An Integrative Data Analysis of Main and Moderated Crossover Effects of Parent-Mediated Interventions on Depression and Anxiety Symptoms in Youth in Foster Care

Abstract

Without preventative intervention, youth with a history of foster care (FC) involvement have a high likelihood of developing depression and anxiety (DA) symptoms. The current study used integrative data analysis to harmonize data across four foster and kinship parent-mediated interventions (and seven randomized control trials) designed to reduce youth externalizing and other problem behaviors to determine if, and for how long, these interventions may have crossover effects on youth DA symptoms. Moderation of intervention effects by youth biological sex, developmental period, number of prior placements, and race/ethnicity was also examined. Youth (N = 1,891; 59% female; ages 4 to 18 years) behaviors were assessed via the Child Behavior Checklist, Parent Daily Report, and Eyberg Child Behavior Inventory at baseline, the end of the interventions (4–6 months post baseline), and two follow-up assessments (9–12 months and 18–24 months post baseline), yielding 4,830 total youth-by-time assessments. The interventions were effective at reducing DA symptoms at the end of the interventions; however, effects were only sustained for one program at the follow-up assessments. No moderation effects were found. The current study indicates that parent-mediated interventions implemented during childhood or adolescence aimed at reducing externalizing and other problem behaviors had crossover effects on youth DA symptoms at the end of the interventions. Such intervention effects were sustained 12 and 24 months later only for the most at-risk youth involved in the most intensive intervention.

Youth with a history of foster care (FC) involvement have a higher prevalence of depression and anxiety (DA) symptoms and disorders relative to their peers without such involvement (Moussavi et al., 2022; Oswald et al., 2010), ranging from 15% to 45% across studies (Oswald et al., 2010). Notably, elevated rates of internalizing symptoms appear in childhood (Minnis et al., 2006). If left untreated, such symptoms can lead to diagnoses of mental health disorders, with accompanying suffering and lost opportunities for these individuals, as well as billions of dollars in health, suicide-related, and workplace costs (e.g., Greenberg et al., 2021). Thus, developing effective preventive interventions for reducing DA symptoms in youth in FC is of paramount importance. However, few such interventions are targeted specifically for these youth and their unique circumstances, such as their high mobility and experiences of multiple early adversities (Kerns et al., 2014; Mersky et al., 2020).

There is increasing interest in examining whether interventions designed to prevent one type of symptom or behavior might also be efficacious in preventing other behaviors not originally targeted by the programs, particularly for populations for which few preventive interventions exist. Showing that an already-extant, evidence-based intervention can have such crossover effects could allow for a greater number of individuals to be served, potentially earlier in their development (Reider et al., 2014). Investigating crossover effects can also help to identify common targets to prevent poor outcomes and promote resilience (Durlak, 1998). Exploring whether interventions shown to prevent other behaviors in youth in FC might also impact internalizing symptoms is thus a potentially cost-effective and beneficial way of identifying efficacious programs to prevent symptoms of DA.

Parent-Mediated Interventions for Youth in Foster Care

Over the past several decades, researchers at the Oregon Social Learning Center and their colleagues have developed several interventions to promote positive outcomes for children and youth in FC, including the Kids In Transition to School (KITS) Program, the Keeping Foster and Kinship Parents Supported and Trained (KEEP) Intervention, the Middle School Success (MSS) Program, and the Treatment Foster Care Oregon (TFCO) Model. Although these programs target different behaviors, they are all based on social learning coercion theory, which posits that poor parenting skills (including inconsistent and harsh parenting) and youths’ dysregulated behavior interact to reinforce and escalate the negativity in each (Patterson et al., 1992), eventually leading to a range of poor outcomes, including internalizing symptoms and delinquency (Wiesner & Kim, 2006). Additionally, the interventions all have a dual focus on youths and their caregivers and are delivered, at least in part, in a group format. The programs differ in the age of the youth targeted (from childhood to adolescence), intervention length (from approximately 4 to 6 months), and the specific components offered (e.g., individual skill-building sessions in addition to group sessions). KITS, KEEP, and MSS are selective interventions, targeting youth in FC to prevent problematic outcomes for which they are at high risk, including low academic achievement, aggressive and delinquent behavior, and poor social interactions (Mersky et al., 2020; Oswald et al., 2010). TFCO is an indicated intervention focused on youth in FC who are already exhibiting severe emotional and behavioral disorders. The interventions have been described in detail in prior publications (Chamberlain, 2003; Chamberlain et al., 2006; Pears et al., 2012; Price et al., 2009); therefore, protocols and findings are summarized briefly here.

