Cortisol and Parenting Predict Pathways to Disinhibited Social Engagement and Social Functioning in Previously Institutionalized Children

Abstract

Previously institutionalized children on average show persistent deficits in physiological and behavioral regulation, as well as a lack of normative reticence towards strangers, or disinhibited social engagement (DSE). Post-adoption parenting, specifically a combination of supportive presence and structure/limit-setting, may protect against DSE over time via better adrenocortical functioning. This study examined the impact of adrenocortical activity and post-adoption parenting on DSE across the first two years post-adoption (age at adoption: 16-36 months) and observed kindergarten social outcomes in previously institutionalized children (n = 94) compared to non-adopted children (n = 52). Path analyses indicated a developmental cascade from institutional care (operationalized as a dichotomous group variable, age at adoption, and months of institutionalization) to blunted adrenocortical activity, increased DSE, and lower kindergarten social competence. Consistent with a permissive parenting style, higher parental support was associated with increased DSE, but only when not accompanied by effective structure/limit-setting. Further, parental structure reduced the association between blunted adrenocortical activity and DSE behaviors.

Children raised in institutional care in infancy and toddlerhood experience inconsistent caregiving, a lack of social stimulation, and in some cases malnutrition (Gunnar, Bruce, & Grotevant, 2000) which is a severe disruption of the typical caregiving environment, resulting in substantial deficits in physiological regulation, cognitive ability, attachment relationships, and social behavior (for a review, see Juffer et al., 2011). Some of these children are eventually adopted into well-resourced families and exhibit considerable recovery in cognitive and social domains (Gunnar et al., 2000); however, many deficits persist or even increase in the years post-adoption. For example, a small but notable group of previously institutionalized (PI) children tend to exhibit a lack of developmentally normative reticence towards adult strangers, otherwise known as disinhibited social engagement (DSE), and the difference between PI children and their never-institutionalized peers in this behavior tends to increase over the first few years post-adoption (Lawler, Koss, Doyle, & Gunnar, 2016).

DSE was originally theorized to be a direct symptom of disrupted caregiving relationships during and following adverse early care (Zeanah & Gleason, 2015). However, this cannot explain why, despite improvements in the attachment relationship over time in the adoptive family (Carlson, Hostinar, Mliner, & Gunnar, 2014), some PI children show consistently higher rates of DSE, especially inappropriate physical contact, several years post-adoption (Lawler et al., 2014; 2016; Miron & Zeanah, 2017). DSE also seems to be driven by a general deficit in inhibitory control or self-regulation more broadly that persists over time in PI children (Bruce, Tarullo, & Gunnar, 2009; Gorter, Helder, Oh, & Gunnoe, 2017; Lehmann et al., 2016). PI children continue to show regulatory deficits compared to typically developing children, namely blunted hypothalamic-pituitary-adrenocortical (HPA) axis reactivity and diurnal regulation and more behavioral problems (Bruce et al., 2009; Koss et al., 2016; McLaughlin et al., 2015). HPA axis functioning is associated both with the quality of the caregiving relationship (Bernard, Dozier, Bick, & Gordon, 2015) and attention and behavior regulation (Isaksson et al., 2012; Koss et al., 2016; Pitula et al., 2017). Furthermore, early exposure to institutional care is strongly associated with both dysregulated HPA activity and increased rates of DSE (Koss et al., 2016; Miron & Zeanah, 2017). Thus, HPA activity might play an important part in the etiology of DSE and its consequences over time for children adopted from institutions, yet only one study to our knowledge has examined the association between the two (Johnson, Bruce, Tarullo, & Gunnar, 2011). That study found that a more dysregulated cortisol rhythm, reflected as higher cortisol levels later in the day, was associated with more DSE symptoms. Likewise, the influence of the caregiving environment on the course of DSE symptoms may be through its impact on HPA regulation, though this is yet to be tested.

Many studies have converged on the idea that supportive, responsive caregivers, particularly in infancy and toddlerhood, play a critical role in the development and regulation of the HPA axis (for a review, see Hostinar, Sullivan, & Gunnar, 2014), which in turn might affect DSE and social competence (Pitula et al., 2017). Additionally, parental structure and effective limit-setting are important for the development of self-regulation (Hammond et al., 2012; LeCuyer-Maus, 2000; Lengua, Honorado, & Bush, 2007), which may as a result improve rates of DSE behaviors. However, there has been some conflicting evidence surrounding the association between parenting and levels of DSE, particularly in children who have experienced institutional care (Dobrova-Krol, Bakermans-Kranenburg, van IJzendoorn, & Juffer, 2010; Garvin, Tarullo, Van Ryzin, & Gunnar, 2012; Lyons-Ruth, Bureau, Riley, & Atlas-Corbett, 2009; Rutter et al., 2007; Zeanah, Smyke, Koga, & Carlson, 2005). Timing of the measure of parenting may be an important factor. When children are adopted as toddlers, DSE behaviors tend to increase over the first two years post-adoption (Lawler et al., 2016) at the same time as language and other social capacities are developing, and persist through adolescence (Kreppner et al., 2010; Rutter et al., 2007). Parenting soon after adoption may be the most critical to disrupt the persistence of DSE behaviors and social problems post-adoption. Additionally, different aspects of parenting behavior, for example support and structure, may moderate each other’s associations with DSE. As outlined by Baumrind (1971), a balance of control/structure and similarly high warmth/support (i.e. authoritative vs. authoritarian or permissive parenting) may ultimately provide the best environment to develop self-regulation (LeCuyer-Maus, 2000; Rinaldi & Howe, 2012), though no study thus far has examined the impact of multiple dimensions of parenting on DSE.

