	0.39 (−0.32, 1.09)	0.42 (0.06, 0.77) *	−0.09 (−0.46, 0.28)
Random Variances
${\tau }_0^2$ (Intercept)	97.11 (SE=14.55)	136.22 (SE=17.83)***	34.16 (SE=5.02)***	65.62 (SE=6.47)***
${\tau }_1^2$ (Slope)	4.83 (SE=1.04)***	4.02 (SE=1.04)***	0.40 (SE=0.28)	1.06 (SE=0.28)***
Covariance between intercept and slope	4.70 (SE=3.02)	−7.69 (SE=3.42)	0.98 (SE=0.90)	1.20 (SE=1.04)

^***p<.0001

^**p<.01

^*p<.05

^aAll estimates are adjusted for child gender, race, study site, maternal depressive symptoms, and income-to-needs ratio.

Figure 2.

Mean socialization skills over time by CPS referral status, controlling for demographic and interpersonal covariates.

3.3.2. Daily Living Skills.

The average daily living skills score at baseline was 96.34 (p<.0001) (age normalized mean=100, SD=15) (Table 3). Significant inter-individual variation (${\tau }_0^2=136.22$, SE = 17.83, p<.0001) indicates between person differences in daily living skills at age 6 years. Mean daily living skills scores for children who were referred to CPS before 6 years of age were significantly lower (−4.83, 95% CI [−7.70, −1.97], p<.01) at age 6 years (Figure 3).

Figure 3.

Mean daily living skills over time by CPS referral status, controlling for demographic and interpersonal covariates.

3.3.3. Internalizing Behaviors.

The average child had an internalizing score of 45.15 (p<.0001) at the baseline age 6 visit (Table 3). Mean internalizing scores remained below the clinical cutoff score of 63 across the study period. Referral to CPS before age 6 years was not significantly associated with differences in internalizing scores at age 6 years (0.01, 95% CI [−1.50, 1.53], p=0.99), but was associated with an increase in the rate of change of internalizing symptoms over time (0.42, 95% CI [0.06, 0.77], p<.05) (Figure 4).

Figure 4.

Mean internalizing behavior score over time by CPS referral status, controlling for demographic and interpersonal covariates.

3.3.4. Externalizing Behaviors.

The average child had an externalizing score of 49.01 (p<.0001) at the age 6 visit and externalizing scores increased at an average rate of 0.54 (95% CI [0.14, 0.94], p<.01) points per year (Table 3). Mean externalizing scores remain below the clinical cutoff score of 63 across the study period. Being referred to CPS before age 6 years was associated with higher baseline externalizing scores at age 6 years (2.96, 95% CI [1.18, 4.74], p<.01), controlling for the income-to-needs ratio and maternal depressive symptoms. However, the non-significant interaction term between visit and referral to CPS indicates that externalizing symptoms increased at the same rate, regardless of referral status (Figure 5).

Figure 5.

Mean externalizing behavior score over time by CPS referral status, controlling for demographic and interpersonal covariates.

4. Discussion

We examined the associations between referral to CPS before age 6 years and the trajectories of four developmental domains through age 12 years: (1) socialization skills; (2) daily living skills; (3) internalizing behaviors; and (4) externalizing behaviors. These four outcomes are distinct but related developmental domains that have all been used in previous studies as measures of resilience constructs (Yoon et al., 2019). Our hypotheses that referral to CPS before 6 years of age would be associated with increased risk of internalizing and externalizing behaviors and decreased socialization skills and daily living skills over time were partially supported.

CPS referral before age 6 was associated with some but not all of the four developmental outcomes at age 6 years after controlling for confounders. Children who were referred to CPS before age 6 years had higher average externalizing behavior scores and lower average daily living skills scores at baseline than children who were never referred to CPS. However, referral to CPS before age six was not significantly associated with baseline internalizing behaviors or socialization skills.

CPS referral was also associated with differences in the slope of linear trends over time in some but not all domains after controlling for confounders. Referral to CPS was associated with a small but significant increase in the rate of change of internalizing behaviors over time. Children who were referred to CPS did not demonstrate significantly different internalizing behavior scores at ages 6- and 8-years compared to children who were not referred to CPS, but did demonstrate significantly higher internalizing behavior scores by age 12. While not statistically significant, trends suggest a small decrease in socialization scores but increase in daily living skills scores over time among children who were referred to CPS. The rate of change of externalizing symptoms did not significantly differ between children who were and were not referred to CPS, but the difference in externalizing behaviors at baseline was sustained. Children who were referred to CPS before age 6 continued to have higher externalizing behavior scores through age 12.

These findings are consistent with previous literature that found maltreatment before age 4 years predicted changes in anxiety/depression symptoms over time but did not predict changes in aggression behavior (Thompson & Tabone, 2010). CPS referral is a proxy measure of early childhood trauma, which may impact children’s ability to self-regulate (van der Kolk & Fisler, 1994). Higher externalizing scores and lower daily living skills scores are both indicative of poorer emotion regulation and independence, while increasing internalizing behaviors and decreasing socialization skills over time may reflect challenges in adapting to age-typical prosocial skills that are expected to develop across middle childhood (Briggs-Gowan, Carter, & Ford, 2012). Examining domain-specific trends in development through middle childhood demonstrates heterogeneity in the patterns of association between referral to CPS and distinct developmental outcomes.