The KITS Program (Pears et al., 2013) is a 4-month long intervention targeted for 4–5-year olds in FC who are transitioning to kindergarten. The intervention prepares the children for school via 24 two-hour long school readiness group sessions for the children and 8 two-hour caregiver group sessions. The program is designed to increase academic achievement, self-regulation, and prosocial skills in the children—as well as foster and kinship caregiver involvement in learning at home and school and positive parenting skills. Outcomes include improvements in children’s self-regulation and literacy skills (Pears et al., 2013; Pears et al., 2012) and reductions in behaviors that are precursors to delinquency and substance use (Pears et al., 2016). The KEEP Interventions (KEEP, Chamberlain et al., 2008; KEEP-Reaching, Price et al., 2015; and KEEP-Safe, Kim et al., 2017) target foster parents of youth aged 5 to 18 years in FC. KEEP is intended to prevent the development of disruptive behaviors in the youth, as well as to strengthen foster and kinship caregivers’ positive parenting skills and placement stability. Ninety-minute weekly KEEP groups are delivered by two group leaders over 4 months using a group-based learning approach. KEEP outcomes include increased placement stability and fewer dysregulated and substance use behaviors for youth, and increased use of positive reinforcement and reduced stress for caregivers (Kim et al., 2017; Price et al., 2009; Price et al., 2012; Price et al., 2015). The MSS Program (MSS-LINKS and MSS-MIDG, Kim & Leve, 2011) is targeted for youth in FC transitioning to middle school and is intended to increase prosocial skills and to decrease delinquency, substance use, and health-risking sexual behaviors. (Note that MSS later informed and was modified into the KEEP-Safe intervention.) MSS includes six parent groups and parallel youth groups prior to the transition to middle school, followed by weekly meetings for the caregivers and skills-coaching sessions for the youth during the first year of middle school. Outcomes include decreases in females’ internalizing, externalizing, and health-risking sexual behaviors and substance use (Kim & Leve, 2011; Kim et al. 2013; Smith et al., 2011). The TFCO Model (TFCO-GIRLS, Chamberlain et al., 2007) is a 6–9-month intervention that targets youth with severe emotional and behavioral disorders. It is designed to decrease antisocial and delinquent behaviors in youth and increase caregiver positive and structured parenting skills. Foster parents attend weekly meetings and receive 24/7 on-call support from a Team Leader, while biological parents receive weekly parenting-skills training. Youth receive close supervision, weekly individual therapy and behavioral skills training, academic support, and psychiatric consultation as needed. The program reduces days in a locked setting (e.g., jail), depressive and psychotic symptoms, deviant peer associations, drug use, and pregnancies, while increasing school attendance and homework time (Chamberlain et al., 2007; Kerr et al., 2009; Leve & Chamberlain, 2005).

Notably, none of these interventions specifically targeted DA symptoms. However, studies have shown that negative, coercive parenting leads to increased risk for DA symptoms in youth across developmental periods, whereas the positive, consistent parenting promoted by the four interventions appears to buffer youth against these symptoms (Hentges et al., 2021; Yap et al., 2014). Further, previous research has shown that the TFCO Model decelerates trajectories of depressive symptoms in young women involved in FC and the juvenile justice system (Harold et al., 2013) and that the MSS intervention decreased internalizing symptoms in girls in FC as they started middle school (Smith et al., 2011). Given that these interventions have been specifically designed for youth involved in FC and directly address factors involved in the development of internalizing symptoms, exploration of potential crossover effects on internalizing symptoms and whether there are subgroups for which any crossover effects are more salient is warranted. This would help determine if these interventions might be efficacious in reducing the high levels of these symptoms in this very vulnerable population.