As children’s social contexts and relationships outside the family expand, dysregulation tends also to be associated with poorer social outcomes for PI children, such as difficulties forming and maintaining friendships (Almas et al., 2015; Pitula et al., 2017). It is possible that DSE behaviors play a role in the cascading effects of physiological and behavioral dysregulation on later social outcomes. Peers may notice when PI children are showing socially-inappropriate DSE behaviors with adults (e.g., teachers, other parents) and be less willing to interact with them. There is also some evidence that DSE behaviors spill over into interactions with similarly-aged peers (Guyon-Harris et al., 2019). Thus, the present study investigates developmental cascades from early adversity to adrenocortical regulation, observed DSE behaviors and caregiver-reported DSE symptoms, and later observed behavior.

As an extension of previous research on trajectories of DSE with this sample (Lawler et al., 2016), this multi-level, multi-method study examined longitudinal associations between adrenocortical activity, post-adoption parenting, observed and caregiver-reported DSE behaviors, and observed kindergarten social competence. Our first aim was to examine differences between PI and non-adopted children on all key variables: adrenocortical activity, observed DSE behaviors, and kindergarten social competence.¹ We hypothesized that PI children would exhibit worse adrenocortical activity, increased DSE, and lower kindergarten social competence compared to non-adopted children, consistent with previously reported findings on this sample (Lawler et al., 2014; Pitula et al., 2017) and others (Almas et al., 2015; McLaughlin et al., 2015; Miron & Zeanah, 2017). Second, we examined whether there was a developmental cascade that includes adrenocortical functioning, DSE behaviors/symptoms, and later social behavior. Specifically, we expected that exposure to institutional care in infancy would predict blunted (i.e. lower than typical) adrenocortical activity, blunted adrenocortical activity would predict higher levels of DSE behaviors, and more DSE behaviors would predict worse kindergarten social functioning. Third, we investigated whether post-adoption parental support and structure/limit-setting would act as promotive factors predicting better adrenocortical activity and fewer observed DSE behaviors and caregiver-reported symptoms. Finally, we examined whether parental support and structure/limit-setting would moderate the association between early institutional care and DSE via adrenocortical functioning (moderated mediation). Thus, the present study informs a potential developmental cascade following early institutional care that includes risk (adrenocortical activity, early DSE behaviors) and promotive or protective factors (post-adoption parenting) toward and away from DSE symptoms and later social dysfunction.

Methods

Participants

Participants in the previously institutionalized group (PI; n = 94, 57% female) were internationally adopted between 16 and 36 months (M = 25.2 months) and spent on average 75% of their pre-adoption life in institutions (SD = 29%, range = 12.5-100%). PI children were primarily adopted from Southeast Asia (40.4%), as well as Africa (29.8%), Russia (17.0%), Latin America (6.4%), and Haiti (6.4%). Non-adopted (NA) participants were of comparable age but were born and raised in their biological families in the United States (n = 52, 50% female). NA families were also demographically similar to the PI adoptive families in terms of income and education level. Median income range for NA families was $75,000-$100,000 and median income range for PI families was $100,000-$125,000, and 77% of the entire sample reported an annual income above $75,000. In both groups, 83% of primary caregivers held a bachelor’s degree or higher. Chi-squared tests revealed that the PI and NA groups were not statistically significantly different in either annual household income range or primary caregiver education level (income: χ² (4, N = 146) = 8.59, p = .38; education: χ² (4, N = 146) = 6.96, p = .14). Prior to arriving at the final sample of 146 children, 14 were excluded due to concern or diagnosis of Fetal Alcohol Syndrome (10 PI), cognitive or congenital disorder (1 PI), Autism Spectrum Disorder (1 NA), maltreatment (1 NA), or medical trauma (1 NA) as these are believed to impact adrenocortical functioning, social behavior, or both in ways unrelated to the present hypotheses.

Procedure

PI children were recruited primarily through local adoption medical clinics and local adoption agencies. NA children were recruited through a departmental participant pool collected via letters to families that had recently given birth using public county records. Of those families who agreed to participate, informed consent was acquired from a parent at each study session. All study procedures were approved by the University of Minnesota Institutional Review Board.