Researchers should consider whether to model resilience as a single composite construct measured at one or more time point(s) or, rather, study the trajectories of individual developmental domains and how promotive factors may be differentially associated with different domains (Vanderbilt-Adriance & Shaw, 2008). The heterogeneity in patterns of association between CPS referral and each outcome domain at baseline and over time supports our position that it may be more informative to examine individual developmental domain trajectories than a single resilience construct that is a composite of multiple developmental domains. Some developmental domains may be more likely than other domains to be affected by experiences of maltreatment. This study suggests that children who are referred to CPS before age 6-years may be more likely to display externalizing behaviors and less developed daily living skills at age 6, whereas referral to CPS before age 6-years is not associated with internalizing behaviors or socialization skills at age 6. While referral to CPS before age 6-years is not associated with differences in internalizing behaviors at age 6 years, it is associated with greater increases in internalizing behaviors over time. Understanding unique patterns of developmental risks in these domains across middle childhood has important implications for how interventions are designed to support children’s developmental needs.

The home and school environments are two key contexts for intervention to support children’s capacity for positive development. Evidence-based home visiting programs provide opportunities to educate families on child health and development, identify strategies for supporting positive social and behavioral development, and address family stressors (Duffee, Mendelsohn, Kuo, Legano, & Earls, 2017). Rigorous implementation of home visiting programs can help to both prevent child maltreatment and support children’s unique developmental needs (Avellar & Supplee, 2013). Schools can also implement trauma-informed approaches to foster children’s capacity for positive development (Wall, 2021). The National Child Traumatic Stress Network suggests that trauma-informed schools provide a systematic approach to addressing social, emotional, and behavioral responses to trauma, including child maltreatment (National Child Traumatic Stress Network, 2008). The Center for Disease Control and Prevention’s Whole School, Whole Community, Whole Child Model provides a framework for centering children’s health, safety and well-being while collaborating across school, health, and other community service sectors (Centers for Disease Control and Prevention, 2021). By centering children’s unique needs and collaborating across service sectors, providers can better tailor services to promote resilience processes and positive development.

4.1. Limitations

This analysis should be interpreted in light of several key limitations. First, the confounders included in this analysis were selected based on a review of previous literature, but there is the potential for bias due to unmeasured confounders. Additionally, we did not include time-varying measures of maternal depressive symptoms or family income-to-needs ratios. Future research is needed to examine how changes in these factors over time influence developmental trajectories. This study did not include measures of other developmental domains that may be influenced by resilience processes, such as cognitive or physical health. Furthermore, this study relied on measures of psychopathology, rather than positive development or flourishing, to assess developmental outcomes.

The analytic sample was drawn from the LONGSCAN study. This is a high-risk sample by design (Runyan et al., 1998) and may not sufficiently distinguish vulnerabilities related to maltreatment from other early childhood stressors. Previous research found that reported maltreatment in the first year of life was a significant predictor of behavioral outcomes at age 4 years, but that a cumulative risk score that included maltreatment as one of many early potential risk factors was a stronger predictor of child outcomes (MacKenzie, Kotch, Lee, Augsberger, & Hutto, 2011). Given the variation in risk factors that determined eligibility across study sites, the differences in externalizing behaviors, socialization skills, and daily living skills that we observe across sites may be artifacts of other risk factors in the samples that we did not account for in this analysis.

Another limitation of this study is the use of CPS referral to determine the exposure and comparison groups. Any child who was referred to CPS, regardless of substantiation status, was included in the CPS referral exposure group, as previous literature has demonstrated that substantiation status does not significantly predict developmental outcomes among children referred to CPS (Hussey et al., 2005). By including all children who were ever referred to CPS before age 6, it is possible that this group includes children who did not experience maltreatment. However, referral to CPS is a conservative measure of maltreatment, as some children who experience maltreatment are never referred to CPS (Sedlak et al., 2010). Therefore, it is also possible that children who were included in the non-CPS referral group had in fact experienced maltreatment, which would bias our results towards the null.

Finally, our analyses were constrained to a linear functional form because we only used three time points, but it is possible that a different functional form may better fit the data if more time points were included. Indeed, previous studies on developmental trajectories among children who have experienced maltreatment have found that quadratic models produce better model fits than linear models (Godinet, Li, & Berg, 2014; Thompson & Tabone, 2010). Future studies examining developmental trajectories over more than three time points would allow for more complex functional forms in trajectory models.

4.2. Conclusion

An examination of the association between CPS referral before age 6 years and developmental trajectories through age 12 years in a high-risk sample of children indicates that children who are referred to CPS may be at greater risk for having poor daily living skills and greater externalizing behaviors at age 6 years and increased internalizing behaviors over time. The heterogeneity in patterns of development across domains supports examining developmental domains individually as outcomes of resilience processes, rather than a composite resilience construct, to better model the nuance of distinct developmental trajectories. A more nuanced understanding of domain-specific trajectories can also better inform the design and implementation of interventions to improve developmental outcomes.