Using Integrative Data Analysis (IDA) to Examine Potential Crossover Effects

Significant barriers to exploring potential crossover effects across multiple interventions include the likelihoods that prior efficacy studies utilized different designs, methodologies, and samples. Fortunately, the IDA framework for aggregating data across studies takes advantage of the heterogeneity in study designs, measures, and sample characteristics that could otherwise limit the applicability and generalizability of individual randomized control trials (RCTs), and allows a more powerful and solidified test of the aggregated intervention effect for similar types of interventions. Further, the combined dataset often represents a more ethnically and racially diverse sample, thus facilitating the generalizability of combined results across populations. Increased diversity and heterogeneity across studies also allows for examination of moderated intervention effects by subgroups that are often considerably underpowered in individual RCTs, particularly for categorical moderators such as biological sex (Brown et al., 2013).

In the current study, the potential moderating effects of biological sex, developmental period, number of FC placements, and race/ethnicity were examined. Studies have indicated that when depressed or anxious, biological females might be more likely than biological males to use and respond positively to social support and emotion regulation as coping strategies (Malooly et al., 2017; Olivier et al., 2022). The interventions examined in this paper promote both supportive parenting skills and youths’ self-regulation strategies, leading us to hypothesize that females might show greater decreases in DA symptoms. In the general population, both the onset and severity of symptoms of DA have often been linked to the developmental period of adolescence (Betts et al., 2016). This effect has been less consistent and less studied in youth in FC (Huguenel et al., 2021), making it an important consideration in studies of internalizing in these youth. Experiencing more placement instability is associated with increased internalizing symptomatology as well as greater symptom chronicity (Hiller & St. Clair, 2018; McGuire et al., 2018). Additionally, such instability appears to have deleterious consequences for youths’ regulatory abilities (Bourne et al., 2022). Thus, interventions with a specific focus on regulation might differentially benefit youth with more placement transitions. Finally, studies have shown that youth of color in FC are less likely to receive services for mental health symptoms, including internalizing symptomatology, even when the severity of their symptoms does not differ from those of White children (Garland et al., 2000). Thus, the inclusion of race/ethnicity as a moderator in intervention studies for this population is particularly warranted.

Research Questions

The current study examined if, when, and for whom preventative interventions for youth in FC implemented during childhood and adolescence may have had positive crossover effects on DA symptoms. Both main and moderated intervention effects were evaluated at the end of the interventions (4–6 months post baseline [BL]), and at two follow-up assessments (9–12 months [F1] and 18–24 months [F2] post BL). Specifically, the interventions were posited to reduce DA symptoms for all youth but be most efficacious (a) for biological females versus males, (b) in early to late adolescence compared to early to late childhood, and (c) for youth who have experienced more FC placements. Potential moderation of intervention effects by youth race/ethnicity were considered to be exploratory due to the lack of prior research on this important potential moderator.

Method

Participants

Sample and demographics.

Data for the current study (N = 1,891 youth in FC in Oregon or California) came from seven RCTs of the four interventions described above (see Table 1 for RCT details). Excluded participants included 3 children from the KITS study, and 1 each from the KEEP, KEEP-Safe, and MSS-MIDG studies due to missing data, and 28 youths from the KEEP-Reaching study because they were not in FC. Youth were an average of 10 years of age at BL and 59% were biological females. They had experienced an average of three prior placements; 34% were in kinship care. The sample identified as 39.6% Hispanic/Latino, 34.5% White, 19.6% Black, 11.3% multiracial, 1.6% Native American, 1.6% Asian/Pacific Islander, 0.1% Middle Eastern, and for 1.7% of the participants race/ethnicity was unknown. Note that the race/ethnicity categories were not mutually exclusive.

Table 1.