Children participated in 5 laboratory study sessions each lasting approximately 1.5 hours, along with a kindergarten classroom observational assessment. Sessions 1-4 occurred relative to PI children’s adoption. These sessions took place approximately 2, 8, 16, and 24 months post-adoption for PI children, respectively. NA sessions were similarly conducted every 8 months but were roughly age-matched to PI sessions. At session 1, children were on average 26.9 months old (SD = 5.3; see Table 1 for demographic information by group). Diurnal and laboratory salivary cortisol, observed parenting behaviors, and observed DSE behaviors were collected at each of the four sessions. Sessions 1 and 4, but not 2 and 3, are included in the present analyses so as to focus on characteristics most proximal to adoption (Session 1) and most proximal to session 5 and kindergarten outcomes (Session 4). Session 5 occurred around 5 years of age, and always occurred after session 4. At this session, a semi-structured interview was conducted to assess caregiver-reported DSE symptoms (PI families only). The kindergarten assessment always occurred after session 5. Children were visited in their kindergarten classroom between October and June (M age = 6.0 years, SD = 0.3) to observe various naturalistic social behaviors.

Table 1.

Demographic information and descriptive statistics by group.

	n PI; n NA	PI (total n = 94)	NA (total n = 52)	t-value	Cohen’s d
Sex, n (%) female	94; 52	54 (57.4)	26 (50.0)	—	—
S1 age in months, M (SD)	68; 52	26.3 (5.0)	28.5 (5.7)	−1.35	−0.25
Hypocortisolism (S1-S4), M(SD)	57; 49	0.51 (0.72)	−1.06 (0.67)	11.59***	2.24
S1 Parental Support, M (SD)	66; 51	5.56 (0.99)	5.69 (0.94)	−0.70	−0.13
S4 Parental Support, M (SD)	81; 47	5.42 (1.01)	5.60 (1.18)	−0.86	−0.16
S1 Parental Structure, M (SD)	66; 51	5.52 (1.00)	5.54 (0.96)	−0.09	−0.02
S4 Parental Structure, M (SD)	81; 47	5.71 (1.05)	5.76 (0.90)	−0.26	−0.05
S4 Physical DSE behaviors, M (SD)	82; 47	1.72 (2.67)	0.34 (0.84)	4.31***	0.63
S5 Caregiver-reported DSE symptoms, M (SD)	74; 0+	2.07 (2.35)	—	—	—
K social skills, M (SD)	51; 36	3.40 (0.57)	3.64 (0.32)	−2.47*	−0.49
K victimization, M (SD)	62; 40	0.23 (0.41)	0.11 (0.24)	1.81	0.33
K aggression, M (SD)	62; 40	0.27 (0.52)	0.07 (0.17)	2.84**	0.48
K behavior regulation, M (SD)	62; 40	3.24 (0.73)	3.68 (0.45)	−3.78***	−0.69

Note. PI: Previously institutionalized; NA: Non-adopted; S1: Session 1; S4: Session 4; S5: Session 5; K: Kindergarten; DSE: Disinhibited social engagement; DSED: Disinhibited social engagement disorder.

^*p < .05,

^**p < .01,

^***p < .001.

Significant group differences are indicated in bold.

⁺Caregiver-reported DSE symptoms were not collected with the non-adopted group, as clinical levels of DSE symptoms were expected to be non-existent in these children.

Measures

Salivary cortisol.

At sessions 1-4, children’s saliva samples were collected to assess two dimensions of children’s HPA functioning: cortisol reactivity to acute stressors and diurnal cortisol. Cortisol reactivity was assessed in response to a series of challenging laboratory tasks including brief separation from a caregiver, interactions with strangers, exposure to arousing stimuli, and electrophysiological assessments. Three saliva samples were collected at each session (25-35 min after arrival, 30 min after first sample, and 35 min after second sample). Diurnal cortisol saliva samples were collected by parents in the home at wake, midday, and bedtime for three days at each session (on average 2-3 weeks after the laboratory session). Trackcaps that record the times when the bottle containing the dental rolls was opened were used to check the accuracy of parents’ written reports of sampling time; parent-reported time averaged within 5 minutes of the trackcap time suggesting parent self-reported time was an accurate reflection of actual sampling time. All saliva samples were collected using dental rolls and a few grains of sweetened drink mix to stimulate the salivary glands (Talge et al., 2005). Samples were stored in the laboratory at −20°C before being shipped to University of Trier, Germany for assay in duplicate using a time-resolved fluorescence immunoassay (DELFIA). Intra- and inter-assay coefficients of variation were all less than 10%. Individual cortisol values that were beyond 4 SDs from the mean were winsorized (n = 16 lab samples, 3.7%; n = 46 diurnal samples, 3.5%).

Separate multilevel structural linear growth models were used to model cortisol reactivity and diurnal cortisol across all four sessions. All analyses controlled for time of sample collection, wake time, and medication use. Medication use was quantified using a method developed by Granger and colleagues (2009) and used frequently in the literature, along with other similar strategies (Shirtcliff et al., 2012). Children were excluded for fevers and corticosteroid use. The intercept and slope values were extracted from the between-subjects portion of the model reflecting child-level or trait-like cortisol. For cortisol reactivity, the intercept reflects children’s initial cortisol at arrival and the slope reflects cortisol reactivity to the laboratory tasks. For diurnal cortisol, the intercept reflects wake levels and the slope reflects diurnal change.