Study Participants, Assessment Schedules, and Sample Size by Assessment

		Assessment schedule (months post BL)
Study	Youth participants (n = intervention/control)	BL	P	F1	F2	N current study	BL	P	F1	F2
KITS	192 children ages 4–5 years entering kindergarten in fall (n = 102/90)	0	4	12	24	189	184	163	152	131
KEEP	700 youth ages 5–12 years in their FC placement at least 30 days (n = 359/341)	0	5	10	NA	699	697	471	327	NA
KEEP-Reaching	335 youth ages 5–12 years in their FC placement at least 30 days (n = 164/171)	0	6	12	18	307*	304	224	144	116
KEEP- Safe	259 adolescents ages 11–17 years in their FC placement at least 30 days (n = 59, 58/142)	0	6	12	18	258	255	200	180	150
MSS-LINKS	173 youth ages 10–12 years transitioning to middle school (n = 87/186)	0	NA	9	18	173	172	NA	152	140
MSS-MIDG	100 females ages 10–12 years transitioning to middle school (n = 48/52)	0	6	12	24	99	95	**	93	89
TFCO-GIRLS	166 adolescent females ages 12–17 years in out-of-home care due to delinquency (n = 81/85)	0	6	12	24	166	161	**	140	90
Total						1891	1868	1058	1188	716

Note.

^*n = 28 KEEP-Reaching youth were not in foster care and thus excluded;

^**CBCL not administered at post-intervention assessment; NA = No assessment; FC = foster care; BL = baseline; P = post intervention; F1 = first follow-up assessment; F2 = second follow-up assessment.

Control conditions.

For all studies except TFCO-GIRLS, the control condition was foster care as usual, which could include some training for the foster caregivers and individual or group therapy services for the youths. In TFCO-GIRLS, adolescent females were court mandated to out-of-home care due to chronic delinquency and were randomly assigned to either receive TFCO or be placed into another out-of-home care, community-based program (typically group-care residential facilities).

Measures

Youth demographic and moderating variables.

Child welfare records or caregiver reports were used to determine youth age and biological sex, the number of FC placements that they had experienced prior to the study, and whether the foster caregiver was kin or nonkin. To test for moderation effects by developmental period, age at BL was categorized into: childhood (ages 4–10 years), early adolescence (ages 11–13 years), or mid-to-late adolescence (ages 14–18 years). Youth race/ethnicity was reported by either the youth or the caregivers, for the younger children. There were a number of differences across the RCTs in how race and ethnicity were assessed, including whether they were measured separately and how many categories participants could select. For the current study, participants were coded as “1” for each racial and ethnic group they selected. Participants who indicated more than one race were coded as “1” for each category as well as “1” for the multiracial category. Moderated intervention effects by race/ethnicity were separately examined for each of four categories (vs. all others): Black/African American, Hispanic/Latino, multiracial, and White. Unfortunately, small sample size for other racial categories precluded examination of their effects.

Depression and anxiety symptoms.

Symptoms of DA were measured using one of three caregiver-reported standardized measures. The Child Behavior Checklist (CBCL; Achenbach, 1991; Achenbach & Rescorla, 2001) was utilized in all RCTs except KEEP, which utilized the Eyberg Child Behavior Inventory (ECBI; Eyberg & Ross, 1978) and Parent Daily Report (PDR; Chamberlain & Reid, 1987). On both the CBCL and ECBI, caregivers reported the frequency with which the youth displayed symptoms over the last 6 months or currently, respectively. Response scales included never = 0, sometimes = 1, and often or almost always = 2 for the CBCL; and never = 0, seldom = 1, sometimes = 2, often = 3, and always = 4 for the ECBI. The PDR (Chamberlain & Reid, 1987), which assesses child internalizing and externalizing behaviors, was administered 3 times over a week at each timepoint. Caregivers were asked to indicate on a binary scale (yes = 1/no = 0) whether any of 31 behaviors had occurred in the last 24 hours. For the current study, the maximum response over the three PDR assessments at each of the timepoints was used for each item.

Item selection and common item harmonization.

DA symptom items were selected a priori and chosen to be mutually exclusive with other aspects of internalizing measured by the scales (i.e., posttraumatic stress disorder symptoms and suicidality). The 13 CBCL items used in the current study for all studies except KEEP included symptoms related to depression (i.e., crying, feeling sad, lonely, guilty, unloved, worthless, and having low energy) and anxiety (i.e., worrying, feeling nervous, anxious, self-conscious, afraid of doing wrong, and perfectionism). The four common harmonized KEEP items included an item related to crying from the ECBI, and nervous or jittery, sluggish, and sad or depressed from the PDR. All response scales were rescaled to indicate the presence or absence of the behavior (i.e., no/never = 0, yes/and all other positive responses = 1). The exception was for the common harmonized item of cries, which was cries easily on the ECBI and cries a lot on the CBCL. ECBI responses of often and always cries easily were mapped on to CBCL responses of sometimes or often or almost always cries a lot (yes = 1). ECBI responses of never, or seldom, or sometimes cries easily were mapped onto the CBCL of not crying a lot (no = 0). All available items at all timepoints were used, except for the low energy item for the KITS sample, as it was not developmentally appropriate (children were ages 4–6 years at BL).

Data Analysis Plan

Data harmonization procedure.

Measurement invariance across study- and person-level covariates (including youth age, biological sex, study membership, intervention condition, and assessment wave) and a subset of two-way interactions was examined using moderated nonlinear factor analysis (Curran et al., 2014). Measurement invariance was evaluated in three steps using a calibration sample of N = 1,891 independent observations by randomly selecting one assessment for individuals with more than one data point. First, impact (IMP) on the factor mean and variance for the set of covariates was examined. Second, all significant predictors of the factor mean and variance were retained in the second set of models, which examined potential differential item functioning (DIF) in the item thresholds and factor loadings. Third, a final model was estimated that simultaneously accounted for all significant IMP and DIF covariate effects; all unattenuated significant predictors were retained in the final scoring model. Further details regarding the subset of two-way interactions tested, the model-building procedure for retaining significant IMP and DIF effects, and adjustments to the Type I error rate for multiple tests using the Benjamini and Hochberg (1995) procedure are given in the Online Resource.

The final step of the data harmonization procedure was to estimate factor scores for all participants at all assessments based on the final scoring model (see Online Resource Table 2), which models invariance rather than dropping items that display invariance, using modal a posteriori (MAP) estimation (Bock & Aitken, 1981). These estimated scores are based on which (rather than simply how many) items were endorsed, while accounting for measurement invariance attributable to study membership and the other covariates.

Hypothesis tests.

Main and moderated aggregated intervention effects were estimated while accounting for differences in DA symptoms attributable to study membership and person-level covariates using a combination of latent class analysis (with known class memberships defined by studies) and mixed-effects modeling (to account for repeated measures across time). Thus, models accounted for fixed and random effects at the study and individual level (i.e., age, age squared, biological sex, intervention condition, and kinship placements; see the Online Resource for example Mplus syntax). In addition, as outlined by Brown and colleagues (2013), as part of our model-building procedure we examined for heterogeneity in efficacy across trials by testing whether the strength of the intervention effects at BL versus P, F1, or F2 significantly varied across individual studies and type of intervention (i.e., KITS, KEEP [3 RCTs], MSS [2 RCTs], and TFCO). In sum, models controlled for the effects of study membership on the mean levels of, and variability in, DA symptoms using known class memberships, as well as mixed effects modeling, to account for repeated measures while controlling for age, nonlinear change in age (i.e., age squared), biological sex, kin versus nonkin placements, and potential differences in the saliency of intervention effects across studies. Finally, main intervention effects were tested by evaluating the significance of the group (intervention vs. control) by time (BL vs. P, F1, or F2) interaction; for moderation, the group by time by moderator three-way interaction was assessed.

Results

Psychometric Scoring Model Results

The significant covariate IMP and DIF effects used in the final scoring procedure are given in the Online Resource Table 2. In the final combined IMP model with multiple covariate effects, differential IMP on the average levels of the factor scores were observed across studies, biological sex, and assessment waves. Regarding DIF, results indicated that 9 of the 13 items had no significant DIF across the set of covariate effects examined. Importantly, this included one of the common harmonized KEEP items, sad or depressed, which is a face valid item that served as an anchor item across studies. This item also had the highest estimated item association with the DA symptoms factor scores in the final scoring model (r = .74, p < .001). Four items (cries, afraid of doing bad, low energy, and nervous) showed significant DIF across 10 covariate effects—all but one of which were on the thresholds. Many of the DIF effects may be explained by a change in measurement across assessments (i.e., the ECBI item cries was not administered to KEEP participants at F1), developmental differences (i.e., crying was endorsed for younger children at lower levels of DA symptoms than older children), or other methodological or sampling differences across studies. The mean estimated DA factor scores for each study by wave and intervention condition are depicted in Figure 1.