Trait-like variation in cortisol production (e.g., abnormally low/blunted) may be best assessed using multiple indicators of activity (diurnal and in response to laboratory challenges) over several time points (in this study, four sessions over two years). Thus, these four growth parameters (laboratory intercept/slope and diurnal intercept/slope) were entered into a principle component analyses with varimax rotation to assess whether they all load onto the same factor of trait-like cortisol activity. However, two factors were extracted from this analysis. The first factor reflected a hypocortisolism factor comprised of diurnal intercept (waking), diurnal slope (change across the day), and laboratory slope (reactivity to challenging tasks). Specifically, lower diurnal intercept, shallower diurnal slope, and shallower laboratory slope all indicated underactivity of the HPA axis, discussed here as hypocortisolism. Values were reversed such that higher scores on this factor reflected greater hypocortisolism. Laboratory intercept (arrival) loaded on its own factor, perhaps indicating a distinct construct like an anticipatory response to the session, and so was not included in this analysis. Detailed descriptions of sample collection, modeling, and creation of the hypocortisolism factor are available in Koss et al. (2016).

Parenting.

At sessions 1-4, each caregiver and child participated in an interaction task consisting of 10 minutes of free play and clean up, where the parent was told to play with their child as they normally would with a wide array of toys, followed by a knock on the door that signaled clean up. This was followed by 10 minutes of more structured play and clean up, in which the parent was told to help the child make a specific picture using Play-doh™ and related tools that were provided, until another knock signaled clean up again. The structured task was selected to be slightly difficult for the children, thus challenging the parents’ abilities to provide appropriate support and limit-setting. Between the free and structured play tasks, one saliva sample was also collected. Interactions were video-recorded to be later coded for caregivers’ supportive presence and structure/limit-setting each on a scale from 1-7, derived from an observational coding method designed by Sroufe and colleagues (Matas, Arend, & Sroufe, 1978). For supportive presence, higher scores indicate more positive involvement, attentiveness, and secure base behavior. A caregiver who scored high on this scale skillfully provided support throughout the session and set up the situation from the beginning as one in which s/he is confident of the child’s efforts. S/he appeared genuinely interested in and attentive to the needs of the child, and was not only emotionally supportive but responded contingently and appropriately to the child’s signals. Structure/effective limit-setting behaviors included clearly communicating expectations for the child’s behavior, following through on directives, providing effective discipline for noncompliance, and providing structure that allows the dyad to complete the task effectively. One score each for supportive presence and structure/effective limit-setting was given across the entire 25-minute task. The two scales were significantly correlated (r’s = .50-.61, p’s < .001). Coders held at least a bachelor’s degree and were blind to adoption history. They were trained to 80% agreement within 1 scale point and subsequently double-coded about 20% of cases to calculate reliability (92.0% agreement within 1 point; ICC = 0.84).

Disinhibited Social Engagement: Observed Behaviors.

At sessions 1-4, DSE behaviors were observed using a ten-minute scripted female stranger interaction derived from Tizard and Rees (1975), videotaped and later coded using ProCoder (Tapp, 2003) by trained undergraduate students blind to adoption history. After consent and before all other tasks, the stranger entered the playroom and made increasing social overtures (i.e. greeting the child, offering toys) at scripted intervals, culminating in interactive play (see Lawler et al., 2014 for details). The caregiver completed paperwork and was discouraged from interacting with the child or influencing his/her behavior toward the stranger.

Children’s verbal and non-verbal initiations and their proximity and physical contact to the stranger were scored. Child initiations were tallied in order to capture the child’s attempts to engage the stranger. Proximity within two feet of the stranger was coded for frequency of approach and duration in seconds. Physical contact was assessed by frequency, duration (seconds), and overall degree throughout the task on a scale from 0 (no physical contact) to 3 (familiar physical intimacy). Latency to first initiation, approach, and touch (in seconds) were also noted. After coding the complete 10-minute video recording, coders re-watched the same 10-minute video and rated the child’s initial reaction to the stranger’s greeting, toy offering, and play engagement, as well as overall verbal and physical intimacy. Following initial training to reliability of kappa = .80, 20% of observations were double-coded (inter-rater reliability averaged kappas of .71-.98, treating scales as ordinal).

Separately, experimenters rated each child for verbal and physical intimacy over the entire 2-hour session using the same coding scale, yielding two overall scores of DSE behavior for each of the four sessions. For example, experimenter-rated physical intimacy across the whole session was rated on a scale from 0 (none) to 5 (invaded privacy). A composite of physical DSED behaviors described above (5 items) has been found to distinguish PI children (see Lawler et al., 2014) and thus is used here. Session 4 DSE behaviors were used in the model, controlling for DSE at session 1, because continued DSE behaviors beyond the initial transition may be most indicative of behavioral patterns detrimental for later social functioning (Lawler et al., 2016).

Disinhibited Social Engagement: Caregiver Interview.