Figure 1.

Mean Estimated Depression and Anxiety Symptoms Factor Scores by Study, Intervention Condition, and Assessment

Results for Hypothesis Tests

BL equivalence.

For the combined sample, there were no significant mean differences on DA symptoms, age, biological sex, race/ethnicity, developmental period, number of prior placements, kinship status between the intervention, and control conditions at BL.

Model building.

The estimated means and variances for DA symptoms for each study are given in Table 2, with the results indicating that all effects were significant, with the exception of the DA symptoms means for the MSS-LINKS and MSS-MIDG studies. In addition, intervention efficacy significantly varied across studies, with intervention effects for TFCO-GIRLS at both follow-up assessments found to be significantly different from the other RCTs (Wald₍₁₎ = 4.342, p = .0372 for F1 and Wald₍₁₎ = 6.532, p = .011 for F2).

Table 2.

Main Intervention Effects Controlling for Study Membership and Person-level Covariates

Effect	BL vs. P	BL vs. F1	BL vs. F2
Aggregated hypothesis tests:
Tx (intervention vs. control)	−0.013 (0.017)	0.007 (0.017)	0.005 (0.018)
Time (BL vs. P, F1, or F2)	−0.091 (0.013)***	−0.088 (0.013)***	−0.126 (0.016)***
Tx X Time	−0.026 (0.013)*	0.003 (0.012)	−0.004 (0.015)
Person-level covariates:
Age	−0.049 (0.010)***	−0.035 (0.010)***	−0.019 (0.010)
Age2	−0.005 (0.001)***	−0.005 (0.001)**	−0.004 (0.001)*
Male (vs. female)	−0.156 (0.034)***	−0.157 (0.034)***	−0.155 (0.034)***
Kin (vs. nonkin) placement	−0.034 (0.017)	−0.033 (0.017)	−0.033 (0.017)
Estimated study means:
KITS	−0.528 (0.072)***	−0.472 (0.072)***	−0.423 (0.074)***
KEEP	−0.365 (0.036)***	−0.343 (0.036)***	−0.348 (0.036)***
KEEP-Reaching	−0.387 (0.059)***	−0.367 (0.059)***	−0.345 (0.059)***
KEEP-Safe	−0.278 (0.063)***	−0.327 (0.063)***	−0.402 (0.066)***
MSS-LINKS	−0.066 (0.061)	−0.047 (0.060)	−0.068 (0.060)
MSS-MIDG	−0.039 (0.082)	−0.014 (0.081)	−0.026 (0.081)
TFCO-GIRLS	0.910 (0.080)***	0.867 (0.082)***	0.749 (0.088)***
Estimated study variances:
KITS	0.377 (0.032)***	0.373 (0.032)***	0.380 (0.033)***
KEEP	0.283 (0.013)***	0.282 (0.013)***	0.282 (0.013)***
KEEP-Reaching	0.257 (0.021)***	0.267 (0.021)***	0.261 (0.021)***
KEEP-Safe	0.390 (0.029)***	0.394 (0.029)***	0.389 (0.028)***
MSS-LINKS	0.364 (0.034)***	0.362 (0.034)***	0.362 (0.033)***
MSS-MIDG	0.360 (0.048)***	0.373 (0.049)***	0.357 (0.047)***
TFCO-GIRLS	0.511 (0.044)***	0.474 (0.042)***	0.492 (0.042)***

Note. Tx = intervention versus control condition; BL = baseline; P = post intervention; F1 = first follow-up; F2 = second follow-up.

^***p < .001.

^**p < .01.

^*p < .05.

Main and moderated intervention effects.

Results for the aggregated main intervention effects are depicted in Table 2. The dual-focused caregiver–youth interventions significantly reduced DA symptoms in the period immediately after the interventions. However, this effect was not sustained at the follow-up assessments 9–12 and 18–24 months later, with the exception that the TCFO program significantly reduced adolescent girls’ DA symptoms 12 and 24 months after BL (b[SE] = −0.091 [0.043], p = 0.036 for F1; b[SE] = −0.128 [0.046], p = 0.006).