An interview adapted from the Disturbances of Attachment Interview (Smyke et al., 2002) and a semi-structured interview on attachment problems in PI children (O’Connor et al., 1999) was conducted at session 5 with parents by phone to measure DSE symptoms in PI participants. Interviewers probed for example behaviors and asked follow-up questions to characterize the child’s behavior in a variety of situations. Interviewers used anchored criteria to make ratings for each item: “0” (not or rarely present), “1” (somewhat or sometimes present), or “2” (definitely or often present). Four ratings of DSE symptoms included: absence of reticence/ willingness to go off with an unfamiliar adult [1], failure to check back with caregiver/tendency to wander off [2], and overly familiar verbal [3] and physical [4] behavior. Inter-rater reliability was calculated on 24% of interviews (ICC range .72-.92). The mean of the four ratings was calculated (α=.76) resulting in scores ranging from 0 (no symptoms in any category) to 2 (high levels of all symptoms).

Peers and Social Skills Questionnaire (PSSQ).

At the kindergarten assessment, an adapted version of the PSSQ (e.g. Kerr, Capaldi, Pears, & Owen, 2009) was completed by a trained observer (all with at least a bachelor’s degree with the exception of two upper-level undergraduate students) in the participating child’s kindergarten classroom for approximately 60 minutes (6 10-minute intervals of structured and unstructured time). Coders were initially trained at the department’s laboratory preschool, which was maintained via double-coding of video-recorded observations. Observers were blind to the child’s adoption history and did not influence the environment, the child, or their peers in any way during the observations. Naturalistic social behaviors were rated on a Likert scale from 1 (Never/Not true) to 5 (Frequently/Very true). Subscales included social skills (7 items, α = .85), victimization (3 items, α = .60), aggression towards others (3 items, α = .67), and behavior regulation (4 items, α = .86). The mean of all items for each subscale was calculated resulting in scores ranging from 0 (low levels of this type of behavior/characteristic) to 5 (frequent/high levels of this type of behavior/characteristic). Thus, high scores on the social skills and behavior regulation scales indicate positive adjustment while high scores on the victimization and aggression scales indicate maladjustment.

Data Analytic Plan

All analyses were conducted using the statistical software program R (R Core Team, 2016). For aim 1, group differences in hypocortisolism, DSE behaviors, and all subscales of kindergarten social competence were calculated using t-tests. Group differences in parenting were also assessed to show that group-level differences in the caregiving environment do not explain the results. Group differences on some of these variables have been reported in previous publications. The analyses were repeated here for context for the focal analyses; previous reports are cited whenever available. Sex was included as a covariate for observed and caregiver-reported DSE and social outcomes in all analyses. Session 1 observed DSE was included as a covariate for session 4 DSE to control for stability over time. Number of months since entering kindergarten was included as a covariate for all kindergarten social variables. Race and ethnicity were different across groups; however, these were not examined as covariates because they are largely confounded with group membership and previous analyses with this sample have shown no within-group differences by race or ethnicity (Koss et al., 2016).

For aim 2, a path analysis was conducted using the ‘lavaan’ package in R (Rosseel, 2012) to investigate the role of hypocortisolism on observed and caregiver-reported DSE behaviors and kindergarten social competence. Mediation was tested using a product-of-coefficients approach (Preacher & Hayes, 2008) with bootstrapped standard errors of the indirect effect of early adversity (i.e. PI vs. NA) on each kindergarten social outcome via adrenocortical functioning, session 4 DSE behaviors, and session 5 DSE symptoms (sequential mediation).

For aim 3, post-adoption parental support and structure/limit-setting were added to the model described in aim 2 to examine whether parenting quality would directly predict reduced hypocortisolism, and observed and caregiver-reported DSE. Session 1 (most proximal to adoption) and session 4 (most proximal to DSE measures) parenting were included simultaneously to investigate whether timing plays a role in parenting/DSE associations. For aim 4, we examined moderated mediation using the ‘mediation’ package in R (Tingley, Yamamoto, Hirose, Keele, & Imai, 2014) where parental support and structure (in separate models) moderated the indirect effect of early institutional care on DSE via adrenocortical functioning.

All covariances between kindergarten social competence scales were included. Group (PI vs. NA) was used as a dichotomous measure of early institutional care, though results remain similar when using age at adoption or duration of institutionalization in months instead of group (see Supplementary Material). Results also remain consistent when excluding participants in the PI group who spent less than 50% of their pre-adoptive life in an institution (results available upon request). Robust standard errors were used. Missingness did not differ based on child age, sex, or group (PI vs. NA), with the exception of session 5 DSE which was missing for all NA participants by design. Additionally, 26 PI participants were enrolled in the study at session 2 primarily due to logistical reasons related to their adoption. Missing data were handled using full-information maximum likelihood, which allows for using all available data from participants even if they have partial missing data.

Results

Aim 1: Group Differences

Descriptive statistics and intercorrelations for all focal variables can be found in Tables 1 and S1, respectively. As previously reported in Koss et al. (2016) and Lawler et al. (2014) respectively, PI children displayed significantly more hypocortisolism and more observed physical DSE behaviors than NA children. In the kindergarten classroom, observers rated PI children significantly lower on social skills and behavior regulation and higher on aggression than NA children. There were no group differences in experiences of victimization. See also Pitula et al. (2017) for related analyses. There were no group differences in parental supportive presence or structure/limit-setting at session 1 or session 4, suggesting that associations between parenting and other focal variables are not due to PIs experiencing generally better or worse parenting than NAs.