For the covariates, as expected, there were a number of significant effects. First, males had significantly lower DA symptoms relative to females. Second, significant nonlinear change in DA symptoms was observed in youth as they matured for all assessments, whereas linear change in DA symptoms was significant for only the BL versus P and F1 assessments. Third, KITS, KEEP, KEEP-Reaching, and KEEP-Safe youth had significantly lower, and TFCO-GIRLS had significantly higher, DA symptoms scores. In contrast, kinship status was nonsignificant.

Moderated intervention results (shown in Table 3) did not support the second set of hypotheses that the interventions would be more efficacious for females versus males—in early to late adolescence versus childhood—and for youth with more prior placements at any assessment timepoint. Likewise, race/ethnicity was not a significant moderator at any timepoint.

Table 3.

Moderated Intervention Effects by Youth Biological Sex, Number of Prior Placements, Developmental Period, and Race/Ethnicity

Moderator	BL vs. P	BL vs. F1	BL vs. F2
Biological sex (male vs. female)	−0.015 (0.022)	0.010 (0.024)	0.007 (0.029)
Number of prior placements	−0.003 (0.004)	0.002 (0.004)	0.002 (0.006)
Developmental period:
Early adolescence vs. childhood	−0.011 (0.013)	−0.003 (0.013)	−0.006 (0.015)
Mid-to-late adolescence vs. childhood	−0.022 (0.017)	−0.008(0.019)	0.014 (0.021)
Race/ethnicity:
Black/African American	0.004 (0.014)	0.010 (0.015)	0.019 (0.018)
Hispanic/Latino	−0.001 (0.011)	−0.010 (0.012)	−0.003 (0.015)
Multiracial	0.010 (0.016)	0.025 (0.020)	0.000 (0.021)
White	−0.016 (0.012)	−0.006 (0.012)	−0.013 (0.015)

Note. Tabled numbers denote b(SE) for the three-way interaction involving each moderator with intervention condition and time. Models also controlled for study membership, biological sex, age, age squared, and kinship placement status.

Discussion

Youth with a history of FC involvement are at heightened risk of developing symptoms of DA without preventative intervention (Moussavi et al., 2022; Oswald et al., 2010). However, there are very few interventions designed to prevent DA in these high-risk youth (Kerns et al., 2014; Mersky et al., 2020). Using an IDA framework to harmonize data across seven RCTs of programs created specifically to improve outcomes other than DA for youth in FC, the current study showed that parent-mediated, dual-focused caregiver–youth interventions reduced DA symptoms in youth at the end of the interventions. This is a hopeful finding, as it shows that interventions designed to prevent externalizing behaviors in youth in FC can also decrease DA symptoms, at least in the short term. The finding that long-term effects were sustained for adolescent females with a history of criminal justice involvement also points to future modifications to potentially increase the durability of effects on DA symptoms in particular. Further, this study highlights the importance of interventions focused on supporting foster caregivers to use positive, noncoercive parenting behaviors in helping promote a number of positive outcomes for youth. Other studies utilizing multiple RCTs of interventions targeted at improving parenting behaviors have demonstrated that such programming can have positive crossover effects for youth from high-risk backgrounds (e.g., Connell et al., 2021).

Although all of the interventions had immediate positive effects on DA symptoms, such effects were not sustained at the intermediate and longer-term follow-up assessments (9–12 months and 18–24 months post BL), with the exception of the TFCO program. This catch-up effect for youth in the foster care as usual condition is in contrast to findings from a number of the RCTs of the interventions showing longer-term effects on disruptive, aggressive, and delinquent behaviors (e.g., Kim & Leve, 2011; Pears et al., 2012). There are at least two differences between TFCO and the other interventions that may help to explain why reductions in DA symptoms specifically were sustained over time for TFCO versus the other interventions, and provide clues about how to sustain effects in future programming. First, the TFCO program is more intensive than the other interventions, providing services and support on a round-the-clock basis to the foster parent; weekly individual therapy and skills training, academic support, and psychiatric consultation to the youth; and training and support for the biological parents if the youth was returning to their care. These services also lasted longer than most of the other interventions. This suggests that crossover effects of interventions may be more likely to be sustained if interventions feature intensive and longer lasting services for both parents and youth.