Aim 2: Precursors and Consequences of Disinhibited Social Engagement

Model results examining a developmental cascade from early institutional care to hypocortisolism, observed and caregiver-reported DSE, and kindergarten social outcomes are visualized as a path diagram in Figure 1. Group was strongly associated with hypocortisolism, such that PI children showed more blunted adrenocortical activity. Hypocortisolism significantly predicted more session 4 DSE behaviors, controlling for session 1 DSE, and subsequently more caregiver-reported DSE at session 5. DSE symptoms were then negatively associated with kindergarten social skills and behavior regulation and positively associated with victimization and aggression. Mediation from institutionalization (group) to each kindergarten social outcome via hypocortisolism, observed DSE, and caregiver-reported DSE were all non-significant.

Fig. 1.

Path diagram representing the developmental cascade model results. N = 146. CFI: .994; TLI: .989; RMSEA: .019. Solid paths are statistically significant and dotted paths are not statistically significant. Analyses are presented including sex as a covariate, as well as controlling for session 1 DSE behaviors and covariation between all kindergarten outcomes (not pictured for clarity)

Aims 3 and 4: Associations with Parental Support and Structure

Neither parental support nor structure at session 1 significantly predicted hypocortisolism; unexpectedly, the same was true for observed DSE and caregiver-reported DSE. Controlling for session 1 parenting, parental support but not structure at session 4 (most proximal to the measurement of observed/caregiver-reported DSE) positively predicted observed DSE behaviors, and these effects were significant (see Figure 2). Further, neither session 1 parenting variable moderated the direct effect of group on DSE nor indirect effect of group on DSE via hypocortisolism (all p’s > .32).

Fig. 2.

Path diagram representing the post-adoption parenting model results. N = 146. CFI: .911; TLI: .845; RMSEA: .058. Solid paths are statistically significant and dotted paths are not statistically significant. Analyses are presented including sex as a covariate, as well as controlling for session 1 DSE behaviors, months since entering kindergarten, and covariation between all kindergarten outcomes (not pictured for clarity)

Session 4 parental support moderated the direct effect of group on DSE (p = .04) but did not significantly moderate the indirect effect of group on DSE via hypocortisolism (p = .24). At −2SD of the mean of parental support, group was not significantly associated with DSE behaviors at session 4 (Estimate = .20, p = .96). At +2SD of the mean of parental support, group was significantly positively associated with session 4 DSE (Estimate = 3.13, p < .001), indicating that the PI group showed significantly more DSE behaviors at session 4 when their parents also showed concurrently higher support. Session 4 parental structure significantly moderated the direct effect of group on DSE (p = .02) and the indirect effect via hypocortisolism (p = .01). At −2SD of the mean of parental structure, group was significantly positively associated with later DSE via hypocortisolism (Estimate = 1.75, p < .001). At +2SD of the mean of parental structure, group still showed a direct positive association with DSE (Estimate = 2.63, p < .001), but the indirect effect of group on DSE via hypocortisolism was nonsignificant (Estimate = −1.75, p = .12). This suggests that, while PI children tend to show more DSE regardless of parental structure, higher structure reduces the association between hypocortisolism and DSE.

Next, we probed the unexpected positive association between session 4 parental support and observed DSE in light of the findings with parental structure. To do this, we examined whether an imbalance between parental support and structure, rather than support itself, was driving the positive association with DSE. It might be expected that high levels of support in the absence of effective structure/limit-setting becomes permissive of socially-inappropriate DSE behaviors and does not allow children to develop the regulatory skills necessary to avoid displaying DSE (LeCuyer-Maus, 2000; Rinaldi & Howe, 2012). Thus, we examined whether the residual of a caregiver’s session 4 support score regressed on their structure/limit-setting score (positive numbers = parenting unequally weighted towards supportive presence, negative numbers = parenting unequally weighted towards structure/limit-setting, zero = equal balance) predicted observed DSE behaviors. We also examined whether this relation was stronger with children exposed to institutional care. To test this, session 4 parenting residual was added to the parenting model described above (Figure 2) in place of the original parental support and structure scores (only predicting observed DSE, as this was the only significant association with parental support in Figure 2) and then moderated by group.

Indeed, session 4 parenting residual was positively associated with DSE behaviors (z = 2.21, p = .03), and this relation was moderated by group (z = −2.70, p = .01). For PI children, the association between parenting residual and DSE behaviors was stronger than for NA children (Figure 3). In other words, supportive parenting in the absence of effective limit-setting is associated with more DSE behaviors for children who experienced greater early adversity. No such associations were found at session 1.

Fig. 3.

Post-hoc moderation analysis depicting the association between parenting residual and DSE behavior, varying by group (previously institutionalized vs. non-adopted). Gray regions represent 95% confidence envelopes

Discussion

This study sought to investigate risk and promotive/protective factors toward and away from the development of DSE and subsequent social difficulties in the school setting following early adversity. Using a developmental cascade approach, this longitudinal study examined exposure to early institutional care, blunted cortisol activity, and early DSE behaviors as risk factors for developing DSE symptoms at age 5. Post-adoption parental support and structure/limit-setting were evaluated as promotive/protective factors. Finally, the association between age 5 DSE and subsequent kindergarten social difficulties was assessed. Overall, we found a risk pathway toward DSE symptoms such that exposure to institutional care predicted more blunted cortisol activity, which predicted increased observed DSE behaviors and in turn more caregiver-reported DSE symptoms at age 5 (Figure 1 and Supplementary Material). Additionally, DSE symptoms predicted more aggression and victimization and less social skills and behavior regulation in kindergarten. These results contribute to the field’s understanding of the etiology of DSE and point to the role of adrenocortical dysfunction as one potential explanation for the persistence in DSE symptoms following adoption.