Second, the TFCO study involved adolescent females who were mandated to out-of-home care due to criminal justice system involvement. Thus, they may have been at the highest risk for experiencing DA symptoms and other mental health problems by virtue of being female, in mid-to-late adolescence, and having dual system involvement (although moderated intervention results for biological sex and developmental period were not significant). This could indicate that crossover effects may be most likely to manifest for youth with multiple vulnerabilities. Taken together, the differences in intervention intensity and sample risk suggest that youth at highest risk for DA symptoms, in this case adolescent females involved in the criminal justice system, might benefit from extended intervention services, or multiple interventions, to sustain long-term positive effects. Future research is needed to determine if this might be the case for other groups of youth with multiple risk factors, such as adolescent boys dually involved in both the FC and criminal justice systems. Such research could also examine how different profiles of risk or trajectories of DA symptoms might affect the long-term sustainability of intervention effects.

Results indicated that the strength of the crossover intervention effect was not moderated by biological sex, developmental period, the number of FC placements the youth had experienced, or race/ethnicity. This may indicate that the interventions generalize to a diverse population of youth in FC who are at risk for developing DA symptoms. It should be noted that because of the small size of some racial groups (e.g., Native American, Pacific Islander, Asian), it was not possible to examine whether there were moderated intervention effects specifically for these youth. Future research should focus on these groups.

Potential Limitations

There were some study limitations. First, only caregiver report of youth DA symptoms was used because only a subset of the RCTs collected youth-reported mental health measures. Although less research on the concordance between caregiver and youth reports of psychological symptoms has been conducted with youth in FC than with their non-welfare-involved peers, evidence suggests that caregivers and youth in FC are more likely to agree on externalizing than internalizing symptoms and that caregivers tend to report higher levels of symptoms across problem domains (Makol et al., 2021; McWey et al., 2018). Further, discrepancies in caregiver–youth reports can predict increases or decreases in symptoms over time (McWey et al., 2015). Future studies should include multi-informant data and examine the potential prediction of intervention outcomes from discrepancies in reports. Second, there were differences in the response time frames across the items. Thus, we cannot determine whether DA symptoms were observed in the youth currently or in the past (i.e., from 1–3 weeks to 6 months). Third, the interventions all targeted caregiver parenting skills that are likely to improve with time and practice. Thus, a group by time design (instead of latent growth curve modeling) was used to evaluate whether the interventions themselves had delayed impacts on youth DA symptoms, which does not account for changes in DA symptoms across time. Finally, although item functioning and the partial information curves were found to be adequate for the four common harmonized KEEP items, less information about the manifestation of DA symptoms was available for these youth, who were the largest sample (N = 699, 37% of 1,891 total); and no second follow-up assessment was collected.

In spite of these limitations, the current study had a number of strengths, including the use of IDA to harmonize data across seven RCTs and four interventions for vulnerable, high-risk samples of youth. Further, the patterns of differences in DA symptoms factor scores across the studies and covariates were as anticipated (e.g., younger children and males showing fewer symptoms than adolescents and females). This provides further evidence of the rigor and reproducibility of the IDA scoring procedure and results. Tests of the intervention effects were also rigorous, including controls for study membership, age, biological sex, kin versus nonkin placement, and differences in the saliency of intervention effects across interventions.

The current study indicates that interventions targeting both caregiver and youth behaviors that were implemented during childhood and adolescence, and originally aimed to reduce youth externalizing and other problem behaviors, had positive short-term crossover effects on DA symptoms for all youth and long-term effects for adolescent females in particular. The current study contributes to the growing literature examining crossover effects of interventions and, consistent with those studies (e.g., Connell et al., 2020; Reider et al., 2013), demonstrates that this may be a cost-effective and efficient way not only of expanding the populations who might be served by extant interventions but also pinpointing targets for future, wide-reaching intervention efforts. This is particularly important for populations at high risk for a variety of poor psychosocial outcomes, such as children in FC. Results of this study add to the knowledge base about ways to effectively promote resiliency in these vulnerable youth.