This study was, to the best of our knowledge, one of the first to find an association between HPA axis activity and DSE behaviors (see also Johnson et al., 2011). Blunted cortisol activity and DSE are two of the most consistent outcomes associated with exposure to institutional care (Koss et al., 2016; Lawler et al., 2016; McLaughlin et al., 2015; Miron & Zeanah, 2017) and there is substantial evidence linking cortisol activity with various social and regulatory outcomes (Isaksson et al., 2012; Koss et al., 2016; Pitula et al., 2017). Furthermore, the HPA axis is profoundly sensitive to the caregiving environment, particularly in infancy and toddlerhood (Hostinar, Sullivan, & Gunnar, 2014). Because DSE symptoms are potentially impacted by both a disrupted early caregiving environment and general deficits in self-regulation (Bruce, Tarullo, & Gunnar, 2009; Gorter et al., 2017; Lehmann et al., 2016), the current findings suggest HPA activity might be an informative biomarker for DSE and its consequences over time. Although the overall mediation analysis was nonsignificant, this developmental pathway was moderated by parental structure. Particularly for children whose parents display lower levels of structure/limit-setting, dysregulated cortisol activity may be a biological mechanism through which DSE behaviors persist, which would provide a promising avenue for prevention and intervention. Future research should further explore how HPA axis dysfunction contributes to the etiology of DSE symptoms in children who have experienced different forms of adversity, and how different parenting behaviors play a role in preventing or mitigating symptoms over time.

Interestingly, parenting immediately following adoption from institutions (~ 2 months) was not associated with hypocortisolism or DSE. Contrary to the hypothesis, later parental support (~2 years after adoption, controlling for early parenting) predicted more DSE behaviors (Figure 2) particularly for PI children. Previous studies examining associations between parenting and DSE found mixed results, sometimes finding a negative association (Lyons-Ruth et al., 2009) while others find a positive (Dobrova-Krol et al., 2010) or no direct association (Garvin et al., 2012; Zeanah et al., 2005). The current study suggests that higher levels of supportive parenting two years after adoption may be permissive of socially inappropriate DSE behaviors, thus reducing the likelihood of remittance.

Moderated mediation analyses showed that parental support did not moderate the indirect effect of institutional care on DSE via hypocortisolism, but parental structure at session 4 did. This is particularly intriguing, because it suggests a reduction of the association between adrenocortical activity and DSE behaviors in the context of high parental structure that is not seen with relatively lower amounts of structure/limit-setting. It is unclear why parental structure would influence this pathway and not support, especially given substantial research showing that parental support predicts better adrenocortical functioning (e.g., Bernard et al., 2015). However, consistent with our previous reports indicating the persistence of hypocortisolism despite marked improvements in caregiving for PI children (Koss et al., 2016), normal variations in parenting are likely not sufficient to ameliorate the impact of chronic deprivation during infancy, a time that appears to be a sensitive period of setting HPA axis activity (McLaughlin et al., 2015; Meaney & Szyf, 2005). It is heartening, however, that even if adrenocortical dysregulation is not improved, it may be possible to reduce its ultimate impact on social and behavioral functioning.

This study highlighted multiple nuances in the influence of caregiving on the etiology of DSE following international adoption. Different patterns may emerge based on the timing of assessment of parenting relative to the exposure to adversity and/or developmental stage. Further, it may be that the combination of support and structure is more important than either construct on its own. This is corroborated by our post-hoc analyses showing that the residual of the supportive presence score regressed on the structure/limit-setting score positively predicted DSE behaviors. Specifically, positive residuals (i.e., greater supportive parenting without similarly high levels of structure and limit-setting) two years post-adoption was associated with more concurrent DSE behaviors, controlling for earlier parenting and DSE. This association was further moderated by exposure to institutional care such that the parenting residual more strongly predicted DSE for PI children compared to NA children. This suggests that, particularly for children who have experienced severe early adversity, supportive parenting that is not coupled with effective structure/limit-setting puts children at risk for higher rates of DSE symptoms.

These findings may be viewed in light of the classic literature by Baumrind and colleagues (1971) arguing that authoritative parenting, defined as a combination of high warmth and high control, is associated with the best developmental outcomes and provides better context for previous studies that have found a positive association between supportive parenting and DSE (e.g., Dobrova-Krol et al., 2010). Children of permissive parents tend to show problems with self-regulation (Piotrowski, Lapierre, & Linebarger, 2013) that may be partially explained by adrenocortical dysfunction, at least in children exposed to adversity, providing a common pathway to DSE. We do not intend to argue that supportive parenting is bad for children; however, it may be particularly important for adoptive parents to balance support and structure in order to promote better socio-emotional outcomes for their children. Parenting interventions for adoptive parents of PI children should promote both support and structure, corroborating evidence from two case studies that behavior management techniques might be useful specifically for children with clinical levels of DSE (Buckner, Lopez, Dunkel, & Joiner Jr., 2008; Dickmann & Allen, 2017). These clinical implications are echoed by a recent conversation about trauma-informed discipline suggesting that common disciplinary strategies like time-out can be harmful for children who have experienced trauma when administered punitively (e.g., high control, low warmth) but may actually promote the parent-child relationship and the child’s mental health when used sensitively (e.g., high control, high warmth; Dadds & Tully, 2019).

We also acknowledge that this pattern may go in the opposite direction, such that DSE behaviors with other adults are interpreted as overly friendly and affectionate, eliciting warmth and support from the caregiver to try and redirect the child’s attention towards themselves. This process might be considered an adaptive response while the child is still in the institution, increasing their likelihood of positive attention from caregivers and their chances of getting adopted out of the institution (Chisholm, 1998). However, upon leaving, this may become maladaptive when children display these behaviors to strangers and other adults outside the caregiver-child relationship. DSE behaviors violate the social norms of children in the U.S. and may strain relationships with peers and other adults. Thus, setting limits for PI children’s behavior (while maintaining high levels of support) may help children develop regulatory skills that over time promote reductions in DSE behaviors and better socio-emotional outcomes.

In this study there were several direct effects that, when examined in a path analysis framework, create a pathway from early institutional care to blunted cortisol activity, to DSE behaviors and symptoms, and finally to kindergarten social outcomes. However, when indirect effects were tested, none emerged as significant. There are several possibilities for why this is the case. First, our modest sample size of 146 and relatively complex path model may have precluded our ability to detect potentially small effects. Second, it is possible that these variables are only associated through a third common predictor and are not actually a part of a single developmental pathway as described above. In other words, cortisol and DSE may only be associated with each other because they are both strongly associated with some other unmeasured variable. Lastly, it may be that a significant mediating pathway is only present for children experiencing poorer post-adoption parenting, as indicated by the moderated mediation with parental structure. Future research, especially with larger sample sizes, is needed to better examine the potential mechanisms through which hypocortisolism and parenting give rise to higher rates of DSE in children who have experienced severe early adversity.

The strengths of this study include its longitudinal approach, which was critical to understanding how pathways toward and away from DSE unfold over time. Similarly, the study began approximately two months after the PI children were internationally adopted from institutions, providing insight into the importance of factors immediately following adoption that promote or hinder resilience in children exposed to institutional care. Having multiple assessments of parenting behavior (2 months to 2 years post-adoption) allowed us to tease apart timing effects of parenting on DSE that may shed light on the mixed extant literature regarding the role of caregiving. Also, this study is strengthened by its multi-method, multi-informant measures including physiological assessments, observations with the parent and in the kindergarten classroom, and structured clinical interviews. This study provides important preliminary information about precursors and consequences of DSE, allowing future researchers and clinicians to apply this knowledge in the development of prevention, intervention, and treatment programs for children who have experienced disrupted early caregiving environments.

While this study has many strengths, there were also several limitations that need to be considered. First, there was no available measure of parenting concurrent to the assessment of caregiver-reported DSE symptoms or kindergarten social outcomes. However, parenting behavior is commonly quite stable over time and is stable across the four available parenting assessments in this study, so session 4 parenting was likely an appropriate proxy for caregiving behavior roughly one year later. Additionally, there is no way to reliably measure prenatal influences or all of the pre-adoptive environmental factors that may contribute to or explain the associations described above. Thus, it is unclear which specific aspects of institutional care increase risk for these maladaptive outcomes. This also limits the generalizability of findings from PI children to children who have experienced other types of adversity. Further, there was significant variability in countries of origin for the children in the PI group, and thus differences in the nature of each child’s institutional care experience, though this was not able to be explored or controlled for with the small sample size of the current study. Notably, this variation was likely to add variability to the data, making it more difficult to identify consistent patterns. Finally, the measure of hypocortisolism was a summary measure of both diurnal and laboratory cortisol across four sessions, spanning two years of data collection. This hinders our ability to make specific conclusions regarding changes in HPA activity over time and its relation to the development of DSE, though previous studies from this group have demonstrated the measure’s predictive value for behavior over time (Koss et al., 2016). The use of this measure may conversely be interpreted as a strength, as it represents trait-like patterns of cortisol activity that were found to be fairly stable over the first few years post-adoption (Koss et al., 2016).

Overall, this study highlights reduced adrenocortical activity as a risk factor and combined parental support and structure as a protective factor in the developmental pathways of DSE following exposure to early institutional care. This study also provides insight into the complex relations between the post-adoption caregiving environment and later DSE behaviors. Furthermore, these findings add to the body of literature demonstrating the impact of DSE on later socio-emotional outcomes. Future research should replicate these developmental pathways and investigate potential mechanisms that can be leveraged for prevention, intervention, and treatment with children who experience various forms of disrupted early caregiving.